Trends in over- and under-screening for cervical cancer after EMR implementation in rural China

Abstract

Inappropriate cervical cancer screening practices, including over- and under-screening, pose significant healthcare burdens in low-resource settings. This study analyzed screening behaviors and determinants among 33,362 women aged 35–64 in Wuxiang County, China, using longitudinal cohort data. Screening events were classified as guideline-adherent, over-screened, under-screened, or unscreened based on prior methods (HPV, cytology, or co-testing) and results, using cause-specific frailty models for analysis. Overall, only 19.9% of events were guideline-adherent, while 29.5% were over-screened and 50.6% were under- or unscreened. Notably, the implementation of a county-wide Electronic Medical Record (EMR) platform in 2022 coincided with a sharp decline in over-screening from 36.7% to 15.7%. Compared with primary HPV testing, prior co-testing increased the hazard of both over- and under-screening, whereas prior cytology was strongly associated with under-screening. Women with low-grade abnormalities (≤CIN1) showed a substantially higher risk of under-screening compared to those with negative results. Additionally, community residents were more prone to over-screening, while village residents faced higher under-screening risks. These findings suggest that transitioning to HPV-based screening and integrating EMR systems effectively reduces unnecessary testing, though enhanced reminder systems are crucial to address persistent under-screening in resource-constrained regions.

Introduction

Cervical cancer remains one of the leading causes of cancer-related deaths among women globally. In 2022, over 662,000 new cases and 349,000 deaths were reported globally, underscoring the substantial threat that cervical cancer poses to women’s health¹. Implementation of cervical cancer screening programs has been widely recognized as a critical intervention for reducing cervical cancer incidence and mortality^2,3. In China, the latest Chinese cervical cancer screening guidelines recommend human papillomavirus (HPV) based testing or co-testing (HPV combines cytology) every five years, or cytology alone screening every three years if with the primary normal screening results (negative for intraepithelial lesion or malignancy (NILM) for cytology, or negative for HPV), for patients with cervical intraepithelial neoplasia grade 1 or less (≤CIN1) one year after diagnosis^4–6. The updated evidence-based guidelines aim to maximize the effectiveness of disease detection and cancer prevention while minimizing the over-treatment of lesions that may regress spontaneously⁷.

With the increasing accessibility of routine preventive screening services, the utilization of low-value preventive services has become increasingly prevalent^8,9. The term “low-value services” primarily refers to over-screening, which manifests as: screening frequency exceeding recommended intervals, misusage of inappropriate screening methods, wastage of screening resources, and screening practices that confer neither clear benefits nor significant harm to patients (yielding only marginal benefits or potential risks) (e.g., screening may cause minor harm without substantially improving survival rates)¹⁰. Over-screening not only elevates healthcare costs but also imposes additional burdens or potential risks on patients¹¹. For instance, premature screening (i.e., screening conducted earlier than recommended intervals) may lead to unnecessary colposcopies and incur additional expenditures due to the detection of transient HPV infections that typically resolve spontaneously, while also causing undue psychological distress^12–16. Evidence from the United States highlights the widespread nature of over-screening. A study of 8000 women aged over 30 found low adherence to cervical cancer screening guidelines: only 34% reported screening as recommended, while 45% were over-screened¹⁷. Further research is needed to understand the factors driving this behavior, including clinical decision-making and patient preferences, to inform interventions that improve guideline adherence.

Inappropriate screening practices in primary care also include under-screening, which is often defined as the failure to conduct timely tests when needed, may result in misdiagnosis, diagnostic delays, deferred effective treatment, ultimately contributing to poor patient prognosis, increased healthcare expenditures, and elevated risks of late-stage cancer diagnosis^7,18,19. Nationwide data from a U.S. study indicated that 18–35% of commercially insured women receive screenings at frequencies below guideline-recommended standards²⁰. Similarly, a study conducted in Mexico revealed that up to 45% of rural women screen at frequencies below recommended guidelines²¹. Although the Chinese government provides a free cervical cancer screening program for women aged 35–64, program coverage remains limited due to the absence of a unified recall system and insufficient clinician awareness, particularly in remote, low-resource counties where lifetime ever-screening coverage reaches merely 48.2%²². This under-screening dilemma coexists with over-screening in low-resource settings: on one hand, over-screening diverts resources away from underserved populations, exacerbating health disparities^23,24; on the other hand, many women experience excessively prolonged screening intervals or delayed follow-up after abnormal results, leading to increased risks of late-stage diagnosis and missed opportunities for early intervention^15,25–27.

Existing studies on screening guideline adherence were mainly cross-sectional and survival studies and were dominated by high-resource countries; there remains a limited understanding of the full spectrum of guideline non-adherence in low-resource regions, especially undergoing transitions in screening modalities^16,28–34. This gap is particularly critical in China, where national cervical cancer screening programs (NCCSP) have increased lifetime coverage to 51.5% yet face implementation challenges in remote and underserved areas²². In addition, current evidence rarely evaluates screening behaviors in the context of individualized follow-up recommendations based on prior screening methods and results. To address this gap, our study examined cervical cancer screening follow-up behaviors in a dynamic, longitudinal cohort from Wuxiang County, Shanxi Province, between 2018 and 2024. Wuxiang County serves as a representative example of a low-resource area. Located in the hinterland of the Taihang Mountains, the county faces a scarcity of high-quality medical resources common to many low-resource regions. The local healthcare system relies heavily on a limited number of county-level facilities to serve a widely dispersed rural population. Additionally, geographical barriers and limited health literacy among residents have exacerbated difficulties in ensuring timely follow-up and case management. These systemic constraints make Wuxiang County an ideal model for evaluating the effectiveness of digital health interventions and optimized screening strategies in resource-constrained environments. Specifically, we investigate the prevalence and predictors of over-screening, under-screening, and guideline adherence screening in relation to prior screening method- and result-specific follow-up recommendations. By leveraging individual-level screening histories and applying frailty models, we aim to provide empirical evidence to inform more efficient and equitable cervical cancer screening practices in low-resource settings.

Results

Demographic characteristics of the study cohort

A total of 57,406 cervical cancer screening events from women aged 21–70 years between 2018 and 2024 were assessed (Fig. 1). After excluding events due to invalid IDs (n = 1436), repeated records (n = 122), participation in clinical trials (n = 1665), and age ≥ 65 years (n = 2209), a total of 51,792 times screening events were included in the combined cohort. The combined cohort was categorized into three groups based on their screening-related risk status: routine screening (46,287 events, 89.1%), Low-risk screening (5505 events,10.6%), and high-risk screening (182 events, 0.4%). Routine screening was more common in younger women, while low-risk classification increased with age. Most screenings occurred in village areas (86.5%), which also had a higher proportion of low-risk screenings than communities (11.3% vs. 6.5%, Table 1). As the high-risk cohort requiring more intensive surveillance (n = 182), and events right-censored beyond the observational period (n = 18,430), 33,362 screening events were ultimately included for frailty analysis.

Fig. 1. Flow diagram of cohort selection.

The flowchart illustrates the step-by-step inclusion and exclusion criteria applied to the initial 57,406 cervical cancer screening events assessed between 2018 and 2024. Ultimately, a final cohort of 33,362 screening events was included for the frailty analysis.

Table 1.

Demographic characteristics of the study cohort

	Routine screeninga		Low-riskb		High-riskc		Combined cohort
	n	%	n	%	n	%	n	%
Total	46,287	89.1%	5505	10.6%	182	0.4%	51,974
Age, years
25–34	383	94.8%	21	5.2%	0	0.0%	404	0.8%
35–39	4227	92.4%	342	7.5%	8	0.1%	4577	8.8%
40–44	7082	91.2%	666	8.6%	35	0.2%	7783	15.0%
45–49	9556	90.7%	959	9.1%	38	0.2%	10,553	20.3%
50–54	10,447	89.1%	1260	10.7%	49	0.2%	11,756	22.6%
55–59	8743	87.5%	1241	12.4%	26	0.1%	10,010	19.3%
60–64	5849	85.1%	1016	14.8%	26	0.2%	6891	13.3%
Residential area
Village	39,765	88.6%	5049	11.3%	150	0.2%	44,877	86.5%
Community	6522	93.3%	456	6.5%	32	0.2%	6993	13.5%

^aRoutine screening cohort included those who did not meet either high-risk or low-risk criteria.

^bThe Low-Risk cohort included women with prior abnormal screening results but ≤CIN1 diagnosis, who may need to be rescreened 1 year after rescreen to manage for cancer progression.

^cThe High-Risk cohort included women with HIV infection, solid organ transplantation, or those with colposcopy or prior abnormal screening results indicating CIN2 + , as well as those with a history of CIN2+ precancerous lesions, cervical cancer, or prior cervical cancer treatment.

Patterns of screening behavior by age, history, and EMR implementation

The distribution of screening behaviors demonstrated clear age-related patterns, with the age group 45-49 years emerging as a turning point between over- and under-screening patterns. In younger women (≤34 years), under- and unscreened behavior predominated, accounting for over 80% of screening events, while guideline-adherent follow-up was rare. As age increased, over-screening rose steadily, reaching its peak in the 45–49 group (approximately 34%), coinciding with a relative stabilization of under-screening and unscreened behaviors. Beyond age 50, over-screening began to decline, while the proportion of under- and unscreened follow-up rose again, particularly in women aged 60–64 (Fig. 2). This inflection pattern suggests a midlife convergence of proactive screening behavior, followed by disengagement or delay in later years.

Fig. 2. Age-specific trends in cervical cancer screening behavior types.

The line chart illustrates the proportion of screening events (%) across different age groups. The green, red, and yellow lines represent guideline-adherent screening, overscreening, and under- and unscreening behaviors, respectively.

These modality- and result-specific distributions reflect subsequent screening behaviors conditional on the year, modality, and result of the preceding screening event, assessed within the available follow-up after that prior screen (Table 2). Over-screening was most frequent following negative results from co-testing (88.6%) and primary HPV testing (75.6%), suggesting a tendency toward early re-testing despite guideline recommendations for extended intervals (Fig. 3). Conversely, follow-up after ≤CIN1 results was dominated by under- and unscreened behavior, especially after cytology (96.2%) and primary HPV testing (85.3%), consistent with recommendations for one-year surveillance in this population. Guideline-adherent follow-up was most commonly observed in ≤CIN1 groups, although overall adherence remained suboptimal across modalities. These findings indicate that both age and preceding screening context interact to influence follow-up behaviors, with significant deviation from recommended intervals at both ends of the age spectrum. Besides that, as shown in Fig. 4, after the EMR-enabled recall system was launched in 2022, screening behavior shifted markedly. The share of over-screening dropped from 36.7% in 2021 to 15.7% in 2022 and stayed low in 2023–2024 (11.1%–14.9%). Unscreened episodes first peaked at 50.4% in 2022 but then declined to 35.5% by 2024, indicating fewer completely missed contacts. Meanwhile, under-screening rose from 9.3% in 2021 to 17.0% in 2023 and 32.7% in 2024, suggesting that many women who would previously have remained unscreened were ultimately recalled, albeit later than recommended. Overall, the system curtailed unnecessary early tests while bringing a substantial proportion of overdue women back into the screening continuum.

Table 2.

Distribution of screening adherence categories based on previous screening modality and outcome (2018–2024)

Prior screened year	Prior screening method	Prior screening resulta	Guideline adherence screen no. (%)	Over-screen no. (%)	Under- and unscreen no. (%)	Total
2018	Co-testing	Negative	45 (5.1)	581 (65.2)	265 (29.7)	891
		≤CIN1	468 (53.8)	2 (0.2)	400 (46.0)	870
	Cytology	Negative	600 (12.6)	3504 (73.4)	669 (14.0)	4773
		≤CIN1	23 (67.6)	2 (5.9)	9 (26.5)	34
	TOTAL		1136 (17.3)	4089 (62.3)	1343 (20.4)	6568
2019	Primary HPV Test	Negative	62 (9.1)	472 (69.1)	149 (21.8)	683
		≤CIN1	3 (6.8)	0 (0.0)	41 (93.2)	44
	Co-testing	Negative	1 (7.7)	8 (61.5)	4 (30.8)	13
		≤CIN1	37 (9.2)	4 (1.0)	361 (89.8)	402
	Cytology	Negative	2586 (25.3)	2718 (26.5)	4935 (48.2)	10,239
		≤CIN1	13 (3.2)	20 (4.9)	373 (91.9)	406
	TOTAL		2702 (22.9)	3222 (27.3)	5863 (49.7)	11787
2020	Co-testing	Negative	21 (1.6)	1312 (98.4)	0 (0.0)	1333
		≤CIN1	49 (10.4)	0 (0.0)	424 (89.6)	473
	Cytology	Negative	1662 (23.6)	67 (1.0)	5320 (75.5)	7049
		≤CIN1	0 (0.0)	0 (0.0)	434 (100.0)	434
	TOTAL		1732 (18.6)	1379 (14.8)	6178 (66.5)	9289
2021	Primary HPV Test	Negative	0 (0.0)	56 (100.0)	0 (0.0)	56
		≤CIN1	0 (0.0)	0 (0.0)	24 (100.0)	24
	Co-testing	Negative	0 (0.0)	701 (100.0)	0 (0.0)	701
		≤CIN1	2 (0.5)	0 (0.0)	427 (99.5)	429
	Cytology	Negative	963 (33.8)	55 (1.9)	1832 (64.3)	2850
		≤CIN1	0 (0.0)	0 (0.0)	608 (100.0)	608
	TOTAL		965 (20.7)	812 (17.4)	2891 (61.9)	4668
2022	Primary HPV Test	Negative	0 (0.0)	18 (100.0)	0 (0.0)	18
		≤CIN1	24 (36.9)	0 (0.0)	41 (63.1)	65
	Co-testing	Negative	0 (0.0)	1 (100.0)	0 (0.0)	1
		≤CIN1	5 (62.5)	0 (0.0)	3 (37.5)	8
	Cytology	Negative	0 (0.0)	212 (100.0)	0 (0.0)	212
		≤CIN1	8 (3.0)	0 (0.0)	255 (97.0)	263
	TOTAL		37 (6.5)	231 (40.7)	299 (52.7)	567
2023	Primary HPV Test	Negative	0 (0.0)	106 (100.0)	0 (0.0)	106
		≤CIN1	47 (12.7)	0 (0.0)	323 (87.3)	370
	Cytology	Negative	0 (0.0)	3 (100.0)	0 (0.0)	3
		≤CIN1	0 (0.0)	0 (0.0)	4 (100.0)	4
	TOTAL		47 (9.7)	109 (22.6)	327 (67.7)	483
TOTAL			6619 (19.8)	9842 (29.5)	16,901 (50.7)	33,362

Percentages are computed within each row (stratum of Prior Screening Method × Result × Year).

^aPrior screening result definitions (applied to HPV-primary, cytology-primary, and co-testing encounters).

Negative include: ①HPV primary: high-risk HPV DNA negative (hr-HPV); ②Cytology primary: smear negative for intra-epithelial lesion or malignancy (NILM); ③Co-testing: NILM cytology and negative for hr-HPV in the same encounter.

≤CIN 1 (low-grade/indeterminate) defined any screening episode not classed as Negative in which no subsequent biopsy confirms CIN 2 or worse, i.e., the most severe finding is low-grade, including: ①hr-HPV-positive (any genotype), with or without cytological abnormality; ②cytology ASC-US, LSIL, ASC-H, HSIL or AGC whose follow-up biopsy is ≤CIN 1 or not performed; ③histopathology CIN 1.

For co-testing, if either component (hr-HPV or cytology) or any follow-up biopsy lies within this low-grade spectrum and no biopsy shows CIN 2 + , the encounter is coded ≤CIN 1.

Fig. 3. The distribution of screening behavior by screening modality and result (2018–2024).

The stacked bar chart illustrates the proportion of screening events (%) stratified by prior screening modality (Primary HPV DNA, Co-testing, or Cytology) and the corresponding test results (Negative or ≤CIN1). The colors indicate the specific screening behavior types: teal represents guideline-adherent screening, red represents overscreening, light yellow represents underscreening, and dark yellow represents unscreening within the observational period.

Fig. 4. Annual proportions of guideline-adherent, over-, under-, and unscreened cervical cancer episodes before (2021) and after (2022-2024) implementation of the electronic health-record recall system.

The stacked bar chart displays the yearly percentage of repeat-screening episodes for each behavior category, with dashed lines indicating the trends over time. The colors represent specific screening behaviors: blue indicates guideline-adherent screening, red indicates overscreening, light yellow indicates underscreening, and orange indicates unscreening within the observational period.

To account for individual-level heterogeneity in repeated screening events, we applied cause-specific frailty models with a gamma-distributed random effect for patient ID. The analysis incorporated each participant’s prior screening method, result, and year as time-varying covariates in a dynamic cohort framework. Descriptive statistics for the distribution of index screening year and age by event status are provided in Supplementary Table 3. Several predictors were significantly associated with over-screening and under- or unscreening (Table 3). Compared with prior primary HPV testing, prior co-testing was associated with a markedly increased hazard of over-screening (adjusted hazard ratio [aHR], 2.09; 95% confidence interval [CI], 1.89-2.32; P < 0.001), and a higher hazard of under- and unscreening (aHR, 3.52; 95% CI, 2.97–4.16; P < 0.001). Prior cytology was also negative but not significantly associated with over-screening (aHR, 0.97; 95% CI, 0.88–1.07; P = 0.57) and was strongly associated with both under- and unscreening (aHR, 19.91; 95% CI, 17.14–23.12; P < 0.001). Women with previous abnormal results (≤CIN1) had significantly lower hazards of over-screening (aHR, 0.06; 95% CI, 0.04–0.09; P < 0.001), but sharply elevated hazards of both under- and unscreening (aHR, 24.81; 95% CI, 22.48–27.38; P < 0.001).

Table 3.

Multivariable analysis of over-screening-associated risk factors

Characteristic	Over-screening			Under-screening/unscreened
	aHR (95% CI)	P-value	Event/N	aHR (95% CI)	P-value	Event/N
Pre-screening method
Primary HPV Testing	Reference		652/714	Reference		149/273
Co-testing	2.09 (1.89–2.32)	<0.001	2603/2670	3.52 (2.97–4.16)	<0.001	269/403
Cytology	0.97 (0.88–1.07)	0.57	6559/12,370	19.91 (17.14–23.12)	<0.001	12,756/24,378
Pre-screen result
Negative	Reference		9814/15,754a	Reference		13,174/25,054
≤CIN 1	0.06 (0.04–0.09)	<0.001	22/707a	24.81 (22.48–27.38)	<0.001	3727/5085
Pre-screening yeara	0.46 (0.44–0.47)	<0.001	—	0.96 (0.93–0.99)	0.007	—
Residence area
Village	Reference		7016/12,853	Reference		15,444/27,118
Community	1.77 (1.68–1.87)	<0.001	2826/3608	0.60 (0.55–0.66)	<0.001	1457/3021
Age (per +1 yr)	0.98 (0.977–0.983)	<0.001	—	0.99 (0.989–0.995)	<0.001	—

Screening type	n	Events	Frailty σ²	Concordance	LR χ² (df)	P
Over-screening	16,461	9842	0.21	0.789	8131 (1,741)	<2 × 10⁻¹⁶
Under-screening/unscreened	23,520	16,901	1.18	0.959	29,301 (10,246)	<2 × 10⁻¹⁶

^aResult-category definitions (applied to HPV-primary, cytology-primary, and co-testing encounters)

Negative include: ①HPV primary: high-risk HPV DNA negative (hr-HPV); ②Cytology primary: smear negative for intra-epithelial lesion or malignancy (NILM); ③Co-testing: NILM cytology and negative for hr-HPV in the same encounter.

≤CIN 1 (low-grade/indeterminate) The ≤CIN1 category included screening episodes initiated by an abnormal screening or triage result (e.g., hr-HPV positivity, abnormal cytology, or abnormal VIA/VILI or colposcopic impression, where applicable), in which no cervical intraepithelial neoplasia grade 2 or higher (CIN2 + ) was identified throughout the episode-level assessment.

Multivariable analysis of factors associated with screening adherence

Calendar time, residential setting, and age were each independently associated with screening-interval behaviors. Relative to 2018, every additional calendar year was linked to a pronounced reduction in the hazard of over-screening (aHR 0.46, 95% CI 0.44–0.47; P < 0.001) and a modest decline in under-screening (aHR 0.96, 95% CI 0.93–0.99; P = 0.007). Women living in community areas, compared with those in villages, were more likely to undergo over-screening (aHR 1.77, 95% CI 1.68–1.87; P < 0.001) but less likely to experience under-screening (aHR 0.60, 95% CI 0.55–0.66; P < 0.001). Increasing age exerted a small protective effect: each additional year of age was associated with lower hazards of both over-screening (aHR 0.98, 95% CI 0.977–0.983; P < 0.001) and under-screening (aHR 0.99, 95% CI 0.989–0.995; P < 0.001). Frailty variance (σ²) values indicated meaningful individual-level heterogeneity across all screening behavior types, with σ² = 0.21 for over-screening and σ² = 1.18 for under- and unscreening. Model concordance was high across outcomes (0.789 for over-screening and 0.959 for under-/unscreening), and both models demonstrated excellent fit as indicated by large likelihood ratio test statistics (P < 0.001 for all outcomes). Sensitivity analyses separating under-screening from completely unscreened outcomes yielded directionally consistent associations and are presented in Supplementary Tables 2.

Discussion

This is the first study to comprehensively evaluate adherence to guideline-recommended cervical cancer screening intervals in a Chinese low-resource county. Among 33,362 analyzable follow-up events, only one in five (19.8%, 9842/33,362) occurred within the recommended window, while over half (50.7%, 16,901/33,362) were either delayed or completely missed. In China, the government provides free cervical cancer screening services for all women aged 35–64³⁵. However, despite these publicly funded initiatives, screening coverage remains suboptimal. Recent data indicated that between 2023 and 2024, 51.5% of women aged 35 to 64 years had received at least one cervical cancer screening, falling short of the WHO’s interim target of 70% coverage by 2030^22,36,37. Additionally, given that cytology remains the predominant screening method in rural China, the 3-year screening coverage in these regions is only 43.1% (90% CI, 40.7%-45.4%), highlighting the need to transform from cytology-based to HPV-based primary screening.

Multivariable analysis revealed that reliance on cytology alone for primary cervical cancer screening was significantly associated with over-screening in primary screening, but delayed or missed follow-up^23,27. This phenomenon primarily stems from two factors: incomplete implementation of updated screening guidelines in primary care, which leads to inconsistent adherence and unnecessary repeat testing, and the exclusion of HPV testing from mainland China’s public screening programs, requiring out-of-pocket payment. This financial barrier, coupled with limited public awareness of recommended screening intervals, results in self-initiated or physician-recommended repeat testing, particularly among health-conscious individuals seeking additional protection despite lacking clear guidance on appropriate screening frequency^10,38.

Notably, within our interval-based outcome definition, women with prior low-grade findings (≤CIN1) exhibited markedly higher rates of under- or unscreening, despite the established risk of progression associated with persistent low-grade abnormalities^39,40. This gap is concerning, as prior studies have shown that women with abnormal screening results often fail to receive guideline-concordant management, elevating their risk of disease progression and delayed diagnosis^15,41,42. The observed under-/unscreening pattern among women with low-grade abnormalities may be shaped by multiple barriers, such as limited health literacy, financial constraints, stigma related to pre-cancerous lesions, and absence of clear clinical guidance or reminders^7,43. Our findings showed that screening episodes following cytology were associated with a substantially higher likelihood of delayed or missed follow-up screening (P < 0.05), thereby compromising the timely identification of disease progression. This limitation is particularly pronounced in resource-constrained settings, where subjective interpretation and shortages of trained personnel lead to elevated rates of false-negative results and misclassification, ultimately undermining the screening efficacy for high-risk populations^44–46.

Paradoxically, women whose previous screening results were negative were more likely to undergo unnecessary repeat screening. Two factors may underlie this pattern: (1) clinicians’ cautions leading them to order additional tests out of an abundance of caution even after a negative result¹⁰; and (2) patients’ distrust of a negative finding, or persistent anxiety about cervical-cancer risk, which prompts them to request screening earlier or more often than guidelines advise⁴⁷; and institutional arrangements using repeat confirmation of negative results as a quality control measure⁴⁸. Mathias et al. reported that 65.7% of low-risk women received unnecessary early Pap tests, resulting in exceeding colposcopy examinations, highlighting the chain of medical resource waste caused by over-screening⁴⁸. Community-dwelling women showed higher over-screening risks. Better accessibility made this population more likely to undergo screening on schedule or even prematurely, while women in remote rural areas faced under-screening risks due to transportation barriers and limited health knowledge^49,50. Nevertheless, the overall over-screening rate showed a declining trend during the observation period, suggesting that educational interventions are gradually improving practice. It is essential to further raise health awareness among this population to better monitor the progression of their condition, thereby reducing incidence and mortality rates.

In 2022, Wuxiang County touched the base of cervical cancer screening eligible women, a total of 38,312 women were identified as cervical cancer screening eligible. Then, to achieve the WHO’s goal of 70% screening coverage, a WeChat-based EMR system was applied in clinical practice. By closing information gaps across all healthcare service tiers, the system shifts cervical cancer prevention from fragmented, paper-based workflows to a continuous, data-driven loop of screening, diagnosis, treatment, and surveillance, markedly improving timeliness, care quality, and resource efficiency within the studied context, suggesting a scalable model for similar settings. In Wuxiang County, before full implementation of the EMR system (2018–2021), over-screening rates peaked at 62.3%, consistent with previously reported rates of 60% and 65.8% in U.S. populations with commercial insurance^12,23,27.

Among all women eligible for routine screening, we identified a significantly higher proportion of over-screening in younger women, with a declining trend as age increased. After accounting for individual differences, the over-screening rate reached 35.2% in the 45–49 age group, gradually decreasing thereafter, which aligns with observations from international multi-center databases^7,10,23,27. Furthermore, our study identified that under-screening was particularly pronounced among women aged 55–59 and 60–64, where more than half of follow-up visits did not occur as advised (53.5% and 65.1%, respectively). Since the HPV vaccine hasn’t been approved for women over 45 in mainland China, getting high-quality cervical cancer screening routinely is the only way for elderly women to prevent cervical cancer. What’s worse, the decreased immunity, immature epithelial metaplasia, and changes in the vaginal micro-environment among this population could contribute to the increasing risk of HR-HPV infection. These older women have repeatedly been identified as bearing higher cervical cancer risk and poorer outcomes, and delayed re-screening in these groups likely contributes to late diagnosis and increased mortality^51,52. And in our cohort, village residential women faced significantly higher under-screening rates compared to the community. This likely reflects limited health awareness of the population, limited HPV testing availability, and weak recall systems in underserved regions⁵³. Given these challenges, more research is needed to identify key barriers preventing timely screening and to assess the broader implications of under-screening for both individual patients and the healthcare system. Additionally, effective health intervention strategies should be explored to enhance follow-up rates and screening adherence, such as digital health education and recall systems, ensuring that cervical cancer screening programs fulfill the intended public health impact.

For future recommendations: at the system level, an EMR-based hierarchical result management and automated reminder system needs to be established to optimize screening intervals, which could limit repeated testing within safe intervals for individuals with negative results, while ensuring timely re-screen reminders for abnormal biopsy results of CIN1 or less. At the physician level, there is a need for strengthened evidence-based education to update knowledge on HPV based primary screening strategies and improve adherence to guideline-recommended practices. At the patient level, strengthen the management of different screening results and education on screening intervals, with targeted outreach to rural women who face access barriers and information gaps^7,38,54. Encouragingly, our study observed a temporal decline in over-screening incidence. This reduction was temporally associated with Wuxiang County’s 2022 EMR pilot screening program. This program established interval thresholds based on screening methods and results, automatically intercepting duplicate tests and significantly reducing over-screening. Similar EMR-based reminder interventions have been shown to reduce unnecessary screening in external studies^10,48,55. However, under-screening issues remain unchanged, consistent with findings from the U.S. national surveys of commercially insured populations⁷. To address the dual challenge of “excessive-insufficient” screening, the proposed system-level interventions (e.g., automated reminders and interval management) can reduce resource waste, while patient- and provider-targeted education can improve follow-up compliance, especially among high-risk groups, thereby maximizing the public health impact of cervical cancer screening.

Strengths and limitations

First, to our knowledge, this is the first study in mainland China to characterize follow-up behavior across the era of extended screening intervals, thereby filling a major evidence gap. Also, this is the first study to systematically document the widespread prevalence of both over- and under-screening in an empirically defined low-resource rural area. By leveraging a near-population-based longitudinal cohort from Wuxiang County—a region representative of systemic resource constraints—we provide critical empirical evidence of the “excessive-insufficient” screening paradox. This shifts the focus beyond mere coverage statistics to the urgent need for addressing screening inefficiencies in under-resourced contexts. In settings where medical capacity is scarce, the diversion of resources toward over-screening directly competes with the need to reach under-screened populations, making these findings vital for optimizing future public health strategies. Second, we analyzed all 2018-2024 screening encounters from the only three hospitals in a low-resource county, thus providing a near-population-based longitudinal cohort. Third, the dataset is event-level and time-stamped, allowing accurate classification of over-, under-, and guideline-adherent screens, which strengthens internal validity. Also, the introduction of EMR not only enhanced the guideline adherence (the over-screening rate declined from 36% in 2021 to 15% in 2022) but also gave the opportunity of precision case management to bridge the information in cervical cancer prevention and treatment phases.

Several limitations should be noted. First, the observation period captured only two complete HPV-based screening intervals (2018–2023 & 2019–2024), limiting the statistical power to assess long-term adherence trends for this specific modality. Second, data were derived from a single rural county without individual-level sociodemographic covariates (e.g., income, education), which limits generalizability; additionally, opportunistic tests may have been missed due to the retrospective nature of data collection. Third, as adherence outcomes are defined relative to guideline-recommended intervals, associations with prior screening modalities reflect programmatic structures rather than causal effects, and estimates from the cohort’s inception phase (2018–2020) should be interpreted with caution due to the interval-based definition. Finally, while the decline in over-screening was temporally associated with the EMR launch, the lack of a concurrent control group precludes definitive causal inference, and future longitudinal studies are needed to confirm that reduced screening frequency does not compromise clinical safety (e.g., CIN2+ detection).

As concluded, future work should, firstly, extend follow-up to cover multiple full HPV cycles and the forthcoming scale-up of self-sampling. Secondly, implement EMR-driven automated reminder systems that flag women overdue for screening, especially those with preceding CIN1 or less abnormal results, who require timely surveillance. Furthermore, future efforts should prioritize establishing county-to-city linked case-management platforms to monitor screening trajectories across healthcare settings. A key component would be developing unified digital health applications that consolidate individuals’ healthcare records by syncing health insurance data and hospital registration systems. Such integration would enable real-time access to comprehensive health profiles, facilitating targeted interventions for under-screened rural populations and improving follow-up tracking in urban areas. By leveraging demographic variables and centralized data, these platforms could dynamically identify gaps, such as missed screenings in remote regions or fragmented urban follow-ups, and prompt tailored outreach.

Methods

Study design and population

Data were retrospectively obtained from the NCCSP database from all three hospitals to cover the entire cervical cancer screening in Wuxiang County, Shanxi Province, China. Screening records between January 1st, 2018, and December 31st, 2024, were consequently included. Data were anonymized before analysis, and the study was approved as non-human subjects research by the Ethics Committee of the Chinese Academy of Medical Sciences, Peking Union Medical College Hospital (No. I-23PJ427). Due to the retrospective nature of data collection from existing medical records, this study was granted a waiver of written informed consent.

Eligible participants were women aged 25–64 years with local household registration and at least one cervical cancer screening recorded during the study period. The high-risk cohort was excluded from the model analysis, whose definition included: (1) a history of hysterectomy, (2) gynecological malignancy, (3) immunosuppressive conditions (e.g., HIV infection, leukemia, or organ transplantation), and (4) prior biopsy results indicating CIN2+ (diagnostic codes detailed in Supplementary Table 1). The average risk cohort consisted of women with negative results from previous cervical screening tests, no history of cervical cancer, and no immunocompromising conditions. The low-risk cohort comprised women with screening abnormal results indicating histologic CIN1 or less, preceded by a low-grade test result (defined as HR-HPV-positive or Cytology ≥ASC-US), consistent with the 2019 American Society for Colposcopy and Cervical Pathology (ASCCP) Risk-Based Management Consensus Guidelines recommending one-year surveillance for this population⁴. Each cervical screening event was analyzed independently to dynamically evaluate adherence to guideline-recommended follow-up intervals. Participants were followed from their initial screening until the end of the study period or censoring due to diagnosis of CIN2 or worse, to distinguish therapeutic management from routine screening behavior assessment.

Introduction of the electronic medical records (EMR) system

In January 2022, an Electronic Medical Record (EMR) system was developed and implemented county-wide via a WeChat mini-program and a web portal for both healthcare providers and cervical cancer screening-eligible women. This EMR system was designed to streamline the entire screening process by linking the women’s interface, primary screening sites, and an upper-tier “green-channel” referral network. On the user side, women complete real-name registration with optical character recognition, book appointments, and receive time-stamped notifications and tailored educational content. Also, primary screening and follow-up colposcopy or biopsy results are delivered instantly on the same page, transforming the traditional “return-to-clinic” model into an on-demand digital service. At primary screening sites, the system provides end-to-end electronic workflows, which include eligibility verification, sample tracking, AI-assisted colposcopic image interpretation, automated quality-control flags, and batch recall lists. For high-risk cases (e.g., HPV16/18-positive or High-grade Squamous Intraepithelial Lesion (HSIL) cytology), the system auto-generates standardized electronic referral orders and uploads imaging in real time to regional hubs, enabling same-day confirmation and remote consultation; surgical completion (LEEP or conization) is likewise tracked to closure.

Screening behavior classification

Screening methods included cytology alone, HPV DNA testing alone, and co-testing. Cytology and HPV DNA tests performed within 45 days of each other were defined as co-testing; tests occurring more than 45 days apart were treated as separate episodes. According to the latest Chinese cervical cancer screening guidelines, recommended intervals were 3 years for cytology, 5 years for HPV testing or co-testing, and 1 year for individuals with ≤CIN1 lesions⁴. A grace window of ±6 months was allowed around the recommended interval to account for real-world scheduling variability (e.g., appointment delays, holidays, and follow-up coordination) and to avoid misclassifying marginal deviations as non-adherent^12,23. This grace window was applied uniformly across all screening modalities (cytology, HPV testing, and co-testing) and result categories (negative and ≤CIN1) to ensure consistent classification.

Each screening episode initiated a new risk interval, with follow-up starting at the date of the index screening. Screening behaviors were classified into three mutually exclusive categories: over-screening, defined as a subsequent screening occurring before the lower bound of the recommended interval; guideline-adherent screening, defined as screening occurring within the recommended interval (including the grace window); and under- or unscreening defined as reaching the upper bound of the recommended interval without a subsequent screening (Table 4), for under- or unscreening, the event time was assigned to the upper bound of the recommended interval rather than the date of a later screening, if any. Screenings occurring after the upper bound were treated as the start of a new screening interval and did not retroactively alter the classification of the prior interval. Screenings expected to occur after December 31, 2024, were administratively censored.

Table 4.

Operational definition of over-screening, guideline-adherent screening, and under-/un-screening based on the recommended follow-up interval

Prior screening modality	Prior screening result	Guideline-recommended interval	Classification of subsequent screening behavior
HPV testing	Negative	5 years	Over-screening: <4.5 years
			Guideline-adherent screening: 4.5–5.5 years
			Under-screening or no rescreening: >5.5 years or no rescreening within the observation period
Cytology (LCT)	Negative	3 years	Over-screening: <2.5 years
			Guideline-adherent screening: 2.5–3.5 years
			Under-screening or no rescreening: >3.5 years or no rescreening within the observation period
Co-testing (HPV + cytology)	Negative	5 years	Over-screening: <4.5 years
			Guideline-adherent screening: 4.5–5.5 years
			Under-screening or no rescreening: >5.5 years or no rescreening within the observation period
HPV /Cytology/Co-testing	≤CIN1	1 year	Over-screening: <0.5 years
			Guideline-adherent screening: 0.5–1.5 years
			Under-screening or no rescreening: >1.5 years or no rescreening within the observation period

Notes:

Definition of negative screening results

Negative screening results were defined as hr-HPV negativity, cytology negative for intraepithelial lesion or malignancy (NILM), or co-testing showing both NILM cytology and hr-HPV negativity at the same clinical visit.

Definition of ≤ CIN1 (abnormal screening episodes without high-grade disease)

The ≤CIN1 category included screening episodes initiated by an abnormal screening or triage result (e.g., hr-HPV positivity, abnormal cytology, or abnormal VIA/VILI or colposcopic impression, where applicable), in which no cervical intraepithelial neoplasia grade 2 or higher (CIN2 + ) was identified throughout the episode-level assessment.

This category corresponds to surveillance after abnormal screening results without high-grade disease, consistent with risk-based management frameworks such as the ASCCP guidelines.

When a biopsy was not performed, episodes were classified as ≤CIN1 if an abnormal screening or triage finding prompted follow-up or surveillance, and no subsequent record within the same episode indicated CIN2 + .

Hierarchical classification rule: When multiple screening, triage, or diagnostic results were available within a single episode, classification was based on the most severe finding, with CIN2+ overriding all other results.

Data analysis

Screening adherence was evaluated iteratively using a rolling classification method for each screening event. The time scale was defined as gap time, measured from the date of the prior screening to the date of the subsequent screening or to the boundary of the guideline-recommended interval, as applicable. Each event was classified based on the preceding screening’s age, screening method (cytology, HPV DNA testing, or co-testing), and screening result (negative or ≤CIN1). After each screening event, individuals re-entered the risk set for the next screening interval, allowing for recurrent events within the same individual. We stratified the analysis by the calendar year of screening to investigate temporal trends in screening behaviors.

We applied a cause-specific frailty model for recurrent events to assess factors associated with screening behavior under guideline-defined intervals. Within this framework, guideline-adherent screenings were treated as the reference outcome and contributed person-time until the occurrence of a non-adherent event (over-screening or under-/unscreening), at which point the corresponding cause-specific event was recorded. Screening outcomes were modeled as a function of demographic and clinical characteristics at the time of the repeated screening, including age, residential area (village vs. community), prior screening method (cytology, HPV DNA testing, or co-testing), and prior result (negative vs. ≤CIN1). We fitted three separate cause-specific frailty models to evaluate the relative likelihood of: (1) guideline-adherent screening versus over-screening; and (2) under-screening versus guideline-adherent screening. Proportional hazards assumptions were assessed using Schoenfeld residuals and visual inspection of log-log survival plots, with no major violations observed. Given that some outcomes were defined by crossing the upper or lower bounds of guideline-recommended intervals, alternative modeling strategies, such as discrete-time survival models or multi-state approaches, were considered. However, the cause-specific frailty model was selected to accommodate recurrent events, irregular screening intervals, and individual-level heterogeneity within the available longitudinal data. All statistical analyses were conducted using R version 4.3.3, with two-sided P < 0.05 considered statistically significant.

Author contributions

H.D. and Y.Q. conceived the study. H.D., Y.Z., and Y.Q. designed the work. H.D., Y.Z., H.W., B.Z., M.C., J.H., and X.S. prepared the data. Y.Z. and H.W. drafted the original manuscript. H.D. and Y.Q. contributed significantly to reviewing and editing the manuscript. All authors read, reviewed, and approved the final manuscript. Y.Z. and H.W. contributed equally to this manuscript.

Data availability

The datasets generated and/or analyzed during the current study are not publicly available due to data use agreements with the NCCSP Database and the Tencent Sustainable Social Value Inclusive Health Lab. Still, they are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-026-02504-w.