Predictive modeling of fall risk in orthogeriatric patients using machine learning techniques

Kraus, Moritz (2025): Predictive modeling of fall risk in orthogeriatric patients using machine learning techniques. Dissertation, LMU München: Medizinische Fakultät

Abstract

Clinical practice has a barrier when assessing physical frailty in older patients, especially those with orthopedic limitations. This is mostly because standard assessment techniques are subjective, unreliable, and time-consuming. They also frequently depend on data relating to mobility, which may not be applicable to people who are immobile. Considering these limitations, two complementary studies were conducted to redefine the evaluation of physical frailty in this demographic group. The aim is to improve the evaluation and assessment of physical frailty. The primary objective of the initial study was to examine and compare the efficacy of utilizing insole data obtained from the Timed-Up-and-Go test in comparison to known benchmark questionnaires and physical tests. The present study employed machine learning algorithms to detect physical frailty, as determined by the Short Physical Performance Battery (SPPB), in a cohort of individuals aged 60 years and above who possessed independent ambulation and did not exhibit any cognitive or neurological disorders. This study conducted a cross-sectional analysis to examine several factors related to physical frailty. The results showed notable disparities in gait metrics between individuals with and without physical frailty. Furthermore, the Timed-Up-and-Go test exhibited superior predictive value, when compared to the SARC-F (Strength, Assistance with walking, Rise from a chair, Climb stairs and Falls) questionnaire, as evidenced by an AUROC of 0.862 versus 0.639. Machine learning algorithms discovered nine critical characteristics, mostly from digital insole gait data, using recursive feature elimination. Robust predictive accuracy was achieved using these settings, with AUROCs ranging from 0.801 to 0.919. In summary, this research shows that machine learning-based gait analysis is superior to conventional evaluations when it comes to accurately detecting physical fragility in elderly individuals. The second study aimed to address the limitations of assessing physical frailty by developing objective machine models that utilize multifactorial non-mobility parameters. This approach dissociates reliance on mobility-related data and predicts the Timed-Up-and-Go test time accurately. Four machine learning methods—a generalized linear model, a support vector machine, a random forest algorithm, and an extreme gradient boost technique—were compared using six distinct feature selection approaches and 67 multifactorial variables. The random forest algorithm demonstrated the highest accuracy in predicting Timed-up-and-Gotest time, with a mean absolute error of 2.7 seconds. The variable selection methodology had minimal influence on the overall model performance. For slower patients, all algorithms tended to underestimate time, whereas for faster individuals, they tended to overestimate it. These results highlight the potential for Timed-Up-and-Go test time prediction in the absence of mobility data, enabling the automated identification and objective evaluation of patients who are physically frail. With this approach to clinical decision-making and tailored interventions, these developments might have the potential to significantly improve patient care and treatment planning in orthogeriatric settings.

• Dokumententyp: Dissertationen (Dissertation, LMU München)
• Themengebiete: 600 Technik, Medizin, angewandte Wissenschaften ; 600 Technik, Medizin, angewandte Wissenschaften > 610 Medizin und Gesundheit
• Fakultäten: Medizinische Fakultät
• Sprache der Hochschulschrift: Englisch
• Datum der mündlichen Prüfung: 24. März 2025
• 1. Bericht­erstatter:in: Böcker, Wolfgang