Abstract
INTRODUCTION: The development of simple clinical tools to identify children at risk of death would enable rapid and rational implementation of lifesaving measures to reduce childhood mortality globally. METHODS: We evaluated the ability of three clinical scoring systems to predict in-hospital mortality in a prospective observational study of Ugandan children with fever. We computed the Lambaréné Organ Dysfunction Score (LODS), Signs of Inflammation in Children that Kill (SICK), and the Pediatric Early Death Index for Africa (PEDIA). Model discrimination was evaluated by comparing areas under receiver operating characteristic curves (AUCs) and calibration was assessed using the Hosmer-Lemeshow goodness-of-fit test. Sub-analyses were performed in malaria versus non-malaria febrile illness (NMFI), and in early (≤48 hours) versus late (>48 hours) deaths. RESULTS: In total, 2089 children with known outcomes were included in the study (99 deaths, 4.7% mortality). All three scoring systems yielded good discrimination (AUCs, 95% confidence interval (CI): LODS, 0.90, 0.88 to 0.91; SICK, 0.85, 0.83 to 0.86; PEDIA, 0.90, 0.88 to 0.91). Using the Youden index to identify the best cut-offs, LODS had the highest positive likelihood ratio (+LR, 95% CI: LODS, 6.5, 5.6 to 7.6; SICK, 4.4, 3.9 to 5.0; PEDIA, 4.4, 3.9 to 5.0), whereas PEDIA had the lowest negative likelihood ratio (-LR, 95% CI: LODS, 0.21, 0.1 to 0.3; SICK, 0.22, 0.1 to 0.3; PEDIA, 0.16, 0.1 to 0.3), LODS and PEDIA were well calibrated (P = 0.79 and P = 0.21 respectively), and had higher AUCs than SICK in discriminating between survivors and non-survivors in malaria (AUCs, 95% CI: LODS, 0.92, 0.90 to 0.93; SICK, 0.86, 0.84 to 0.87; PEDIA, 0.92, 0.90 to 0.93), but comparable AUCs in NMFI (AUCs, 95% CI: LODS, 0.86, 0.83 to 0.89; SICK, 0.82, 0.79 to 0.86; PEDIA, 0.87, 0.83 to 0.893). The majority of deaths in the study occurred early (n = 85, 85.9%) where LODS and PEDIA had good discrimination. CONCLUSIONS: All three scoring systems predicted outcome, but LODS holds the most promise as a clinical prognostic score based on its simplicity to compute, requirement for no equipment, and good discrimination.