Mammography attendance in China’s national breast cancer screening program: a county-level analysis of the Fujian province

Abstract

Background

Missed mammography attendance among women screened positive is one of the main challenges facing breast cancer screening program. We sought to analyze county-level mammography non-attendance rates and the spatiotemporal differences in Fujian, China, and to explore factors associated with non-attendance.

Methods

We used county-level non-attendance rates calculated from the Fujian Provincial Maternal and Child Health Information System (2021–2023), complemented by information from the Fujian Provincial Statistical Yearbooks. Theil Indices were used to analyze county-level disparities of non-attendance. Multivariable regressions were used to identify variables associated with non-attendance, with spatiotemporal heterogeneity examined using a geographic temporal weighted regression.

Results

The average county-level mammography non-attendance rate in Fujian decreased from 43.85% in 2021 to 26.89% in 2023. The Theil-T and Theil-L Indices of county-level non-attendance rates over the three years were 0.4 and 0.29, respectively. Per capita GDP was negatively associated with mammography non-attendance in most counties in Fujian, while urbanization rate and in-house mammography capacity showed a positive association in the southwest. Over time, the spatial effects of these factors weakened.

Conclusion

While mammography non-attendance rates declined in recent years in Fujian, persistent geographical disparity shown in this study indicates the importance of addressing service and information gaps. It is likely helpful to integrate the information system and service delivery across counties in breast screening programs to address effects of potentially higher mobility in the western and poverty in the northern parts of Fujian. Considering the geographically varied effects of such factors, we also suggested a classified approach be adopted to improve mammography attendance. These findings underscore the need to incorporate local contextual factors, such as socioeconomic conditions and health resource availability, into both research and policy-making to better understand and reduce spatial disparities in screening adherence.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12905-025-04074-7.

Introduction

Breast cancer is the most commonly diagnosed cancer and a leading cause of cancer-related mortality among women worldwide [1]. Mammography screening stands as a cornerstone of global efforts to mitigate this public health challenge [2]. Since 1990, mortality rates from breast cancer have declined steadily in many high-income countries (e.g., a 44% reduction in the US from 1989 to 2022) [3], largely attributable to increased uptake of mammography screening [4]. However, non-attendance among women who screen positive remains a critical barrier to the effectiveness of screening programs.

In high-income countries, mammography screening is generally accessible to eligible women in high-income countries, participation gaps persist [5, 6]. While individual-level factors such as low income [7], immigrant status [8], and social isolation [9] have been extensively studied, emerging evidence highlights the role of regional disparities in mammography non-attendance. For instance, a French study revealed that census blocks with higher screening participation exhibited significantly lower poverty rates [10], and a US study demonstrated that reduced mammography capacity (i.e. the number of mammography machines per 10,000 women aged 40 or older) correlated with longer appointment wait times [11]. Countries offering fully funded mammography screening, such as Denmark and Finland, reported an average non-attendance rate of 16% [12]. Where partial fees applied (e.g., 22% in Sweden) or out-of-pocket payments existed (e.g., 23% in Norway and 39% in Iceland), rates increased substantially [13].

In China, organized free breast cancer screening began with the 2009 Rural Women's Screening Project, targeting rural women aged 35–59 [14]. Subsequently, the National Health Commission issued the detailed Work Plan for Breast Cancer Screening in 2022, expanding the coverage of the breast cancer screening project to women aged 35–64, prioritizing rural women as well as urban women receiving subsistence allowances [15]. In China, the government has provided free breast cancer screening for women aged 35–64 since 2009. However, regional implementation gaps persist: a mammography screening program in a district of Beijing, reported an average attendance rate of only 29.2% from 2013 to 2016 [16]. Breast cancer screening in China is carried out at the county level, where disparities in socio-economic conditions and healthcare infrastructure could contribute to differing non-attendance rates for mammography across counties.

To date, the spatial patterns of mammography non-attendance or their underlying drivers have not been studied in China. To address these gaps, we investigated the county-level spatiotemporal distribution and factors contributing to mammography non-attendance in Fujian Province, China. Additionally, we analyzed the geographically varied influence of these factors on non-attendance. Fujian Province was selected for this study primarily owing to the availability of county-level data from the Fujian Provincial Maternal and Child Health Information System. The province exhibits substantial internal heterogeneity, including both economically developed coastal areas and lower-income inland regions, providing an ideal context to examine spatial disparities in mammography non-attendance and the differential effects of various factors. Moreover, Fujian follows the national breast cancer screening guidelines, enhancing the generalizability of our findings to elsewhere in China.

Methods

Study setting and data sources

In China, the latest implementation plan for breast cancer screening was issued by the National Health Commission in 2021 [17]. The plan encompasses two key screening phases [18, 19] (see Appendix Figure A1). The first phase includes clinical breast examination (visual inspection and palpation) to identify any abnormal signs, the results of which are integrated with ultrasound findings to determine the need for further assessment. The ultrasound results are evaluated according to the Breast Imaging Reporting and Data System (BI-RADS classification). Based on the BI-RADS classification, individuals with ultrasound results classified as BI-RADS 1 or 2 are recommended to receive screening every 2–3 years; those with results of BI-RADS 0 or 3 should undergo mammography.

The second phase involves mammography: individuals with BI-RADS 1 or 2 are advised for regular follow-up, those with BI-RADS 0 or 3 are evaluated by senior specialists for short-term follow-up, biopsy, or other examinations, and those with BI-RADS 4 or 5 undergo histopathological examination. Mammography is either conducted by the primary screening institutions with such capacity, or at another designated institution in the local county or municipality through referral where in-house mammography is unavailable at the primary screening institutions.

This study extracted de-identified county-level breast cancer screening data and Breast Imaging Reporting and Data System (BI-RADS) classification data (categories 0 or 3) from 2021 to 2023 from the Fujian Provincial Maternal and Child Health Information System, which was approved by Fujian Maternity and Child Health Hospital Ethics Board (2025KY004). Data from Xiamen Municipality were not available in the system and were therefore not included in the analysis. The study involved 77 districts and counties (see Appendix Figure A2), complemented with additional socioeconomic and health system information from the Fujian Statistical Yearbooks. Female participants were divided into six age groups (35–39, 40–44, 45–49, 50–54, 55–59, 60–64 years). This led to a total of 1,386 observations (age group-county-year) over the three years studied.

Variables and measurement

The outcome of interest was mammography non-attendance rate, defined as the proportion of women who missed mammography among those who were eligible (individuals with ultrasound results classified as BI-RADS 0 or 3). It should be noted that among the 128,956 women who underwent ultrasound over the three studied years the vast majority (128,924) were classified as BI-RADS 3, with only 32 classified as BI-RADS 0.

Guided by the Anderson Behavioral Model for health service utilization (see Appendix Table A1), we conceptualized mammography attendance as a health behavior shaped by environmental, predisposing, enabling, and need-related factors [20, 21]. Using such conceptualization and building on prior studies, we investigated the association of mammography non-attendance rate with: (1) environmental factors including health system characteristics (health professionals per 10,000 population and in-house mammography capacity) and external conditions (transportation network density and urbanization rate); (2) predisposing characteristics represented by age group, population density; (3) enabling resources measured as county-level GDP per capita (reflecting financial access), (4) evaluated need represented by the implicit variable of clinical referral based on BI-RADS classification from ultrasound screening (the criteria for inclusion into the denominator of non-adherence rate calculation).

Statistical analysis

After conducting descriptive statistical analyses of participant characteristics, socioeconomic status, and county-level mammography non-attendance rate. To quantify spatiotemporal disparities in county-level mammography non-attendance, we computed the Theil-T and Theil-L indices across counties, as well as between-municipality and within-municipality disparities [22–24]. The Theil-T Index measures absolute difference (i.e. the numerical difference), whereas the Theil-L Index reflects relative difference (i.e. the proportional difference). This approach aimed to identify whether there were spatiotemporal patterns of inequality between and within municipalities.

As the analysis demonstrated substantial inter-county disparity dominated by disparities within municipalities rather than between municipalities, we conducted multivariate ordinary least squares (OLS) regression focusing on the county level associations between non-attendance rates and covariates. We used R 4.2.1 for data analysis mentioned above.

Then, we went on to employ Moran’s Index to assess pattern global spatial autocorrelation of mammography non-attendance rates using ArcGIS software (version 10.8, ESRI, Redlands, CA, USA). Spatial neighbors were defined under the “queen” criterion, which identifies neighboring spatial units as those sharing at least a common edge or vertex, thereby establishing adjacency relationships for spatial autocorrelation analysis [25].

To further examine the potential temporal and spatial heterogeneities of parameters, we applied Geographical and Temporal Weighted Regression (GTWR), widely used in medicine, geography, environmental science and other fields [26, 27]. The GTWR model used in this study can be expressed as follows:

1

For each county , is the dependent variable (i.e. mammography non-attendance rate), whereas is the kth explanatory variable.  represents the space–time coordinates of county i, where and are the projected spatial coordinates in the map, and is the projected temporal coordinate. represents the intercept county i situated its space–time coordinates, and denotes the space–time-specific regression coefficient.

Based on local weighted least squares, we estimated parameters can be given as follows:

2

is a weight matrix, with each observation point i corresponding to a weighted matrix. is the design matrix, representing a set of independent variables, where each row of the matrix represents an observation point (e.g., the feature values of the independent variables for the spatial location of each district). epresents the transpose of the matrix . Details about the GTWR approach is available elsewhere [28]. Moreover, the GTWR model relaxes the assumption of spatial and temporal stationarity inherent in global regression (e.g., OLS), allowing regression coefficients to vary across geographic locations and time points. This approach assumes that: (1) Parameters are spatially and temporally non-stationary. In other words, the relationships between mammography non-attendance and its determinants (e.g., GDP, urbanization) may differ across counties and years. (2) Observations are weighted by spatiotemporal proximity. Data points closer in space and time to location exert greater influence on its local parameter estimates, governed by a distance-decay kernel function.

Results

Table 1 shows the characteristics of the sample counties and the mammography non-attendance rate. The average population density for each sample county was 0.13 per 10,000 people/km². The average per capita GDP was 118.4 thousand Chinese yuan. Mean urbanization rate was 64.53%. The average transportation network density was 1.15 km/km². 29.87% of counties had in-house mammography capacity. The average non-attendance screening rate for mammography decreased from 43.85% in 2021 to 26.89% in 2023.

Table 1.

Characteristics of sample counties and mammography non-attendance rate

Characteristic (N = 1386)	Mean (SD)
Population density (10,000 people/km2)	0.133 (0.357)
Per capita GDP (10,000 Chinese yuan)	11.844 (4.775)
Urbanization (%)	64.53 (15.349)
Transportation network densitya (km/km2)	1.151 (0.608)
Per capita number of healthcare professionals (per 10,000 population)	73.777 (44.24)
In-house mammography capacityb
Yes	414 (29.87%)
No	972 (70.13%)
Rate of mammography non-attendance (%)
2021	43.852 (36.716)
2022	40.908 (34.031)
2023	26.893 (27.365)
All three years	37.218 (33.738)

Abbreviation: SD, Standard deviation

The value in italics in parentheses represents the proportion of institutions with in-house mammography capacity relative to the total number of institutions

^aTransportation network density is defined as the total length of all types of roads (e.g., urban roads, rural roads, highways, etc.) per square kilometer of area

^bIn-house mammography capacity refers to the ability of primary screening institutions in a county to provide mammography services without requiring patients to be referred to other primary screening institutions in the local county or municipality

Significant spatiotemporal variations in the mammography non-attendance rate were observed at the county level in Fujian (see Fig. 1). From 2021 to 2022, counties with a non-attendance rate above 80% were mainly concentrated in the northern and southwestern. The central region had a lower non-attendance rate, where most counties had rates below 40%. By 2023, areas with a non-attendance rate above 80% were limited to the northern region. The number of counties with a non-attendance rate below 20% increased from 23 in 2021 to 41 in 2023, while the number of counties with non-attendance rates above 80% decreased from 16 to 4.

Fig. 1.

Spatial distribution of mammography non-attendance rate in Fujian China, 2021–2023

The differences in the mammography non-attendance rate primarily existed within municipalities (see Table 2). The Theil-T Index for the three years studied was 0.4, while the and Theil-L Index for the three years was 0.29. Within-municipalities differences were the dominant share in overall differences using either index, namely 86% using Theil-T and 83.56% using Theil-L. The Theil-T Index increased from 0.35 in 2021 to 0.45 in 2023. Similarly, the Theil-L Index rose from 0.23 in 2021 to 0.36 in 2023.

Table 2.

Theil-T Index and Theil-L Index of county-level mammography non-attendance rate, in 2021–2023

Theil index	2021	2022	2023	All three years
Theil-T	0.348	0.325	0.453	0.4
Theil-T within municipalities	0.283	0.251	0.371	0.344
Theil-T between municipalities	0.065	0.074	0.082	0.056
Theil-L	0.234	0.249	0.358	0.292
Theil-L within municipalities	0.177	0.185	0.288	0.244
Theil-L between municipalities	0.057	0.064	0.07	0.048

Table 3 shows that per capita GDP was negatively associated with the mammography non-attendance rate (−0.72, 95% CI: −1.18 to −0.26, P = 0.002), as were urbanization rate (−0.26, 95% CI: −0.42 to −0.1, P < 0.001) and in-house mammography capacity (−6.96, 95% CI: −10.87 to −3.05, P < 0.001). Notably, no significant difference was observed across age groups (See Appendix Table A2).

Table 3.

Multiple variable regression on factors potentially associated with non-attendance

Exploratory variables	Coef	95% CI	P-value
Population density	−2.923	[−12.284, 6.437]	0.54
Per capita GDP	−0.723	[−1.183, −0.264]	0.002
Urbanization rate	−0.26	[−0.419, −0.101]	0.001
Transportation network density	1.174	[−2.222, 4.57]	0.498
Per capita number of healthcare professionals	0.017	[−0.059, 0.094]	0.658
In-house mammography capacity (reference: No)	−6.961	[−10.869, −3.053]	0.000
Age group (reference: 35–39 years old)
40–44	−0.597	[−6.654, 5.46]	0.847
45–49	−0.648	[−6.705, 5.409]	0.834
50–54	1.531	[−4.526, 7.588]	0.62
55–59	2.402	[−3.655, 8.459]	0.437
60–64	−0.662	[−6.719, 5.395]	0.83
Intercept	62.076	[51.677, 72.476]	0.000

Abbreviation: CI Concentration index, GDP Gross domestic product

We found significantly positive spatial autocorrelation in county-level mammography non-attendance rates (Moran's I range: 0.238–0.392 during 2021–2023, all P < 0.01; see Appendix Table A3). In other words, counties with similar non-attendance rates (whether high or low) tend to cluster geographically, rather than being randomly distributed. The coefficients of the three factors significantly associated with county-level mammography non-attendance rates in the OLS model showed notable regional heterogeneity in the GTWR model (see Fig. 2, also see Appendix Table A4 AICc = −17.53 and Adjusted R² = 0.394 for the improved model fitness and Appendix Figure A3 for the significance of the county-level model). In most counties in Fujian, a higher per capita GDP is associated with a lower mammography non-attendance rate (Coef: −0.114 to 0, P < 0.001), which is consistent with the OLS results (Fig. 2A). In northern counties, an increase in per capita GDP is associated with a higher non-attendance rate, but the effect is weak (Coef: 0 to 0.018, P < 0.001). Urbanization exhibited a consistent negative association in most counties (Coef. −1.24 to 0, P < 0.001) (Fig. 2B), with coefficients in northern counties (Coef. −1.24 to −0.93) being four times stronger than the OLS (Coef. −0.26). In-house mammography capacity reduced non-attendance in most areas (Coef. −42.72 to 0, P < 0.001) (Fig. 2C), but had the opposite effect in southern regions (Coef. 0 to 8.01, P < 0.001).

Fig. 2.

Spatiotemporal distribution of correlation coefficients between per capita GDP (A), urbanization rate (B), in-house mammography capacity (C) and non-attendance mammography rate from GTWR modeling

Discussion

This is the first study on county-level mammography non-attendance following an abnormal breast cancer screening result in China, as well as spatiotemporally specific factors associated with non-attendance. We found that the average county-level mammography non-attendance rate in Fujian decreased from 2021 to 2023. The increase in both the Theil-T and Theil-L indices from 2021 to 2023 indicates a widening of absolute and relative inequalities in non-attendance across counties over time. This suggests that the overall reduction in non-attendance rates was not evenly distributed. The decomposition of the indices reveals that the majority of the inequality stemmed from within municipalities rather than between them, pointing to localized socioeconomic or health system factors as primary drivers of the disparity.

Counties with similar non-attendance rates tended to agglomerate geographically. We also identified per capita of GDP, urbanization rate and in-house mammography capacity as key area-level factors influencing mammography non-attendance rates, with spatially varied patterns of association. The reduction of overall mammography non-attendance in Fujian over the three years studied could be related to improvement under the dynamics of COVID-19 control activities which were rolled back in 2023. In terms of age differences, we found no significant difference between age and mammography non-attendance rates, which suggested that patients in different age groups at the county level performed similarly in terms of participation in screening program. Meanwhile, we observed an association in mammography non-attendance rates between neighboring areas, indicating that the significant disparities at the county level were not entirely related to the pandemic, but were also influenced by socio-economic factors.

Previous studies have found a significant association between low-income areas and high mammography non-attendance rates [5, 29]. In our study, we also observed a negative correlation between GDP per capita and mammography non-attendance rates in a majority of counties, suggesting that economic development may increase participation in mammography. This indicates that poverty is a key barrier to mammography non-attendance, likely reflecting healthcare inequalities caused by factors such as income, education levels, and geographical barriers to healthcare access across counties [30–32]. These results also support previous studies from France [33], Europe [34], and Canada [35].

Contrasting previous findings that urbanization rate negatively affects mammography non-attendance, indicates that higher urbanization rates may be associated with better healthcare accessibility for patients. However, when considering spatial differences, we observed a positive correlation between urbanization rate and mammography non-attendance in the southwestern region. This phenomenon may involve more complex underlying factors, which warrants further investigation in future studies.

Furthermore, previous studies found the relationship between mammography capacity, patient wait times, and mammography screening [11, 36]. In our study, we observed a positive association between in-house mammography capacity and mammography attendance rates. Such lack of capacity under the fragmented relationship between primary screening providers and specialist care is also seen in the primary care sector of China. As China actively explore ways of vertical integration between specialist medical services and primary care, the results suggest the potential of such integrated delivery services to reduce non-attendance by enhancing primary care's capacity of colposcopy (e.g. specialist outreach or seamless referral) particularly in areas with limited resources. In addition, payment structures for screening programs should incorporate process indicators like non-compliance rates [37] and establish patient reminder systems [38] to improve follow-up completeness.

By revealing the geographical heterogeneity in mammography non-attendance, the study enriches the theoretical perspectives of the Anderson model with a more spatially sensitive view. The fact that the same factors exhibited diverse association with non-attendance implies complex interplays between the observed factors and non-observed, area-related factors. Such geographical insights challenge the conventional assumptions that effects of factors were either constant or probabilistic. The study thus opens doors to people-centered, localized investigation into how non-attendance was affected by how the women participating in the scheme interact with local geographical, socio-economic or health system factors.

Several implications may be drawn from our study. First, continuous monitoring and evaluation regarding mammography non-attendance is needed in breast cancer screening programs in Fujian, China and elsewhere. Second, it is important to tailor health resource allocation and policy making to the specific geographical context of local areas, particularly in the northern regions of Fujian, screening efforts should address the likely challenge of low mammography participation by increasing awareness of breast cancer prevention among residents in economically underdeveloped areas. Third, establishing a continuous service process is crucial for cancer screening, especially for primary screening institutions that lack mammography capabilities. Efforts should focus on supporting the improvement of primary healthcare facilities (including in-house mammography capacity), reducing patient waiting times and transportation costs, and ensuring the accessibility and convenience of screening services. Fourth, integrating breast cancer screening data, including mammography follow-up, into existing health information systems (or ensuring interoperability between linked systems) within and across regions could improve the efficiency and accessibility of screening services, facilitate accurate monitoring of mammography non-attendance, and support collaborative care to improve adherence.

This study reveals spatial disparities of non-attendance, offering new insights into systemic gaps in screening continuity, a challenge shared by many LMICs. While China’s government-led screening system demonstrated resilience during public health crises such as COVID‑19, colposcopy non-attendance persists, reflecting entrenched care fragmentation, especially in resource-limited and socioeconomically disadvantaged areas. Mexico’s experience in employing digital follow-up tools, including mobile health units and algorithm-supported proactive nursing, has improved service accessibility and continuity [39]. China could adapt such strategies by introducing mobile screening services and strengthening integrated digital follow-up systems in areas with limited in-house mammography capacity or high population mobility, thereby mitigating service fragmentation and advancing geographic equity.

Limitations

Our study has several limitations. First, due to the lack of a comprehensive information system to trace women with positive screening results, those who traveled to neighboring counties for mammography screenings may not have been recorded. This could result in an overestimation of the mammography non-attendance rate. Second, because of the insufficient detailed information regarding the implementation of screening programs in each county, we probably have missed important factors related to mammography non-attendance during the screening process. Third, constrained by data collected in the administrative datasets of the screening program, the ecological data structure, we could not measure some key individual level predisposing factors (e.g., health beliefs, self-efficacy) in the Andersen framework, caution should also be taken in drawing implications from this ecological study which focusses on county-level variables. Fourth, the absence of data from Xiamen may affect the analysis of spatial heterogeneity in non-attendance, potentially influencing our findings for regions near Xiamen. Finally, our study is restricted to the years when cancer screening programs were disrupted by COVID-19. Therefore, caution should be taken in generalizing our findings to other periods.

Conclusion

Our study provides the first evidence of mammography non-attendance rates in Fujian Province. Although the overall non-attendance rates declined from 2021 to 2023, significant and persistent county-level disparities were observed. Counties with higher rates were concentrated in the southwestern and northern. Lower per capita GDP, lower urbanization rate, and the lack of in-house mammography capacity at primary screening facilities were significantly associated with higher mammography rates, while the associations varied across different regions and time periods. Improving the accessibility and continuity of service delivery, as well as enhancing information tracking, may help reduce mammography non-attendance and regional disparities. These findings highlight the importance of integrating local contextual factors, such as socioeconomic conditions and health resource availability, into both research and policy efforts to better understand and reduce spatial disparities in screening adherence.

Authors’ contributions

Jin Xu, Qihang Zhao, Guanyu Ruan, Xinxin Huang and Pengming Sun conceptualized the study; Qihang Zhao, Jin Xu and Zi Ye reviewed the literature; Guanyu Ruan and Xinxin Huang obtained original data; Qihang Zhao, Zi Ye, and Xinxin Huangwere responsible for data entry and management; Qihang Zhao and Jin Xu developed detailed methodology, analyzed data and drafted the manuscript; Xinxin Huang, Yu Wang and Guanyu Ruan provided critical comments on the analysis framework and drafts of the manuscript; Qihang Zhao and Jin Xu revised the manuscript; Jin Xu, Pengming Sun and Yu Wang supervised and obtained funding for the study. All authors contributed to the data interpretation.

Funding

JX’s participation in this research was supported by grant number 71804004 sponsored by the National Natural Science Foundation of China. The funding sources had no involvements in study design, in the collection, analysis, and interpretation of data, in the writing of the article, or in the decision to submit it for publication. This study was also funded by the Basic Public Health Service Fund of Fujian Province (225,007,008), which contributed to the study design and data collection.

Data availability

The data used to support the findings of this study are available from the corresponding author upon request.

Declarations

Ethics approval and consent to participate

This study was performed in line with the principles of the Declaration of Helsinki. And this study was approved by the Ethics Committee of Fujian Maternity and Child Health Hospital (Reference No. 2025KY004). Informed consent was waived as the research involved the analysis of fully de-identified data with minimal risk to participants.

Consent for publication

Not applicable.

Competing of interests

The authors have no conflicts of interest to declare.