Data collaboration for causal inference from limited medical testing and medication data

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 21, 2025

Observational studies enable causal inferences when randomized controlled trials (RCTs) are not feasible. However, integrating sensitive medical data across multiple institutions introduces significant privacy challenges. The collaboration quasi-experiment (DC-QE) framework addresses these concerns by sharing "intermediate representations"—dimensionality-reduced derived from raw data—instead of the data. Although DC-QE can estimate treatment effects, its application to remains unexplored. aim this study was apply a single institution simulate distributed environments under independent and identically (IID) non-IID conditions. We propose method for generating intermediate representations within framework. Experimental results show that consistently outperformed individual analyses various accuracy metrics, closely approximating performance centralized analysis. proposed further improved performance, particularly These outcomes highlight potential as robust approach privacy-preserving in healthcare. Broader adoption increased use could grant researchers access larger, more diverse datasets while safeguarding patient confidentiality. This may ultimately aid identifying previously unrecognized relationships, support drug repurposing efforts, enhance therapeutic interventions rare diseases.

Language: Английский

Enhancing Type 2 Diabetes Treatment Decisions With Interpretable Machine Learning Models for Predicting Hemoglobin A1c Changes: Machine Learning Model Development

JMIR AI, Journal Year: 2024, Volume and Issue: 3, P. e56700 - e56700

Published: July 18, 2024

Background Type 2 diabetes (T2D) is a significant global health challenge. Physicians need to assess whether future glycemic control will be poor on the current trajectory of usual care and usual-care treatment intensifications so that they can consider taking extra measures prevent outcomes. Predicting from trends in hemoglobin A1c (HbA1c) levels difficult due influence seasonal fluctuations other factors. Objective We sought develop model accurately predicts among patients with T2D receiving care. Methods Our machine learning (HbA1c≥8%) using transformer architecture, incorporating an attention mechanism process irregularly spaced HbA1c time series quantify temporal relationships past at each point. assessed 7787 seeing specialist physicians University Tokyo Hospital. The training data include instances occurring during intensifications. compared prediction accuracy, area under receiver operating characteristic curve, precision-recall accuracy rate, LightGBM. Results rate (95% confidence limits) proposed were 0.925 CI 0.923-0.928), 0.864 0.852-0.875), 0.86-0.869), respectively. achieved high comparable or surpassing LightGBM’s performance. prioritized most recent for predictions. Older slightly more influential predictions good control. Conclusions care, including intensifications, allowing identify cases warranting extraordinary If used by nonspecialist, model’s indication likely may warrant referral specialist. Future efforts could incorporate diverse large-scale clinical improved accuracy.

Language: Английский