Post-market analysis of an AI-powered clinical decision support tool for ovarian stimulation

Study question

How do treatment decisions and patient outcomes change with physician utilization of artificial intelligence (AI) to help determine FSH starting dose and trigger injection timing?

Summary answer

With physician use of an AI platform, patients were treated with significantly less starting and total FSH without statistically significant changes in outcomes.

What is known already

Stim Assist is an AI clinical decision support software providing physicians with adjunctive information for the prediction of the number of mature (MII) oocytes that may be retrieved from an ovarian stimulation cycle. The software is intended to assist with the selection of the starting dose of FSH as well as the decision of when to administer the trigger injection. Prior studies have retrospectively estimated the benefit of using AI for treatment decisions during ovarian stimulation, but there is a lack of post-market analyses evaluating the true efficacy of these tools when integrated in clinical practice.

Study design, size, duration

This study was conducted by multiple physicians at a single IVF clinic in the U.S. The treatment arm consisted of 292 patients seen by physicians using Stim Assist between December 2022 - December 2023. The control arm consisted of 292 matched historical control patients, treated between May 2019 - May 2022 at the same clinic without use of Stim Assist.

Participants/materials, setting, methods

Matching for each patient was performed for each physician individually, comparing each treatment-arm patient to a historical control-arm patient who was seen by the same physician. Patients were matched 1-to-1 based on age, baseline AMH, and baseline antral follicle count. The primary endpoints were the starting FSH and total FSH prescribed to the patient, as well as MIIs retrieved at the end of stimulation.

Main results and the role of chance

Comparing the treatment arm to the control arm, the average MIIs were 11.17 vs. 11.45 (p = 0.68), the average starting FSH was 397.09 vs. 443.58 (p < 0.01), and the average total FSH was 4181.77 vs 4601.8 (p < 0.01). The results were stratified by age groups of < 35, 35-40, and >40. For patients in the <35 group (N = 122), the average MIIs were 13.99 vs. 16.06 (p = 0.06), the average starting FSH was 346.72 vs. 380.53 (p < 0.01), and the average total FSH was 3525.84 vs. 3927.25 (p < 0.01). For patients in the 35-40 group (N = 129), the average MIIs were 9.46 vs. 8.71 (p = 0.34), the average starting FSH was 420.35 vs. 486.05 (p < 0.01), and the average total FSH was 4568.6 vs. 5063.37 (p < 0.01). For patients in the >40 group (N = 41), the average MIIs were 8.17 vs. 6.34 (p = 0.12), the average starting FSH was 473.78 vs. 497.56 (p = 0.16), and the average total FSH was 4916.46 vs. 5156.71 (p = 0.34).

Limitations, reasons for caution

This preliminary post-market analysis was performed at a single clinic in the U.S. We could not calculate live birth outcomes. Due to the limited sample size, differences in outcomes were not statistically significant.

Wider implications of the findings

These findings suggest that physician use of Stim Assist helped significantly reduce the starting and total FSH prescribed to patients. Additionally, MII outcome differences showed a trend towards an increase in MIIs for patients >35 years old, and a slight reduction in MIIs for younger, better prognosis patients.