We conducted the study in Kasese District located in Western Uganda. Kasese is one of the districts where the Integrated Community Case Management for Childhood Illnesses including diarrhoea, malaria, and pneumonia were being implemented [11]. There were 139 health facilities in Kasese District, 81 health centre IIs, 48 health centre IIIs, 6 health centre IVs and 4 general hospitals. As per the revised package of basic health services for Uganda, 2014, severe pneumonia can only be managed at health centre IIIs, health centre IV, and general hospitals [12]. We specifically conducted the study in five health facilities reporting the highest number of severe pneumonia cases. These included Kagando Hospital, Bwera Hospital, Kasanga PHC Health Centre III, Kyarumba HC III, and St Francis of Assis-Kitabu Health Centre III (Fig. 1). These facilities offer both in-patient and outpatient childcare services.

Health facilities where study was conducted in Kasese District, Uganda, January to April 2023

We conducted an unmatched hospital-based case-control study between January to April 2023. We defined severe pneumonia cases as presentation with pneumonia with any of the following danger signs: oxygen saturation <90%, central cyanosis, severe respiratory distress, inability to drink or breastfeed or vomiting everything, altered consciousness, and convulsions in a child aged 2–59 months [13] attending five high-volume facilities in Kasese District from January to April 2023. A control was a child aged 2–59 months presenting with lower chest wall indrawing or fast breathing (respiratory rate ≥ 50 breaths per min if aged 2–11 months; ≥40 breaths per min if aged 12–59 months) and without signs of severe pneumonia at the respective health facilities during the same period as cases.

We determined the sample size using the Fleiss et al. formula [14] using Open Epi software. We assumed a two-sided confidence level (CI) = 95%, power = 80%, 1:1 ratio of cases to controls, taking non-exclusive breastfeeding as a main predictor for severe pneumonia with an odds ratio of 2.7 [15], and percentage of exposed controls at 9% [16]. We obtained a total sample of 153 cases and 153 controls. Controls were selected from those registered after a case at the same facility. We reviewed medical records at facilities and used an interviewer-administered questionnaire with caregivers in their homes to obtain information on clinical and non-clinical characteristics.

Using a structured questionnaire, we collected data regarding socio-demographics, child’s health, health-seeking behaviour, and environmental-related factors from both the cases and controls. We further reviewed medical records and child health cards to obtain more details on the clinical characteristics including the immunization status of the children.

Socio-demographic factors included age, sex, birth order, caregiver’s education level, and wealth status. We derived wealth status as a composite variable using principal component analysis of data on ownership of consumer items and livestock, characteristics of the dwelling unit, water sources and sanitation facilities.

Child health factors included nutritional status categorised into normal stunting, wasting, underweight and overweight. This was assessed by the following MUAC (in cm), weight for age Z score (WAZ), weight for height/length Z score (WHZ), and height for age Z score (HAZ). Breastfeeding status was categorised into exclusively breastfed for ≥ 6 months and those who were not as defined by WHO classification of breastfeeding [17]. Underlying illness was defined as having diarrhoea, measles, malaria or HIV during the current pneumonia illness. Immunisation status was defined as age-appropriate receipt of vaccines (including Pneumococcal Conjugate Vaccine (PCV), DPT-HebB + Hib2, measles and polio vaccines) as indicated in the Uganda Immunisation schedule [18].

Environmental factors included type of cooking fuel used at home, exposure to household air pollution from cooking fuel assessed using proximity to cooking combustion sources during home cooking with the child in the kitchen and exposure to tobacco smoke defined as staying with a cigarette smoker in the same household. Health-seeking behaviour factors included time to seeking care following caregiver’s recognition of illness, type of healthcare sought for the episode of illness prior to enrolment in the study and place where healthcare was sought for this episode of illness prior to enrolment in the study: home remedies, village health team, clinic, drug shop, health centre II, health centre III, and health centre IV.

We downloaded data in an Excel file and imported in STATA 16 software (StataCorp, Texas USA) for analysis. We tabulated categorical variables as frequencies with corresponding percentages. For continuous variables with a skewed distribution, such as the time to care-seeking following the caregiver’s recognition of illness, we reported medians along with their interquartile ranges (IQR). We used logistic regression to obtain independent variables significantly associated with having severe pneumonia. The independent variables with p-values ≤ 0.2 at bi-variate analysis were used to develop a multivariable logistic regression model using a forward stepwise approach. The strength of association between outcome variable and the independent variables of interest were assessed by calculating odds ratios (ORs) with 95% confidence intervals. Variables with p < 0.05 were considered statistically significant.