Medical texts, particularly electronic medical records (EMRs), are a cornerstone of modern healthcare, capturing critical information about patient care, diagnoses, and treatments. These texts hold immense potential for advancing clinical decision-making and healthcare analytics. However, their unstructured nature, domain-specific language, and variability across contexts make automated understanding an intricate challenge. Despite the advancements in natural language processing, existing methods often treat all data as equally challenging, ignoring the inherent differences in complexity across clinical records. This oversight limits the ability of models to effectively generalize and perform well on rare or complex cases. In this paper, we present TACL (Threshold-Adaptive Curriculum Learning), a novel framework designed to address these challenges by rethinking how models interact with medical texts during training. Inspired by the principle of progressive learning, TACL dynamically adjusts the training process based on the complexity of individual samples. By categorizing data into difficulty levels and prioritizing simpler cases early in training, the model builds a strong foundation before tackling more complex records. By applying TACL to multilingual medical data, including English and Chinese clinical records, we observe significant improvements across diverse clinical tasks, including automatic ICD coding, readmission prediction and TCM syndrome differentiation. TACL not only enhances the performance of automated systems but also demonstrates the potential to unify approaches across disparate medical domains, paving the way for more accurate, scalable, and globally applicable medical text understanding solutions.

Recent progress in large language models (LLMs) has enabled the automated processing of lengthy documents even without supervised training on a task-specific dataset. Yet, their zero-shot performance in complex tasks as opposed to straightforward information extraction tasks remains suboptimal. One feasible approach for tasks with lengthy, complex input is to first summarize the document and then apply supervised fine-tuning to the summary. However, the summarization process inevitably results in some loss of information. In this study we present a method for processing the summaries of long documents aimed to capture different important aspects of the original document. We hypothesize that LLM summaries generated with different aspect-oriented prompts contain different \textit{information signals}, and we propose methods to measure these differences. We introduce approaches to effectively integrate signals from these different summaries for supervised training of transformer models. We validate our hypotheses on a high-impact task -- 30-day readmission prediction from a psychiatric discharge -- using real-world data from four hospitals, and show that our proposed method increases the prediction performance for the complex task of predicting patient outcome.

Hospital readmission prediction is critical for clinical decision support, aiming to identify patients at risk of returning within 30 days post-discharge. High readmission rates often indicate inadequate treatment or post-discharge care, making effective prediction models essential for optimizing resources and improving patient outcomes. We propose PT, a Transformer-based model that integrates Electronic Health Records (EHR), medical images, and clinical notes to predict 30-day all-cause hospital readmissions. PT extracts features from raw data and uses specialized Transformer blocks tailored to the data's complexity. Enhanced with Random Forest for EHR feature selection and test-time ensemble techniques, PT achieves superior accuracy, scalability, and robustness. It performs well even when temporal information is missing. Our main contributions are: (1)Simplicity: A powerful and efficient baseline model outperforming existing ones in prediction accuracy; (2)Scalability: Flexible handling of various features from different modalities, achieving high performance with just clinical notes or EHR data; (3)Robustness: Strong predictive performance even with missing or unclear temporal data.

Developing novel predictive models with complex biomedical information is challenging due to various idiosyncrasies related to heterogeneity, standardization or sparseness of the data. We previously introduced a person-centric ontology to organize information about individual patients, and a representation learning framework to extract person-centric knowledge graphs (PKGs) and to train Graph Neural Networks (GNNs). In this paper, we propose a systematic approach to examine the results of GNN models trained with both structured and unstructured information from the MIMIC-III dataset. Through ablation studies on different clinical, demographic, and social data, we show the robustness of this approach in identifying predictive features in PKGs for the task of readmission prediction.

The use of Large Language Models (LLMs) in medicine is growing, but their ability to handle both structured Electronic Health Record (EHR) data and unstructured clinical notes is not well-studied. This study benchmarks various models, including GPT-based LLMs, BERT-based models, and traditional clinical predictive models, for non-generative medical tasks utilizing renowned datasets. We assessed 14 language models (9 GPT-based and 5 BERT-based) and 7 traditional predictive models using the MIMIC dataset (ICU patient records) and the TJH dataset (early COVID-19 EHR data), focusing on tasks such as mortality and readmission prediction, disease hierarchy reconstruction, and biomedical sentence matching, comparing both zero-shot and finetuned performance. Results indicated that LLMs exhibited robust zero-shot predictive capabilities on structured EHR data when using well-designed prompting strategies, frequently surpassing traditional models. However, for unstructured medical texts, LLMs did not outperform finetuned BERT models, which excelled in both supervised and unsupervised tasks. Consequently, while LLMs are effective for zero-shot learning on structured data, finetuned BERT models are more suitable for unstructured texts, underscoring the importance of selecting models based on specific task requirements and data characteristics to optimize the application of NLP technology in healthcare.

With the increasing availability of patients' data, modern medicine is shifting towards prospective healthcare. Electronic health records contain a variety of information useful for clinical patient description and can be exploited for the construction of predictive models, given that similar medical histories will likely lead to similar progressions. One example is unplanned hospital readmission prediction, an essential task for reducing hospital costs and improving patient health. Despite predictive models showing very good performances especially with deep-learning models, they are often criticized for the poor interpretability of their results, a fundamental characteristic in the medical field, where incorrect predictions might have serious consequences for the patient health. In this paper we propose a novel, interpretable deep-learning framework for predicting unplanned hospital readmissions, supported by NLP findings on word embeddings and by neural-network models (ConvLSTM) for better handling temporal data. We validate our system on the two predictive tasks of hospital readmission within 30 and 180 days, using real-world data. In addition, we introduce and test a model-dependent technique to make the representation of results easily interpretable by the medical staff. Our solution achieves better performances compared to traditional models based on machine learning, while providing at the same time more interpretable results.

The intensive care unit (ICU) comprises a complex hospital environment, where decisions made by clinicians have a high level of risk for the patients' lives. A comprehensive care pathway must then be followed to reduce p complications. Uncertain, competing and unplanned aspects within this environment increase the difficulty in uniformly implementing the care pathway. Readmission contributes to this pathway's difficulty, occurring when patients are admitted again to the ICU in a short timeframe, resulting in high mortality rates and high resource utilisation. Several works have tried to predict readmission through patients' medical information. Although they have some level of success while predicting readmission, those works do not properly assess, characterise and understand readmission prediction. This work proposes a standardised and explainable machine learning pipeline to model patient readmission on a multicentric database (i.e., the eICU cohort with 166,355 patients, 200,859 admissions and 6,021 readmissions) while validating it on monocentric (i.e., the MIMIC IV cohort with 382,278 patients, 523,740 admissions and 5,984 readmissions) and multicentric settings. Our machine learning pipeline achieved predictive performance in terms of the area of the receiver operating characteristic curve (AUC) up to 0.7 with a Random Forest classification model, yielding an overall good calibration and consistency on validation sets. From explanations provided by the constructed models, we could also derive a set of insightful conclusions, primarily on variables related to vital signs and blood tests (e.g., albumin, blood urea nitrogen and hemoglobin levels), demographics (e.g., age, and admission height and weight), and ICU-associated variables (e.g., unit type). These insights provide an invaluable source of information during clinicians' decision-making while discharging ICU patients.

Recent advances in large language models have led to renewed interest in natural language processing in healthcare using the free text of clinical notes. One distinguishing characteristic of clinical notes is their long time span over multiple long documents. The unique structure of clinical notes creates a new design choice: when the context length for a language model predictor is limited, which part of clinical notes should we choose as the input? Existing studies either choose the inputs with domain knowledge or simply truncate them. We propose a framework to analyze the sections with high predictive power. Using MIMIC-III, we show that: 1) predictive power distribution is different between nursing notes and discharge notes and 2) combining different types of notes could improve performance when the context length is large. Our findings suggest that a carefully selected sampling function could enable more efficient information extraction from clinical notes.

Traditional evaluation metrics for classification in natural language processing such as accuracy and area under the curve fail to differentiate between models with different predictive behaviors despite their similar performance metrics. We introduce sensitivity score, a metric that scrutinizes models' behaviors at the vocabulary level to provide insights into disparities in their decision-making logic. We assess the sensitivity score on a set of representative words in the test set using two classifiers trained for hospital readmission classification with similar performance statistics. Our experiments compare the decision-making logic of clinicians and classifiers based on rank correlations of sensitivity scores. The results indicate that the language model's sensitivity score aligns better with the professionals than the xgboost classifier on tf-idf embeddings, which suggests that xgboost uses some spurious features. Overall, this metric offers a novel perspective on assessing models' robustness by quantifying their discrepancy with professional opinions. Our code is available on GitHub (https://github.com/nyuolab/Model_Sensitivity).

Deep learning models have demonstrated superior performance in various healthcare applications. However, the major limitation of these deep models is usually the lack of high-quality training data due to the private and sensitive nature of this field. In this study, we propose a novel textual data augmentation method to generate artificial clinical notes in patients' Electronic Health Records (EHRs) that can be used as additional training data for patient outcomes prediction. Essentially, we fine-tune the generative language model GPT-2 to synthesize labeled text with the original training data. More specifically, We propose a teacher-student framework where we first pre-train a teacher model on the original data, and then train a student model on the GPT-augmented data under the guidance of the teacher. We evaluate our method on the most common patient outcome, i.e., the 30-day readmission rate. The experimental results show that deep models can improve their predictive performance with the augmented data, indicating the effectiveness of the proposed architecture.