Effectiveness of a diagnostic algorithm for dengue based on an artificial neural network.

Area of Science:

• Medical Informatics
• Epidemiology
• Infectious Diseases

Background:

• Dengue presents a broad clinical spectrum, making early and accurate diagnosis challenging for healthcare professionals.
• Effective diagnostic tools are essential for timely intervention and improved patient outcomes in dengue cases.
• Artificial intelligence, specifically artificial neural networks (ANNs), offers potential for enhancing diagnostic capabilities.

Purpose of the Study:

• To evaluate the diagnostic effectiveness of an artificial neural network (ANN) algorithm for dengue.
• To compare the performance of an ANN algorithm against a direct diagnostic algorithm in an endemic area.

Main Methods:

• A single-center case-control study involving 233 dengue cases and 233 controls was conducted.
• Two algorithms were developed: a 'direct algorithm' based on official operational definitions and an ANN algorithm using the brain.js library.
• Diagnostic accuracy was assessed using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and Cohen's kappa index.

Main Results:

• The ANN algorithm achieved a sensitivity of 0.90, specificity of 0.82, NPV of 0.91, PPV of 0.81, and a kappa of 0.72.
• The direct algorithm demonstrated higher diagnostic accuracy with a sensitivity of 0.97, specificity of 0.96, NPV of 0.97, PPV of 0.96, and a kappa of 0.93.
• The direct algorithm significantly outperformed the ANN in all measured diagnostic parameters.

Conclusions:

• The direct algorithm, based on established operational definitions, proved superior to the artificial neural network (ANN) for dengue diagnosis in this study.
• Traditional diagnostic algorithms remain highly effective and may be preferable to ANNs for dengue diagnosis in resource-limited or endemic settings.
• Further research could explore hybrid approaches or optimized ANN models for dengue diagnosis.