In Saudi Arabia, the machine learning model helps reduce outpatient no-shows
Two years ago, the Riyadh-based hospital of the Ministry of National Guard of the Kingdom of Saudi Arabia, King Abdulaziz Medical City, became the the first in the world to reach Phase 7 in four different HIMSS models. (And it has been that way ever since recently become a pioneer (with impressive work to reach Phase 6 on a different model.) The advanced use of health information and technology has been a boon to the health care system’s 1.3 million patients.
Since then, the 3,720-bed MNGHA has continued its digital health transformation efforts for a variety of specific use cases, including a seemingly simple application that has long vexed provider organizations around the world: no shows in outpatient settings.
They are disruptive, they add unnecessary costs to the care delivery process – and they can have real consequences for care management and patient outcomes.
But the Department of Health National Guard has been able to make notable gains in reducing no-shows by applying artificial intelligence to its analytics, said Huda Al Ghamdi, director of data and business intelligence management at MNGHA, which uses AI to proactively predict which patients are more likely to miss their appointments in an outpatient setting.
The healthcare system uses machine learning to extract data from its electronic health records (patient summaries, clinical information, appointment history) and process and train it for AI models that can alert physicians within the EHR, allowing them to send the necessary reminders to can direct their patient and even book appointments within their own workflows.
MNGHA includes more than 30 hospitals, specialty hospitals and primary care centers in Saudi Arabia, with all facilities linked to a unified EHR system called BESTCare.
That gives the “advantage of having a huge amount of data,” Al Ghamdi explained. “Advanced Analytics, Predictions, and Machine Learning.”
Innovative approaches to analytics have helped the health care system in many areas, she said, but no-shows were a particular concern.
“The reason for addressing this issue in particular is because the outpatient setting is considered the largest channel through which MNGHA delivers medical services to the patients,” she said. “Unlike inpatient or emergency care, the outpatient is considered the largest because we are talking about an average of about 20,000 visits per day.”
That amounts to 5 to 6 million visits per year.
“So when you have an issue like a no-show, it certainly has an impact on the healthcare providers, on the resources and on the patient themselves,” Al Ghamdi said.
The fact that MNGHA is a government hospital means it is sometimes difficult to measure the costs of patients not showing up for their appointments, she notes, but there are costs involved, “and we need to be aware of that and start thinking about savings. .”
Fortunately, MNGHA has a “huge amount of data that we can start analyzing and studying and trying to figure out the factors that influence this,” Al Ghamdi said. “We have a uniform electronic medical record system with different modules for registration, admission and outpatient treatment.
“When it comes to the datasets we use in this project, it is mainly about demographic information, very simple information, mainly gender, age, in addition to the information related to the clinic itself, because there is a variation of no-show from one clinic to another,” she explained. “And the third part of the data sets is the history of the patients themselves. With some patients we notice that they have a high percentage because they miss their appointments, just like the other patients. So that kind of history gives us insight about those kinds of patients.”
Importantly, we “did not address any kind of clinical data” for this project, she added, because that would require expert physicians to decide what kinds of clinical factors might influence a no-show.
But using a basic dataset of patient information allowed some initial models to be created, which were then validated to ensure which was the best and most accurate.
“The project started two years ago. It will take phases to ensure that we are ready to (incorporate the model) into the electronic medical record system,” Al Ghamdi said. “So in the first year the model was created, and I can say that we are in the phase of validating the model, this validation phase lasts about four to six months.
“Some of that validation has been done within data science, and then we’re launching it for a small group of physicians and nursing and patient services staff,” she added. “And that phase lasted about another six months. At that point, we spend a year validating and making sure that the model is reliable and that we can really trust the results of that model.”
Once the data science experts were satisfied with the algorithm, MNGHA took the step to incorporate the model into its EHR system and integrate it into clinical workflows.
“The doctor can see that the patient scheduled for that day may not attend the appointment. And by having this kind of flag in the medical record system, the doctor can send additional reminders, or for example ask patient services for an kind of call to remind the patient,” said Al Ghamdi.
The plan is ultimately to implement the model across all MNGHA facilities, across all regions.
For healthcare systems looking to try something similar for their own organizations, Al Ghamdi offers some advice.
“Even if they start with a small data set, it is better to do these kinds of implementations even in a small ball of data or a small list of parameters, because we know for sure that data tells us a lot about our patients, and there are a kind of hidden patterns that we can discover by using machine learning and artificial intelligence techniques.
“Taking the steps forward to engage with the data and extract knowledge from it is something very important,” she said. “It’s a very simple model to create. But it has a huge impact on the organization.”
Read a more in-depth case study on MNGHA’s use of machine learning for predictive analytics here.
Mike Miliard is editor-in-chief of Healthcare IT News
Email the writer: mike.miliard@himssmedia.com
Healthcare IT News is a HIMSS publication.