Remote Monitoring of Chemotherapy-Induced Peripheral Neuropathy by the NeuroDetect iOS App: Observational Cohort...

Chemotherapy-induced peripheral neuropathy (CIPN) is a common and debilitating adverse effect of neurotoxic chemotherapy characterized by symptoms such as numbness, tingling, and weakness. Effective monitoring and detection of CIPN are crucial for avoiding progression to irreversible symptoms. Due to the inconvenience of clinic-based objective assessment, CIPN detection relies primarily on patients' reporting of subjective symptoms, and patient-reported outcomes are used to facilitate CIPN detection. Our previous study found evidence that objective functional assessments completed within a smartphone app may differentiate patients with and those without CIPN after treatment.

This prospective, longitudinal observational cohort study aimed to determine the feasibility and accuracy of app-based remote monitoring of CIPN in patients with cancer undergoing neurotoxic chemotherapeutic treatment and to conduct exploratory comparisons of app-based functional CIPN monitoring versus patient-reported outcome-only monitoring.

The NeuroDetect app (Medable Inc) includes subjective EORTC (European Organization for Research and Treatment of Cancer) Quality of Life Questionnaire (QLQ)-20-item scale (CIPN20) and 6 objective functional assessments that use smartphone sensors to mimic neurological examinations, such as walking, standing, and manual dexterity tests. The functional assessment data were collected from patients with cancer undergoing neurotoxic chemotherapy, and a neurological examination was conducted at the end of treatment to diagnose CIPN in the feet (CIPN-f) or CIPN in the hands (CIPN-h). Various classification models including NeuroDetect features only (NeuroDetect Model) CIPN20-only (CIPN20 Model) or a combination of both (Combined Model) were trained and evaluated for accuracy in predicting CIPN probability.

Of the 45 patients who completed functional assessments and neurological examinations, 24 had CIPN-f, and 29 had CIPN-h. The NeuroDetect Model could discriminate between patients with and those without CIPN-f (area under the curve=83.8%, comparison with no information rate P=.02) but not CIPN-h (area under the curve=67.9%, P=.18). The rolling rotation features from the eyes-closed phase of the Romberg Stance assessment showed the greatest contribution to CIPN-f (40% of total variable importance) and the Finger Tapping assessment showed the greatest contribution to CIPN-h (85% of total variable importance). The NeuroDetect Model had numerically, and at some time points statistically, superior performance to the CIPN20 Model in both CIPN-f and CIPN-h, particularly before and early in treatment. The Combined Model numerically, though not statistically, outperformed either assessment strategy individually, indicating that the combination of functional and patient-reported assessment within a smartphone may be optimal to CIPN detection.

Our findings demonstrate the feasibility of integrating subjective and objective CIPN assessment into a smartphone app for remote, longitudinal CIPN monitoring. Studies of larger patient cohorts are needed to refine the app-based CIPN detection models and determine whether their use in practice improves CIPN detection.