Identifying predictors of independence in toileting activities using machine learning

Introduction

A critical factor determining the eligibility of patients with stroke to return home following hospitalization is their ability to perform activities of daily living independently, together with social interaction, dressing, changing of upper garments, defecating, and defecation control1,2.
As such, it is essential to focus on the degree of independence in defecation behavior. This is to a great extent determined by the interaction between trunk and cognitive functions, and trunk function and age3. However, prior reports on independence in toileting behavior have mainly focused on the lower limb, trunk, and cognitive functions 3-5; nevertheless, their findings indicate that inter alia endurance to move to the toilet, balance to perform undergarment manipulation, and upper limb and hand functions to handle incontinence pads and press the nurse call button all play important roles. In addition, an appropriate nutritional status is also essential for maintaining the high motor skills required to perform activities of daily living6, although laboratory tests are necessary to confirm that adequate nutrition is being received. While it is clear that independence is influenced by various physical and cognitive factors, other types may play a role. To date, there remains no consensus on the factors contributing to toileting independence, presumably because many reports have confined their analysis to a specific range or potential influencing factors. Moreover, the range of influencing factors differs from one report to another. Hence, to better understand the intricacies involved in toileting independence, it is necessary to analyze explanatory variables from multiple factors.
One particularly effective approach for such holistic analysis involves the use of machine learning, which can enhance the precision of predictions by concurrently considering multiple factors7. Machine learning technology enables computers to learn from experience and data, and make predictions and decisions based on new information; algorithms are used to extract patterns and trends, and create models that can be used to solve problems. Therefore, the aim of this study was to use the random forest machine learning method to identify the factors that predict the level of independence in toileting activities in patients with stroke.

Materials and methods

The study included 30 patients with stroke (male: 19, female: 11, mean ± standard deviation age: 71.5 ± 13.4 years) who were admitted to the Medical Corporation Wako Kai Yamada Hospital between January 20, 2023 and June 30, 2023. Patients who were transferred to acute care hospitals owing to their deteriorating condition or who died during this period were excluded from the study. Their electronic medical records were searched to collect and construct the dataset by two research collaborators. The data set included the following items, based on previous research: functional Independence Measure (FIM) in toileting; basic attributes of age; cognitive function involved in decision-making for physical and excretory activities (revised version of Hasegawa's Dementia Scale; HDS-R); grip strength, knee extensor strength, and Berg Balance Scale (BBS), which are involved in maintaining standing posture and lower limb movement on the toilet; and the 6-minute walk test (6MWT), which is an indicator of mobility needed to move the toilet.
The following laboratory tests were performed: red blood cell count (RBC), hematocrit (Ht), hemoglobin level (Hb), platelet count (PLT), mean corpuscular hemoglobin (MCH), mean corpuscular volume (MCV), mean corpuscular hemoglobin concentration (MCHC), white blood cell count (WBC), albumin (Alb), blood urea nitrogen (BUN), total protein (TP), creatinine (Cre), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and C-reactive protein (CRP) (Table 1).
When integrating the data, the FIM result was dichotomized based on whether or not it was independent. Established cutoff and reference values, were used as the basis for dichotomization for the HDS-R, 6MWT, BBS and blood data. For age, grip strength and knee extensor strength, for which reference values have not been established, dichotomization was performed based on the median method.
For statistical analysis, a random forest algorithm was used to create a predictive model that classified FIM activities as independent or dependent, to extract the two categories of toilet defecation, and their characteristic factors. Cross-validation was performed to evaluate the model. Specifically, to generate an algorithm for classifying the level of independence of toileting behavior, the training data were divided into five cross-validation folds, and the training and test data in each fold were then classified. Subsequently, the test data were introduced into the constructed algorithm, and the classification was predicted. The accuracy was calculated for each fold, and the mean value was computed to establish the accuracy of the prediction model. The mean value of the Gini importance for each of the five folds was calculated and used as the characteristic factor index of the classification. To reduce dimensionality, the important factors were calculated using the square root of the total number of variables as an indicator.
The data analysis was performed using the “RandomForest” package of R, version 4.3.0. Data were collected retrospectively from electronic medical records. The study was conducted in accordance with the Declaration of Helsinki and the data obtained were anonymized so that personal information could not be identified. In conducting this study, the subjects were given a written explanation of the study and signed a consent form. The Research Ethics Review Committee of Kansai University of Health Sciences approved the study (approval number: 22-16) before the experiments were conducted.

Results

The top five Gini importance factors were HDS-R, 6MWT, BBS, Alb, and age. Based on the Mean decrease Gini, the critical characteristics for predicting the independence of voiding behavior were calculated as HDS-R (1.0), 6MWT (0.9), BBS (0.8), Alb (0.8), and age (0.8) (Figure 1). The mean accuracy value calculated for the FIM score in predicting toileting independence was 0.75.
ALT- alanine aminotransferase, Alb- albumin, AST- aspartate aminotransferase, BBS- Berg Balance Scale, BUN- blood urea nitrogen, Cre- creatinine, CRP- C-reactive protein, Hb- hemoglobin, HDS-R- revised version of Hasegawa’s Dementia Scale, Ht- hematocrit, MCH- mean corpuscular hemoglobin, MCHC- mean corpuscular hemoglobin concentration, MCV- mean corpuscular volume; PLT- platelet, RBC- red blood cell, TP- total protein, WBC- white blood cell, 6MWT- 6-min walk test

Discussion

This study utilized machine learning to identify the factors that predict toileting independence in patients with stroke. The Gini importance results revealed that HDS-R, 6MWT, BBS, Alb, and age are critical factors in predicting independence in toileting activities. The novelty of this research lies in its creation of a dataset based on various explanatory variables and identification of the most influential factors for achieving toileting independence. Furthermore, its findings showed an accuracy of 75%. This is a reasonable value when compared to previous studies8,9.
To adequately consider independence in toileting, it is essential to address both excretory control and transfer/defecation activities. Voiding control is subdivided into several components, such as the patient's ability to perceive the urge to urinate or defecate, communicate the intention to urinate or defecate, and be aware of the toilet's location. Many patients with stroke demonstrate symptoms associated with impaired urinary storage, including nocturia, daytime frequency and urgency, resulting in an increased need for toilet use 10. Furthermore, post-stroke cognitive decline can result in an inability to communicate the intention to defecate10,11 ,12, and hence an increased need to use the toilet and a greater risk of incontinence. This is consistent with our finding that the HDS-R, which indicates overall cognitive function, was a crucial factor for predicting independence in toileting activities.
In addition to cognitive factors, physical factors likewise influence the degree of independence in toileting. Among these, a particularly crucial role is played by lower limb function, as impaired lower limb function prevents the performance of transferring, dressing, and wiping in toileting activities13. Specifically, standing up and changing direction are involved in the transfer motion, and these are closely related to movement of the hip and ankle joints; furthermore, bending the knee and reaching downwards play important roles in lower garment manipulation, and hip joint movement is needed to enable sitting down and cleaning one side of the pelvis while bearing weight on that side.
To summarize, we can say that lower limb function is closely related to motor skills in toilet activities. Furthermore, motor impairment also affects incontinence in patients with stroke with normal bladder and rectal function; this type of incontinence is considered to be associated with the prolonged time taken to perform the related movements14,15. Accordingly, lower limb dysfunction can exacerbate functional incontinence by prolonging the time required for transfer and voiding activities. Hence, it is possible that 6MWT and BBS, which reflect lower limb function at the activity level, are related to the degree of independence in toileting activities, as supported by our present findings: 6MWT and BBS were found to be predictive of voiding behavior independence.
Our findings also indicate that age was a predictive factor for toileting independence. Previous studies have reported that older age increases the risk of incontinence, regardless of the presence or absence of disease16,17. Moreover, it is widely recognized that older age increases the risk of cognitive decline and muscle weakness18,19. Cognitive function influences the ability to express the intention to defecate, associated with defecation control. Therefore, we infer that age influences toileting behavior through mediating factors, such as cognitive decline and muscle weakness. Indeed, prior research has shown that cognitive function, motor function, and age affect toileting behavior [3-5], and the present results are in agreement with these findings.
Additionally, Alb, an indicator of nutritional status, was identified as an essential factor for predicting toileting independence. It has been suggested that a low nutritional status after a stroke inhibits the ability to perform activities of daily living20, and that malnutrition following a stroke increases the risk of cognitive decline, sarcopenia, and physical decline21. Furthermore, in patients with stroke, a better nutritional status is associated with a greater ability to perform activities of daily living, which is strongly associated with physical function6. Nevertheless, nutritional status is not an independent factor for predicting independence in toileting; however, it is indirectly associated with the degree of independence in toileting through mediating factors such as cognitive function and muscle strength.
These findings have clinical implications. Firstly, they suggest that actively intervening in lower limb function may play an important role in achieving independence in toileting. They also highlight the importance of cognitive function; however, this does not just mean that there is a need for interventions aimed at improving cognitive function, as this is extremely difficult, and not always possible. Therefore, our findings indicate the importance of strengthening inter-professional collaboration with nurses and therapists, and creating an environment that makes it easier to express the intention to defecate.
One of the limitations of this study is that the sample size is very small, and further studies with more participants are needed to increase the generalizability of the results. In addition, further research should use Structural Equation Modeling to examine the causal relationships and interactions between the important factors obtained from the Random Forest.

Conclusions

Cognitive and lower limb functions are crucial factors in reaching independence in toileting. Age and nutritional status also impact the achievement of independent toileting; that said, these factors may exert indirect effects that are mediated by cognitive and lower limb functions. Given that this was a retrospective observational study, further research is required to clarify the significance of mediating factors.

Funding

This research received no external funding.

Conflicts of interest

The authors declare no conflict of interest.

References