JMIR Publications

ABSTRACT

Background:

Functional impairment is one of the most decisive prognostic factors in Complex Chronic Patients (CCP). A more significant functional impairment indicates that the disease is progressing, making adapting the diagnostic and therapeutic efforts necessary to avoid worsening.

Objective:

To predict alterations in the clinical condition of CCP by predicting the Barthel Index (BI) in order to assess their clinical and functional status utilizing an Artificial Intelligence (AI) model and data collected through an Internet of Things (IoT) mobility device.

Methods:

A two-phase pilot prospective single-center observational study was designed. During both phases, patients were recruited and a wearable activity tracker (WAT) was allocated to gather physical activity data. Data preprocessing and Machine learning (ML) techniques were used for analyzing mobility data. A decision tree was used to achieve a robust and interpretable model. To assess the quality of the predictions, several metrics were considered, including the Mean Absolute Error (MAE), Median Absolute Error (MAD), and Root Mean Squared Error (RMSE). Statistical analysis was performed using IBM SPSSv25 and Python 3.10.9 for the ML modeling.

Results:

90 CCP were included, 50 during phase 1, (10 for class A, 20 for class B and 20 for class C); 40 patients were enrolled during phase 2 (20 class B and 20 class C). 94.4% of patients had a caregiver. The mean value of BI was 58.31±24.5. Concerning mobility aids, 59.5% of patients required no aids, 20.2% required walkers, 16.9% wheelchairs, 6.7% cane, and 1.1% crutches. Regarding clinical complexity, 85% met poly-pathological patients criteria with a mean of 2.7±1.25 (2 RIC 6) categories, 68.5% frailty criteria and 21% CCP criteria. The most characteristic symptoms were dyspnea (82%), chronic pain (70.4%), asthenia (68.5%) and anxiety (46.1%). Polypharmacy was presented in 87% of patients. The most important variables for predicting the BI were identified as the maximum step count during both the evening and morning periods and the absence of a mobility device. The model exhibits consistency in the median prediction error with a MAD close to 5 in the training, validation and production-like test sets. Model accuracy for identifying the BI class was 91% in training, 88% in validation and 90% in test.

Conclusions:

Using commercially available mobility recording devices makes it possible to identify different mobility patterns and to relate them to functional capacity in multi-pathological patients according to the BI without using clinical parameters.