Study setting
The outbreak occurred in a tri-district area in Central Uganda comprising Lyantonde, Lwengo, and Rakai districts. The estimated total population was approximately 926,000 (101,200 in Lyantonde, 281,400 in Lwengo, and 543,400 in Rakai), based on the 2017 projected populations from the 2014 census (10). The three districts border each other and share several public hospitals in Lyantonde District.
Case definition and case finding
We defined a probable case as sudden onset of fever and generalized rash in a resident of the tri-district area from 1 June to 30 September 2017, plus ≥1 of the following: coryza, conjunctivitis, or cough. A confirmed case was a probable case with measles-specific IgM positivity.
For case-finding, we reviewed outpatient and inpatient records at health facilities in the tri-district area, and actively searched for cases with the help of members of village health teams and community leaders. We collected data on patients’ symptoms, onset dates of symptoms, treatment outcomes, demographic characteristics, place of residence, receipt of care, and vaccination status.
Descriptive epidemiology
We analyzed the line-listed cases by onset of symptoms, age, sex, and place of residence. To calculate attack rates (AR) by age and sex, we used the estimated population in the tri-district area, provided by Uganda Bureau of Statistics (9). We used a choropleth map to describe the ARs by sub-county for Lyantonde District, where most (68%) of the cases came from.
Hypothesis generation
Using a semi-structured questionnaire, we interviewed a convenience sample of 15 caretakers for case-patients regarding their potential exposures during their likely exposure period (i.e., 7-21 days before their rash onset, or between minimum and maximum incubation periods). The exposures of interest included attending social gatherings, attending worship places, visiting health facilities, visiting communal gathering points, and immunization status. We generated hypotheses about exposures based on findings from the descriptive epidemiology analysis and hypothesis-generation interviews.
Case-control study
We conducted a case-control study to test the hypothesis on potential exposures. At the time of the case-control study, 38 case-patients were line-listed. We recruited 34 of those case-patients aged ≥1 year to participate in the case-control study. For each case, we selected 4 controls in the same immediate neighborhood (i.e., within three homes of the case-patient’s) who had no measles symptoms from 1 July to 30 September 2017. We individually matched controls to the case by age (±2 years). We assessed the exposure risk factors for both the case-patient and the matched controls during the case-patient’s likely exposure period, using a structured questionnaire. Vaccination status was determined by either reviewing the vaccination records or, if unavailable, by asking whether the child had received an injection on the upper arm at 9 months of age (which is the standard practice for measles vaccine in Uganda). Cases and controls were considered vaccinated only if they were vaccinated prior to the onset of the outbreak. We also collected data on demographic characteristics (e.g., age and sex) of both case-patients and controls.
To account for individual matching in the study design, we used conditional logistic regression to analyze the data, using the matched set as the matching variable. We first assessed the association between each individual risk factor and measles. Risk factors that were statistically significant at the p<0.05 level during the univariate analysis were included in the multivariate conditional logistic regression model to calculate the adjusted odds ratios (ORadj) and their associated 95% confidence intervals (CI). Non-significant variables in the multivariable model (p≥0.05) were backward-eliminated until all were significant. Multivariate logistic regression analysis was used to control for potential confounding variables that could be included in the model by holding all the other variables constant.
Estimation of vaccine effectiveness (VE) and vaccination coverage (VC)
We estimated measles VE using the following formula (11):
VE = 1 – ORadj
where ORadj is the odds ratio associated with having been vaccinated for at least one dose of measles vaccine, adjusted for risk factors that were significant during the univariate analysis, using conditional logistic regression.
We estimated the VC using the percentage of controls with a history of measles vaccination in the case-control study, assuming the controls to be representative of the general population. We also obtained the administrative data from the district surveillance officer on VC in Lyantonde District.
Environmental assessment
We observed the layout of the pediatric department at Lyantonde Hospital, especially the pediatric wards in question, and examined the ventilation system. We also reviewed the patient log to assess the type of illnesses admitted in the pediatric department.