Equitable Modeling for Persistent Opioid Use Prediction & Personalization
Description: Oftentimes, datasets used for decision-making, by design or default, do not adequately capture under-represented populations. Inference, predictive or causal, drawn from algorithms trained on these datasets has, at best, limited utility for these populations. At worst, it can lead to decisions with adverse outcomes. These costs are further exacerbated when applied in sensitive domains – 1) crime prediction 2) job hiring decisions and 3) clinical decision-making. In this context, opioid over/misuse, while continuing to be one of the US’ most pressing public health problems also represents a highly stigmatized behavior. Given the highly sensitive nature of this problem, if under-represented populations suffer from higher risks of misclassification, that has strong implications for equity and fairness in public health. In this project, we aim to estimate costs of these modeling ‘inequities’ in the context of predicting persistent opioid use (POU), and adapt machine learning models to address them. To accomplish this we will: (a) quantify the costs of applying three existing models for predicting POU to under-served populations, and (b) adapt transfer learning approaches to generalize predictions to these populations to calibrate a model which is robust to dataset and sub-population shifts, for unbiased identification of at-risk individuals across sub-groups.
Under the primary mentorship of Dr. Ladhania, the student hire will work on a data analysis project for estimating costs of modeling these ‘inequities’ in the context of predicting POU, and adapting machine learning models to address them.
Student Research Assistant Responsibilities: If you are a student interested in a paid opportunity to work on a project on Equitable Modeling for Persistent Opioid Use Prediction & Personalization, please email Dr. Rahul Ladhania or Dr. Anne Fernandez. This opportunity is available through a MIDAS PODS grant. MIDAS PODS at Michigan has generously provided funding to support a collaboration among Dr. Rahul Ladhania (Assistant Professor of Health Management & Policy), Dr. Anne Fernandez (Assistant Professor, Department of Psychiatry), Dr. Karandeep Singh (Assistant Professor, Learning Health Sciences), and Dr. Paramveer Dhillon (Assistant Professor, School of Information).
We have funds to help support work from up to 1 Masters or rising Senior student through the grant till June 2022. This work will be compensated at the temp hourly rate at the level of GSRA hourly rate, and the number of hours would vary from 10-20 per week (i.e., this is not a GSRA position).
Qualifications: The ideal applicant should have:
1. Strong programming skills in a statistical language – e.g., R (preferred), python, Julia
2. Desired (but not required) background in statistical and ML methods – logistic regression, recursive partitioning trees and forests, ensemble methods
3. Creativity, enthusiasm, and good verbal and written communication skills
4. Strong interest in working on problems in healthcare and machine learning (prior experience is not required)
How to apply: Interested applicants should submit their curriculum vitae, cover letter indicating relevant qualifications and research interests, to Dr. Rahul Ladhania (email@example.com), using the subject line “Equitable Models for POU RA Application.” University of Michigan at Ann Arbor is an equal opportunity/affirmative action employer. There is no citizenship requirement for this position. The application process may involve a programming exercise to evaluate candidate fit.
Applications Due: Open until filled, applications will be reviewed on a rolling basis
Contact: Please email Dr. Rahul Ladhania (firstname.lastname@example.org) or Dr. Anne Fernandez (email@example.com) if you have any questions.