Akshat Dasula

Akshat Dasula, Prasanna Desikan, Jaideep Srivastava

Incomplete or inconsistent discharge documentation is a primary driver of care fragmentation and avoidable readmissions. Despite its critical role in patient safety, auditing discharge summaries relies heavily on manual review and is difficult to scale. We propose an automated framework for large-scale auditing of discharge summaries using locally deployed Large Language Models (LLMs). Our approach operationalizes core transition-of-care requirements such as follow-up instructions, medication history and changes, patient information and clinical course, etc. into a structured validation checklist of questions based on DISCHARGED framework. Using adult inpatient summaries from the MIMIC-IV database, we utilize a privacy-preserving LLM to identify the presence, absence, or ambiguity of key documentation elements. This work demonstrates the feasibility of scalable, automated clinical auditing and provides a foundation for systematic quality improvement in electronic health record documentation.

Akshat Dasula, Prasanna Desikan, Jaideep Srivastava

Large language models (LLMs) are increasingly applied to structured clinical data, yet whether they can recognize the limits of their own knowledge on such tasks remains unexplored. We study this question through the lens of cross-model attribution divergence with the goal of reducing epistemic uncertainty for structured tasks, comparing Qwen 2.5 7B and XGBoost on a prediction task via attribution divergence analysis. We report four findings. First, LLM verbalized confidence is epistemically vacuous, it outputs a near-constant (0.856-0.937) regardless of whether accuracy is 49% or 75.3%, tracking prompt format rather than prediction quality. Second, the LLM exhibits an inverse difficulty effect: accuracy drops to 64.8% when XGBoost is 99% correct, but matches XGBoost (73.8% vs. 73.1%) when it is moderately uncertain. Third, few-shot examples and SHAP-derived feature evidence are orthogonal, super-additive interventions: they reduce the Attribution Disagreement Score (ADS) from 1.54 to 0.38 and improve accuracy from 49% to 75.3% without training. Fourth, a cross-model calibrator that determined LLM reliability using attribution divergence signals reduces expected calibration error from 0.254 to 0.080, replacing uninformative verbalized confidence with patient-specific reliability estimates, without accessing model internals or requiring repeated inference. We frame these findings as a cold start problem for LLMs on structured data and outline a path toward genuine epistemic self-awareness.