x axis = time since prostatectomy, the y axis = predicted sexual function level,
scaled by the sexual function level. Young patients (below 55 years) reporting
poor sexual function prior to treatment
For all the advances in both medicine and technology, patients still face a bewildering array of advice and information when trying to weigh the possible consequences of certain medical treatments. But a hands-on, data-driven tool I have developed with some colleagues can now help patients obtain personalized predictions for their recovery from surgery. This tool can help patients better manage their expectations about their speed of recovery and long-term effects of the procedure.
People need to be able to fully understand the possible effects of a medical procedure in a realistic and clear way. Seeking to develop a model for recovery curves, we developed a Bayesian modeling approach to recovery curve prediction in order to forecast sexual function levels after prostatectomy, based on the experiences of 300 UCLA clinic patients both before radical prostatectomy surgery and during the four years immediately following surgery. The resulting interactive tool is designed to be used before the patient has a prostatectomy in order to help the patient manage expectations. A central predicted recovery curve shows the patient’s average sexual function over time after the surgery. The tool also displays a range of lighter-colored curves illustrating the broader range of possible outcomes.
This model not only shows people what they can expect about their recovery on average, based on their own specific characteristics, but it also clarifies the uncertainty in the shape of the recovery curve. It shows a range of possible realistic outcomes. We want to help patients who are considering this particular surgery to understand what they could expect. We can’t tell them exactly what their recovery will look like, but at least we can now forecast a personalized recovery curve and show them an informed prediction of their possible outcomes. The model can be used in an interactive way. For example, patients could adjust their reported age or reported sexual function levels to see how their predicted recovery curves change.
Medical data used in the paper was provided by one of the paper’s co-authors, Dr. John Gore, an assistant professor of urology at the University of Washington. He plans to have the interactive tool approved and available for patient use within the clinic in a few months. The other members of my team are MIT PhD student Fulton Wang and Tyler McCormick at the University of Washington.
While this effort is based on prostatectomy, the same statistical model can be used to model recovery curves from other types of surgery and recovery from other medical conditions, such as stroke.
Patients can greatly benefit by being able to leverage the vast amounts of medical data that are now being collected in order to make data-driven decisions. Predictive medicine is getting to be a really big deal. Until now, when people had questions about possible treatment effects, their doctors might give them vague, textbook-type responses and they’d get a range of answers based on whom they ask. It’s time for patients to be able to make decisions based on data. And this type of work helps the data speak.
Cynthia Rudin is Associate Professor of Operations Research at the MIT Sloan School of Management