DEI Projects & Initiatives

DBMI supports the overall Columbia University mission to foster an environment of diversity, equity, inclusion, and anti-racism for students, faculty and staff. This page highlights several of the ongoing or past projects and initiatives focused on DEI within our department. Any trainees or faculty members who are interested in any of the projects below are invited to reach out to the lead.
 
More information about student, DBMI and Columbia University efforts towards ensuring DEI can be found here.

Principal Investigator: Sarah Rossetti Collins (sac2125@cumc.columbia.edu)

Description
CONCERN is a multi-site study (Columbia University Medical Center and MassGeneral Brigham) that is developing and evaluating an early warning score system to predict and provide clinical decision support when patients are at increased risk of deterioration. The early warning system is based on the variability in nurses’ electronic health record (EHR) documentation patterns, which implies nursing surveillance and reflects nurses’ changing degrees of concern. Due to the inherent biases in EHR data, prediction models developed from EHR data are highly likely to be biased. As a part of our ongoing effort to identify and mitigate implicitly embedded biases in our model, we are conducting analyses to monitor the differences in nursing documentation patterns associated with patients’ demographic characteristics. Our recent work focuses on examining the differences in nursing documentation patterns associated with patients’ race, socioeconomic status, and primary language and seeking solutions to mitigate these biases in our predictive model.

Opportunities
Students interested in working with large EHR data sets can assist in cleaning and analysis of those data, including machine learning and logistic regression.

Principal Investigator: George Hripcsak (gh13@cumc.columbia.edu)

Description
Many clinical algorithms include race as a predictor. However, the appropriate use and the implications of including a patient’s race in clinical predictive algorithms remain unclear. In this work, we study the impact of race on the performance of predictive algorithms for GFR. We compare the prediction error of the estimated GFR with and without the variable race between Black patients and White patients. Our results showed that the prediction error for patients coded as Black was higher compared to those coded as White, regardless of inclusion of race as a variable. Using a large amount of information represented in electronic health record variables achieved a more accurate prediction of GFR and the least difference in prediction error across racial groups.

Opportunities
Potential directions include developing and applying fairness assessment pipeline to other clinical predictive algorithms. Prerequisites: Master’s or above, basic programming in python or R, SQL. Alternatively, medical students or fellows with expertise in a clinical domain.

Principal Investigator: George Hripcsak (gh13@cumc.columbia.edu)

Description
Fairness in clinical decision-making is a critical element of health equity, but assessing fairness of clinical decision-making from observational data is challenging. Recently, many fairness notions have been proposed to quantify fairness in decision-making, among which causality-based fairness notions have gained increasing attention. In this work, we explore a causal fairness notion called principal fairness as a potential metric for assessing fairness of treatment allocation. We develop a probabilistic machine learning algorithm for estimating principal fairness, and show how principal fairness can measure fairness in medical decisions using electronic health records (EHR) data. 

Opportunities
PhD student or post-doc, probability and statistical inference, programming in python or R, SQL. Alternatively, medical students or fellows with expertise in a clinical domain.

Principal Investigator: Pierre Elias (pae2115@cumc.columbia.edu)

Description
Importance: Transthyretin Amyloid Cardiomyopathy (ATTR-CM) is a cause of heart failure that disproportionately affects black patients. Despite this epidemiologic prevalence, systemic racial disparities exist in the diagnosis of ATTR-CM.

Opportunities
Students should have a general comfort with Python programming and ideally prior experience in one of the following domains: image-based analysis, machine learning, deep learning, or EHR phenotyping.