Discrimination of Normal Cognition Versus Mild Cognitive Impairment in Ambulatory Older Adults Using the 4-Square Step Test: A Machine Learning Approach

Abstract

Background and Purpose: Physical therapists use balance tests to assess postural control and fall risk, and cognitive tests to screen for cognitive impairment. Evidence shows that declines in balance accompany cognitive decline; it may be possible to detect both balance and cognitive impairment using only balance tests. This study has 2 aims: (1) to investigate the relationship between balance tests and the Mini-Mental State Examination (MMSE); (2) to determine whether balance test scores can accurately discriminate between older adults with and without mild cognitive impairment (MCI). Methods: Independently ambulatory older adults who walked without an assistive device (N = 128; 57 females and 71 males; mean age: 76.69 [6.26] years) residing in nursing homes were included in this cross-sectional study. The 4-Square Step Test (FSST), Timed Up-and-Go Test (TUG), Single Leg Stance Test (SLS), and Functional Reach Test were used to assess balance. The MMSE was used to evaluate cognitive function. This study employed a machine learning (ML) approach, which is a type of artificial intelligence that enables computers to learn from data and make predictions without being explicitly programmed. The ML approach was used to discriminate cognitive decline as defined by MMSE, utilizing demographics, clinical features, and balance test scores to classify older adults into normal cognition versus a combined group of participants with mild and severe cognitive impairment. Results and Discussion: A negative correlation between FSST and MMSE scores was found (r: -0.43; P < .001). The cutoff score of FSST between older adults with normal cognition (MMSE >= 24; n = 92) and older adults with mild or severe cognitive impairment (MMSE <24; n = 36) was defined as 14.56 seconds. Using this cutoff score, the FSST demonstrated an area under the curve of 0.71 (95% CI: 0.61-0.82, P: 196 < .001) resulting in a sensitivity of 0.77 and a specificity of 0.56, with a positive likelihood ratio of 1.75, and negative likelihood ratio of 0.41. This study yielded promising results, with 1 ML model achieving an accuracy of 92% in MMSE-based cognitive impairment classification. Older adults with MCI required more time to complete the FSST compared to those with normal cognitive function. Artificial neural architectures based on MMSE data outperformed other ML algorithms in using FSST scores to classify cognitive impairment. Conclusion: Cognitive impairments in older adults can be predicted using the FSST. Additionally, it has been shown that various ML models have the potential to effectively analyze these outcomes.