Community-directed mass drug administration is undermined by status seeking in friendship networks and inadequate trust in health advice networks

Abstract

Over 1.9 billion individuals require preventive chemotherapy through mass drug administration (MDA). Community-directed MDA relies on volunteer community medicine distributors (CMDs) and their achievement of high coverage and compliance. Yet, it is unknown if village social networks influence effective MDA implementation by CMDs. In Mayuge District, Uganda, census-style surveys were conducted for 16,357 individuals from 3,491 households in 17 villages. Praziquantel, albendazole, and ivermectin were administered for one month in community-directed MDA to treat Schistosoma mansoni, hookworm, and lymphatic filariasis. Self-reported treatment outcomes, socioeconomic characteristics, friendship networks, and health advice networks were collected. We investigated systematically missed coverage and noncompliance. Coverage was defined as an eligible person being offered at least one drug by CMDs; compliance included ingesting at least one of the offered drugs. These outcomes were analyzed as a two-stage process using a Heckman selection model. To further assess if MDA through CMDs was working as intended, we examined the probability of accurate drug administration of 1) praziquantel, 2) both albendazole and ivermectin, and 3) all drugs. This analysis was conducted using bivariate Probit regression. Four indicators from each social network were examined: degree, betweenness centrality, closeness centrality, and the presence of a direct connection to CMDs. All models accounted for nested household and village standard errors. CMDs were more likely to offer medicines, and to accurately administer the drugs as trained by the national control programme, to individuals with high friendship degree (many connections) and high friendship closeness centrality (households that were only a short number of steps away from all other households in the network). Though high (88.59%), additional compliance was associated with directly trusting CMDs for health advice. Effective treatment provision requires addressing CMD biases towards influential, well-embedded individuals in friendship networks and utilizing health advice networks to increase village trust in CMDs.

Keywords: Compliance; Coverage; Mass drug administration; Social networks; Uganda.

Figure 1

Treatment outcomes for individuals in 17 study villages. The distribution of praziquantel (PZQ), albendazole (ALB), and ivermectin (IVM) is provided for individuals aged one year and older. During mass drug administration, community medicine distributors informed 815 participants of ineligibility due to severe illness, pregnancy status (first trimester for IVM), or age (under five years for PZQ). No response was available for individuals where treatment offer was unknown. Stage 1 and 2 represent the sequence of actions for drug distribution. Individuals provided one or more reasons for noncompliance. For the lack of health education, individuals stated that they did not know the benefit and purpose of the drugs (71.32%, 470/659), did not need to take the drug because they had no symptoms (25.95%, 171/659), believed the drugs did not work (2.12%, 14/659), and witchcraft caused infections (1.67%, 11/659). Only 27.04% (258/954) of noncompliers deliberately refused treatment, despite knowledge of the drug benefits. Their responses noted bad side effects (66.28%, 171/258), lack of food or drink to accompany treatment (27.52%, 71/258), repurposing treatment for livestock (3.49%, 9/258), few friends and neighbors accepting treatment (3.49%, 9/258), and clan and tribe differences with the drug distributor (1.50%, 4/258). Only 3.88% (37/954) of individuals did not comply with treatment because of both the lack of health education and reasons of deliberate refusal.

Figure 2

Association of coverage and compliance across 17 villages. The proportion of eligible individuals offered treatment (coverage) within each village was positively correlated to the proportion of those individuals who ultimately accepted and swallowed treatment (compliance). The fit of an ordinary least squares regression (Coeff. 0.408, p-value = <0.001; 95% CI 0.293, 0.523) is overlaid with the raw data points. The range of village coverage and compliance was 10.9%–86.6% and 67.8%–99.1%, respectively.

Figure 3

Predicted probabilities of coverage at values of friendship degree and friendship closeness. Predicted probabilities of coverage (the offer of at least one drug) against A) friendship degree and B) friendship closeness centrality are shown from the results of the Heckman selection models of Table 1, Table 2.

Figure 4

Predicted probabilities of the accuracy of medicine delivery by CMDs at values of friendship degree and friendship closeness. Panels A–C show the predicted probabilities from the bivariate probit model that includes friendship degree (Supplementary Table 8). Panels D–F present the predicted probabilities from the bivariate probit model with friendship closeness (Supplementary Table 9). Panels C and F from each model show the probabilities for both binary outcomes, 1) praziquantel and 2) both albendazole and ivermectin, equaling one.

Figure 5

Community medicine distributors and health advice networks. Nodes represent households and edges represent trust for health advice. The largest component of each network is presented. Each ID represents one village. Two community medicine distributors (CMDs) are shown in red. The networks were drawn with a force-directed layout, so the most connected and embedded nodes were placed in the centre of the graph. IDs 1–3, 12, and 16 show two CMDs with similar network location.