An exploration of mortality risk factors in non-severe pneumonia in children using population-based clinical data from Kenya.

Background

Childhood pneumonia is the leading infectious cause of mortality in children aged less than five years, with 96% of the incidences worldwide each year occurring in LMICs where based on severity, <13% of these cases are deemed life-threatening and require hospitalisation. Recent updates to guidelines provided by World Health Organisation’s to guide management of pneumonia recommend outpatient care for a population of children previously classified as high risk. This has been challenged by policy makers in Sub-Saharan Africa where pneumonia is often complicated by comorbidities such as malaria, diarrhoea, and malnutrition, and is associated with high mortality.

The aim of this study was, using robust modelling techniques, to identify factors that best discriminate mortality risk in non-severe pneumonia paediatric cases and explore whether these factors offer any added benefit over the current severity criteria used to admit paediatric pneumonia patients.

Methods and Findings

We undertook a retrospective cohort study of children aged 2 – 59 months admitted with pneumonia at 14 public hospitals in Kenya between February 2014 to February 2016. Using machine learning modelling techniques (partial least squares -discriminant analysis, support vector machines, random forests and elastic net) in addition to logistic regression - we analysed whether demographic characteristics (age and sex), clinical signs (e.g. respiratory rate, fever, pallor etc.) and common comorbidities (malaria, dehydration, and malnutrition) increased risk mortality in non-severe pneumonia in children. The topmost risk factors were applied to decision curve analysis to explore if using them in the decision to admit non-severe pneumonia cases in children had any net benefit above the current criteria (admit severe pneumonia).

Out of the 16,162 pneumonia cases admitted to hospital between February 2014 and February 2016, 11,318 were eligible for subsequent analysis. Inpatient mortality within this non-severe mortality group was 267 / 11225 (2.36%), with 93/11,318 (0.89%) missing data on mortality outcome. In terms of AUC, all models had moderately good performance ranging from 0.725 – 0.781. However, partial least squares discriminant analysis model offered the most improved performance as compared to logistic regression model in prediction of mortality in this patient population.

The four risk factors that were highly discriminative of mortality in children classified as having non-severe pneumonia were respiratory rate ≥ 70 breathes/minute, age < 12 months, comorbidities and child’s sex. These factors were consistent across the different models. From the decision curve analysis, we found that for a risk threshold probability ≥ 6%, there is a net benefit to admitting the patient sub-population with these features as an additional criterion in addition to those with severe pneumonia. Sensitivity analyses indicated that the overall results are not significantly affected by variations in pneumonia severity classification criteria.

Conclusions

Non-severe pneumonia in patients aged 2-11 months with respiratory rate ≥ 70 breaths/minute and are comorbid have a risk of mortality comparable to those with severe pneumonia. This relatively high-risk group ought to be considered for inpatient care that is provided to severe pneumonia patients.