Severity prediction markers in dengue: a prospective cohort study using machine learning approach.

Area of Science:

• Medical research
• Infectious diseases
• Clinical diagnostics

Background:

• Dengue virus infection presents a spectrum of illness, ranging from mild to severe complications.
• Prognostic indicators for severe dengue (SD) outcomes remain unclear, complicating early clinical management.
• Accurate prediction of dengue severity is crucial for effective patient care and resource allocation.

Purpose of the Study:

• To identify clinical and laboratory parameters predictive of severe dengue (SD) in adults.
• To evaluate the utility of statistical and machine learning (ML) models in predicting dengue severity.
• To establish early warning signs for severe dengue complications.

Main Methods:

• Analysis of clinical and laboratory data from 102 adult dengue patients, categorized into severe dengue (SD), warning signs, and no warning signs groups.
• Application of statistical models and machine learning (ML) algorithms to identify significant predictors of dengue severity.
• Examination of parameters during both early febrile and critical phases of the illness.

Main Results:

• Classical statistical analysis identified abnormal ultrasound findings, low lymphocyte counts, and platelet levels as significant indicators of SD during the febrile phase.
• During the critical phase, low creatinine, high sodium levels, and elevated AST/ALT were associated with SD.
• Machine learning models pinpointed AST/ALT levels and lymphocyte counts as key discriminators between severe and non-severe dengue cases.

Conclusions:

• Liver enzymes (AST/ALT), platelet counts, and ultrasound (USG) findings are significantly associated with severe dengue (SD).
• Early ultrasound testing and point-of-care quantification of AST/ALT may enable earlier prediction of SD.
• These findings can inform the development of improved diagnostic and prognostic tools for dengue management.