**Authors and Affiliations**
Taylor, B1., Paul, F2., Filler, C2., Enriquez, D., 2 Song, Z., 3 Vimmerstedt, J., 2 Moore, K., 2 Tomah, M. 2, Fernandes, R., 2 Studer, M., 4 Scales, R. 2
1 Mayo Clinic-Florida, 2 Mayo Clinic-Arizona, 3 Midwestern University, 4 Touro University.

**Background**
Fall risk (FR) screening in cardiac rehabilitation (CR) with computerized posturography (CP) has the potential to enhance workflow efficiency and improve patient outcomes by using artificial intelligence to individualize recommendations for balance training (BT). Our aim was to evaluate the clinical application of the Rehabilitation Artificial Physical Intelligence Database (RAPID) portable foot pressure mat with CP machine learning (ML) software to quantify balance and identify recommendations for therapeutic exercise.

**Methods**
On entry to early outpatient CR, static balance was assessed in 44 adults (86% male, 68 + 13 years) with the modified Clinical Test of Sensory Interaction in Balance (mCTSIB). Two trials of four 20-second progressively demanding test conditions were conducted while standing on a foot pressure mat. RAPID software decoded peak foot pressure, weight distribution and Center of Pressure (CoP) postural sway. A higher composite CoP score in centimeters of sway across the four test conditions indicated a limited ability to remain steady. Prescriptive and predictive ML algorithms identified patterns that categorized CoP scores into a percentile based on age-sex norms. This information was linked to one of ten corrective BT protocols with the potential to mitigate FR (Group 1: BT prescribed to those with the 'worst' balance; Group 10: BT prescribed to those with the 'best' balance). This included a group of five exercise demonstration videos that was generated from a video library.

**Results**
Thirty-eight participants (86%) completed all four test conditions. The ‘best’ (i.e., the lowest) group mean composite CoP score was 140 + 99 cm. Group assignment to the BT protocols was: Groups 1-2 = 18.2%; Groups 3-4 = 4.5%; Groups 5-6 = 4.5%; Groups 7-8 = 27.3%; Groups 9-10 = 45.5%). Those unable to complete all test conditions were assigned Group 1.

**Conclusion**
Screening with CP identified objective FR metrics and corrective exercise recommendations. This provided feedback to augment the clinician’s decision making in the customization an individualized exercise plan. A rigorous research methodology is needed to evaluate the effect of CP in CR.



Computerized posturography with machine learning for fall risk prevention in cardiac rehabilitation