SAM is a significant public health problem in many developing countries, including Ethiopia. Various factors determine the treatment outcome of these children admitted to the stabilization centers. In this study, we developed and internally validated a model to predict mortality (the most devastating outcome of SAM) among children younger than 59 months who were admitted to the in-patient treatment centers. Predicting the risk of mortality among SAM children is essential to provide appropriate care/clinical decisions (admitting high-risk children in ICU) that can avert the risk of death among those children having the high predictive probability of mortality.
The incidence of mortality among these children throughout the course of treatment was 14.03% (11.6, 16.8), which is higher than the expected death rate from SAM according to the international sphere standard's reference (6). In our study, a combination of eight factors that related to the child demographic characteristics (age), medical complications (HIV/AIDS, heart failure, and anemia), type of SAM (children with mixed forms of SAM), and the medical care being provided (children who didn’t receive antibiotics, folic acid, and vitamin A) predict mortality among SAM children with a classification accuracy (AUROC) of 0.81, which is a good accuracy according to diagnostic accuracy classification. Likewise, our model consists of easily obtainable information in routine clinical practice to be used by both mid and lower-level health professionals in the primary care settings. Almost all (8) of the characteristics can be found registered in the medical recording of SAM children and/or from history taking.
A study by Van den Brink and his colleagues predict mortality among SAM children by fecal volatile organic compound (VOC) analysis through grouping VOC profiles according to other important clinical characteristics such as degree of edema, medical complications, comorbidities, antibiotic prescription, HIV, very low anthropometry, and age with an AUC of 0.71 (23). The former study predicts mortality with fair accuracy, which is lower than the current study. This could be due to the addition of different clinical parameters in the current model that have an additive effect in impairing the normal physiology of SAM children, resulting in greater AUC in our study. Moreover, examining the existence of VOC in the fecal components of SAM children may not be easily obtainable information in routine clinical and public health practice, particularly in a low-resource setting, which makes the model less practical.
This study has different strengths. Firstly, we used an adequate number of participants with the outcome (mortality) that ease the process to construct the model using arguably enough predictor variables. Secondly, we internally validated our model using bootstrapping technique resulting in a small optimism coefficient, indicating our model is less sample dependent and not overfitted. Thirdly, our prediction model is constructed from easily obtainable demographic and clinical characteristics of the children that make it applicable in primary care settings. However, the following limitations of this study should be considered while interpreting the findings. External validation of the model would be required before using it for the clinical decision-making process in another context. Although we have an adequate number of participants with the outcome, having a relatively low number of overall participants did not enable us to validate the model in separate datasets. Lastly, since the data were extracted from the medical recording of SAM patients, some deviation in data quality is expected. Nevertheless, the model will provide its maximum benefit provided that all the required predictor information is collected.
The study implies that providing special care/clinical decisions (admitting high-risk children in ICU) using the model has a higher net benefit than not providing special care (not admitting all SAM children in the ICU) regardless of their risk threshold.