Based in Albany, Georgia, Phoebe Physician Group, a subsidiary of Phoebe Putney Health System, operates in a predominantly rural 41-county service area where caregivers say it has become increasingly acceptable for patients to miss a doctor’s appointment.
THE PROBLEM
So the physician group’s overall no-show rate was 12%, more than double the national average of about 5% reported by the Medical Group Management Association. In urban markets, for example, provider organizations can pay for taxi vouchers, but in southern Georgia, that’s not an option. Automated text messages and reminder calls didn’t help.
Phoebe Physician’s size, the breadth of its clinic markets, and difficulties recruiting in rural areas only compounded the problem. Frequent staff turnover and lack of staff experience led to inconsistent schedules, double bookings, and variable appointment confirmation practices.
“You think your staff is sending these reminders, but often they’re not, or not effectively,” said Matthew Robertson, chief administrative officer of Phoebe Physician Group. “That’s why we decided to explore how AI technology could help increase patient volumes while minimizing disruption for providers and improving the patient experience.”
PROPOSAL
Things started with a conversation with Berkeley Research Group staff informed BRG of the problem and reported that text messages and reminder calls were not working. The organization also needed to eliminate the human element to free up staff time to ensure their work was actually getting done.
“BRG came up with an AI tool, Tool Development and Piloting MelodyMD, which they developed in parallel with Trajum ML,” Robertson explained. “The tool leverages machine learning to analyze years of patient visit data and predict the likelihood that a given patient will not show up for their appointment.
“As new patients are scheduled, MelodyMD communicates with Phoebe Physician’s scheduling system to analyze patient no-shows and automatically creates an adjacent appointment slot if the probability of a no-show exceeds defined thresholds,” he added.
TRY THE CHALLENGE
The tool’s developers looked at the data to see which data had the strongest correlation with the likelihood that a patient would not show up for an exam. These included patient demographics, provider specialty, appointment time, past appointment history, and insurance. As more patient visit data was added, the developers continued to refine their model.
“One of the key things we worked on over time was ensuring that double bookings were limited each day,” Robertson noted. “That is, ensuring that only patients with a high probability of not showing up were considered for double bookings. The exclusions were then applied to specific clinics and appointment types.”
“As the model was rolled out and tested, we made adjustments to the reminder process to improve patient communications and ensure our team had adequate time to fill newly vacant appointment slots,” he continued.
The AI tool, he added, also enabled the organization to measure performance and implement improvements at the following levels:
-
Patient access. Regularly tracks usage, no-show volumes, completed visits, cancellations, cancellations within 24 hours, and rescheduled visits.
-
Reference management. Allows regular monitoring of reference volume, patient leakage and retention rates, as well as distributor and competitor volumes.
-
Supplier evaluation sheet. Provides regular tracking of relative work value units, visit types, assessment and management coding, average number of visits per session, median number of days to schedule new and established patients, no-show rates, and payer mix by provider.
-
Productivity of physicians and advanced care providers. Provides regular tracking of relative work value units, visit type, assessment and management coding, and current procedural terminology details by provider.
-
Non-service personnel. Ensures regular monitoring of full-time equivalent salaries, productivity and overtime to ensure workforce on demand.
RESULTS
From January 2023 to February 2024, Phoebe Physician saw an average increase of 168 visits per week. This represents approximately 7,800 additional visits and $1.4 million in new net patient revenue.
Even though people are still not showing up and the rural perception of making doctor appointments remains a barrier, the double-booking element has gone a long way to reducing the impact of such events, Robertson said.
ADVICE FOR OTHERS
“It’s really important to talk to your providers early in the process and be transparent about what’s happening,” Robertson advised. “You need to clearly articulate the problem, the goals, and the potential impacts you expect AI technology to have. When we talked to our 20 primary care providers and told them they were having 84 absences a day, they were shocked. It helped get them started on trying a new solution.”
“We also need to be careful about health equity,” he continued. “How do we, for example, ensure that the AI tool doesn’t introduce implicit bias into appointment scheduling? How can we refine the algorithm to ensure that underinsured or uninsured patients or those with certain types of insurance aren’t disproportionately impacted by having their appointment slots always double-booked, which could lead to longer wait times in offices?”
And of course, AI is only as good as the underlying data, he added.
“We worked with BRG to clean and organize three years of patient data that we could then use to build an effective model,” he concluded.
Follow Bill’s coverage of HIT on LinkedIn: Bill Siwicki
Send him an email: bsiwicki@himss.org
Healthcare IT News is a publication of HIMSS Media.
