Surgical Informatics Lab

Seeking a post-doctoral fellow with skills to develop clinical prediction tools at the intersection of surgery and clinical informatics and the interest to translate algorithms to action in under a year. The Surgical Informatics Lab at Harvard Medical School and Beth Israel Deaconess Medical Center has a funded position for 2 projects with unique data assets, access to faculty in surgery and machine learning, and a robust set of initial algorithms.

The Surgical Informatics Lab is a new hospital–university data and research collaborative of informatics faculty in the Department of Biomedical Informatics (DBMI) at Harvard Medical School and surgeon researchers at Beth Israel Deaconess Medical Center (BIDMC). Our program has graduate students, surgical residents, and research and clinical faculty who are developing collaborative tools to transform surgical care using prediction/classification modeling and surgeon- and patient-centered informatics tools. Projects will cover two significant areas:

1. Hospital Opioid Use Project: Identify variations and patterns in peri-surgical treatment that predict opioid consumption after surgical discharge. We have developed a database of consumption from numerous hospitals across the country and the ability to design web tools that can be scaled to multiple hospitals.
2. Ulcerative Colitis Surgical Trajectories Project:  Explainable machine learning models that identify trajectories of ulcerative colitis care and predict transitions from chronic medical management to surgical planning. We have both hospital and national curated datasets and robust preliminary models.

Skills required include proficiency in R, python, and SQL. Statistical needs are regression, cohort matching, inference, and Bayesian techniques. Machine learning methods of clustering, deep learning and temporal techniques are highly desired.

As we expand into multiple in the areas of surgical prediction, the fellow will be part of the team to develop new models and determine future directions of analysis.

Data Resources

1. Aetna, Inc. medical and pharmacy claims for over 70M patients over 10 years.
2. BIDMC medical and pharmacy claims data organized to 1997.
3. BIDMC medical record data, including all medication administrations, lab, surgical notes, OR data, and cost and reimbursement information over 7 years.
4. Opioid use data for 7000 patients across 7 hospital systems from the US.

Projects

Hospital Opioid Use Project

Nearly 70% of surgical patients receive opioids after discharge and 10% become chronic users. We are building a tool to allow post-surgical opioid prescribing to be a personalized, standardized, and optimized process based on patient risk profiles. We have leveraged national claims and opioid use data to build predictive and classification models for post-surgical opioid use and prescribing by patient and surgeon. Multiple ongoing studies are leveraging causal inference in observational data to better understand opioid use as well as machine learning techniques to classify high risk groups. 

Working with experts at DBMI and our data resources, the fellow will identify predictors of outpatient opioid consumption and risks for misuse from the pre-surgical, intra-operative, and in-hospital record. The algorithms will identify concerning temporal consumption markers within BIDMC and partner data. The team will leverage machine learning models to extract latent risk profiles from the pharmacy and clinical data. They will then use techniques of representation learning to train model for affiliated hospitals.

Ulcerative Colitis Surgical Trajectories Project

More than 70% of patients with ulcerative colitis ultimately require surgery. With the advent of expensive medical treatments, the timing of surgery has become an incredibly important question; patients treated medically for too long have worse outcomes. We currently have a prediction model that leverages elements of explainable machine learning methods to predict and visualize transitions to surgery for these patients. The fellow will work with our multi-institutional team to analyze and create clinically relevant and rapidly deployable prediction models. They will use the existing cohort of temporal medical and pharmacy claims data and integrate new tranches of data, as they arrive.

How to Apply

Submit Curriculum Vitae, including list of publications, or questions to Dr. Gabriel Brat at gabriel_brat@hms.harvard.edu