Lightweight Language Models are Prone to Reasoning Errors for Complex Computational Phenotyping Tasks

TL;DR

Lightweight language models frequently make reasoning errors in complex computational phenotyping tasks, and an expanded evaluation framework helps identify and analyze these faults to improve model reliability.

Contribution

This work introduces an extension to the PHEONA framework to evaluate reasoning errors in lightweight LLMs during complex phenotyping tasks, highlighting the prevalence of reasoning faults.

Findings

Reasoning errors are common across all tested models.

Prompt modifications have limited impact on reducing errors.

DeepSeek showed the smallest accuracy decline after prompt changes.

Abstract

Objective: Although computational phenotyping is a central informatics activity with resulting cohorts supporting a wide variety of applications, it is time-intensive because of manual data review. We previously assessed the ability of LLMs to perform computational phenotyping tasks using computable phenotypes for ARF respiratory support therapies. They successfully performed concept classification and classification of single-therapy phenotypes, but underperformed on multiple-therapy phenotypes. To understand issues with these complex tasks, we expanded PHEONA, a generalizable framework for evaluation of LLMs, to include methods specifically for evaluating faulty reasoning. Materials and Methods: We assessed the responses of three lightweight LLMs (DeepSeek-r1 32 billion, Mistral Small 24 billion, and Phi-4 14 billion) both with and without prompt modifications to identify explanation correctness and unfaithfulness errors for phenotyping. Results: For experiments without prompt modifications, both errors were present across all models although more responses had explanation correctness errors than unfaithfulness errors. For experiments assessing accuracy impact after prompt modifications, DeepSeek, a reasoning model, had the smallest overall accuracy impact when compared to Mistral and Phi. Discussion: Since reasoning errors were ubiquitous across models, our enhancement of PHEONA to include a component for assessing faulty reasoning provides critical support for LLM evaluation and evidence for reasoning errors for complex tasks. While insights from reasoning errors can help prompt refinement, a deeper understanding of why LLM reasoning errors occur will likely require further development and refinement of interpretability methods. Conclusion: Reasoning errors were pervasive across LLM responses for computational phenotyping, a complex reasoning task