Differentiated effects of sources of malaria messages on mosquito bed net use among women of reproductive age in Uganda: a propensity score matched analysis

Abstract

Introduction

Malaria remains a significant public health challenge, particularly in high-burden settings such as Uganda. Social and Behaviour Change Communication (SBCC) interventions play a vital role in influencing knowledge, attitudes, and practices related to malaria prevention. However, despite the widespread promotion of mosquito bed net (MBN) use, there is limited evidence on the comparative effectiveness of different media platforms in encouraging their adoption. This study estimated the effect of exposure to malaria messages from different media platforms on MBN use among women of reproductive age in Uganda using propensity score–matched analysis.

Methods

We conducted a secondary analysis of nationally representative data from the 2018–2019 Uganda Malaria Indicator Survey (UMIS), restricted to 3488 women aged 15–49 years who reported exposure to at least one malaria message. The primary outcome was MBN use, and the key exposures were nine distinct SBCC platforms. One-to-one nearest neighbour propensity score matching was applied, adjusting for key socio-demographic characteristics including age, education level, household wealth index, place of residence, and region. Propensity score matching analysis (PSMA) was applied to estimate the average treatment effect on the treated (ATT) for each platform, with ATT chosen to quantify effects among women who were actually exposed to malaria-related messages, adjusting for relevant observed covariates. Data analysis was done in Stata V14.0.

Results

Of the 3488 women included in the study, 73.2% reported using MBN the previous night before the survey. Propensity score matched analysis revealed that exposure to malaria messages through community health workers (ATT = 0.322, 95% CI 0.111, 0.533), community events (ATT = 0.296, 95% CI 0.085, 0.507), and social mobilization (ATT = 0.185, 95% CI 0.008, 0.362) significantly increased MBN use. Other effective channels included social media, radio, interpersonal communication, and billboards. Television exposure (ATT = 0.051; 95% CI  − 0.062, 0.164) and exposure from other unspecified sources were not significantly associated with MBN use.

Conclusion

Among women exposed to malaria messages, exposure through community health workers and community events showed the strongest associations MBN use, followed by exposure through social mobilisation and other SBCC platforms. Exposure through television was not significantly associated with MBN use, suggesting that community-based and interpersonal communication channels may be more strongly associated with MBN utilisation than some mass media platforms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12936-026-05806-2.

Introduction

Malaria remains one of the most pressing global public health challenges, with an estimated 263 million new cases and 597,000 deaths in 2023 [1] and 282 cases and 61,000 death in 2024 [2] were reported worldwide. The World Health Organisation (WHO) African Region bears a disproportionate burden, accounting for 95% of these cases and deaths [3]. Uganda is among the most affected countries, contributing 5.1% of global malaria cases and 2.9% of global malaria deaths, with nearly 12.6 million cases and 15,945 deaths in 2023 [4]. Malaria remains the leading cause of morbidity and mortality in Uganda, particularly among children, with over 90% of the population at risk. Despite interventions such as mosquito bed nets (MBNs), indoor residual spraying (IRS), and vaccination campaigns, malaria continues to impose a health and economic toll exceeding $500 million annually [4].

Pregnant women are especially vulnerable due to reduced immunity during pregnancy, facing risks such as severe illness, anaemia, low birth weight, premature delivery, and neonatal mortality [5, 6]. In 2023, an estimated 12.7 million pregnancies in the WHO African Region were exposed to malaria, with Uganda contributing significantly to this statistic [3]. Prevalence of malaria among pregnant women varies widely, ranging from 4.4% in areas with effective control, like Lira District [7] to 26.1% in high-burden areas such as Arua District [8]. Vulnerable groups include younger, less-educated, primigravida women in rural setting [9]. Placental malaria, though declining due to interventions like IRS, still threatens maternal and fetal outcomes. Without effective control, malaria in pregnancy could result in nearly 914,000 neonates with low birth weight annually in the region [3]. In addition, women in general are primary caregivers for children under five years, who are a vulnerable group to malaria, and they are key decision makers on the use of preventive measures such as insecticide-treated nets [10].

Malaria is recognised as both a cause and perpetuator of poverty at the individual and household level, with direct costs associated with malaria treatment estimated to be 2–3% of household income [11]. A study in Zimbabwe showed that the average cost of treatment of malaria was $17.48 per patient, with $7.59 direct and $9.90 indirect household costs [12]. A more recent study in Uganda found that the societal economic cost of treating suspected malaria cases was $15.2 per outpatient and $27.21 per inpatient, with households incurring 72% of inpatient and 81% of outpatient costs [13]. A woman’s knowledge of malaria prevention has a ripple effect on children's protection against malaria and the adoption of effective methods, such as MBN use [14], possibly at the household level. This would reduce household and individual direct and indirect costs of malaria, increase household productivity, income and quality of life.

Mosquito bed nets (MBNs) remain a cornerstone of malaria prevention, functioning both as physical barriers and insecticidal agents. Their effectiveness in reducing malaria incidence and mortality is well-documented, yet disparities persist in utilisation. Although 68% of households in sub-Saharan Africa owned at least one insecticide-treated net in 2020 [3], consistent use is low. In Uganda, while 78% of women reported sleeping under a net the previous night [15], only 49.8% reported consistent nightly use [16]. Barriers include misconceptions about efficacy, discomfort, heat, sociocultural sleeping arrangements, and concerns about net durability [17]. These challenges highlight that distribution alone is insufficient and that education, counselling at antenatal clinics, and community and culturally sensitive interventions are essential to close the gap between ownership and utilisation. The Government of Uganda and international partners, through National Malaria Control Program (NMCP) has prioritised malaria control. Uganda’s malaria prevention policies focus on an integrated approach that include widespread of insecticide-treated nets, IRS in malaria high-burden districts, intermittent preventive treatment during pregnancy (IPTp) and prompt diagnosis and treatment [18]. Recently, malaria vaccine has been introduced into immunization routine of young children [19, 20] and multi-sectoral campaign for prevention, through a whole government approach for malaria prevention.

Behavioural and contextual barriers to MBN use necessitate complementary strategies, particularly behaviour change communication (BCC). In the context of malaria prevention, social mobilisation refers to coordinated community-level processes that engage local structures, leaders, and social networks to raise awareness, generate collective demand, and sustain participation in preventive behaviours [25]. Social mobilization typically involves activities such as community dialogues, engagement of religious and local leaders, and organised group sensitisation, with an emphasis on influencing shared norms and collective practices rather than individual behaviour alone. While social mobilization may overlap in practice with other communication approaches, it is conceptually distinct from interpersonal communication such as one-to-one counselling during antenatal care visits or household visits by community health workers and from mass media communication, which delivers standardised messages to large audiences with limited interaction. Although these modes are not mutually exclusive, they represent different pathways through which information and motivation for malaria prevention may be transmitted. Clarifying these distinctions is essential for meaningful interpretation of their relative associations with mosquito bed net utilisation in observational studies.

In Uganda, the Ministry of Health leverages mass media platforms such as radio, television, and interpersonal communication to promote malaria prevention. The Uganda’s Ministry of Health, Division of Health Education and Promotion has developed and use various malaria-related messages to educate and enhance malaria-related awareness, including MBN use for malaria prevention, malaria treatment, and control [21]. Approximately 39% of women of reproductive age report exposure to malaria prevention messages, though coverage is much lower in refugee settlements (16%) [22]. The recent UMIS opines that MIS report show that 59% of households owned a radio and 19% owned a television [22]. The exposure to these platforms are necessary for health information dissemination for malaria prevention. Media campaigns have proven effective in improving knowledge and mosquito bed net use, as seen in Nigeria, where radio interventions increased insecticide-treated net (ITN) uptake among pregnant women [23]. However, the relative influence of different media platforms radio, television, print, and social media, community health workers, social mobilisation among others, on MBN utilisation remains underexplored, particularly among women of reproductive age in Uganda. Previous studies have examined the effects of malaria message exposure on knowledge of malaria prevention [21] and insecticide treated net use [24] among women in Uganda. These studies did not isolate the effects of different media platforms on MBN use. This study covered this research gap.

To generate policy-relevant evidence using existing observational data, robust methods such as propensity score matching (PSM) are needed to account for confounding factors that influence both media exposure and MBN use. Characteristics such as age, education, wealth, and rural–urban residence affect both access to media and preventive practices [26]. As a quasi-experimental approach, propensity score matching helps balance observed covariates, mimicking randomised controlled trials where experiments are infeasible in this context. [27]. Applying PSM in Uganda allows isolation of the impact of specific media platforms on MBN uptake, offering a nuanced understanding of how communication translates into behaviour change. By identifying the most effective platforms in reaching women of reproductive age, malaria control programs can prioritise limited resources, enhance MBN use, and reduce the disease burden and its economic costs. Ultimately, such evidence-driven approaches will strengthen behaviour change strategies and contribute to sustainable malaria prevention in Uganda.

Methods and materials

Study design and setting

This study utilised secondary data from the 2018–2019 UMIS, a nationally representative cross-sectional survey conducted as part of the Demographic and Health Surveys (DHS) program [22]. The UMIS employed a two-stage stratified sampling design to ensure regional and urban–rural representativeness. The survey collected data on malaria prevention practices, knowledge, and treatment. A total of 8868 women aged 15–49 years were interviewed. For this analysis, the sample was restricted to 3488 women who reported exposure to at least one source of malaria messaging, as these individuals were eligible for assessing the impact of specific SBCC platforms on MBN use. In recent Malaria Indicator Survey (MIS), only women who had reported to have seen/heard a malaria message, were further asked the source of the message. Thus, only these women were eligible for this study.

Study variables

Outcome Variable: The primary outcome variable was MBN use, defined as whether the respondent reported sleeping under a mosquito bed net the night before the survey. This variable was coded as binary (1 = yes, 0 = no).

Treatment Variables: The treatment variables were defined as exposure to any of the various malaria message delivery platforms. These platforms were coded as binary variables indicating whether a respondent reported having seen/heard a malaria message (exposed: yes = 1) or not (unexposed: no = 0) within six months preceding the survey, as provided in MIS. The platforms assessed included: radio, television, flyers, posters/billboards, community health workers (CHWs), Community events, social mobilisation activities, interpersonal communication, social media, and other sources. These sources were used to estimate the effect of each platform on MBN utilisation. Interpersonal communication involves one-to-one or face-to-face interaction between a health worker and women on issues related to MBN use while social mobilisation involves bringing women together with other stakeholders to raise awareness on MBN use. Community events are gatherings organised in communities to inform women about MBN use. These approaches are no mutually exclusive as personal interactions may happen during community events and social mobilisation may be need in organising such events.

Matching Variables: Covariates included in the propensity score model were selected based on their empirical association with exposure to malaria messages and MBN use. The matching variables included age of the respondent (15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49), education level (no education, primary, secondary/higher), household wealth index (poorest, poorer, middle, richer, richest), residence (urban, rural, refugee), and region (Central, Eastern, Northern, Western). These variables were selected based on previous literature on factors associated with media exposure [28, 29]. These were also selected because they fit the propensity score models well.

Statistical analysis

The study used counts and percentages to present categorical covariates of the sample. Cross–tabulations with their respective chi-square tests were used to identify the distribution and association of these covariates on MBN use. Propensity score matching analysis (PSMA) was used to estimate the effect of exposure to each SBCC platform on MBN use. This analytical approach helps address selection bias in observational data by creating balanced groups based on observed covariates. [27, 30–32]. Propensity scores were estimated using logistic regression models including the treatment and matching variables. The matching variables included age group, education level, household wealth index, place of residence, and region, selected based on their empirical relevance to exposure to media platforms and MBN use. To address the potential curse of dimensionality associated with multiple covariates, propensity scores were used as a scalar balancing score, reducing the multidimensional covariate space into a single index for matching.

A propensity score refers to the probability of assigning a woman, “i” (i = 1, 2, …, n), to either the exposed group (women who had seen/heard a malaria message from a certain source/media platform (s, s = 1, 2, …, k)) or the unexposed group (women who had not seen/heard a malaria message from a certain source/media platform), based on a set of observed woman characteristics “X” [30]. This is presented as (Eq. 1)

$$e\left(\mathit{X}\right)=\text{Pr}\left[Z_i,=,1\right|{\mathit{X}}_i=x_i]$$ 1

where $e(X)$ is the conditional probability of exposure to a certain source of malaria messages, given the vector of observed women characteristics ${\mathit{X}}_i=x_i$. The variable Z is the treatment variable that takes the values of 0 and 1, where 0 is for unexposed group and 1 is for exposed group. The vector ${\mathit{X}}_i$ of exposure to a certain source of malaria messages [24, 27]. This study estimated the Average Treatment Effect of the Treated (ATT) related to MBN use among women who had seen/heard a malaria messages prior to the survey. To ensure fair comparisons between women, one-to-one (1:1) nearest neighbour matching without replacement was implemented, using caliper widths of 0.01 to 0.0425 to minimise poor matches and residual bias. These were obtained based on guidance from previous study which observed that a caliper of 0.25 of standard deviations of logit of the propensity score removes 90% of the bias [33]. The ATT is estimated using a counterfactual framework shown in Eq. 2 below

$$ATT=E\left[Y_i,\left(1\right),-,Y_i,\left(0\right)\right|Z_i=1]$$ 2

where $Y_i\left(1\right)$ is the outcome in women who were exposed, $Y_i\left(0\right)$ is the outcome in women who were unexposed, $E\left[Y_i,\left(1\right)\right|Z_i=1]$ represents the expected outcome of MBN use, assuming that all women have been exposed. $E\left[Y_i,\left(0\right)\right|Z_i=1]$ represents the expected outcome of MBN use assuming that all women have not been exposed (unobserved) [24]. The ATT denotes the expected difference in MBN use between women in the exposed group and the same women if they were not in the exposed group [34]. The standard errors were calculated using bootstrapping with 150 repetitions to aid in the estimation of 95% confidence intervals of average treatment effects of the treated.

The psmatch2 command in STATA version 14.0 was used to perform the analysis. Covariate balance before and after matching was assessed using standardised percentage bias and pseudo R-squared values. Balancing quality was evaluated using the pstest command. A reduction in mean and median biases of less than 10% in matched samples were considered successful. In addition, successful match was defined by non-significant likelihood ratio chi-squares p-values and very small pseudo R-squared in matched samples, implying absence of systematic differences in exposed and unexposed groups after matching. The average treatment effect on the treated (ATT) was estimated for each media platform. ATT was chosen because the primary objective was to assess the effect of exposure to sources of malaria messages among women who were actually exposed to malaria-related messages, providing a policy-relevant estimate of the impact of existing communication strategies rather than a hypothetical effect in the entire population. Treated units were women exposed to a given source of malaria message platform, while controls were women not exposed to that platform; this was not a traditional case–control design but a matched comparison within a cross-sectional sample. A forest plot (Fig. 11) was generated to visually summarise the ATT estimates and their corresponding 95% confidence intervals (CI). Sampling weights, clustering and stratification were applied to cater for complex survey designs.

Fig. 11.

Forest plot of treatment effects and their 95% confidence intervals

Results

Descriptive characteristics of the study population by MBN use status

Among the 3488 women included in the analysis, 73.2% reported using a mosquito bed net (MBN) the night before the survey. Among these women, 20.8% were aged 20–24 years, and majority had attained a primary education (51.4%). Approximately 26.1% of the women belonged to the poorest wealth quintile, and majority (68.1%) of the resided in rural areas. One-third of women were from Northern Uganda (33.1%). Results from cross-tabulations of MBN and different socio-demographic reveal that, women that used MBN, the proportion of those aged 20–24 years (20.8%) and had a primary education were (51.4%). Notably, while the poorest quintile was most prevalent in the overall sample, most users of MBN were from richest wealth quintile (24.8%). Geographically, 67.2% of users lived in rural areas, and the largest proportion of users was located in Western Uganda (33.2%). Different socio-demographic factors demonstrated a statistically significant association with MBN usage (p < 0.001), including maternal age, education level, wealth index, and region. Conversely type of residence (urban, rural, refugee) did not show a statistically significant association with MBN use. These findings are detailed in Table 1.

Table 1.

Distribution of women who slept under the mosquito bed net by selected characteristics

Covariates	Categories	Overall	Slept under a mosquito bed net		Chi-square p-value
			No	Yes
		N = 3488	935 (26.81%), col	2553 (73.19%), col
Age in 5-year groups	15–19	673 (19.3%)	251 (26.8%)	422 (16.5%)	< 0.001
	20–24	725 (20.8%)	194 (20.7%)	531 (20.8%)
	25–29	626 (17.9%)	153 (16.4%)	473 (18.5%)
	30–34	512 (14.7%)	116 (12.4%)	396 (15.5%)
	35–39	387 (11.1%)	82 (8.8%)	305 (11.9%)
	40–44	348 (10.0%)	87 (9.3%)	261 (10.2%)
	45–49	217 (6.2%)	52 (5.6%)	165 (6.5%)
Highest educational level	No education	423 (12.1%)	162 (17.3%)	261 (10.2%)	< 0.001
	Primary	1793 (51.4%)	480 (51.3%)	1313 (51.4%)
	Secondary	973 (27.9%)	233 (24.9%)	740 (29.0%)
	Higher	299 (8.6%)	60 (6.4%)	239 (9.4%)
Wealth index combined	Poorest	909 (26.1%)	311 (33.3%)	598 (23.4%)	< 0.001
	Poorer	648 (18.6%)	169 (18.1%)	479 (18.8%)
	Middle	540 (15.5%)	138 (14.8%)	402 (15.7%)
	Richer	572 (16.4%)	132 (14.1%)	440 (17.2%)
	Richest	819 (23.5%)	185 (19.8%)	634 (24.8%)
Type of place of residence	Urban	971 (27.8%)	236 (25.2%)	735 (28.8%)	0.120
	Rural	2377 (68.1%)	661 (70.7%)	1716 (67.2%)
	Refugee	140 (4.0%)	38 (4.1%)	102 (4.0%)
Region	Central	477 (13.7%)	121 (12.9%)	356 (13.9%)	< 0.001
	Eastern	755 (21.6%)	207 (22.1%)	548 (21.5%)
	Northern	1155 (33.1%)	354 (37.9%)	801 (31.4%)
	Western	1101 (31.6%)	253 (27.1%)	848 (33.2%)

Association between sources of malaria messages and MBN use among women in the sample

The results reveal that, of the women included in the study, 66.9% received malaria messages from radios, 41.8% from interpersonal communication, 37.7% from community health workers, 23.0% from community events, 20.4% from televisions, 15.8% from social mobilisation, 15.3% from posters/billboards, 8.7% from flyers and 7.8% from social media. Sixteen percent of the women received malaria messages from elsewhere. Exposure to malaria messages via radio demonstrated a statistically significant positive association with bed net usage (p < 0.001), with 69.2% of those exposed to radio messages using bed nets. Similarly, television (p = 0.020) was significantly associated with MBN usage. Conversely, posters/billboards (p = 0.400), community health workers (p = 0.540), community events (p = 0.073), interpersonal communication (p = 0.310), flyers (p = 0.078), social mobilization (p = 0.680), and social media (p = 0.880) did not show a statistically significant association with MBN usage Table 2.

Table 2.

Distribution of sources of malaria messages and women who slept under the mosquito bed net

Source of malaria message	Categories	Overall	Slept under a mosquito bed net		Chi-square p-values
			No	Yes
		N = 3488	935 (26.81%), col	2553 (73.19%), col
Radio	No	1156 (33.1%)	369 (39.5%)	787 (30.8%)	< 0.001
	Yes	2332 (66.9%)	566 (60.5%)	1766 (69.2%)
Television	No	2777 (79.6%)	769 (82.2%)	2008 (78.7%)	0.020
	Yes	711 (20.4%)	166 (17.8%)	545 (21.3%)
Poster/billboard	No	2,955 (84.7%)	800 (85.6%)	2155 (84.4%)	0.400
	Yes	533 (15.3%)	135 (14.4%)	398 (15.6%)
Community health worker	No	2174 (62.3%)	575 (61.5%)	1599 (62.6%)	0.540
	Yes	1314 (37.7%)	360 (38.5%)	954 (37.4%)
Community event	No	2685 (77.0%)	700 (74.9%)	1985 (77.8%)	0.073
	Yes	803 (23.0%)	235 (25.1%)	568 (22.2%)
Interpersonal communication	No	2030 (58.2%)	531 (56.8%)	1499 (58.7%)	0.310
	Yes	1458 (41.8%)	404 (43.2%)	1054 (41.3%)
Flyers	No	3186 (91.3%)	867 (92.7%)	2319 (90.8%)	0.078
	Yes	302 (8.7%)	68 (7.3%)	234 (9.2%)
Social mobilisation	No	2936 (84.2%)	791 (84.6%)	2145 (84.0%)	0.680
	Yes	552 (15.8%)	144 (15.4%)	408 (16.0%)
Social media	No	3216 (92.2%)	861 (92.1%)	2355 (92.2%)	0.880
	Yes	272 (7.8%)	74 (7.9%)	198 (7.8%)
Anywhere else	No	2927 (83.9%)	807 (86.3%)	2120 (83.0%)	0.020
	Yes	561 (16.1%)	128 (13.7%)	433 (17.0%)

Descriptive estimates average treatment effects of different sources of malaria message on mosquito bed net use

The propensity score graphs, presented in Figs. 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10, were used to assess the matching quality of exposed to malaria messages through various channels (radio, television, billboards, community health workers, community events, interpersonal communication, flyers, social mobilisation, social media, and any other source) and those unexposed, in relation MBN use. Across all figures, the distributions of propensity scores for exposed (1 = Exposed) and unexposed (0 = Unexposed) groups are well-balanced and overlap exits. These similar distributions indicate effective matching, suggesting that the exposed and unexposed groups are suitable for comparison when evaluating the of effects of different sources of malaria messages on MBN use. Off-support was 22 women on exposure to malaria messages through radios, 2 women on interpersonal communication and these were excluded in analysis in analysis of effect of these sources on MBN use. Off-support also called “region of common support” occurs when a woman in the study has a propensity score that cannot be matched with another woman for the other comparison group. There were no off supports for other sources of malaria messages on MBN use. These minimal off-support (< 1%) and having most sources of malaria messages having no off-support ensured robust and reliable comparisons for each communication channel.

Fig. 1.

Distribution of propensity scores by exposure through radio

Fig. 2.

Distribution of propensity scores by exposure through television

Fig. 3.

Distribution of propensity scores by exposure through billboards

Fig. 4.

Distribution of propensity scores by exposure through community health workers

Fig. 5.

Distribution of propensity scores by exposure through community events

Fig. 6.

Distribution of propensity scores by exposure through interpersonal communication

Fig. 7.

Distribution of propensity scores by exposure through flyers

Fig. 8.

Distribution of propensity scores by exposure through social mobilisation

Fig. 9.

Distribution of propensity scores by exposure through social media

Fig. 10.

Distribution of propensity scores by exposure through any other source

After matching, the average treatment effect on the treated (ATT) was estimated for MBN use among women of reproductive age in Uganda, based on exposure to malaria messages from various sources. The descriptive statistics indicated that the ATT for MBN use varied by source of malaria messages, as shown in Table 3. The ATT estimates are as follows: radio (ATT = 0.173, SE = 0.057), television (ATT = 0.051, SE = 0.077), billboards (ATT = 0.101, SE = 0.050), community health workers (ATT = 0.322, SE = 0.132), community events (ATT = 0.296, SE = 0.129), interpersonal communication (ATT = 0.152, SE = 0.069), flyers (ATT = 0.139, SE = 0.054), social mobilisation (ATT = 0.185, SE = 0.108), social media (ATT = 0.151, SE = 0.078), and other sources (ATT = 0.144, SE = 0.069). All sources show a positive effect on mosquito bed net use among exposed women of reproductive age in Uganda, however, exposure through televisions and anywhere else are not statistically significant.

Table 3.

Estimation of ATT of exposure to different sources of malaria messages on mosquito bed net use

Source of malaria message	Sample	Exposed	Unexposed	Difference	SE	T-stat
Radio	Unmatched	0.757	0.681	0.076	0.016	4.81
	ATT	0.757	0.584	0.173	0.057	3.05
Television	Unmatched	0.767	0.723	0.044	0.019	2.34
	ATT	0.767	0.716	0.051	0.077	0.66
Billboards	Unmatched	0.747	0.729	0.017	0.021	0.84
	ATT	0.747	0.645	0.101	0.050	2.03
Community health workers	Unmatched	0.726	0.736	− 0.009	0.015	− 0.61
	ATT	0.726	0.404	0.322	0.132	2.45
Community events	Unmatched	0.707	0.739	− 0.031	0.018	− 1.79
	ATT	0.707	0.411	0.296	0.129	2.29
Interpersonal communication	Unmatched	0.723	0.738	− 0.016	0.015	− 1.02
	ATT	0.723	0.571	0.152	0.069	2.20
Flyers	Unmatched	0.775	0.728	0.047	0.027	1.76
	ATT	0.775	0.636	0.139	0.054	2.55
Social mobilisation	Unmatched	0.739	0.731	0.009	0.021	0.42
	ATT	0.739	0.554	0.185	0.108	1.71
Social media	Unmatched	0.728	0.732	− 0.004	0.028	−0.15
	ATT	0.728	0.577	0.151	0.078	1.93
Anywhere else	Unmatched	0.772	0.724	0.048	0.020	2.33
	ATT	0.772	0.627	0.144	0.069	2.08

ATT and their 95% confidence intervals for different sources of malaria message on mosquito bed net use

The results reveal that women exposed to malaria messages through social mobilisation had 18.5%% higher likelihood to use MBNs (ATT = 0.185, 95% CI 0.008, 0.362) compared to the same women who were not exposed to malaria messages through social mobilisation. Those exposed to flyers had 13.9% higher likelihood to use MBN (ATT = 0.139, 95% CI 0.025, 0.254), compared to similar individuals had they not been exposed to these messages through flyers. Significant estimates were also observed for exposure to malaria messages from community health workers (ATT = 0.322, 95% CI 0.111, 0.533), community events (ATT = 0.296, 95% CI 0.085, 0.507), social media (ATT = 0.151, 95% CI 0.009, 0.293), radio (ATT = 0.173, 95% CI 0.076, 0.270), billboards (ATT = 0.101, 95% CI 0.014, 0.189), and interpersonal communication (ATT = 0.152, 95% CI 0.018, 0.286). Exposure to malaria messages from television and other sources (elsewhere) showed a non-significant estimates (ATT = 0.051, 95% CI − 0.062, 0.164) and (ATT = 0.144, 95% CI − 0.004, 0.292) respectively on MBN use among women of reproductive age. This suggests that exposure to malaria messages through community health workers and community events increases MBN use to prevent malaria by up to 32.2% and 29.6%, respectively, compared to women with similar characteristics who were not exposed to these malaria messages from these sources. These results indicate that exposure to malaria messages significantly increases MBN use among women across most communication channels. The quality of matching was very high, with pseudo R² and LR Chi² values of 0.000 to 0.002, p = 1.000, and mean and median biases ranging from 0 to 1.2 across all sources. Results are shown in Table 4.

Table 4.

Quality of matching and average treatment effects on the treated (ATT) of access to malaria messages from different sources on mosquito bed net use

Sources of malaria messages		Model diagnostics					ATT [95% CI]	p-value
		Pseudo R2	LR chi2	p > Chi2	Mean bias	Median bias
Radio	Unmatched	0.074	326.18	0.000	11.1	6.0
	CC(0.0425)	0.000	3.02	0.999	1.0	0.8	0.173 (0.076, 0.270)	< 0.001
Television	Unmatched	0.431	1493.08	0.000	48.2	33.4
	NN(1)	0.002	3,03	1.000	1.6	1.1	0.051 (− 0.062, 0.164)	0.381
Billboards	Unmatched	0.071	231.88	0.000	17.7	9.3
	CC(0.05)	0.000	1.60	1.000	1.2	1.2	0.101 (0.014, 0.189)	0.023
Community health workers	Unmatched	0.015	70.65	0.000	7.1	4.7
	CC(0.018)	0.000	0.000	1.000	0.0	0.0	0.322 (0.111, 0.533)	0.003
Community events	Unmatched	0.016	61.14	0.000	6.4	4.5
	CC(0.015)	0.000	0.000	1.000	0.0	0.0	0.296 (0.085, 0.507)	0.006
Interpersonal communication	Unmatched	0.014	66.72	0.000	4.3	2.8
	CC(0.023)	0.000	0.01	1.000	0.1	0.4	0.152 (0.018, 0.286)	0.027
Flyers	Unmatched	0.048	98.35	0.000	15.5	8.6
	CC(0.0115)	0.001	0.57	1.000	1.3	1.0	0.139 (0.025, 0.254)	0.017
Social mobilisation	Unmatched	0.007	21.51	0.011	5.7	5.5
	CC(0.01)	0.000	0.000	1.000	0.0	0.0	0.185 (0.008, 0.362)	0.041
Social media	Unmatched	0.124	236.51	0.000	21.5	11.6
	CC(0.0165)	0.000	0.000	1.000	0.0	0.0	0.151 (0.009, 0.293)	0.038
Anywhere else	Unmatched	0.060	184.64	0.000	13.0	4.3
	CC(0.021)	0.000	0.000	1.000	0.0	0.0	0.144 (− 0.004, 0.292)	0.057

CC Common Caliper, NN Nearest Neighbour, ATT Average Treatment Effect of the Treated, LR Likelihood Ratio, Chi2 Chi-square

To visualise the ATTs, a forest plot (Fig. 11) was used to visually represent the estimated treatment effects (ATT) and their 95% confidence intervals for various malaria message sources, with the dashed blue line at ATT = 0.00 signifying no effect. The most effective sources of malaria messages were community health workers and community events with statistically significant estimates on MBN use. The figure further shows that mass media varies widely, with radio the most effective, while television shows no significant impact. Exposure to malaria messages through billboards shows a small but significant impact on MBN use. Exposure of malaria messages through digital and print media (social media and flyers) shows moderate, but significant impact on MBN use among women of reproductive. All sources except television and anywhere else have confidence intervals that do not cross zero, implying that they are statistically significant in predicting MBN use.

Discussion

This study investigated the effectiveness of various sources of malaria messages in promoting MBN utilization among Ugandan women of reproductive age, who reported exposure to one or more source of malaria messages. By utilizing Propensity Score Matching Analysis (PSMA) with data from the 2018–2019 Uganda Malaria Indicator Survey (UMIS), this study provided a robust assessment of estimated treatment effects, ATT, while effectively controlling for observable confounding variables. The findings of this study reinforce the crucial role of SBCC in encouraging malaria prevention practices. A high proportion of women who were exposed to malaria messages reported using mosquito bed nets (73.19%). This finding is consistent with regional estimates of MBN use among women in East Africa [35], but slightly lower than average prevalence of MBN use in seven high malaria endemic countries in SSA [36]. Initial findings highlight that maternal age, education, wealth index and region are significantly associated with MBN use. The findings also show that, although the poorest wealth quintile had the highest representation in the sample (26.1%), the richest quintile showed a high proportion of net usage (24.8%). This suggests that while malaria messages reach a broad audience, economic barriers to MBN use persist, indicating that malaria messaging should be integrated with economic support to ensure equity.

The findings also show inconsistency between bivariate correlations between different sources of malaria messages and MBN use and PSM results. While radio and television showed significant associations in the initial cross-tabulations, several other platforms were not significantly associated with MBN use, but revealed their potential impact through PSM. This occurs because bivariate analyses do not account for self-selection bias, where women already predisposed to using nets might also be more likely to seek out or remember health messages. By matching exposed and unexposed women on observable covariates, PSM reduces this bias [37] to better estimate the specific contribution of each communication platform.

The study findings reveal that community-based strategies emerged as powerful drivers of behaviour change, supporting earlier evidence from Nigeria that highlights the success of community-based initiatives in promoting preventive behaviours [38]. Specifically, exposure to malaria messages through Community Health Workers (CHWs) and Community Events was associated with an estimated 32.2% and 29.6% increase in the probability of MBN use, respectively. These findings demonstrate the strong evidence that the personalised, face-to-face communication provided by trusted local figures, in communities effectively overcomes common barriers to the adoption and use of MBN [39]. CHWs can leverage their professional position and community relationships to build trust, making their messages more acceptable, credible and influential [40, 41] and can provide thorough information on the importance of mosquito bed nets, correct hanging techniques, and how to maintain them to prevent damage, solving issues that often lead to underutilisation [42]. The findings also showed interpersonal communication as a source of malaria messages is associated with MBN use among women in Uganda, underscoring the critical role of personal interactions in health behaviour change [41]. Interpersonal communication facilitates tailored, context-specific messaging that resonates with women in both rural and urban settings [43]. It builds trust, foster understanding, and allow for two-way conversations leading to more effective health promotion and improved health literacy and outcomes [44]. In Uganda, interpersonal interactions effectively address barriers to MBN use, such as misconceptions about malaria or practical challenges in net maintenance and improve utilization [42, 45]. For instance, village health teams (VHTs) and women’s groups are instrumental in delivering hands-on demonstrations and reinforcing MBN benefits, fostering sustained behavioural adoption as they leverage existing community structures to educate women in familiar settings [46]. These findings show that community-based sources of malaria messages can enhance the MBN uptake and support Uganda’s malaria elimination efforts.

Social mobilisation emerged as one of the effective platform for increasing MBN use, supporting earlier evidence from Nigeria that highlights the success of community-based initiatives in promoting preventive behaviours [38]. The participatory nature of social mobilisation, which fosters trust and community engagement, likely accounts for its effectiveness, particularly in rural contexts where interpersonal relationships play a critical role in behaviour change [35]. Social mobilisation involves a comprehensive process that engages and involves all stakeholders to create an enabling environment that leads to not only primary benefits to a group of people [47], and in this context, related to the ownership, correct use, and maintenance of MBNs. It increases MBN use by providing key information on mosquito net use, fostering community ownership, and informing women about the burden of malaria and prevention measures [48]. The significant effect highlights social mobilisation’s role in boosting MBN adoption, offering a scalable strategy to strengthen Uganda’s malaria control efforts. Television, had no significant association on MBN use in this study.

The study also reveals the relevance of traditional print and digital media. Exposure to flyers showed a significant positive ATT, suggesting printed materials remain impactful due to their visual and tactile nature. Similarly, social media and radio demonstrated significant estimated effects. Radios are among the most common sources of malaria messages or information [21, 49, 50]. This finding underscores the effectiveness of radio campaigns in the region’s health communication landscape, due to their wide reach and accessibility in rural and low-resource settings. Studies show that exposure to radio messages is associated with increased knowledge of MBN benefits, and thus, those exposed are more likely to use MBN and discuss its use with others [23, 39, 51]. The significant impact suggests that leveraging radio as a medium for targeted health messaging can enhance MBN adoption, supporting broader malaria control efforts in Uganda. Also, literature shows that flyers and posters, and other print media, improve awareness and adoption of malaria prevention strategies [52, 53]. Flyers impact mosquito bed net use through the provision of information about the benefits and effectiveness of net use. They can be used to educate women and households on the benefits of MBN use, such as protection against mosquitoes and malaria [54], and offer avenues for reaching specific groups based on languages and literacy levels, since pictures can also be used to display ways of MBN use and provide wider geographical coverage for a long period of time [55, 56]. Exposure to malaria messages from billboards showed effectiveness in MBN use among women in Uganda. Normally, billboards, strategically placed along major roads and in urban centres to deliver concise visuals and high-impact messages in local languages to reach diverse audiences. This finding indicates that billboard exposure enhances MBN adoption by increasing visibility and recall of prevention strategies, suggesting that integrating visual campaigns with community efforts could optimise malaria elimination goals.

Exposure to malaria messages through social media enhances MBN use highlighting the transformative role of digital platforms in promoting preventive behaviours against malaria in endemic regions, potentially bridging gaps in traditional health communication channels. Consistent with these results, a study among women in Uganda revealed that exposure to mass media messages, significantly boosted knowledge of MBN use as a prevention measure [21]. These convergent findings suggest that integrating social media into malaria control strategies could yield scalable impacts on public health outcomes, warranting further investment in tailored digital interventions. In Uganda, limited access to television, particularly in rural areas, combined with economic barriers and geographic disparities, may reduce its effectiveness as an sources of malaria [57]. These findings suggest that investments in malaria communication campaigns should be strategically focused on platforms that combine reach with interpersonal influence and cultural relevance. While television is a powerful medium, its limited effectiveness in this study may be a reflection of low access and economic barriers in rural areas rather than the quality of the messages themselves. For maximum effectiveness, malaria communication should prioritize interactive, culturally relevant platforms that combine broad reach with interpersonal influence.

Strengths and limitations

The application of PSMA is a major methodological strength, as it mimicked a randomized approach to reduce bias in cross-sectional data [58, 59]. However, several limitations must be acknowledged. The analysis excluded women with no reported exposure to malaria messaging; given that only 3,488 out of 8,868 interviewed women (approx. 40%) reported exposure to at least one source, these findings apply only to the reached population and may not generalize to all women in Uganda. This highlights a significant gap in message coverage that requires urgent attention. Both exposure and bed net use were self-reported, which could lead to recall or social desirability bias. While PSM strengthens the analysis by controlling for observable covariates, it does not fully guarantee causality due to potential unobserved confounders such as cultural norms or specific attitudes toward media exposure. Also, this study did not formally analyse how media platforms impact is mediated by knowledge of malaria, and the cross-sectional nature of the data prevents the evaluation of long-term sustainability, which remain vital areas for future longitudinal research. Matching was done for separate PSM models for each media platform. This may introduce multiple testing bias. However, since each media platform was considered as a separate exposure, to test the effect of each, it was necessary that matching be done on each of them. Few categorical variables were used in matching and drawing associations and conclusions. The matching was done on only variables, yet in presence of many matching variables, it’s possible that more efficient estimates may be obtained.

The study utilised secondary data from the 2018–19 Uganda Malaria Indicator Survey (UMIS). Whereas this is a population-based survey, some confounders, such as cultural norms, attitudes and knowledge towards the media that might affect exposure to SBCC platforms, were not collected and thus were not controlled in PSM analysis. Also, this study used cross-sectional data and could not evaluate the long-term effects of exposure to SBCC platforms, which can be evaluated through longitudinal studies. In addition, the study did not evaluate how the impact of exposure to malaria messages through these media platforms on MBN use is mediated by knowledge of malaria cause and prevention methods. This could be evaluated in future studies. Future studies could carry out the mediation analysis of comprehensive knowledge of malaria in the relationship between exposure to malaria messages through these platforms and MBN use. Future qualitative studies could be carried out to examine other covariates of media exposure and MBN use such as socio-cultural attitudes access to MBNs and other barriers.

Whereas this study evaluated the effects of malaria messages for different media platforms on MBN use, the contents, frequency and language of these messages were not known, yet this might affect MBN use. Future studies and Malaria Indicator Surveys should extend beyond measuring access to malaria messages through these platforms and capture both the sources and characteristics of malaria messages. Identifying these aspects would improve examination of how and why media exposure influences MBN use, as access malaria messages alone may not adequately reflect impactful exposure or malaria message effectiveness. Furthermore, this study could not clearly differentiate information sources, communication channels, and programmed communication or mechanisms, that are often employed for specific communicators and audiences to use specific communication channels that allow them to exchange information about malaria. The available data capture general exposure to malaria message sources rather than specific modes or contexts of communication. Hence, the estimates should be interpreted as reflection to broad media platforms exposure.

Conclusion and recommendations

This study demonstrates that various sources of malaria messages, particularly those involving community health workers, social mobilization, and flyers, are significantly associated with increased MBN utilization among women in Uganda. While radio had a broad reach, its effect was modest, and television showed no significant impact, suggesting the importance of considering platform interactivity, cultural relevance, and access within the Ugandan context. The study recommends prioritization of community-based platforms of malaria messaging and maximize the use of CHWs for personalized outreach. Given that about 60% of the surveyed population reported no exposure to malaria messaging, it is critical to scale up these platforms to reach the unexposed majority. Furthermore, radio messaging should be optimized to be more interactive, and investments in television should be context-specific based on local access levels. Future research should examine the multi-platform impact of combined channels to create a high-intensity communication environment for sustained behavioral change.

The main objective of this study was to investigate the effect of different SBCC (radio, television, poster/billboards, community events, social mobilisation, social media, community health workers, interpersonal communication, fliers) platforms on MBN use among women of reproductive age in Uganda using propensity score matched analysis. This study also evaluated the binary association of these platforms with MBN use among women in Uganda. This study found that only exposure to malaria messages from the radio and television was significantly related to use of MBN among women. Despite statistically insignificant associations in the other platforms (a part from television and anywhere else) and MBN use, PSM analysis revealed a statistically significant effect of these platforms, and exposure to malaria messages from radio as well on MBN use. This is because correlation analyses do not account for confounding or control for observable covariates, which can obscure the true relationship between exposure to malaria messages from these platforms and MBN use [30]. PSM addresses this by matching exposed and unexposed women on observable confounders, isolating the causal impact of exposure to malaria messages from these platforms [58, 59]. The significant ATT estimates from PSM indicates that exposure to these platforms promotes MBN use when biases are controlled, emphasising the value of causal inference methods in public health research and the need for targeted malaria messaging to enhance MBN adoption in Uganda.

Abbreviations

ATT

Average treatment effect of the treated

AU

African Union

BCC

Behavioural change communication

CC

Common caliper

CHWs

Community health workers

CI

Confidence interval

DHS

Demographic and Health Survey

ICF

International Classification of Functioning, Disability and Health

IPTp

Intermittent preventive treatment during pregnancy

IRS

Indoor residual spraying

ITN

Insecticide-treated net

MBN

Mosquito bed net

MBNs

Mosquito bed nets

LR

Likelihood ratio

MIS

Malaria indicator survey

MM

Malaria messages

NMCD

National malaria control division

PSMA

Propensity score-matched analysis

PSM

Propensity score matching

SE

Standard error

SMS

Short messages

UBOS

Uganda Bureau of Statistics

UMIS

Uganda Malaria Indicator Survey

WHO

World Health Organisation

Author contributions

EM conceptualised the research idea, obtained and analysed the data. DMA wrote the first draft of the manuscript. LOA, DKK and DTI did thorough edits of the draft manuscript. GKK and AM supervised and proofread the final document. All authors read and approved the final manuscript.

Funding

None.

Data availability

The datasets analysed during the current study were obtained from the MEASURE DHS program. https://dhsprogram.com/data/dataset_admin/index.cfm

Declarations

Ethics approval and consent to participate

This study analysed data from Malaria Indicator Surveys (MIS), obtained from the DHS Program, which had no identifiable participants’ data; as such, ethical approval was not needed.

Consent for publications

Not applicable.

Competing interests

The authors declare no competing interests.