The researchers analyzed the 2018/2019 Uganda Malaria Indicator Survey, which used a two-stage stratified cluster design covering 15 subregions and nine refugee settlements. Data were collected from 320 clusters in the subregions and 22 clusters in refugee settlements between December 2018 and January 2019. The study included 8,868 women aged 15–49 from selected households: 8,231 non-refugee women and 637 refugees. Survey weights were applied to descriptive statistics and statistical models, while cluster-robust inference accounted for the primary sampling units. The analysis therefore addressed a national cross-sectional snapshot rather than a longitudinal intervention or a randomized campaign evaluation.

Among the 8,868 participants, 6,143, or 69.3% with a 95% confidence interval of 67.2% to 71.2%, reported sleeping under an insecticide-treated net on the night before the survey. A total of 3,334 women, or 37.6% with a 95% confidence interval of 35.5% to 39.7%, reported seeing or hearing a malaria message during the preceding six months. Comprehensive malaria knowledge was recorded for 5,427 women, or 61.2% with a 95% confidence interval of 58.7% to 63.6%. The weighted sample was predominantly rural: 5,867 women, or 66.2%, lived in rural areas, while 2,364, or 26.7%, lived in urban areas and 637, or 7.2%, were refugees. Western Uganda accounted for 32.9% of the sample and the eastern region for 30.8%.

Comprehensive knowledge was a strict composite measure requiring correct answers to all seven questions. These covered mosquito bites or mosquitoes as a cause of malaria, the possibility of avoiding malaria, whether sleeping under an ITN was the best way to prevent it, whether pregnant women should take antimalarial medication, whether a child with fever should visit a health provider the same or next day, and whether treatment for a child with fever should be sought the same day. Individual response levels were high: 93.6% said malaria could be avoided, 94.3% agreed that sleeping under an ITN was the best preventive method, 91.3% said pregnant women should take antimalarial drugs, 96.6% said a child with fever should be taken to a provider on the same or next day, and 91.4% selected same-day treatment. The requirement to answer every item correctly meant that 38.8% nevertheless fell short of the study’s comprehensive-knowledge definition.

To partition the association, the study used a counterfactual causal mediation framework with two logistic regression models: one for the mediator, comprehensive knowledge, and one for the outcome, ITN use. The total effect represented the modeled difference in ITN use if all women were exposed to malaria messages rather than none. The natural direct effect captured the portion not operating through knowledge, while the natural indirect effect represented the portion transmitted specifically through the mediator. Both models incorporated the survey weights and adjusted for age, education, total births, wealth quintile, sex and age of the household head, household membership, number of children under five, residence type and region. The models also used cluster-robust inference to reflect the survey sampling structure.

The estimated total effect of malaria-message exposure on ITN use was 0.065, equivalent to 6.5 percentage points, with a 95% confidence interval of 0.034 to 0.094 and p < 0.001. The natural direct effect was 0.064, with a 95% confidence interval of 0.034 to 0.092 and p < 0.001, indicating that the association was not reduced appreciably when the knowledge-mediated component was separated from the model. The natural indirect effect through comprehensive knowledge was 0.001, with a 95% confidence interval spanning zero. The corresponding estimated mediated proportion was 0.015, with a 95% confidence interval of −0.033 to 0.078 and p = 0.462. These estimates do not establish that knowledge is unrelated to net use in every context; rather, they did not detect a statistically significant knowledge-mediated component in this population under the model’s assumptions.

The authors said the pattern is consistent with communication affecting net use through influences other than changes in the study’s comprehensive-knowledge measure. Possibilities include perceived risk, reminders, social norms, motivation and behavioral cues, none of which was measured individually as a mediator. The results also fit the high baseline knowledge levels in the sample: more than 90% of women endorsed several major statements about malaria prevention and treatment, leaving relatively little room for additional messages to move the strict composite measure. In such a setting, communications could reinforce an existing practice without producing a detectable shift in comprehensive knowledge. The direct estimate should therefore not be interpreted as proof of one particular behavioral mechanism, and the analysis cannot determine which individual message components were responsible for the observed association.

Causal interpretation depended on several assumptions that could not all be tested directly. The framework required no unmeasured confounding between message exposure and ITN use, between exposure and knowledge, or between knowledge and ITN use after adjustment for the measured covariates. It also required that no confounder of the mediator–outcome relationship had itself been affected by message exposure. ITN access was deliberately excluded from the adjustment set because messages might influence net acquisition, or because campaigns could coincide with net distribution; controlling for access could therefore block part of the total effect and change the quantity being estimated. As a result, the natural direct effect includes all modeled pathways not operating through comprehensive knowledge, including potential pathways involving access to nets, rather than isolating a purely psychological or behavioral mechanism.

The study has additional limitations imposed by its cross-sectional design and self-reported measures. Because all three central variables were assessed within the same survey, the temporal sequence among hearing a message, acquiring knowledge and using a net could not be established. Participants might also have misreported message exposure or net use, creating recall or measurement error. Residual confounding remains possible, including the influence of pre-existing net access on both knowledge and use. An E-value analysis suggested moderate rather than complete robustness to unmeasured confounding: an unmeasured factor would have needed risk-ratio associations of at least 1.640 with both exposure and outcome, conditional on the observed covariates, to explain the association fully. To shift the confidence interval to include the null, the corresponding minimum association with each would have been 1.491.

For malaria-control communication, the results indicate that informational content alone may not capture how messages relate to preventive behavior. The authors recommend combining information with motivational and behavioral reinforcement, while also ensuring access to insecticide-treated nets and addressing contextual barriers to consistent use. Those recommendations are interpretations of the mediation pattern rather than interventions tested in this analysis. The study’s narrowest conclusion is that, among Ugandan women in the 2018/2019 national survey, recent exposure to malaria messages was associated with greater ITN use, while comprehensive malaria knowledge accounted for a small and statistically indistinguishable share of the modeled total association.

Subject of Research: Causal mediation of malaria-message exposure, malaria knowledge and insecticide-treated net use among Ugandan women

Article Title: Does exposure to malaria prevention messages influence insecticide-treated net use through knowledge of malaria? Evidence from a causal mediation analysis among women in Uganda

Article References: Mwebesa, E., Agbi, D. M., Natuhamya, C., Musinguzi, B., Izudi, J., Kerich, G., & Mwangi, A. (2026). Does exposure to malaria prevention messages influence insecticide-treated net use through knowledge of malaria? Evidence from a causal mediation analysis among women in Uganda. Frontiers in Public Health, 14, Article 1948476. https://doi.org/10.3389/fpubh.2026.1948476

Image Credits: AI Generated

DOI: 10.3389/fpubh.2026.1948476

Keywords: malaria prevention, insecticide-treated nets, health communication, causal mediation, malaria knowledge, Uganda, cross-sectional survey, women’s health, social and behavior change communication, Does, exposure, malaria