Artificial intelligence for predicting clinical response and therapeutic value in interventional pain: a scoping review

Chronic pain is one of the leading causes of disability worldwide and represents a constantly increasing health, social, and economic burden. According to the Global Burden of Disease According to a 2019 study , musculoskeletal disorders, led by low back pain, consistently remain among the top causes of years lived with disability in all regions of the world, affecting more than 600 million people (1,2). In economic terms, the impact of chronic pain translates into direct and indirect costs exceeding hundreds of billions of dollars annually, which has led to its management being considered a strategic priority for contemporary health systems (3). In this context, a significant proportion of patients with refractory chronic pain are referred to specialized units, where interventional pain management has experienced sustained growth over the past few decades. This growth has been reflected in a considerable increase in the use of procedures such as epidural injections, image-guided blocks, radiofrequency ablation, and neuromodulation techniques (4,5). However, this expansion has not been uniformly accompanied by a systematic evaluation of the sustained clinical benefit, functionality, or cost-effectiveness of these interventions in real-world clinical practice settings. Consequently, multiple studies have documented marked variability in the utilization rates of interventional procedures across geographic regions, institutions, and even among professionals within the same center—variability that exceeds what would be expected based on epidemiological or clinical differences (6,7). This heterogeneity in practice has therefore become an indirect indicator of potential patterns of overuse in certain contexts and underuse in others, raising important questions about the appropriateness and value of these interventions. In this context, overuse occurs when interventional procedures are performed on patients with a low probability of sustained clinical benefit or when interventions are repeated despite limited or nonexistent previous responses. Conversely, underuse occurs when potentially effective interventions are not offered to patients who could benefit from them, reflecting inequities in access, biases in candidate selection, and deficiencies in clinical decision-making processes (8,9). Furthermore, low-value practices include interventions that provide marginal or inconsistent benefits on clinically relevant outcomes such as functionality, quality of life, or sustained pain relief. Therefore, the impact of these practices transcends the clinical setting and extends to the sustainability of health systems. It is estimated that between 20 and 30 percent of total health expenditure in high-income countries is associated with low-value interventions, with a particularly significant impact in areas characterized by high technology intensity (10,11). Consequently, the identification and reduction of these practices has become a central objective of value-based healthcare models (12). In response to this challenge, international initiatives such as Choosing Wisely have promoted recommendations aimed at reducing unnecessary procedures and encouraging evidence-based clinical decisions (13). However, in the field of interventional pain management, the implementation of these approaches faces substantial limitations. The subjective nature of pain, clinical heterogeneity, the coexistence of comorbidities, and interindividual variability in response to procedures make it difficult to apply uniform criteria based solely on clinical guidelines or randomized controlled trials (14,15). Due to these limitations, traditional approaches are insufficient to capture the complexity of the decision-making process in interventional pain management. In this scenario, artificial intelligence emerges as a tool with transformative potential. Machine learning models allow for the analysis of large volumes of longitudinal clinical data, the integration of multidimensional information, and the detection of complex patterns of response, non-response, and unnecessary repetition of procedures that are not easily identifiable using conventional analytical methods (16,17). In other areas of medicine, artificial intelligence has demonstrated the ability to identify inappropriate variability in clinical practice, predict therapeutic outcomes, and support strategies aimed at reducing low-value interventions (18,19). Consequently, its application to interventional pain management offers the opportunity to identify patterns of overuse, detect clinically relevant scenarios of underuse , and recognize low-value practices through the integrated analysis of clinical, procedural, and patient-reported outcome data (20). However, despite this potential, the available evidence on the use of artificial intelligence to identify overuse, underuse , and low-value practices in interventional pain management is scattered and exhibits significant methodological heterogeneity. To date, there is no structured synthesis that allows for a comprehensive mapping of this emerging field. Therefore, a scoping The review is configured as the most appropriate methodological design to systematically explore the scope, characteristics and gaps of the available literature, allowing the description of the current state of knowledge and guiding future lines of research and clinical development (21,22). The period between January 2015 and January 2026 was selected because it corresponds to the stage in which modern machine learning and deep learning models began to be systematically applied in clinical decision support systems and in the analysis of outcomes in interventional pain . The objective of this scoping The review aims to map and characterize the existing evidence on the use of artificial intelligence to identify overuse, underuse , and low-value practices in interventional pain management.