Artificial Neural Network and Falls in Community-Dwellers: A New Approach to Identify the Risk of Recurrent Falling?

A comprehensive review of machine learning algorithms and their application in geriatric medicine: present and future

The increasing access to health data worldwide is driving a resurgence in machine learning research, including data-hungry deep learning algorithms. More computationally efficient algorithms now offer unique opportunities to enhance diagnosis, risk stratification, and individualised approaches to patient management. Such opportunities are particularly relevant for the management of older patients, a group that is characterised by complex multimorbidity patterns and significant interindividual variability in homeostatic capacity, organ function, and response to treatment. Clinical tools that utilise machine learning algorithms to determine the optimal choice of treatment are slowly gaining the necessary approval from governing bodies and being implemented into healthcare, with significant implications for virtually all medical disciplines during the next phase of digital medicine. Beyond obtaining regulatory approval, a crucial element in implementing these tools is the trust and support of the people that use them. In this context, an increased understanding by clinicians of artificial intelligence and machine learning algorithms provides an appreciation of the possible benefits, risks, and uncertainties, and improves the chances for successful adoption. This review provides a broad taxonomy of machine learning algorithms, followed by a more detailed description of each algorithm class, their purpose and capabilities, and examples of their applications, particularly in geriatric medicine. Additional focus is given on the clinical implications and challenges involved in relying on devices with reduced interpretability and the progress made in counteracting the latter via the development of explainable machine learning.

XGBoost based machine learning approach to predict the risk of fall in older adults using gait outcomes

This study aimed to identify the optimal features of gait parameters to predict the fall risk level in older adults. The study included 746 older adults (age: 63–89 years). Gait tests (20 m walkway) included speed modification (slower, preferred, and faster-walking) while wearing the inertial measurement unit sensors embedded in the shoe-type data loggers on both outsoles. A metric was defined to classify the fall risks, determined based on a set of questions determining the history of falls and fear of falls. The extreme gradient boosting (XGBoost) model was built from gait features to predict the factor affecting the risk of falls. Moreover, the definition of the fall levels was classified into high- and low-risk groups. At all speeds, three gait features were identified with the XGBoost (stride length, walking speed, and stance phase) that accurately classified the fall risk levels. The model accuracy in classifying fall risk levels ranged between 67–70% with 43–53% sensitivity and 77–84% specificity. Thus, we identified the optimal gait features for accurate fall risk level classification in older adults. The XGBoost model could inspire future works on fall prevention and the fall-risk assessment potential through the gait analysis of older adults.

Comparison of Machine Learning Models to Predict Risk of Falling in Osteoporosis Elderly

Falls are a multifactorial cause of injuries for older people. Subjects with osteoporosis are more vulnerable to falls. The focus of this study is to investigate the performance of the different machine learning models built on spatiotemporal gait parameters to predict falls particularly in subjects with osteoporosis. Spatiotemporal gait parameters and prospective registration of falls were obtained from a sample of 110 community dwelling older women with osteoporosis (age 74.3 ± 6.3) and 143 without osteoporosis (age 68.7 ± 6.8). We built four different models, Support Vector Machines, Neuronal Networks, Decision Trees, and Dynamic Bayesian Networks (DBN), for each specific set of parameters used, and compared them considering their accuracy, precision, recall and F-score to predict fall risk. The F-score value shows that DBN based models are more efficient to predict fall risk, and the best result obtained is when we use a DBN model using the experts’ variables with FSMC’s variables, mixed variables set, obtaining an accuracy of 80%, and recall of 73%. The results confirm the feasibility of computational methods to complement experts’ knowledge to predict risk of falling within a period of time as high as 12 months.

A new tool for the evaluation of the rehabilitation outcomes in older persons: a machine learning model to predict functional status 1 year ahead

PURPOSE

To date, the assessment of disability in older people is obtained utilizing a Comprehensive Geriatric Assessment (CGA). However, it is often difficult to understand which areas of CGA are most predictive of the disability. The aim of this study is to evaluate the possibility to early predict-1 year ahead-the disability level of a patient using machine leaning models.

METHODS

Community-dwelling older people were enrolled in this study. CGA was made at baseline and at 1 year follow-up. After collecting input/independent variables (i.e., age, gender, schooling followed, body mass index, information on smoking, polypharmacy, functional status, cognitive performance, depression, nutritional status), we performed two distinct Support Vector Machine models (SVMs) able to predict functional status 1 year ahead. To validate the choice of the model, the results achieved with the SVMs were compared with the output produced by simple linear regression models.

RESULTS

218 patients (mean age = 78.01; SD = 7.85; male = 39%) were recruited. The combination of the two SVMs is able to achieve a higher prediction accuracy (exceeding 80% instances correctly classified vs 67% instances correctly classified by the combination of the two linear regression models). Furthermore, SVMs are able to classify both the three categories, self sufficiently, disability risk and disability, while linear regression model separates the population only in two groups (self-sufficiency and disability) without identifying the intermediate category (disability risk) which turns out to be the most critical one.

CONCLUSIONS

The development of such a model can contribute to the early detection of patients at risk of self-sufficiency loss.

Falls Risk Prediction for Older Inpatients in Acute Care Medical Wards: Is There an Interest to Combine an Early Nurse Assessment and the Artificial Neural Network Analysis?

BACKGROUND

Identification of the risk of falls is important among older inpatients. This study aims to examine performance criteria (i.e.; sensitivity, specificity, positive predictive value, negative predictive value and accuracy) for fall prediction resulting from a nurse assessment and an artificial neural networks (ANNs) analysis in older inpatients hospitalized in acute care medical wards.

METHODS

A total of 848 older inpatients (mean age, 83.0±7.2 years; 41.8% female) admitted to acute care medical wards in Angers University hospital (France) were included in this study using an observational prospective cohort design. Within 24 hours after admission of older inpatients, nurses performed a bedside clinical assessment. Participants were separated into non-fallers and fallers (i.e.; ≥1 fall during hospitalization stay). The analysis was conducted using three feed forward ANNs (multilayer perceptron [MLP], averaged neural network, and neuroevolution of augmenting topologies [NEAT]).

RESULTS

Seventy-three (8.6%) participants fell at least once during their hospital stay. ANNs showed a high specificity, regardless of which ANN was used, and the highest value reported was with MLP (99.8%). In contrast, sensitivity was lower, with values ranging between 98.4 to 14.8%. MLP had the highest accuracy (99.7).

CONCLUSIONS

Performance criteria for fall prediction resulting from a bedside nursing assessment and an ANNs analysis was associated with a high specificity but a low sensitivity, suggesting that this combined approach should be used more as a diagnostic test than a screening test when considering older inpatients in acute care medical ward.

Falling in the elderly: Do statistical models matter for performance criteria of fall prediction? Results from two large population-based studies

OBJECTIVE

To compare performance criteria (i.e., sensitivity, specificity, positive predictive value, negative predictive value, area under receiver operating characteristic curve and accuracy) of linear and non-linear statistical models for fall risk in older community-dwellers.

METHODS

Participants were recruited in two large population-based studies, "Prévention des Chutes, Réseau 4" (PCR4, n=1760, cross-sectional design, retrospective collection of falls) and "Prévention des Chutes Personnes Agées" (PCPA, n=1765, cohort design, prospective collection of falls). Six linear statistical models (i.e., logistic regression, discriminant analysis, Bayes network algorithm, decision tree, random forest, boosted trees), three non-linear statistical models corresponding to artificial neural networks (multilayer perceptron, genetic algorithm and neuroevolution of augmenting topologies [NEAT]) and the adaptive neuro fuzzy interference system (ANFIS) were used. Falls ≥1 characterizing fallers and falls ≥2 characterizing recurrent fallers were used as outcomes. Data of studies were analyzed separately and together.

RESULTS

NEAT and ANFIS had better performance criteria compared to other models. The highest performance criteria were reported with NEAT when using PCR4 database and falls ≥1, and with both NEAT and ANFIS when pooling data together and using falls ≥2. However, sensitivity and specificity were unbalanced. Sensitivity was higher than specificity when identifying fallers, whereas the converse was found when predicting recurrent fallers.

CONCLUSIONS

Our results showed that NEAT and ANFIS were non-linear statistical models with the best performance criteria for the prediction of falls but their sensitivity and specificity were unbalanced, underscoring that models should be used respectively for the screening of fallers and the diagnosis of recurrent fallers.

Predicting prolonged length of hospital stay in older emergency department users: use of a novel analysis method, the Artificial Neural Network

OBJECTIVE

To examine performance criteria (i.e., sensitivity, specificity, positive predictive value [PPV], negative predictive value [NPV], likelihood ratios [LR], area under receiver operating characteristic curve [AUROC]) of a 10-item brief geriatric assessment (BGA) for the prediction of prolonged length hospital stay (LHS) in older patients hospitalized in acute care wards after an emergency department (ED) visit, using artificial neural networks (ANNs); and to describe the contribution of each BGA item to the predictive accuracy using the AUROC value.

METHODS

A total of 993 geriatric ED users admitted to acute care wards were included in this prospective cohort study. Age >85years, gender male, polypharmacy, non use of formal and/or informal home-help services, history of falls, temporal disorientation, place of living, reasons and nature for ED admission, and use of psychoactive drugs composed the 10 items of BGA and were recorded at the ED admission. The prolonged LHS was defined as the top third of LHS. The ANNs were conducted using two feeds forward (multilayer perceptron [MLP] and modified MLP).

RESULTS

The best performance was reported with the modified MLP involving the 10 items (sensitivity=62.7%; specificity=96.6%; PPV=87.1; NPV=87.5; positive LR=18.2; AUC=90.5). In this model, presence of chronic conditions had the highest contributions (51.3%) in AUROC value.

CONCLUSIONS

The 10-item BGA appears to accurately predict prolonged LHS, using the ANN MLP method, showing the best criteria performance ever reported until now. Presence of chronic conditions was the main contributor for the predictive accuracy.