Personalization in AI-Driven mHealth Systems

Personalization in AI-Driven mHealth Systems

Why Do We Need Personalization in Health Apps?

• Every individual’s health journey is unique.
• We cannot design one-size-fits-all mHealth interventions.
• We need personalization systems that can adjust what, when, and how support is provided.

Generic mHealth App

Personalized mHealth App

What is Personalization?

Characteristics:

• User answers a questionnaire or sets preferences.
• App tailors content accordingly, but doesn’t adapt over time.

• Logic is rule-based, not data-driven.

Non-AI–Driven Personalization (Static)

Title

Use this side to give more information about a topic.

Personalization based on manual setup, user input, or static rules, rather than automated learning.

Personalization is the process of adapting app features, content, and interventions to an individual’s characteristics, behavior, and context using AI algorithms.

Subtitle

Example: A fitness app that asks for your age, gender, and weight, then recommends calorie goals.

Characteristics:

• Continuously analyzes sensor and usage data.
• Learns patterns and adapts interventions automatically.
• Uses algorithms like clustering, reinforcement learning, or predictive modeling.

User Data

AI Algorithms

Personalized Output

AI-Driven Personalization (Adaptive)

Title

Use this side to give more information about a topic.

Personalization is where the system evolves dynamically through machine learning or context-aware sensing to deliver relevant feedback or interventions.

Subtitle

Example: A sleep coach that changes bedtime notifications based on recent sleep quality.

AI-Driven Personalization

Non-AI-Driven Versus AI-Driven Personalization

Drag the words and place them in the correct column

Non-AI Personalization

AI-Driven Personalization

Complexity

Adaptation

Basis

Maintenance

Aspect

Example

High

App suggests general tips for women aged 50+

Continuous data learning

Manual Updates

Low

Static

User Input or Fixed Rules

Dynamic

Self-Improving Model

App detects stress via HRV → sends breathing prompt

Dimensions of Personalization

Personalization in mHealth is multi-dimensional, evolving from who you are (demographics), to how you behave (behavioral patterns), to what your body signals (physiological states)

Physiological

Contextual

Behavioral

From Static to Dynamic Personalization

Demographic

Difference between Static & Dynamic Personalization

• Static personalization = preset tailoring
• Rules are written once
• Example: “Send a motivational message every morning.”

→ Works well for predictable needs or early-stage design.

• Dynamic personalization = adaptive intelligence
• Rules evolve as AI learns what works best for this individual.
• Example: “System detects that a user has slept poorly. Based on this knowledge, it delays motivational message to afternoon when users are more receptive.”

→ Essential when user states, environments, or goals change over time.

Mechanism of Dynamic Personalization

• The engine behind dynamic personalization is the adaptive loop
• It refers to a continuous cycle of sensing, analyzing, deciding, acting, and evaluating.

• Each loop iteration helps the system personalize a little better for this user.

Act

Static Versus Dynamic Personalization (Again)

Drag the words and place them in the correct column

Static Personalization

Dynamic Personalization

Adaptation

Who defines behavior

How it works

Data dependency

Aspect

Example approach

Example scenario

Requires only initial setup data (demographics, preferences).

Rule-based or profile-based tailoring.

Reinforcement Learning, Bayesian Optimization, or Just-in-Time Adaptive Interventions (JITAIs).

None — fixed logic throughout use.

Requires ongoing data streams (behavioral, contextual, physiological).

Continuous — system refines timing, content, or delivery as it learns what works.

The AI system updates its own parameters as it observes user patterns (e.g., “If reminders at 8 PM are ignored, try 7 PM instead”).

Human designers or clinicians manually define all rules (e.g., “If user is female and over 60, show these tips”).

Uses learning algorithms that adapt based on user data over time.

A sleep app that always reminds users at 10 PM to relax.

Uses predefined rules or conditions set by designers.

A sleep app that learns each user’s bedtime patterns and adjusts notification time dynamically.

Benefits of Personalized Health

Personalization transforms engagement into adherence, feedback into trust, and real-world data into precision health.

Trust & Peceived Relevance

Behavor Change Effectiveness

Improved User Engagement

Supports Percision Health

Tailored interventions are more successful at changing health behaviors because they target the right barriers, triggers, and motivators for each person.

When an app consistently offers meaningful and accurate recommendations, users begin to trust its intelligence and perceive it as credible.

Personalization is the behavioral expression of precision health — tailoring care not just to clinical profiles but to everyday life data.

When an app feels personally relevant, users are more likely to stay engaged and follow through with recommended actions. Otherwise, messages can turn into "notification fatigue".

Precision Personalization vs. Precision Health vs. Precision Medicine

mHealth lives in the space of precision personalization; it’s where AI, sensors, and everyday behavior meet.

PERSONALIZATION

mHealth can be thought of as the precision layer that personalizes care in real time. It does not deliver genetic treatments (precision medicine) but it delivers behavioral adaptation.

Congratulations, you have completed this activity.

Generic mHealth App

A one-size-fits-all mhealth app that delivers the same content and reminders to every user, regardless of their goals, routines, or context.

Characteristics

• Fixed reminders and tips (e.g., “Drink 8 glasses of water daily!”)
• No adaptation based on user data or feedback
• Ignores personal factors like schedule, stress level, or preferences
• Limited engagement and short-term adherence

Example A step-tracking app that sends the same “Time to walk!” notification at 7 PM to all users every day.

Sense / Collect

The app gathers input data from sensors, user behavior, or self-reports. Data collection is an ongoing action. For example: Sleep, Steps, User Satisfaction Metric

Personalized mHealth App

An health app that adapts its recommendations, reminders, and feedback based on each user’s unique data, behaviors, and context.

Characteristics

• Learns from user inputs, habits, and sensor data.
• Adjusts goals dynamically (e.g., lowers daily step target after poor sleep).
• Sends context-aware prompts (e.g., “You have been sitting for 2 hours—time for a stretch?”)
• Builds engagement by matching tone, timing, and content to the individual.

Example A mindfulness app that detects elevated heart rate variability (HRV) and offers a short breathing exercise when stress is likely.

Evaluate / Update

• The system observes user response and updates its decision rules.

• Example: If user ignores the prompt, system delays future reminders, such as sending it the next day.

Decide / Adapt

• The model chooses an appropriate intervention (action) or modifies timing/content of a message.
• The sytem is always ready to act (“Decide/Act”) when certain thresholds or patterns are detected.

• Example: Decides a 5-min breathing break at 4 p.m. is the best option for the user under the circumstances.

Benefits & Design Responsibilites

This closed feedback loop lets the app adjust automatically over time, forming the basis of dynamic and adaptive interventions.

Benefits

• Continuous personalization leads to higher engagement.
• Learns optimal timing and message tone.
• Supports precision health in everyday life.

Design Responsibilities

• Maintain user control: allow opting-out or pausing loops.
• Be transparent: explain why an action was triggered.
• Protect data: loops depend on continuous sensing.

Behavioral Ring

• Includes observed routines and interactions.
• More dynamic than demographic information as people can change their behaviors.

• Example Data: Steps, app use, dietary intake, routines.

Physiological Ring

• Based on biometric and physiological signals that reflect the body’s real-time state.
• Enables real-time, adaptive support

• Example Data: HRV, PPG, SpO₂, sleep sensors.

Context Ring

• Incorporates environmental and situational cues (location, time, weather).
• More dynamic, since contextual factors are subject to change.

• Example Data: GPS, calendar, notifications, weather details

Analyze / Learn

• AI or rules detect patterns or changes.
• Conceptually, this is a continuous process, the system is always looking for changes or trends.
• However, practically, it is interval-based, i.e. the system does this after a certain time period (time-triggered) or when a certain event happens (event triggered).
• Example: The system learns that the user walks less on stressful days, or stress levels spike at 4 p.m.

Dimensions of Personalization

Independent examples:

• Language localization → Demographic only

• Goal adjustment → Behavioral only.

• Location-triggered message → Contextual only

• HRV-based stress alert → Physiological only.

• The inner rings (e.g., Demographic) represent the simplest and most stable (static) forms of personalization, i.e. rules decide what kind of personalization will be offered.
• The outer rings (e.g., Physiological) represent more dynamic, adaptive, and privacy-sensitive forms. (Adaptive Loop)
• Depending on app's design, AI can personalize using any one or subset of rings.

• Combining rings creates richer and more adaptive interventions.
• Each ring introduces new capabilities — and new ethical responsibilities.

Combined examples:

• Personalized stress intervention → Contextual + Physiological.
• Lifestyle coaching → Behavioral + Demographic.

Demographic Ring

• The base layer of personalization that is foundational and necessary for all personalizations.
• Least dynamic, since a person's demographics do not change.

• Example Data: age, gender, language, culture

Act / Deliver

• This is when the system delivers the intervention to the user.
• The app provides personalized feedback, notification, or suggestion.
• Example: Sends an activity prompt on smartwatch, such as “Try a 5-minute walk before your next meeting.”