Monitoring Opioid-Use-Disorder Treatment Adherence Using Smartwatch Gesture Recognition

Abstract

The opioid epidemic in the United States has significantly impacted pregnant women with opioid use disorder (OUD), leading to increased health and social complications. This study explores the feasibility of using machine learning algorithms with consumer-grade smartwatches to identify medication-taking gestures. The research specifically focuses on treatments for OUD, investigating methadone and buprenorphine taking gestures. Participants (n = 16, all female university students) simulated medication-taking gestures in a controlled lab environment over two weeks, with data collected via Ticwatch E and E3 smartwatches running custom ASPIRE software. The study employed a RegNet-style 1D ResNet model to analyze gesture data, achieving high performance in three classification scenarios: distinguishing between medication types, separating medication gestures from daily activities, and detecting any medication-taking gesture. The model's overall F1 scores were 0.89, 0.88, and 0.96 for each scenario, respectively. These findings suggest that smartwatch-based gesture recognition could enhance real-time monitoring and adherence to medication regimens for OUD treatment. Limitations include the use of simulated gestures and a small, homogeneous participant pool, warranting further real-world validation. This approach has the potential to improve patient outcomes and management strategies.

Keywords: context-aware environments; ecological momentary assessment; human activity recognition; machine learning; medication detection; neural networks; smart healthcare; wearable sensors.

Figure A1

Step-by-step protocol for medication-taking with a bottle. This protocol outlines the standardized procedure for medication intake using a bottle. Steps include verifying the bottle’s contents (Step 0), securely holding the bottle (Step 1), opening the cap (Step 2), lifting the bottle for ingestion (Step 3), securely closing the cap (Step 4), and repeating the process for both hands to ensure consistent data collection (Step 5).

Figure A2

Step-by-Step Protocol for Medication-Taking with a Packet. This protocol details the steps for medication intake using a sublingual film or powdered packet. The procedure involves tearing open the packet and pouring out the medication (Step 0), securely holding the packet (Step 1), tearing it open (Step 2), placing the medication under the tongue (Step 3), and repeating the process for both hands to ensure comprehensive data collection (Step 4).

Figure 1

Example smartwatch accelerometer data for the buprenorphine gesture. This figure illustrates accelerometer data collected during a simulated buprenorphine-taking gesture. The motion pattern highlights distinct sub-actions: (A) tearing open the top of the packet and bringing the medication toward the mouth; (B) placing the sublingual film under the tongue; and (C) lowering the hand back down.

Figure 2

Example smartwatch accelerometer data for the methadone gesture. This figure presents accelerometer recordings from a smartwatch worn during a simulated methadone-taking gesture. The motion trajectory illustrates key phases of the action: (A) unscrewing the cap (not visible in accelerometer data, as the hand without the watch performed this action); (B) raising the bottle to the mouth and ingesting the liquid; (C) lowering the bottle back down; and (D) the completion of the gesture.

Figure 3

The distribution of bout durations for methadone and buprenorphine. The histogram represents the frequency of bout durations (in seconds), with overlaid kernel density estimates (KDEs) for both substances. Dashed vertical lines indicate the median bout duration for each group. Methadone is shown in blue, while buprenorphine is shown in orange.

Figure 4

Model performance and gesture classification results for Scenario 1. In this scenario, the model performs binary classification to differentiate between two medication-taking gestures: methadone (liquid) and buprenorphine (sublingual film). (A) Training and validation loss curves, showing model convergence and generalization over epochs. (B) F1 score curves for training and validation sets, indicating classification performance across medication types. (C) Precision–recall curve demonstrating the model’s ability to distinguish between methadone and buprenorphine gestures. (D) Precision matrix of the test set, illustrating the proportion of correctly classified gestures for each medication type. (E) Recall matrix of the test set, highlighting model sensitivity to each medication-taking gesture. (F) t-SNE visualization of latent space representations, showing the clustering of methadone and vuprenorphine gestures.

Figure 5

Model performance and gesture classification results for Scenario 2. In this scenario, the model classifies gestures into three categories: methadone, buprenorphine, and daily living. (A) Training and validation loss curves demonstrate model convergence over epochs. (B) F1 score curves for training and validation sets indicate classification performance across classes. (C) Precision–recall curves highlight the model’s ability to distinguish between the three gesture categories. (D) Precision matrix of the test set, showing the proportion of correct predictions for each class. (E) Recall matrix of the test set, illustrating the sensitivity of the model to each class. (F) t-SNE visualization of latent gesture representations, demonstrating the separation between the methadone, buprenorphine, and daily living gestures.

Figure 6

Model performance and gesture classification results for Scenario 3. In this scenario, the model performs binary classification to distinguish between medication-taking gestures and daily living activities. (A) Training and validation loss curves showing model convergence and generalization over epochs. (B) F1 score curves for training and validation sets, representing classification performance. (C) Precision–recall curve illustrating the trade-off between recall and precision for detecting medication-taking gestures. (D) Precision matrix of the test set, showing the proportion of correct predictions for each class. (E) Recall matrix of the test set, highlighting model sensitivity to medication-taking gestures. (F) t-SNE visualization of latent space representations, demonstrating separation between medication-taking and daily living gestures.