SISA

SISA predicts the likelihood of hospitalization for individuals suspected of arboviral infections (dengue, chikungunya, and Zika) using intake demographic and symptom data.

Key Features:

• Prediction target: Predicts hospitalization outcome for patients with suspected arboviral infections (dengue, chikungunya, and Zika).
• Input data: Uses subject intake demographic and symptom data collected at sentinel medical centers in Machala, Ecuador between November 2013 and September 2017.
• SISAL (extended model): SISAL (Severity Index for Suspected Arbovirus with Laboratory) incorporates additional laboratory data to improve prediction accuracy.
• Algorithms evaluated: Seven different machine learning algorithms were evaluated to identify the optimal predictive models.
• SISA model algorithm and performance: The SISA model uses a generalized boosting algorithm and achieved an area under the receiver operating characteristic curve (AUC) of 0.91.
• SISAL model algorithm and performance: The SISAL model uses an elastic net algorithm and achieved an AUC of 0.94.
• Sensitivity analysis: A sensitivity analysis indicated SISA and SISAL are not directly comparable due to differences in data inputs and model structures.

Scientific Applications:

• Clinical decision support: Assist clinical decision-making regarding hospitalization for patients presenting with suspected dengue, chikungunya, or Zika.
• Triage and resource allocation: Support triage and allocation of hospital resources in settings where laboratory diagnostics may be limited.
• Clinical management insights: Provide data-driven insights to inform clinical management strategies for arboviral infections.

Methodology:

Seven machine learning algorithms were evaluated and a generalized boosting algorithm was selected for SISA while an elastic net algorithm was selected for SISAL based on predictive performance using subject intake (demographic and symptom) data, with SISAL incorporating additional laboratory data.