Skip to content

Adaptive Text Messaging System Allows Personalized Approach to Health Promotion

Key findings

  • Researchers developed the first adaptive text messaging system for psychological and physical health promotion that uses a machine learning algorithm to personalize messages to each patient's needs and preferences
  • Text messaging works well with inexpensive cell phones, making it accessible to low-income populations
  • The system is also the first to both include content in English and Spanish and combine health-related messaging with content to promote facets of psychological well-being that are linked to cardiac health behaviors

Mobile apps for delivering health education and motivational messages have advantages over in-person counseling, but they can be time-consuming and complex. Text messaging is simpler and, unlike mobile apps, is suitable for inexpensive phones. Text messaging may therefore be a great way to support low-income populations in improving their health behaviors.

However, text messaging has shown mixed results for promoting cardiovascular health in previous research. Jeff C. Huffman, MD, director of the Cardiac Psychiatry Research Program, Christopher M. Celano, MD associate director of the Cardiac Psychiatry Research Program, and colleagues have created a text messaging system that has several enhancements compared with previous efforts. It concurrently targets a number of different elements:

  • Psychological well-being
  • Health education
  • Health support

It also uses a machine algorithm that learns over time how to personalize messages for each patient.

In The Primary Care Companion for CNS Disorders, the research team describes the initial rollout of the system to an urban primary care population.

Text Message Content

The researchers used numerous sources to develop a variety of text messages, including prior studies of positive psychology and text message systems in healthy and medically ill populations, national health organizations, traditional psychotherapeutic interventions (e.g., cognitive-behavioral therapy) and their own previous research on delivery of positive psychology and health-related content.

The team selected 320 brief messages that fell into five categories:

  1. Positive psychology activities performed alone (e.g., writing a gratitude list)
  2. Positive psychology activities performed with others (e.g., expressing thanks to another person)
  3. Physical activity
  4. Healthy eating
  5. Stress reduction

Algorithm Creation and Message Adaptation

The program was designed such that patients would receive one text message each day (in English or Spanish) and send feedback about its utility (helpful/not helpful). The team created an algorithm that recorded the proportion of messages each patient liked or disliked for each message type, then more frequently provided messages of the favored types.

The algorithm can incorporate more than five message types. In addition, multiple attributes (e.g., positive psychology-based, type of health behavior) can be assigned to a single message. For example, the algorithm might send a message that encourages performing physical activity with a friend, which is both prosocial and exercise-related.

Pilot Demonstration

The text messaging system was piloted at an urban primary care clinic where 71% of patients are Hispanic/Latino and 60% identify as being best served in a language other than English (Spanish in nearly all cases). The program was initiated in January 2017 for patients with at least one chronic medical condition, then was expanded to include any patient identified as needing support to improve their health behaviors.

To test the logistics of enrolling participants, the research team started by sending 74 patients a fixed set of text messages. Then they began using the adaptive system. All participants in both phases were able to join the program on their cell phone and receive text messages.

Toward the Future

The researchers plan to determine the rates of successful message delivery and rates of feedback to the daily messages to assess the feasibility and acceptability of the program. In addition, they plan to interview a representative sample of participants to obtain detailed data on patient experiences.

Even if the intervention does prove to be feasible in the primary care setting, the authors say, more work will be needed to assess whether the text messaging system affects patient well-being, changes health behaviors and influences specific health metrics such as lipid levels.

Learn more about the Cardiac Psychiatry Research Program

Explore psychiatry research at Mass General


Massachusetts General Hospital researchers are using big data and machine learning to find patterns or predictive profiles that may indicate risk of suicide.


Previous research has shown that diagnostic codes routinely collected in electronic health records can help predict domestic abuse an average of two years in advance. Could EHR systems also be used to predict suicidal behavior?