Prediction of caregiver quality of life in amyotrophic lateral sclerosis using explainable machine learning

Abstract

Amyotrophic Lateral Sclerosis (ALS) is a rare neurodegenerative, fatal and currently incurable disease. People with ALS need support from informal caregivers due to the motor and cognitive decline caused by the disease. This study aims to identify caregivers whose quality of life (QoL) may be impacted as a result of caring for a person with ALS. In this study, we worked towards the identification of the predictors of a caregiver's QoL in addition to the development of a model for clinical use to alert clinicians when a caregiver is at risk of experiencing low QoL. The data were collected through the Irish ALS Registry and via interviews on several topics with 90 patient and caregiver pairs at three time-points. The McGill QoL questionnaire was used to assess caregiver QoL-the MQoL Single Item Score measures the overall QoL and was selected as the outcome of interest in this work. The caregiver's existential QoL and burden, as well as the patient's depression and employment before the onset of symptoms were the features that had the highest impact in predicting caregiver quality of life. A small subset of features that could be easy to collect was used to develop a second model to use it in a clinical setting. The most predictive features for that model were the weekly caregiving duties, age and health of the caregiver, as well as the patient's physical functioning and age of onset.

Figure 1

Barplot of MQoL-SIS scores. The vertical line between 6 and 7 demonstrates the way the data was split. The value 6 was the median of the distribution and was included in the low QoL class.

Figure 2

Precision-Recall Curves for the predictive models that were developed to identify the predictors of QoL.

Figure 3

Bar plot and summary plot of model M7 for the identification of predictors of caregiver QoL. (A) Bar plot showing features in order of importance based on the mean absolute value of the SHAP values for each feature. (B) Summary plot where features appear in order of their sum of SHAP value magnitudes, and SHAP values show the impact each feature has on the model output. The colour represents the feature value (red high, blue low). The suffix “.C” represents a caregiver characteristic.

Figure 4

Bar plot and summary plot of model “M10-CDSS”. (A) Bar plot showing features in order of importance based on the mean absolute value of the SHAP values for each feature. (B) Summary plot where features appear in order of their sum of SHAP value magnitudes, and SHAP values show the impact each feature has on the model output. The colour represents the feature value (red high, blue low).

Figure 5

Precision-Recall Curves of CDSS models.