Although recent work has described the microbiome in solid tumors, microbial content in hematological malignancies is not well-characterized. Here we analyze existing deep DNA sequence data from the blood and bone marrow of 1870 patients with myeloid malignancies, along with healthy controls, for bacterial, fungal, and viral content. After strict quality filtering, we find evidence for dysbiosis in disease cases, and distinct microbial signatures among disease subtypes. We also find that microbial content is associated with host gene mutations and with myeloblast cell percentages. In patients with low-risk myelodysplastic syndrome, we provide evidence that Epstein-Barr virus status refines risk stratification into more precise categories than the current standard. Motivated by these observations, we construct machine-learning classifiers that can discriminate among disease subtypes based solely on bacterial content. Our study highlights the association between the circulating microbiome and patient outcome, and its relationship with disease subtype.
Objective: Case–control study designs are commonly used in retrospective analyses of real-world evidence (RWE). Due to the increasingly wide availability of RWE, it can be difficult to determine whether findings are robust or the result of testing multiple hypotheses. Materials and Methods: We investigate the potential effects of modifying cohort definitions in a case–control association study between depression and type 2 diabetes mellitus. We used a large (>75 million individuals) de-identified administrative claims database to observe the effects of minor changes to the requirements of glucose and hemoglobin A1c tests in the control group. Results: We found that small permutations to the criteria used to define the control population result in significant shifts in both the demographic structure of the identified cohort as well as the odds ratio of association. These differences remain present when testing against age- and sex-matched controls. Discussion: Analyses of RWE need to be carefully designed to avoid issues of multiple testing. Minor changes to control cohorts can lead to significantly different results and have the potential to alter even prospective studies through selection bias. Conclusion: We believe this work offers strong support for the need for robust guidelines, best practices, and regulations around the use of observational RWE for clinical or regulatory decision-making.
The growing use of large language models for health communication raises important questions about patient preferences, trust, and satisfaction with AI-generated content. We conducted a qualitative survey comparing patient perceptions of clinician versus AI answers to questions about Alzheimer’s disease and genetics. Twenty-six participants scored responses on relevance, trustworthiness, and coherence, and additionally selected a preferred response and provided free-text comments. We found that participants generally preferred AI responses over human-written ones and rated them higher on all evaluation axes. Qualitative analysis of free-text comments identified key factors influencing patient preferences, including clarity, verbosity, certainty, empathy, and numeracy, with patients showing diverse and sometimes contradictory preferences along these dimensions. These findings suggest the AI-generated responses were well-received by patients and may have a role in triaging patients with information-seeking queries. Future work should focus on dynamically identifying patient communication preferences and tailoring communication styles accordingly.
The 5th Machine Learning for Health (ML4H) symposium was held in person on December 1-2, 2025, in San Diego, California. The symposium included research roundtable sessions to foster discussions between participants and senior researchers on timely and relevant topics for the ML4H community (Adibi et al., 2025; Jeong et al., 2024; Hegselmann et al., 2023). The organization of the research roundtables at the conference involved 12 senior, 18 junior, and 3 sponsor chairs across 13 tables, including 12 program roundtables coordinated by the organizing committee and one sponsored roundtable coordinated by a corporate sponsor. Each program roundtable consisted of an invited senior chair (with substantial experience in the field), junior chairs (responsible for facilitating the discussion), and attendees from diverse backgrounds with interest in the session’s topic. The sponsored roundtable was co-chaired by 3 employees of the sponsoring company and held concurrently with the program roundtables.