An artificial intelligence platform to optimize workflow during ovarian stimulation and IVF: process improvement and outcome-based predictions.

RESEARCH QUESTION

Can workflow during IVF be facilitated by artificial intelligence to limit monitoring during ovarian stimulation to a single day and enable level-loading of retrievals?

DESIGN

The dataset consisted of 1591 autologous cycles in unique patients with complete data including age, FSH, oestradiol and anti-Müllerian concentrations, follicle counts and body mass index. Observations during ovarian stimulation included oestradiol concentrations and follicle diameters. An algorithm was designed to identify the single best day for monitoring and predict trigger day options and total number of oocytes retrieved.

RESULTS

The mean error to predict the single best day for monitoring was 1.355 days. After identifying the single best day for evaluation, the algorithm identified the trigger date and range of three oocyte retrieval days specified by the earliest and the latest day on which the number of oocytes retrieved was minimally changed with a variance of 0-3 oocytes. Accuracy for prediction of total number of oocytes with baseline testing alone or in combination with data on the day of observation was 0.76 and 0.80, respectively. The sensitivities for estimating the total number and number of mature oocytes based solely on pre-IVF profiles in group I (0-10) were 0.76 and 0.78, and in group II (>10) 0.76 and 0.81, respectively.

CONCLUSIONS

A first-iteration algorithm is described designed to improve workflow, minimize visits and level-load embryology work. This algorithm enables decisions at three interrelated nodal points for IVF workflow management to include monitoring on the single best day, assign trigger days to enable a range of 3 days for level-loading and estimate oocyte number.