Mortality prediction models developed in high-income country intensive care units (ICUs) may perform poorly in low–middle-income country (LMIC) ICUs due to differences in case mix and provision of life-supporting interventions. This study aims to develop a novel prognostic score and investigate its precision, discrimination, and calibration properties in predicting ICU mortality. Predictor variables were identified by reviewing relevant published studies from LMICs. The most frequently cited variables were selected to develop A Mortality Predictor Score (AMPS). Subsequently, the tool was assessed for accuracy, discrimination, and calibration using the Brier score, area under receiver operating characteristic curve (AUROC), reliability diagram and Decision curve analysis by using data gathered from the ICU of Black Lion Hospital between September 2019 and September 2020. Its prognostic ability was compared with the mortality prediction model II (MPM II) and ICU priority level. P p-value less than or equal to 0.05 was considered significant. The commonly identified model variables were Altered Mental status, Mechanical ventilation, More than two organ systems diagnosis, Pressure (systolic) less than 90, Potentially irreversible condition, Suspected infection, and Severe hypoalbuminemia (serum albumin <2 g/dl) (A(MPS)2. Each of these variables was found to be predictors of mortality when tested on 265 ICU patients admitted to the Adult Intensive Care Unit of Black Lion Hospital. A(MPS)2 was able to correctly predict mortality in 86.4% of the cases with a sensitivity of 87.6% and specificity of 84.6%. The area under the curve for the predictor scores was AMPS (Area 0.92; CI 0.88–0.93), MPM II score (Area 0.86; CI 0.82–0.91), and ICU priority level (Area 0.76; CI 0.68–0.83). AMPS is a promising prognostic score for ICU mortality in low-resource settings. External validation and comparison to other scores are needed.
Debebe et al. (Thu,) studied this question.