Classification of dementia risk in the elderly through gait analysis with machine learning algorithms

Artificial neural network to classify cognitive impairment using gait and clinical variables

The Impact of Peripheral Neuropathy and Cognitive Decrements on Gait in Older Adults With Type 2 Diabetes Mellitus

Mild cognitive impairment (MCI) poses significant challenges in early diagnosis and timely intervention. Underdiagnosis, coupled with the economic and social burden of dementia, necessitates more precise detection methods. Machine learning (ML) algorithms show promise in managing complex data for MCI and dementia prediction. This study assessed the predictive accuracy of ML models in identifying the onset of MCI and dementia using the Korean Longitudinal Study of Aging (KLoSA) dataset. This study used data from the KLoSA, a comprehensive biennial survey that tracks the demographic, health, and socioeconomic aspects of middle-aged and older Korean adults from 2018 to 2020. Among the 6171 initial households, 4975 eligible older adult participants aged 60 years or older were selected after excluding individuals based on age and missing data. The identification of MCI and dementia relied on self-reported diagnoses, with sociodemographic and health-related variables serving as key covariates. The dataset was categorized into training and test sets to predict MCI and dementia by using multiple models, including logistic regression, light gradient-boosting machine, XGBoost (extreme gradient boosting), CatBoost, random forest, gradient boosting, AdaBoost, support vector classifier, and k-nearest neighbors, and the training and test sets were used to evaluate predictive performance. The performance was assessed using the area under the receiver operating characteristic curve (AUC). Class imbalances were addressed via weights. Shapley additive explanation values were used to determine the contribution of each feature to the prediction rate. Among the 4975 participants, the best model for predicting MCI onset was random forest, with a median AUC of 0.6729 (IQR 0.3883-0.8152), followed by k-nearest neighbors with a median AUC of 0.5576 (IQR 0.4555-0.6761) and support vector classifier with a median AUC of 0.5067 (IQR 0.3755-0.6389). For dementia onset prediction, the best model was XGBoost, achieving a median AUC of 0.8185 (IQR 0.8085-0.8285), closely followed by light gradient-boosting machine with a median AUC of 0.8069 (IQR 0.7969-0.8169) and AdaBoost with a median AUC of 0.8007 (IQR 0.7907-0.8107). The Shapley values highlighted pain in everyday life, being widowed, living alone, exercising, and living with a partner as the strongest predictors of MCI. For dementia, the most predictive features were other contributing factors, education at the high school level, education at the middle school level, exercising, and monthly social engagement. ML algorithms, especially XGBoost, exhibited the potential for predicting MCI onset using KLoSA data. However, no model has demonstrated robust accuracy in predicting MCI and dementia. Sociodemographic and health-related factors are crucial for initiating cognitive conditions, emphasizing the need for multifaceted predictive models for early identification and intervention. These findings underscore the potential and limitations of ML in predicting cognitive impairment in community-dwelling older adults.

This study examined the relationship between sleep disorders and the risk of dementia in patients with newly diagnosed type 2 diabetes. This study used the Korean Health Screening Cohort data and included 39 135 subjects aged ≥40 years with new-onset type 2 diabetes between 2004 and 2007, with follow-up throughout 2013. Sleep disorders were measured by F51(sleep disorders not due to a substance or known physiological condition) or G47(sleep disorders) under International Classification of Diseases, Tenth Revision (ICD-10) codes as a primary diagnosis, and the adjusted hazard ratio (AHR) and 95% CI of all-cause dementia, Alzheimer disease, vascular dementia, and other dementia were estimated using multivariable Cox proportional hazards regression models. In the patients with type 2 diabetes with an age range between 42 and 84 years (M = 57.8, SD = 9.5), this study identified 2059 events of dementia during an average follow-up time of 5.7 years. In patients with type 2 diabetes, subjects with sleep disorders were associated with an increased risk of all-cause dementia (AHR, 1.46; 95% CI, 1.19-1.80), Alzheimer disease (AHR, 1.39; 95% CI, 1.02-1.88), and other dementia (AHR, 1.69; 95% CI, 1.23-2.31) compared to those without sleep disorders. Men (AHR, 1.93; 95% CI, 1.42-2.62) and older adults (AHR, 1.70; 95% CI, 1.35-2.15) with sleep disorders were associated with an increased risk of dementia than their counterparts without sleep disorders among patients with type 2 diabetes. These findings suggest that sleep disorders are significantly associated with an increased risk of dementia in patients with new-onset type 2 diabetes.

When someone says, “I can't hear myself think,” it usually means that background noise is making it difficult to concentrate. But another meaning of this expression has emerged from recent research examining the relationship between hearing and cognition in aging. Several studies have demonstrated the association between aspects of auditory abilities and cognitive function in older adults (Laryngoscope Investig Otolaryngol. 2017;2[2]:69). It has long been understood that the perception of complex everyday sounds, such as music and speech, involves much more than auditory detection and discrimination. However, the relationship between hearing and cognition has only recently become the focus of intensive scientific inquiry (Ear Hear. 2016; 37 Suppl 1:5S).hearing loss, ethnicity, race, health careFigure: Kaplan–Meier failure curve showing the cumulative incidence of dementia according to observed hearing loss status (unadjusted for covariates) (N = 1,881). (Republished with permission from J Am Geriatr Soc. 2017;65:1691).One striking finding of this work is that the rate of age-related cognitive decline is significantly greater in older adults with a hearing loss than among their normal-hearing peers, even when controlling for other known risks (Laryngoscope Investig Otolaryngol. 2017). While the presence and magnitude of this effect vary across studies, overall results demonstrate that the negative impact of hearing loss extends well beyond quality of life issues. Understanding the nature of the relationship between hearing loss and cognition can potentially lead to the design of effective interventions to benefit an individual's well-being and reduce the disease burden on society. Several general mechanisms have been proposed to account for a relationship between hearing and cognition (JAMA Intern Med. 2013;173[4]:293). First, the basis could arise from a common cause, such as the general neural degeneration associated with aging that affects both hearing and cognitive function. Second, cascading consequences of the chronic sensory deprivation due to hearing loss may prevent the appropriate stimulation of higher central auditory processing structures, leading to a subsequent decline in cognitive status. The increased listening effort used to compensate for sensory deprivation may itself affect cognitive processing, resulting in suboptimal allocation of limited cognitive resources and reduced performance on other tasks. Finally, hearing loss may indirectly affect cognitive status through a concomitant reduction in the extent of social interaction and an increase in clinical depression. ETHNICITY AND RACE FACTORS A recent study by Golub and colleagues highlighted an important but less-explored aspect of the relationship between hearing loss and cognition in aging–the influence of ethnicity and race (J Am Geriatr Soc. 2017;65[8]:1691). The researchers examined the prevalence of incident dementia and hearing loss in a large longitudinal sample of aging adults from an ethnically diverse neighborhood of New York City. Among 1,881 baseline participants (40% Hispanic, 31% black, and 29% white), 377 developed incident dementia during the average of 7.4 (+/- 4.6) years of follow-up visits. Overall, there was a 1.69 greater risk of incident dementia among participants with hearing loss than normal-hearing participants, even when controlling for other potential contributors such as cardiovascular risk and stroke. When the sample was stratified by ethnicity and race, however, hearing impairment increased the risk of incident dementia only for black participants (2.62, P <.01). The risk of dementia among Hispanics and whites with hearing loss was also greater than that for those with normal hearing (1.43 and 1.61, respectively), though this trend was not statistically significant. Differences among the three groups persisted even when factors such as education and income were considered in statistical modeling. Golub, et al., observed that the specific reasons for the greater risk of dementia in older black adults with hearing loss are not known. The findings are generally consistent with previous research that demonstrated how ethnicity and race can be significant predictors of cognitive status in older adults (J Int Neuropsychol Soc. 2016;22[1]:66). In our previous work, tests of spectral-temporal processing in which listeners were asked to discriminate changes in the phase of a spectral-ripple and compare brief (0.5 sec) spectro-temporal frequency patterns showed reduced performance among black participants compared with white participants (PLoS One. 2015;10[8]:e0134330). Performance on these tests was significantly correlated with global cognition assessed using a battery of 12 neuropsychological tests, more specifically with scores on working memory tests. The authors suggested that differences in the response strategies used by white and black participants in tasks with a high degree of uncertainty may have contributed to the performance differences between the groups. Another study, however, found that black adults demonstrate a greater resilience to age-related hearing loss (J Gerontol A Biol Sci Med Sci. 2011; 66[5]:582). Therefore, the relationship between hearing and cognition in black adults is more complex than a simple group-wise association. Interpretation of the findings of Golub, et al., on racial and ethnic differences requires further caution for several reasons. First, the criterion used to determine the presence of hearing loss in the study was quite lax. The determination was based on the examiner's observations of a participant's hearing status, including whether the examiner needed to speak loudly or if the participant wore a hearing aid. As the authors acknowledged, this assessment approach may potentially confound the effects of hearing loss and dementia because the ability to repeat and follow the experimenter's instruction—part of the basis for hearing loss determination—also involves cognitive processing of linguistic information. In addition, the performance of participants with hearing loss on cognitive tests may in part reflect their inability to hear instructions well. HEARING AIDS AND COGNITIVE DECLINE The reliability of hearing assessment becomes a critical question as an increasing number of studies use examiner observations and patient reports to evaluate the relationship between hearing and cognition. For example, a study by Amieva and colleagues reported that older adults with self-reported hearing loss and who used hearing aids had the same rate of cognitive decline as their normal-hearing peers, while those who had hearing loss but did not use a hearing aid showed a significantly steeper rate of cognitive decline (J Am Geriatr Soc. 2015;63[10]:2099). This finding, based on the analysis of 3,414 participants in a 25-year longitudinal study, is encouraging because it suggests that the negative impact of hearing loss on cognition may be a modifiable risk factor that can be effectively addressed through hearing aid use. On the other hand, the determination of hearing loss in that study was made based on a single question (“Do you have hearing trouble?”) that was asked only once at the baseline assessment. Participants who were either unaware of their hearing loss or felt uncomfortable admitting it, would not be correctly categorized. Similarly, hearing aid use was also assessed only at baseline, thus reflecting neither the frequency nor consistency of hearing aid use over time. Notably, the study findings revealed that the steeper rate of cognitive decline among participants with hearing loss was no longer different from normal-hearing participants after controlling for psychosocial variables. Amieva, the lead author of the study, later observed that “it is highly unlikely that hearing aids have a direct effect on cognition,” and hypothesized that depression and social isolation associated with hearing loss may mediate the relationship (J Am Geriatr Soc. 2015). IMPACT OF HEARING LOSS Indeed, hearing loss affects many aspects of a person's life. In a recent study, Vas and colleagues examined the complaints of those affected by hearing loss and their communication partners (Trends Hear. 2017). Their analysis documented that in addition to complaints directly related to auditory function such as listening and communication, both hearing-impaired individuals and their communication partners presented with a large range of complaints related to social interactions and individual well-being. Given that social engagement and self-appraisal are known factors in the cognitive decline of older adults, it is likely that these may also play a mediating role between hearing and cognition in older adults. Although there is yet a full understanding of the relationship between hearing and cognition, our current knowledge of the mediating factors can already inform the development of a wide range of effective interventions. Among these, hearing instruments as well as cognitively-focused and healthy lifestyle interventions may serve as effective tools to improve the social circumstances and overall well-being of people with hearing loss and those of their families. Journal Club Highlight Observed Hearing Loss and Incident Dementia in a Multiethnic Cohort Golub, JS, et al. J Am Geriatr Soc. 2017; 65:1691.Valeriy Shafiro, PhDStanley Sheft, PhD

Gait variability in the context of a deterministic dynamical system may be quantified using nonlinear time series analyses that characterize the complexity of the system. Pathological gait exhibits altered gait variability. It can be either too periodic and predictable, or too random and disordered, as is the case with aging. While gait therapies often focus on restoration of linear measures such as gait speed or stride length, we propose that the goal of gait therapy should be to restore optimal gait variability, which exhibits chaotic fluctuations and is the balance between predictability and complexity. In this context, our purpose was to investigate how listening to different auditory stimuli affects gait variability. Twenty-seven young and 27 elderly subjects walked on a treadmill for 5 min while listening to white noise, a chaotic rhythm, a metronome, and with no auditory stimulus. Stride length, step width, and stride intervals were calculated for all conditions. Detrended Fluctuation Analysis was then performed on these time series. A quadratic trend analysis determined that an idealized inverted-U shape described the relationship between gait variability and the structure of the auditory stimuli for the elderly group, but not for the young group. This proof-of-concept study shows that the gait of older adults may be manipulated using auditory stimuli. Future work will investigate which structures of auditory stimuli lead to improvements in functional status in older adults.

While several psychotropic and cardiovascular drugs have been identified as fall-risk-increasing drugs (FRIDs) in older adults, the intervening mechanisms linking FRIDs and falls are unclear. It is plausible that gait performance is an intermediate variable on the causal pathway between FRIDs and falls. The current evidence on the relationship between medication use and gait performance in older adults is scarce. We aimed to assess the association between FRIDs and gait performance in community-dwelling older adults. This was a cross-sectional analysis using data from the Gait and Brain Study, a study of community-dwelling older adults aged 65years old and over (N=345). The following drug classes were assessed: antidepressants, benzodiazepines, alpha-blockers, beta-blockers, vasodilators, diuretics, statins and aspirin. Medication use was ascertained through validated questionnaires and electronic medical records. Multiple linear regression models were used to assess the association between each of the drug classes and gait speed and gait variability. Gait variability was expressed as the coefficient of variation (CV=mean/standard deviation) of stride time. Models were adjusted for age, sex, education, body mass index (BMI), mini-mental status exam (MMSE) score, Geriatric Depression Scale (GDS) score, general activity level, use of other FRIDs and comorbidity propensity score. Diuretic use was associated with significantly reduced gait speed (B=-7.97cm/s, 95% CI: -13.94, -2.00, P=0.009). Statin use was associated with significantly increased stride time CV (B=0.13, 95% CI: 0.02, 0.24, P=0.026). Other drugs did not have a statistically significant relationship with gait speed or variability. The association between diuretic use and reduced gait speed is consistent with existing evidence on diuretic use and increased fall risk. The association between statins and increased stride time variability is notable given inconclusive evidence in previous studies. Our results provide initial estimates of the association between FRIDs and gait performance in older adults for future longitudinal studies.

Depressive symptoms, balance, and gait in older adults: Mediation and moderation analysis.

Gait changes are potential markers of cognitive disorders (CDs). We developed a model for classifying older adults with CD from those with normal cognition using gait speed and variability captured from a wearable inertial sensor and compared its diagnostic performance for CD with that of the model using the Mini-Mental State Examination (MMSE). We enrolled community-dwelling older adults with normal gait from the Korean Longitudinal Study on Cognitive Aging and Dementia and measured their gait features using a wearable inertial sensor placed at the center of body mass while they walked on a 14-m long walkway thrice at comfortable paces. We randomly split our entire dataset into the development (80%) and validation (20%) datasets. We developed a model for classifying CD using logistic regression analysis from the development dataset and validated it in the validation dataset. In both datasets, we compared the diagnostic performance of the model with that using the MMSE. We estimated optimal cutoff score of our model using receiver operator characteristic analysis. In total, 595 participants were enrolled, of which 101 of them experienced CD. Our model included both gait speed and temporal gait variability and exhibited good diagnostic performance for classifying CD from normal cognition in both the development (area under the receiver operator characteristic curve [AUC] = 0.788, 95% CI 0.748-0.823, p < 0.001) and validation datasets (AUC = 0.811, 95% CI 0.729-0.877, p < 0.001). Our model showed comparable diagnostic performance for CD with that of the model using the MMSE in both the development (difference in AUC = 0.026, standard error [SE] = 0.043, z statistic = 0.610, p = 0.542) and validation datasets (difference in AUC = 0.070, SE = 0.073, z statistic = 0.956, p = 0.330). The optimal cutoff score of the gait-based model was >-1.56. Our gait-based model using a wearable inertial sensor may be a promising diagnostic marker of CD in older adults. This study provides Class III evidence that gait analysis can accurately distinguish older adults with CDs from healthy controls.

To the Editor: Age-related changes in spatial and temporal gait parameters have been associated with an adaptation to a safer gait and an increased risk of falling.1 In particular, high stride-to-stride variability was found to be a dependable and potent predictor of falls in community-dwelling older adults. In fact, a small increase in stride-to-stride deviation for stride length of 0.017 m doubled the likelihood of future falling.2 Inability to walk and talk at the same time ("stops walking while talking") is also associated with an increased risk of falling.3 Little information is available about the effect of an attention-splitting task on the gait variability of older adults.4 The aim of this study was to compare stride-to-stride variation in stride length and velocity of community-dwelling older and young adults who were asked to walk while counting backward. Twelve older adults (11 women and 1 man, mean age ± standard deviation (SD) = 83.4 ± 7.7) and 12 young control subjects (11 women and 1 man, mean age ± SD = 22.5 ± 2.4) volunteered to participate. All older adults met the following inclusion criteria: aged 70 and older, no fall history, Mini-Mental State Examination (MMSE) score of 20 or greater, able to walk without a walking aid, and no major neurological or orthopedic problems. The participants were asked to perform two different tasks: walking as a single task and walking while counting backward out loud from 50. Subjects completed one trial for each of the walking conditions on a 15-meter walkway in a well-lit environment, at their own speed, and wearing their own footwear. The human investigation committee approved the project, and written informed consent was obtained before testing. Stride length (m) and stride velocity (m/s) were determined during steady-state walking using an ambulatory device (Physilog, BioAGM, Lausanne, Switzerland) and sensors attached on both lower extremities.5 Stride-to-stride variability was determined by calculating the coefficient of variation (CV = (SD/mean) × 100) for stride length and stride velocity. The Wilcoxon rank-sum test was used to compare coefficients of variation between the two walking conditions. P-values less than .05 were considered statistically significant. Stride-to-stride variability of stride length and stride velocity increased significantly in older subjects with the interfering task of counting. No significant change in gait variability was found in young subjects (Table 1). These results represent, to the best of our knowledge, the first report demonstrating that a dual task increases stride-to-stride variability in older subjects. Walking is a highly automated, rhythmic motor behavior that is mostly controlled by subcortical locomotor brain regions.6 The added task of counting backward interfered with gait regularity in older adults only. This strongly suggests the involvement of higher cortical regions for the motor control of gait in this population. This is consistent with other evidence that walking requires a higher level of attention and thus more cortical involvement with increasing age.7 The inability to allocate attention properly between walking and counting backward could explain the high stride-to-stride variability in these older sample. A dual-task-dependent increase in stride-to-stride variability appears to be a marker for age-related decline in gait control in older adults and might prove to be a particularly sensitive method for the identification of older subjects prone to falling. Further research is needed to define the potential association between falls and increased gait variability under dual-task conditions and to explore interventions that could favorably affect stride-to-stride variability of older adults.

This symposium brings together four studies that employ national data collected from older Americans to examine longitudinal changes in older adults’ cognitive, social, and psychological health. The first study focuses on the role of self-perception of aging in memory decline and social mechanisms linking aging perception and memory. It found that negative perceptions of aging predicted memory decline 8 years later and that loneliness and social isolation were pathways. The second study is concerned with neuropsychological changes before the onset of dementia. Using latent growth modeling, it found that the trajectory of depression increased over time in a quadratic fashion before the onset of incident dementia. Further, there were racial/ethnic group differences in levels of depression. Using fixed-effect linear regression, the third study reports that changes in cognitive status were associated with changes in social isolation in older adults. Specifically, it found that transitioning from cognitive intact to cognitive impairment status was associated with increased social isolation. The fourth and final paper examines changes in psychological well-being related to the transition to widowhood. It found that during the year right after the spouse died, older adults experienced increased depressive symptoms and reduced positive well-being. But adaptation is evident—the surviving spouse returned to pre-loss levels of psychological well-being around two years after the death. Together, the four studies show that attitude, cognitive status, and life course transition shape the dynamics of cognitive, social, and psychological health in later life.

Separating the effects of age and walking speed on gait variability

Association between fear of falling and spatial and temporal parameters of gait in older adults: the FIBRA-RJ study.

Advances in Vascular Cognitive Impairment

Research background. Action observation therapy has been successfully applied in treatment of gait and balance problems in neurological patients, however its effects on gait and balance in older adults remains equivocal. Research aim. The aim of this study was to determine the effects of action observation therapy on gait and balance in older adults. Research methods. The study included 20 older adults. They were randomly assigned to control (usual physiotherapy plus nature recordings observation) and intervention (usual physiotherapy plus action recordings observation) groups. The interventions consisted of a 60-min program fve times a week, for 8 weeks. Gait and balance were evaluated before and after interventions. Research results. Both interventions signifcantly improved (p &lt; 0.05) gait and balance. Greater gait improvement (p &lt; 0.05) was observed after usual therapy plus action recordings observation compared with usual therapy plus nature recordings observation, whereas no differences in intervention effects on balance were observed. Conclusions. Action observation therapy can be used as an effective intervention to improve gait in older adults, whereas it has no effect on balance.Keywords: action observation, motor function, elderly, mirror neurons.