Meta-analysis and Time Series Projection of Human Papillomavirus Prevalence Among Women in Honduras (1990-2023)

2.1 Study design

This study utilized a two-phase methodology: (1) a systematic review and meta-analysis of studies on HPV prevalence, and (2) a time series forecasting of age-specific prevalence trends. The systematic review was executed and documented in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines.⁶ The completed PRISMA 2020 checklist is available at Zenodo.⁷ This research entailed secondary analysis of published aggregate data and did not necessitate ethics committee approval. No individual patient data was accessed or analyzed.

2.2 Search strategy and information sources

We conducted a comprehensive literature search in four electronic databases: PubMed,⁸ Google Scholar,⁹ Tz’ibalnaah (National Autonomous University of Honduras repository),¹⁰ and SciELO (Scientific Electronic Library Online).¹¹ The search took place in January 2025 and included works published between January 1990 and December 2024.

Search terms employed were: “HPV Honduras”, “human papillomavirus Honduras”, “papilomavirus humano Honduras”, and “VPH Honduras” without language restrictions. The search strategy was deliberately broad to maximize sensitivity and minimize the risk of missing relevant studies. We manually searched the reference lists of included studies to identify additional relevant publications. All database searches were documented with dates and results counts to ensure reproducibility. The complete search strategy is available in the PRISMA checklist.

To address potential retrieval bias, we included both international databases (PubMed, Google Scholar, SciELO) and a local Honduran repository (Tz’ibalnaah) to capture studies that might not be indexed in major international databases. This approach aimed to minimize geographic and language-related publication bias common in Latin American epidemiological research.

2.3 Eligibility criteria

Studies were eligible for inclusion if they met all the following criteria:

Exclusion criteria were:

2.4 Study selection and data collection

Two independent reviewers (MM, AM) screened all titles and abstracts identified in the database searches. Studies flagged as potentially eligible by either reviewer proceeded to full-text evaluation. Full-text articles were independently assessed by both reviewers using a standardized eligibility form. Disagreements regarding inclusion were resolved through discussion and, if necessary, arbitration by a third reviewer (SD).

Data extraction was performed independently by two reviewers using a standardized, pilot-tested data extraction form. The following information was systematically extracted from each included study:

Discrepancies in extracted data were resolved by consensus between extractors. When information was unclear or missing, we attempted to contact study authors. If authors could not be reached or data could not be clarified, we used conservative assumptions documented in our analysis.

2.5 Population-based prevalence adjustment

To enable comparison of prevalence rates across studies conducted in different years and settings, we calculated population-adjusted prevalence rates per 10,000 inhabitants. Age-specific population denominators were obtained from the National Statistics Institute of Honduras (INE) Population and Housing Census.¹² For each study, we selected census data from the year closest to the study midpoint.

This adjustment approach has inherent limitations; these should be considered when interpreting population-adjusted prevalence estimates:

2.6 Quality assessment

We assessed study quality using the following criteria adapted for prevalence studies:

Due to the limited number of studies meeting basic inclusion criteria (n=4), we included all studies that met minimal methodological standards rather than excluding studies based on quality scores. However, we noted quality concerns and incorporated them into our interpretation of results and assessment of evidence certainly.

Several sources of bias were identified and addressed:

2.7 Statistical analysis

Meta-analysis of proportions was conducted using MedCalc statistical software v19.7.1 (64-bit).¹³ We calculated pooled prevalence estimates with 95% confidence intervals (CI) using random-effects models (DerSimonian-Laird method), anticipating heterogeneity across studies. Random-effects models were chosen a priori given expected variability in study populations, settings, diagnostic methods, and time periods. Statistical heterogeneity was quantified using the I² statistic and Cochran’s Q test. I² values of 25%, 50%, and 75% were interpreted as representing low, moderate, and high heterogeneity, respectively. Forest plots were generated to visually present individual study estimates with confidence intervals and the pooled effect estimate. Time series analysis and forecasting were performed using Python programming language v3.10.3 with the following libraries: NumPy v1.18 (numerical computing), Pandas v2.1 (data manipulation), SciPy v1.10.1 (scientific computing), Matplotlib v3.7 (visualization), and Seaborn v0.12.2 (statistical graphics).^14,15 The Holt-Winters exponential smoothing method¹⁶ was employed for prevalence forecasting from 2025 to 2035. This method decomposes time series into three components: level (average value), trend (directional change), and seasonality (periodic fluctuation). Given our data structure, comprising only four temporal observations at irregular intervals without cyclical patterns; we applied the non-seasonal variant focusing exclusively on level and trend components. The method employs exponential smoothing with two smoothing parameters: α (level smoothing, range 0-1) and β (trend smoothing, range 0-1). These parameters were optimized automatically using least-squares minimization to fit historical data. Forecasts were generated for each age group independently.

Critical limitations of the forecasting approach:

Given these limitations, forecasts should be interpreted as illustrative of general directional trends (increasing vs. decreasing) rather than precise point estimates. Confidence intervals for forecasts were not calculated due to insufficient data points, further limiting interpretability. We recommend viewing projections as hypothesis-generating rather than definitive predictions suitable for policy planning.

Forecast accuracy was evaluated using root mean square error (RMSE) and mean absolute error (MAE),^17,18 with following formulas:

Lower RMSE and MAE values indicate superior model fit. RMSE penalizes larger errors more heavily, while MAE provides an average absolute deviation metric. Both metrics are reported in prevalence percentage units for interpretability. All analyses were conducted in a Microsoft Windows 10 (64-bit) environment.^13–15 Statistical significance was set at α = 0.05 for all tests. Analysis scripts are available upon reasonable request to promote reproducibility.

2.8 Data availability and reproductibility

All data underlying the results are available as part of the article and no additional source data is required. The four primary studies included in the meta-analysis are publicly available through peer-reviewed publications: Ferrera et al. 1999,¹⁹ Tabora et al. 2009,²⁰ Avilez et al. 2017,²¹ and Montoya 2023.¹

Population denominator data were obtained from publicly accessible census records maintained by the National Statistics Institute of Honduras (INE).¹² Age-specific population estimates used for prevalence rate calculations are provided in supplementary materials to facilitate replication.

Python analysis scripts, data extraction forms, and complete search documentation are available from the corresponding author upon reasonable request. The PRISMA 2020 checklist with item-by-item reporting locations has been deposited in Zenodo with DOI: 10.5281/zenodo.17429814 under a CC0 1.0 Universal (CC0 1.0) Public Domain Dedication license.⁷

4.1 Interpretation of age-specific trends

Among younger women (15–34 years), the sharp rise, particularly in those aged 15–24 years, from 0% in 2017 to 10.65% in 2023; it is concerning and may reflect declining vaccination coverage following health-system disruptions; earlier sexual debut in more recent cohorts; reduced awareness of STI prevention due to waning public health campaigns; expanded access to diagnostic services that is detecting previously missed infections; and random variability related to small sample sizes in the most recent study.⁵ In contrast, the sustained decline among women ≥35 years likely reflects the cumulative impact of STI prevention campaigns implemented during 1980–2000; cohort-level benefits of HPV vaccination (with today’s 35–44-year-olds having been adolescents or young adults at program initiation); age-related changes in sexual behavior and partner numbers; natural immune clearance of infections acquired earlier in life; and cohort effects arising from differential exposure to risk factors.

4.2 Comparison with regional and global Data

Our findings correspond with regional trends indicating increased prevalence in the age groups under 34 and over 55 years, and decreased rates in intermediate ages, when juxtaposed with HPV Information Centre data for Central America.^22,23 Nonetheless, direct comparison is constrained by variations in diagnostic methodologies, study cohorts, and HPV type-specific detection.

When compared to official SESAL data ( Figure 5), there is some agreement, especially for the 25–34 age group, but not for other age groups. SESAL projections indicate a greater prevalence in the 35-54 age demographic compared to our meta-analysis predictions, potentially highlighting discrepancies between clinical reporting systems (SESAL) and research study samples.

Figure 5. Comparison of meta-analysis versus SESAL administrative data projections (2025-2035), Comparison of age-specific projections from research-based meta-analysis (solid lines) versus Honduras Ministry of Health (SESAL) administrative surveillance data (dashed lines).

Agreement observed for 25-34 age group projections. Divergence is evident for other age groups, particularly 15-24 (meta-analysis shows steep rise, SESAL shows stability) and 35-54 (meta-analysis predicts decline, SESAL predicts persistence). Discrepancies likely reflect differences in data sources, populations, and detection methods.

4.3 Methodological limitations and forecast reliability

The main methodological problem is that there aren’t enough time observations, only four (1999, 2009, 2017, and 2023) for time-series forecasting. Standard practice says that there should be at least 10–20 observations for stable trend estimation. With such limited data, it is impossible to tell the difference between short-term stochastic variation and long-term trends. Parameter estimates are very uncertain, projections beyond 2–3 years become less reliable, and the model can’t handle non-linear dynamics or structural breaks. The appearance of negative prevalence forecasts for individuals aged 35 years and older further illustrates model misspecification; prevalence cannot decrease below zero, and these figures represent artifacts of linear extrapolation approaching a low asymptote. This pattern shows that we need bounded models that set natural limits (0–100%), that Holt-Winters is not a good fit for behavior close to zero, and that long-horizon projections (2030–2035) are hard to understand.

In addition to forecasting constraints, significant heterogeneity between studies (I² = 96.8%) restricts the generalizability of pooled estimates. This heterogeneity is caused by differences in diagnostic methods (clinical vs. PCR), study populations (clinic-based vs. community-based), geographic settings (e.g., Tegucigalpa vs. Danlí), changes in HPV epidemiology and testing practices over time, and sample sizes (100 to 2,148 participants). There is also a possibility of publication and selection bias due to the limited number of published studies and the tendency to favor reporting positive results or specific age groups, while some clinic-based studies remain in institutional reports. Lastly, population adjustments based on national census data assume that the population is evenly spread out, which may not be true for the populations at risk in the study areas.

The forecasting model produced biologically implausible negative prevalence estimates for older age groups (35+) in later projection years (2030-2035). These negative values represent artifacts of linear trend extrapolation when data approach lower bounds and indicate model misspecification rather than true predictions. In practice, prevalence in these age groups is expected to stabilize near zero rather than become negative. This artifact underscores the limitations of applying Holt-Winters exponential smoothing to sparse temporal data with only four observations and highlights the need for bounded forecasting approaches (e.g., logistic models, exponential decay) when modeling processes with natural constraints.

4.4 Implications for Honduran Public Health Policy

Even with methodological flaws, these results have clear implications for policy. The sharp rise in HPV cases among women aged 15 to 24 years necessitates an immediate escalation of HPV vaccination efforts, preferably before the initiation of sexual activity, coupled with thorough school- and community-based sexual health education that highlights prevention and the advantages of vaccination. To enhance decision-making, a systematic annual surveillance system employing standardized age strata is essential for facilitating comprehensive trend analysis and forecasting. The ongoing declines seen in older women show that past prevention efforts have worked, which means that more money should be spent on broad STI control programs. Finally, it is important to strengthen the resilience of the health system by fixing the weaknesses that were shown during times of political instability (for example, the 2009 constitutional crisis (curfew)²⁴ and the failed 2014 health reform²⁵) to keep public health gains.

4.5 Sociopolitical context

Honduras has experienced periods of political and economic instability with direct and persistent effects on the delivery of health services. The constitutional crisis of 2009 messed up public health programs and messed up supply chains, vaccination schedules, and the normal running of health facilities.²⁴ In 2014, efforts to reform health care didn’t fix the problems with financing, governance, and stewardship that were built into the system.²⁵ These shocks probably led to inconsistent vaccination coverage, less screening and treatment for STIs, a smaller budget for prevention campaigns, and a loss of skilled workers due to migration and staff changes. At the operational level, higher opportunity costs for users, restrictions on mobility and security, and a more informal labor force may have reduced the demand for preventive services, while staff turnover and institutional fragmentation compromised the continuity of interventions.

These contextual conditions provide a reasonable basis for elucidating the observed epidemiological trends: increases or variations in prevalence among young women may signify accumulated deficiencies in primary and secondary prevention, generational shifts in sexual behavior, and inequitable access to diagnosis, alongside methodological discrepancies across studies. The decline in women aged ≥35 years may indicate delayed effects of prior interventions and immunological clearance, as well as potential biases stemming from the selection of the served population or alterations in test sensitivity. The COVID-19 pandemic and other macroeconomic shocks, along with the fact that systems are different, mean that we should be careful when trying to figure out trends.²⁶ This shows how important it is to have standardized periodic surveillance by age group, better information systems, and evaluations of implementation. To keep up the progress and close the gaps in HPV prevention and control, it’s important to come up with strong strategies that include protected financing, strong supply chains, keeping staff, and working together across sectors.²⁷

4.6 Recommendations for future research

Research priorities should encompass prospective cohorts with age-stratified follow-up to elucidate the natural history and persistence of HPV infection; type-specific genotyping analyses that differentiate oncogenic from non-oncogenic variants and facilitate the estimation of vaccination impact; and a standardized surveillance system featuring annual surveys, uniform sampling frames and laboratory methods, and consistent age bands to enhance the comparability and validity of trends.^28,29 Simultaneously, mixed-methods studies investigating barriers to vaccination and healthcare access would inform the optimization of implementation, while comprehensive economic evaluations contrasting expanded vaccination with diverse screening strategies would facilitate effective resource allocation. Finally, it is important to have multicenter collaborations at the Central American level to understand how diseases spread in the region and how they vary from place to place.