DeepaMed: Deep learning-based medication adherence of Parkinson's disease using smartphone gait analysis

Objectives

Parkinson's disease (PD) is a neurodegenerative chronic disorder with multiple motor and non-motor symptoms. As PD has no ultimate cure, physicians aim to delay PD complications, especially those that degrade the patient's quality of life such as motor symptoms and dyskinesia. Patients' lack of adherence to prescribed medication is a major challenge for physicians, especially for patients suffering from chronic conditions. The Centers for Disease Control and Prevention (CDC) estimates that medication non-adherence causes 30 to 50 percent of chronic disease treatment failures and 125,000 deaths per year in the USA (U.S. Foods and Drugs Administration (FDA) “Why You Need to Take Your Medications as Prescribed or Instructed” https://www.fda.gov/drugs/special-features/why-you-need-take-your-medications-prescribed-or-instructed. June 2021). In PD patients particularly, adherence varies between 10% and 67% (Straka et al., 2019Straka, Igor, et al. "Adherence to pharmacotherapy in patients with Parkinson's disease taking three and more daily doses of medication." Frontiers in neurology 10 (2019): 799).

Objective

The goal of this work is to remotely determine whether PD patients have taken their medication, by analyzing gait data gathered from their smartphone sensors. Using this approach, physicians can track the level of medication adherence of their PD patients.

Methodology

Using data from the mPower study (Bot et al., 2016), we selected 152 PD patients who recorded at least 3 walks before and 3 after taking medications and 304 healthy controls (HC) who recorded 3 walks at minimum. We extracted each subject's gait cycle from their accelerometer and gyroscope sensors data. The sensor data corresponding to gait cycles were fed to DeePaMed; a multilayer Conventional Neural Network (CNN), crafted for patches of gait strides. DeePaMed classified 30 s of a walk as either PD patient “On” vs. “Off” medication, or if the gait data belongs to an HC.

Results

Our DeePaMed model was able to discriminate PD patients on-vs off-medication and baseline HC walk with an accuracy of 98.2%. The accuracy of our CNN model surpassed that of traditional Machine Learning methods by over 17%. We also found that our model performed best with inputs containing a minimum of 10 full gait strides.

Conclusion

Medication non-adherence can be accurately predicted using smartphone sensing of the motor symptoms of PD, suggesting that PD patients’ medication response and non-adherence can be monitored remotely via smartphone-based measures.