Real-world Impact of HPV Vaccination on CIN2+ Reduction: A Meta-analysis of Observational Studies Assessing Variation by Age, Coverage, and Vaccine Type

Abstract

Background

Human papillomavirus (HPV) vaccination reduces the incidence of high-grade cervical lesions, but its population-level impact on cervical intraepithelial neoplasia grade 2 or worse (CIN2+) may vary by age at vaccination, program coverage, and vaccine type. The extent to which these factors explain variation in CIN2+ outcomes across real-world vaccination programs remains incompletely quantified.

Methods

Using data from a prior systematic review of observational studies across HPV-related endpoints, we restricted this meta-analysis to studies reporting histologically confirmed CIN2+. Reported ratio measures were analyzed on the natural log scale using random-effects models fitted by restricted maximum likelihood with Hartung–Knapp–Sidik–Jonkman inference. Prespecified subgroup analyses examined age at vaccination, vaccination coverage, and vaccine type. Exploratory univariable meta-regressions assessed follow-up duration, age-at-vaccination category, study design, and publication year. Sensitivity analyses evaluated the influence of effect-measure type, alternative between-study variance estimation, and trim-and-fill adjustment. A frequentist network meta-analysis (NMA) compared vaccine types versus no vaccination.

Results

Twenty-two studies were included. HPV vaccination was associated with a pooled ratio estimate of 0.38 (95% CI 0.29–0.50), corresponding to an approximate 62% reduction in CIN2+ occurrence. Heterogeneity was very high (I² = 97.0%), and the 95% prediction interval crossed 1 (0.12–1.23), indicating substantial variation in expected effects across real-world vaccination programs. Point estimates were generally protective across subgroup strata, but subgroup differences by age at vaccination, vaccination coverage, and vaccine type were not statistically significant. Sensitivity analyses were directionally consistent with the primary analysis. The NMA suggested lower CIN2+ occurrence for all vaccine-type categories compared with no vaccination; however, all comparisons were indirect through a common no-vaccination comparator, so vaccine-type differences should be interpreted as exploratory.

Conclusions

HPV vaccination was associated with a substantial average reduction in CIN2+ occurrence across observational studies. However, high heterogeneity, wide prediction intervals, nonsignificant subgroup tests, and indirect network comparisons indicate that subgroup and vaccine-type findings should be interpreted cautiously.

Cervical cancer remains a significant global health issue, primarily caused by persistent infection with high-risk human papillomavirus (HPV) types [1, 2]. Cervical intraepithelial neoplasia grade 2 or higher (CIN2+) represents a critical precancerous lesion that, if left untreated, can progress to invasive cervical cancer [3]. Consequently, CIN2+ is widely used in epidemiological studies as a surrogate marker for evaluating cervical-cancer prevention strategies [4].

Observational studies play a pivotal role in assessing population-level impact, as they reflect real-world variations in vaccine uptake, adherence to vaccination schedules, healthcare infrastructure, and population behaviors. Previous meta-analyses have documented substantial reductions in CIN2+ incidence (30%–60%) following HPV-vaccine introduction, particularly among adolescents vaccinated before sexual debut [5, 6]. More recently, a systematic review [7] updated evidence on HPV-vaccine effectiveness (individual-level VE, by dose number), reinforcing strong protection at the individual level even in catch-up cohorts. Similarly, large observational studies have demonstrated reductions in invasive cervical cancer following HPV vaccination [8], while modeling analyses predict that high-coverage vaccination programs could eliminate cervical cancer as a public health problem within decades [9].

Despite this growing evidence base, important uncertainties remain regarding how the real-world impact of HPV vaccination varies across populations. In particular, the quantitative influence of age at vaccination, population-level coverage, and vaccine type on population-level CIN2+ reduction has not been fully characterized. To address these gaps, this meta-analysis synthesizes evidence from diverse programmatic contexts and evaluates population-level reductions in CIN2+ across age groups, coverage levels, and vaccine types. In addition, to examine comparative population impact across vaccine products in the absence of direct observational comparisons, we applied a network meta-analysis (NMA) framework informed by emerging data on newer vaccine schedules [10].

RATIONALE FOR A CIN2+ META-ANALYSIS

Previous reviews of HPV-vaccine impact have combined a wide range of outcomes, including HPV infection, genital warts, and cervical abnormalities, making quantitative synthesis challenging because of differences in effect measures, screening eras, and diagnostic practices. In addition, infection-based outcomes often exhibit greater diagnostic and reporting variability and longer latency, limiting their utility for comparative, program-level synthesis [5–7]. CIN2+ is a clinically relevant and policy-aligned endpoint that enables cross-study comparability [2, 11, 12] and aligns with World Health Organization (WHO) cervical-cancer elimination goals [1, 13], making it suitable for monitoring progress.

A focused analysis restricted to CIN2+ therefore enables clearer interpretation of population-level trends and more valid cross-study comparison. Accordingly, this CIN2+-focused meta-analysis quantitatively synthesizes CIN2+ outcomes and examines effect modification by age at vaccination, vaccination coverage, and vaccine type.

OBJECTIVES

The primary objective was to estimate the pooled population-level impact of HPV vaccination programs on CIN2+. Secondary objectives were to (1) assess effect modification by age at vaccination, coverage, and vaccine type, and (2) compare vaccine types versus no vaccination using a frequentist NMA.

METHODS

Design and Registration

This meta-analysis is a prespecified CIN2+-focused extension of our broader published systematic review, which synthesized real-world HPV vaccination impact across multiple HPV-related outcomes [14]. Using the same search, screening, eligibility, and risk-of-bias procedures as the companion review, we restricted this meta-analysis to studies reporting histologically confirmed CIN2+, a clinically relevant endpoint for population-level evaluation, and performed endpoint-specific quantitative syntheses. The companion review was registered in PROSPERO (CRD420251042433), and this CIN2+-focused analysis was reported in accordance with PRISMA 2020 [15].

Primary Outcome of Current Meta-analysis

The prespecified primary outcome was histologically confirmed CIN2+, which yields comparable effect measures with sufficient data for robust pooling.

Eligibility Criteria

Studies were eligible if they: (1) reported histologically confirmed CIN2+ as an outcome; (2) compared vaccinated and unvaccinated females or assessed CIN2+ incidence before and after HPV-vaccine program implementation; (3) provided effect estimates (risk ratio (RR), odds ratio (OR), incidence rate ratio (IRR), hazard ratio (HR), prevalence ratio (PR) or a derived/inverted estimate) with 95% confidence intervals (CI); (4) employed an ecological, cross-sectional, or cohort design; and (5) achieved moderate to high quality based on Joanna Briggs Institute (JBI) tools [16].

Quality Appraisal

Quality appraisal was conducted using design-specific JBI tools because the included evidence comprised cohort, cross-sectional, and ecological studies. Each study was independently assessed by 2 reviewers from a team of 3, with disagreements resolved through discussion. Studies were eligible for inclusion if they met the prespecified moderate-to-high quality threshold, defined as ≥75% of applicable JBI items rated Yes. Appraisal results were not used to weight studies statistically but informed eligibility and interpretation of risk of bias.

Subgroup Definitions

We examined effect modification by age at vaccination, coverage, and vaccine type.

• Age at vaccination. We prespecified age bands (<15, 15–18, ≥19 years at first dose) to reflect biologic plausibility (greater likelihood of HPV-naïve status at younger ages) and programmatic practice (routine vs catch-up). Each study was assigned to an age stratum (ie, subgroup category) based on the age group comprising ≥50% of vaccinated person-time (or ≥50% of vaccinated participants when person-time was unavailable). If age distributions spanned multiple bands and a majority could not be determined, we inferred the stratum from program descriptions (routine school-based: <15 years; adolescent school-based: 15–18 years; adult catch-up: ≥19 years).

• Vaccine coverage. Coverage was defined as the proportion of the target population completing the recommended series within the study's observation window (series completion preferred; initiation used if completion was unavailable). When coverage varied over time, we used the value representing most of the person-time or the midperiod value when unclear. Coverage was classified a priori as <50%, 50%–79%, or ≥80%. If coverage was not directly reported, studies with insufficient information were coded as not available (NA).

• Vaccine type. Studies were grouped by the vaccine type predominantly administered: bivalent, quadrivalent, or mixed (transitional or multiproduct programs). One included study evaluated the nonvalent vaccine (9vHPV); because only 1 study contributed data, it was not analyzed as a separate vaccine-type category and was grouped within the mixed/other vaccine-type stratum for descriptive synthesis. Mixed vaccine type refers to study-level programs in which more than 1 HPV-vaccine product was used during the study period, including transitional programs or multiproduct regional or national programs. It does not necessarily imply that individual participants received mixed vaccine products across doses unless explicitly reported by the original study.

• Follow-up period was recorded as a key contextual characteristic because CIN2+ incidence and detection depend on screening uptake, lesion latency, program maturity, and timing of outcome ascertainment after vaccine roll-out. Follow-up refers to time from vaccine program implementation or vaccination exposure to outcome ascertainment, as reported by each study, not necessarily individual longitudinal follow-up. Because follow-up definitions were heterogeneous, we summarized follow-up descriptively and examined follow-up duration as an exploratory study-level moderator in univariable meta-regression.

STATISTICAL ANALYSIS

Effect Scale and Extraction

For each study, we extracted 1 selected ratio estimate for the CIN2+ meta-analysis. The selected estimate was the adjusted study-level ratio measure most closely aligned with the CIN2+ outcome and the main vaccinated-versus-unvaccinated, vaccine-offered-versus-prevaccination, or postvaccination-versus-prevaccination comparison. When multiple estimates were reported, we prioritized adjusted estimates, estimates corresponding to completed or program-relevant vaccination when available and estimates that matched the prespecified CIN2+ endpoint. When the original comparison was reported in the opposite direction, the estimate was inverted so that values below 1 consistently indicated lower CIN2+ occurrence after HPV vaccination.

Reported RRs, ORs, HRs, IRRs, PRs, and derived or inverted estimates were analyzed on the natural log scale as reported ratio measures; we did not recalculate RRs from participant-level data. Because these measures do not estimate identical causal quantities, meta-analytic results are reported as pooled ratio estimates rather than pooled RRs. The original effect-measure type used from each study is reported in Table 1 and the supplementary extraction Table S1.

Table 1.

Characteristics and Quality of Included Studies on Impact of HPV Vaccination on Reduction of CIN2+ (N = 22)

Study	Age (Years) or Birth Cohort, Participant Number (N), Country	Design	Vaccine (Type, Doses, Coverage)	Comparator and Year	Data Source and Follow-up	Outcome and Key Findings	Effect Measure	Selected Ratio Estimate (95% CI)a	Study Quality
Hariri et al., 2015 [17]	≤ 26 (4678), USA	Cohort	4vHPV; ≥1 dose; 25.4%	Unvaccinated	HPV-IMPACT; 2008–2012 (∼4 yrs)	↓ HPV16/18-attributable CIN2+	PR	0.28 (0.14–0.55)	High
Niccolai et al., 2017 [18]	21–39 (15 218), USA	Ecological	2vHPV; 3 doses; 50% coverage	Unvaccinated (2006)	Public surveillance DB; 2008–2015 (∼8 yrs)	↓ CIN2+ by 30%–74% in young adults	Ecological RR/derived rate ratio	0.30 (0.20–0.43)	High
Gargano et al., 2022 [19]	20–39 (18 344 cases), USA	Cohort	4vHPV; 3 doses; NR	Unvaccinated (2006)	HPV-IMPACT; 2008–2016 (∼8 yrs)	↓ CIN2+ by 77% in 20–24 y	IRR/derived incidence ratio	0.23 (0.11–0.49)	High
Rodriguez et al., 2020 [20]	9–26 (133 082), USA	Cohort	4vHPV; 1–3 doses; ∼50%	Unvaccinated (2006)	Claims data; 2006–2016 (∼10 yrs)	↓ CIN2/3, consistent across doses	HR	0.66 (0.55–0.80)	Moderate
Donken et al., 2021 [21]	16–28 (NR), Canada	Ecological	4v/9v; 2–3 doses; ∼60%–69%	Pre-vax period	Screening registry; 2004–2017 (∼13 yrs)	↓ CIN2 by 62% in women aged 16–23. RR = 0.38. No decline in older, unvaccinated cohorts.	RR	0.38 (0.32–0.46)	High
Racey et al., 2020 [22]	15–29 (38 304), Canada	Cohort	4vHPV; 3 doses; 66.9%	Unvaccinated (2008)	Health and immunization registries; 10 yrs	↓ CIN2+ incidence, strongest in early vaccination.	RR	0.42 (0.31–0.57)	Moderate
Ogilvie et al., 2015 [23]	15–22 (NR), Canada	Ecological	4vHPV; 3 doses; 58%–62%	Pre-vax	Screening registry; 2004–2012 (∼8 yrs)	↓ CIN2+ in 15–17 yrs	IRR	0.31 (0.13–0.75)	High
Brotherton et al., 2015 [24]	≤26 yrs at screening (289 478), Australia	Cohort	4v; 1–3 doses; routine coverage ≥70%–80%, catch-up lower (∼30%–50%)	Unvaccinated (2007)	Cytology registry; ∼5 yrs	↓ CIN2+ across all doses; dose–response observed	HR	0.71 (0.64–0.80)	High
Gertig et al., 2013 [25]	≤ 17 yrs at screening (38 956), Australia	Cohort	4v; 3 doses; ∼85% routine coverage	Unvaccinated (2007)	Cytology registry; ∼5 yrs	↓ CIN2 by ∼30% in fully vaccinated	HR	0.70 (0.52–0.94)	High
Goodman et al., 2024 [26]	28–33 (50 333), Germany	Claims-based analysis	4vHPV; 3 doses; 54.1%	Pre-vax cohort	SHI database; ∼14 yrs	↓ CIN2+ prevalence by 51.1% postvaccination; selected prevalence ratio = 0.49	PR/impact ratio	0.49 (0.38–0.63)	Moderate
Osmani et al., 2022 [27]	9–28 (433 346), Germany	Cohort	All vaccines; 2–3 doses; 13%–41%	Unvaccinated (2007)	SHI data; 2008–2018 (∼10 yrs)	↓ CIN2+; limited by low uptake	HR	0.77 (0.71–0.84)	High
Dong et al., 2023 [28]	Born 1975–96 (832 732), Norway	Cohort	4vHPV; 3 doses; 5.6%	Unvaccinated (2006)	National registry; 2006–2016 (∼10 yrs)	↓ CIN2+ if vaccinated <20 yrs. No benefit seen for ≥20 yrs at vaccination.	IRR	0.62 (0.46–0.84)	High
Mikalsen et al., 2024 [29]	20–25 (15 328), Norway	Ecological	2v/4v; ≥1 dose; 67%–86%	Unvaccinated (2009)	Cancer registry; 15 yrs	Prevaccine cohorts had higher CIN2+ risk; estimate inverted for vaccinated-direction comparison	Inverted OR	0.11 (0.07–0.17)	High
Palmer et al., 2019 [30]	Born 1988–96 (138 692), Scotland	Cohort	2vHPV; 3 doses; >85%	Pre-vax (1988–1990)	Screening/vax registry; ∼7–8 yrs	↓ CIN2+ by 88% at age 12–13 vaccination	OR	0.12 (0.08–0.17)	High
Pollock et al., 2014 [31]	Born 1988–92 (106 042), Scotland	Cohort	2vHPV; 3 doses; 71.8%	Pre-vax cohort	Screening registry; 2008–2013 (∼5 yrs)	↓ CIN2 risk. Significant reduction in both low- and high-grade abnormalities.	RR	0.50 (0.40–0.63)	High
Wu et al., 2025 [32]	10–35 (2.2 M), Sweden	Cohort	4vHPV; 1–3 doses; 83.5%	Unvaccinated (2007)	National registries; 2006–2022 (∼17 yrs)	↓ CIN2+ across doses; stronger in younger people. Three doses required for protection in older groups.	IRR	0.50 (0.47–0.53)	High
Herweijer et al., 2016 [33]	13–29 (1.3 M), Sweden	Cohort	4vHPV; 3 doses; 17.7%	Unvaccinated (2006)	National registry; 2006–2013 (∼7 yrs)	↓ CIN2+ by 75% in <17 yrs. Effectiveness declined with age at vaccination.	IRR	0.25 (0.18–0.35)	High
Acuti Martellucci et al., 2022 [34]	25–30 (4665), Italy	Cohort	2vHPV; 3 doses; 24%	Unvaccinated	Local registry; 2015–2020 (∼5 yrs)	↓ CIN2+ risk, greater when younger at vaccination	OR	0.33 (0.11–0.96)	High
Dehlendorff et al., 2018 [35]	13–30 (2.25 M), Denmark/Sweden	Cohort	4vHPV; 3 doses; 32%	Unvaccinated (2006)	National registries; 2006–13 (∼7 yrs)	Large reduction in <17 y group	IRR	0.23 (0.11–0.49)	High
Thamsborg et al., 2020 [36]	Born 1983 & 1993 (35 699), Denmark	Cohort	4vHPV; 3 doses; 91%–93%	1983 birth cohort	National registers; ∼10 yrs	↓ CIN2+ overall; largest in vaccinated cohort	RR	0.74 (0.66–0.82)	High
Dvořák et al., 2024 [10]	22–49 (147), Czech Republic	Cohort	9vHPV; 3 doses postexcision	Unvaccinated match	Clinic data; 30 m	↓ CIN2+ recurrence by 90%	IRR	0.10 (0.01–0.88)	High
Shiko et al., 2020 [37]	Mean age ∼22 (20–29) (34 281), Japan	Cross-sectional	2v or 4v; ≥1 dose; coverage ∼10.9%	Unvaccinated (2010–13)	Cancer society DB; 2 yrs	↓ CIN2+ in vaccinated versus unvaccinated	RR	0.24 (0.10–0.60)	High

Symbols: ↓ indicates reduction; ∼ indicates approximate value. N refers to the participant number or analytic denominator used for the selected CIN2+ estimate where reported or derivable. For ecological or surveillance studies, denominators may represent screened women, registry records, or CIN2+ cases rather than individual participants. NR, not reported as a single participant-level analytic sample size.

Abbreviations: CIN2+, cervical intraepithelial neoplasia grade 2 or worse; N, number of participants; CI, confidence interval; HPV, human papillomavirus; 2vHPV, bivalent HPV vaccine; 4vHPV, quadrivalent HPV vaccine; 9vHPV, nonavalent HPV vaccine; RR, risk ratio or rate ratio, as reported by individual studies; PR, prevalence ratio; OR, odds ratio; HR, hazard ratio; IRR, incidence rate ratio; vax, vaccinated; SHI, statutory health insurance; HPV-IMPACT, Human Papillomavirus Vaccine Impact Monitoring Project; DB, database; NR, not reported; NA, not available/not applicable.

^aSelected ratio estimate refers to the single study-level effect estimate extracted for the CIN2+ meta-analysis, prioritizing the adjusted estimate most closely aligned with the CIN2+ outcome and the main vaccinated-versus-unvaccinated, vaccine-offered-versus-prevaccine, or postvaccination-versus-prevaccination comparison. Because included studies reported different ratio measures, including RR, OR, HR, IRR, PR, and derived or inverted estimates, selected estimates were analyzed on the natural log scale. Values below 1 indicate lower CIN2+ occurrence after HPV vaccination. One study had vaccination coverage not reported and was excluded from coverage subgroup pooling.

Primary Meta-analysis and Heterogeneity

The primary meta-analysis included all selected ratio estimates. Pooled ratio estimates were calculated using random-effects models fitted with restricted maximum likelihood (REML) and Hartung–Knapp–Sidik–Jonkman inference. Between-study heterogeneity was assessed using Cochran's Q and summarized using I² and τ² on the natural log scale. To convey expected real-world dispersion, we reported 95% prediction intervals, which reflect the expected range of true effects across comparable programs and are distinct from the 95% CI around the pooled mean.

Prespecified Subgroup Analyses

We fit separate random-effects models by age at vaccination (<15, 15–18, and ≥19 years), vaccination coverage (<50%, 50%–79%, and ≥80%), and vaccine type (bivalent, quadrivalent, and mixed). Tests for subgroup differences were used to assess whether pooled ratio estimates differed across strata. Subgroup analyses were interpreted cautiously because of the limited number of studies within several strata and substantial between-study heterogeneity.

Sensitivity Analyses

To assess the influence of combining different ratio measures, we conducted sensitivity analyses by effect-measure type. These included analyses excluding OR and inverted OR studies, excluding rate- or hazard-based measures, excluding PR or impact-ratio studies, restricting to cumulative-type measures only (RR/OR/PR), and restricting to RR-only studies. Additional robustness analyses included re-estimating τ² using the DerSimonian–Laird estimator and comparing the primary pooled ratio estimate with the trim-and-fill adjusted estimate.

Exploratory Meta-regression

To explore residual heterogeneity, we conducted exploratory univariable mixed-effects meta-regressions for follow-up duration, age-at-vaccination category, study design, and publication year. Follow-up duration was modeled as years from vaccine program implementation or vaccination exposure to outcome ascertainment, as reported by each study. Age at vaccination was modeled as an ordinal study-level moderator using the prespecified categories of <15, 15–18, and ≥19 years. Study design was modeled as cohort versus noncohort because only 1 cross-sectional study was included and the number of ecological studies was limited. Publication year was modeled as a continuous contextual moderator. We did not fit multivariable meta-regression models because the number of included studies was small, covariate completeness varied, and key moderators such as follow-up duration, screening era, program maturity, and coverage were conceptually correlated. All meta-regression analyses were interpreted as exploratory and hypothesis-generating.

Small-study Effects and Publication Bias

We visually assessed funnel plots and formally tested asymmetry using Egger's regression. We explored the potential impact of missing studies using trim-and-fill analysis [38], with adjusted estimates compared against the primary pooled ratio estimate.

NMA Comparing Vaccine Types

Because no direct head-to-head observational comparisons between vaccine types were available, we conducted a frequentist NMA comparing vaccine types with no vaccination as the common reference. Vaccine rankings were summarized using rank-probability histograms. Because the network was star-shaped and relied primarily on indirect comparisons, NMA findings were interpreted as exploratory rather than definitive comparative evidence. Potential violations of transitivity were considered possible because studies differed by age at vaccination, coverage, screening era, follow-up duration, and program context.

RESULTS

Study Characteristics

From 13 549 screened records, 22 studies met inclusion criteria for this CIN2+ meta-analysis (PRISMA, Figure 1). Most were cohort studies from high-income countries, with limited evidence from middle-income regions and none from low-income settings. Vaccination coverage ranged from 32% to 92%, with follow-up of 2–17 years. Shorter follow-up generally reflected earlier program phases, whereas longer follow-up captured more mature programs. Follow-up and screening context are summarized in Table 1.

Figure 1.

PRISMA flow diagram.

Quality Assessment

All 22 studies met the predefined moderate-to-high quality threshold on design-specific JBI checklists (≥75% of applicable items rated Yes), including 16 cohort, 1 cross-sectional, and 5 ecological studies. Common concerns were residual confounding (eg, sexual behavior), misclassification of vaccination status and, among ecological studies, the lack of individual-level comparators. Study-level scores and limitations are summarized in Table 1, with full appraisal details in Supplementary Table 1.

Pooled CIN2+ Reduction

In the primary meta-analysis of 22 studies, HPV vaccination was associated with a pooled ratio estimate of 0.38 (95% CI 0.29–0.50), corresponding to an approximate 62% reduction in CIN2+ occurrence. Heterogeneity was very high (I² = 97.0%; τ² = 0.2973; Q = 311.71, P < .001; Figure 2). The 95% prediction interval was 0.12 to 1.23, indicating that the expected effects may vary substantially across comparable vaccination program contexts. Given this high heterogeneity, the pooled estimate should be interpreted as an average association rather than a precise universal benchmark, consistent with published guidance [39, 40]; random-effects meta-analysis can yield meaningful average effects when heterogeneity reflects real contextual variation rather than bias.

Figure 2.

Forest plot from the random-effects meta-analysis of selected ratio estimates for CIN2+ after HPV vaccination. Study-specific selected ratio estimates, and 95% confidence intervals are shown on the ratio scale. The diamond represents the pooled ratio estimate from a random-effects model fitted using restricted maximum likelihood with Hartung–Knapp adjustment. Values below 1 indicate lower CIN2+ occurrence after HPV vaccination, whereas values above 1 would indicate higher occurrence. The 95% prediction interval reflects the expected range of effects in comparable future vaccination program contexts and may cross 1 even when the pooled estimate is below 1.

Subgroup Analysis

To explore whether the impact of HPV vaccination on CIN2+ varied across key program dimensions, we conducted prespecified subgroup analyses stratified by age at vaccination, vaccination coverage, and vaccine type. Results are summarized in Table 2 and Figures 3–5.

Table 2.

Subgroup Meta-Analyses of CIN2+ After HPV Vaccination

Subgroup Category	Subgroup	No. of Studies	Pooled Ratio Estimate (95% CI)	Percent Reduction	95% Prediction Interval	I2 (%)	Tau2	Q	P Value
Overall	All studies	22	0.38 (0.29–0.50)	62%	0.12 to 1.23	97.0	0.2973	311.71	<0.001
Age at vaccination	<15 y	7	0.34 (0.16–0.70)	66%	0.04 to 2.59	98.8	0.6059	170.02	<0.001
Age at vaccination	15–18 y	11	0.43 (0.33–0.57)	57%	0.18 to 1.01	89.0	0.1298	97.58	<0.001
Age at vaccination	≥19 y	4	0.40 (0.14–1.14)	60%	0.05 to 2.93	69.4	0.2857	11.55	0.009
Vaccine coverage	<50%	7	0.38 (0.23–0.61)	62%	0.10 to 1.38	87.9	0.2426	64.94	<0.001
Vaccine coverage	50%–79%	9	0.46 (0.35–0.60)	54%	0.22 to 0.94	85.9	0.0852	60.67	<0.001
Vaccine coverage	≥80%	5	0.33 (0.10–1.06)	67%	0.02 to 5.58	99.3	0.8658	151.17	<0.001
Vaccine coverage	NR/NA coverage	1	—	—	—	—	—	—	—
Vaccine type	Quadrivalent	13	0.47 (0.37–0.62)	53%	0.20 to 1.10	94.0	0.1351	105.64	<0.001
Vaccine type	Bivalent	4	0.28 (0.10–0.75)	72%	0.03 to 2.45	90.6	0.3724	40.76	<0.001
Vaccine type	Mixed	5	0.29 (0.10–0.80)	71%	0.02 to 3.25	97.4	0.6375	118.07	<0.001

Tests for subgroup differences were not statistically significant for age at vaccination (Q_M = 0.22, df = 2, P = .805), vaccination coverage (Q_M = 0.36, df = 2, P = .704), or vaccine type (Q_M = 1.72, df = 2, P = .205). Subgroup analyses were exploratory. Although point estimates were generally protective across age, coverage, and vaccine-type strata, prediction intervals were wide, particularly for the <15 y, ≥19 y, ≥80% coverage, bivalent, and mixed vaccine-type strata.

Figure 3.

Forest plot of CIN2+ reduction stratified by age at vaccination. Top: <15 y; middle: 15–18 y; bottom: ≥19 y. Squares denote study-specific selected ratio estimates with 95% CIs; diamonds show pooled ratio estimates from random-effects models fitted using REML with Hartung–Knapp–Sidik–Jonkman inference. The vertical line marks a ratio estimate of 1. Reported I², τ², Q, and P values summarize within-stratum heterogeneity.

Figure 5.

Forest plot of CIN2+ reduction stratified by HPV-vaccine type. (Top) Quadrivalent vaccine, (middle) bivalent vaccine, (bottom) mixed schedule.

Table 2 summarizes pooled ratio estimates with 95% CIs from random-effects models, stratified by age at vaccination (<15, 15–18, and ≥19 years), vaccination coverage (<50%, 50%–79%, and ≥80%), and vaccine type (bivalent, quadrivalent, and mixed). Because included studies reported different ratio measures, subgroup results are presented as pooled ratio estimates rather than pooled RRs.

Effect by Age at Vaccination

Vaccination was associated with reduced CIN2+ occurrence across all prespecified age-at-vaccination strata (Table 2; Figure 3). The pooled ratio estimate was 0.34 (95% CI 0.16–0.70) for vaccination before age 15, 0.43 (95% CI 0.33–0.57) for vaccination at ages 15–18 years, and 0.40 (95% CI 0.14–1.14) for vaccination at age ≥19 years. Although point estimates were generally protective, prediction intervals were wide, particularly for the <15 years and ≥19 years strata. The test for subgroup differences was not statistically significant (Q_M = 0.22, df = 2, P = .805), indicating no clear evidence of differential pooled effects across age-at-vaccination strata.

Effect by Vaccination Coverage

Point estimates were generally protective across vaccination coverage strata, but uncertainty varied substantially across groups (Table 2; Figure 4). For studies with <50% coverage, the pooled ratio estimate was 0.38 (95% CI 0.23–0.61), corresponding to an approximate 62% reduction in CIN2+ occurrence. For studies with 50%–79% coverage, the pooled ratio estimate was 0.46 (95% CI 0.35–0.60), corresponding to an approximate 54% reduction. For studies with ≥80% coverage, the pooled ratio estimate was 0.33 (95% CI 0.10–1.06), corresponding to an approximate 67% reduction, but the CI was wide and crossed 1. One study had vaccination coverage not reported and was excluded from coverage subgroup pooling.

Figure 4.

Forest plot of CIN2+ reduction stratified by vaccination coverage. (Top) Coverage <50%, (middle) coverage 50%–79% and (bottom) coverage ≥ 80%.

The test for subgroup differences was not statistically significant (Q_M = 0.36, df = 2, P = .704), indicating no clear evidence that pooled effects differed across coverage strata. The ≥80% coverage stratum should be interpreted cautiously because it included few studies and showed very high heterogeneity (I² = 99.3%).

Effect by Vaccine Type

All vaccine-type strata were associated with lower CIN2+ occurrence, although between-stratum differences were not statistically significant (Table 2; Figure 5). The pooled ratio estimate was 0.47 (95% CI 0.37–0.62) for quadrivalent vaccine studies, 0.28 (95% CI 0.10–0.75) for bivalent vaccine studies, and 0.29 (95% CI 0.10–0.80) for mixed vaccine-type studies. Although the bivalent and mixed categories had lower point estimates than the quadrivalent category, CIs were wide and the test for subgroup differences was not statistically significant (Q_M = 1.72, df = 2, P = .205). These contrasts should therefore be interpreted cautiously because they are based on comparisons across heterogeneous programs that differ in age at vaccination, vaccination coverage, screening context, follow-up duration, study design, and calendar period.

Sensitivity and Robustness Analyses

As shown in Table 3, sensitivity and robustness analyses were directionally consistent with the primary analysis. In the primary analysis, including all selected ratio estimates, the pooled ratio estimate was 0.38 (95% CI 0.29–0.50), with a 95% prediction interval of 0.12 to 1.23. Re-estimating between-study variance using the DerSimonian–Laird estimator yielded a similar pooled ratio estimate of 0.40 (95% CI 0.31–0.52). Excluding OR and inverted OR studies yielded 0.46 (95% CI 0.37–0.57), while excluding rate- or hazard-based measures yielded 0.32 (95% CI 0.21–0.48). Excluding PR or impact-ratio studies produced an estimate like the primary analysis, 0.38 (95% CI 0.29–0.51). Restricting the analysis to cumulative-type measures only (RR/OR/PR) yielded 0.32 (95% CI 0.21–0.48), and the RR-only analysis yielded 0.44 (95% CI 0.30–0.64). The trim-and-fill adjusted estimate was also like the primary analysis, 0.39 (95% CI 0.30–0.50).

Table 3.

Sensitivity and Robustness Analyses of CIN2+ Meta-analysis

Analysis	No. of Studies	Pooled Ratio Estimate (95% CI)	95% Prediction Interval	I 2 (%)	Tau2	Q	Q P Value
Primary: all selected ratio estimates	22	0.38 (0.29–0.50)	0.12 to 1.23	97.0	0.2973	311.71	<0.001
DerSimonian–Laird tau2 estimator	22	0.40 (0.31–0.52)	0.18 to 0.88	93.3	0.1268	311.71	<0.001
Excluding OR/inverted OR studies	19	0.46 (0.37–0.57)	0.21 to 1.02	94.3	0.1336	191.36	<0.001
Excluding rate/hazard-based studies	11	0.32 (0.21–0.48)	0.08 to 1.23	94.5	0.3322	172.75	<0.001
Excluding PR/impact-ratio studies	20	0.38 (0.29–0.51)	0.11 to 1.30	97.4	0.3253	306.33	<0.001
Cumulative-type measures only: RR/OR/PR	11	0.32 (0.21–0.48)	0.08 to 1.23	94.5	0.3322	172.75	<0.001
RR-only studies	6	0.44 (0.30–0.64)	0.17 to 1.10	88.9	0.1077	60.88	<0.001
Trim-and-fill adjusted estimate	23	0.39 (0.30–0.50)	0.13 to 1.16	96.9	0.2963	312.44	<0.001

Although all sensitivity and robustness analyses remained protective, heterogeneity remained high across models, with I² ranging from 88.9% to 97.4%. These findings support the robustness of the overall protective association but also indicate that interpretational differences across effect-measure types and residual between-study heterogeneity remain important. Therefore, the pooled estimate should be interpreted as a pooled ratio estimate rather than a common RR.

Meta-regression

As shown in Table 4, exploratory univariable meta-regressions did not identify statistically significant moderators of the pooled ratio estimate. Follow-up duration was not associated with the pooled estimate (β = .0071, 95% CI −0.0643 to 0.0784; P = .838), and age-at-vaccination category also showed no statistically significant association (β = .1015, 95% CI −0.3009 to 0.5039; P = .605). Publication year was not associated with effect size (β = −.0173, 95% CI −0.0921 to 0.0575; P = .635). Study design showed a suggestive but nonsignificant association (β = −.4727, 95% CI −1.0137 to 0.0683; P = .083; pseudo-R² = 12.99%). However, between-study heterogeneity remained high across all models, and these aggregate-level analyses were underpowered. Therefore, the meta-regression findings should be interpreted as exploratory and hypothesis-generating rather than definitive evidence that these moderators do or do not explain heterogeneity.

Table 4.

Exploratory Univariable Meta-regression Results

Moderator	K	Beta	95% CI	P Value	Tau2	I2 (%)	QM	QM P Value	Pseudo-R2 (%)
Follow-up duration, years	22	0.0071	−0.0643 to 0.0784	0.838	0.3156	96.3	0.0429	0.838	0.00
Age-at-vaccination category	22	0.1015	−0.3009 to 0.5039	0.605	0.3043	96.5	0.2768	0.605	0.00
Study design: cohort versus noncohort	22	−0.4727	−1.0137 to 0.0683	0.083	0.2587	96.6	3.3216	0.083	12.99
Publication year	22	−0.0173	−0.0921 to 0.0575	0.635	0.3133	96.6	0.2328	0.635	0.00

Small-study Effects and Publication Bias

Funnel plots (Supplementary Figures 1 and 2) showed asymmetry consistent with possible small-study effects. Egger's regression was statistically significant (z = −3.59; P = .0003). Because funnel plot asymmetry can also arise from substantial between-study heterogeneity, selective outcome reporting, or differences in study design and program context, we assessed the potential impact of missing studies using trim-and-fill analysis (Supplementary Figure 2). The trim-and-fill adjusted estimate was materially unchanged from the primary analysis, with an absolute difference in the pooled ratio estimate of <0.03. These findings suggest that small-study effects are unlikely to fully explain the overall protective association, although asymmetry should still be interpreted cautiously given the high heterogeneity.

NMA of Vaccine Types

The evidence network was star-shaped, with each vaccine type connected only through the common comparator of no vaccination and no direct head-to-head comparisons between vaccine types (Figure 6). We conducted a random-effects NMA comparing vaccine types with no vaccination as the reference. All vaccine types were associated with lower CIN2+ occurrence compared with no vaccination. The pooled ratio estimate was lowest for the bivalent vaccine, 0.28 (95% CI 0.18–0.43), followed by mixed vaccine programs, 0.32 (95% CI 0.21–0.49), and the quadrivalent vaccine, 0.47 (95% CI 0.37–0.60). Rank-probability histograms showed that the bivalent vaccine had the highest probability of being ranked most protective, defined as having the lowest CIN2+ ratio estimate. Mixed vaccine programs were most often ranked second, whereas quadrivalent vaccine was most often ranked below bivalent and mixed programs. However, because the network lacked direct head-to-head comparisons between vaccine types, these rankings should be interpreted as exploratory. Apparent differences between vaccine types may reflect differences in age at vaccination, vaccination coverage, screening context, follow-up duration, study design, and calendar period rather than intrinsic differences between vaccine products.

Figure 6.

Network meta-analysis of HPV-vaccine types versus no vaccination for preventing CIN2+. Panel (A) shows the star-shaped evidence network, with all vaccine types connected only through no vaccination. Panel (B) shows random-effects pooled ratio estimates versus no vaccination. Panel (C) shows rank-probability histograms; Rank 1 indicates the lowest CIN2+ ratio estimate. Rankings should be interpreted cautiously because vaccine-type comparisons were indirect.

DISCUSSION

Main Findings

This meta-analysis found that HPV vaccination was associated with a substantial reduction in CIN2+ occurrence across real-world observational studies. In the primary analysis of 22 studies, the pooled ratio estimate was 0.38 (95% CI 0.29–0.50), corresponding to an approximate 62% reduction in CIN2+ occurrence. Given that the bivalent and quadrivalent vaccines used in most included programs primarily target HPV16 and HPV18, which together account for approximately 70% of cervical cancers and a large proportion of CIN2+ lesions [2, 41], this magnitude of reduction represents an important population-level public-health impact. At the same time, the pooled estimate should be interpreted as an average association across heterogeneous programs rather than a single universal effect.

Point estimates were generally protective across age-at-vaccination, vaccination coverage, and vaccine-type strata. The lowest pooled estimate was observed among studies in which vaccination occurred before age 15, consistent with the expected benefit of vaccination before HPV exposure. Protective estimates were also observed for vaccination at ages 15–18 years and ≥19 years, although the ≥19-year stratum was less precisely estimated. This pattern is broadly consistent with cohort, surveillance, and modeling evidence showing the greatest impact when vaccination is delivered before sexual debut, while still demonstrating reductions in high-grade cervical lesions and, in some settings, invasive cervical cancer among those vaccinated in later adolescence or young adulthood [5, 8, 28, 32, 42–45]. Protective estimates were also evident across vaccination coverage bands, although the ≥80% coverage stratum had wide uncertainty and very high heterogeneity. Tests for subgroup differences were not statistically significant for age at vaccination, vaccination coverage, or vaccine type; therefore, these subgroup patterns should be interpreted as exploratory rather than definitive evidence of effect modification. By focusing specifically on CIN2+, stratifying program impact by key implementation dimensions, and reporting prediction intervals, this analysis extends prior syntheses and better reflects the variability expected across real-world settings.

The NMA suggested that all vaccine types were associated with lower CIN2+ occurrence compared with no vaccination. Rank-probability histograms indicated that the bivalent vaccine had the highest probability of being ranked most protective, followed by mixed vaccine programs and quadrivalent vaccine. However, the evidence network was star-shaped, with vaccine types connected only through the no-vaccination comparator and no direct head-to-head comparisons. Therefore, apparent vaccine-type differences may reflect differences in age at vaccination, vaccination coverage, screening context, follow-up duration, study design, or calendar period rather than intrinsic differences between vaccine products.

Interpreting Heterogeneity

Between-study heterogeneity was very high (I² = 97.0%), and the 95% prediction interval ranged from 0.12 to 1.23. This indicates that although HPV vaccination was associated with a substantial average reduction in CIN2+ occurrence, the magnitude of effect is expected to vary considerably across comparable real-world settings. The prediction interval crossing 1 is important because some settings may observe smaller or less certain reductions, even when the overall average association is strongly protective. This heterogeneity likely reflects real-world variation in age at vaccination, vaccination coverage, vaccine type, screening practices, follow-up duration, program maturity, and study design. The evidence base included individual-level cohort studies alongside ecological, surveillance-based, claims-based, and cross-sectional evaluations. These designs differ in confounding control, vaccination linkage, outcome ascertainment, and susceptibility to ecological bias. Differences in screening context may also contribute because CIN2+ detection depends not only on underlying disease risk but also on screening participation, diagnostic algorithms, and timing of outcome ascertainment.

Exploratory univariable meta-regressions did not identify statistically significant study-level moderators. Follow-up duration, age-at-vaccination category, study design, and publication year did not clearly explain the observed heterogeneity, although study design showed a suggestive but nonsignificant association. These analyses were limited by the small number of studies, aggregate-level covariates, and correlation among program characteristics such as follow-up duration, screening era, program maturity, and coverage. Therefore, the pooled estimate should be interpreted as an average association rather than a precise universal benchmark, with τ² and the prediction interval providing important information about expected real-world variability.

Biological Plausibility and Consistency With External Evidence

The observed protection at the CIN2+ endpoint is biologically plausible because HPV vaccination prevents persistent infection with oncogenic HPV types, thereby reducing the pool of precursor lesions that can progress to high-grade cervical disease. This causal pathway is supported by randomized trial evidence summarized in a recent Cochrane NMA, which reported substantial reductions in CIN2+, particularly for lesions attributable to vaccine HPV types, among vaccinated women [46]. The findings are also consistent with population-level evidence showing large declines in HPV infection and cervical precancer after HPV-vaccine program implementation and, in some settings, reductions in invasive cervical cancer [5, 8, 32].

Comparative Effectiveness and Exploratory Vaccine Rankings

The NMA supported the overall finding that HPV vaccination was associated with lower CIN2+ occurrence compared with no vaccination across vaccine-type categories. In the NMA, the bivalent vaccine had the lowest pooled ratio estimate, and rank-probability histograms indicated that it had the highest probability of being ranked most protective. However, these findings arise from a star-shaped evidence network in which vaccine types were compared indirectly through the common no-vaccination comparator, with no direct head-to-head comparisons between vaccine products. The apparent advantage of the bivalent vaccine may therefore reflect study mix, including younger age at vaccination, higher coverage, differences in screening context, longer or shorter follow-up, or earlier program eras, rather than intrinsic product effects. Prior evidence has suggested broader cross-protection of the bivalent vaccine against some nonvaccine oncogenic HPV types, including HPV31, HPV33, and HPV45, compared with quadrivalent vaccines [47, 48]. However, indirect observational evidence should not substitute for direct comparative trials. Recent randomized head-to-head comparisons of bivalent versus quadrivalent vaccines against CIN3+ [49], large publicly funded 1- versus 2-dose trials such as ESCUDDO [50], and WHO SAGE evidence reviews on reduced-dose HPV schedules [51] highlight the importance of rigorous comparative and dosing studies for refining HPV-vaccine policy. The expanded genotype coverage of the 9-valent vaccine suggests potential additional benefit, although real-world CIN2+ data remain sparse [12, 52]. Overall, these findings support programmatic flexibility: licensed HPV-vaccines reduce CIN2+ occurrence, and product choice should be guided by supply, affordability, delivery feasibility, local program context, and coverage goals. Given the scale and cost of national HPV vaccination programs, further investment in rigorous head-to-head, reduced-dose, and long-term real-world effectiveness studies remains a high-value priority for research and policy.

Alignment With Population-level Evaluations

The direction of effect observed in this meta-analysis is consistent with population-level evaluations showing reductions in CIN2+ and other high-grade cervical outcomes after HPV-vaccine program implementation, particularly when vaccination is delivered at younger ages and within high-coverage or multicohort programs [5, 43]. National evaluations show similar patterns. In England, the largest reductions were observed among cohorts offered vaccination at ages 12–13 years [42]. In Sweden, HPV vaccination was associated with a markedly lower risk of invasive cervical cancer, with the strongest reductions among women vaccinated at younger ages [8]. Scotland also documented substantial reductions in high-grade cervical disease following bivalent vaccination at ages 12–13 years [30]. Our synthesis differs from many prior evaluations in its endpoint-specific focus on real-world CIN2+ outcomes. Rather than combining HPV infection, cytological abnormalities, genital warts, and cervical precancer, this analysis focused on a clinically proximate cervical precancer endpoint and explicitly stratified results by age at vaccination, vaccination coverage, and vaccine type. This approach improves policy relevance while also showing that apparent subgroup patterns should be interpreted alongside substantial heterogeneity, wide prediction intervals, and the absence of statistically significant subgroup differences.

Programmatic Implications and Evidence Gaps

The protective estimates observed in older age-at-vaccination strata support the continued value of catch-up vaccination, although effects were less precise than for younger vaccination. This is consistent with U.S. Advisory Committee on Immunization Practices (ACIP) recommendations, as published by the Centers for Disease Control and Prevention (CDC), which prioritize routine adolescent vaccination while supporting catch-up vaccination through age 26 and shared clinical decision-making for adults aged 27–45 years [44]. Recent U.S. surveillance showing large declines in cervical precancers among age groups most likely to have been vaccinated further supports the population-level benefits of HPV vaccination [45]. Maximizing benefit now depends not only on vaccine effectiveness but also on implementation. Programs should prioritize high and equitable coverage through school-based delivery, community partnerships, culturally appropriate outreach, and strategies to reach high-burden and underserved populations. Linkage between immunization and screening registries will also be important for monitoring long-term program performance, identifying coverage gaps, and evaluating CIN2+ outcomes across population subgroups. Expanding access to HPV DNA-based screening, including self-collection where appropriate, may further strengthen the combined benefits of vaccination and screening.

The dominance of evidence from high-income countries limits generalizability to LMICs, where cervical-cancer burden is highest and where vaccination delivery, screening infrastructure, registry linkage, diagnostic timing, and treatment access may differ substantially. In LMICs, lower screening coverage, delayed diagnosis, incomplete vaccination or screening registries, and differential access to treatment may modify the observed relationship between HPV vaccination and CIN2+ detection. Therefore, the pooled estimate should not be assumed to directly predict the magnitude or timing of impact in LMIC settings. Priority evidence gaps include real-world CIN2+ outcomes in LMICs, durability of protection by age at vaccination, product-specific impact of 9-valent vaccine programs, performance of reduced-dose schedules, and the combined impact of vaccination with HPV-based screening. Dedicated implementation studies and linked vaccination-screening surveillance systems are needed to quantify real-world CIN2+ reductions in under-represented settings and to support progress toward cervical-cancer elimination.

Policy Implications

From a policy perspective, these findings reinforce HPV vaccination as a cornerstone of cervical-cancer prevention. Programs should prioritize early adolescent vaccination with high coverage, support equitable catch-up through age 26 (with shared decision-making for adults aged 27–45 years), and maintain flexibility in product choice to safeguard supply and affordability. Integration can be implemented by offering HPV vaccination at the same clinical or community encounters where HPV self-collection kits are distributed or returned, or during screening events [41]; electronic health-record prompts, bidirectional registry linkages, and mobile outreach for clients who miss either service. Addressing equity requires closing gaps in both HPV vaccination coverage and screening access (including treatment pathways), especially for underserved populations and LMICs [53], will be essential to achieving WHO's 2030 elimination targets [54] and to provide a practical roadmap for maximizing population-level benefit. Our findings suggest that real-world monitoring of CIN2+ incidence can provide an early indicator of program performance alongside coverage, enabling early identification of equity gaps.

From a health economics perspective, the findings support continued investment in early adolescent HPV vaccination because preventing CIN2+ reduces downstream costs associated with diagnostic colposcopy, treatment of high-grade lesions, surveillance, and invasive cervical-cancer care [55]. However, cost-effectiveness depends on vaccine price, delivery costs, coverage, screening intensity, and local cervical-cancer burden [56]. In high-burden settings, especially LMICs, the economic value of vaccination is likely to be greatest when vaccination is integrated with scalable HPV-based screening and treatment pathways. Program decisions should therefore consider not only comparative vaccine efficacy but also affordability, supply reliability, delivery feasibility, and the marginal benefit of increasing coverage among underserved groups [56].

Strengths and Contributions of This Review

This review extends prior syntheses [5, 6] and recent large-scale national evidence [32] by focusing specifically on CIN2+, a clinically proximate and policy-relevant cervical precancer endpoint. We quantified subgroup patterns by age at vaccination, vaccination coverage, and vaccine type; reported heterogeneity using I² and τ²; and included 95% prediction intervals to communicate the expected range of real-world effects across settings. We also explicitly addressed the challenge of combining different reported ratio measures by analyzing selected study-level estimates on the natural log scale and reporting pooled ratio estimates rather than pooled RRs.

A further contribution is the exploratory NMA of vaccine types. Although limited by the star-shaped evidence network and the absence of direct head-to-head comparisons, the NMA provides a transparent summary of available observational evidence comparing vaccine types with no vaccination. Importantly, the analysis emphasizes uncertainty and indirectness rather than making definitive claims about vaccine-type superiority.

Limitations

Several limitations warrant caution. First, between-study heterogeneity was very high (I² = 97.0%), and the 95% prediction interval crossed 1. The pooled estimate should therefore be interpreted as an average association rather than a precise universal benchmark. Second, the evidence base included individual-level cohort studies, ecological studies, and 1 cross-sectional study. These designs differ in confounding control, vaccination linkage, outcome ascertainment, and susceptibility to ecological bias, which could inflate or attenuate study-specific estimates.

Third, included studies reported different effect measures, including RR, OR, HR, IRR, PR, and derived or inverted estimates. These measures do not estimate identical causal quantities. We analyzed them on the natural log scale as reported ratio measures and did not recalculate common RRs from participant-level data. Although sensitivity and robustness analyses were directionally consistent, residual interpretational differences across effect-measure types remain possible.

Fourth, most studies came from high-income settings, limiting generalizability to LMICs, where cervical-cancer burden is highest and where vaccination delivery, screening coverage, registry infrastructure, diagnostic pathways, and treatment access may differ substantially [53]. Addressing these gaps with targeted implementation studies will be critical to achieving the WHO 2030 cervical-cancer elimination goals [54]. Finally, the NMA compared vaccine types only indirectly through the common no-vaccination comparator. Vaccine rankings are therefore exploratory and hypothesis-generating rather than definitive comparative effectiveness evidence.

CONCLUSION

HPV vaccination was associated with an approximate 62% average reduction in CIN2+ occurrence across real-world observational studies. The pooled ratio estimate was protective, but between-study heterogeneity was very high, and the prediction interval crossed 1, indicating that program-level effects may vary substantially across settings. Point estimates were generally protective across age-at-vaccination, vaccination coverage, and vaccine-type strata, although subgroup differences were not statistically significant. These findings are consistent with the established role of early adolescent HPV vaccination as a cornerstone of cervical cancer prevention, while also supporting catch-up vaccination and equity-focused implementation. Addressing evidence gaps in LMICs and for newer vaccine types, including the 9-valent vaccine, will be important for advancing cervical-cancer elimination goals. Ongoing monitoring of CIN2+ incidence, vaccination coverage, screening participation, and treatment access will be critical for identifying equity gaps and accelerating progress toward WHO cervical-cancer elimination targets.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.