Perceived barriers to Aedes aegypti larval habitat control among surveyed residents in five Panamanian districts: a cross-sectional health belief model analysis

1 Introduction

Dengue is a rapidly expanding mosquito-borne viral disease, and Aedes aegypti control remains essential where multiple serotypes circulate. The Americas experienced unprecedented transmission during 2024–2025, and contemporary surveillance continues to emphasize early detection, integrated vector management, and community participation (1, 2). Panama reported more than 32,000 dengue cases during the 2024 outbreak, while 25 years of national surveillance demonstrate marked spatial heterogeneity in incidence, reporting, hospitalization, and inferred transmission intensity (3). These patterns make local context central to prevention.

Source reduction in and around homes depends on more than awareness. Water storage, waste management, infrastructure, access to public services, vector-control activity, risk communication, and residents’ opportunity to act can shape larval-habitat control. Systematic evidence from Latin America and other endemic settings suggests that community engagement and education may improve entomological and behavioral outcomes, but implementation, sustainability, and contextual barriers vary (4–6).

The Health Belief Model (HBM) organizes perceived susceptibility, severity, benefits, barriers, self-efficacy, and cues to action as factors that may influence health decisions (7, 8). Dengue surveys have applied the HBM and knowledge-attitudes-practices approaches to understand household prevention (9–14). However, favorable attitudes and stated intention do not necessarily indicate routine practice; a recent meta-analysis found a substantially higher prevalence of positive attitudes than of good preventive practices (15). Self-reported responses may also be affected by acquiescence, social desirability, and scale ceiling effects.

Evidence from Panama and other Latin American and Caribbean settings further indicates that preventive practices are shaped by sociocultural context, community participation, risk communication, perceived opportunity, and the framing of public-health messages (16–20).

Panama lacks a large, district-comparative analysis of perceived barriers to household larval-habitat control. The central contribution of this study is therefore epidemiological rather than merely psychometric: it quantifies contextual heterogeneity in perceived barriers among surveyed residents of five districts despite near-universal intention to remove observed breeding sites. Psychometric analyses are used to characterize the measurement instrument and the uncertainty surrounding its scores.

We aimed to describe HBM perceptions, knowledge, self-efficacy, and behavioral intention; estimate adjusted associations of district, essential-service access, knowledge, and sociodemographic characteristics with perceived barriers; and assess whether conclusions were robust to ordinal modeling and disaggregation of the services index. The survey did not measure the frequency or observation of actual mosquito-control behavior.

2 Materials and methods

2.1 Study design, setting, and data source

We conducted a secondary cross-sectional analysis of the Panama-adapted MosquitoWise survey. Data were collected from October 2024 through February 2025 in the districts of Panamá, San Miguelito, Chepo, Arraiján, and Chitré. The source files contained 2,115 records. The questionnaire and a bilingual item-by-item mapping are provided in Supplementary Table S2.

2.2 Recruitment, participants, and analytical sample

Participants were recruited through non-probability convenience sampling in community settings because no comprehensive sampling frame was available. Eligible respondents were residents aged at least 15 years who had lived in the community for at least one year and agreed to anonymous use of their information. Only one participant was enrolled per household: the person responsible for checking for or eliminating larval habitats.

The archived secondary dataset does not retain exact recruitment-site identifiers, venue type, dates or periods by site, identity or cadre of recruiters, number approached before the survey, or documentation showing that procedures were identical across districts. Twenty-one respondents declined data use; this is the only recorded refusal count. Recruitment-site clustering could not be modeled because site identifiers were unavailable. Within-household clustering was not expected because the design allowed one respondent per household.

We excluded 21 records without data-use agreement and three records from residents with less than one year in the community, leaving 2,091 eligible participants. The adjusted models used 2,079 complete cases because education was missing for 12 respondents.

2.3 Questionnaire, constructs, and scoring

The Spanish questionnaire included sociodemographic, housing, service-access, geographic, HBM, knowledge, care-seeking, and intention items. HBM statements used three response categories (1=disagree, 2=indifferent, 3=agree). Susceptibility (three items; 3–9), severity (three; 3–9), benefits (four; 4–12), barriers (three; 3–9), self-efficacy (two; 2–6), and cues to action (two; 2–6) were unweighted sums. Higher scores indicated more of the named construct; the barrier items were negatively framed, so higher barrier scores indicated more perceived barriers without reverse scoring.

The behavioral-intention item asked whether an observed breeding site should be eliminated and was described as a single ordinal item; it was not treated as observed participation. Breeding-site knowledge and care-seeking knowledge each summed three correct/incorrect items (0–3). The essential-services index summed access to potable water, sewerage, electricity, and garbage collection (0–4; yes=1). No data-derived high/low cutoffs were used.

The source instrument was administered in Spanish. The English translation supplied with this revision was prepared from the original Spanish questionnaire for reproducibility. Archived materials did not document a formal forward/back translation, expert-panel reconciliation, cognitive interviews, or pilot linguistic validation; therefore, we do not claim that the bilingual version is a formally validated cross-cultural instrument (21).

2.4 Outcome hierarchy

The designation of perceived barriers as the primary analytical outcome was introduced during manuscript revision after inspection of the outcome distributions; it was not prespecified in the field protocol. Barriers retained sufficient variability for adjusted analysis, whereas self-efficacy showed an 88.4% ceiling. Self-efficacy was therefore treated as an exploratory secondary outcome, and behavioral intention remained descriptive. This revision-stage hierarchy limits multiplicity but does not convert the primary analysis into a prospectively prespecified test.

2.5 Psychometric evaluation

We evaluated internal consistency using Cronbach alpha, ordinal alpha from polychoric correlations, composite omega, and Spearman–Brown coefficients for two-item constructs. After a reproducible random split, parallel analysis and minimum-residual exploratory factor analysis with oblique rotation were conducted in one half. A six-factor confirmatory model was evaluated in the other half using weighted least squares with mean- and variance-adjusted estimation. These analyses characterize performance in this sample and are not evidence of full external validation.

2.6 Statistical analysis

We summarized continuous variables with means and standard deviations or medians and interquartile ranges and categorical variables with counts and percentages. The primary multivariable linear regression estimated adjusted mean differences in the barriers score. Covariates were age group, sex, completed years of education, marital status, district, essential-services score, breeding-site knowledge, and care-seeking knowledge. District was modeled as a fixed effect to capture observed and unobserved contextual differences. The exploratory self-efficacy model used the same covariates.

HC3 heteroscedasticity-robust standard errors were used because the Breusch–Pagan test indicated heteroscedasticity. We report 95% confidence intervals, two-sided p values, and Benjamini–Hochberg false-discovery-rate q values. The primary complete-case model included 2,079 respondents. Sensitivity analyses omitted education to include all 2,091 respondents, replaced district with province, removed cases with Cook’s distance greater than 4/n, and fitted cumulative-logit models to ordered scores.

For the barriers cumulative-logit model, we tested the proportional-odds assumption using nominal-effect likelihood-ratio tests. Because the assumption was not supported for education, district, and breeding-site knowledge, common odds ratios are presented as exploratory robustness summaries rather than definitive effect estimates. We assessed whether the directions and substantive pattern of district, service, and knowledge associations agreed with the linear model.

To examine the unexpected essential-services association, we replaced the 0–4 index with the four binary components in a mutually adjusted exploratory model while retaining the primary covariates. These component results are descriptive and potentially affected by sparse exposure categories, measurement limitations, residual confounding, and contextual differences. Analyses used R version 4.6.1; the reproducible script, session information, analytical dataset, and output tables accompany the revision.

2.7 Ethics

The present study was a secondary analysis of a deidentified database and was approved by the Comité de Bioética de la Investigación (UMECIT) and the Ministry of Health of the Republic of Panama. Under the Panamanian regulations applicable to secondary research using deidentified data, written informed consent was not required for this analysis, including for records from participants aged 15–17 years; separate written permission from a parent or guardian was likewise not required for the secondary analysis. The original survey recorded voluntary agreement to participate and to anonymous governmental use of information at its start. The ethics committee reviewed and approved the secondary-data design, inclusion of records from participants aged 15–17 years, and the applicable consent procedures. No identifiable images or data are reported.

2.8 Use of generative artificial intelligence

The corresponding author used OpenAI Codex (GPT-5 model, OpenAI, San Francisco, CA, United States; accessed August 2026) to assist with manuscript restructuring, language editing, drafting of reproducible R code, and preparation of document files. The analytical source data were not collected by the AI tool. All AI-assisted text, references, code, figures, and numerical results were reviewed and verified by the authors; the authors retain full responsibility for accuracy, originality, analysis decisions, and conclusions.

3 Results

3.1 Participant characteristics

Among 2,091 eligible participants, mean age was 46.4 years (SD 16.6), 61.2% were female, and 62.7% were married or cohabiting. Panamá district contributed 37.4%, San Miguelito 24.4%, Chepo 16.0%, Arraiján 6.4%, and Chitré 15.8%. Potable water, sewer, electricity, and garbage-collection access were reported by 91.4%, 68.0%, 98.9%, and 88.1%, respectively (Table 1).

Table 1. Characteristics of the eligible analytical sample (N = 2,091).

Values are n (%) unless otherwise specified. Available denominators: education 2,079; household size 2,089; community residence 2,064; optional 2024 dengue diagnosis 2,070.

3.2 Distribution of HBM and related constructs

The barriers score averaged 5.24 (SD 2.34); 42.0% of participants were at its theoretical minimum and 19.6% at its maximum. Self-efficacy averaged 5.77 (SD 0.69), with 88.4% at the maximum. Severity, benefits, cues to action, and behavioral intention also had substantial ceiling effects. Overall, 2,021 participants (96.7%) agreed that they should eliminate observed breeding sites, but the survey did not measure whether or how often they did so (Table 2; Supplementary Figure S1).

Table 2. Distribution of HBM and related scores (N = 2,091).

Floor and ceiling denote theoretical minima and maxima. Behavioral intention is one ordinal item and is not measured participation.

3.3 Psychometric properties

Parallel analysis retained six factors in the exploratory half-sample (n=1,045). In the confirmatory half (n=1,046), the six-factor model had CFI 0.997, TLI 0.996, RMSEA 0.032 (95% CI 0.026–0.038), and SRMR 0.079. Ordinal reliability was 0.84–0.91 across constructs, although conventional alpha was modest for ceiling-concentrated two-item scales. Two severity items were nearly redundant (polychoric r=0.976). These findings support cautious within-sample use but not a claim of full instrument validation (Table 3).

Table 3. Reliability and construct-validity summary.

ω=composite omega; SB, Spearman–Brown coefficient; AVE, verage variance extracted. Six-factor CFA: CFI 0.997; TLI 0.996; RMSEA 0.032 (0.026–0.038); SRMR 0.079.

3.4 Factors associated with perceived barriers

The primary model included 2,079 complete cases and explained 14.8% of the barriers-score variance. Compared with Panamá district, the adjusted mean was 0.98 points lower in San Miguelito (95% CI −1.28 to −0.68), 1.10 points higher in Chepo (0.80–1.41), and 1.52 points higher in Arraiján (1.09–1.96); Chitré did not differ clearly (−0.10, −0.40 to 0.20). Relative to the barriers-score SD, the Arraiján, Chepo, and San Miguelito contrasts corresponded to approximately 0.65, 0.47, and −0.42 SD, respectively.

Each one-point increase in breeding-site knowledge was associated with 0.17 fewer barrier points (95% CI −0.28 to −0.06), and each one-point increase in care-seeking knowledge with 0.51 fewer points (−0.63 to −0.38). The district contrasts were therefore larger than the per-point breeding-knowledge association and comparable to or larger than the care-seeking association. Essential-service access was associated with more barriers (0.36 per service, 0.21–0.51). Other demographic estimates were small or imprecise after false-discovery-rate adjustment (Table 4; Figure 1).

Table 4. Adjusted associations with perceived barriers and exploratory self-efficacy.

AMD, adjusted mean difference; CI, confidence interval; q=Benjamini–Hochberg false-discovery-rate value. Both models used the full covariate set, complete cases (n=2,079), and HC3 robust standard errors.

3.5 Exploratory analysis of service components

When the four service indicators replaced the aggregate index in one mutually adjusted model, sewer access was associated with a 0.67-point higher barriers score (95% CI 0.45–0.90; q<0.001). Potable water (0.22, −0.17 to 0.60), electricity (−0.01, −1.08 to 1.05), and garbage collection (0.03, −0.30 to 0.35) did not show clear associations. The result suggests that the aggregate association was primarily carried by sewer access in these data, but it does not establish a causal infrastructure effect (Supplementary Table S6 and Supplementary Figure S2).

3.6 Factors associated with self-efficacy

Because 88.4% of respondents had the maximum self-efficacy score, adjusted differences were interpreted as exploratory and of uncertain practical importance. Female participants had a 0.08-point lower adjusted mean than male participants (95% CI −0.14 to −0.02). Chepo was 0.10 points higher than Panamá (0.03–0.17), and Arraiján was 0.19 points lower (−0.34 to −0.04). Breeding-site and care-seeking knowledge were associated with 0.12 (0.08–0.15) and 0.06 (0.02–0.11) higher points per index point. These magnitudes are small relative to the severe ceiling and should not be interpreted as demonstrated behavioral differences.

3.7 Sensitivity analyses

Cumulative-logit models produced the same directions for all principal district, service, and knowledge associations: San Miguelito OR 0.44, Chepo OR 2.45, Arraiján OR 3.31, essential services OR 1.32, breeding-site knowledge OR 0.85, and care-seeking knowledge OR 0.67 for a higher barriers category. However, nominal-effect tests indicated possible proportional-odds violations for education (p=0.003), district (p=0.013), and breeding-site knowledge (p<0.001); ordinal odds ratios are therefore robustness summaries. Omitting education, substituting province, and excluding 66 influential observations did not overturn the central pattern of district and knowledge associations (Supplementary Tables S7-S11).

4 Discussion

The principal finding is substantial district-level heterogeneity in perceived barriers among surveyed residents despite near-universal stated intention to eliminate breeding sites. Differences of approximately one to one-and-a-half points on a 3–9 scale were larger than most demographic associations and the per-point association with breeding-site knowledge. This contextual signal is the study’s main contribution beyond applying an HBM questionnaire: willingness was high, yet perceived feasibility varied sharply across districts.

These district coefficients must not be interpreted as causal effects of residence. They may reflect infrastructure, environmental conditions, service reliability, water storage, vector-control operations, local communication, recruitment venue, interviewer administration, or other unmeasured characteristics. Contemporary Panamanian surveillance also demonstrates spatial heterogeneity in dengue transmission and reporting (3), and regional evidence emphasizes interacting ecological, biological, household, community, and structural determinants (6). The observed differences justify district-specific investigation rather than ranking districts or generalizing to their populations.

The positive association between the services index and perceived barriers was counterintuitive. Disaggregation showed that sewer access carried most of this association, while water, electricity, and garbage collection estimates were imprecise or near zero. Sewer access may be a marker of urban form, population density, drainage, tenure, or other contextual features not measured by the survey. Service presence also does not measure continuity or quality. Residual confounding, district composition, sparse categories, and selection into recruitment sites remain plausible; the result should not be read as evidence that infrastructure increases barriers.

Greater breeding-site and care-seeking knowledge were associated with fewer barriers. This may indicate that respondents who understand mosquito ecology and appropriate care perceive prevention as more feasible. Nevertheless, knowledge alone is rarely sufficient for sustained source reduction. Reviews of Aedes control and community participation identify multilevel implementation barriers, and KAP evidence shows that favorable attitudes can coexist with weak practice (4, 5, 15). Because this survey measured neither frequency nor observation of larval-habitat control, it cannot test whether knowledge or beliefs translated into action.

Self-efficacy and several other constructs showed severe ceiling effects. Differences of 0.08–0.19 points may be statistically detectable in a large sample yet behaviorally trivial and poorly discriminated by the scale. Acquiescence and social-desirability processes may also inflate agreement. Future instruments should use documented translation and cultural adaptation, cognitive testing, more discriminating response options, and items that separate confidence from actual behavior (21). Direct observation, household inspections, repeated diaries, or frequency-based practice measures would better evaluate the intention–behavior gap.

Public-health implications follow from the contextual pattern rather than a causal prescription. Local programs could combine district-specific qualitative work, service-reliability assessment, community mapping, and observed household practices before tailoring communication or source-reduction support. Knowledge messages should be paired with practical opportunity, reliable services, and sustained community engagement. Probability-based or clearly enumerated sampling, recorded recruitment-site identifiers, longitudinal follow-up, and multilevel analyses would permit stronger population and contextual inference.

4.1 Strengths and limitations

Strengths include a large multi-district dataset, explicit scoring, split-sample psychometric evaluation, robust standard errors, presentation of effect sizes, disaggregation of service components, full ordinal-model results, proportional-odds diagnostics, and reproducible analytical materials. The bilingual questionnaire and item mapping improve transparency.

The study also has important limitations. Convenience sampling and an unknown approach denominator create selection bias and prevent population prevalence estimates. Exact recruitment venues, recruiters, schedules, and site identifiers were unavailable, so recruitment-site clustering could not be modeled. Although only one respondent was enrolled per household, community-level dependence may remain. Cross-sectional data cannot establish temporality or causality. All substantive measures were self-reported, actual preventive behavior was not measured, and several scales had floor or ceiling concentration. The instrument lacked a documented formal translation/adaptation protocol and external validation. Knowledge indices cover distinct content and are better interpreted as correct-answer counts than reflective scales. Proportional-odds assumptions were not satisfied for all predictors. Finally, district coefficients may combine context with differential recruitment or administration.

5 Conclusion

Among participants recruited by convenience in five Panamanian districts, perceived barriers to larval-habitat control varied substantially by district and were lower with greater dengue-related knowledge. Although nearly all participants reported an intention to eliminate observed breeding sites, intention was not a measure of action. The findings identify contextual hypotheses for local investigation and tailored prevention; they do not support representative, behavioral, or causal conclusions.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The corresponding author used OpenAI Codex (GPT-5 model, OpenAI, San Francisco, CA, United States; accessed August 2026) to assist with manuscript restructuring, language editing, drafting of reproducible R code, and preparation of document files. The authors reviewed and verified all AI-assisted text, references, code, figures, and numerical results and retain full responsibility for the work.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fitd.2026.1864749/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Allassane Foungoye Ouattara, University of Nangui Abrogoua, Côte d’Ivoire

Reviewed by: Moises Apolaya Segura, Scientific University of the South, Peru

Zahra Montaseri, Fasa University of Medical Sciences, Iran