Smarter health predictions from electronic medical records
Stochastic Deep Learning for Electronic Health Records: Localizing Learning with Massive and Fragmented Data
The team is building a new type of computer model to make more reliable health predictions from messy, incomplete electronic medical records for patients with varied care histories.
Quick facts
| Grant type | R01 grant |
|---|---|
| Study type | NIH-funded research |
| Funding institution | Purdue University NIH-funded |
| Lab location | 1 site (West Lafayette, United States) |
| Project ID | NIH-11163561 on NIH RePORTER |
What this research studies
Researchers are creating a new kind of neural network called StoNet that is designed to handle very large but messy electronic health records, including lots of missing or inconsistent information. They plan to combine EHR data with imaging and omics information when available and train the model using an advanced stochastic gradient MCMC algorithm to improve learning from fragmented data. StoNet links simple regression methods with deep learning so it can adapt to different patient subgroups and data types. The goal is a rigorous statistical framework that produces predictions that work across diverse and incomplete medical records.
Who could benefit from this research
Good fit: People whose medical records—such as clinic notes, lab results, imaging, or genetic data—are stored in electronic health record systems could be included in or benefit from this work.
Not a fit: Patients without electronic records, without data-sharing agreements, or whose conditions are not represented in the datasets used may not see direct benefit.
Why it matters
Potential benefit: If successful, this could make computer-based health predictions more accurate and fair for patients whose records are incomplete or come from varied care settings.
How similar studies have performed: Previous deep-learning approaches on EHRs have shown promise but often struggle with missing and heterogeneous data, so this StoNet approach is relatively novel and less tested at large scale.
Where this research is happening
West Lafayette, United States
- Purdue University — West Lafayette, United States (Active)
Researchers
- Principal investigator: Liang, Faming — Purdue University
- Study coordinator: Liang, Faming
About this research
- This is an active NIH-funded research project — typically early-stage science, not a clinical trial accepting patient enrollment.
- Some NIH-funded labs run parallel clinical studies or seek volunteers for related work. To check, contact the principal investigator or institution listed above.
- For full project details, budget, and progress reports, visit the official NIH RePORTER page below.