Open research questions in Machine Learning in Healthcare

What the literature leaves open

• Dealing with the gradual scale of complications, such as bleeding, which can range from slight to life-threatening. Addressing the limited number of cases and complications, particularly for rare events like phrenic nerve palsy.

  Development of an artificial intelligence algorithm for automated detection of cardiovascular intervention-related complications · 2026 · DOI

• Another challenge is the need for external validation, which is often performed retrospectively and does not test whether a model can function within real-world clinical environments. The paper also highlights the challenge of establishing clear responsibilities for post-deployment surveillance and maintaining models over time.

  Editorial: Translating machine learning innovations into clinical practice · 2026 · DOI

• Limited size of available cohorts. Scarcity of events. Overfitting associated with the limited number of mortality events.

  Mortality Prediction in Imbalanced Pediatric Oncology Data: A Cost-Sensitive Hybrid Resampling Framework Using Recurrent Deep Learning · 2026 · DOI

• The limited number of mortality events increases uncertainty in performance estimates and limits model generalizability. Severe class imbalance increases the risk of overfitting and limits the robustness of cross-validation estimates. The lack of richer clinical, pathological, and disease-specific predictors limits model generalizability.

  Mortality Prediction in Imbalanced Pediatric Oncology Data: A Cost-Sensitive Hybrid Resampling Framework Using Recurrent Deep Learning · 2026 · DOI

• Heterogeneous clinical information, which includes demographic information, clinical history, and item-level Symptom Checklist-90 data. Ambiguous clinical histories, which may contain uncertain or missing information. The need to model directional cross-source interactions and hierarchical comorbidity relationships between different mental disorders.

  A multisource clinical deep learning model for early identification of mental disorders in first-visit psychiatric patients · 2026 · DOI

• Integrating heterogeneous, variably structured reports across institutions with differing documentation practices is a significant challenge. The study identifies the challenge of addressing rare CPT codes with limited training data. The study highlights the need for machine learning approaches that can adapt to different institutions and documentation practices.

  Adapting transformer-based encoder models for automated billing code assignment of current procedural terminology from pathology reports in multi-institutional settings · 2026 · DOI

• The study identifies a gap in the existing literature, where most publications include other types of clinical variables, risk factors, patient history, or treatments. The study highlights the need for a more limited set of data to obtain results similar to those of published studies.

  Application of Machine Learning Algorithms in Urinary Tract Infections Diagnosis Based on Non-Microbiological Parameters · 2025 · DOI

• Multimodal approaches combining EHR, computed tomography (CT) imaging, and free-text clinical reports provide complementary information, but the optimal fusion strategy for prognosis prediction remains unclear.

  Multimodal clinical data integration for prognosis of pulmonary embolism: A comparative study · 2026 · DOI

• Specific verifier feedback clears fatal leakage in a hosted agent, and a three-seed ablation finds principle-only feedback clears decision-level defects, though these mitigation results are limited.

  LeakBench: Auditing Reward Hacking through Prediction-Time Leakage in Clinical Modeling Agents · 2026

• Despite its disruptive potential, the application of TabPFN in surgical data science remains unexplored.

  Precision vs recall in surgical AI: Performance and limitations of tabular foundation model for perioperative classification and risk prediction · 2026 · DOI

• Despite the increasing availability of personal health records (PHRs), comprehensive personalized health indicators remain scarce.

  An integrative framework for estimating personal healthy life expectancy from personal health records · 2026 · DOI

• However, limited by the design of the pretraining task and the frequency of knowledge occurrence, it lacks understanding of medical knowledge.

  Leveraging Large Language Models to Enhance Medical Text Representation for Lung Diagnosis Prediction via Knowledge Infusion · 2025 · DOI

• Whether survival at extreme ages can be accurately predicted remains unclear.

  Machine learning prediction of survival in centenarians after age 100: a retrospective, population-based cohort study · 2025 · DOI

• A secondary analysis was performed after filtering inconsistent data points to test the model in an idealized scenario.

  Hybrid triaging assistance algorithm for continuous patient monitoring · 2025 · DOI

• The lack of a holistic, interpretable, and scalable solution for clinical decision support.

  ARTIFICIAL INTELLIGENCE DRIVEN CLINICAL DECISION SUPPORT SYSTEMS: OPPORTUNTIES, FUTURE AND DIRECTIONS · 2026 · DOI

• None of these studies evaluate machine learning models for predicting hospital length of stay across diverse patient populations or healthcare systems; existing work demonstrates high performance on single EHR datasets but lacks external validation or cross-institutional generalization testing.

  Explainable Machine Learning Framework for Predicting Hospital Length of Stay to Enhance Healthcare Resource Management · 2026 · DOI

• Conventional diagnostic approaches have limitations in detecting prostatitis. There is a lack of publicly available longitudinal datasets specific to prostatitis.

  An LSTM-Based Time-Series Framework for Early Detection of Prostatitis Using Longitudinal Clinical Indicators · 2026 · DOI

• Healthy life expectancy is conventionally calculated at the population level, with no standardized approach for estimating it for individuals.

  An integrative framework for estimating personal healthy life expectancy from personal health records · 2026 · DOI

• Using sparse data from an open ICU database, this study shows the feasibility of vital sign-based MRP use at care facilities, which is a big step towards the practical use of the MRP in those facilities.

  Exploring the feasibility of vital signs-based mortality risk prediction in a care facility setting · 2025 · DOI

• The algorithm missed two cases of bleeding, resulting in a sensitivity of 94.4% - The number of cases and complications is limited, particularly for phrenic nerve palsy - The algorithm may generate excessive false-positive cases if made too…

  Development of an artificial intelligence algorithm for automated detection of cardiovascular intervention-related complications · 2026 · DOI

• Retrospective validation can be completed using existing datasets within conventional research timelines, but shadow testing and prospective studies would require substantially greater coordination and resources - The prospect of uncovering problems may further discourage investigators from pursuing prospective study designs - Most of the articles in this Research Topic did not address the later stages of model…

  Editorial: Translating machine learning innovations into clinical practice · 2026 · DOI

• Exploring the use of generative language models in clinical NLP. Investigating the application of transformer-based models to other medical coding tasks. Developing more advanced techniques for interpreting model predictions and verifying alignment between influential keywords and established CPT code terminology.

  Adapting transformer-based encoder models for automated billing code assignment of current procedural terminology from pathology reports in multi-institutional settings · 2026 · DOI

• After filtering out inconsistent data, the hybrid model's accuracy improved to over 90%.

  Hybrid triaging assistance algorithm for continuous patient monitoring · 2025 · DOI

• While multiple studies employ explainable AI techniques (SHAP, feature importance) for model interpretability, none evaluate whether clinicians actually adopt or trust these explanations in real clinical workflows, or measure the impact of explainability on clinical decision-making outcomes.

  Explainable Machine Learning Framework for Predicting Hospital Length of Stay to Enhance Healthcare Resource Management · 2026 · DOI

• For AIM to progress from a prototype toward a validated clinical tool, the following steps are required: (1) Prospective, controlled clinical studies in which the system is operated by clinicians independent of its development team, with diagnostic accuracy assessed against an independently established reference standard; (2) Formal measurement of LLM hallucination rates and safety-critical error rates…

  Architecture and Design of a Prototype Multi-Modal Clinical Decision Support System for Integrative Medicine · 2026 · DOI

Most-cited papers in Machine Learning in Healthcare

• The Inevitable Application of Big Data to Health Care · JAMA · 2013 · 1,353 citations
• Med-BERT: pretrained contextualized embeddings on large-scale structured electronic health records for disease prediction · npj Digital Medicine · 2021 · 818 citations
• PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods · BMJ · 2025 · 727 citations
• Multimodal deep learning models for early detection of Alzheimer’s disease stage · Scientific Reports · 2021 · 673 citations
• BEHRT: Transformer for Electronic Health Records · Scientific Reports · 2020 · 539 citations
• Machine Learning in Healthcare · Current Genomics · 2021 · 518 citations
• Evaluation of clinical prediction models (part 1): from development to external validation · BMJ · 2024 · 510 citations
• Reinforcement Learning in Healthcare: A Survey · ACM Computing Surveys · 2021 · 488 citations
• Early prediction of circulatory failure in the intensive care unit using machine learning · Nature Medicine · 2020 · 467 citations
• Interpreting Incremental Value of Markers Added to Risk Prediction Models · American Journal of Epidemiology · 2012 · 450 citations

Most recent work

• Multimodal Machine Learning Approaches in Predictive Healthcare Analytics: A Comprehensive Survey · Archives of Computational Methods in Engineering · 2026
• Scaling medical AI across clinical contexts · Nature Medicine · 2026
• HEAL-LLMF: A hierarchical explainable AI framework using LLM fusion for multi-modal and personalized healthcare decision making · Information Systems · 2026
• Multimodal graph neural networks in healthcare: a review of fusion strategies across biomedical domains · Frontiers in Artificial Intelligence · 2026
• Structured clinical approach to enable large language models to be used for improved clinical diagnosis and explainable reasoning · Communications Medicine · 2026
• A Bilayer Feature Fusion Framework for Pan-Cancer Survival Prediction Based on Multihead Attention and Adaptive Differential Privacy: Model Development and Validation Study · JMIR Medical Informatics · 2026
• MentalQLM: A Lightweight Large Language Model for Mental Healthcare Based on Instruction Tuning and Dual LoRA Modules · IEEE Journal of Biomedical and Health Informatics · 2026
• A scalable framework for evaluating health language models · npj Digital Medicine · 2026
• A transformer-based survival model for prediction of all-cause mortality in patients with heart failure: a multi-cohort study · npj Digital Medicine · 2026
• Foundation Models Meet Medical Image Interpretation · Research · 2026

Related topics in Computer Science