Minimal benefit of co-testing over HPV primary screening with cytology triage from resource-limited settings in China

Abstract

Background

Co-testing with human papillomavirus (HPV) DNA testing plus liquid-based cytology is still used in parts of China, although many screening programmes are moving toward HPV-based strategies. We aimed to compare co-testing with HPV-based and cytology-only approaches in routine county services in resource-limited areas.

Methods

We analysed a screening cohort of 33,387 women aged 35–64 years from four primary care sites. Because all women received both HPV testing and cytology, we reconstructed four strategies within the same population: co-testing, HPV primary screening with cytology triage, HPV-only, and cytology-only. For each strategy we estimated detection of cervical intraepithelial neoplasia grade 2 or worse (CIN2 + ), referrals for specialist examination of the cervix, and cytology workload per 1000 women screened.

Results

Here we show that co-testing detects 6.7 CIN2+ cases per 1000 women screened, compared with 6.5 for HPV primary screening with cytology triage, 4.3 for HPV-only, and 4.9 for cytology-only. However, co-testing requires more resources than HPV primary screening with cytology triage, including 33.1 additional colposcopy referrals and 888.8 extra cytology slides per 1,000 women screened, with little gain in detection. Cytology-only increases referrals while detecting fewer CIN2+ cases, whereas HPV-only reduces referrals but detects fewer CIN2 + .

Conclusions

In resource-limited county programmes, HPV primary screening with cytology triage provides the most favourable balance between detecting cervical pre-cancer and limiting unnecessary procedures. These findings support transitioning from routine co-testing to HPV-based screening tailored to local capacity.

Plain Language Summary

Regular screening can prevent cervical cancers, but programmes must choose which tests to use. In parts of China, women still have both a human papillomavirus (HPV) test and a cervical cell test at the same visit, which increases workload. We studied 33,387 women from four county clinics where all women had both tests, and compared four ways of using these results. Here we show that doing both tests on everyone found almost the same number of important cell changes as using the HPV test first and then a cervical cell test only for HPV-positive women, but needed many more referrals and examinations. Cytology-only increased referrals while finding fewer problems, and HPV-only reduced referrals but missed some changes. In resource-limited areas, HPV-first screening with cytology triage offers the best balance between benefits and workload.

Jia et al. compared the performance of 4 counterfactual screening strategies: co-testing, HPV primary with liquid-based cytology (LBC) triage, HPV-only, and LBC-only. They show that LBC-only detects fewer CIN2+ cases yet still increased referrals while HPV-only reduces referrals but had a lower CIN2+ detection rate.

Introduction

Cervical cancer remains a major global public health challenge, with approximately 661,021 new cases and 348,189 deaths reported in 2022¹. China alone accounts for nearly 22.7% of global incidence and 16% of cervical cancer-related deaths². The burden is highly inequitable, as countries in the lowest Human Development Index (HDI ≤ 0.55) strata experience double the incidence and five-fold higher mortality relative to high-HDI countries^3,4. Decades of programme evaluations showed that organized screening can reduce cervical cancer incidence by up to 80% and mortality by almost 70% compared with no screening^5,6. Meta-analyses further demonstrated that high-risk human papillomavirus (HPV) DNA testing detects cervical intraepithelial neoplasia (CIN)3+ with ≥92–95% sensitivity, whereas routine liquid-based cytology (LBC) rarely exceeds 50–70%^7,8. On the basis of this evidence, the World Health Organization (WHO) updated its guidelines in 2021 to recommend HPV DNA as the preferred primary test⁹. In 2020, WHO launched the “90-70-90” elimination strategy to be achieved by 2030¹⁰. By 2024, at least twelve high-income countries had fully implemented national HPV-primary screening programmes, and around 20 more were gradually transitioning.

Earlier reports indicated that HPV-primary screening detects more precancerous lesions and may allow longer screening intervals, thereby freeing resources for underserved populations¹¹. China’s National Cervical & Breast Cancer Screening Programme for Rural Women (established 2009) initially relied on LBC or visual inspection with acetic acid/Lugol’s iodine (VIA/VILI) for women aged 35–64 years¹². LBC is labour-intensive and relies on experienced cytotechnologists, which are scarce at county and township levels¹³. Although China is progressing toward adoption of HPV DNA as the primary screening test, many regions have implemented a dual screening strategy, performing both HPV DNA testing and LBC for each screened woman.

Here, we show that, in a multicentre real-world cohort of 33,387 women screened in county-level programmes in central and western China, co-testing provides no meaningful gain in CIN2+ detection compared with HPV primary screening with cytology triage, but requires substantially greater cytology and colposcopy resources. Co-testing yields 6.7 CIN2+ detections per 1000 women screened versus 6.5 with HPV primary screening with cytology triage, while requiring 33.1 additional colposcopy referrals and 888.8 extra cytology slides per 1000 women. Overall, HPV primary screening with cytology triage provides the most favourable balance between detection and resource burden in health-resource-limited county programmes.

Methods

Study design and setting

We selected four county or city level sites participating in the National Cervical Cancer Screening Programme in China, located in the central and western regions, to reflect real-world implementation in primary care settings (Fig. 1). These county-level programmes primarily serve rural catchment populations in health-resource-limited areas, where cervical cancer screening coverage before programme initiation had been modest and often irregular. In 2023, these sites were equipped with HPV DNA testing platforms (Dalton hybrid-capture HPV assay; Hangzhou Dalton Biosciences), which has been approved by the National Medical Products Administration (NMPA) for primary screening. LBC slides were interpreted by local cytopathologists or provincial reference cytopathologists. Women were referred for colposcopy if they tested positive for HPV16/18 or had LBC results of Atypical Squamous Cells of Undetermined Significance (ASC-US) or worse. The study protocol was approved by the Institutional Review Board of the Chinese Academy of Medical Sciences & Peking Union Medical College (CAMS&PUMC-IEC-2022-059-1), and all participants provided written informed consent.

Fig. 1. Study sites, real-world screening flow, and site-year summary.

The base map of China was obtained from the Standard Map Service of the Ministry of Natural Resources of China (http://bzdt.ch.mnr.gov.cn/); overlays and layout were created by the authors. LBC liquid-based cytology, ASC-US atypical squamous cells of undetermined significance, hrHPV high-risk human papillomavirus, other-12 hrHPV types other than 16/18, PPV positive predictive value, NNR number needed to refer. Samples were co-collected at the same visit and tested subsequently.

HPV DNA testing

DH2 HPV DNA Detection Kit (Hangzhou Dalton BioSciences, China) is a hybrid capturebased assay designed for the detection of 14 high-risk HPV types (HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68), without providing genotyping information. DH2 positive samples were further classified for HPV16/18 using DH1 (Hangzhou Dalton BioSciences, China), which is designed for the detection of HPV16 and HPV18.

The testing procedure involves three main steps: denaturation of double-stranded DNA (dsDNA) into single-stranded DNA (ssDNA), hybridization of RNA probes with ssDNA to form RNA: DNA hybrids, and subsequent capture of these hybrids by antibodies immobilized on a microplate. Chemiluminescence is used to detect and measure the hybrids. For the determination of positive and negative results in clinical samples, if the ratio of the sample’s relative light unites (RLU) to the cut-off (CO) value (1 pg/mL HPV DNA) is ≥1.0, the result is considered positive; if the ratio of RLU/CO value is <1.0, the result is considered negative.

Study population and eligibility criteria

Inclusion criteria was women with an intact cervix who were permanent residents of the study counties. Exclusion criteria were: (i) a history of cervical cancer; (ii) prior total hysterectomy; (iii) current pregnancy or breastfeeding; (iv) acute pelvic inflammatory disease or other conditions precluding pelvic examination; (v) inability to comprehend or sign the informed consent form; and (vi) an unsatisfactory LBC specimen. In total, 33,387 women with valid screening results were included in the final analysis.

Data collection and outcomes

All women in this screening cohort received both HPV testing (12-type panel with HPV16/18 genotyping) and LBC. Using the same underlying individuals, we reconstructed four screening strategies deterministically from the observed test results:

1. Co-testing (HPV + LBC): referral to colposcopy if either HPV16/18 is positive or LBC ≥ ASC-US.

2. HPV primary with LBC triage (12 + 2): colposcopy for HPV16/18 positive; other 12-type positive undergo LBC triage and are referred if LBC ≥ ASC-US.

3. HPV-only (12 + 2): colposcopy for HPV16/18 positive; other 12-type positive receive a one-year recheck; LBC is not used for referral.

4. LBC-only: referral if LBC ≥ ASC-US.

For each strategy, we calculated the following outcomes: number screened, number referred to colposcopy, number completing colposcopy, number biopsied, histopathologic diagnoses among biopsied women (normal/benign, CIN1, CIN2, CIN3, carcinoma), CIN2+ and CIN3+ detection rates per 1000 screened, and the positive predictive value (PPV) for CIN2+ among those who completed colposcopy. LBC workload was expressed as the number of LBC slides per 1000 screened (1 slide per woman when LBC is used by the strategy; HPV-only contributes 0). The number needed to refer (NNR) was defined as referred to colposcopy divided by CIN2+ detected. Unless otherwise stated, detection rates used the entire screened population as the denominator, and histopathologic proportions used the biopsied population as the denominator.

Statistics and reproducibility

Baseline characteristics were summarized as mean ± SD for continuous variables and n (%) for categorical variables. Baseline comparability between groups was assessed using standardised mean differences (SMDs) to quantify the magnitude of differences (in units of the pooled standard deviation for continuous variables, or indicator contrasts for categorical variables), which are not sensitive to sample size¹⁴. Following methodological recommendations for observational studies, absolute SMDs <0.10 were considered to indicate negligible imbalance, whereas values ≥ 0.10 were interpreted as potentially meaningful imbalance^14,15.

To quantify the effects of strategies, we first calculated, for each strategy, outcome proportions per 1000 women screened (e.g. CIN2+ detection, referrals to colposcopy, and cytology workload). Using HPV primary screening (12 + 2) with LBC triage as the pre-specified policy reference strategy, we then computed absolute differences as:

$$\mathrm{\Delta }\text{CIN}2+/1000=1000\times \left(p_s-p_{\mathrm{ref}}\right)$$ 1

$$\mathrm{\Delta }\text{referrals}/1000=1000\times \left(r_s-r_{\mathrm{ref}}\right)$$ 2

$$\mathrm{\Delta }\mathrm{slides}/10=0.1\times \left({\mathrm{slides}}_s-{\mathrm{slides}}_{\mathrm{ref}}\right),$$ 3

where $p_s$ and $r_s$ are, respectively, the CIN2+ detection proportion and referral proportion under strategy s, with the screened population as denominator, and $p_{\mathrm{ref}}$, $r_{\mathrm{ref}}$ and ${\mathrm{slides}}_{\mathrm{ref}}$ denote the corresponding outcome measures under the reference strategy (HPV primary + LBC triage). For differences in proportions (e.g. ΔCIN2 + , Δreferrals), 95% confidence intervals (CIs) were computed using the large-sample Wald approximation to the difference of two independent proportions, with the screened population as the denominator under each strategy. Histopathology yields among biopsied women (normal/benign or CIN1, CIN2, CIN3 + /cancer) were summarised using counts and percentages only; no additional hypothesis testing was performed for these conditional distributions because the biopsied subsets are determined by the strategy-specific referral rules and are not independent samples across strategies.

All analyses were pre-specified and two-sided with α = 0.05. Data management and analyses were conducted in R (version 4.4) using packages tidyverse, janitor, gtsummary, gt, and ggplot2 (and ggh4x for figure facets).

Results

Participants and baseline characteristics

We analyzed 33,387 women who were simultaneously eligible for all four counterfactual strategies (co-testing, LBC-only, HPV-only, and HPV primary with LBC triage). Across measured domains the strategy-determined referral subsets were similar to each other and to the overall cohort. Standardized mean differences (SMDs) were small (all ≤0.18-for example, for site the largest SMD versus co-testing was 0.18), details in Table 1.

Table 1.

Baseline characteristics of women referred (positive) under each screening strategy and overall, with standardized mean differences (SMD) versus co-testing

Variable	Overall	Co-testing referred	LBC-only referred	SMD (Co vs LBC)	HPV-only referred	SMD (Co vs HPV)	HPV + LBC triage referred	SMD (Co vs HPV + LBC triage)
Year				0.035		0.114		0.063
2023	16,639 (49.84%)	1380 (53.89%)	1054 (52.13%)		468 (59.54%)		830 (57.01%)
2024	16,748 (50.16%)	1181 (46.11%)	968 (47.87%)		318 (40.46%)		626 (42.99%)
Unknown	0
Site				0.066		0.176		0.083
Mangshi	5527 (16.55%)	374 (14.60%)	294 (14.54%)		111 (14.12%)		231 (15.87%)
Xinpin	11,260 (33.73%)	747 (29.17%)	647 (32.00%)		173 (22.01%)		378 (25.96%)
Yanting	4644 (13.91%)	442 (17.26%)	319 (15.78%)		162 (20.61%)		275 (18.89%)
Zezhou	11,956 (35.81%)	998 (38.97%)	762 (37.69%)		340 (43.26%)		572 (39.29%)
Unknown	0
Marital status				0.012		0.041		0.046
Unmarried	316 (0.98%)	27 (1.09%)	22 (1.12%)		10 (1.32%)		16 (1.15%)
Married	31,516 (97.43%)	2393 (96.65%)	1891 (96.58%)		731 (96.69%)		1348 (96.63%)
Divorced	205 (0.63%)	20 (0.81%)	15 (0.77%)		6 (0.79%)		11 (0.79%)
Widowed	311 (0.96%)	36 (1.45%)	30 (1.53%)		9 (1.19%)		20 (1.43%)
Unknown	1039	85	64		30		61
Ethnicity				0.037		0.108		0.055
Han	23,516 (72.70%)	1861 (75.16%)	1440 (74.10%)		602 (79.63%)		1067 (76.50%)
Other	8832 (27.30%)	615 (24.84%)	518 (25.90%)		154 (20.37%)		328 (23.50%)
Unknown	1039	85	64		30		61
Religion				0.009		0.051		0.041
Yes	438 (1.36%)	24 (0.98%)	18 (0.93%)		11 (1.47%)		15 (1.08%)
No	31,656 (98.64%)	2433 (99.02%)	1925 (99.07%)		739 (98.53%)		1370 (98.92%)
Unknown	1293	104	79		36		71
Smoking				0.015		0.040		0.049
Yes	193 (0.60%)	19 (0.77%)	13 (0.66%)		8 (1.06%)		13 (0.93%)
No	32,155 (99.40%)	2457 (99.23%)	1945 (99.34%)		748 (98.94%)		1382 (99.07%)
Unknown	1039	85	64		30		61
Alcohol use				0.009		0.035		0.053
Yes	1940 (6.00%)	147 (5.94%)	116 (5.93%)		49 (6.48%)		92 (6.60%)
No	30,408 (94.00%)	2329 (94.06%)	1842 (94.07%)		707 (93.52%)		1303 (93.40%)
Unknown	1039	85	64		30		61
Education				0.015		0.045		0.057
≤Middle school	28,285 (88.65%)	2263 (92.29%)	1793 (92.33%)		687 (92.21%)		1264 (91.79%)
≥High school	3623 (11.35%)	189 (7.71%)	149 (7.67%)		58 (7.79%)		113 (8.21%)
Unknown	1479	109	80		41		79
Occupation				0.037		0.104		0.110
Unemployed/low-labour	9908 (30.63%)	798 (32.23%)	598 (30.55%)		280 (37.04%)		510 (36.56%)
Manual labour	18,552 (57.35%)	1474 (59.53%)	1198 (61.23%)		415 (54.89%)		761 (54.53%)
Non-manual/professional	3888 (12.02%)	204 (8.24%)	162 (8.29%)		61 (8.07%)		124 (8.90%)
Unknown	1039	85	64		30		61
Contraception				0.031		0.113		0.088
None	11,230 (44.31%)	891 (46.99%)	713 (46.81%)		263 (48.35%)		505 (49.08%)
Short-acting	3,412 (13.46%)	206 (10.86%)	168 (11.02%)		63 (11.58%)		117 (11.37%)
Long-acting	10,678 (42.14%)	799 (42.14%)	643 (42.18%)		218 (40.07%)		407 (39.56%)
Other/unspecified	22 (0.09%)	0 (0.00%)	0 (0.00%)		0 (0.00%)		0 (0.00%)
Unknown	8045	665	498		242		427
Family history of cancer				0.016		0.063		0.055
Yes	325 (1.01%)	23 (0.93%)	18 (0.92%)		5 (0.67%)		12 (0.87%)
No	31,910 (98.99%)	2443 (99.07%)	1935 (99.08%)		743 (99.33%)		1374 (99.13%)
Unknown	1152	95	69		38		70
Menopause				0.01		0.03		0.05
Postmenopausal	13,564 (41.93%)	1203 (48.59%)	953 (48.67%)		364 (48.15%)		676 (48.46%)
Premenopausal	18,784 (58.07%)	1273 (51.41%)	1005 (51.33%)		392 (51.85%)		719 (51.54%)
Unknown	1039	85	64		30		61
Age (years)				<0.01		<0.01		0.01
Mean ± SD	50.00 ± 8.00	51.87 ± 7.75	52.00 ± 8.00		52.00 ± 8.00		52.00 ± 8.00
Unknown	77	7	5		3		5
BMI (kg/m²)						0.01		0.03
Mean ± SD	24.20 ± 3.30	24.11 ± 3.22	24.10 ± 3.20		24.10 ± 3.20		24.00 ± 3.20
Unknown	1039	85	64		30		61
Household size				0.01		<0.01		0.03
Mean ± SD	4.12 ± 1.47	4.05 ± 1.47	4.04 ± 1.46		4.05 ± 1.48		4.00 ± 1.51
Unknown	1644	120	94		42		84
Age at first birth (years)				0.02		0.09		0.05
Mean ± SD	22.90 ± 3.00	22.58 ± 2.87	22.63 ± 2.96		22.34 ± 2.49		22.46 ± 2.73
Unknown	3,172	215	154		83		143

Values are n (%) for categorical variables and mean ± SD for continuous variables. Percentages are column-wise within strategy. “Referred” denotes women meeting that strategy’s referral criteria (co-testing: HPV16/18 positive or LBC ≥ ASC-US; HPV-only: HPV16/18 positive; LBC-only: LBC ≥ ASC-US; HPV primary + LBC triage: HPV16/18 positive or HPV other 12-type positive with LBC ≥ ASC-US). “Unknown” is displayed where available and is excluded from percentage denominators and from standardized mean difference (SMD) calculations. SMDs compare each group with co-testing: for continuous variables, SMD = |μ₁–μ₂|/SD_pooled; for categorical variables, SMDs are computed from indicator contrasts, with the largest absolute SMD across levels reported. Absolute SMDs ≥0.10 were interpreted as potentially meaningful imbalance; hypothesis tests were not performed for baseline comparisons. Minor discrepancies may occur due to rounding.

LBC liquid-based cytology, HPV human papillomavirus.

Clinical performance and resource use (per strategy)

Using observed data and applying each decision rule to the same population, co-testing referred 2561 women to colposcopy (7.67%), LBC-only referred 2022 (6.06%), HPV-only (12 + 2) referred 786 (2.35%), and HPV primary with LBC triage referred 1,456 (4.36%). Among those referred, 2082, 1641, 656, and 1213 completed colposcopies, respectively. 1436, 1057, 584, and 1039 underwent biopsy, respectively-details in Table 2.

Table 2.

Clinical performance of different cervical cancer screening strategies

Indicator	Co-testing	HPV only	LBC only	HPV primary + LBC triage
Screened (n)	33,387	33,387	33,387	33,387
Referred to colposcopy (n, %)	2561 (7.67%)	786 (2.35%)	2022 (6.06%)	1456 (4.36%)
Completed colposcopy (n, %)	2082 (81.30%)	656 (83.46%)	1641 (81.16%)	1213 (83.31%)
Biopsied (n, %)	1436 (68.97%)	584 (89.02%)	1057 (64.41%)	1039 (85.66%)
CIN1 detected (n, % of biopsied)	350 (24.37%)	142 (24.32%)	257 (24.31%)	278 (26.76%)
CIN2 detected (n, % of biopsied)	117 (8.15%)	65 (11.13%)	85 (8.04%)	114 (10.97%)
CIN3 + /cancer detected (n, % of biopsied)	106 (7.38%)	77 (13.18%)	77 (7.28%)	104 (10.01%)
CIN2+ detection rate (% of screened)	0.67	0.43	0.49	0.65
PPV for CIN2+ (% among completed colposcopy)	10.71	21.65	9.87	17.97
Number needed to refer (NNR)	11.48	5.54	12.48	6.68
Number needed to complete colposcopy (NNC)	9.34	4.62	10.13	5.56
Δ CIN2+ vs HPV + LBC triage (per 1000)	0.15	-2.28	-1.68	0
Δ referrals vs HPV + LBC triage (per 1000)	33.1	-20.07	16.95	0
Δ cytology slides vs HPV + LBC triage (per 1000)	888.76	-111.24	888.76	0

Counts are n; percentages are column-wise unless otherwise specified. “Referred to colposcopy” is among all screened. “Completed colposcopy (n, %)” is the number and percentage of women with a completed colposcopy examination among those referred. “Biopsied (n, %)” is the number and percentage of women who received a biopsy among those with completed colposcopy. histologic yields (CIN1, CIN2, CIN3 + /cancer) are shown as n and % among biopsied. CIN2+ detection rate uses the screened population as the denominator. PPV for CIN2+ uses completed colposcopy as the denominator. NNR (number needed to refer) is the number of referrals per CIN2+ detected (referrals/CIN2 + ), calculated on the screened denominator; NNC (number needed to complete colposcopy) is completed colposcopies per CIN2+ detected. Δ rows show absolute differences per 1000 screened compared with the HPV-primary + LBC triage strategy (positive values indicate more events than the triage strategy, negative values fewer). LBC slide counts are the number of LBC slides processed per 1000 screened.

CIN cervical intraepithelial neoplasia, PPV positive predictive value, NNR number needed to refer, NNC number needed to complete colposcopy, LBC liquid-based cytology.

CIN2+ detection rates per 1000 screened were 6.7 for co-testing, 4.9 for LBC-only, 4.3 for HPV-only, and 6.5 for HPV primary with LBC triage. CIN3+ detection was 3.2, 2.3, 2.3, and 3.1 per 1000, respectively. The positive predictive value (PPV) for CIN2+ among women completing colposcopy was highest for HPV-only (21.65%) and triage (17.97%), and lowest for co-testing (10.71%) and LBC-only (9.87%). Correspondingly, the NNR (colposcopies per CIN2+ detected) was lowest for HPV-only (5.54) and triage (6.68), and highest for LBC-only (12.48) and co-testing (11.48) (Table 2). The overall resource yield trade-off is visualised in Fig. 2.

Fig. 2. Resource–benefit trade-off across cervical cancer screening strategies.

Each point represents one screening strategy; the x-axis shows colposcopy referrals per 1000 screened, and the y-axis shows CIN2+ detected per 1000 screened. Bubble area encodes LBC slides per 1000 screened (larger bubbles indicate greater cytology workload). The crosshair marks the reference strategy (HPV primary + LBC triage). The dashed grey segment illustrates the Pareto frontier between HPV only and the reference strategy (fewer referrals for less CIN2+ yield). HPV only (smallest bubble; lower detection with fewer referrals), HPV primary + LBC triage (reference; crosshair), LBC only (intermediate bubble; higher referral burden with lower detection than triage), and Co-testing (HPV + LBC) (largest bubble; highest referral burden with similar CIN2+ yield to triage). Rates are calculated with the screened population as the denominator; no error bars are displayed because the plot summarizes observed operating points rather than interval estimates. Colour and bubble-size keys are shown at right. CIN2 +  cervical intraepithelial neoplasia grade 2 or worse, LBC liquid-based cytology, NNR number needed to refer.

Histopathology among biopsied women

Among biopsied women, normal/benign or CIN1 comprised 60.10% and 24.37% under co-testing, 60.36% and 24.31% under LBC-only, 51.37% and 24.32% under HPV-only, and 52.26% and 26.76% under triage. CIN2 and CIN3 proportions were higher for HPV-only (11.13% and 11.47%) and triage (10.97% and 9.05%) than for co-testing (8.15% and 6.62%) and LBC-only (8.04% and 6.81%). Across strategies, the proportion of “false-positive” endpoints (≤CIN1 among biopsied) was highest for co-testing and LBC-only (84.47% and 84.67%), and lower for triage (79.02%) and HPV-only (75.68%). Completion of colposcopy among referred were 81.3%, 81.16%, 83.46%, and 83.31% across the four strategies-details in Table 3.

Table 3.

Histopathological outcomes and detection rates by screening strategy

Indicator	Co-testing	HPV only	LBC only	HPV primary + LBC triage
Pathological outcomes (among biopsied women)
Normal / benign, n (%)	863 (60.10%)	300 (51.37%)	638 (60.36%)	543 (52.26%)
CIN1, n (%)	350 (24.37%)	142 (24.32%)	257 (24.31%)	278 (26.76%)
CIN2, n (%)	117 (8.15%)	65 (11.13%)	85 (8.04%)	114 (10.97%)
CIN3, n (%)	95 (6.62%)	67 (11.47%)	72 (6.81%)	94 (9.05%)
Carcinoma, n (%)	11 (0.77%)	10 (1.71%)	5 (0.47%)	10 (0.96%)
Total biopsied, n	1436	584	1057	1039
Detection rates in the total screened population
CIN2+ detection rate (% of screened)	0.67	0.43	0.49	0.65
CIN3+ detection rate (% of screened)	0.32	0.23	0.23	0.31
≤CIN1 histology (n, % of biopsied)	1213 (84.47%)	442 (75.68%)	895 (84.67%)	821 (79.02%)
Colposcopies per CIN2+ detected (NNC)	9.34	4.62	10.13	5.56
Follow-up completion among referred (%)	81.3	83.46	81.16	83.31

Pathological outcomes are shown as n with column-wise percentages among biopsied women; “Carcinoma” includes invasive carcinoma. “≤CIN1 histology” counts women with biopsy results of normal/benign or CIN1 following referral. CIN2+ and CIN3+ detection rates use the screened population as the denominator. NNC is completed colposcopies per CIN2+ detected (lower values indicate greater efficiency). Follow-up completion among referred is completed colposcopy divided by number referred to colposcopy. LBC positivity was defined as ASC-US or worse.

CIN cervical intraepithelial neoplasia, LBC liquid-based cytology, NNC number needed to complete colposcopy.

Incremental comparisons versus HPV primary + LBC triage (12 + 2)

Using the triage strategy as the policy reference, co-testing produced no marginal gain in CIN2+ detection (+0.15 per 1000; 95% CI −1.08 to 1.38) but required 33.1 additional colposcopy referrals per 1000 (95% CI 29.50–36.70) and 888.76 additional LBC slides per 1000 (95% CI 885.39–892.13). LBC-only detected 1.68 fewer CIN2+ per 1,000 (95% CI −2.82 to −0.54) and still referred 16.95 more women per 1000 (95% CI 13.58 to 20.32) than triage. HPV-only (16/18 direct referral) detected 2.28 fewer CIN2+ per 1000 (95% CI −3.39 to −1.17) but required 20.07 fewer referrals per 1000 (95% CI −22.80 to −17.34) than triage and used no LBC (Table 4). These absolute differences (with 95% CIs) are also displayed in a centred difference plot (Fig. 3). The relative efficiency mirrored these differences: ΔNNR versus triage was +4.83 for co-testing (worse), +5.85 for LBC-only (worse), and −1.14 for HPV-only (better).

Table 4.

Incremental effects vs HPV primary + cytology triage (12 + 2)

	Δ Co-testing	95% CI	Δ LBC-only	95% CI	Δ HPV-only (12 + 2)	95% CI	Note
CIN2+ detection (/1000)	0.15	–1.08, 1.38	–1.68	–2.82, –0.54	–2.28	–3.39, –1.17	per 1000
CIN3 + /cancer (/1000)	0.06	–0.79, 0.91	–0.81	–1.60, –0.02	-0.81	–1.60, –0.02	per 1000
Colposcopy referrals (/1000)	33.1	29.50, 36.70	16.95	13.58, 20.32	−20.07	−22.80, −17.34	per 1000
Cytology slides (/1000)	888.76	885.39, 892.13	888.76	885.39, 892.13	0		per 1000 (0 for HPV-only)
ΔNNR	4.83	4.19, 5.63	5.85	4.66, 7.28	−1.14	−1.71, −0.60	Lower is better

Δ values are absolute differences (strategy - HPV-primary + LBC triage). Unless stated, units are per 1000 screened; for LBC slides, the unit is per 1000 slides processed (0 by design for HPV-only). ΔNNR is the difference in the number needed to refer (referrals/CIN2 + ); lower is better. Negative Δ values indicate fewer events than the triage strategy; positive Δ values indicate more.

CIN cervical intraepithelial neoplasia, CI confidence interval, NNR number needed to refer, LBC liquid-based cytology, HPV human papillomavirus.

Fig. 3. Incremental differences vs HPV primary + LBC triage.

From top to bottom, panels display absolute differences (Δ = strategy - reference) in (A) CIN2+ detected per 1000 screened, B LBC slides per 10 screened, and C colposcopy referrals per 1000 screened, all relative to HPV primary screening with LBC triage. Points represent the point estimate (measure of centre) of the absolute difference for each strategy, and horizontal bars are 95% Wald confidence intervals. Numeric labels give the point estimates (two decimals); “(ns)” indicates intervals that include 0 (i.e., not statistically significant at α = 0.05). Positive values indicate a higher count than the HPV primary + LBC triage strategy; negative values indicate fewer. For the LBC panel, the HPV-only strategy adds 0 LBC slides and may be omitted or plotted at 0. Δ absolute difference, LBC liquid-based cytology, CI confidence interval.

Discussion

In this multicentre real‑world study of 33,387 women screened in resource‑limited counties, the four competing strategies yielded strikingly different balances between high-grade lesion detection and resource use. Compared with HPV primary screening with LBC triage (the reference strategy), co‑testing only marginally increased CIN2+ detection (+0.15/1000) while requiring 33.1 extra colposcopies and about 889 additional LBC slides per 1000 screened. LBC‑only missed more CIN2+ cases (−1.68/1000) and still generated excess referrals (+16.95/1000), whereas HPV (12 + 2) genotyping alone reduced referrals by 20.07/1000 and eliminated LBC but at the cost of 2.28 fewer CIN2+ detections. These findings suggest that HPV primary with LBC triage offers the most balanced trade‑off between efficacy and resource burden.

The limited incremental benefit of co‑testing over HPV primary with LBC triage reflects the biological and operational realities of cervical carcinogenesis. High‑risk HPV DNA testing detects oncogenic infection earlier and more consistently than LBC¹⁶. Adding LBC alongside HPV (co‑testing) increases the number of women with HPV‑negative but LBC‑positive results, which have low predictive value and result in many unnecessary colposcopies. In our cohort, co‑testing produced the highest false‑positive proportion (≥84% of biopsied lesions were ≤CIN1) and the lowest PPV (10.71%). Conversely, HPV primary with LBC triage focused resources on HPV‑positive women and achieved higher PPV (17.97%) and lower NNR (6.68). The HPV-only (12 + 2) strategy-immediate referral for HPV16/18 and annual re‑testing for other 12 types-reduced referrals but inevitably delayed detection of some high-grade lesions emerging from other high-risk types, explaining its slightly lower CIN2+ yield.

Our results align closely with high-quality trials and comparative cohorts showing that HPV-based screening yields superior benefit–harm profiles versus LBC or co-testing. In pooled long-term analyses of European randomized trials, HPV-based screening reduced invasive cervical cancer compared with LBC¹⁷; in the Canadian HPV FOCAL RCT, primary HPV screening yielded a significantly lower 48-month CIN3+ risk than liquid-based LBC, demonstrating a more durable low-risk interval after a negative HPV test¹⁸. Population-based data from China similarly support HPV-based strategies, as a nationwide programme study reported the effectiveness of HPV testing at scale¹⁹, and a prospective cohort directly comparing primary HPV, co-testing, and LBC found that HPV-primary achieved favourable clinical performance versus co-testing and clearly outperformed LBC-only²⁰. A recent synthesis further concludes that any incremental detection gained by co-testing is small and comes at the cost of substantially more downstream procedures, yielding an unfavourable benefit-to-harm ratio relative to primary HPV testing, which is also consistent with our incremental analyses²¹.

Contemporary guidelines point in the same direction. The WHO 2021 guideline recommends HPV DNA testing as the preferred primary test within either screen-and-treat or screen-triage-treat pathways²². This indicates that when HPV testing is used, 5-year intervals are appropriate for average-risk populations. Partial genotyping with 16/18 direct referral and triage of other high-risk types is an acceptable approach. The American Cancer Society (2020) likewise prefers primary HPV testing every 5 years from age 25 to 65²³. In the U.S. transitional context, the USPSTF (2018) gives A-grade recommendations to primary hrHPV (q5y), co-testing (q5y), or LBC (q3y) for ages 30–65²⁴, and professional advisories (e.g., ACOG 2021) mirror these three options²⁴. Taken together, the guideline direction of travel favours HPV-primary algorithms, with co-testing viewed largely as a transitional compromise rather than a strategy with superior clinical value.

These findings inform national initiatives to eliminate cervical cancer in low‑resource settings. Adopting primary HPV screening with LBC triage (12 + 2) can sustain high detection while substantially reducing colposcopy procedures and LBC workload. Our data indicate that, per 10,000 women screened, replacing co‑testing with HPV primary screening plus LBC triage would prevent approximately 331 unnecessary referrals and avoid preparation of roughly 8,887 LBC slides without any meaningful loss in CIN2+ detection. LBC‑only strategies miss a significant proportion of high‑grade disease and are therefore suboptimal. An HPV‑only strategy (direct referral for HPV16/18) may be considered where LBC services are unavailable. However, policymakers should acknowledge its lower immediate detection and plan robust follow‑up to capture lesions in subsequent rounds. Implementation programmes should integrate HPV testing with efficient recall systems, interoperable data platforms, and self‑sampling options to ensure high coverage and adherence to follow‑up.

In county-level programmes constrained by budgets and workforce, the objective is to maximise CIN2+ yield while minimising laboratory and referral workload. Recent economic evaluations consistently favour HPV-based pathways on value. A general-population modelling analysis in Nature Medicine reported that primary HPV strategies are the most effective and cost-effective, and that adding triage (including genotyping and cytology) preserves effectiveness while reducing unnecessary downstream procedures, which lowers programme costs²⁵. Complementary regional modelling shows similar patterns in vaccinated and unvaccinated settings, with primary HPV with partial genotyping lying on or near the efficiency frontier when compared with cytology-based screening and co-testing²⁶. By contrast, co-testing yields minimal additional detection but escalates downstream resource use; contemporary comparative analyses show an unfavourable benefit-to-harm profile versus primary HPV, implying higher referrals and added costs without commensurate gains²¹. Middle-income country evaluations reach the same conclusion: in Vietnam, a 2025 BMJ Open cost-effectiveness study found co-testing not cost-effective under prevailing prices and capacity²⁷. Taken together, these data indicate that as HPV assay prices decline, cytology throughput and quality assurance become the binding bottlenecks, whereas primary HPV with selective triage concentrates resources on HPV-positive women and improves operational metrics such as PPV and number needed to refer^21,25.

Strengths of this study include: (1) a large, real‑world sample spanning multiple county‑level sites; (2) use of a single cohort in which four screening strategies were compared by applying distinct decision rules to the same individuals, thereby minimising between‑group confounding; and (3) reporting of outcomes as absolute differences per 1000 women, along with PPV and NNR (number needed to refer), metrics that are policy‑relevant and not materially influenced by the low verification fraction. By applying four pre-defined screening algorithms to the same co-tested cohort across multiple county-level programmes, this study provides a multi-centre, within-cohort, real-world comparison that complements previous trial and modelling evidence on the limited incremental value of co-testing and quantifies resource use with policy-relevant metrics. Several limitations warrant consideration. First, partial verification (only women referred for colposcopy/biopsy received full assessment) precluded calculation of true negative predictive values. Nevertheless, current WHO guidance emphasises PPV and resource efficiency metrics under incomplete verification. Second, the quality of LBC and colposcopy may have varied across sites. Third, follow‑up data for women scheduled for one‑year re‑testing in the HPV‑only strategy are not yet complete. Some apparently “missed” CIN2+ cases may be detected in subsequent screening rounds. Fourth, all four study sites were county- or city-level programmes in central and western China that primarily serve rural women in health-resource-limited settings. As such, our findings may not fully generalise to highly screened urban populations or settings with abundant cytology and colposcopy capacity. The relative performance of co-testing versus primary HPV screening in those high-resource environments may therefore warrant separate evaluation.

Conclusion

In resource‑limited county programmes, HPV primary screening (12 + 2) with LBC triage offers the most favourable balance between CIN2+ detection and resource use, whereas co-testing provided minimal additional detection while substantially increasing cytology and colposcopy workload. HPV-only screening may be a pragmatic alternative where cytology cannot be delivered, albeit with some loss of immediate detection that can be mitigated by subsequent screening rounds. In this multi-centre, real-world analysis from county-level programmes in health-resource-limited areas of China, applying four screening algorithms to the same screened cohort provides empirical confirmation that co-testing adds little incremental benefit over HPV-based strategies. Policies should therefore pivot away from routine co-testing towards HPV-based screening tailored to local capacity, alongside investment in follow-up infrastructure to achieve WHO’s elimination targets.

Author contributions

Study conception, design, and overall coordination: X.J., C.G., A.N., F.Z., Y.Q. Site-level data collection and curation: X.D., J.S., R.D., T.Z., Z.L., Y.Z., Y.W., C.T., S.Q., Y.Y. Data processing and analysis: X.J. Manuscript drafting and coordination: X.J. Critical revision of the manuscript: C.G., F.Z. Data access and verification: X.J., Y.Q. Study supervision and corresponding author: Y.Q.

Peer review

Peer review information

Communications Medicine thanks Christine Bergeron and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Data availability

Source data underlying the main figures are provided in the Supplementary Data file. Additional de-identified data underlying the analyses are available from the corresponding author upon reasonable request. Source data for Figs. 1–3 are available in Supplementary Data.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s43856-026-01467-z.