Healthcare Analytics

MIT Sloan researchers strive to solve a variety of digital tool implementation problems. These problems include using analytical methods that maximize the interpretability (i.e., trustworthiness) of recommendations made with clinical decision support tools, and mitigating privacy and equity issues arising from implementing digital tools. In addition, in the workplace, we are investigating maximizing adoption of new tools, and rigorously testing the clinical and cost impacts through randomized controlled trials.

Dimitris Bertsimas has built more than 20 web-based applications to aid in clinical decision making for a variety of indications, including diabetes, prostate cancer screening, and emergency surgery risks. Many have been proven to be more accurate than prior methods because, in essence, rather than working from experience based on hundreds of cases, a provider is seeing the accumulated knowledge from millions.

Catherine Tucker works on projects in which she studies the health impacts of privacy, performance, and equity-related challenges of implementing electronic medical records, digital health and machine-learning tools.

Joseph Doyle recently published a study with Aurora Health that randomized the introduction of a clinical decision support system across physicians to test its effectiveness versus existing practices when providers must decide among high-cost imaging options, such as CT scans and MRIs.

MIT Sloan researchers are combining data analytics with their disciplinary expertise to create new ways to solve health care problems. In drug development, this includes applying machine learning tools to clinical trials, including selecting which therapeutics to test. Another project applies financial portfolio optimization methods to candidate drug portfolios to improve the probability of finding successful therapies. Other work includes applying novel analytics techniques to support development of liquid biopsy methods to detect cancers more reliably.

Dimitris Bertsimas develops machine-learning tools using massive datasets to optimize clinical trials to discover new cancer treatments, including for gastric cancer. Learn more

Andrew Lo uses data-driven methods informed by financial-portfolio construction in clinical trial design and trial success forecasting. Both the design and the forecasting methods are aimed at improving prediction of the probability of success and understanding the risks underlying the therapies under trial. Ultimately, a better understanding of risk may bring more investors to the table and bring more money into the drug development space.

Vivek Farias received funding through HSI to expand his efforts in using machine learning to develop more cost-efficient liquid biopsies to detect early stage cancer. Specifically, he is using machine learning methods to account for locality in gene mutations. This makes it possible to get higher accuracy for the same cost or similar accuracy for lower cost than status quo sequencing.

Predictive and prescriptive analytics encompass a powerful set of tools and methods that can benefit health delivery in a variety of ways. Our goal is to develop digital tools along with workforce capabilities to improve diagnosis and treatment pathways.

Two HSI-funded studies are enabling Georgia Perakis (with BI Lahey Health) and Jónas Jónasson and Nikos Trichakis (with the Staten Island Performing Provider System) to develop predictive models using both medical data, as well as non-medical data such as from housing and law enforcement agencies, to personalize interventions for those with substance use disorders.

Jónas Jónasson and Erez Yoeli are investigating optimization of treatment pathways by studying when to use human intervention, rather than relying on only a digital platform, to promote TB treatment adherence.