Prediction models of COVID-19 fatality in nine Peruvian provinces: A secondary analysis of the national epidemiological surveillance system

Abstract

There are initiatives to promote the creation of predictive COVID-19 fatality models to assist decision-makers. The study aimed to develop prediction models for COVID-19 fatality using population data recorded in the national epidemiological surveillance system of Peru. A retrospective cohort study was conducted (March to September of 2020). The study population consisted of confirmed COVID-19 cases reported in the surveillance system of nine provinces of Lima, Peru. A random sample of 80% of the study population was selected, and four prediction models were constructed using four different strategies to select variables: 1) previously analyzed variables in machine learning models; 2) based on the LASSO method; 3) based on significance; and 4) based on a post-hoc approach with variables consistently included in the three previous strategies. The internal validation was performed with the remaining 20% of the population. Four prediction models were successfully created and validate using data from 22,098 cases. All models performed adequately and similarly; however, we selected models derived from strategy 1 (AUC 0.89, CI95% 0.87-0.91) and strategy 4 (AUC 0.88, CI95% 0.86-0.90). The performance of both models was robust in validation and sensitivity analyses. This study offers insights into estimating COVID-19 fatality within the Peruvian population. Our findings contribute to the advancement of prediction models for COVID-19 fatality and may aid in identifying individuals at increased risk, enabling targeted interventions to mitigate the disease. Future studies should confirm the performance and validate the usefulness of the models described here under real-world conditions and settings.

Fig 1. Validation of the selected models for the prediction of death in confirmed cases of COVID-19.

(A) Performance of the model derived from strategy 1 and 4 in the dataset used for the model construction (N = 17 678) and (B) performance of the model derived from strategy 1 and 4 in the dataset for validation (N = 4 420).