Researchers took 18,000 medical appointments, where the reasons for non-attendance were analyzed to predict possible absences. The software created by researchers from the Center for Mathematical Modeling at the University of Chile was able to increase management efficiency in several health centers.
Requesting a medical appointment and not showing up generates losses in time and resources for the health system, especially in public sector hospitals. That is why a group of researchers from the Center for Mathematical Modeling generated a system based on artificial intelligence to reduce no-shows to medical appointments, and thus increase efficiency in the health system.
A phone call, a text message or a whatsapp: what is the most effective way to remind patients of their doctor’s appointment and thus prevent them from not attending? Researchers at the Center for Mathematical Modeling (CMM) at the University of Chile set out to offer a solution to this question through science. By means of artificial intelligence, the academic group trained machines capable of optimizing the management of hospital schedules.
Thanks to a scientific development fund, the team led by Jocelyn Dunstan and Héctor Ramírez devised a software capable of increasing efficiency in hospital care. This, by determining through machine learning what is the probability that a patient does not show up for his or her medical appointment, either by the attendance history of each person, as well as other geographical and social factors that are processed by this model.
In Chile, 1 out of every 5 patients who request a medical appointment do not show up in the end, which generates losses of up to 35 million dollars a year, without considering the time lost in this process. This is why this project uses cost/efficiency measures to reduce these costs. “We are concerned with making this model try to call as little as possible, but with the greatest possible effectiveness in confirming attendance,” says Héctor Ramírez, Ph.D. in Applied Mathematics and CMM researcher, about this project.
Jocelyn Dunstan, Ph.D. in Applied Mathematics and CMM researcher, explains that this method consists of supervised machine learning. “We use a lot of historical information to train these models to understand the pattern or profile of the person who is missing for a certain specialty and in a certain hospital”. They also integrate other factors that can influence the feasibility of a patient not showing up for his or her medical appointment: the day, the time, the geographical area, previous behavior, among others.
The project has already been tested in simulated models, and later the CMM scientists wanted to know how this prototype worked in situ. This is how this method was tested for three months in three hospitals in the country: the Luis Calvo Mackenna Pediatric Hospital, the Regional Hospital of Talca and the Cordillera Oriente Health Reference Center, located in Peñalolén.
“The work we carried out at the experimental level consisted of intervening in the medical agenda of these three hospitals, in order to test the real effectiveness of our predictive method. This also allowed us to better understand the reasons why people do not attend their medical appointments, which in turn feeds our models and makes them more effective,” explained Héctor Ramírez.
Different hospitals, different data
One of the main challenges to be considered in the learning process was the heterogeneity of each hospital’s operation and the profile of its patients. “That is why we decided not to make the same model for all the hospitals, but to differentiate it by establishment, with its own data. Apart from the fact that the specialties also have different patient behavior,” added Ramírez. Added to this is the fact that each patient’s profile also urges us to establish the most optimal way to remind them of their visit to the doctor: with some it is more effective to do so by telephone, others by whatsapp and even in some cases where connectivity is not so good, by text message (SMS).
In addition, in the test pilot, differences were detected between the different types of medical consultation that each patient requests. “There are discrepancies between specialties, which is why we decided not to group them together. The behavior of a pediatric patient is not the same as one who goes to the ophthalmologist, or the gynecologist. We let the data speak to us and tell us what are the main characteristics that determine the non-attendance of patients,” added Dunstan.
This is how around 18,000 medical appointments were processed with artificial intelligence, where the reasons for non-attendance were analyzed to predict possible absences. From this exercise, it was possible to reduce from 20.3% to 12.5% the non-attendance of patients who were reminded via telephone call. On the other hand, the percentage of absenteeism of people notified by whatsapp and text messages was reduced by 5.4%. In this way, the software created by the CMM researchers managed to increase the efficiency in the management of these health centers.
Despite the diversity of hospitals and specialties studied, in all the cases studied it was possible to reduce the number of patients who did not show up for their medical appointments. With this, the CMM scientists hope to develop tools that will serve the different hospital profiles in the country, and thus contribute to reducing the waiting lists that afflict patients in the public sector.
“The natural thing for us is that in the future this would be something used by most public hospitals, and in some private institutions, where efficiency when contacting patients is quite relevant,” concludes Ramírez. The importance of optimizing the medical agenda will also take center stage with the explosive increase in telemedicine during the pandemic, which has even higher rates of absenteeism. This team has set this issue as a new challenge for the future.
Previously published on La Tercera