Declared a public health emergency of international concern by the World Health Organization, coronavirus disease 2019 (COVID-19) has continuously posed a threat to both physical and psychological health.1 Beginning in December 2021, the Omicron variant triggered the fifth wave of the pandemic in Hong Kong, endangering psychological well-being.1 2 Filipina domestic workers (FDWs), the primary group of migrant domestic workers, constitute >2.5% of the Hong Kong population3 and are considered a vulnerable population. Before the Hong Kong government reiterated the rights of migrant workers, many FDWs faced mistreatment, including abuse, exploitation, and illegal dismissal upon infection with severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2).4 5 6 Filipina domestic workers were susceptible to both direct and indirect consequences of the COVID-19 pandemic.

Migrant workers often experience poor psychosocial conditions and substandard working environments.4 5 6 7 8 However, few studies have consistently examined the well-being of FDWs.8 9 10 Anxiety, a key indicator of well-being, commonly coexists with other psychological conditions. Considering the large number of domestic workers in Hong Kong, efforts to safeguard the psychological health of this minority population are essential to prevent excessive strain on the healthcare system.11 Additionally, various aspects of working conditions should be investigated in relation to anxiety.12

This study aimed to estimate the prevalence of anxiety and examine its relationships with perceived social support and decent work among FDWs after the peak of the Omicron wave during the COVID-19 pandemic in Hong Kong. Insights regarding the psychosocial conditions encountered by FDWs during the aftermath of the pandemic may contribute to existing literature.

A cross-sectional survey, written in English, was administered between June and August 2022. The target population comprised FDWs. Eligibility criteria included age ≥18 years, ability to read and understand English, and ability to provide informed consent. Filipina domestic workers who began employment on or after 1 February 2022 in Hong Kong, as well as male FDWs, were excluded from the present study. Because the majority of FDWs are women (97.8%), the inclusion of a small sample of male FDWs could compromise representativeness.3

Convenience sampling was utilised. Recruitment was conducted at gathering places in Central, Hong Kong, where a large proportion of FDWs spend their days off. Data collection was performed on rest days (Sundays and statutory holidays). Support and clarifications were provided to respondents who required assistance in understanding the questions. Respondents were offered a gratuity of HK$20 in cash as a token of appreciation for their time and assistance. According to Yeung et al,10 the prevalence of anxiety among FDWs in Hong Kong at the beginning of the pandemic was 25%. With a 95% confidence interval (95% CI) and a desired margin of error of ±5%, the minimum required sample size was estimated to be 289.

The questionnaire consisted of four sections, namely, anxiety, perceived social support, working conditions, and potential covariates (eg, socio-demographic and health-related factors). The questionnaire was developed based on validated instruments and a literature review of similar contexts.12 13 14 15 16 17

The Generalised Anxiety Disorder-7 (GAD-7) scale was adopted to assess anxiety levels.13 The total score ranges from 0 to 21; a threshold score of ≥10 to identify self-reported anxiety provides optimal sensitivity (89%) and specificity (82%).13 The GAD-7 has demonstrated high internal consistency in the general population (Cronbach’s alpha=0.92) and among FDWs working in Chinese regions (Cronbach’s alpha=0.80).18 19

The Multidimensional Scale of Perceived Social Support, using a 7-point Likert scale, was used to measure perceived social support across three domains, namely, significant others, family, and peers.14 Each domain comprises four items. We calculated a mean score for each domain ranging from 1 to 7 and a total mean score averaged from the three concerned domains to represent the total score of perceived social support. A higher score indicates a greater level of perceived social support. The authors of the scale proposed multiple approaches for interpreting perceived social support, one of which involved analysing continuous data for the three domains and the overall score.14 This scale has been validated with high internal consistency among Southeast Asian domestic workers in Hong Kong (Cronbach’s alpha=0.96).16 20

The Decent Work Scale was adopted to evaluate working conditions, including 15 items grouped into five components, namely, physically and interpersonally safe working conditions, access to essential healthcare support, sufficient income, adequate rest time, and alignment of working settings with social values.12 Each item scored from 1 to 7, resulting in component scores ranging from 3 to 21 and a total score ranging from 15 to 105, with higher scores indicating better working conditions. This scale has been validated with high internal consistency among the working population in the US (Cronbach’s alpha=0.86).12

Statistical analysis was performed using SPSS (Windows version 27.0; IBM Corp, Armonk [NY], US). Confidence intervals were established at the 95% level, and P values <0.05 were considered statistically significant. We computed 95% CIs for anxiety prevalence. Socio-demographic variables were compared between anxiety statuses using the Chi squared test, whereas scores for perceived social support and working conditions were compared using the independent samples t test.

Odds ratios (ORs) with 95% CIs were computed using a binary logistic regression model. For univariable analysis, simple logistic regression was conducted; perceived social support and working conditions constituted the main independent variables. Multivariable logistic regression analysis was performed to estimate the independent effects of these variables while adjusting for potential confounders.

The GAD-7 scores were categorised into four levels of anxiety severity: minimal (0-4), mild (5-9), moderate (10-14), and severe (15-21).13 We conducted sensitivity analysis using the GAD-7 score as an ordinal outcome and constructed an ordinal logistic regression model to assess the robustness of previously identified anxiety-associated factors.

Among the 441 FDWs approached, 71 declined to participate, yielding a response rate of 83.9% (Fig 1). Primary reasons for refusal were survey length and time constraints. The distribution of GAD-7 scores was positively skewed (Fig 2). The estimated prevalence of probable anxiety (GAD-7 score ≥10) was 8.6% (95% CI=5.8-11.5). Among the 370 respondents, approximately half were aged 35 to 44 years (51.1%) and married (48.4%). Most respondents had attained a university-level education or higher (60.3%), reported a monthly income ranging from HK$4630 to HK$4999 (68.6%), and had children residing in their home country (82.7%). The proportions of respondents residing on Hong Kong Island, in Kowloon and the New Territories were evenly distributed. The median (interquartile range) duration of employment in Hong Kong was 5.0 years (interquartile range, 3.0-9.0). Most respondents had no history of COVID-19 (81.9%) and no chronic diseases (97.8%) [Table 1]. Table 2 shows that the mean scores for the three domains of perceived social support ranged from 5.5 to 5.7 out of 7, whereas the mean score for decent work was 78.1 out of 105. Among the five components of working conditions measured by the Decent Work Scale, the lowest mean score was observed for rest periods (14.1); access to healthcare had the highest mean score (17.1).

Participants with probable anxiety had a higher proportion of chronic diseases relative to those without anxiety (9.4% vs 1.5%; P=0.024) [Table 1]. Respondents with probable anxiety reported worse perceptions of social support and working conditions; they had lower scores across all domains relative to those of respondents without anxiety (Table 2).

Simple logistic regression analysis indicated that one domain of perceived social support and multiple subscales of working conditions were significantly associated with anxiety (Table 3). Filipina domestic workers with higher perceived social support from significant others, better access to healthcare, greater satisfaction with compensation and salary, increased free time and rest periods, and higher satisfaction with their employer’s value orientation exhibited a lower likelihood of experiencing probable anxiety. Multivariable logistic regression analysis—adjusted for all relevant socio-demographic variables, health status, and subscales of perceived social support and working conditions—identified three variables that remained statistically significant (Table 3). Greater satisfaction with compensation and salary (adjusted odds ratio [aOR]=0.825, 95% CI=0.728-0.935), increased free time and rest periods (aOR=0.878, 95% CI=0.780-0.987), and higher satisfaction with value orientation (aOR=0.887, 95% CI=0.796-0.989) were associated with lower anxiety risk. Sensitivity analysis, which examined the four levels of anxiety as an ordinal outcome using an ordinal logistic regression model, showed that effect estimates were slightly attenuated. However, the findings confirmed the association between anxiety levels and inadequate compensation, while also identifying a history of chronic diseases as a risk factor for increased anxiety severity (Table 4).

The observed prevalence of anxiety among FDWs was 8.6%, representing a lower proportion compared with previous studies.10 11 21 22 The Omicron variant led to an unprecedented surge in cases, which peaked in early March 2022. Compared with a local study conducted at the onset of the COVID-19 pandemic,10 the prevalence of probable anxiety among FDWs declined from 25% to 8.6%. A remarkably lower prevalence of anxiety was observed when using the official cut-off score of ≥7 for the Anxiety subscale of the Depression, Anxiety, and Stress Scale-21 Items (DASS-21-A) in both the general population of Hong Kong (14%)11 and the Philippines (38.4%).21 In Singapore, 17.5% of migrant workers exhibited probable anxiety (DASS-21-A score ≥8).22 The discrepancy in anxiety prevalence across studies may be attributed to differences in study contexts and timeframes. Although the fifth wave of COVID-19 had nearly subsided in Hong Kong during the present study period, other regions were still experiencing high caseloads. The relatively low prevalence of anxiety among FDWs may indicate the development of psychological resilience after the Omicron pandemic. Additionally, information dissemination and vaccine availability were more established compared with the second and third waves of the pandemic.10

In response to the fifth wave of the COVID-19 pandemic, the local government implemented comprehensive public health policies to safeguard rights and facilitate risk communication among minority populations in Hong Kong. Coronavirus disease 2019 and vaccine-related information were made available in multiple languages, including Tagalog and English, thereby improving access to formal and accurate health information for FDWs. Access to adequate and accurate health information is essential for mitigating psychological distress and reducing anxiety levels associated with the pandemic, as demonstrated by the findings of a study conducted in the Philippines.21

Access to COVID-19 vaccines may partially explain the findings. In Hong Kong, domestic workers were designated as a priority group for vaccination within 1 month of launching the COVID-19 vaccination programme.23 Furthermore, the initial procurement of 22.5 million vaccine doses ensured sufficient supply for the entire population, allowing domestic workers to choose between Sinovac and BioNTech vaccines at no cost. The high effectiveness of COVID-19 vaccination may have contributed to anxiety reduction. As of August 2021, the majority of sampled domestic workers (80%) had received at least one dose of COVID-19 vaccine.24 A study by McMenamin et al25 demonstrated the substantial protective effect of COVID-19 vaccines against severe or fatal outcomes (BioNTech: two doses=83.9%; three doses=97.9%). Vaccination significantly reduces the risk of severe COVID-19 complications, hospitalisation, and mortality, which may have indirectly alleviated probable anxiety among FDWs. This assumption is supported by the results of a study examining the psychological impact of COVID-19 vaccination, which revealed lower anxiety levels among vaccinated individuals.26 However, the aforementioned local10 11 20 and Singapore studies22 assessing the anxiety of migrant workers were conducted during periods when no pharmaceutical preventive measures were available. Therefore, access to COVID-19 vaccines is a plausible explanation for the lower prevalence of probable anxiety among FDWs.

Additionally, job security may explain the decline in probable anxiety. Some FDWs expressed concerns regarding job insecurity and experienced distress due to job loss.4 Amid increasing reports of illegal contract terminations, the government intervened to uphold FDWs’ employment rights.27 On 5 March 2022, a government spokesperson emphasised zero tolerance for employers who illegally dismissed FDWs exhibiting SARS-CoV-2 infection.27 Any violation of the Employment Ordinance and related laws was subject to prosecution and fines.27 Filipina domestic workers exhibiting SARS-CoV-2 infection or identified as close contacts of individuals with COVID-19 receive the same assistance and support as other Hong Kong citizens, including quarantine and isolation arrangements.27 Greater institutional support for their employment may have contributed to the lower prevalence of anxiety among FDWs.

The significant others domain of perceived social support was negatively associated with anxiety in univariable analysis but was no longer significant according to multivariable regression. Significant others are individuals that the respondents regard as special persons.12 This finding contrasts with previous studies that identified perceived social support as an essential factor in coping with psychological distress among migrant workers.8 9 This discrepancy may be attributable to the small sample size. However, the finding is consistent with results from a local study conducted in a similar context.10

Filipina domestic workers migrate to foreign countries to support their families’ livelihoods; they are often portrayed as resilient and independent figures by the Philippine Government. This narrative may subtly reinforce the perception among FDWs that they are the sole breadwinners responsible for their families’ well-being.28 Consequently, although FDWs may seek informal social support from significant others, their self-disclosure remains selective. Psychological concerns, in particular, may be considered sensitive topics, leading to avoidance of such discussions in an effort to protect their self-esteem. This avoidance may explain the absence of an observed association between perceived social support and anxiety.

Another key finding was that better working conditions—including greater satisfaction with compensation and salary, increased free time and rest periods, and higher satisfaction with value orientation—were associated with a lower likelihood of probable anxiety. Working conditions are recognised as social determinants of mental health. Findings from the World Health Organization suggest that jobs offering high rewards and a greater sense of control serve as protective factors for mental well-being, thereby reinforcing the importance of favourable working conditions for employees.29 Consistent with the previous findings,30 high and regular monetary compensation was linked to lower probable anxiety in our study. According to the Occupational Wages Survey in the Philippines,30 the median monthly income was PHP13 646 (HK$1865, US$239), whereas the minimum monthly wage in Hong Kong was HK$4630 (US$594) during the study period.31 Filipina domestic workers in Hong Kong earned at least 2.48-fold more than their counterparts in the Philippines. Higher monthly earnings are often allocated toward property purchases in the Philippines, meeting family obligations, and fulfilling roles and responsibilities. Thus, greater satisfaction with compensation and salary may have contributed to lower probable anxiety among FDWs. Although this factor may explain the observed association, a qualitative study would provide deeper insights into the relationship between higher compensation and reduced psychological distress.

Additionally, increased free time and rest periods were associated with a lower risk of probable anxiety. An occupational health study32 established an inverse relationship between working hours and sleep duration, where anxiety and depression scores were higher among individuals working longer hours. These findings suggest that increased free time and rest periods can help reduce anxiety risk.

Notably, greater alignment between FDWs’ working environments and their social values was associated with lower anxiety risk. Value orientation refers to the principles an individual upholds, including ethics, morality, and attitudes toward work. In the workplace, each aspect of the working environment is interconnected with FDWs and their employers, influencing the likelihood of psychological distress. Employers are encouraged to engage in discussions with FDWs regarding working conditions—such as job demands and task restructuring—to ensure alignment in value orientation between both parties.

While chronic disease was not a statistically significant predictor of anxiety in multivariable logistic regression model, sensitivity analysis using an ordinal outcome revealed that it remained a risk factor for increased anxiety severity. Despite the inconclusive findings regarding this association, a systematic review33 indicated that a history of chronic diseases is linked to higher anxiety levels. The presence of chronic diseases has a negative impact on mental health.33

Some limitations were inherent in our sampling method and study design. First, we could not establish causality. Because cross-sectional study designs provide only short-term data regarding associations, longitudinal studies are needed to examine temporal sequences and causal relationships. Second, the use of convenience sampling may introduce selection bias; therefore, generalisations of the findings to the entire FDW population should be made with caution. However, this bias is likely minimal because all FDWs were approached, and none were selectively invited based on specific characteristics; also, the demographic distribution of the sample closely resembled that of domestic workers recorded in the Hong Kong Population Census.34 The age distributions in the Census data34 and the study sample were comparable: 18-34 years (29.8% vs 27.8%), 35-44 years (48.2% vs 51.1%), and ≥45 years (22.0% vs 21.1%). Additionally, the respondents’ residence areas were evenly distributed across Hong Kong Island, Kowloon, and the New Territories. These findings suggest high representativeness and generalisability in the study sample. Furthermore, monetary incentives were provided, which may have contributed to higher-quality responses.

This study identified associations between optimal working conditions and lower probable anxiety among FDWs. The findings update the estimated prevalence of anxiety in this population and suggest that favourable working conditions may serve as protective factors. The study provides insights for the development and refinement of public health measures and occupational policies related to migrant workers, including compensation for overtime work, job security, and adequate rest periods. Psychological interventions tailored to domestic workers should be developed to address diverse mental health needs while incorporating labour protection. Regular review and refinement of occupational policies may be necessary. The Labour Department could consider conducting large-scale quantitative surveys and qualitative interviews with domestic workers to assess and accommodate their occupational needs. Future studies should aim to include domestic workers of various nationalities and other migrant worker populations.

Concept or design: KKY Lai, ELY Wong.
Acquisition of data: KKY Lai.
Analysis or interpretation of data: All authors.
Drafting of the manuscript: KKY Lai.
Critical revision of the manuscript for important intellectual content: ELY Wong.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

The authors thank Prof Marc KC Chong, Ms Annie WL Cheung and Mr Jonathan CH Ma from the Centre for Health Systems and Policy Research, The Jockey Club School of Public Health and Primary Care, The Chinese University of Hong Kong for their valuable comments on the study and support in data analysis. The authors also thank all study respondents for their valuable time in completing the questionnaires and for their contributions as migrant workers in Hong Kong.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This research was approved by the Survey and Behavioural Research Ethics Committee of The Chinese University of Hong Kong, Hong Kong (Ref No.: 018-22). The study was conducted in accordance with the principles of the Declaration of Helsinki. Informed consent was obtained from the participants prior to commencement of the survey.

1. Centre for Health Protection and Hospital Authority, Hong Kong SAR Government. Statistics on 5th wave of COVID- 19 (from 31 Dec 2021 up till 31 May 2022 00:00). Available from: https://www.coronavirus.gov.hk/pdf/5th_wave_statistics/5th_wave_statistics_20220531.pdf. Accessed 5 Dec 2022.

2. Xiong J, Lipsitz O, Nasri F, et al. Impact of COVID-19 pandemic on mental health in the general population: a systematic review. J Affect Disord 2020;277:55-64. Crossref

3. Census and Statistics Department, Hong Kong SAR Government. 2021 Population Census. Main Results. Available from: https://www.census2021.gov.hk/doc/pub/21c-main-results.pdf. Accessed 1 Apr 2025.

4. Chow Y. No home away from home for domestic workers terminated after contracting coronavirus amid Hong Kong’s fifth wave. Young Post. South China Morning Post; 2022 May 16. Available from: https://www.scmp.com/yp/discover/news/hong-kong/article/3177657/no-home-away-home-domestic-workers-terminated-after. Accessed 4 Dec 2022.

5. Cheung JT, Tsoi VW, Wong KH, Chung RY. Abuse and depression among Filipino foreign domestic helpers. A cross-sectional survey in Hong Kong. Public Health 2019;166:121-7. Crossref

6. Choy CY, Chang L, Man PY. Social support and coping among female foreign domestic helpers experiencing abuse and exploitation in Hong Kong. Front Commun 2022;7:1015193. Crossref

7. Sterud T, Tynes T, Mehlum IS, et al. A systematic review of working conditions and occupational health among immigrants in Europe and Canada. BMC Public Health 2018;18:770. Crossref

8. Ioannou M, Kassianos AP, Symeou M. Coping with depressive symptoms in young adults: perceived social support protects against depressive symptoms only under moderate levels of stress. Front Psychol 2019;9:2780. Crossref

9. Straiton ML, Aambø AK, Johansen R. Perceived discrimination, health and mental health among immigrants in Norway: the role of moderating factors. BMC Public Health 2019;19:325. Crossref

10. Yeung NC, Huang B, Lau CY, Lau JT. Feeling anxious amid the COVID-19 pandemic: psychosocial correlates of anxiety symptoms among Filipina domestic helpers in Hong Kong. Int J Environ Res Public Health 2020;17:8102. Crossref

11. Choi EP, Hui BP, Wan EY. Depression and anxiety in Hong Kong during COVID-19. Int J Environ Res Public Health 2020;17:3740. Crossref

12. Duffy RD, Allan BA, England JW, et al. The development and initial validation of the Decent Work Scale. J Couns Psychol 2017;64:206-21. Crossref

13. Spitzer RL, Kroenke K, Williams JB, Löwe B. A brief measure for assessing generalized anxiety disorder: the GAD-7. Arch Intern Med 2006;166:1092-7. Crossref

14. Zimet GD, Dahlem NW, Zimet SG, Farley GK. The Multidimensional Scale of Perceived Social Support. J Pers Assess 1988;52:30-41. Crossref

15. International Organization for Migration. The Determinants of Migrant Vulnerability. Geneva: United Nations; 2019. Available from: https://www.iom.int/sites/g/files/tmzbdl486/files/our_work/DMM/MPA/1-part1-thedomv.pdf. Accessed 5 Dec 2022.

16. Garabiles MR, Lao CK, Yip P, Chan EW, Mordeno I, Hall BJ. Psychometric validation of PHQ-9 and GAD-7 in Filipino migrant domestic workers in Macao (SAR), China. J Pers Assess 2020;102:833-44. Crossref

17. Mendoza NB, Mordeno IG, Latkin CA, Hall BJ. Evidence of the paradoxical effect of social network support: a study among Filipino domestic workers in China. Psychiatry Res 2017;255:263-71. Crossref

18. Yeung NC, Kan KK, Wong AL, Lau JT. Self-stigma, resilience, perceived quality of social relationships, and psychological distress among Filipina domestic helpers in Hong Kong: a mediation model. Stigma Health 2021;6:90-9. Crossref

19. Pan American Health Organization. Questionnaire to Assess the Diagnosis and Treatment of Chronic Diseases. Geneva: World Health Organization. Available from: https://www.paho.org/hq/dmdocuments/2009/cncd_mgt_questionnaire.pdf. Accessed 4 Dec 2022.

20. Leung DD, Tang EY. Correlates of life satisfaction among Southeast Asian foreign domestic workers in Hong Kong: an exploratory study. Asian Pac Migr J 2018;27:368-77. Crossref

21. Tee ML, Tee CA, Anlacan JP, et al. Psychological impact of COVID-19 pandemic in the Philippines. J Affect Disord 2020;277:379-91. Crossref

22. Saw YE, Tan EY, Buvanaswari P, Doshi K, Liu JC. Mental health of international migrant workers amidst large-scale dormitory outbreaks of COVID-19: a population survey in Singapore. J Migr Health 2021;4:100062. Crossref

23. Labour Department, Hong Kong SAR Government. Foreign domestic helpers. Vaccination priority groups to be expanded to cover people aged 30 or above. 2021 Mar 15. Available from: https://www.fdh.labour.gov.hk/en/news_detail.html?year=2021&n_id=190. Accessed 28 Mar 2025.

24. Sumerlin TS, Kim JH, Wang Z, Hui AY, Chung RY. Determinants of COVID-19 vaccine uptake among female foreign domestic workers in Hong Kong: a cross-sectional quantitative survey. Int J Environ Res Public Health 2022;19:5945. Crossref

25. McMenamin ME, Nealon J, Lin Y, et al. Vaccine effectiveness of one, two, and three doses of BNT162B2 and CoronaVac against COVID-19 in Hong Kong: a population-based observational study. Lancet Infect Dis 2022;22:1435-43. Crossref

26. Babicki M, Malchrzak W, Hans-Wytrychowska A, Mastalerz-Migas A. Impact of vaccination on the sense of security, the anxiety of COVID-19 and quality of life among polish. A nationwide online survey in Poland. Vaccines (Basel) 2021;9:1444. Crossref

27. Hong Kong SAR Government. Government’s response on situation of foreign domestic helpers affected by COVID-19 (with photos) [press release]. 2022 Mar 5. Available from: https://www.info.gov.hk/gia/general/202203/05/P2022030500399.htm. Accessed 4 Dec 2022.

28. Rich GJ. Filipina migrant domestic workers in Asia: mental health and resilience. In: Rich GJ, Jaafar JL, Barron D, editors. Psychology in Southeast Asia: Sociocultural, Clinical, and Health Perspectives. London: Routledge, Taylor & Francis Group; 2020. Crossref

29. World Health Organization. Social determinants of mental health. 2014 May 18. Available from: https://www.who.int/publications/i/item/9789241506809. Accessed 1 Apr 2025.

30. Mapa DS. Average monthly wage rates of selected occupations: 2018 and 2020 [Internet]. 2020 Occupational Wages Survey (OWS). Philippine Statistics Authority; 2022. Available from: https://psa.gov.ph/statistics/occupational-wages-survey/node/168472. Accessed 4 Dec 2022.

31. Hong Kong SAR Government. Minimum allowable wage and food allowance for foreign domestic helpers [press release]. 2021 Sep 30. Available from: https://www.info.gov.hk/gia/general/202109/30/P2021093000329.htm. Accessed 4 Dec 2022.

32. Afonso P, Fonseca M, Pires JF. Impact of working hours on sleep and mental health. Occup Med (Lond) 2017;67:377-82. Crossref

33. Clarke DM, Currie KC. Depression, anxiety and their relationship with chronic diseases: a review of the epidemiology, risk and treatment evidence. Med J Aust 2009;190:S54-60. Crossref

34. Census and Statistics Department, Hong Kong SAR Government. 2021 Population Census: Summary Results. 2022. Available from: https://www.censtatd.gov.hk/en/data/stat_report/product/B1120106/att/B11201062021XXXXB01.pdf. Accessed 4 Dec 2022.

Low back pain (LBP) is a prevalent health condition that can have disabling effects on individuals of all ages.1 This condition also imposes substantial socio-economic costs, as evidenced by studies demonstrating its impacts on healthcare systems and workforce productivity worldwide.2 3

Psychological treatments, particularly when combined with medication and physical therapy, are effective in managing LBP.4 Mindfulness-based interventions (MBIs; ie, evidence-based psychological approaches) have been shown to reduce pain, disability, and psychological distress associated with LBP.5 Moreover, studies have emphasised the cost-effectiveness of MBIs in reducing chronic pain–related healthcare expenses and productivity losses.6 7 Although the exact mechanisms through which MBIs alleviate pain have not been elucidated, there is evidence that they may alter pain signal processing in the brain, fostering acceptance and non-judgemental awareness. These outcomes enhance pain tolerance and reduce emotional reactivity to pain.8

Other commonly used social and psychotherapeutic modalities include cognitive-behavioural therapy and acceptance and commitment therapy. Cognitive-behavioural therapy targets maladaptive thought patterns and behaviours,9 whereas acceptance and commitment therapy focuses on promoting psychological flexibility despite the presence of pain.10 Mindfulness-based interventions uniquely emphasise cultivating present-moment awareness and acceptance of pain sensations.11 Key advantages of MBIs include their accessibility and cost-effectiveness: they can be efficiently delivered in group settings (either online or face-to-face), facilitating scalability for public healthcare initiatives.12 13 Moreover, they have the potential to enhance self-management skills for sustainable pain management.14 Acceptance and commitment therapy has limited empirical support and mixed results regarding its effectiveness in terms of improving pain intensity among patients with chronic pain.15 16 Cognitive-behavioural therapy is a widely used and well-researched therapeutic approach for chronic pain.12 However, it is considered suitable for one-on-one (rather than group-based) formats because it requires personalised treatment plans that address the unique needs and concerns of each patient.17 Furthermore, MBIs have demonstrated greater cost-effectiveness relative to cognitive-behavioural therapy among patients with chronic LBP.18

In Hong Kong, approximately 90% of specialist and inpatient care services and 30% of primary care services are provided by the public sector.19 Given the absence of universal health insurance or co-payment, the majority of chronic diseases (eg, LBP) are managed within the public healthcare system.20 The incorporation of MBIs into standard LBP treatment within this system requires an understanding of patients’ willingness to pay (WTP) and preferences. Relatively few studies have explored WTP or preferences for MBIs among patients with chronic LBP. An understanding of WTP is crucial for efforts to assess the perceived value of healthcare interventions, inform policy decisions, and guide resource allocation.21 22 Consideration of patient preferences in healthcare service decisions can improve uptake, adherence, efficiency, and patient satisfaction while reducing costs.23 24

This study aimed to estimate WTP and preferences for MBIs among patients with chronic LBP in the public healthcare sector and to explore factors associated with WTP and preferences for MBIs.

Chronic LBP is significantly influenced by psychological factors; social determinants play a crucial role in the interpretation of chronic LBP and the ways that individuals seek and receive pain treatment.25 26 The socio-psychobiological model of chronic pain represents a paradigmatic shift from the conventional biopsychosocial model.27 28 Whereas the latter model recognises the interplay of social, psychological, and biological factors, it tends to prioritise biological determinants over social and psychological aspects.27 28 In contrast, the socio-psychobiological model primarily emphasises social determinants, followed by psychological and biological factors.27 28

Our research, which assesses WTP and preferences for MBIs in the context of chronic LBP, aligns with the socio-psychobiological model for pain management. The examination of WTP and preferences can provide valuable insights into the socio-economic backgrounds of individuals with chronic LBP, which may strongly influence their experiences of pain and responses to pain management interventions. The findings may also clarify patients’ abilities to access and afford pain management strategies.29 This aspect is particularly important because it underscores the social dimensions of chronic pain management, highlighting disparities and barriers that may exist in pain experiences and access to effective interventions. Furthermore, MBIs constitute a psychological and group-based approach to chronic pain management, addressing both psychological and social factors emphasised within the socio-psychobiological model.12 30 These interventions provide individuals with skills to manage psychological distress linked to chronic LBP while also promoting social support and connectivity in group settings.31 32 By fostering mindfulness practices, MBIs equip individuals with coping mechanisms to navigate the psychological distress often associated with chronic LBP, while also enhancing social support and connectivity within group settings.31 32

We conducted a prospective cross-sectional survey using convenience sampling to recruit eligible patients with chronic LBP from two Hong Kong public hospitals between September 2022 and February 2023. We utilised a discrete choice experiment (DCE) design to examine preferences for MBIs. This study adhered to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines.

The inclusion criteria for this study were age ≥18 years, chronic non-specific LBP, and the ability to speak and understand Chinese. Chronic non-specific LBP was defined as pain in the lumbosacral region, with or without sciatica, that persisted for >3 months and lacked a clearly identifiable cause or pathology based on clinical evaluation, imaging, or laboratory tests. Exclusion criteria were chronic LBP with a specific identifiable cause or pathology, such as inflammatory diseases, tumours, infections, fractures, structural abnormalities, or other spinal pathologies evident on clinical evaluation, imaging, or laboratory tests. Patients who did not provide written informed consent, were pregnant, or were <6 months postpartum or post-weaning were also excluded.

To determine the sample size for evaluating WTP, we used the payment card elicitation format sample size formula established by Mitchell and Carson.33 The formula is:

where n is the minimum required sample size, Z_1-α/2 represents the desired confidence interval, Z_1-β corresponds to the value for power, V denotes the coefficient of variation (ie, ratio of estimated standard deviation of WTP to estimated mean WTP), and D is the designed effect (ie, percentage difference between true WTP and mean of estimated WTP bids). For this study, assuming α=0.05, β=0.20, V=0.98 (based on a previous study evaluating WTP for reduced pain intensity among patients with chronic pain),34 and D=0.20, the calculated minimum sample size was 470, considering a 20% non-response rate.

To explore preferences for MBI receipt using a DCE design, we applied the rule of thumb described by Orme35 and Johnson and Orme.36 The minimum sample size required for the main effects was calculated as follows:

Under conditions of two alternatives, a maximum of four attribute levels and eight scenarios per patient, a minimum of 125 patients was required. Considering two subgroups with different characteristics and a 20% non-response rate, the adjusted minimum sample size was 312.

A self-administered questionnaire was used to collect data. An onsite research assistant invited patients in the clinic waiting area to participate in the survey and was available to provide assistance if needed.

The independent variables of the study are as follows:

1. Socio-demographic characteristics: Age, gender, education level, employment status, and personal and family income were recorded.
2. General self-reported health status: A single-item self-rated health scale was used to assess participants’ self-rated health, with response options ranging from ‘Very good’ to ‘Very poor.’37 Studies have shown that this scale is associated with patients’ WTP for pain treatments.38 39 40 41
3. Knowledge and usage of MBIs: Knowledge of mindfulness was assessed using two items adapted from a previous study that investigated health professionals’ and health profession students’ knowledge of and attitudes toward mindfulness.42 The items were as follows: (1) What is the extent of your knowledge of MBIs? (2) Might MBIs be useful for treating chronic pain? Usage of MBIs was determined using two items adapted from a previous study that evaluated employees’ preferences for accessing MBIs.43 The items were as follows: (1) Have you ever participated in mindfulness courses? (2) How many mindfulness sessions have you attended?
4. Pain-related characteristics: Pain-related characteristics included pain duration, pain intensity, disability, and frequency of treatment for chronic LBP. Pain intensity was measured using an 11-point Numeric Rating Scale (NRS).44 Disability was assessed using the Roland-Morris Disability Questionnaire.45 Pain duration was determined by asking participants to report the number of months they had experienced an ongoing LBP problem. Frequency of treatment was evaluated by asking participants to report how many times they had consulted a doctor or other healthcare professional for LBP in the past 12 months.
5. Satisfaction with current treatment: An item was adapted from a previous study that assessed treatment satisfaction in patients with osteoarthritis and LBP.46 This item asked participants to rate their satisfaction with the effectiveness of current treatment in controlling LBP.
6. Monthly expenses on current treatment: Participants were asked to report their monthly expenses with respect to chronic LBP treatment.

Willingness to pay and preferences for MBIs were the two dependent variables of the current study. The payment card method was used to assess WTP for MBIs.47 This approach minimises starting point bias and reduces the high rate of item non-response relative to other elicitation methods.48 To ensure that participants were familiar with MBIs, we provided an introduction using a text description and a video before each participant responded to the WTP question (online supplementary Fig). Participants were presented with a range of monetary values (HK$0 to HK$10 000) and asked to select the value that best represented the amount they would be willing to pay for MBIs. Additionally, WTP for pain reduction was evaluated using two items adapted from a previous study that assessed WTP for reductions in chronic LBP and neck pain using the payment card method.49 These items asked participants to indicate the amount they would be willing and able to pay out-of-pocket per month for their chronic LBP to be reduced by half or entirely eliminated. Participants unwilling to pay any amount were asked to specify their reasons.

Participants were invited to respond to eight choice sets evaluating patient preferences for MBIs. In each choice task, they were asked to select their most preferred option from two hypothetical MBIs with different attribute levels. To ensure comprehension, we included a test scenario with a dominant alternative. If participants did not choose the dominant option, research staff provided clarification. Internal validity was assessed by including a choice set with dominant pairs, in which one alternative was clearly superior across all attributes.

Complete-case analysis was utilised for the dependent variable of WTP for MBIs. The Tobit regression model was used to estimate the associated factors.50 This model was selected because WTP measures exhibited left-censoring (ie, a substantial proportion of zero values [46.6% of the sample]; the remaining responses indicated positive WTP for MBIs). Multicollinearity was examined using tolerance and the variance inflation factor (VIF). Continuous variables were presented as mean±standard deviation. The level of statistical significance was set at 5%.51

A DCE design was used in this study to examine the preferences of individuals with chronic LBP for MBIs. The DCE comprised four key steps: (1) conducting a literature review to identify conceptual attributes and levels; (2) conducting qualitative research to determine contextual attributes and levels; (3) integrating attributes and levels into choice sets, conducting pilot tests, and refining the questionnaire; and (4) collecting experimental data and performing data analysis.

A systematic review of DCEs examining patient preferences for non-surgical treatments in chronic musculoskeletal pain was conducted.52 Studies that used DCEs to evaluate patient preferences for the management of chronic musculoskeletal pain were included.

Participants with chronic LBP were invited to discuss characteristics of MBIs they might consider valuable when deciding whether to participate in MBIs. These valued characteristics were summarised. A panel of experts from relevant fields (chronic pain, DCE methodology, and psychology) then reviewed and refined the attributes and levels, selecting six to eight attributes for inclusion.

A D-efficient experimental design was used to generate choice sets, which were randomly assigned to five blocks. A pilot DCE survey was conducted to assess cognitive difficulty and questionnaire length. Twenty patients with chronic LBP participated in the pilot study; they provided feedback and suggestions for improvement.

Discrete choice experiment data were collected as part of the cross-sectional survey. Respondents’ relative preferences were estimated using a mixed logit model with panel specification to adjust for correlated choices within individuals. The coefficients of four variables—‘improvement in capacity to perform daily life activities’, ‘risk of adverse events’, ‘improvement in pain relief’, and ‘out-of-pocket costs’—were assumed to be random, following a zero-bounded triangular distribution because the distribution of these random parameters should comprise only positive or negative values. ‘Out-of-pocket costs’ was specified as a continuous variable in the mixed logit model. The marginal WTP for different levels within each attribute was calculated through division of the negative estimated beta coefficient for each level by the estimated beta coefficient for ‘out-of-pocket costs’. The log-likelihood and adjusted McFadden’s pseudo–R-squared were calculated to assess model goodness of fit. Higher log-likelihood and adjusted McFadden’s pseudo–R-squared values indicate a better-fitting model.53 54 Subgroup analyses were conducted to assess preference heterogeneity across characteristics, including age, gender, family monthly income, and education.

Of the 589 participants invited, 488 questionnaires were returned, yielding a response rate of 82.9%. The study sample had a mean age of 60.06±12.72 years; 69.5% of the participants were women. The average pain duration was 6.46±8.16 years; mean NRS and Roland-Morris Disability Questionnaire scores were 4.70±2.12 and 7.58±5.63, respectively. Participant characteristics are summarised in Table 1.

Regarding knowledge and usage of MBIs, 77.3% of participants were unfamiliar with MBIs, 84.5% were uncertain about their effectiveness in treating chronic LBP, and 94.5% had never attended an MBI session. Knowledge and usage of MBIs are summarised in Table 2.

The mean monthly WTP values for MBIs to reduce pain by half and to entirely eliminate pain were HK$684.68±1347.43 and HK$1102.70±1983.83, respectively. The overall mean WTP for an eightsession MBI programme was HK$258.75±508.11. Among the participants, 237 were not willing to pay for MBIs, citing reasons such as limited knowledge of MBIs, unwillingness to spend money on treatment, lack of time, and scepticism regarding MBI effectiveness (online supplementary Table 1).

Multicollinearity among the independent variables was assessed; all tolerance values were >0.25 and VIF values were <4, except for two similar variables (ie, usage of MBIs measured as a binary variable [‘Yes’ or ‘No’] and number of MBI sessions attended). Given that only a small number of participants had attended MBIs, the variable measuring the number of MBI sessions was selected for inclusion in the Tobit regression model (online supplementary Table 2).

Factors associated with WTP for MBIs are summarised in Table 3. Participants with a higher NRS score (β=81.26; P=0.003), family monthly income of ≥HK$30 000 (β=320.1; P=0.035), high school education (β=242.94; P=0.045), and higher monthly expenses on chronic LBP treatment (β=0.11; P=0.003) were more willing to pay for MBIs. Conversely, participants who did not believe in the usefulness of MBIs (β=-528.88; P=0.033) were less willing to pay for them.

In total, 15 eligible studies were included.52 The attributes most frequently cited were ‘capacity to realize daily life activities’, ‘risk of adverse events’, ‘effectiveness in pain reduction’, and ‘out-of-pocket costs’, which were also ranked among the top three most important attributes. Other attributes, cited less frequently but revealing important preferences, included ‘treatment frequency’ and ‘onset of treatment efficacy’.52

Eight patients with chronic LBP participated in this stage of developing contextual attributes through patient-public involvement. Two focus group interviews were conducted to identify contextual attributes. Valued characteristics of MBIs were summarised, including effectiveness in pain reduction, mood regulation, and sleep improvement; treatment environment; reliability of mindfulness instructors; reputation of the organisation; safety; affordability; flexibility (availability of online resources at all times); availability of follow-up courses; and a group-based course format. Three experts finalised the selection of seven attributes for inclusion (Table 4).

Only minor changes in terminology were applied to attribute levels after the pilot study. This pilot study verified the attributes and their levels, as presented in Table 4. The pilot study also indicated that most patients understood the instructions and attributes. Only minor layout adjustments were made—some participants reported that the font size was too small.

After the exclusion of participants who declined to answer DCE questions due to difficulties in comprehension or unwillingness to respond (n=69, 14.1%) and those with missing DCE responses (n=4, 0.8%), the final participant count was reduced to 415. Among these participants, six (1.4%) did not pass the dominance test; thus, 409 participants were included in the analysis. The results of the DCE examining factors associated with patients’ preferences for MBIs are presented in Table 5. Participants were more likely to choose MBIs with lower out-of-pocket costs, higher levels of pain relief, and greater improvements in capacity to perform daily life activities. Face-to-face treatment modes were preferred over online formats. Regarding model fit, the log-likelihood and adjusted McFadden’s pseudo–R-squared for the mixed logit model were -1502.8 and 0.330, respectively.

The results of subgroup analyses are presented in online supplementary Tables 3 to 6. Preferences differed substantially between age-groups, family income levels, and education levels, but showed no gender-based significant differences. Improvement in the capacity to perform daily life activities was an important attribute when selecting MBIs for older participants, those with lower family monthly income, and those with higher education level; this attribute was not important for younger participants and those with higher family monthly income and lower education level. Group size was an important attribute for younger participants and those with higher family monthly income but not for older participants or those with lower family monthly income. Younger participants and those with higher family monthly income preferred MBIs with a group size of one person, rather than 7 to 12 people. Treatment mode was an important attribute for participants with lower family monthly income and higher education level but not for those with higher family monthly income and lower education. Participants with lower family monthly income and higher education preferred face-to-face treatment over online treatment. Furthermore, participants with lower family monthly income and older age placed greater priority on out-of-pocket costs for MBIs, as indicated by substantially larger regression coefficients for out-of-pocket costs in subgroup analyses.

Consistent with previous studies,34 49 we found that patients with higher pain scores, higher family income, and higher monthly expenses on LBP treatment were more willing to pay for MBIs. Comparison of WTP for MBIs in this study to a national survey on WTP for complementary and alternative medicine treatments in England55 revealed that participants in the present study had a lower WTP. One possible explanation for this discrepancy is that complementary and alternative medicine practices, such as acupuncture and herbal medicine, are more established in some cultures; MBIs are relatively new and may be less familiar to our study population.

In Hong Kong’s public healthcare system, physiotherapy and occupational therapy for chronic pain cost HK$80 per visit. If MBIs followed this fee structure, eight sessions would cost a total of HK$640. However, the current WTP for MBIs is HK$258.75, approximately 40% of this cost. Notably, WTP was calculated in a population with limited knowledge of MBIs. Increased awareness of their efficacy may enhance WTP, aligning it more closely with the existing fee structure.

Our study evaluating preferences for MBIs confirmed previous findings that chronic pain treatment preferences are significantly influenced by treatment effectiveness and out-of-pocket costs.52 56 57 However, in contrast to prior studies,52 56 57 we found that the risk of adverse events was not an attribute considered important by patients with chronic LBP during MBI selection. One possible explanation is that the risk of adverse events from psychological interventions is lower and less severe than the risk of such events associated with pharmacological or exercise-based interventions.58 59 60 Additionally, we observed that treatment mode constituted an important attribute of MBIs, consistent with investigations of exercise therapy preferences among patients with chronic pain.39

Our study focused on assessing WTP and preferences for MBIs in chronic LBP, following the socio-psychobiological model that prioritises social and psychological factors over biological factors.27 28 This approach provides insights into the socio-economic backgrounds of patients with chronic LBP and highlights their pain experiences and access to pain management strategies, emphasising the social dimension of chronic pain management. Mindfulness-based interventions, as a psychological and group-based approach, equip individuals with skills to manage psychological distress related to chronic LBP while fostering social support and connectivity through group interaction.

The current approach to chronic pain care often results in the underutilisation of high-value care (eg, psychological therapies) and overuse of low-value care, including invasive procedures and opioid medications.4 28 The adoption and implementation of a socio-psychobiological model could serve as an effective strategy for establishing pain care systems that prioritise high-value care.27 28

Despite the recognised value of MBIs in chronic pain management, their limited integration into clinical practice may be attributed to patients’ unfamiliarity and lack of knowledge about these interventions, coupled with insufficient investment in primary care resources. Additionally, economic incentives often favour high-volume practice models in primary care settings.28 Thus, there is an urgent need for educational initiatives to enhance awareness and knowledge of MBIs among individuals with chronic LBP, as well as increased investment in primary care resources.

This study provided critical insights into the integration of MBIs for chronic LBP management within the Hong Kong public healthcare system. In the context of Hong Kong’s public healthcare settings, we propose integrating MBIs as an intermediary step between primary care and specialist care for chronic LBP management. Primary care providers could identify patients experiencing psychological and social distress who may benefit from MBIs and facilitate their referral for MBI treatment. Patients whose condition does not improve after an MBI could then be referred to specialist clinics. This approach could substantially reduce waiting times for chronic LBP treatment within the Hong Kong public healthcare system.

This study has several strengths. To our knowledge, it is the first investigation to assess WTP and preferences for MBIs in chronic pain management; it included a comprehensive list of independent variables covering key factors that influence WTP. Additionally, the study utilised a mixed logit model to consider preference heterogeneity within the sample. Furthermore, a rigorous systematic review and qualitative interviews informed the attributes and levels used in the DCE. However, certain limitations should be acknowledged. First, participants’ limited knowledge of MBIs may have influenced WTP and preferences. Second, participants were recruited through convenience sampling from outpatient clinics in two Hong Kong public hospitals, which may have introduced selection bias that skewed the sample composition and limited its representativeness. This limitation may affect the generalisability of the findings beyond the specific group sampled. Third, the cross-sectional design of the study precluded establishment of causal relationships between WTP and preferences for MBIs, as well as associated factors.

Although WTP and preferences are essential considerations for MBI implementation, they should not be the sole determinants. Factors such as cost-effectiveness, impact on quality of life, and infrastructure availability must also be considered. Further research is required to provide additional evidence for implementation within the Hong Kong public healthcare system. Nevertheless, this study established a rationale for assessing WTP and preferences for MBIs, with a methodology that can be adapted for healthcare evaluations in other countries.

This study highlights the need to increase awareness of MBIs for chronic LBP management within the public healthcare system. The findings indicate low WTP among participants, suggesting a gap in understanding and utilisation. Notably, individuals with higher pain scores, higher family income, and higher monthly LBP treatment expenses, as well as a stronger belief in MBIs, were more willing to pay for such interventions; these observations indicate targeted demand. Patient preferences favoured lower costs, face-to-face treatment, and enhanced effectiveness. These findings provide practical insights for designing patient preference–aligned MBIs and will serve as valuable references for future healthcare evaluations.

Concept or design: M Zhu, PKH Mo, RWS Sit.
Acquisition of data: M Zhu.
Analysis or interpretation of data: All authors.
Drafting of the manuscript: M Zhu.
Critical revision of the manuscript for important intellectual content: All authors.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

As an editor of the journal, RWS Sit was not involved in the peer review process. Other authors have disclosed no conflicts of interest.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This research was approved by the Joint Chinese University of Hong Kong–New Territories East Cluster Clinical Research Ethics Committee, Hong Kong (Ref. No.: 2022.279). The research was conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent before completing the questionnaire.

The datasets generated during and/or analysed during the current study are not publicly available due to ethics restrictions. A request for the code can be made directly to the corresponding author.

The supplementary material was provided by the authors and some information may not have been peer reviewed. Accepted supplementary material will be published as submitted by the authors, without any editing or formatting. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by the Hong Kong Academy of Medicine and the Hong Kong Medical Association. The Hong Kong Academy of Medicine and the Hong Kong Medical Association disclaim all liability and responsibility arising from any reliance placed on the content.

1. Hoy D, March L, Brooks P, et al. The global burden of low back pain: estimates from the Global Burden of Disease 2010 study. Ann Rheum Dis 2014;73:968-74. Crossref

2. Dieleman JL, Cao J, Chapin A, et al. US health care spending by payer and health condition, 1996-2016. JAMA 2020;323:863-84. Crossref

3. Hong J, Reed C, Novick D, Happich M. Costs associated with treatment of chronic low back pain: an analysis of the UK General Practice Research Database. Spine (Phila Pa 1976) 2013;38:75-82. Crossref

4. Foster NE, Anema JR, Cherkin D, et al. Prevention and treatment of low back pain: evidence, challenges, and promising directions. Lancet 2018;391:2368-83. Crossref

5. Anheyer D, Haller H, Barth J, Lauche R, Dobos G, Cramer H. Mindfulness-based stress reduction for treating low back pain: a systematic review and meta-analysis. Ann Intern Med 2017;166:799-807. Crossref

6. Herman PM, Anderson ML, Sherman KJ, Balderson BH, Turner JA, Cherkin DC. Cost-effectiveness of mindfulness-based stress reduction versus cognitive behavioral therapy or usual care among adults with chronic low back pain. Spine (Phila Pa 1976) 2017;42:1511-20. Crossref

7. Pérez-Aranda A, D’Amico F, Feliu-Soler A, et al. Cost-utility of mindfulness-based stress reduction for fibromyalgia versus a multicomponent intervention and usual care: a 12-month randomized controlled trial (EUDAIMON study). J Clin Med 2019;8:1068. Crossref

8. Day MA, Jensen MP, Ehde DM, Thorn BE. Toward a theoretical model for mindfulness-based pain management. J Pain 2014;15:691-703. Crossref

9. Ehde DM, Dillworth TM, Turner JA. Cognitive-behavioral therapy for individuals with chronic pain: efficacy, innovations, and directions for research. Am Psychol 2014;69:153-66. Crossref

10. Hayes SC, Luoma JB, Bond FW, Masuda A, Lillis J. Acceptance and commitment therapy: model, processes and outcomes. Behav Res Ther 2006;44:1-25. Crossref

11. Zhang D, Lee EK, Mak EC, Ho CY, Wong SY. Mindfulness-based interventions: an overall review. Br Med Bull 2021;138:41-57. Crossref

12. Khoo EL, Small R, Cheng W, et al. Comparative evaluation of group-based mindfulness-based stress reduction and cognitive behavioural therapy for the treatment and management of chronic pain: a systematic review and network meta-analysis. Evid Based Ment Health 2019;22:26-35. Crossref

13. Liu Z, Jia Y, Li M, et al. Effectiveness of online mindfulness-based interventions for improving mental health in patients with physical health conditions: systematic review and meta-analysis. Arch Psychiatr Nurs 2022;37:52-60. Crossref

14. Khusid MA, Vythilingam M. The emerging role of mindfulness meditation as effective self-management strategy, part 2: clinical implications for chronic pain, substance misuse, and insomnia. Mil Med 2016;181:969-75. Crossref

15. Veehof MM, Trompetter HR, Bohlmeijer ET, Schreurs KM. Acceptance- and mindfulness-based interventions for the treatment of chronic pain: a meta-analytic review. Cogn Behav Ther 2016;45:5-31. Crossref

16. Hughes LS, Clark J, Colclough JA, Dale E, McMillan D. Acceptance and commitment therapy (ACT) for chronic pain: a systematic review and meta-analyses. Clin J Pain 2017;33:552-68. Crossref

17. Gryesten JR, Poulsen S, Moltu C, Biering EB, Møller K, Arnfred SM. Patients’ and therapists’ experiences of standardized group cognitive behavioral therapy: needs for a personalized approach. Adm Policy Ment Health 2024;51:617-33. Crossref

18. Zhang L, Lopes S, Lavelle T, et al. Economic evaluations of mindfulness-based interventions: a systematic review. Mindfulness (N Y) 2022;13:2359-78. Crossref

19. Census and Statistics Department, Hong Kong SAR Government. Thematic Household Survey Report No. 50. Jan 2013. Available from: https://www.statistics.gov.hk/pub/B11302502013XXXXB0100.pdf. Accessed 7 Apr 2025.

20. Health Bureau, Hong Kong SAR Government. The healthcare challenges in Hong Kong. 2022. Available from: https://www.primaryhealthcare.gov.hk/bp/en/supplementary-documents/challenges/. Accessed 18 Mar 2025.

21. Abbas SM, Usmani A, Imran M. Willingness to pay and its role in health economics. JBUMDC 2019;9:62-6.

22. Liu S, Yam CH, Huang OH, Griffiths SM. Willingness to pay for private primary care services in Hong Kong: are elderly ready to move from the public sector? Health Policy Plan 2013;28:717-29. Crossref

23. Krist AH, Tong ST, Aycock RA, Longo DR. Engaging patients in decision-making and behavior change to promote prevention. Stud Health Technol Inform 2017;240:284-302. Crossref

24. Ostermann J, Brown DS, de Bekker-Grob EW, Mühlbacher AC, Reed SD. Preferences for health interventions: improving uptake, adherence, and efficiency. Patient 2017;10:511-4. Crossref

25. Alhowimel A, AlOtaibi M, Radford K, Coulson N. Psychosocial factors associated with change in pain and disability outcomes in chronic low back pain patients treated by physiotherapist: a systematic review. SAGE Open Med 2018;6:2050312118757387. Crossref

26. Karran EL, Grant AR, Moseley GL. Low back pain and the social determinants of health: a systematic review and narrative synthesis. Pain 2020;161:2476-93. Crossref

27. Carr DB, Bradshaw YS. Time to flip the pain curriculum? Anesthesiology 2014;120:12-4. Crossref

28. Mardian AS, Hanson ER, Villarroel L, et al. Flipping the pain care model: a sociopsychobiological approach to high-value chronic pain care. Pain Med 2020;21:1168-80. Crossref

29. Allen-Watts K, Sims AM, Buchanan TL, et al. Sociodemographic differences in pain medication usage and healthcare provider utilization among adults with chronic low back pain. Front Pain Res (Lausanne) 2022;2:806310. Crossref

30. Majeed MH, Ali AA, Sudak DM. Mindfulness-based interventions for chronic pain: evidence and applications. Asian J Psychiatr 2018;32:79-83. Crossref

31. Smith SL, Langen WH. A systematic review of mindfulness practices for improving outcomes in chronic low back pain. Int J Yoga 2020;13:177-82. Crossref

32. Petrucci G, Papalia GF, Russo F, et al. Psychological approaches for the integrative care of chronic low back pain: a systematic review and metanalysis. Int J Environ Res Public Health 2021;19:60. Crossref

33. Mitchell RC, Carson RT. Using Surveys to Value Public Goods: The Contingent Valuation Method. New York and London: Resources for the Future; 1989.

34. Chuck A, Adamowicz W, Jacobs P, Ohinmaa A, Dick B, Rashiq S. The willingness to pay for reducing pain and pain-related disability. Value Health 2009;12:498-506. Crossref

35. Orme BK. Sample size issues for conjoint analysis studies. In: Orme BK, editor. Getting Started with Conjoint Analysis: Strategies for Product Design and Pricing Research. 4th ed. Madison [WI]: Research Publishers LLC; 1998: 57-65.

36. Johnson R, Orme B. Sawtooth Software Research Paper Series. Getting the most from CBC. WA: Sawtooth Software; 2003. Available from: https://sawtoothsoftware.com/resources/technical-papers/getting-the-most-from-cbc. Accessed 24 Mar 2025.

37. Hanmer J. Measuring population health: association of self-rated health and PROMIS measures with social determinants of health in a cross-sectional survey of the US population. Health Qual Life Outcomes 2021;19:221. Crossref

38. Copsey B, Buchanan J, Fitzpatrick R, Lamb SE, Dutton SJ, Cook JA. Duration of treatment effect should be considered in the design and interpretation of clinical trials: results of a discrete choice experiment. Med Decis Making 2019;39:461-73. Crossref

39. Cranen K, Groothuis-Oudshoorn CG, Vollenbroek-Hutten MM, IJzerman MJ. Toward patient-centered telerehabilitation design: understanding chronic pain patients’ preferences for web-based exercise telerehabilitation using a discrete choice experiment. J Med Internet Res 2017;19:e26. Crossref

40. Ferreira GE, Howard K, Zadro JR, O’Keeffe M, Lin CC, Maher CG. People considering exercise to prevent low back pain recurrence prefer exercise programs that differ from programs known to be effective: a discrete choice experiment. J Physiother 2020;66:249-55. Crossref

41. Laba TL, Brien JA, Fransen M, Jan S. Patient preferences for adherence to treatment for osteoarthritis: the MEdication Decisions in Osteoarthritis Study (MEDOS). BMC Musculoskelet Disord 2013;14:160. Crossref

42. McKenzie SP, Hassed CS, Gear JL. Medical and psychology students’ knowledge of and attitudes towards mindfulness as a clinical intervention. Explore (NY) 2012;8:360-7. Crossref

43. Lau MA, Colley L, Willett BR, Lynd LD. Employee’s preferences for access to mindfulness-based cognitive therapy to reduce the risk of depressive relapse—a discrete choice experiment. Mindfulness 2012;3:318-26. Crossref

44. Atisook R, Euasobhon P, Saengsanon A, Jensen MP. Validity and utility of four pain intensity measures for use in international research. J Pain Res 2021;14:1129-39. Crossref

45. Yamato TP, Maher CG, Saragiotto BT, Catley MJ, McAuley JH. The Roland-Morris Disability Questionnaire: one or more dimensions? Eur Spine J 2017;26:301-8. Crossref

46. Turk D, Boeri M, Abraham L, et al. Patient preferences for osteoarthritis pain and chronic low back pain treatments in the United States: a discrete-choice experiment. Osteoarthritis Cartilage 2020;28:1202-13. Crossref

47. Tian X, Yu X, Holst R. Applying the payment card approach to estimate the WTP for green food in China. In: IAMO Forum 2011; 2011 Jun 23-24; Halle, Germany; 2011: No.23.

48. Soeteman L, van Exel J, Bobinac A. The impact of the design of payment scales on the willingness to pay for health gains. Eur J Health Econ 2017;18:743-60. Crossref

49. Herman PM, Luoto JE, Kommareddi M, Sorbero ME, Coulter ID. Patient willingness to pay for reductions in chronic low back pain and chronic neck pain. J Pain 2019;20:1317-27. Crossref

50. Pavel MS, Chakrabarty S, Gow J. Assessing willingness to pay for health care quality improvements. BMC Health Serv Res 2015;15:43. Crossref

51. Kanter G, Komesu Y, Qaedan F, Rogers R. 5: Mindfulness-based stress reduction as a novel treatment for interstitial cystitis/bladder pain syndrome: a randomized controlled trial [abstract]. Am J Obstet Gynecol 2016;214(4 Suppl 1):S457-8. Crossref

52. Zhu M, Dong D, Lo HH, Wong SY, Mo PK, Sit RW. Patient preferences in the treatment of chronic musculoskeletal pain: a systematic review of discrete choice experiments. Pain 2023;164:675-89. Crossref

53. UCLA: Statistical Consulting Group. FAQ: How are the likelihood ratio, wald, and lagrange multiplier (score) tests different and/or similar? Available from: https://stats.oarc.ucla.edu/other/mult-pkg/faq/general/faqhow-are-the-likelihood-ratio-wald-and-lagrange-multiplier-score-tests-different-andor-similar/#:~:text=The%20log%20likelihood%20(i.e.%2C%20the,model%20with%20a%20likelihood%20function. Accessed 12 May 2024.

54. Hu B, Shao J, Palta M. PSEUDO-R 2 in logistic regression model. Stat Sin 2006;16:847-60.

55. Sharp D, Lorenc A, Morris R, et al. Complementary medicine use, views, and experiences: a national survey in England. BJGP Open 2018;2:bjgpopen18X101614. Crossref

56. Al-Omari B, McMeekin P, Bate A. Systematic review of studies using conjoint analysis techniques to investigate patients’ preferences regarding osteoarthritis treatment. Patient Prefer Adherence 2021;15:197-211. Crossref

57. Poder TG, Beffarat M. Attributes underlying non-surgical treatment choice for people with low back pain: a systematic mixed studies review. Int J Health Policy Manag 2021;10:201-10. Crossref

58. Ho EK, Chen L, Simic M, et al. Psychological interventions for chronic, non-specific low back pain: systematic review with network meta-analysis. BMJ 2022;376:e067718. Crossref

59. Els C, Jackson TD, Kunyk D, et al. Adverse events associated with medium- and long-term use of opioids for chronic non-cancer pain: an overview of Cochrane Reviews. Cochrane Database Syst Rev 2017;10:CD012509. Crossref

60. Geneen LJ, Moore RA, Clarke C, Martin D, Colvin LA, Smith BH. Physical activity and exercise for chronic pain in adults: an overview of Cochrane Reviews. Cochrane Database Syst Rev 2017;4:CD011279. Crossref

The implementation of early gastric cancer screening in community populations and performance of endoscopic examinations in high-risk groups represents a feasible, cost-effective, and efficient strategy to address the challenges of gastric cancer diagnosis and treatment in China.1 More than 80% of early-stage gastric cancer cases are identified in asymptomatic community populations aged ≥40 years. Thus, community-based screening programmes are important for increased detection of early-stage cancer. Gastroscopy remains the gold standard for diagnosing upper gastrointestinal diseases. High-quality intragastric visibility is essential for ensuring diagnostic accuracy, minimising the risks of misdiagnosis and missed diagnosis, and improving the detection of minimal-change gastric lesions. However, air bubbles and mucus in the stomach often reduce gastroscopic field visibility, leading to missed diagnoses and prolonged examination times. Pretreatment with defoaming agents and mucolytic agents enhances gastroscopic field visibility.2 Pronase, a proteolytic enzyme isolated from the culture filtrate of Streptomyces griseus, effectively cleaves the peptide bonds of glycoproteins, thereby dissolving and eliminating gastric mucus.3 This study aimed to evaluate the impact of pronase on the detection rate of precancerous lesions and early cancer, clarifying its utility in early gastric cancer screening. The findings will provide foundational evidence for the incorporation of pronase in endoscopic screening for upper gastrointestinal tract cancers and clinical diagnostic examinations.

This study enrolled 1450 individuals aged 40 to 70 years from a community population who participated in the 2020-2021 Upper Gastrointestinal Cancer Screening Programme in Wuwei, Gansu Province, China. The inclusion criteria were: (1) ability to cooperate with the gastroscopic procedure; (2) ability to discontinue anticoagulant medications 1 week prior to endoscopy; and (3) voluntary participation and provision of written informed consent. The exclusion criteria were: (1) contraindications to gastroscopy; (2) severe heart disease or heart failure; (3) severe respiratory disease; (4) posterior pharyngeal abscess or severe spinal deformity; (5) other serious illnesses or physical conditions that precluded tolerance of endoscopy; and (6) bleeding tendency.

Using a random number table, all 1450 participants from the community population were randomly assigned to either an experimental group (n=725) or a control group (n=725). All participants underwent gastroscopy and tissue biopsy. In the experimental group, 1 sachet (20 000 U) of pronase (Beijing Tide Pharmaceutical, Beijing, China) and 1 g of sodium bicarbonate were dissolved in 50 to 80 mL of drinking water (20-40°C) by shaking. The solution was orally administered 15 to 30 minutes before gastroscopy (GIF-H290; Olympus, Tokyo, Japan). Dimethicone was also given orally to lubricate the cavity and remove gastric bubbles. To ensure that pronase reached all areas of the stomach, participants laid flat on a bed under a nurse’s guidance, then turned sideways three to five times. Subsequently, routine gastroscopy was performed. In the control group, participants received oral dimethicone 15 to 30 minutes before routine gastroscopy (GIF-H290).

The gastroscopy examinations were performed by two physicians holding the title of associate chief physician or higher, each having >10 years of experience in gastroscopy. The visibility of each part of the visual field was evaluated during the procedure; pathological examinations were conducted on tissue biopsies collected from minimal-change lesions.

Endoscopic visibility scores were compared between the two groups. Scoring criteria were as follows4: 1 point, no mucus; 2 points, a small amount of mucus but no blurring of the visual field; 3 points, a large amount of mucus with a blurred visual field, requiring <30 mL of water for rinsing; and 4 points, very thick mucus with a blurred visual field, requiring ≥30 mL of water for rinsing. Lower scores indicated better endoscopic visibility. To minimise errors during the scoring process, each visibility score was recorded as the average of scores assigned by the two physicians who performed gastroscopy. The lesion detection rate was defined as the percentage of subjects within a group in whom lesions were identified. Gastroscopy time was measured from entry of the gastroscope into the oesophagus until its removal. Adverse reactions included nausea, vomiting, difficulty breathing, facial flushing, and other symptoms.

R software (version 4.0.5) was used for statistical analysis. Quantitative data were expressed as mean±standard deviation; intergroup differences were analysed using independent sample t tests. Qualitative data were expressed as frequency and percentage; intergroup differences were assessed using the Chi squared test or Fisher’s exact test. Multivariable linear regression analysis was performed to evaluate the effect of group assignment on visibility scores after adjustment for confounding factors. Differences in early cancer detection rates between the two groups were analysed using multivariable binary logistic regression analysis. All statistical tests were two-sided, and P values <0.05 were considered statistically significant.

A summary of the baseline characteristics of the experimental and control groups is provided in Table 1. Among the 1450 patients in the cohort, 416 (28.7%) had a family history of gastrointestinal disease, 172 (11.9%) had a history of smoking, 91 (6.3%) had a history of alcohol consumption, and 335 (23.1%) had a history of gastrointestinal disease. Significant differences between the two groups were observed in the proportions of patients with a history of smoking, alcohol consumption, and gastrointestinal disease.

Average visibility scores for the oesophagus, cardia, gastric fundus, gastric body, gastric antrum, gastric angle, and duodenum were significantly lower in the experimental group than in the control group (P<0.001 for all comparisons) [Table 2]. The visibility of different regions of the stomach under gastroscopy substantially differed between the two groups (Fig).

Multiple linear regression analysis was performed with the visibility score for each site as the dependent variable and group assignment as the independent variable; adjustments were conducted for sex, age, marital status, education level, smoking status, alcohol consumption, history of gastrointestinal disease, and family history of gastrointestinal disease. After adjustment for these confounding factors, the visibility scores for all regions of the stomach remained significantly higher in the control group than in the experimental group (P<0.001 for all visibility scores) [Table 3].

Chi squared test analyses revealed that the detection rates of precancerous lesions (including atrophic gastritis, intestinal metaplasia, and low-grade intraepithelial neoplasia5) and early cancer were significantly higher in the experimental group than in the control group (77.5% vs 62.5%; P<0.001) [Table 4].

Multivariable binary logistic regression analysis was performed with early cancer detection as the dependent variable and group assignment as the independent variable; adjustments were conducted for sex, age, marital status, education level, smoking status, alcohol consumption, history of gastrointestinal disease, and family history of gastrointestinal disease. The likelihood of early cancer detection was significantly higher in the experimental group compared with the control group, with an odds ratio of 3.840 (95% confidence interval=1.204-12.241; P=0.023) [Table 5].

Average gastroscopy times were 6.52±2.51 minutes in the experimental group and 10.03±1.23 minutes in the control group. Gastroscopy time significantly differed between the two groups (t=33.81; P=0.001).

No adverse reactions, such as nausea, vomiting, dyspnoea, or facial flushing, were reported in either group.

Currently, approximately 90% of primary gastric cancers in China are diagnosed at an advanced stage.6 The prognosis of affected patients is closely related to the timing of diagnosis and treatment. Despite surgical intervention, the 5-year survival rate for patients with advanced gastric cancer remains <30%.7 After treatment, the 5-year survival rate for patients with early gastric cancer exceeds 90%, and cure may be achieved.8 However, the rates of early diagnosis and treatment of gastric cancer in China are <10%, substantially lower than rates reported in Japan (70%) and South Korea (50%).9 In Wuwei, the incidence and mortality rates of gastric cancer remain among the highest in the country; gastric cancer ranks first among malignant tumours in the city.10 Screening for upper gastrointestinal cancer is one of the most effective methods for population-level detection of early-stage cancer. Since 2010, Wuwei Tumour Hospital has implemented an upper gastrointestinal cancer screening programme (endoscopy combined with tissue biopsy) in Wuwei. Improvements in the detection rates of precancerous lesions and upper gastrointestinal cancer are key objectives of this screening initiative.

Gastroscopy is currently a widely used method for the clinical diagnosis and treatment of gastrointestinal diseases. A clear endoscopic field of vision is essential for accurate diagnosis and effective treatment by endoscopists. To optimise gastroscopy outcomes and enhance visibility within the stomach, bubbles and mucus must be removed. The use of pronase in combination with defoaming agents is recommended by the Consensus on Early Gastric Cancer Screening and Endoscopic Diagnosis and Treatment in China11 and the Guidelines for Endoscopic Diagnosis of Early Gastric Cancer (2019 edition) developed by the Japan Gastroenterological Endoscopy Society.12

Lee et al13 demonstrated that administering pronase 10 to 20 minutes before gastroscopy significantly improved the visibility of the endoscopic visual field and reduced the number of water washes required. Similarly, a multicentre randomised controlled study by Liu et al14 indicated that the combination of pronase and dimethicone significantly enhanced the visibility of the upper gastrointestinal mucosa. Pronase has also been utilised in narrow-band imaging endoscopy. A randomised controlled study by Cha et al15 compared the effects of orally administering pronase and simethicone 10 minutes before narrow-band imaging endoscopy on mucosal visibility and diagnostic performance. The results showed that mucosal visibility within the proximal stomach was significantly better in the pronase group than in the simethicone group.15 In the present study, the visibility scores for all sites in patients who received pronase were approximately 1 point, indicating minimal mucus adhesion. After adjustment for confounding factors, multiple linear regression analysis confirmed that visibility scores remained significantly lower in the pronase group than in the control group at all sites; this finding further validated the effectiveness of pronase. The present study also revealed that the average endoscopic examination time was significantly shorter (approximately 5 minutes) in the pronase group than in the control group. This reduced examination time was attributed to the near-complete absence of mucus adhesion after pronase administration, which decreased the number of rinses needed during the procedure. The shorter examination also enhanced patient comfort and increased compliance for subsequent screenings.

Zhang et al16 and Gao et al17 conducted retrospective analyses of 25 314 patients who underwent gastroscopy at Nanfang Hospital of Southern Medical University and 166 260 patients at Bazhong Central Hospital, revealing early cancer detection rates of 0.2% and 0.62%, respectively. Zhang et al1 performed a follow-up analysis of individuals in Liangzhou District in Wuwei who underwent upper gastrointestinal cancer screening in 2017; they observed an early cancer detection rate of 2.8%.1 In the present study, lesion detection rates for the experimental and control groups were 77.5% and 62.5%, respectively; corresponding early cancer detection rates were 3.0% and 2.1%. These percentages align with findings from the previous study in Wuwei1 and are substantially higher than those reported for other regions.16 17 The present results suggest that in Wuwei, a region displaying one of the highest incidences of upper gastrointestinal cancer in China, early cancer screening should be actively promoted. Furthermore, the detection rates of precancerous lesions and early cancer can be improved by using endoscopy combined with tissue biopsy.

The efficacy of pronase in improving the endoscopic visual field is well established, but studies investigating its impacts on the detection rates of precancerous lesions and early cancer have yielded inconsistent results.14 18 19 Chen et al18 conducted a randomised controlled trial that enrolled older patients undergoing gastroscopy; they found that the detection rate of minimal-change lesions was higher in the pronase group than in the control group (45.2% vs 27.5%; P<0.05).18 Lee et al19 demonstrated that the use of pronase when rinsing a lesion during endoscopy significantly increased the tissue depth of endoscopic biopsies and improved the anatomical localisation of biopsy sites, thereby enhancing the accuracy of disease diagnosis. In the present study, the detection rates of precancerous lesions and early cancer were significantly higher in the experimental group than in the control group (P<0.001). After adjustment for confounding factors, multivariable logistic regression showed that the likelihood of detecting early cancer was significantly greater in the experimental group than in the control group (odds ratio=3.840; P=0.023) [Table 5]. This finding indicates that pronase pretreatment before gastroscopy can enhance the detection rates of precancerous lesions and early cancer. The enhancement may be attributed to the clear visual field provided by pronase, which facilitates accurate selection of biopsy sites and improves recognition of minimal-change lesions. Gastroscopy physicians have substantial daily workloads and manage large numbers of patients requiring treatment. The use of pronase reduced the time required for endoscopy, potentially improving patient compliance with clinical microscopy.

As an early cancer screening study, this investigation had a relatively small sample size; therefore, the findings require further validation in large-scale clinical studies. Cluster randomisation was used in this study, leading to baseline differences between groups; however, adjustments for these factors were included in the statistical analyses. The gastroscopy procedures were performed by highly skilled endoscopists. The generalisability of the findings to all endoscopists warrants additional investigation.

Pronase pretreatment before gastroscopy improves visual field clarity, reduces examination time, increases the detection rates of precancerous lesions and early cancer, and demonstrates good safety. This approach is beneficial for early cancer screening in regions with a high incidence of upper gastrointestinal cancer. The practical value of this method requires confirmation in large-scale clinical studies.

Concept or design: Z Wu, S Li, G Wang.
Acquisition of data: L Lu, G Zhao, J Liu, S Li.
Analysis or interpretation of data: T Qin.
Drafting of the manuscript: Z Zhang.
Critical revision of the manuscript for important intellectual content: Z Wu.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

This research was supported by the National Key Research and Development Program of China (Ref No.: 2017YFC0908302). The funder had no role in study design, data collection, analysis, interpretation, or manuscript preparation.

This research was approved by the Medical Ethics Committee of Wuwei Cancer Hospital, Wuwei, Gansu, China (Ref No.: 2019-Ethical review-11). The trial was registered with the Chinese Clinical Trial Registry (Ref No.: ChiCTR2200064855). Informed consent was obtained from all study participants, including consent for the publication of their anonymised data and clinical photos.

1. Zhang Z, Wu Z, Lu L, et al. Analysis of the upper gastrointestinal cancer screening and follow-up results in Liangzhou District of Wuwei City from 2009 to 2017 [in Chinese]. Chin J Cancer Prev Treat 2019;23:1750-5.

2. Choi IJ. Gastric preparation for upper endoscopy. Clin Endosc 2012;45:113-4. Crossref

3. Kim GH, Cho YK, Cha JM, Lee SY, Chung IK. Effect of pronase as mucolytic agent on imaging quality of magnifying endoscopy. World J Gastroenterol 2015;21:2483-9. Crossref

4. Beg S, Ragunath K, Wyman A, et al. Quality standards in upper gastrointestinal endoscopy: a position statement of the British Society of Gastroenterology (BSG) and Association of Upper Gastrointestinal Surgeons of Great Britain and Ireland (AUGIS). Gut 2017;66:1886-99. Crossref

5. Gomceli I, Demiriz B, Tez M. Gastric carcinogenesis. World J Gastroenterol 2012;18:5164-70. Crossref

6. Committee of Laboratory Medicine of Chinese Association of Integrative Medicine. Chinese Expert Consensus on Detection Technologies for Early-stage Gastric Cancer Screening [in Chinese]. Chin J Lab Med 2023;46:347-59.

7. Katai H, Ishikawa T, Akazawa K, et al. Five-year survival analysis of surgically resected gastric cancer cases in Japan: a retrospective analysis of more than 100,000 patients from the nationwide registry of the Japanese Gastric Cancer Association (2001-2007). Gastric Cancer 2018;21:144-54. Crossref

8. Sumiyama K. Past and current trends in endoscopic diagnosis for early-stage gastric cancer in Japan. Gastric Cancer 2017;20(Suppl 1):20-7. Crossref

9. Ren W, Yu J, Zhang Z, Song Y, Li Y, Wang L. Missed diagnosis of early gastric cancer or high-grade intraepithelial neoplasia. World J Gastroenterol 2013;19:2092-6. Crossref

10. Lu L, Nie P, Zhang Z. Analysis of incidence and mortality of stomach cancer from 2011 to 2015 in Wuwei City, Gansu Province [in Chinese]. China Cancer 2020;29:677-81.

11. Chinese Society of Digestive Endoscopy; Chinese Anti-Cancer Association The Society of Tumor Endoscopy. Chinese Consensus on Screening and Endoscopic Diagnosis and Management of Early Gastric Cancer (Changsha, April 2014) [in Chinese]. Chin J Gastroenterol 2014;19:408-27.

12. Yao K, Uedo N, Kamada T, et al. Guidelines for endoscopic diagnosis of early gastric cancer. Dig Endosc 2020;32:663-98. Crossref

13. Lee GJ, Park SJ, Kim SJ, Kim HH, Park MI, Moon W. Effectiveness of premedication with pronase for visualization of the mucosa during endoscopy: a randomized, controlled trial. Clin Endosc 2012;45:161-4. Crossref

14. Liu X, Guan CT, Xue LY, et al. Effect of premedication on lesion detection rate and visualization of the mucosa during upper gastrointestinal endoscopy: a multicenter large sample randomized controlled double-blind study. Surg Endosc 2018;32:3548-56. Crossref

15. Cha JM, Won KY, Chung IK, Kim GH, Lee SY, Cho YK. Effect of pronase premedication on narrow-band imaging endoscopy in patients with precancerous conditions of stomach. Dig Dis Sci 2014;59:2735-41. Crossref

16. Zhang Q, Chen Z, Chen C, et al. Training in early gastric cancer diagnosis improves the detection rate of early gastric cancer: an observational study in China. Medicine (Baltimore) 2015;94:e384. Crossref

17. Gao Z, Liang S, Li M, et al. Clinicopathological features and trends of 1025 cases of early gastric cancer, 2006-2020 [in Chinese]. J Cancer Control Treat 2021;34:649-54.

18. Chen L, Feng Y, Wang W, Zheng P. Clinical value of pronase combined with sodium bicarbonate in gastroscopy of elderly patients [in Chinese]. Zhejiang JITCWM 2018;28:225-7.

19. Lee SY, Han HS, Cha JM, Cho YK, Kim GH, Chung IK. Endoscopic flushing with pronase improves the quantity and quality of gastric biopsy: a prospective study. Endoscopy 2014;46:747-53. Crossref

The global twin birth rate has increased by one-third since the 1980s, rising from 9.1 to 12 per 1000 deliveries, resulting in approximately 1.6 million twin pairs born annually.1 One major factor contributing to this trend is the growing use of assisted reproductive techniques in recent decades.1 Relative to singleton pregnancies, twin pregnancies are associated with higher incidences of maternal and fetal complications, which may require earlier delivery.2 3 Even in uncomplicated cases, the National Institute for Health and Care Excellence (NICE) guidelines recommend delivery at 37 weeks for dichorionic twin pregnancies and 36 weeks for monochorionic twin pregnancies.4 Consequently, earlier delivery is frequently required in twin pregnancies. The Twin Birth Study,5 a large multicentre randomised controlled trial published in 2013, demonstrated the safety of both vaginal and caesarean birth in twin pregnancies where the first twin presented in cephalic position at 32 weeks of gestation or later. These findings have supported an increase in vaginal deliveries for twin pregnancies through induction of labour.

The success rate, benefits, and complications associated with induction of labour in singleton pregnancies have been extensively studied.6 7 However, only a limited number of studies have compared the success rate of induction of labour in twin pregnancies relative to singleton pregnancies.8 9 10 11 According to Loscul et al8 and Okby et al,9 induction of labour in twin pregnancies increases the likelihood of caesarean section. In contrast, Fausett et al10 and Taylor et al11 reported that the risk of caesarean delivery in twin pregnancies is comparable to the risk in singleton pregnancies undergoing induction of labour. These conflicting findings may arise from variations in induction methods, ethnicity-related factors, selection biases, and differences in study designs.

To provide appropriate counselling to patients, obstetricians must understand the likelihood of vaginal delivery after induction of labour. Reliance on data from singleton deliveries to estimate this likelihood for twin pregnancies may be inappropriate due to inherent differences between twin and singleton pregnancies.12 Considering the current lack of robust evidence regarding induction of labour in twin pregnancies, this study aimed to evaluate the rate of caesarean section (including classical or lower segment caesarean sections) and associated outcomes in twin pregnancies undergoing induction of labour, compared with singleton pregnancies.

Our institution, a regional public hospital in Hong Kong, provides obstetric services for 3000 to 5000 deliveries annually. All cases of twin pregnancies are recorded in a specialised twin pregnancy clinic registry and managed by a dedicated team of obstetricians and midwives in the Twin Pregnancy Clinic. The medical professionals overseeing this clinic have specialised expertise in maternal fetal medicine.13 In accordance with departmental protocol, patients attend regular follow-up appointments and undergo ultrasound examinations. When a patient approaches term or requires earlier delivery, the attending obstetrician discusses the mode of delivery with the patient. For uncomplicated dichorionic-diamniotic and monochorionic-diamniotic twin pregnancies, vaginal delivery is encouraged if the first twin presents in cephalic position.

The same induction of labour protocol is applied to both twin and singleton pregnancies. Patients are admitted to the hospital and a cervical examination is conducted to assess the modified Bishop score. If the cervix is unfavourable with a modified Bishop score (Calder score) <6, cervical priming is performed using either dinoprostone tablets or a cervical ripening balloon (Cook Medical, Bloomington [IN], United States). If the cervix is favourable with a modified Bishop score ≥6, the patient is transferred to the labour ward for artificial rupture of membranes and administration of synthetic oxytocin. The induction of labour protocol used in this study aligns with NICE recommendations, except that the modified Bishop score was used instead of the Bishop score to assess cervical readiness for induction.14 15

Medical records of patients with twin pregnancies who underwent induction of labour and delivered at our institution between January 2012 and December 2020 were retrospectively identified using the International Classification of Diseases codes through the Clinical Data Analysis and Reporting System of Hospital Authority. The identified medical records were individually reviewed. Study participants were required to meet all of the following inclusion criteria: gestational age ≥24 weeks, intact membranes, and planned induction of labour. Exclusion criteria included premature rupture of membranes, labour in the latent or active phase, threatened preterm labour resulting in spontaneous labour, and intrauterine fetal death.

Each twin pregnancy patient who underwent induction of labour was matched with a singleton pregnancy patient at a 1:1 ratio in the same hospital during the same study period. Matching was based on specific criteria, including parity (nulliparous or multiparous),16 maternal age (advanced maternal age of ≥35 years, or not),16 17 and the indication for induction of labour.18 These criteria were selected to minimise confounding factors that could affect the success rate of induction of labour. To further reduce the impact of variations in medical practice during the study period, the singleton pregnancy patient with the delivery date closest to that of the twin pregnancy patient was selected.

For both twin and singleton pregnancies, demographic data, past obstetric history, parity, modified Bishop score, method of cervical priming, indication for induction of labour, and use of epidural analgesia were recorded. The primary outcome was the mode of delivery. Secondary outcomes included the time from oxytocin infusion to vaginal delivery or caesarean section, indications for caesarean section or instrumental delivery, and maternal and neonatal outcomes. Postpartum haemorrhage was defined as blood loss of ≥500 mL, regardless of the mode of delivery. Patients who underwent caesarean section for the second twin after vaginal delivery of the first twin were considered to have undergone caesarean section.

To calculate the required sample size, it was assumed that the incidences of caesarean section were 25% in the control group and 40% in the study group. The proportion of discordant pairs was assumed to be 0.45. A two-sided significance level of 0.05 was selected, and the study aimed to achieve a 1:1 comparison between the groups. Calculations in G*Power software (version 3.1.9.6; Erdfelder, Faul, & Buchner, Germany) indicated that a total sample size of 160 pairs would provide 80% power for the analysis.

Categorical variables are reported as numerator and denominator values (%), whereas continuous variables are presented as mean±standard deviation. McNemar’s test was used to analyse the primary outcome for twin pregnancies and their matched controls. Data for the matched pairs are presented in a 2×2 table, showing concordant and discordant study pairs. Odds ratios (ORs) were calculated as the ratio of discordant pairs, and the test statistic was derived from McNemar’s test. For the remaining outcomes, paired t tests, Wilcoxon signed rank test and analyses of variance were used to analyse continuous variables, whereas McNemar’s tests or Fisher’s exact tests were utilised for categorical variables when comparing the case and control groups. For non-matched data, the Chi squared test was applied for categorical data, and unpaired t tests were used for normally distributed continuous data.

Statistical analyses of data using McNemar’s test were performed with Epi Info (version 7.2.5.0; Centers for Disease Control and Prevention, Atlanta [GA], US). All other analyses were conducted with SPSS (Windows version 27.0; IBM Corp, Armonk [NY], US). Two-sided P values <0.05 were considered statistically significant.

During the study period, 760 women with twin pregnancies were recorded out of 42 280 maternity cases. Of these, 160 women met the inclusion criteria for this study. The incidence of twin pregnancies was 1.8% and the rate of induction of labour in twin pregnancies was 21.1%.

The study group consisted of women with twin pregnancies who underwent induction of labour, whereas the control group comprised women with singleton pregnancies who delivered at the same hospital during the same period. The two groups were matched in terms of age, parity and indication for induction of labour, and were well balanced with respect to these matching factors. Patients with twin pregnancies had a significantly lower body mass index (21.1 kg/m² vs 22.2 kg/m²; P=0.008) and a significantly higher modified Bishop score (6.2 vs 5.4; P<0.001) relative to the control group. The mean gestational age at delivery was significantly earlier in twin pregnancies than in the control group (37.1 weeks vs 40.2 weeks; P<0.001). Other baseline characteristics, including ethnicity, prior caesarean section, and use of epidural anaesthesia, did not significantly differ between the two groups (Table 1).

Out of 160 pairs, 44 were discordant (ie, one member of the pair had a caesarean section and the other had a vaginal delivery), and 116 were concordant (ie, both members of the pair had either a caesarean section or a vaginal delivery) [Fig]. Patients with twin pregnancies who underwent induction of labour had a significantly higher risk of caesarean section relative to those with singleton pregnancies (OR=2.14, 95% confidence interval [CI]=1.14-4.04; P=0.024).

Among the patients with twin pregnancies, there were 118 vaginal deliveries, 41 caesarean sections, and one case in which the first twin was delivered vaginally and the second twin was delivered by caesarean section. Among the patients with singleton pregnancies, there were 133 vaginal deliveries and 27 caesarean sections. Instrumental deliveries were performed in 32 patients with twin pregnancies and eight patients with singleton pregnancies (Table 2).

There was no significant difference between the groups in the time from oxytocin administration to vaginal delivery (P=0.143) or the time from oxytocin administration to caesarean section (P=0.054). In total, eight patients with twin pregnancies and three patients with singleton pregnancies delivered after dinoprostone insertion without requiring artificial rupture of membranes or oxytocin infusion. Three patients with twin pregnancies and one patient with a singleton pregnancy had an unfavourable cervix after repeated doses of dinoprostone; thus, caesarean section was performed (Table 2).

Twin pregnancies were associated with significantly greater blood loss relative to singleton pregnancies (median: 400 mL vs 250 mL; P<0.001) and a higher incidence of postpartum haemorrhage (35.0% vs 10.6%; P<0.001). However, there was no significant difference between the groups in blood transfusion rates (8.1% vs 3.8%; P=0.115). The aetiology of postpartum haemorrhage and need for second-line treatments were also comparable between the two groups (Table 2).

Among those 118 vaginal deliveries, 45 had vaginal cephalic deliveries of both twins, whereas 42 had a vaginal cephalic delivery of the first twin followed by a vaginal breech delivery of the second twin. Additionally, 32 patients required instrumental delivery with vacuum or forceps for at least one twin (Table 3).

In cases where vaginal breech delivery was required for the second twin, most babies were in breech presentation. Internal podalic version was performed in 16 patients (13.6%) to facilitate delivery of the second twin (Tables 3 and 4). Notably, even in five cases where the second twin was in cephalic presentation, internal podalic version was performed by manual upward displacement of the fetal head to expedite delivery due to fetal bradycardia or cord presentation.

When neonatal outcomes were compared between the two groups, no significant differences were observed in Apgar scores at 1 and 5 minutes or in rates of admission to neonatal intensive care units. However, both the first and second twins were significantly lighter in weight relative to neonates in singleton pregnancies (Table 5).

This case-control study, utilising matched controls, demonstrated that the rate of failed induction of labour was significantly higher in twin pregnancies than in singleton pregnancies. Nevertheless, 73.8% of patients with twin pregnancies achieved successful vaginal deliveries. This study represents the largest cohort investigation of its kind in a predominantly Chinese population and differs from previous studies conducted in Western countries.8 9 10 11 A previous study19 revealed that ethnic variation can influence the success of induction of labour; thus, our findings provide valuable insights for counselling and managing Chinese patients with twin pregnancies.

The literature on induction of labour in twin pregnancies compared with singleton pregnancies remains limited. The present findings are consistent with those reported by Loscul et al8 and Okby et al,9 both of which identified an increased risk of caesarean delivery after induction of labour in twin pregnancies. Loscul et al8 reported an adjusted OR of 1.8 (95% CI=1.4-2.2), whereas Okby et al9 reported an adjusted OR of 2.2 (95% CI=1.7-2.7). Similarly, the current study demonstrated an OR of 2.14, reinforcing the notion that induction of labour in twin pregnancies is associated with a higher rate of caesarean section relative to singleton pregnancies. However, limitations existed in these studies. For instance, the large cohort study by Loscul et al,8 which included 1995 twin deliveries and 2771 singleton deliveries, did not consider chorionicity, and the methods of induction were not described; considering the multicentre retrospective design of that study, interhospital variations may have existed in terms of induction methods, intrapartum assessment, and decisions regarding caesarean section. Furthermore, Okby et al9 included 191 twin deliveries and 25 913 singleton deliveries, but did not provide details regarding induction methods, cervical status prior to induction, Bishop score, or the chorionicity of twin pregnancies. Conversely, Fausett et al10 and Taylor et al11 included smaller cohorts of twin pregnancies (62 and 100 patients, respectively), and their findings may have been influenced by the small sample sizes. The method of random patient selection used in the control group of the study by Taylor et al11 may have introduced potential bias. Another factor potentially contributing to differences in findings among these studies is ethnic variation.

Physiological differences in the myometrium between twin and singleton pregnancies may explain the higher incidence of failed labour induction in twin pregnancies. Research has shown that myometrial activity in twin pregnancies is characterised by shorter and more frequent contractions compared with singleton pregnancies, particularly at term.20 Shortened contraction duration may result in ineffective and dysfunctional contractions, increasing the likelihood of failed labour induction. Additionally, the uterus undergoes greater distension and stretching in twin pregnancies. Physiological studies have indicated that increased myometrial stretching is associated with reduced uterine contraction in response to oxytocin stimulation.21 One potential mechanism for this phenomenon is that prolonged stretching enhances the expression or activity of TWIK-related K+ channels, which subsequently diminish myometrial contraction in response to oxytocin.21 Further physiological and molecular investigations are warranted to explore the differences between singleton and twin pregnancies in greater detail.

In conjunction with the primary outcomes, the secondary outcomes of this study provide important clinical insights and have substantial implications. Notably, 26.3% of twin pregnancies delivering vaginally required a vaginal breech delivery for the second twin (Tables 3 and 4). Therefore, we recommend that senior obstetricians with expertise in internal podalic version and breech extraction be present during such deliveries.

This study identified one patient who required a caesarean section for the second twin after the first twin had been delivered vaginally. When the first twin was delivered vaginally in our cohort, the probability of caesarean section for the second twin was 0.8%. Patients should be informed of this potential risk prior to induction of labour. Previous studies have shown that the risk of caesarean section for the second twin after vaginal delivery of the first twin ranges from 4.3% to 10.7%.5 22 23 24 In our cohort, the percentage of caesarean deliveries for second twins was much lower than that in other series, including another retrospective study conducted in Hong Kong with the same ethnic population.22 23 24 This discrepancy may be attributed to selection bias because the present study included only patients undergoing induction of labour, whereas other studies included patients with both induction of labour and spontaneous onset of labour. Furthermore, elective induction of labour for twin pregnancies in our unit is typically scheduled during daytime hours. This practice ensures the availability of experienced staff proficient in internal podalic version, potentially improving the likelihood of successful vaginal delivery for the second twin. Our findings showed that 13.6% of cases involving second twin deliveries required internal podalic version, primarily due to transverse or oblique lie (Table 4). Even when the second twin presented in cephalic position (as observed in five cases), internal podalic version was required to expedite delivery because of complications such as cord presentation or fetal bradycardia. The presence of experienced staff skilled in performing internal podalic version can significantly increase the likelihood of achieving successful vaginal delivery for the second twin. At our hospital, vaginal twin deliveries during daytime hours are typically supervised by experienced obstetric consultants or associate consultants.

To minimise the impact of variations in medical practice during the study period, we utilised a rigorous matching approach in which the singleton pregnancy patient with the delivery date closest to that of the twin pregnancy patient was selected. This approach effectively reduced the potential for confounding factors, including variations in medical practices, and ensured that the same induction of labour protocol was applied to both patient groups. Also, we recorded detailed information concerning chorionicity, indications for induction of labour, and the modified Bishop score. Finally, the induction of labour protocol used in this study aligns with NICE recommendations; thus, the results are applicable to other centres that use a similar protocol.

However, this study had some limitations. The earlier gestational age at delivery in the twin pregnancy group, as recommended by international guidelines4 even for uncomplicated twin pregnancies, may have affected the efficacy of induction of labour and influenced the overall outcomes. Additionally, patients with twin pregnancies had a higher initial modified Bishop score relative to those with singleton pregnancies. This difference may be due to selection bias because obstetricians often discourage vaginal delivery in patients with low initial modified Bishop scores; instead, they recommend caesarean section. Despite the higher initial modified Bishop score in twin pregnancies, the success rate of vaginal delivery remained lower in this group than in singleton pregnancies, suggesting that this factor did not significantly influence the study’s results.

Although the induction of labour protocol was consistent for both twin and singleton pregnancies, variations in obstetricians’ assessments of cervical dilatation, labour progression, and confidence in managing vaginal twin deliveries may have influenced study outcomes. Obstetricians with less experience or confidence in vaginal twin delivery may have been more likely to diagnose failed induction of labour and proceed with caesarean section. However, no statistically significant difference was observed in the time from initiation of oxytocin to selection of caesarean section between the two groups. The retrospective nature of the study introduced potential biases, and the limited incidence of twin pregnancies in a single regional hospital restricted the sample size. A larger sample size and multicentre design would enhance the generalisability of the findings. Furthermore, because the study primarily included Chinese patients, the applicability of these conclusions to other ethnic groups is limited; there is a need for further research in this area.

The failure rate of induction of labour was higher in twin pregnancies than in singleton pregnancies. Nevertheless, 73.8% of patients with twin pregnancies achieved successful vaginal deliveries; approximately 20% required instrumental delivery for at least one twin. Furthermore, twin pregnancies were associated with a higher incidence of postpartum haemorrhage. These findings can help facilitate informed decision making for patients and obstetricians when considering induction of labour and selecting the most appropriate mode of delivery for patients with twin pregnancies.

Concept or design: CK Wong, WL Lau.
Acquisition of data: CK Wong.
Analysis or interpretation of data: CK Wong.
Drafting of the manuscript: CK Wong, CMW Hung.
Critical revision of the manuscript for important intellectual content: All authors.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

The authors thank the Kwong Wah Hospital Clinical Research Centre and Mr Steven Lau, Biostatistician from the Centre for Clinical Research and Biostatistics of The Chinese University of Hong Kong, for their statistical advice.

This work was posted on Authorea as a registered online preprint (https://doi.org/10.22541/au.169320065.53300017/v1).

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This research was approved by the Kowloon Central Cluster/Kowloon East Cluster Research Ethics Committee of Hospital Authority, Hong Kong (Ref No.: KC/KE-22-0113/ER-3). The requirement for informed patient consent was waived by the Committee due to the retrospective nature of the research and the use of anonymised data in the research.

1. Monden C, Pison G, Smits J. Twin peaks: more twinning in humans than ever before. Hum Reprod 2021;36:1666-73. Crossref

2. Santana DS, Cecatti JG, Surita FG, et al. Twin pregnancy and severe maternal outcomes: the World Health Organization multicountry survey on maternal and newborn health. Obstet Gynecol 2016;127:631-41. Crossref

3. Cheong-See F, Schuit E, Arroyo-Manzano D, et al. Prospective risk of stillbirth and neonatal complications in twin pregnancies: systematic review and meta-analysis. BMJ 2016;354:i4353. Crossref

4. National Institute for Health and Care Excellence. NICE guideline [NG137]. Twin and triplet pregnancy. London: National Institute for Health and Care Excellence; 2019.

5. Barrett JF, Hannah ME, Hutton EK, et al. A randomized trial of planned cesarean or vaginal delivery for twin pregnancy. N Engl J Med 2013;369:1295-305. Crossref

6. Sotiriadis A, Petousis S, Thilaganathan B, et al. Maternal and perinatal outcomes after elective induction of labor at 39 weeks in uncomplicated singleton pregnancy: a meta-analysis. Ultrasound Obstet Gynecol 2019;53:26-35. Crossref

7. Stock SJ, Ferguson E, Duffy A, Ford I, Chalmers J, Norman JE. Outcomes of elective induction of labour compared with expectant management: population-based study. BMJ 2012;344:e2838. Crossref

8. Loscul C, Schmitz T, Blanc-Petitjean P, Goffinet F, Le Ray C; JUMODA and MEDIP study groups. Risk of cesarean after induction of labor in twin compared to singleton pregnancies. Eur J Obstet Gynecol Reprod Biol 2019;237:68-73. Crossref

9. Okby R, Shoham-Vardi I, Ruslan S, Sheiner E. Is induction of labor risky for twins compare to singleton pregnancies? J Matern Fetal Neonatal Med 2013;26:1804-6. Crossref

10. Fausett MB, Barth WH Jr, Yoder BA, Satin AJ. Oxytocin labor stimulation of twin gestations: effective and efficient. Obstet Gynecol 1997;90:202-4. Crossref

11. Taylor M, Rebarber A, Saltzman DH, Klauser CK, Roman AS, Fox NS. Induction of labor in twin compared with singleton pregnancies. Obstet Gynecol 2012;120:297-301. Crossref

12. Amikam U, Hiersch L, Barrett J, Melamed N. Labour induction in twin pregnancies. Best Pract Res Clin Obstet Gynaecol 2022;79:55-69. Crossref

13. Yung WK, Liu AL, Lai SF, et al. A specialised twin pregnancy clinic in a public hospital. Hong Kong J Gynaecol Obstet Midwifery 2012;12:21-32.

14. National Institute for Health and Care Excellence. NICE guideline [NG207]. Inducing labour. London: National Institute for Health and Care Excellence; 2021.

15. Thomas J, Kavanagh J, Kelly A, editors. RCOG Evidence-based Clinical Guidelines Induction of labour. RCOG Press; 2001.

16. Batinelli L, Serafini A, Nante N, Petraglia F, Severi FM, Messina G. Induction of labour: clinical predictive factors for success and failure. J Obstet Gynaecol 2018;38:352-8. Crossref

17. Jeong Y, Choo SP, Yun J, Kim EH. Effect of maternal age on maternal and perinatal outcomes including cesarean delivery following induction of labor in uncomplicated elderly primigravidae. Medicine (Baltimore) 2021;100:e27063. Crossref

18. Chan YY, Lo TK, Yu EL, Ho LF. Indications for induction of labour and mode of delivery in nulliparous term women with an unfavourable cervix. Hong Kong J Gynaecol Obstet Midwifery 2021;21:69-75. Crossref

19. Papoutsis D, Antonakou A, Tzavara C. The effect of ethnic variation on the success of induced labour in nulliparous women with postdates pregnancies. Scientifica (Cairo) 2016;2016:9569725. Crossref

20. Turton P, Arrowsmith S, Prescott J, et al. A comparison of the contractile properties of myometrium from singleton and twin pregnancies. PLoS One 2013;8:e63800. Crossref

21. Yin Z, He W, Li Y, et al. Adaptive reduction of human myometrium contractile activity in response to prolonged uterine stretch during term and twin pregnancy. Role of TREK-1 channel. Biochem Pharmacol 2018;152:252-63. Crossref

22. Mok SL, Lo TK. Vaginal delivery of second twins: factors predictive of failure and adverse perinatal outcomes. Hong Kong Med J 2022;28:376-82. Crossref

23. Tang HT, Liu AL, Chan SY, et al. Twin pregnancy outcomes after increasing rate of vaginal twin delivery: retrospective cohort study in a Hong Kong regional obstetric unit. J Matern Fetal Neonatal Med 2016;29:1094-100. Crossref

24. Kong CW, To WW. The predicting factors and outcomes of caesarean section of the second twin. J Obstet Gynaecol 2017;37:709-13. Crossref

Mammography is the principal modality used for breast cancer screening and detection in women worldwide.1 However, 10% to 30% of breast cancers may be undetected during mammography due to factors such as dense breast tissue, poor imaging technique, perceptual error, and subtle mammographic abnormalities.2

Conventional computer-aided diagnosis systems have been developed for more than two decades; however, large-scale studies have shown no significant benefit of such systems in enhancing radiologists’ diagnostic performance.3 4 Such systems do not facilitate differentiation between benign and malignant breast lesions, resulting in numerous false-positive results that require radiologist review, which may lead to reader fatigue and unnecessary additional investigations.

Currently, artificial intelligence–based computer-assisted diagnosis (AI-CAD) is widely implemented in mammography to improve diagnostic accuracy and reduce radiologist workload.5 6 The AI-CAD systems developed using deep-learning algorithms make independent decisions and self-learn without the need for feature engineering and computation.7 Artificial intelligence algorithms have been applied to multiple aspects of breast cancer screening, including risk stratification, triage, lesion interpretation, and patient recall.8 As of 2022, the US Food and Drug Administration has approved >15 AI tools for mammography applications, including density assessment, triage, lesion detection, and classification.9 Most commercial AI-CAD programs provide heatmaps with abnormality scores. Generally, higher abnormality scores indicate more suspicious radiological features and a greater likelihood of cancer.

Most existing evidence in the literature is derived from population-based screening studies.5 10 11 However, unlike other developed Asian countries such as Singapore and Korea, Hong Kong lacks a large-scale population screening programme.12 13 Our patient population primarily consists of symptomatic individuals. Evidence concerning the application of AI-CAD in symptomatic breast imaging is limited. This study aimed to evaluate the potential of AI-CAD in Hong Kong, focusing on the impact of radio-pathological phenotypes of breast cancer on AI-CAD performance. We analysed the distinctive characteristics of high-score versus low-score breast cancers. We also discuss observed strengths and limitations of AI-CAD in identifying breast cancer.

In total, 488 consecutive patients with histology-confirmed breast cancers were identified from a prospectively maintained database managed by two tertiary referral centres in Hong Kong during the period between January 2020 and September 2022. In our centres, all patients referred for diagnostic mammography were symptomatic, presenting with various clinical symptoms. We included patients with breast cancers confirmed by core needle biopsy under ultrasound guidance or stereotactic-guided vacuum-assisted breast biopsy performed at our centres. We excluded patients with diagnostic mammography performed at outside facilities (n=6), chest wall recurrence after mastectomy (n=14), cancers identified only in axillary nodes (n=3), tumour locations not feasible for mammography (n=3), and mammographically occult breast cancers undetectable by both reporting radiologists and AI-CAD (n=64) [Fig 1]. Finally, 398 patients with 414 breast cancers and 347 unaffected breasts were included in the study. Sixteen patients were diagnosed with bilateral breast cancer. Among the 382 patients with unilateral breast cancer, 35 had previously undergone contralateral mastectomy.

Full-field digital mammography (MAMMOMAT Inspiration; Siemens, Erlangen, Germany or Selenia Dimensions; Hologic, Newark [DE], US) was performed prior to each biopsy. The included mammograms were exported and processed by a commercial AI-CAD program (INSIGHT MMG, version 1.10.2; Lunit, Seoul, South Korea), which is approved by the US Food and Drug Administration for lesion detection and classification in breast imaging.9

The AI-CAD algorithm used in the current study was developed and validated through multinational studies.14 15 This algorithm provides a heatmap that highlights mammographic abnormalities and generates a score ranging from 0 to 100 for each view (craniocaudal and mediolateral oblique views). The abnormality score is the maximum value for each breast, reflecting the likelihood of malignancy.

All mammograms were interpreted by radiologists subspecialising in breast radiology (with 4 to 20 years of experience in breast imaging). Mammography reports from the time of breast cancer diagnosis were retrieved from the radiology information system and retrospectively reviewed for breast density, dominant mammographic features of breast cancer, and any axillary lymphadenopathy. The clinical findings, pathological results, and molecular profiles of breast cancers were also recorded. Breast density was categorised from 1 to 4 using the BI-RADS (Breast Imaging Reporting and Data System) classification.16 The cancers were classified according to their dominant mammographic features as asymmetrical density, mass (with or without calcifications), calcifications alone, or architectural distortion.

For breast cancers presenting as a mass without calcifications, the tumour distances from the chest wall and the tumour-to–breast area ratio (TBAR) were measured in mammograms using the picture archiving and communication system by a radiologist with 2 years of experience in breast imaging. Tumour distance from the chest wall was defined as the shortest distance between the tumour and the pectoralis major in the mediolateral oblique view (Fig 2a). Tumours partially visible within the lower breast in the mediolateral oblique view, where the pectoralis muscle is not discernible, were assigned a chest wall distance of 0 cm. The TBAR was calculated via division of the tumour area by the breast area, as measured using the freehand region-of-interest tool (Fig 2b).

The radiologists matched the index lesion to the AI-CAD heatmap to determine whether the AI-CAD correctly localised the known cancer. When the cancer was correctly localised by the AI-CAD, an abnormality score of ≥30 was regarded as a true positive, whereas a score &LT;30 was considered a false negative. When the cancer was undetected or incorrectly localised by the AI-CAD, this result also was regarded as a false negative. Breast cancers with abnormality scores of ≥90 and <30 were designated as ‘high-score tumour’ and ‘low-score tumour’, respectively.

Abnormality scores are presented as medians with interquartile ranges. The scores were analysed according to patient symptoms, breast density, mammographic findings, cancer histology, and molecular profile using the Mann-Whitney U test or Kruskal–Wallis H test. The AI-CAD abnormality scores were divided into three intervals: 0 to <30, 30-90, and >90 to 100. The Chi squared test and Mantel-Haenszel test for trend were used to analyse associations with different factors. For cancers presenting as a mass, mean abnormality scores across various TBARs and distances to the chest wall were evaluated using analysis of variance with pairwise comparisons. Statistical analyses were performed using SPSS (Windows version 26; IBM Corp, Armonk [NY], US). P values <0.05 were considered statistically significant.

In total, 398 patients (mean age, 62.4 years; range, 35-100) with 414 breast cancers and 347 unaffected breasts were included in the study. The cohort consisted of two men and 396 women. Among the 414 breast cancer cases, 284 (68.6%) were palpable (Table 1).

The median and mean abnormality scores for the 414 breast cancers were 95.6 and 80.6, respectively (range, 0.4-99.9). The distribution of breast cancers according to abnormality score interval is presented in Figure 3. The sensitivity of the AI-CAD algorithm in detecting breast cancers was 91.5%, based on breast cancer identification using an abnormality score of ≥30. Overall, 65.7% of breast cancers were classified as high-score tumours, whereas 8.5% were classified as low-score tumours with abnormality scores <30; these low-score tumours were regarded as false-negative cases. Table 1 presents the medians and interquartile ranges of abnormality scores according to clinical, radiological, and pathological phenotypes.

Palpable lesions, cancers in entirely fatty or scattered fibroglandular breasts, cancers presenting as masses with or without calcifications and architectural distortion, and larger cancers were associated with higher abnormality scores (all P<0.001) [Table 1]. Invasive cancers had higher abnormality scores compared with ductal carcinoma in situ (P=0.010). Axillary nodal status (P=0.078) and cancer molecular subtype (P=0.820) were not associated with abnormality scores (Table 2).

High-score breast cancers had higher prevalences of palpable disease, cancers presenting as masses with or without calcifications, invasive cancers, and larger cancers (>1 cm) [Table 2].

The false-negative rate for AI-CAD was 8.5% (35/414). These cancers had higher prevalences of non-palpable disease, cancers presenting as asymmetrical densities, small cancers (<1 cm), and locations in heterogeneously dense or extremely dense breast tissue.

Overall, 158 cancers presenting as masses without calcifications were included in this analysis. The mean abnormality score for cancers with a TBAR of ≥30% was significantly lower than for those with a TBAR of <30% (86.7 vs 54.4; P<0.001). Tumours bordering the chest wall (ie, distance of 0 cm from chest wall) demonstrated significantly lower abnormality scores compared with those located 1 cm and ≥2 cm away from the chest wall (mean, 65.5 vs 89.2 vs 87.2; P<0.001).

In the analysis of 347 unaffected breasts (regarded as negative findings by reporting radiologists), the median abnormality score was 0 (mean, 3.5; range, 0-81). Using a threshold score of 30, the false-positive rate was 3.7% (13/347), indicating 96.3% specificity. Most of these false positives (11/13) scored between 30 and 50; none scored >90. One case with known postoperative changes from breast conservative surgery showed stable mammographic finding for 10 years, scored 81 by AI-CAD. One case with a breast cyst scored 73, which was confirmed via fine needle aspiration cytology.

Most AI-CAD algorithms provide heatmaps with abnormality scores ranging from 0 to 100; a higher score generally implies a greater likelihood of cancer. Previous AI-CAD studies have used various threshold scores; some set a threshold of 10 for population screening,18 19 20 whereas Weigel et al21 set a threshold of 28 for detecting malignant calcifications. However, the clinical implications of the abnormality score itself have not been clarified; a score range from 10 to 100 may be too broad for distinguishing malignancies in clinical practice. These aspects highlight the need for further validation of the appropriate reference score provided by AI-CAD algorithms. In this study, we set the threshold at 30 because, unlike population screening approaches, our patients were symptomatic individuals. A higher threshold score appears more practical in the clinical setting of symptomatic patients.

In our study, the AI-CAD algorithm detected 91.5% (379/414) of breast cancers with an abnormality score of >30; of these 379 cancers, 71.7% exhibited a high abnormality score of >90. The false-negative rate of 8.5% is comparable to previously reported rate for this AI-CAD algorithm.5

All cancers presenting as architectural distortion in our study were correctly localised by the AI-CAD, with abnormality scores >30; 87.5% of them were assigned high abnormality scores of >90 (Fig 4a and b). Unlike cancers presenting as masses or calcifications, cancers presenting as architectural distortion remain challenging for radiologists to detect and interpret.22 23 24 Wan et al25 showed that a standalone AI algorithm did not outperform radiologists; however, with AI assistance, junior radiologists demonstrated significant improvements in diagnostic accuracy for architectural distortion.

One case of breast cancer presenting as asymmetric density in heterogeneously dense breast tissue was missed by the reporting radiologist but detected by AI-CAD, which assigned an abnormality score of 68. The cancer was later identified by the radiologist via ultrasound, which is part of routine workup for symptomatic patients in our centre. Retrospective review indicated that the asymmetric density was visible on mammography (Fig 4c and d). In a study by Kim et al,26 40 of 128 mammographically occult breast cancers were correctly identified by the AI algorithm, demonstrating its added value in detecting such cancers.

The 64 cases of mammographically occult breast cancer not detected by either the AI-CAD or the radiologists were excluded from the study. Of these cases, 84.3% were found in heterogeneously dense and extremely dense breast tissue (BI-RADS 3 and 4).16 Dense breast tissue is recognised as a significant feature associated with mammographically occult and missed cancers.27 28 29 30 We suspect that mammographic signs of cancer are masked or obscured by dense breast parenchyma, thus evading detection by the AI-CAD. Conversely, both radiologists and the AI-CAD tended to more effectively detect cancers in fatty breasts.18

In our study, the AI-CAD correctly localised a small breast cancer with a high abnormality score (>90) in a patient with polyacrylamide hydrogel (PAAG)–injected augmentation mammoplasty (Fig 4e and f). The diagnosis of breast cancer after PAAG-injected augmentation mammoplasty is challenging. Lesion visualisation may be masked by the presence of polyacrylamide gel, and extravasated polyacrylamide gel may mimic a lesion on mammography, potentially delaying early cancer detection. In such cases, assessments of suspicious calcifications and parenchymal distortion within visible breast parenchyma are considered the main goals of screening mammography.31 32 The effectiveness of AI-CAD in detecting breast cancer among patients with augmentation mammaplasty remains uncertain, warranting further studies.

Isolated cases of large, clearly visible lesions that evaded AI detection have been described by Lång et al33 and Choi et al.18 To our knowledge, our study is the first to investigate factors contributing to such evasion. In this study, the AI algorithm tended to underscore cancers presenting as large masses (Fig 5a and b). Cancers with a TBAR of ≥30% had significantly lower mean abnormality scores relative to those with a ratio of <30%. Tumours bordering the chest wall (0 cm distance) also showed significantly lower abnormality scores than those located away from the chest wall. The underlying cause remains unclear; however, these findings highlight concerns regarding the use of AI-CAD as a standalone tool for triaging cases in symptomatic populations. We also noted that the AI-CAD missed certain cancers with obvious findings, such as nipple retraction and diffuse dermal thickening (Fig 5c to f).

Moreover, the inability of AI-CAD to compare mammograms with previous studies may hinder its effectiveness in specific scenarios, such as the detection of subtle developing symmetries and identification of early recurrence in postoperative cases (Fig 5g to i). In contrast, radiologists can compare mammograms with previous studies, improving mammogram interpretation accuracy.

Studies have shown that the diagnostic performances of AI algorithms are comparable to those of radiologists in terms of assessing screening mammograms; the use of AI to triage screening mammograms could potentially reduce radiologists’ workload.5 34 35 We identified potential limitations and weaknesses of AI-CAD in diagnosing breast cancers under certain conditions, highlighting the need for further large-scale studies to investigate clinical applications of AI-CAD in symptomatic patients.

This study had several key strengths. To our knowledge, it is the first to evaluate AI-CAD for breast cancer detection in Hong Kong, using an AI-CAD system that had not previously been exposed to images from our centres during their product development. Additionally, all digital mammograms were obtained before biopsies, avoiding any biopsy-related changes which could potentially affect AI-CAD performance. Limitations of the study include its retrospective design and inclusion of cancer-enriched datasets, which may lead to overestimation of AI-CAD performance; the use of a single AI vendor, hindering applicability to other AI algorithms; and the lack of BI-RADS correlation. Furthermore, there was a lack of information concerning progression in unaffected breasts over an extended follow-up interval (≥2 years), which could impact the false-positive rate of the AI-CAD. An extended observation period is needed to identify potential malignancies that may have been initially missed by radiologists.

Unlike other developed cities or countries, Hong Kong does not have population-based screening programmes. The adoption and implementation of AI programs in Hong Kong for breast imaging remains in early stages, mainly due to ongoing debates about efficacy and a lack of sufficient local data to support widespread application. Current literature is almost entirely based on population screening data, which may not be applicable to cities without screening programmes. In our study, AI-CAD demonstrated promising accuracy in detecting breast cancers within symptomatic settings; its performance varied according to radio-pathological characteristics. To translate these research findings into practical clinical applications, further validation studies with larger sample sizes are required; these would confirm the reliability of AI-CAD systems. The development of protocols for integrating AI-CAD into existing clinical workflows, formulation of usage guidelines, and initiation of training programmes for radiologists to effectively utilise AI as a second reader are essential elements of this process. Collaborations with information technology departments and hospital management are necessary to ensure successful integration. Although further investigation is needed, this study provides encouraging evidence to support the use of AI-CAD as a breast cancer detection tool in symptomatic settings, ultimately benefitting patients.

Concept or design: SM Yu, MNY Choi, EHY Hung, HHL Chau.
Acquisition of data: SM Yu, MNY Choi, TH Chan, CYM Young, YH Chan, YS Chan, C Tsoi.
Analysis or interpretation of data: SM Yu, TH Chan, J Leung.
Drafting of the manuscript: SM Yu, CYM Young.
Critical revision of the manuscript for important intellectual content: SM Yu, CYM Young, WCW Chu, EHY Hung, HHL Chau.

All authors had full access to the data, contributed to the study, approved the final version for publication, and take responsibility for its accuracy and integrity.

All authors have disclosed no conflicts of interest.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

This research was approved by the New Territories East Cluster Research Ethics Committee/Institutional Review Board of Hospital Authority, Hong Kong (Ref No.: NTEC-2023-074). The requirement for informed patient consent was waived by the Committee due to the retrospective nature of the research.

1. Tabár L, Vitak B, Chen HH, Yen MF, Duffy SW, Smith RA. Beyond randomized controlled trials: organized mammographic screening substantially reduces breast carcinoma mortality. Cancer 2001;91:1724-31. Crossref

2. Majid AS, de Paredes ES, Doherty RD, Sharma NR, Salvador X. Missed breast carcinoma: pitfalls and pearls. Radiographics 2003;23:881-95. Crossref

3. Fenton JJ, Taplin SH, Carney PA, et al. Influence of computer-aided detection on performance of screening mammography. N Engl J Med 2007;356:1399-409. Crossref

4. Lehman CD, Wellman RD, Buist DS, et al. Diagnostic accuracy of digital screening mammography with and without computer-aided detection. JAMA Intern Med 2015;175:1828-37. Crossref

5. Dembrower K, Wåhlin E, Liu Y, et al. Effect of artificial intelligence–based triaging of breast cancer screening mammograms on cancer detection and radiologist workload: a retrospective simulation study. Lancet Digit Health 2020;2:e468-74. Crossref

6. Raya-Povedano JL, Romero-Martín S, Elías-Cabot E, Gubern-Mérida A, Rodríguez-Ruiz A, Álvarez-Benito M. AI-based strategies to reduce workload in breast cancer screening with mammography and tomosynthesis: a retrospective evaluation. Radiology 2021;300:57-65. Crossref

7. Erickson BJ, Korfiatis P, Kline TL, Akkus Z, Philbrick K, Weston AD. Deep learning in radiology: does one size fit all? J Am Coll Radiol 2018;15(3 Pt B):521-6. Crossref

8. Schünemann HJ, Lerda D, Quinn C, et al. Breast cancer screening and diagnosis: a synopsis of the European breast guidelines. Ann Intern Med 2020;172:46-56. Crossref

9. Bahl M. Artificial intelligence: a primer for breast imaging radiologists. J Breast Imaging 2020;2:304-14. Crossref

10. Hickman SE, Woitek R, Le EP, et al. Machine learning for workflow applications in screening mammography: systematic review and meta-analysis. Radiology 2022;302:88-104. Crossref

11. Leibig C, Brehmer M, Bunk S, Byng D, Pinker K, Umutlu L. Combining the strengths of radiologists and AI for breast cancer screening: a retrospective analysis. Lancet Digit Health 2022;4:e507-19. Crossref

12. Intelligence Unit, The Economist. Breast cancer in Asia—the challenge and response. A report from the Economist Intelligence Unit. 2016. Available from: https://www.eiuperspectives.economist.com/sites/default/files/EIU Breast Cancer in Asia_Final.pdf. Accessed 19 Nov 2017.

13. Lim YX, Lim ZL, Ho PJ, Li J. Breast cancer in Asia: incidence, mortality, early detection, mammography programs, and risk-based screening initiatives. Cancers (Basel) 2022;14:4218. Crossref

14. Salim M, Wåhlin E, Dembrower K, et al. External evaluation of 3 commercial artificial intelligence algorithms for independent assessment of screening mammograms. JAMA Oncol 2020;6:1581-8. Crossref

15. Kim HE, Kim HH, Han BK, et al. Changes in cancer detection and false-positive recall in mammography using artificial intelligence: a retrospective, multireader study. Lancet Digit Health 2020;2:e138-48. Crossref

16. D’Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BIRADS® Atlas, Breast Imaging Reporting and Data System. Reston, VA: American College of Radiology; 2013.

17. McKinney SM, Sieniek M, Godbole V, et al. International evaluation of an AI system for breast cancer screening. Nature 2020;577:89-94. Crossref

18. Choi WJ, An JK, Woo JJ, Kwak HY. Comparison of diagnostic performance in mammography assessment: radiologist with reference to clinical information versus standalone artificial intelligence detection. Diagnostics (Basel) 2022;13:117. Crossref

19. Lee SE, Han K, Yoon JH, Youk JH, Kim EK. Depiction of breast cancers on digital mammograms by artificial intelligence–based computer-assisted diagnosis according to cancer characteristics. Eur Radiol 2022;32:7400-8. Crossref

20. Koch HW, Larsen M, Bartsch H, Kurz KD, Hofvind S. Artificial intelligence in BreastScreen Norway: a retrospective analysis of a cancer-enriched sample including 1254 breast cancer cases. Eur Radiol 2023;33:3735-43. Crossref

21. Weigel S, Brehl AK, Heindel W, Kerschke L. Artificial intelligence for indication of invasive assessment of calcifications in mammography screening [in English, German]. Rofo 2023;195:38-46. Crossref

22. Suleiman WI, McEntee MF, Lewis SJ, et al. In the digital era, architectural distortion remains a challenging radiological task. Clin Radiol 2016;71:e35-40. Crossref

23. Babkina TM, Gurando AV, Kozarenko TM, Gurando VR, Telniy VV, Pominchuk DV. Detection of breast cancers represented as architectural distortion: a comparison of full-field digital mammography and digital breast tomosynthesis. Wiad Lek 2021;74:1674-9. Crossref

24. Alshafeiy TI, Nguyen JV, Rochman CM, Nicholson BT, Patrie JT, Harvey JA. Outcome of architectural distortion detected only at breast tomosynthesis versus 2D mammography. Radiology 2018;288:38-46. Crossref

25. Wan Y, Tong Y, Liu Y, et al. Evaluation of the combination of artificial intelligence and radiologist assessments to interpret malignant architectural distortion on mammography. Front Oncol 2022;12:880150. Crossref

26. Kim HJ, Kim HH, Kim KH, et al. Mammographically occult breast cancers detected with AI-based diagnosis supporting software: clinical and histopathologic characteristics. Insights Imaging. 2022;13:57. Crossref

27. Lian J, Li K. A review of breast density implications and breast cancer screening. Clin Breast Cancer 2020;20:283-90. Crossref

28. Freer PE. Mammographic breast density: impact on breast cancer risk and implications for screening. Radiographics 2015;35:302-15. Crossref

29. Arora N, King TA, Jacks LM, et al. Impact of breast density on the presenting features of malignancy. Ann Surg Oncol 2010;17 Suppl 3:211-8. Crossref

30. Ma L, Fishell E, Wright B, Hanna W, Allan S, Boyd NF. Case-control study of factors associated with failure to detect breast cancer by mammography. J Natl Cancer Inst 1992;84:781-5. Crossref

31. Cheng NX, Liu LG, Hui L, Chen YL, Xu SL. Breast cancer following augmentation mammaplasty with polyacrylamide hydrogel (PAAG) injection. Aesthetic Plast Surg 2009;33:563-69. Crossref

32. Teo SY, Wang SC. Radiologic features of polyacrylamide gel mammoplasty. AJR Am J Roentgenol 2008;191:W89-95. Crossref

33. Lång K, Dustler M, Dahlblom V, Åkesson A, Andersson I, Zackrisson S. Identifying normal mammograms in a large screening population using artificial intelligence. Eur Radiol 2021;31:1687-92. Crossref

34. Rodriguez-Ruiz A, Lång K, Gubern-Merida A, et al. Stand-alone artificial intelligence for breast cancer detection in mammography: comparison with 101 radiologists. J Natl Cancer Inst 2019;111:916-22. Crossref

35. Rodríguez-Ruiz A, Krupinski E, Mordang JJ, et al. Detection of breast cancer with mammography: effect of an artificial intelligence support system. Radiology 2019;290:305-14. Crossref