Research by Margrét Bjarnadóttir
Merging Prediction Modeling with Optimization to Improve Appointment Outcomes
When patients fail to show up for their scheduled medical appointments, not only are they potentially hurting themselves, they are negatively impacting the health care system. For individuals, skipping appointments can lead to worsening medical conditions, an increased risk of death, and more frequent hospitalizations and emergency room visits. For the health care system, cancellations and no-shows can result in increasing costs, capacity issues, rescheduling costs, decreases in revenue from empty time slots, and patient and staff dissatisfaction.
So what can be done to make sure patients show up? And which patients should be contacted? That’s the problem Margrét Bjarnadóttir, assistant professor of management science and statistics, addressed in her research, “Optimal Intervention Programs in Health Care Systems.” Bjarnadóttir’s work weighs the costs and benefits of different options for reducing patient no-shows in the context of prediction models.
Bjarnadóttir says there has been a lot of work in the last decade to create models for predicting health care incidents, such as readmission of hospital patients or survival probabilities of cancer patients. “But prediction models don’t answer a very critical question: Now what?,” she says. “It’s fine to build a prediction model, but you need to do something with the prediction model outcomes.”
In health care, deciding what to do has always been about weighing the costs and benefits of different options. But traditionally, those options have been in the context of entire populations, says Bjarnadóttir. Her research considers the individual characteristics of people through prediction models based on patient characteristics to improve operational decisions.
“We took the prediction models and put them to work,” she said. “One extreme is to do nothing and just deal with the no-shows and cancellation costs. The other extreme is to apply the intervention or prevention to the entire population. But then you’re incurring costs from contacting people who don’t really need reminders or assistance. The right balance is probably somewhere in the middle. The question is how do we find the sweet spot? How do we find the correct cohort of patients to maximize the overall benefit of the system. That’s what we did.”
Bjarnadóttir took the case study of appointment adherence using data from a cancer clinic. Bjarnadóttir and her co-researchers used 104 variables to build their models for predicting which patients would cancel or not show up for appointments, covering such things as types of treatment, severity of disease, and the day and time of the visit. The researchers looked at two intervention options: Having a skilled nurse call people to remind them of their appointments, and using an automated text reminder system.
“You don’t just have to choose one,” she said. “If you have a portfolio of options, you should have the option to mix your interventions, in our case call some patients while texting others, choosing an appropriate option for each segment of the population.”
The researchers found the optimal strategy was to use a mix of telephone reminders for 56.6 percent of the population and text messages for 34.1 percent. Surprisingly, the researchers found that remainder of the people needed no intervention at all and would show up with a very high probability.
“That is something we are looking into in more detail – figuring out who those people are by analyzing their characteristics,” Bjarnadóttir said. “That’s something the industry can really learn from.”
Bjarnadóttir said this research could be applied to numerous other health care scenarios to determine the most effective interventions for the right patients at the right times for a more efficient health care system. She said this research stream could also be extended to fields beyond health care, such as marketing – “anywhere where your action has complicated cost structures associated with it.”
Bjarnadóttir’s research has been invited for a revision for publication in the Journal of Manufacturing and Service Operations. It received a second place award for best paper from the POMS College of Healthcare Operations Management.
|Previous Article||Table of Contents||Next Article|