Digital therapeutics are moving from static software interventions towards adaptive models that respond to patient needs over time. Dynamic Personalized Optimization, a framework for continuously adjusting treatment to an individual patient, defines how artificial intelligence can support real-time personalisation and ongoing treatment refinement. It does this by combining patient characteristics, clinical status, treatment content and post-treatment feedback to guide each next intervention. The framework is particularly relevant for conditions that require ongoing adjustment, including tinnitus, depression and mild cognitive impairment. By setting out the functions needed for dynamic personalisation, it provides a structured model for more responsive digital care.
From Timed Interventions to Content Optimisation
Dynamic Personalized Optimization sets out the core artificial intelligence functions needed to support adaptive treatment in digital therapeutics. It focuses on integrating patient data, treatment history, usage patterns and outcome signals to refine therapeutic strategies continuously. Instead of fixed pathways, treatment becomes an iterative process in which each patient response informs the next step.
Must Read: Digital Health Moves into Routine Care
The framework builds on just-in-time adaptive interventions but shifts emphasis from timing to treatment content. Rather than only determining when to intervene, it prioritises selecting the most appropriate therapeutic content at each stage.
Four data types structure this process. User data include demographics, medical history and lifestyle information. Status measurement data capture objective indicators such as symptom severity and test results. Treatment content data describe the intervention itself, including format and type. Feedback data reflect the patient’s response, such as reported feelings or task performance scores. These elements form a continuous sequence, beginning with baseline information and progressing through repeated cycles of treatment delivery, feedback collection and status updates. In this model, therapy evolves through ongoing adjustment rather than remaining static.
Two Routes to Selecting the Next Best Intervention
The framework outlines two methods for selecting treatment content. The first approach predicts feedback for each candidate intervention. Artificial intelligence models analyse historical sequences of user data, status, treatment content and feedback to estimate the likely response to each option. The system then selects the intervention expected to generate the strongest feedback.
This method treats feedback as an indicator that can inform future treatment decisions. Although feedback does not directly measure clinical outcomes, it can guide optimisation by reflecting engagement and response patterns.
The second approach predicts future patient status. Using prior data and a candidate intervention, the model estimates the resulting patient condition. By comparing predicted outcomes across options, the system identifies the intervention most likely to improve status relative to baseline.
Both methods remove the need to define the optimal treatment content in advance. Because effectiveness varies between individuals and contexts, prediction replaces predefined selection. Treatment is therefore determined dynamically, with each step informed by accumulated patient-specific data.
Mild Cognitive Impairment as an Example of Iterative Care
A cognitive intervention pathway for mild cognitive impairment illustrates how the framework operates in practice. The process begins with user data such as age, education level, daily activity and sleep patterns, alongside baseline measurements reflecting cognitive status, including Mini-Mental State Examination scores.
The system tracks previously delivered treatment content and associated feedback. In this context, treatment may include cognitive exercises targeting memory, attention and problem-solving. Feedback may include performance scores reflecting engagement and task outcomes. Using these historical sequences, the model predicts expected feedback or postsession status for each candidate intervention.
These predictions guide selection of the next treatment element. After delivery, new feedback is incorporated and patient status is updated, allowing the process to repeat. This iterative cycle enables continuous refinement of therapy based on individual response patterns.
Through repeated adaptation, treatment becomes increasingly aligned with patient needs. The framework positions this cycle as a mechanism to support both engagement and cognitive function by tailoring interventions over time.
Dynamic Personalized Optimization provides a structured model for integrating artificial intelligence into digital therapeutics. By combining multiple data types in a continuous feedback loop, it enables treatment that adapts in real time to patient response. The framework also highlights the role of advanced models in processing diverse data formats and supporting integrated analysis. At the same time, it emphasises the need to address privacy, security, transparency, fairness and accountability in implementation. Clinical validation and real-world deployment remain essential next steps. If applied effectively, this approach could support more responsive and personalised digital care while improving engagement and therapeutic outcomes.
Source: JMIR Medical Informatics
Image Credit: iStock
References:
Rim D (2026) Dynamic Personalized Optimization: An AI Functionality Framework for Digital Therapeutics. JMIR Med Inform;14:e75256.
