Success of automated algorithmic scheduling in an outpatient setting

Clinical scheduling often struggles to balance high patient volume with provider workload. Traditional booking methods may lead to underutilized time or excessive provider stress. Prior research has shown that double-booking can increase patient throughput but risks overburdening staff. No prior work had resolved how to implement double-booking without increasing physician workload. This gap motivated the development of algorithmic scheduling tools. Predictive models have been used in other healthcare contexts to forecast no-shows. However, their application to outpatient dermatology remains underexplored. This study aimed to test whether algorithmic recommendations could improve efficiency. The study focused on outpatient dermatology due to its high demand and regular session structure.

The researchers aimed to assess if algorithmic scheduling could increase patient volume without overburdening physicians. They hypothesized that predictive modeling could optimize double-booking. The study tested this in a real-world dermatology setting. The primary goal was to compare patient arrival rates between algorithmic and standard scheduling. A secondary aim was to evaluate physician workload perceptions. The study used a randomized controlled trial design. Each session was assigned to either the intervention or control group. The researchers sought to determine if the algorithm could scale to other outpatient settings.

A randomized controlled trial was conducted with 519 clinical sessions. Sessions were assigned to Smart-Booking or standard scheduling. The Smart-Booking algorithm used a predictive model for no-shows. The model had a c-statistic of 0.71. Physicians were unaware of the assignment method. Patient arrival data were collected for each session. Physician workload was measured via post-session surveys. A generalized multivariate linear model compared outcomes between groups.

Smart-Booking increased the average number of arrived patients per session. The intervention group had 15.7 arrivals versus 15.2 in the control. The difference was statistically significant (P = 0.014). The 95% confidence interval ranged from 0.08 to 0.75. Variance in arrivals was higher in the intervention group (3.72 vs 3.33). However, the difference was not statistically significant (P = 0.38). Physician workload was measured using post-session surveys. The survey response rate was 92%. Physicians reported similar levels of busyness in both groups.

The study found that algorithmic double-booking recommendations increased patient volume. This was achieved without increasing physician workload. The results suggest that predictive modeling can optimize outpatient scheduling. The Smart-Booking system is likely scalable to other clinical settings. The algorithm's performance was supported by a c-statistic of 0.71. The study did not find a significant increase in variability of patient arrivals. Physicians did not report greater stress in the intervention group. These findings support the use of algorithmic tools in outpatient scheduling.