Study Area
The survey was conducted in Sussundenga village. Sussundenga village is approximately 42 km from the Provincial capital of Chimoio City in central Mozambique (Fig. 1). Sussundenga village is the district municipality of Sussundenga District and is a rural agrarian community with an approximate population of 19,112 residents and shares a border with Zimbabwe51,52. From September to March, Manica Province has high temperatures and increased rainfall which contributes to the reported seasonally high malaria incidence. The main rural health center (RHC) in the village is the Sussundenga-Sede health center, with 14 smaller RHCs located in the district. The study sampled households from this village because of the high malaria incidence, proximity to the Zimbabwe border, rurality, impacts of Cyclone Idai, and accessibility to several RHCs for referral care. The primary malaria prevention intervention in the region at the time of the survey were insecticide treated nets (ITNs) for households distributed through the antenatal care center.
Study Design and Data Collection
A cross-sectional community-based survey was administered from December 2019 to February 2020 in Sussundenga village. This analysis focused on household level damage incurred during Cyclone Idai and malaria risk in the aftermath. Additional data regarding household structure, demographics, and use of malaria prevention were collected. Satellite imagery was used to enumerate 2,889 households in Sussundenga village. A random household sampling method selected 125 households for screening and 100 households for enrollment to allow for potential misclassification of household structures and refusals. This was determined from estimated 5–6 total residents per household at the time of survey administration. As a pilot study the sample size was determined to detect differences in specific risk factors between those with and without P. falciparum infection detected by rapid diagnostic test. GPS coordinates were used to locate households for enrollment. Data collectors determined eligibility through a notification visit with the head of the household and assigned each household member a unique identifier.
Data collectors also obtained informed consent for all adult residents and parent/guardian permission for children between 3 months to 13 years old and assent for minors between 13 to 17 years old. Enrollment eligibility criteria was any full-time resident older than 3 months. After the notification visit, data collectors administered the electronic survey and recorded household GPS coordinates on a tablet computer using a REDCap® (Research Electronic Data Capture) mobile application. All participants present at the time of survey administration who were older than 13 years old completed the survey and parents provided responses for children 3 months old to 13 years old.
A study nurse collected a finger prick blood sample and administered a malaria rapid diagnostic test (RDT) [Right Sign Malaria Pf (Biotest, Hangzhou Biotest Biotech Co, China]. All participants with positive results were referred to Sussundenga-Sede RHC for confirmation of diagnosis and treatment. All data were collected and stored using the REDCap® server hosted at University of Minnesota School of Public Health and was treated confidentially (11,12). Ethical review and approval for the study was completed by the Institutional Review Board (IRB) at the University of Minnesota [STUDY00007184] and from A Comissão Nacional de Bioética em Saúde (CNBS) at the Ministry of Health of Mozambique [IRB00002657] and was preformed within their guidelines and regulations. Informed consent was obtained from all adults and parental permission and assent were obtained from all minor children.
Data Analysis
All data analyses were preformed using Stata (StataCorp. Version 15.1). Household damage was collected as a categorical variable by self-report by the head of household as no damage, minor damage, significant damage, or destruction of the home. ITN use was collected as a binary variable of whether the participant slept under an ITN the night before or not. Household structure data was collected as the materials used to construct the floor, walls, and roof. Floor, walls, and roof structure were collected as categorical variables based on the type of material as natural, rudimentary, or modern. For floors, natural materials were soil or mud, rudimentary materials were bamboo, wood, or sticks, and modern materials were cement, brick, or tile. For walls, natural materials were straw or mud, rudimentary materials were mud blocks, sticks, or metal, modern materials were cement, fired brick, or treated wood. For roofs, natural materials were straw or leaves, rudimentary materials were bamboo, wood, or sticks, and modern materials were zinc, metal, or asbestos. For analyses these were recategorized as rudimentary and modern, with natural and rudimentary materials being aggregated into a single category. Windows were collected as a categorical variable of whether they were open or able to close partially or fully. This variable was analyzed as whether windows were open or able to close or not (either partially or fully). Age and number of residents per household were analyzed as continuous variables.
Descriptive statistical analysis was conducted comparing the primary outcome variable (malaria infection by RDT) and primary exposure variable (household damage due to Cyclone Idai) by comparing proportions of those with and without malaria infection across the levels of household damage. Precision around these proportions was determined by 95% confidence intervals. Potential confounders were compared similarly across the levels of household damage, with proportions used for binary and categorical variables (ITN use, wall structure, roof structure, floor structure, windows closing, and head of household education level) and medians and interquartile ranges used for continuous variables (age and number of household residents).
Generalized estimating equations (GEE) logistic regression models were used to quantify the association between Cyclone Idai household damage and malaria risk. GEE models were chosen to account for household level correlation and non-independence of the primary exposure variable (household damage) and the shared environmental conditions related to the non-independence of the outcome variable (malaria infection). A univariable model of the association between Cyclone Idai household damage and malaria infection was built. Univariable models of the association between potential confounders (age, ITN use, wall structure, roof structure, floor structure, windows closing, head of household education level, and residents per household) were also constructed to quantify the association between these variables and malaria infection. As these variables were determined a priori as potential confounders, they were also included in the multivariable analysis regardless of the precision around the effect estimates.
A multivariable GEE logistic regression model was constructed to determine the relationship between Cyclone Idai household damage and malaria infection while adjusting for measured confounders. Odds ratios were used to determine the effect size of the association with 95% confidence intervals for the precision around these estimates.