MIT Sloan Health Systems Initiative
HSI Spring 2022 research updates
The study team includes Principal Investigators Profs. David Rand and Charles Senteio as well as two Sloan MBA Research Assistants. Prof. Senteio, Assistant Professor of Library and Information Science, Rutgers University, was a Sloan MLK Scholar for the 2020-2021 academic year. His work during that time focused on health equity. We previously wrote about his research here.. Along with researchers from University of Regina and Rutgers University, Prof. Senteio invited a non-profit organization The Hood Medicine Initiative, which is focused on improving the health of communities of color.
Thus far, the team has conducted six national online surveys with thousands of participants to understand the perceptions and actions associated with COVID-19 vaccination. High-level findings include:
- Acculturation - the extent to which Black Americans are aligned with African American versus White culture - is a strong predictor of attitudes toward COVID-19 vaccination.
- Of those who did get the vaccine, the primary reason was government regulation.
- Those who did not get the vaccine gave a variety of reasons, such as propaganda, mistrust, and safety concerns.
Earlier this year, Prof. Senteio partnered with the Boston Public Health Commission (BPHC) to design and conduct a survey at a free vaccination clinic in Roxbury, MA. Only three of the 245 respondents had not been vaccinated against COVID-19. The research team pivoted their focus from vaccine hesitancy to barriers. They are currently designing a project to investigate barriers to COVID vaccination for children aged five to eleven.
Prof. Vivek Farias’ research continues to investigate innovative methods to analyze very large, highly complex, yet noisy and incomplete data sets. His focus is on observational data related to proteomics, that is the study of proteins in a cell, with the goal of diagnosing illness and discovering new therapeutics.
The previous update for this project reported very promising results when the team, in concert with the Broad Institute, tested the research model with a large data set. Their approach yields an approximately 30% improvement in ‘covered’ entries when the ’budget’ on the total width of confidence intervals is small.
Most recently, Prof. Farias and his team wanted to quantify the uncertainty, specifically characterizing the uncertainty in any specific imputation and test the model on real-world array data. Their work on this topic has been accepted to AISTAT 22, the premier conference at the interface of Machine Learning and High-dimensional statistics. Together with Sloan HSI Prof. Jónas Jónasson, they also received funding through the MIT-Takeda alliance to apply this technique to a healthcare network optimization-related problem at Takeda.
This long-running research project is investigating the best ways to facilitate medication adherence over a long course of medication. The disease that is at the center of the study, TB, requires patients to take medication for several months and often after they already “feel better”. The primary mode of communication to remind patients to take their medication is sent via Keheala, a mobile phone platform. The team’s goal is to use machine learning to enable Keheala and its field partners to provide differentiated care, that is provide more contact for patients who are at risk of not successfully completing the treatment, and, more importantly, to provide a proof-of-concept for others.
The team is collecting data from a one-year RCT and a three-year RCT to form the basis of their analyses and insights. The previous update mentioned some initial highlights from the research and listed papers to be written and submitted for publication. Most of those papers have been completed.
During the past research period, the team refined their analysis of data from the three-year study. New insights include:
- Keheala’s estimated effect was positive and statistically different from zero for 100% of individuals.
- All interventions are roughly 50% more powerful when administered in the first month of TB treatment.
The researchers also noted two themes that are recurring throughout all of their work on this project.
- Machine Learning can help assess why an intervention is effective and serve as a basis for making it more effective.
- Machine Learning can help assess whether interventions are beneficial for treatment and outcome equality, and be used in the design of interventions that reduce inequality.
Prof. Georgia Perakis is leading a team (composed of Babson Prof. and MIT Affiliate Dessislava Pachamanova and two PHD students from the Operations Research Center, Omar Skali Lami
And Asterios Tsiourvas) investigating the knotty problem of predicting length of stay (LOS) in the Emergency Department (ED). Their initial goal is to create a model that routes patients through the ED to discharge both effectively and equitably. When the research team first analyzed the hospital’s data, they saw a pattern of potential fairness issues. With this insight, health equity was added as a factor in their model.
A key insight that facilitated the project’s success thus far was dividing LOS into three distinct components (arrival-to room, room-to-disposition and disposition-to-discharge) and developing predictive models for each. Traditionally, LOS Is taken as a whole. This team’s success is evidence that successful models will also have to divide LOS into separate periods.
Most recently, using actual hospital data, the team has noted impressive results from their model, including:
- near-optimal patient allocation to the right resources with more than 50% increase in patient throughput.
- significant decrease in average patient waiting time, i.e., the average waiting time of patients with their method is less than half of the patient average waiting time realized by the hospital.
In a new development this past period, the team is developing sophisticated machine learning models for predicting LOS based on the chief complaint(s) of the patient when they arrive in the ED. They are also in discussion to adapt this model to other areas of hospital operations.
The team’s innovative work in understanding resource and capacity constraints clearly helps the hospital staff prioritize process improvement initiatives. Their method further takes advantage of more granular patient metrics and corrects for bias. These novel components in their continuing research have the potential to change how patients are treated and save lives.
- Held monthly meetings with the Geisinger research team to monitor outcome measurement and keep abreast of any modifications to the program. They decided to stay with every-other-week food delivery, but otherwise the program is continuing as designed.
- Received weekly data pulls to facilitate this monitoring
- Received monthly data pulls on a wider set of outcomes so that they can prepare their results for the end of the study period
- Held monthly meetings with economist investigators on the project to go over interim results and propose new directions for investigation
- Held quarterly meetings with the clinical investigators on the project
The team plans to conclude data collection by the end of September 2022. At that time, they will unblind the results from the trial, analyze the data and publish the results. They have received early interest from the New England Journal of Medicine.