Post-discharge mortality prediction in under 5s with acute infectious diseases: a prospective cohort study

Abstract

Background: Acute infectious diseases are an important contributor to under-5 mortality. Mortality following discharge is an important but poorly recognised contributor to overall mortality. The identification of at-risk children is critical in developing efficient and effective post-discharge interventions. The objective of this study was to derive a model of post-discharge mortality after acute infectious illness. Methods: This prospective observational cohort study was conducted at two hospitals in Mbarara, Uganda, between March, 2012, and December, 2013. We included children aged between 6 months and 60 months who were admitted with a proven or suspected infection. Baseline clinical, laboratory, and sociodemographic variables were collected at admission. Children received usual care during their admission and received follow-up to 6 months after discharge to determine vital status. Primary outcome was death at 6 months. We modelled candidate predictor variables against the outcome of death at 6 months using logistic regression. The most promising (p<0·05) candidate predictors were incorporated into a multivariable logistic regression model using a stepwise backwards selection process balancing Aikaike's information criterion, area under the receiver operator curve (AUC), and parsimony. Findings: We enrolled 1307 consecutive participants over the study period. During hospitalisation, 65 (5·0%) participants died, thus there were 1242 live discharges. During follow-up we noted 61 deaths (4·9%), of which 31 (51%) occurred within the first 30 days. The follow-up rate was 98·5%. Age, mid-upper arm circumference, admission temperature, admission oxygen saturation, admission systolic blood pressure, length of hospital stay, previous hospitalisation within 7 days, abnormal Blantyre coma score, duration of illness before admission, parasitaemia, and HIV status were identified in the univariate analysis as being associated with post-discharge mortality. The final adjusted model included the variables mid-upper arm circumference (OR 0·95 [95%CI 0·94–0·97] per 1 mm increase), time since last hospitalisation (0.76 [0·61–0·93] for each increased period of no hospitalisation, categorized as <7 days, 7–30 days, 30–365 days, and never), oxygen saturation (0·96 [0·94–0·99] per 1% increase), abnormal Blantyre coma score (2·41 [1·19–4·87]), and HIV positive status (2·67 [1·19–6·00]). This model produced a receiver operating characteristic curve with an AUC of 0·815 (p<0·0001). Using a probability cut-off of 3·5%, our model would have a sensitivity of 80% (95% CI 70–90) and specificity of 65% (95% CI 62–68). Approximately 35% of children would be identified as high risk (10% mortality risk) and the remaining would be classified as low risk (1·5% mortality risk), in a cohort similar to this study cohort. Interpretation: A simple prediction tool that uses five easily collected admission variables could be used to identify children at high risk of death after discharge. Improved discharge planning and post-discharge care could be provided for these high-risk children. Further external validation of this model is required before implementation. Funding: Center for International Child Health, BC Childrens Hospital, Vancouver, BC, Canada.