Anticipatory Mobile Digital Health: Towards Personalised Proactive Therapies and Prevention Strategies.

Veljko Pejovic, Abhinav Mehrotra, Mirco Musolesi

TL;DR

This paper introduces the concept of Anticipatory Mobile Digital Health (AMDH), a paradigm shift from reactive to proactive healthcare using mobile and wearable sensors combined with machine learning. By predicting a user's future health state and modeling the impact of potential interventions, AMDH systems aim to deliver personalized preventative therapies before health issues manifest.

What Problem Does This Paper Address?

Modern healthcare is largely reactive and relies on a "one size fits all" approach, which is both economically unsustainable at scale and inefficient for managing chronic or behavioral conditions (such as depression or obesity). While mobile sensing has advanced significantly, most current applications are limited to diagnosing or inferring a user's current state (e.g., counting steps or detecting current stress levels).

The authors address the gap between passive state inference and active, preventative intervention. They propose a conceptual framework to move beyond simple prediction toward anticipation—using forecasts of a patient's health trajectory to actively steer them toward a healthier future.

Methods

The paper outlines a conceptual architecture for an anticipatory healthcare system driven by a closed feedback loop between the user and a mobile device.

Fig. 1. Anticipatory mobile systems predict context evolution and the impact that current actions can have on the predicted context. The feedback loop consisting of a mobile and a human enables the system to affect the future.

The proposed architecture relies on four core building blocks:

  1. Mobile Sensing: Managing adaptive sampling of physiological (e.g., heart rate, galvanic skin response) and environmental sensors to balance data granularity with battery life.
  2. Therapy and Prevention Toolbox: A curated set of interventions and behavioral change strategies overseen by medical experts.
  3. Machine Learning Core: This requires two distinct models: a context evolution model (predicting how the health state naturally evolves) and a therapy/prevention-effect model (predicting how a specific intervention will alter that trajectory).
  4. User Interaction Interface: A system designed to deliver therapies seamlessly based on personal interruptibility models.

To train the therapy-effect models, the authors suggest using Reinforcement Learning (RL), where the system acts as an agent maximizing a "reward" (positive health change). To mitigate the risks of testing unproven interventions on humans, they introduce the concept of Latent Learning—harnessing multimodal sensing to observe how a user naturally reacts to real-world environmental changes (e.g., tracking a user's mood before and after they voluntarily go out with friends) to model potential therapy outcomes without direct interference.

Fig. 2. Anticipatory mobile systems predict context evolution and the impact their actions can have on the predicted context. The feedback loop consisting of a mobile and a human enables the system to affect the future.

Datasets

Not discussed in the paper. (As a position paper, it outlines a theoretical framework rather than evaluating a specific empirical dataset).

Key Insights and Ideas

  • Prediction vs. Anticipation: Prediction merely forecasts the "original future" (e.g., a depressive episode is coming). Anticipation combines this forecast with reasoning and action to create a "modified future" (e.g., intervening early to prevent the episode entirely).
  • The Exploration vs. Exploitation Dilemma in Health: Standard RL algorithms learn by experimenting with unknown actions. In a digital health context, experimenting with the wrong therapy could cause irreversible psychological or physical harm, necessitating safer learning paradigms like latent learning.
  • Incorporating Genetics: True personalization cannot rely on sensor data alone. The authors emphasize that integrating static a-priori data, such as genetic backgrounds and pharmacogenomics, is essential for predicting how different demographics will react to identical digital therapies.

Key Results

Because this is a position paper defining a new research direction, it does not present quantitative experimental results or performance metrics. Instead, it synthesizes state-of-the-art advancements in mobile sensing (such as inferring depression from GPS mobility traces or stress from microphone data) to demonstrate that the building blocks for AMDH are technically viable.

Limitations and Future Work

The authors highlight several open challenges that must be addressed before AMDH can be integrated into clinical practice:

  • Resource Constraints: Continuous sensing rapidly drains mobile batteries, requiring more sophisticated, context-aware sensor scheduling.
  • Predictive Accuracy: While predicting location or basic mobility is an active field, reliable predictive models for complex, long-term human health states do not yet exist.
  • HCI and Timing: Interventions delivered at inappropriate times (e.g., during a meeting) cause annoyance and non-compliance. Better predictive models of user interruptibility are required.
  • Ethics and Accountability: The responsibility chain is completely undefined. If an automated intervention fails or causes harm, it is legally and ethically unclear who is to blame: the clinician who designed the therapy, the software engineer who wrote the ML algorithm, or the patient.
  • Privacy: To protect user privacy, sensor data should ideally remain on-device. However, siloed data complicates the training of robust, joint machine learning models across wider populations.

Summary updated

Abstract

The last two centuries saw groundbreaking advances in the field of healthcare: from the invention of the vaccine to organ transplant, and eradication of numerous deadly diseases. Yet, these breakthroughs have only illuminated the role that individual traits and behaviours play in the health state of a person. Continuous patient monitoring and individually-tailored therapies can help in early detection and efficient tackling of health issues. However, even the most developed nations cannot afford proactive personalised healthcare at scale. Mobile computing devices, nowadays equipped with an array of sensors, high-performance computing power, and carried by their owners at all time, promise to revolutionise modern healthcare. These devices can enable continuous patient monitoring, and, with the help of machine learning, can build predictive models of patient's health and behaviour. Finally, through their close integration with a user's lifestyle mobiles can be used to deliver personalised proactive therapies. In this article, we develop the concept of anticipatory mobile-based healthcare - anticipatory mobile digital health - and examine the opportunities and challenges associated with its practical realisation.
BibTeX
@misc{pejovic2015anticipatory,
  author = {Veljko Pejovic and Abhinav Mehrotra and Mirco Musolesi},
  title = {{Anticipatory Mobile Digital Health: Towards Personalised Proactive Therapies and Prevention Strategies.}},
  year = {2015},
  eprint = {1508.03722},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1508.03722},
}
Figure