Rema Padman

Yubo Li, Rema Padman, Ramayya Krishnan

A retrieval-augmented generation (RAG) system deployed over a multi-author institutional corpus can give a different answer to the same question depending on which source it retrieves -- a failure mode the dominant single-gold-answer paradigm cannot diagnose. We argue that source-dependence is a missing axis of NLP evaluation, and that auditing it means shifting the unit of evaluation from answer correctness to the inter-source relationship. We make this concrete in transplant patient education, where institutional sources demonstrably disagree, releasing three artefacts: TransplantQA, a benchmark of real patient questions, each answered by grounding generation in multiple institutional handbooks as candidate sources; HERO-QA, a hierarchical retrieval strategy that grounds and audits each answer; and a structured-output judge that scores inter-source relationships on a validated 5-label taxonomy. At scale, better retrieval reveals far more disagreement than prior estimates suggested -- understating its prevalence, not its intensity. The framework is domain-agnostic and transfers to legal and educational RAG: measuring source-dependence is a responsibility for deployed multi-source NLP generally.

Yubo Li, Ramayya Krishnan, Rema Padman

Reasoning models are evaluated on single-turn benchmarks but deployed in multi-turn dialogue, where users push back on correct answers. Under sustained adversarial pressure we find a previously undocumented failure mode: the chain-of-thought stays factually correct from first turn to last while the emitted answer flips wrong. We call this unfaithful capitulation (UC) and isolate it with a $2\times 2$ latent-versus-behavioral framework that flip-rate metrics and single-turn faithfulness probes both miss. Across three datasets (MT-Consistency, MMLU-Pro, GSM8K), the latent-correct rate at the behavioral flip clusters near 50% in think mode and collapses to 11-15% under no_think -- paired, within-model causal evidence that reasoning creates the gap. Across models the effect tracks the reasoning channel (high in Qwen3-32B and GPT-OSS-20B, low in inline-CoT Gemma-4-31B-it). An independent GPT-4o judge corroborates $86\%$ of UC labels; a token-level probe shows the answer-slot argmax is correct in $84\%$ of UC cells; and a naive trace-anchored defense backfires. We release all trajectories, traces, and judge labels.

Yubo Li, Saba Al-Sayouri, Rema Padman

This study explores the potential of utilizing administrative claims data, combined with advanced machine learning and deep learning techniques, to predict the progression of Chronic Kidney Disease (CKD) to End-Stage Renal Disease (ESRD). We analyze a comprehensive, 10-year dataset provided by a major health insurance organization to develop prediction models for multiple observation windows using traditional machine learning methods such as Random Forest and XGBoost as well as deep learning approaches such as Long Short-Term Memory (LSTM) networks. Our findings demonstrate that the LSTM model, particularly with a 24-month observation window, exhibits superior performance in predicting ESRD progression, outperforming existing models in the literature. We further apply SHapley Additive exPlanations (SHAP) analysis to enhance interpretability, providing insights into the impact of individual features on predictions at the individual patient level. This study underscores the value of leveraging administrative claims data for CKD management and predicting ESRD progression.

Yubo Li, Ramayya Krishnan, Rema Padman

Large reasoning models with reasoning capabilities achieve state-of-the-art performance on complex tasks, but their robustness under multi-turn adversarial pressure remains underexplored. We evaluate nine frontier reasoning models under adversarial attacks. Our findings reveal that reasoning confers meaningful but incomplete robustness: most reasoning models studied significantly outperform instruction-tuned baselines, yet all exhibit distinct vulnerability profiles, with misleading suggestions universally effective and social pressure showing model-specific efficacy. Through trajectory analysis, we identify five failure modes (Self-Doubt, Social Conformity, Suggestion Hijacking, Emotional Susceptibility, and Reasoning Fatigue) with the first two accounting for 50% of failures. We further demonstrate that Confidence-Aware Response Generation (CARG), effective for standard LLMs, fails for reasoning models due to overconfidence induced by extended reasoning traces; counterintuitively, random confidence embedding outperforms targeted extraction. Our results highlight that reasoning capabilities do not automatically confer adversarial robustness and that confidence-based defenses require fundamental redesign for reasoning models.

Jinsong Liu, Yuhang Jiang, Ramayya Krishnan et al.

Clinical decision support requires not only correct answers but also clinically valid reasoning. We propose Differential Reasoning Learning (DRL), a framework that improves clinical agents by learning from reasoning discrepancies. From reference reasoning rationales (e.g., physician-authored clinical rationale, clinical guidelines, or outputs from more capable models) and the agent's free-form chain-of-thought (CoT), DRL extracts reasoning graphs as directed acyclic graphs (DAGs) and performs a clinically weighted graph edit distance (GED)-based discrepancy analysis. An LLM-as-a-judge aligns semantically equivalent nodes and diagnoses discrepancies between graphs. These graph-level discrepancy diagnostics are converted into natural-language instructions and stored in a Differential Reasoning Knowledge Base (DR-KB). At inference, we retrieve top-$k$ instructions via Retrieval-Augmented Generation (RAG) to augment the agent prompt and patch likely logic gaps. Evaluation on open medical question answering (QA) benchmarks and a Return Visit Admissions (RVA) prediction task from internal clinical data demonstrates gains over baselines, improving both final-answer accuracy and reasoning fidelity. Ablation studies confirm gains from infusing reference reasoning rationales and the top-$k$ retrieval strategy. Clinicians' review of the output provides further assurance of the approach. Together, results suggest that DRL supports more reliable clinical decision-making in complex reasoning scenarios and offers a practical mechanism for deployment under limited token budgets.

Krishna Pothugunta, Xiao Liu, Anjana Susarla et al.

Studies suggest that one in three US adults use the Internet to diagnose or learn about a health concern. However, such access to health information online could exacerbate the disparities in health information availability and use. Health information seeking behavior (HISB) refers to the ways in which individuals seek information about their health, risks, illnesses, and health-protective behaviors. For patients engaging in searches for health information on digital media platforms, health literacy divides can be exacerbated both by their own lack of knowledge and by algorithmic recommendations, with results that disproportionately impact disadvantaged populations, minorities, and low health literacy users. This study reports on an exploratory investigation of the above challenges by examining whether responsible and representative recommendations can be generated using advanced analytic methods applied to a large corpus of videos and their metadata on a chronic condition (diabetes) from the YouTube social media platform. The paper focusses on biases associated with demographic characters of actors using videos on diabetes that were retrieved and curated for multiple criteria such as encoded medical content and their understandability to address patient education and population health literacy needs. This approach offers an immense opportunity for innovation in human-in-the-loop, augmented-intelligence, bias-aware and responsible algorithmic recommendations by combining the perspectives of health professionals and patients into a scalable and generalizable machine learning framework for patient empowerment and improved health outcomes.

Yubo Li, Xiaobin Shen, Xinyu Yao et al. · cmu

Recent advancements in large language models (LLMs) have revolutionized their ability to handle single-turn tasks, yet real-world applications demand sophisticated multi-turn interactions. This survey provides a comprehensive review of recent advancements in evaluating and enhancing multi-turn interactions in LLMs. Focusing on task-specific scenarios, from instruction following in diverse domains such as math and coding to complex conversational engagements in roleplay, healthcare, education, and even adversarial jailbreak settings, we systematically examine the challenges of maintaining context, coherence, fairness, and responsiveness over prolonged dialogues. The paper organizes current benchmarks and datasets into coherent categories that reflect the evolving landscape of multi-turn dialogue evaluation. In addition, we review a range of enhancement methodologies under multi-turn settings, including model-centric strategies (contextual learning, supervised fine-tuning, reinforcement learning, and new architectures), external integration approaches (memory-augmented, retrieval-based methods, and knowledge graph), and agent-based techniques for collaborative interactions. Finally, we discuss open challenges and propose future directions for research to further advance the robustness and effectiveness of multi-turn interactions in LLMs. Related resources and papers are available at https://github.com/yubol-cmu/Awesome-Multi-Turn-LLMs.

Yubo Li, Yidi Miao, Xueying Ding et al.

Large Language Models (LLMs) have shown remarkable capabilities across various tasks, but their deployment in high-stake domains requires consistent and coherent behavior across multiple rounds of user interaction. This paper introduces a comprehensive framework for evaluating and improving LLM response consistency, making three key contributions. Code and data are available at: https://github.com/yubol-bobo/MT-Consistency. First, we introduce Position-Weighted Consistency (PWC), a metric designed to capture both the importance of early-stage stability and recovery patterns in multi-turn interactions. Second, we present MT-Consistency, a carefully curated benchmark dataset spanning diverse domains and difficulty levels, specifically designed to evaluate LLM consistency under various challenging follow-up scenarios. Third, we introduce Confidence-Aware Response Generation (CARG), a framework that significantly improves response stability by explicitly integrating internal model confidence scores during the generation process. Experimental results demonstrate that CARG significantly improves response stability without sacrificing accuracy, offering a practical path toward more dependable LLM behavior in critical, real-world deployments.

Yubo Li, Lu Zhang, Tianchong Jiang et al.

Large language models systematically fail when a salient surface cue conflicts with an unstated feasibility constraint. We study this through a diagnose-measure-bridge-treat framework. Causal-behavioral analysis of the ``car wash problem'' across six models reveals approximately context-independent sigmoid heuristics: the distance cue exerts 8.7 to 38 times more influence than the goal, and token-level attribution shows patterns more consistent with keyword associations than compositional inference. The Heuristic Override Benchmark (HOB) -- 500 instances spanning 4 heuristic by 5 constraint families with minimal pairs and explicitness gradients -- demonstrates generality across 14 models: under strict evaluation (10/10 correct), no model exceeds 75%, and presence constraints are hardest (44%). A minimal hint (e.g., emphasizing the key object) recovers +15 pp on average, suggesting the failure lies in constraint inference rather than missing knowledge; 12/14 models perform worse when the constraint is removed (up to -39 pp), revealing conservative bias. Parametric probes confirm that the sigmoid pattern generalizes to cost, efficiency, and semantic-similarity heuristics; goal-decomposition prompting recovers +6 to 9 pp by forcing models to enumerate preconditions before answering. Together, these results characterize heuristic override as a systematic reasoning vulnerability and provide a benchmark for measuring progress toward resolving it.

Yubo Li, Ramayya Krishnan, Rema Padman

Patient education materials for solid-organ transplantation vary substantially across U.S. centers, yet no systematic method exists to quantify this heterogeneity at scale. We introduce a framework that grounds the same patient questions in different centers' handbooks using retrieval-augmented language models and compares the resulting answers using a five-label consistency taxonomy. Applied to 102 handbooks from 23 centers and 1,115 benchmark questions, the framework quantifies heterogeneity across four dimensions: question, topic, organ, and center. We find that 20.8% of non-absent pairwise comparisons exhibit clinically meaningful divergence, concentrated in condition monitoring and lifestyle topics. Coverage gaps are even more prominent: 96.2% of question-handbook pairs miss relevant content, with reproductive health at 95.1% absence. Center-level divergence profiles are stable and interpretable, where heterogeneity reflects systematic institutional differences, likely due to patient diversity. These findings expose an information gap in transplant patient education materials, with document-grounded medical question answering highlighting opportunities for content improvement.

Xinyu Yao, Aditya Sannabhadti, Holly Wiberg et al.

Medical knowledge graphs (KGs) are essential for clinical decision support and biomedical research, yet they often exhibit incompleteness due to knowledge gaps and structural limitations in medical coding systems. This issue is particularly evident in treatment mapping, where coding systems such as ICD, Mondo, and ATC lack comprehensive coverage, resulting in missing or inconsistent associations between diseases and their potential treatments. To address this issue, we have explored the use of Large Language Models (LLMs) for imputing missing treatment relationships. Although LLMs offer promising capabilities in knowledge augmentation, their application in medical knowledge imputation presents significant risks, including factual inaccuracies, hallucinated associations, and instability between and within LLMs. In this study, we systematically evaluate LLM-driven treatment mapping, assessing its reliability through benchmark comparisons. Our findings highlight critical limitations, including inconsistencies with established clinical guidelines and potential risks to patient safety. This study serves as a cautionary guide for researchers and practitioners, underscoring the importance of critical evaluation and hybrid approaches when leveraging LLMs to enhance treatment mappings on medical knowledge graphs.

Yubo Li, Xinyu Yao, Rema Padman

Despite the outstanding performance of deep learning models in clinical prediction tasks, explainability remains a significant challenge. Inspired by transformer architectures, we introduce the Temporal-Feature Cross Attention Mechanism (TFCAM), a novel deep learning framework designed to capture dynamic interactions among clinical features across time, enhancing both predictive accuracy and interpretability. In an experiment with 1,422 patients with Chronic Kidney Disease, predicting progression to End-Stage Renal Disease, TFCAM outperformed LSTM and RETAIN baselines, achieving an AUROC of 0.95 and an F1-score of 0.69. Beyond performance gains, TFCAM provides multi-level explainability by identifying critical temporal periods, ranking feature importance, and quantifying how features influence each other across time before affecting predictions. Our approach addresses the "black box" limitations of deep learning in healthcare, offering clinicians transparent insights into disease progression mechanisms while maintaining state-of-the-art predictive performance.

Yubo Li, Rema Padman

Modeling clinical time-series data is hampered by the challenge of capturing latent, time-varying dependencies among features. State-of-the-art approaches often rely on black-box mechanisms or simple aggregation, failing to explicitly model how the influence of one clinical variable propagates through others over time. We propose $\textbf{Chain-of-Influence (CoI)}$, an interpretable deep learning framework that constructs an explicit, time-unfolded graph of feature interactions. CoI leverages a multi-level attention architecture: first, a temporal attention layer identifies critical time points in a patient's record; second, a cross-feature attention layer models the directed influence from features at these time points to subsequent features. This design enables the tracing of influence pathways, providing a granular audit trail that shows how any feature at any time contributes to the final prediction, both directly and through its influence on other variables. We evaluate CoI on mortality and disease progression tasks using the MIMIC-IV dataset and a private chronic kidney disease cohort. Our framework significantly outperforms existing methods in predictive accuracy. More importantly, through case studies, we show that CoI can uncover clinically meaningful, patient-specific patterns of disease progression that are opaque to other models, offering unprecedented transparency into the temporal and cross-feature dependencies that inform clinical decision-making.

Yubo Li, Ramayya Krishnan, Rema Padman

Large Language Models (LLMs) have revolutionized conversational AI, yet their robustness in extended multi-turn dialogues remains poorly understood. Existing evaluation frameworks focus on static benchmarks and single-turn assessments, failing to capture the temporal dynamics of conversational degradation that characterize real-world interactions. In this work, we present the first comprehensive survival analysis of conversational AI robustness, analyzing 36,951 conversation turns across 9 state-of-the-art LLMs to model failure as a time-to-event process. Our survival modeling framework-employing Cox proportional hazards, Accelerated Failure Time, and Random Survival Forest approaches-reveals extraordinary temporal dynamics. We find that abrupt, prompt-to-prompt(P2P) semantic drift is catastrophic, dramatically increasing the hazard of conversational failure. In stark contrast, gradual, cumulative drift is highly protective, vastly reducing the failure hazard and enabling significantly longer dialogues. AFT models with interactions demonstrate superior performance, achieving excellent discrimination and exceptional calibration. These findings establish survival analysis as a powerful paradigm for evaluating LLM robustness, offer concrete insights for designing resilient conversational agents, and challenge prevailing assumptions about the necessity of semantic consistency in conversational AI Systems.

Yingtao Luo, Reza Skandari, Carlos Martinez et al. · cmu

Decisions about managing patients on the heart transplant waitlist are currently made by committees of doctors who consider multiple factors, but the process remains largely ad-hoc. With the growing volume of longitudinal patient, donor, and organ data collected by the United Network for Organ Sharing (UNOS) since 2018, there is increasing interest in analytical approaches to support clinical decision-making at the time of organ availability. In this study, we benchmark machine learning models that leverage longitudinal waitlist history data for time-dependent, time-to-event modeling of waitlist mortality. We train on 23,807 patient records with 77 variables and evaluate both survival prediction and discrimination at a 1-year horizon. Our best model achieves a C-Index of 0.94 and AUROC of 0.89, significantly outperforming previous models. Key predictors align with known risk factors while also revealing novel associations. Our findings can support urgency assessment and policy refinement in heart transplant decision making.

Faezeh Movahedi, Rema Padman, James F. Antaki

Objective: This study illustrates the ambiguity of ROC in evaluating two classifiers of 90-day LVAD mortality. This paper also introduces the precision recall curve (PRC) as a supplemental metric that is more representative of LVAD classifiers performance in predicting the minority class. Background: In the LVAD domain, the receiver operating characteristic (ROC) is a commonly applied metric of performance of classifiers. However, ROC can provide a distorted view of classifiers ability to predict short-term mortality due to the overwhelmingly greater proportion of patients who survive, i.e. imbalanced data. Methods: This study compared the ROC and PRC for the outcome of two classifiers for 90-day LVAD mortality for 800 patients (test group) recorded in INTERMACS who received a continuous-flow LVAD between 2006 and 2016 (mean age of 59 years; 146 females vs. 654 males) in which mortality rate is only %8 at 90-day (imbalanced data). The two classifiers were HeartMate Risk Score (HMRS) and a Random Forest (RF). Results: The ROC indicates fairly good performance of RF and HRMS classifiers with Area Under Curves (AUC) of 0.77 vs. 0.63, respectively. This is in contrast with their PRC with AUC of 0.43 vs. 0.16 for RF and HRMS, respectively. The PRC for HRMS showed the precision rapidly dropped to only 10% with slightly increasing sensitivity. Conclusion: The ROC can portray an overly-optimistic performance of a classifier or risk score when applied to imbalanced data. The PRC provides better insight about the performance of a classifier by focusing on the minority class.

Wen Wang, Rema Padman, Nirav Shah

Stratifying patients at risk for postoperative complications may facilitate timely and accurate workups and reduce the burden of adverse events on patients and the health system. Currently, a widely-used surgical risk calculator created by the American College of Surgeons, NSQIP, uses 21 preoperative covariates to assess risk of postoperative complications, but lacks dynamic, real-time capabilities to accommodate postoperative information. We propose a new Hidden Markov Model sequence classifier for analyzing patients' postoperative temperature sequences that incorporates their time-invariant characteristics in both transition probability and initial state probability in order to develop a postoperative "real-time" complication detector. Data from elective Colectomy surgery indicate that our method has improved classification performance compared to 8 other machine learning classifiers when using the full temperature sequence associated with the patients' length of stay. Additionally, within 44 hours after surgery, the performance of the model is close to that of full-length temperature sequence.

Kaung Khin, Philipp Burckhardt, Rema Padman

De-identification is the process of removing 18 protected health information (PHI) from clinical notes in order for the text to be considered not individually identifiable. Recent advances in natural language processing (NLP) has allowed for the use of deep learning techniques for the task of de-identification. In this paper, we present a deep learning architecture that builds on the latest NLP advances by incorporating deep contextualized word embeddings and variational drop out Bi-LSTMs. We test this architecture on two gold standard datasets and show that the architecture achieves state-of-the-art performance on both data sets while also converging faster than other systems without the use of dictionaries or other knowledge sources.

Xiao Liu, Bin Zhang, Anjana Susarla et al.

YouTube presents an unprecedented opportunity to explore how machine learning methods can improve healthcare information dissemination. We propose an interdisciplinary lens that synthesizes machine learning methods with healthcare informatics themes to address the critical issue of developing a scalable algorithmic solution to evaluate videos from a health literacy and patient education perspective. We develop a deep learning method to understand the level of medical knowledge encoded in YouTube videos. Preliminary results suggest that we can extract medical knowledge from YouTube videos and classify videos according to the embedded knowledge with satisfying performance. Deep learning methods show great promise in knowledge extraction, natural language understanding, and image classification, especially in an era of patient-centric care and precision medicine.