Prediction model for unfavorable treatment outcome for complicated sever acute malnutrition (SAM) in under five children admitted in hospitals at Amhara Region

TL;DR

This study creates a prediction model and risk score to identify under-five children with severe acute malnutrition likely to have poor treatment outcomes, helping clinicians prioritize care.

Contribution

The study introduces a validated prediction model and risk score for unfavorable treatment outcomes in SAM children in low-income settings.

Findings

The incidence of unfavorable treatment outcomes was 27.8% among hospitalized SAM children.

The model includes seven predictors like baseline edema, pneumonia, and anemia with an AUC of 91.3%.

Decision curve analysis showed higher net benefit for threshold probabilities above 21%.

Abstract

Severe acute malnutrition (SAM) affects 45 million children worldwide, with 14.89% of Ethiopian children under five suffering from it. This study validates a prediction model and develops risk scores for unfavorable treatment outcomes in SAM patients, addressing the scarcity of risk assessment tools in low-income settings and providing clinicians with a practical tool to improve decision-making.

A cohort study was conducted among 915 SAM children hospitalized with SAM hospitals in Amhara Region. Data analysis was conducted using STATA 17 and R 4.4.1. A lasso-selected multivariable model developed a nomogram for clinical utility. Model performance was assessed via AUC, calibration plot and validated with bootstrapping. Decision curve analysis evaluated the model’s clinical and public health utility.

The incidence of unfavorable treatment outcomes of SAM cases was 27.8% (95% CI: 25, 31). Majority of admitted children in stabilization center were complicated Severe Acute Malnutrition (cSAM) under-five children a magnitude of 89.52% (95% CI: 80.5–99.82). The developed nomogram comprised seven predictors: baseline Oedema, Diarrhea, CBC test results (Anemia), Pneumonia, Folic Acid supplementation, Vitamin A supplementation and IV antibiotic treatment. The AUC of the original model was 91.3% (95% CI: 89.0, 93.5), whereas the risk score model produced prediction accuracy of an AUC of 90.86 (95% CI: 88.6, 92.9). It was internally validated by bootstrapping method, and it has a relatively corrected discriminatory performance. Decision curve analysis indicated a higher net benefit compared to treating all or none, especially for threshold probabilities above 21%.

Our model and risk score demonstrate excellent discrimination and calibration, with minimal accuracy loss from the original, ensuring robust predictive performance. The models can have the potential to improve care and treatment outcomes in the clinical settings. Healthcare professionals prioritize the management of cSAM cases in children, particularly those presenting with baseline edema and co-morbidities such as pneumonia, anemia and diarrhea. Emphasis should be placed on timely interventions, including the administration of folic acid and Vitamin A supplementation, as well as intravenous antibiotics. Implementing a comprehensive care plan that addresses these factors will significantly improve treatment outcomes and enhance recovery in this vulnerable population.

Figures