Active Learning for Forecasting Severity among Patients with Post Acute Sequelae of SARS-CoV-2

TL;DR

This paper presents an active learning approach using a novel Active Attention Network and Llama-3.1-70B-Instruct features to improve prediction of PASC progression events, with fewer annotations and enhanced clinical decision-making.

Contribution

It introduces the first public PASC cohort with Llama-based text features and proposes an active attention network that combines human expertise for better risk prediction.

Findings

Enhanced prediction accuracy with fewer annotations

Successful integration of Llama-3.1-70B-Instruct features

Improved identification of progression events

Abstract

The long-term effects of Postacute Sequelae of SARS-CoV-2, known as PASC, pose a significant challenge to healthcare systems worldwide. Accurate identification of progression events, such as hospitalization and reinfection, is essential for effective patient management and resource allocation. However, traditional models trained on structured data struggle to capture the nuanced progression of PASC. In this study, we introduce the first publicly available cohort of 18 PASC patients, with text time series features based on Large Language Model Llama-3.1-70B-Instruct and clinical risk annotated by clinical expert. We propose an Active Attention Network to predict the clinical risk and identify progression events related to the risk. By integrating human expertise with active learning, we aim to enhance clinical risk prediction accuracy and enable progression events identification with fewer number of annotation. The ultimate goal is to improves patient care and decision-making for SARS-CoV-2 patient.