AIMI: Leveraging Future Knowledge and Personalization in Sparse Event Forecasting for Treatment Adherence

TL;DR

The paper tackles forecasting medication adherence for chronic disease management by introducing AIMI, a knowledge-guided system that integrates smartphone-sensor data, historical adherence, and known future information about prescriptions. It compares CNN and LSTM architectures, showing that LSTM with future-knowledge features achieves high performance (accuracy $0.932$, macro F1 $0.936$), and demonstrates significant gains from including future knowledge and personalization. An incremental learning approach enables on-device training in memory-limited environments, highlighting practical feasibility for personalized, real-time interventions. Clinically, this work offers a path toward proactive, context-aware adherence support, with potential to reduce adverse outcomes and healthcare costs, while providing code and dataset resources for reproducibility.

Abstract

Adherence to prescribed treatments is crucial for individuals with chronic conditions to avoid costly or adverse health outcomes. For certain patient groups, intensive lifestyle interventions are vital for enhancing medication adherence. Accurate forecasting of treatment adherence can open pathways to developing an on-demand intervention tool, enabling timely and personalized support. With the increasing popularity of smartphones and wearables, it is now easier than ever to develop and deploy smart activity monitoring systems. However, effective forecasting systems for treatment adherence based on wearable sensors are still not widely available. We close this gap by proposing Adherence Forecasting and Intervention with Machine Intelligence (AIMI). AIMI is a knowledge-guided adherence forecasting system that leverages smartphone sensors and previous medication history to estimate the likelihood of forgetting to take a prescribed medication. A user study was conducted with 27 participants who took daily medications to manage their cardiovascular diseases. We designed and developed CNN and LSTM-based forecasting models with various combinations of input features and found that LSTM models can forecast medication adherence with an accuracy of 0.932 and an F-1 score of 0.936. Moreover, through a series of ablation studies involving convolutional and recurrent neural network architectures, we demonstrate that leveraging known knowledge about future and personalized training enhances the accuracy of medication adherence forecasting. Code available: https://github.com/ab9mamun/AIMI.

Figures (5)

• Figure 1: An overview of the AIMI system with the architecture of the LSTM-based forecasting model. The shapes of the data for a batch size 32 and forecasting with sensor, event, and knowledge features are indicated in parentheses.
• Figure 2: An illustration of the AIMI system's data processing method and separation of training and test data.
• Figure 3: Comparison of the CNN and LSTM models' accuracies and macro average F-1 scores in different training phases. The training phase labels, 1-1, 5-4, 9-4, 13-4, 17-4, and 22-5, indicate the numbers of training and test participants for those phases.
• Figure 4: The performance of the model drops for the test data of Participants 10 to 13 when the model is trained with 9 more additional participants. When the model is trained with the training data of Participants 10 to 13 again, the performance recovers on the test data of those participants.
• Figure 5: The architecture of the CNN model.