Development and validation of a simple machine learning tool to predict mortality in leptospirosis

Area of Science:

• Infectious Diseases
• Medical Informatics
• Public Health

Background:

• Leptospirosis poses a significant global health challenge, with high mortality rates in severe cases.
• Accurate prediction of mortality risk is crucial for timely intervention and resource allocation.
• Existing scoring systems like SPIRO and quick SOFA have limitations in predicting leptospirosis outcomes.

Purpose of the Study:

• To develop and validate a novel, accurate, and simple predictive tool for mortality in leptospirosis patients.
• To compare the performance of the new tool against existing scoring systems.
• To identify key risk factors associated with leptospirosis mortality.

Main Methods:

• A machine-learning approach, specifically Lasso regression, was employed using admission data from 295 leptospirosis patients.
• A predictive model, LeptoScore, was derived from the Lasso regression coefficients.
• A simplified score, QuickLepto, was developed using key predictors with high importance values.

Main Results:

• The Lasso regression model achieved an Area Under the Receiver Operating Characteristic Curve (AUC-ROC) of 0.776 for mortality prediction.
• The simplified QuickLepto score, comprising five variables (age > 40 years, lethargy, pulmonary symptoms, MAP < 80 mmHg, hematocrit < 30%), demonstrated strong predictive accuracy (AUC-ROC = 0.788).
• Both LeptoScore and QuickLepto outperformed the SPIRO score (AUC-ROC = 0.500) and showed comparable or better accuracy than the quick SOFA score (AUC-ROC = 0.782).

Conclusions:

• The QuickLepto score is a simple, accurate, and valuable tool for predicting in-hospital mortality in leptospirosis patients upon admission.
• This new scoring system can aid clinicians in rapidly identifying high-risk individuals, facilitating prompt initiation of critical supportive care.
• The development of QuickLepto addresses the need for improved risk stratification in leptospirosis management.