Development, evaluation, and validation of machine learning models for COVID-19 detection based on routine blood...

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Hematology

Background:

• The real-time reverse transcription polymerase chain reaction (rRT-PCR) test is the gold standard for COVID-19 detection but has limitations.
• These limitations include long turnaround times, reagent shortages, significant false-negative rates (15-20%), and high equipment costs.
• Routine hematochemical blood test values present a potentially faster and more cost-effective alternative for COVID-19 diagnosis.

Purpose of the Study:

• To develop and validate machine learning (ML) models for identifying COVID-19 positive patients using hematochemical values from routine blood exams.
• To assess the performance of ML models trained on different feature sets derived from blood tests.

Main Methods:

• Developed five ML models using three distinct training datasets from 1,624 patients (52% COVID-19 positive).
• Datasets included a comprehensive set (72 features: CBC, biochemical, coagulation, hemogasanalysis, CO-Oxymetry, demographics, symptoms), a COVID-specific set (32 features), and a Complete Blood Count (CBC) set (21 features).
• Validated models using internal-external (58 cases) and external (54 negative cases) validation sets.

Main Results:

• ML models achieved high performance across datasets, with Area Under the Curve (AUC) ranging from 0.83 to 0.90 for the complete OSR dataset.
• AUC values for the COVID-specific and CBC datasets ranged from 0.83-0.87 and 0.74-0.86, respectively.
• Validation results demonstrated good performance, with AUC from 0.75 to 0.78 and specificity from 0.92 to 0.96.

Conclusions:

• Machine learning applied to blood tests can serve as an adjunct or alternative to rRT-PCR for rapid and cost-effective COVID-19 identification.
• This approach is particularly beneficial for developing countries or regions experiencing surges in infections.
• Hematochemical analysis via ML offers a promising strategy for improving COVID-19 diagnostic accessibility and efficiency.