INCLAIR: Inception-Based Longitudinal Clinical Anomaly Detection with Informed Reasoning

2026-07-29

AI Summary

This work addresses the challenges of anomaly detection in longitudinal clinical data, where anomalies are sparsely evidenced, patient histories vary substantially in length, and expert annotations are costly to obtain. The authors propose a novel approach that integrates multi-historical context scoring, profile-level evidence aggregation, and natural language explanation generation under limited supervision. To manage computational complexity, they introduce a U-statistic–based subsequence scoring mechanism coupled with an incomplete subset approximation. Furthermore, they mitigate local anomaly attenuation through an Inception-style architecture, variance decomposition of U-statistics, and top-k pooling. Evaluated on three real-world clinical datasets, the method significantly outperforms existing techniques and demonstrates strong clinical interpretability and alignment with expert assessments, as illustrated by a steroid profile case study.

Abstract

Detecting anomalies in longitudinal clinical profiles is clinically important but difficult: abnormal evidence is often sparse, patient histories have unequal length, and expert explanations are costly. We propose INCLAIR, a framework that scores each observation against multiple historical contexts, aggregates evidence at the profile level, and generates grounded natural-language explanations under limited expert supervision. Under stated within-profile exchangeability assumptions, the complete mean subsequence score takes an order-$l$ U-statistic form, yielding a variance decomposition and an incomplete-subset approximation that controls combinatorial inference cost independently of profile length. The same analysis shows that mean aggregation attenuates localized anomalies by a factor set by the anomaly support and profile length, motivating validation-selected top-$k$ pooling. Across three clinical datasets, INCLAIR consistently outperforms state-of-the-art baselines. We further validate practical relevance through a case study on longitudinal steroid profiles, comparing INCLAIR's predictions and explanations against domain-expert assessments supported by DNA analysis. The results show that INCLAIR enables clinically actionable anomaly detection under limited expert supervision.

Problem

• longitudinal clinical anomaly detection
• anomaly detection
• clinical profiles
• expert supervision
• sparse abnormal evidence

Innovation

• longitudinal anomaly detection
• U-statistic
• incomplete-subset approximation
• top-k pooling
• explainable AI