Review

Materials and methods

We conducted a narrative review and critical synthesis of the literature. The interdisciplinary nature of the five thematic domains (clinical maturity, patient experience and therapeutic alliance, algorithmic equity and bias, regulatory governance, and economic viability) necessitated a broad synthesis rather than a narrow quantitative systematic review. Inclusion criteria focused on peer-reviewed articles, implementation studies, and ethical analyses addressing AI applications in chronic pain, musculoskeletal (MSK) rehabilitation, patient-reported outcomes, and algorithmic bias.

Results

Spectrum of Clinical Maturity

AI interventions in pain rehabilitation range from pilot applications to integrated digital therapeutics (DTx) [7]. For example, a recent retrospective cohort study evaluated medical motion, an AI-composed exercise program for personalized pain management. The study demonstrated significant real-world efficacy, showing a 1.78-point reduction on the numeric rating scale (NRS) for pain intensity and a 3.11-point improvement in patient well-being over eight weeks [8]. Similarly, advanced ML models utilizing large retrospective cohorts have achieved high accuracy in predicting individual pain relief and identifying key clinical indicators that affect chronic pain outcomes [9].

The collection of real-world subjective and objective markers via mobile health (mHealth) applications is also proving vital for understanding chronic pain in daily life [10]. However, AI adoption success is determined less by algorithmic sophistication and more by operational fit. Clinicians must prioritize systems that seamlessly integrate into existing clinical workflows and offer clear explainability to support shared decision-making, ensuring that the technology augments multidisciplinary care without alienating the user.

Patient Experience and Digital Fatigue

While mHealth tools show great potential in facilitating adherence to chronic musculoskeletal pain (CMP) management [11], the patient experience is often complex. Qualitative reviews of digital health in chronic disease management reveal a dichotomy: users appreciate the reassurance of continuous monitoring, but frequently experience "ambivalence" and "digital fatigue" [12]. The burden of continuous tracking, poor usability in certain applications, and the anxiety caused by an influx of health data can lead to therapy dropout. In fact, ML models are now being specifically deployed to predict which patients are at the highest risk of dropping out of chronic pain treatments so that providers can intervene earlier [13].

The Digital Therapeutic Alliance

A major concern in digital rehabilitation is the loss of the human-clinician bond. However, conversational agents (CAs) and chatbots are increasingly utilized for health interventions and cognitive-behavioral techniques [14]. Tools such as Wysa, an AI conversational agent, have shown surprising efficacy in building rapport. In studies evaluating the therapeutic alliance of users interacting with Wysa, individuals reported a mean working alliance bond score of 3.98 out of 5, demonstrating that users can form a perceived therapeutic alliance with an AI agent that is comparable to human-delivered psychotherapy [15]. While patients appreciate the consistent availability of the AI agent, it must be positioned as an auxiliary tool that supports, rather than replaces, embodied care.

Algorithmic Bias and Corrective Initiatives

Bias in AI is a documented reality in healthcare. A landmark study demonstrated that a widely used predictive algorithm in the US systematically underestimated the illness severity of Black patients compared to White patients [16]. The algorithm used healthcare costs as a proxy for health needs; because unequal access to care means less money is historically spent on Black patients, the algorithm falsely predicted they needed less care, reducing the percentage of Black patients identified for necessary extra care by more than half [16].

Corrective efforts are vital. The ability to mitigate algorithmic bias requires deliberate action across the AI life cycle, moving toward models of "shared responsibility" between developers, healthcare facilities, and regulatory bodies [17]. Furthermore, there is an urgent need to apply a health equity lens throughout the AI development cycle, intentionally designing algorithms that address the needs of historically excluded or marginalized populations [18]. Some researchers argue for a consequentialist framework that directly evaluates how algorithmic outputs impact distinct populations, rather than simply relying on equalized error rates [19].

Economic Viability and Systemic Barriers

AI development and maintenance costs are substantial. Current reimbursement schemes, which are typically episodic and procedure-based, are fundamentally incompatible with the continuous monitoring required for AI-driven rehabilitation. Value-based, remote models show considerable promise in this space. Fully remote digital care programs for musculoskeletal conditions have been shown to maintain high completion rates and significant clinical improvements, even for high-risk patient populations with complex comorbidities such as severe obesity [20].

Discussion

The promise of AI in chronic pain rehabilitation lies not only in computational power but in its alignment with the core values of person-centered care. Our findings suggest that the success of AI tools depends less on technical sophistication and more on contextual relevance, operational realism, and ethical design. Regulatory frameworks and ethical governance ecosystems must evolve to address the complexities of algorithmic transparency and dynamic risk, ensuring safe, inclusive innovation [21].