Aetiology of hospitalized fever and risk of death at Arua and Mubende tertiary care hospitals in Uganda from August 2019 to August 2020 - BMC Infectious Diseases - BMC Infectious Diseases

Study design

From August 2019 to August 2020, we recruited adults 18 years of age or older presenting to Mubende and Arua Regional Referral Hospitals (RRHs) emergency or outpatient departments designated to be hospitalized with a measured or reported temperature ≥ 38.0 °C occurring within the past 48 h or a clinical history consistent with fever within 48 h of presentation. There were 255 participants consecutively screened with 97 enrolled at Mubende RRH and 35 enrolled at Arua RRH (Additional file 1: Figure S1). Participants with COVID-19 diagnoses were in separate designated areas, required scarce personal protective equipment, and were not enrolled. Exclusion criteria included being hospitalized for ≥ 72 h, receipt of antibiotics, previous participation in the study, being a prisoner, and being a psychiatric patient.

Study setting

The two tertiary care hospitals were chosen as study sites because both sites are located in regions with different ecology and serve different populations. Mubende RRH (bed capacity 173) is located in central Uganda and serves a district population of over 688,000 people but the catchment area goes beyond the district it serves [17]. Mubende is situated on a major road between Democratic Republic of the Congo and the capital of Uganda, Kampala. Arua RRH (bed capacity 323) is located in the more arid northwest Uganda (proximal to borders with Democratic Republic of the Congo and with South Sudan), serves over 785,000 people in the districts, and receives referrals from neighbouring countries [18].

Data collection

After informed consent, participants provided information about sociodemographic and comorbid medical conditions at enrolment. History suggestive of exposure to zoonotic infections was obtained including area of residence, occupation, environmental exposures, and healthcare exposure [19]. Also, recent antibiotic use was documented, relying on patient self-report or any clinical documents they possessed. Vital signs were recorded including temperature, heart rate, respiratory rate, blood pressure, and pulse oximetry; a Glasgow Coma Score (GCS) was calculated. Participants were prospectively followed until hospital discharge. Inpatient antibiotic usage, vital signs monitoring, intensive care unit admission, length of stay, timing of receipt of infectious disease test results, discharge diagnosis, and vital status at discharge were recorded. Upon discharge, participants returned after 3–8 weeks and were interviewed on duration of illness and antimicrobial use and serum, whole blood, and plasma stored for future assessment of seroconversion. If participants were unable to follow-up in-person, they were contacted via telephone to determine vital status.

Clinical laboratory testing

Laboratory tests that were conducted included complete blood count, serum creatinine, serum alanine transferase (ALT), serum aspartate transferase (AST), blood cultures (one aerobic bottle, Bactec 9050, BD, NJ, United States), hepatitis A IgM (Vaxpert, Inc., Deira Dubai, UAE), hepatitis B sAg (SD Bioline, Abbott diagnostics, Gyeonggi-do, Republic of Korea), and HIV testing with consent (Determine HIV1/2: Abbott, OK, USA; Stat-Pak, Chembio Diagnostics Systems, Inc, NY, USA; SD Bioline: Gyeonggi-do, Republic of Korea), malaria rapid test (CareStart malaria Pf HRP2 antigen RDT), and thick malaria smears. A malaria smear result was stratified by any parasitaemia or ≥ 3000 parasites/μL [20, 21]. If there was clinical suspicion, testing for tuberculosis was performed using PCR sputum testing (Xpert MTB/RIF Ultra, Cepheid, Sunnyvale, CA, USA). If HIV testing was positive, serum cryptococcal antigen, urine lipoarabinomannan (LAM; Alere Waltham, MA, United States), and plasma CD4 count (BD Facs Calibur, BD, NJ, United States) were measured. HIV testing and counselling were performed per Uganda's national guidelines.

Data analysis

We estimated that a target sample size of 365 patients across sites would estimate the frequency of uncommon endemic viral infections with a 5% prevalence using a precision of 0.025 at 95% confidence. Here we report initial findings from an accrued sample size of 132 participants enrolled to date to provide clinical epidemiologic estimates, particularly in the context of the development of the Ebola virus disease outbreak in Mudende district, Uganda that started in September 2022. Summary statistics were performed for baseline demographics and prevalence of difference diagnoses. Baseline demographics, symptom prevalence, physiologic parameters, clinical laboratory results, quick sequential organ failure assessment (qSOFA; range, 0 [best] to 3 [worst] points) scores [22], and subsequent diagnoses made by the research physicians on-site were summarized and stratified by HIV diagnosis and the presence of a laboratory-confirmed acute infectious disease diagnosis. We stratified these descriptive statistics between those living with HIV or were newly diagnosed with HIV and those that were HIV-negative. Kaplan–Meier curve of survival stratified by HIV status. The logrank test was also performed to evaluate for a survival different with HIV. Individuals lost to follow-up after discharge were censored at hospital discharge. After checking the proportional hazards assumption, exploratory univariate Cox regression was performed to evaluate for a survival difference associated with baseline demographics, HIV status (including both new and known), physiologic parameters, and laboratory parameters. The qSOFA (quick Sepsis Related Organ Failure Assessment) score dichotomized at a score of 2 and components of the qSOFA score were also independently fit. Dichotomized parameters were used at qSOFA score cut offs (i.e., respiratory rate > 21 breaths per minute, Glasgow coma scale < 15, and systolic blood pressure 100 mmHg) [22]. Exploratory bivariate model p-values were not corrected for multiple comparisons due to a limited sample size. Adjusted regression models were performed using parameters with bivariate regression model p-values < 0.002, a Bonferroni correction cutoff. Harrell's C-statistic was calculated to determine the performance of these unadjusted and adjusted final models [23]. The sample set was not split into derivation and validation sets due to limited sample size. Analyses were performed using Stata, version 16.0 (StataCorp LLC, College Station, TX, USA), and figures were created using Stata or R, version 4.0.1 (R Foundation). A map of districts represented was created using ArcGIS ArcGIS Online (Redlands, CA) [24].