Have you ever gone on a trip and unexpectedly found yourself in need of medical care? What if your condition could have been predicted? Better yet, what if you already had the medicine needed to treat that condition in your luggage?
The Hierarchical Association Rule Model (HARM), which I co-developed with Tyler McCormick of the University of Washington and David Madigan of Columbia University, can help patients be better prepared by warning them (and their doctors) about the conditions they may likely experience next. The predictive modeling tool checks data about an individual patient against other patients in the database with similar situations to help determine future conditions. It also alerts patients about any higher risks they may have for certain types of conditions.
For example, a patient or doctor would input the patient's medical history into HARM's interface and HARM would then combine that information with other information in the database to rank likely future medical conditions. It would say something like: patients like you who have experienced X and Y tend to experience Z next. HARM is not just a black box -- it can explain its predictions in simple easy-to-understand terms.
While HARM can be used by anyone -- doctors, nurses, patients -- it was originally intended to supplement direct care in places where access to healthcare is more difficult. The tool was designed so that when patients experience a new condition, they could input that information themselves to obtain a ranked list of possible future conditions.
Doctors might also use HARM while they are seeing a patient to check the most likely conditions experienced by other patients in similar situations. Doctors always draw from their own experience and medical references when they are seeing patients, but no doctor can truly keep an enormous database and calculate statistics from it in their head. That's why tools like HARM can be helpful to access large databases and make truly data-driven predictions. The tool also could help make medical care more consistent between doctors if they have access to the same data-centered tools.
Regardless of where it's used, the goal is to give increased warning of future conditions, which ultimately means better healthcare results, better allocation of resources, and more advance warning and peace of mind for patients. It also could encourage patients to seek care earlier from their primary physician instead of waiting until emergency care is required.
With the healthcare industry placing more emphasis on access to care and decreasing costs, tools like HARM will be in more demand. Already, we're seeing patient groups seeking the capability to make predictions using patient databases rather than solely relying on doctors' experiences, and insurance companies have an interest in providing preventative health tools to patients.
We also might see this tool used for predictive modeling outside of the healthcare setting, as it can be used for any problem where there are sequences or events that happen over time. A good example is a recommender system like the one used by Amazon.com to indicate which other products customers purchased after buying the one you are considering. The tool also could be used in grocery stores to identify a subset of ingredients that usually go together and warn you if you've forgotten to pick up an item.
The possibilities are great, but in the healthcare setting our hope is that this tool will become freely available to improve access to care. Since it's purely data-driven and not engineered, it's possible that it could be utilized in the not-too-far future.
Cynthia Rudin is an assistant professor of statistics. The paper, Bayesian Hierarchical Rule Modeling for Predicting Medical Conditions, co-authored with Tyler McCormick from the University of Washington and David Madigan from Columbia University, was recently published in the Annals of Applied Statistics and can be viewed here.
How will Donald Trump’s first 100 days impact YOU? Subscribe, choose the community that you most identify with or want to learn more about and we’ll send you the news that matters most once a week throughout Trump’s first 100 days in office. Learn more