Prediction of inpatient mortality in hospitalised children in low- and middle-income countries: An external validation of paediatric mortality risk scores.

Abstract

Background: Risk prediction tools for acutely ill children have been developed in high- and low-income settings, but few are validated or incorporated into clinical guidelines. We aimed to assess the performance of existing paediatric early warning scores for use in low- and middle-income countries using clinical data from a recent large multi-country study in Africa and South-Asia.

Methods: We used data (children across three nutritional strata) from the Childhood Acute Illness and Nutrition (CHAIN) Network cohort study (n = 3101). We assessed 10 scores where similar predictor variables were available in the CHAIN cohort. We evaluated performance using the area under the receiver operating curve (AUC) (primary outcome), sensitivity, specificity, positive and negative predictive value, and positive and negative likelihood ratio (secondary outcomes).

Results: Most scores showed poor discrimination, and all scores had low sensitivity. The paediatric early death index for Africa (AUC = 0.80; 95% confidence interval (CI) = 0.77-0.83), respiratory index of severity in children (AUC = 0.77; 95% CI = 0.74-0.81), and respiratory index of severity in children in Malawi (AUC = 0.78; 95% CI = 0.75-0.82) showed acceptable/good overall discrimination. Among children without wasting, most scores had acceptable/good performance, some even excellent. Poor discrimination was found for most scores among children with moderate and severe wasting or kwashiorkor.

Conclusions: All scores demonstrated lower validation performance than originally reported. Among children without wasting, most risk prediction scores performed acceptably whilst in malnourished children they performed poorly. There is a need for a malnutrition specific score. Further research is needed on specific actions in responding to scores. Integration into future guidelines will require acknowledging staffing, resources and workflows.