Impact of six-month COVID-19 travel moratorium on *Plasmodium falciparum* prevalence on Bioko Island, Equatorial Guinea

Plasmodium falciparum malaria remains endemic in many countries, and increasing human mobility complicates control and elimination efforts. On Bioko Island, Equatorial Guinea, malaria burden has declined over two decades, yet several urban areas with substantial travel to the higher-burden mainland continue to experience sustained prevalence. Prior analyses indicated imported infections in returning travelers likely contribute to persistent transmission, but estimates relied on retrospective travel histories and lacked a causal framework to quantify the impact of importation on prevalence. In 2020, a national COVID-19 travel moratorium halted movement between the mainland and Bioko for six months, creating a natural experiment. The study’s research question is to quantify the contribution of imported infections to malaria prevalence by comparing changes in infection prevalence in historically high-travel areas versus low-travel areas before (2019) and after (2020) the moratorium using a difference-in-differences design. This addresses an important evidence gap for designing interventions targeting imported malaria while maintaining local control measures.

Previous studies and modeling on Bioko Island and other settings suggest mobility drives malaria importation and sustains transmission, particularly in urban areas with typically lower baseline prevalence. Simulation models indicated that reducing imported infections between Bioko and the mainland could substantially lower prevalence in high-travel areas. Earlier Bioko analyses linked higher infection risk to residence in areas with more travelers and highlighted human mobility patterns as a major determinant of transmission heterogeneity. However, prior evidence largely relied on cross-sectional survey travel histories and did not directly estimate the prevalence impact of imported cases via an exogenous interruption of travel. The COVID-19 travel restrictions provided the first opportunity to directly evaluate these hypotheses in Bioko.

Design and setting: Natural experiment leveraging a six-month national travel moratorium (March–September 2020) that halted movement between Bioko Island and mainland Equatorial Guinea. A difference-in-differences (DID) approach compared changes in malaria prevalence between historically high-travel and low-travel areas from 2019 (pre-moratorium) to 2020 (post-moratorium). Data sources: Annual Malaria Indicator Surveys (MIS) conducted August–September 2019 and 2020 across Bioko Island, with household-based sampling and individual rapid diagnostic testing (CareStart HRP2/pLDH). Consent and treatment followed national guidelines; data were collected electronically. Classification of travel areas: Historical off-island travel prevalence (reported travel to the mainland in the prior 8 weeks) for 2015–2018 was smoothed by map area using R-INLA methods; EA-level values were derived via weighted averages. Enumeration areas (EAs) in the top quartile of smoothed travel prevalence (≥12.2%) were classified as high-travel; bottom quartile (≤4.3%) as low-travel; middle quartiles were excluded. Sample: In 2019 and 2020, 109 EAs were sampled; 56 fell into high or low travel and were included. Analytic sample comprised 12,128 of 13,195 individuals (92%) with complete covariate data. Survey design accounted for stratification, weights, and clustering at household and EA levels. Outcome: Pf positivity by RDT (binary). Primary analysis: Survey-weighted generalized linear models estimated prevalence by group and year, and DID contrasts measured the change in high-travel areas relative to low-travel areas from 2019 to 2020. Unadjusted and adjusted models were fit; covariates included if they changed meaningfully between 2019 and 2020 within travel groups (absolute change ≥5% and proportional change ≥10%). Final adjusted model covariates: IRS coverage, going indoors before 7 pm, within-island travel (overnight), and household air conditioning. Logistic regression with a logit link estimated odds ratios for relative comparisons. Model specification: Pr(RDT+) = β0 + β1 POST + β2 hightravel + β3 POST×hightravel + β4 Covariates + ε. The DID effect is β3. Validity assessment: Parallel trends were assessed using 2015–2019 data via visual inspection and a generalized linear mixed effects model with year×travel interactions; a bootstrap-like robustness check repeatedly fit models on 80% subsamples (50 iterations). Sensitivity analyses: (1) Excluded three EAs with known land-use changes among seven with substantial ecological changes during the period; (2) Adjusted for an EA-level care-seeking variable (proportion seeking care among those reporting illness); (3) Stratified analyses by urban/rural stratum and a subset comparison of high-travel rural vs low-travel urban EAs to probe potential urban–rural confounding. Software: R v3.6.2; survey package v4.1-1.

• Sample and classification: From 2015–2018, EA-level smoothed travel prevalence ranged 1.5%–39.9%; top quartile (≥12.2%) labeled high travel, bottom quartile (≤4.3%) low travel; 56 EAs included. Analytic sample: 12,128 individuals with complete data. - Prevalence changes: 2019 prevalence: low-travel 7.3% (95% CI 4.5, 10.1), high-travel 13.6% (12.4, 14.9). 2020 prevalence: low-travel 12.8% (7.2, 18.5), high-travel 11.8% (10.0, 13.5). Differences 2020 vs 2019: low-travel +5.5% (0.9, 10.1) unadjusted; +5.8% (0.5, 11.0) adjusted. High-travel −1.9% (−3.2, −0.5) unadjusted; −3.4% (−5.1, −1.8) adjusted. - Difference-in-differences (DID): Unadjusted DID for high-travel vs low-travel: −7.4% (−12.1, −2.6). Adjusted DID: −9.2% (−14.5, −3.9), indicating prevalence in high-travel areas was substantially lower than expected absent the moratorium. - Relative effects (odds): Unadjusted OR for high-travel post vs counterfactual: 0.45 (95% CI 0.29, 0.71); adjusted OR 0.38 (0.22, 0.63). - Covariate shifts 2019→2020: Earlier time indoors increased (low: +2.5%; high: +6%); IRS coverage fell in low-travel (46.9%→35.1%) but rose in high-travel (28.3%→57.2%); within-island travel decreased in both groups; air conditioning increased in high-travel (+6%). Care-seeking decreased in both groups. - Parallel trends: 2015–2019 trends in high vs low travel were similar, with a roughly constant ~7% difference; only 2017 showed a smaller difference, plausibly due to construction-related outbreaks in low-travel EAs. - Sensitivity analyses: Excluding three EAs with land-use change yielded DID unadjusted −5.4% (−9.0, −1.7), adjusted −6.8% (−10.5, −3.1). Adding EA-level care-seeking yielded adjusted DID −9.3% (−14.5, −4.0). Stratified DID: rural −10.4% (−19.5, −1.2); urban −8.3% (−14.8, −1.8). Subset (high-travel rural vs low-travel urban): unadjusted −6.9% (−15.0, 1.5); adjusted −8.5% (−18.1, 1.9).

The travel moratorium provided a quasi-experimental setting to isolate the contribution of imported infections to malaria prevalence on Bioko Island. In historically high-travel areas, prevalence decreased while it increased in low-travel areas over the same period, yielding a significant negative DID effect. This supports the hypothesis that imported infections substantially sustain prevalence in high-travel communities. The adjusted DID of −9.2% and adjusted OR of 0.38 indicate a large reduction in infection risk relative to the expected counterfactual without the moratorium. However, persistence of measurable prevalence and the disappearance of the pre-moratorium risk differential between high- and low-travel areas during the moratorium suggest importation is not the sole driver; local transmission likely persists, potentially amplified by higher receptivity in rural areas. Stratified analyses indicated larger DID effects in rural settings, consistent with greater onward transmission potential from imported cases. Programmatically, results imply that combining interventions targeting imported infections (e.g., traveler screening/prophylaxis or targeted treatment) with sustained local control (IRS, LLINs, larval source management) can synergistically reduce burden, particularly in high-travel communities.

This study provides causal evidence, via a difference-in-differences natural experiment, that imported infections significantly contribute to malaria prevalence in high-travel areas on Bioko Island. The six-month COVID-19 travel moratorium was associated with a 7–9% absolute reduction in prevalence relative to expected trends and a 55–62% reduction in odds of infection in high-travel areas. Nonetheless, residual prevalence during the moratorium underscores ongoing local transmission, especially in receptive rural settings. Future work should integrate granular mobility data, rainfall and ecological change metrics, and molecular or genomic tools to distinguish imported versus locally acquired infections. Evaluations of targeted traveler-focused interventions alongside strengthened local vector control and case management are warranted to accelerate progress toward elimination.

• Parallel trends assumption: DID validity relies on similar pre-trends in high vs low travel areas; while broadly supported (2015–2019), 2017 deviated, potentially due to construction-related outbreaks in low-travel EAs. - Ecological changes: Land use changes and environmental modifications were not precisely measured; while sensitivity analyses excluding known changed EAs attenuated effects slightly, unmeasured ecological shifts could bias estimates. - Rainfall and vector ecology: Lack of granular rainfall data and comprehensive entomological measures may confound temporal changes in transmission potential. - Residual travel and within-island movement: Some off-island travel persisted in 2020 and within-island travel was common; these movements could contribute to ongoing transmission, complicating attribution solely to importation. - Care-seeking behavior: Decreased care-seeking in both groups could elevate prevalence by reducing treatment rates; inclusion of an EA-level care-seeking covariate did not materially change estimates but may not capture individual-level effects. - EA classification and spillover: Potential misclassification at EA borders and mosquito movement between adjacent high- and low-travel EAs could cause spillover; though not supported by observed trends, this remains a possibility. - Missing covariate data: The adjusted analysis excluded individuals with missing covariates (8%), which could introduce selection bias if missingness was differential.