Predicting death from kala-azar: construction, development, and validation of a score set and accompanying software

• Published in: Revista da Sociedade Brasileira de Medicina Tropical
• Main Authors: Dorcas Lamounier Costa, Regina Lunardi Rocha, Eldo de Brito Ferreira Chaves, Vivianny Gonçalves de Vasconcelos Batista, Henrique Lamounier Costa, Carlos Henrique Nery Costa
• Format: Article in Journal/Newspaper
• Language: English
• Published: Sociedade Brasileira de Medicina Tropical (SBMT)
• Subjects: Kala-azar; Visceral leishmaniasis; Prediction; Mortality; Software; Arctic medicine. Tropical medicine; RC955-962; Arctic; Azar
• Online Access: https://doi.org/10.1590/0037-8682-0258-2016; https://doaj.org/article/ced3a9dbb5cb4dc8acc3ef48449228b5

Abstract INTRODUCTION Early identification of patients at higher risk of progressing to severe disease and death is crucial for implementing therapeutic and preventive measures; this could reduce the morbidity and mortality from kala-azar. We describe a score set composed of four scales in addition to software for quick assessment of the probability of death from kala-azar at the point of care. METHODS: Data from 883 patients diagnosed between September 2005 and August 2008 were used to derive the score set, and data from 1,031 patients diagnosed between September 2008 and November 2013 were used to validate the models. Stepwise logistic regression analyses were used to derive the optimal multivariate prediction models. Model performance was assessed by its discriminatory accuracy. A computational specialist system (Kala-Cal(r)) was developed to speed up the calculation of the probability of death based on clinical scores. RESULTS: The clinical prediction score showed high discrimination (area under the curve [AUC] 0.90) for distinguishing death from survival for children ≤2 years old. Performance improved after adding laboratory variables (AUC 0.93). The clinical score showed equivalent discrimination (AUC 0.89) for older children and adults, which also improved after including laboratory data (AUC 0.92). The score set also showed a high, although lower, discrimination when applied to the validation cohort. CONCLUSIONS: This score set and Kala-Cal(r) software may help identify individuals with the greatest probability of death. The associated software may speed up the calculation of the probability of death based on clinical scores and assist physicians in decision-making.