Artigos

Conteúdo (article):

Potencialidades e desafios do uso da Inteligência Artificial no Manejo das Dores Crônicas: Uma Revisão de Escopo
Potentialities and Challenges of Using Artificial Intelligence in Chronic Pain Management: A Scoping Review
Frederico de Oliveira Meirelles – Universidade Estácio de Sá / Programa de Pós-graduação em Saúde da Família UNESA/IDOMED – fredericomeirelles@hotmail.com – ORCID: 0000-0001-6075-2106.
Mariana Inocêncio Matos – Programa de Pós-graduação em Ciências do Exercício e do Esporte (PGCEE) Universidade do Estado do Rio de Janeiro - mariana.i.matos@gmail.com - ORCID: 0000-0002-2717-062X.
Katia Maria Braga Edmundo - Programa de Pós-graduação em Saúde da Família UNESA/IDOMED - katiaedmundo@gmail.com – ORCID: 0000-0001-5726-1998.
Claudia Leite de Moraes - Instituto de Medicina Social Hésio Cordeiro e Programa de Pós-graduação em Saúde da Família UNESA/IDOMED - clmoraes.uerj@gmail.com – ORCID: 0000-0002-3223-1634.
Resumo
Objetivo: Mapear o que vem sendo produzido na literatura científica sobre o uso da inteligência artificial (IA) no manejo das dores crônicas, principal motivo de incapacidade no mundo. Métodos: Revisão de escopo realizada seguindo as diretrizes PRISMA-ScR. A busca de artigos foi realizada nas bases de dados: MEDLINE; LILACS; SPORTDiscus; SCOPUS; CENTRAL; Science Direct; IEEE Xplore; e Association for Computing Machinery Digital Library and arXiv. Combinou-se os descritores, “Chronic Pain” e seus sinônimos com os termos “Artificial Intelligence” ou “Machine Learning” ou seus sinônimos. Resultados: Inicialmente, identificou-se 2767 artigos, sendo selecionados e analisados 109 estudos. Percebe-se que o volume de produção científica vem crescendo ao longo dos anos. A maioria dos algoritmos de IA utilizados focalizaram a classificação de indivíduos de acordo com o tipo de dor crônica e a predição de grupos de risco. Support Vector Machine, Random Forest, Redes Neurais, Regressão logística e K-means foram os algoritmos mais usados. Poucos estudos focalizaram o uso da IA na Atenção Primária (APS). Conclusão: Destaca-se o potencial da IA em prevenir e melhorar o manejo das dores crônicas, mas ainda há importante carência de estudos focados na APS, porta de entrada do SUS e setor privilegiado para a prevenção e manejo adequado do problema.
Palavras-chave: Dor crônica; Inteligência Artificial; Aprendizado de máquina; Atenção Primária à Saúde; Saúde da Família.

Abstract
Objective: To map the scientific literature on using artificial intelligence (AI) in managing chronic pain, the leading cause of disability worldwide. Methods: A scoping review was conducted following PRISMA-ScR guidelines. We searched for articles in MEDLINE, LILACS, SPORTDiscus, SCOPUS, CENTRAL, Science Direct, IEEE Xplore, Association for Computing Machinery Digital Library, and arXiv databases. Descriptors “Chronic Pain” and its synonyms were combined with “Artificial Intelligence” or “Machine Learning” or their synonyms. Results: Initially, 2767 articles were identified, and 109 studies were selected and analyzed. The volume of scientific production has been increasing over the years. Most adopted AI algorithms focused on classifying individuals by type of chronic pain and predicting risk groups. Support Vector Machine, Random Forest, Neural Networks, Logistic Regression, and K-means were the most adopted algorithms. Few studies focused on AI use in Primary Health Care (PHC). Conclusion: We underscore AI’s potential to prevent and improve chronic pain management. However, there is still a significant lack of studies focused on PHC, the gateway to the SUS and a privileged sector for prevention and proper management of the problem.
Keywords: Chronic pain; Artificial Intelligence; Machine Learning; Primary Health Care; Family Health.

INTRODUCTION
Chronic pain is a common and complex health problem with a significant impact on society(1). It is one of the biggest global public health problems and the leading cause of disability worldwide, according to the Global Burden of Disease Study(2). Despite being one of the most researched health topics, evidence to date has been unable to change its status as the principal cause of disability on the planet(3).
According to the International Association for the Study of Pain (IASP), pain is an unpleasant sensory and emotional experience associated with actual or potential tissue damage(4). Chronic pain is when this pain is continuous or intermittent for more than 12 weeks. Primary chronic pain (PCP) is defined as pain with no apparent cause that persists for more than three months in one or more anatomical regions, causing significant emotional distress (anxiety, anger/frustration, or depressed mood) or functional disability (interference with daily activities and reduced participation in social roles)(5). The process that can lead to PCD is multifactorial. Biological, psychological, and social factors contribute to the pain syndrome(6). This classification describes that secondary chronic pain includes chronic pain related to cancer(7), chronic post-surgical or posttraumatic pain(8), chronic neuropathic pain(9), orofacial pain or secondary chronic headache(10), secondary chronic visceral pain(11), and chronic secondary musculoskeletal pain(6,12) (Supplementary Material 1).
The most significant difficulty in treating chronic pain is the multiplicity of factors complexly interconnected in the genesis of the problem(13). As the causes are diffuse, the management of the problem ends up being non-specific. This situation ends up reducing the effectiveness of therapeutic proposals(14). Encouraging studies on interventions in which the treatment or prevention of pain in individuals is based on an individual’s absolute risk and not a general risk could hypothetically generate evidence to better support the choice of therapeutic strategies(15). The observation of the low effectiveness of interventions, the syndrome’s recurrence, and its managing challenges seem to point to a necessary investment in new diagnostic and therapeutic tools.
Recently, the literature has suggested several uses of artificial intelligence (AI) in health. Among the possibilities for using AI, we underscore screening and diagnosing health problems, supporting treatment decisions, indicating specialists, estimating prognosis, analyzing medical records, building knowledge bases, extracting information, and providing descriptive information. AI tools target healthcare professionals, healthcare managers, and patients or guardians(16). As a tool that, based on previous data, can understand patterns and rules and thus estimate the risk of a given outcome(17), AI has been increasingly used to support decision-making at different healthcare stages in recent years. However, we still do not have sufficient knowledge to indicate the potential and challenges of its use in health. Thus, research on the subject should be encouraged.
Exploring the potential of artificial intelligence algorithms used in the management of individuals with chronic pain can provide evidence to promote health and prevent new cases, increase the effectiveness of therapeutic proposals in the treatment of cases, and reduce the adverse consequences of such situations to the well-being and quality of life of individuals who suffer from the problem(18).
The Unified Health System (SUS), established by the 1988 Constitution and based on the principles of universality, comprehensiveness, and equity, ensures free access to healthcare for all Brazilians(19). Primary Health Care (PHC) holds a central position in the structure of the SUS and is the main gateway to the Healthcare Network (RAS)(20). In Brazil, PHC is predominantly organized by the Family Health Strategy (ESF), aiming at health promotion, disease prevention, and continuous and comprehensive care(19,21). Given the SUS complexity and scope, adopting emerging technologies, such as Artificial Intelligence (AI) tools, is a promising opportunity to streamline management and service delivery. In PHC, AI can improve screening, monitoring of chronic diseases, and care personalization, enhancing health promotion and disease prevention actions.
In order to expand knowledge about the use of AI in health services, this study aimed to conduct a comprehensive scoping review on the use of AI in the management of chronic pain in health services, especially PHC/ESF, identifying the AI application scenarios studied, the types of algorithms used for managing chronic pain and the most promising ones, the main predictors used in these algorithms, the population that benefits from the results of the AI application (managers, health professionals, and users), which chronic pain conditions have been most studied and those with growth potential, besides pointing out knowledge gaps on the subject.
METHODS
Drafting and registration
This review was planned, conducted, and written under the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR)(22,23). The project was registered in the Open Science Framework (OSF) under Identification N° DOI 10.17605/OSF.IO/2GE63, available at https://osf.io/2ge63/.
Study design
This Scoping Review complies with the guidelines of the JBI Manual for Evidence Synthesis(24), available at https://synthesismanual.jbi.global.
Review question
What scientific literature on using artificial intelligence tools in managing chronic pain is currently available?
Eligibility criteria
Inclusion criteria
The inclusion criteria are a) sample/population – people with chronic musculoskeletal pain and healthy individuals; b) use of artificial intelligence technique at any time during rehabilitation; c) analysis of any type of outcome, preferably clinical outcomes; and d) no restrictions based on ethnicity, gender, age, or religion.
Exclusion criteria
As exclusion criteria, we considered a) studies published in journals without peer review, b) research protocols or other publications that do not bring original research results, and c) studies that have used any data analysis method other than Artificial Intelligence.
Bibliographic search strategy
On 28/03/2023, a bibliographic search was conducted without time or language filter in the following databases: MEDLINE (US National Library of Medicine); LILACS (Latin American and Caribbean Literature in Health Sciences); SPORTDiscus; SCOPUS; CENTRAL; Science Direct; IEEE Xplore; Association for Computing Machinery Digital Library; and arXiv.
Two independent authors developed the search key, and disagreements were discussed and resolved with the entire group of authors. Boolean operators “OR” between synonyms and “AND” between descriptors were used to create the search key. The descriptors found in the Health Sciences Descriptors (DeCS) and Medical Subject Headings (MeSH) databases, “Chronic Pain” and its synonyms, were used in combination with “Artificial Intelligence” or “Machine Learning” or their synonyms. The descriptors were adapted for use in all databases, with only the search key being specific for better clarity of results and easy reproducibility. The complete search strategy used in the databases is available in Supplementary Material 2.
After searching all databases, studies were analyzed, and duplicate studies were excluded. After excluding duplicate studies, two independent authors analyzed the studies by reading titles and abstracts, excluding those not meeting the previously defined selection criteria. Any disagreement between authors was discussed and resolved with the entire group of authors.
Two independent evaluators read the selected articles in full. The information of interest was extracted using a data extraction form focused on the following characteristics: author, year, country, journal publication area, chronic pain type, pain measurement, age of participants, care type, study type, sample size, study objective, tool target, AI type, AI objective, data sources, and data types.
RESULTS
The search for articles in the nine databases resulted in a collection of 2,767 retrieved articles. Five hundred seventeen were duplicated in different databases and excluded. Of the remaining 2,250 articles, 2,087 were excluded after reading the abstracts because they did not address the topic. One hundred sixty-four articles were selected for full reading and eligibility analysis per the previously stipulated exclusion and inclusion criteria.
At this screening stage, 75 articles were excluded for the following reasons: inadequate population (n=31), inadequate outcome (n=23), inadequate intervention (n=15), and inadequate study type (n=6). Thus, 89 articles were selected after passing the eligibility screening in the full reading. Then, 26 additional articles were retrieved by manual citation search, six of which were excluded in eligibility due to inadequate outcome analysis, leaving 20 additional articles. Therefore, 109 articles met the stipulated eligibility criteria and were included in this scoping review (Figure 1).
FIGURE 1
The studies selected in this scoping review were quite comprehensive per the pre-established criteria (sample, use of artificial intelligence technique at any time during rehabilitation, analysis of any outcome without restrictions based on ethnicity, gender, age, and religion). The selected studies were published from 2002 to March 2023 (search date). The countries with the most significant number of publications in the field were the United States, with 44 studies; England, with 16 studies; and Canada, with 10. Only two studies were conducted by Brazilian researchers. The gradual growth in the number of publications in the field is visible, especially since 2018 (Figure 2).
FIGURE 2
The use of AI-based tools had different objectives, with significant variation regarding the level of healthcare intended. Most studies did not specify whether the tool would be helpful in primary, secondary, or tertiary care (n=60). Most studies reporting where the AI-based tool would be intended focused on tertiary care (n=31), followed by secondary care (n=12). Only five studies indicated PHC as the destination for AI-based tools. Furthermore, one study was developed for multiple destinations (n=1) (Supplementary Material 3).
Four of the five studies aimed at PHC were longitudinal (3 observational and 1 Randomized Controlled Study), and one was cross-sectional. Fodeh et al.(29) (2017) focused on analyzing patterns in unstructured assessment data from the medical records of patients with chronic pain, seeking to understand trends in information such as medical record notes and patient reports and offer a broader and more integrated view of the patient’s health status.
Nijeweme-d’Hollosy et al.(28) (2018) aimed to explore the possibilities of using Machine Learning to support clinical decision-making in individuals with chronic low back pain, involving creating tools that would guide professionals in their therapeutic choice. Tsai et al.(26) (2021) analyzed the performance of predictors related to movement and sleep hours to identify factors that could prevent chronic pain or shape the response to treatment.
The study by Nephew et al.(25) (2022) investigated associations of individual variables with depression and chronic pain to identify relationships between specific factors and health outcomes, facilitating an understanding of how some characteristics of patients and treatments are connected to specific results. Liew et al.(27) (2023) focused on using AI to classify subgroups of individuals with low back pain to find specific treatment strategies, increasing therapeutic efficacy.
Using the current classification of chronic pain suggested by IASP(6), which classifies chronic pain into seven subtypes, we observed that most studies were directed at chronic musculoskeletal pain (n=39). Primary chronic pain was also the target of most studies (n=25). Fewer studies were found focusing on chronic visceral pain (n=9), chronic cancer pain (n=7), chronic headache or orofacial pain (n=4), chronic post-surgical and posttraumatic pain (n=4), and chronic neuropathic pain (n=2). Notably, some publications did not report the chronic pain type for which the study was conducted (n=19) (Figure 03).
FIGURE 3
When analyzing the studies by body topography of the pain they target, it is clear that the body regions differ between the studies. Although most selected studies did not specify the chronic pain location (n=57), a large part of the studies targeted chronic low back pain (n=32), followed by chronic back pain (where the location of the pain in the spine was not specified) (n=7), chronic pelvic pain (n=5), chronic neck pain, and chronic knee pain with the same number of studies (n=3); lastly, chronic headache or orofacial pain (n=2) (Supplementary material 4).
When analyzing the tool’s target population, we highlight the large number of studies focused on the patient; that is, the individual with a chronic condition (n=97) would be the primary user of the tool. A few studies were developed to support health professionals (n=9) and health managers (n=2) in chronic pain management.
For better understanding, the study’s results were systematically organized into two large tables: general and methodological characteristics of the studies (Table 1) and Characteristics of the tools and data used in the research (Table 2).
This scoping review identified and systematized several important aspects of the scientific literature related to AI tools applied to chronic pain. As can be seen in Table 1, most studies were conducted in North American and European countries, followed by Asian countries. South America had only two studies, both conducted by Brazilian researchers. Most journal publications fields focused on medical sciences, technology, and diagnostic imaging.
The most prevalent chronic pains were, respectively, chronic musculoskeletal pain (n=39), chronic primary pain (n=25), chronic visceral pain (n=9), chronic cancer pain (n=7), chronic headache or orofacial pain (n=4), chronic post-surgical or posttraumatic pain (n=4), and, lastly, chronic neuropathic pain (n=2). The most used tools to measure pain were the Visual Analog Scale (n=24), the Numerical Estimation Scale (n=19), and specific questionnaires.
The mean age of the total sample reported was 50 years. Most studies did not report the healthcare level (n=60). Studies emphasizing tertiary care had the most significant publications (n=31), and PHC had the smallest number (n=5). Observational study designs were predominant in the review (n=102), with a small number of experimental studies (n=7). The total sample of data used for analysis was 733,106,571. The most prevalent study objectives were classifying individuals by type of chronic pain and predicting risk groups (Table 1).
Table 1
Besides the general and methodological characteristics, we found that the main targets of these tools are patients, but some studies targeted health professionals, and very few targeted managers. The AI types adopted include supervised, semi-supervised, and unsupervised learning. Supervised learning was the most common in the studies of this review (n=92), notably Support Vector Machine (n=32), Random Forest (n=31), Logistic Regression (n=26), Neural Networks (n=13) and K-means (n=10). AI algorithms are aimed at classifying, predicting, and grouping variables.
The data sources used by AI tools are diverse, including electronic medical records, patient reports, biometric sensors, and monitoring devices. Data from clinical trials and longitudinal studies are also integrated to provide a comprehensive and robust knowledge base. The types of data processed by AI tools range from structured to unstructured. These data can be in text, audio, or image format. Textual data include patients’ verbal descriptions of pain, clinical notes, and questionnaires. Images, such as medical imaging scans or photographs of affected areas, provide complementary visual information. Audio, such as recordings of pain reports, are analyzed to capture nuances in patients’ verbal communication about their pain experience (Table 2).
Table 2
DISCUSSION
This comprehensive scoping review reveals that the volume of scientific production on AI in chronic pain has been growing, especially after 2018. The first selected study was published in 2002, and the last was published in 2023 (search date). This massive increase in recent studies on the subject is possibly due to the popularization of AI tools, greater technical capacity, and computational capacity to process an ever-increasing amount of data(30,31). Also, the high prevalence of chronic pain worldwide and health professionals’ difficulties in improving the management of individuals with these conditions encourage the production of new studies and the incorporation of new technologies, including AI(32). Using AI in different ways, as shown in this review, can be a path to address such a complex and challenging issue.
Contrary to expectations, only five studies focused on activities developed in PHC, a setting that is usually the gateway for patients with these conditions. As indicated in the Results section, the five studies differ regarding their objectives and show methodological differences that hinder quantitative synthesis, suggesting that this is still an incipient area of research. The scarcity of studies focusing on PHC also reveals that although AI tools have great potential to assist in screening populations at risk for chronic pain at the first moment of contact with the health system, we still have a long way to go. Encouraging researchers with AI and experience in PHC seems to be a viable path to increasing publications in the field and, consequently, advancing practical applications of new technologies in PHC.
Chronic musculoskeletal pain was the most prevalent in the studies (n=39), followed by primary chronic pain (n=25). According to the specialized literature, this finding was unexpected because primary chronic pain is the most prevalent worldwide(5). Many studies selected in this review focused on chronic low back pain (n=32) since it is the world’s most prevalent chronic musculoskeletal condition. Thus, we should avoid generalizing the results of these studies to all chronic pains. AI algorithms in the selected studies were almost entirely focused on patient use. This characteristic of the target to be reached by the tools may translate into a perception that managing chronic pain conditions by patients is genuinely an important focus in the care of this field of research.
The 109 studies had different objectives and designs and employed different algorithms. This thematic and methodological diversity indicates that AI in chronic pain is a broad field of study with its thematic underlying areas increasingly explored in the coming years. Another important result is that most AI algorithms focused on classifying individuals and predicting risk groups (n=93). Few studies effectively focused on treating chronic pain (n=5), converging with the findings of reviews already published on the subject (33, 34). This appears to be a significant gap despite the algorithms potentially being of great use in assisting clinical support decisions in treatment and prognosis.
One of the most researched topics is using biomarkers and neuroimaging to understand which variables can predict responses to treatment in patients with chronic pain(35-38). In this field, the analysis of functional brain images, such as Functional Magnetic Resonance Imaging, has been used to identify brain patterns that correlate with the response to pain and treatment with drugs, such as opioid analgesics(39-41).
Another important theme was assessing the performance of the integration of movement sensors and electromyography (EMG) with Machine Learning algorithms in identifying anomalous patterns in patients with chronic pain. Analysis of data captured by sensors can distinguish subtle motor characteristics that may contribute to pain(42, 43). The issue of prolonged use of opioids and the prediction of abuse of these substances is also an area of focus. Studies suggest that predictive algorithms have been effective in identifying patients at risk of prolonged opioid use after surgical procedures or chronic pain (44, 45). This aspect is directly related to the growing effort to mitigate the opioid epidemic, as evidenced by Jones et al. (46) (2018), who argue that predicting inappropriate use behaviors may be key to reducing the impact of this global health crisis.
Most studies selected in this scoping review were observational (n=102). This methodological characteristic reveals that data from individuals were used for some indirect analysis with AI, such as classifying individuals into subgroups, associating variables that may be related in search of risk and prognostic factors, identifying genes involved in chronification processes, creating new means of assessment for individuals with chronic pain, creating models to predict chronic post-surgical pain, predicting the amount of post-surgical opioid, among others.
Few intervention studies and analyses of therapeutic efficacy, such as randomized controlled clinical trials, were found (n=7) and should be further encouraged. On the other hand, one of the strengths of the studies included in this review is that most works (n=56) used pain scales commonly adopted in the literature on the subject to assess pain intensity in individuals, following a pattern already identified in chronic pain studies(49).
Another notable finding was that many of the selected studies used public databases and/or patient self-reported data (n=25), in which data quality may not be ideal. AI requires a large amount of data with known origin and reliable quality to make good predictions(50). In health, structured data is generated in more significant amounts than unstructured data(51), which was reflected in the findings of this review since the number of studies in which structured data was used as a basis for AI analysis (n=89) was much higher than those that used unstructured data.
As indicated in the Results section, supervised algorithms were the most frequently found in the studies that participated in this review (n=92), notably Support Vector Machine (n=32), Random Forest (n=31), Logistic Regression (n=26), Neural Networks (n=13) and K-means (n=10). This characteristic aligns with (structured) data types most accessible in health systems and that much of the research aims to classify individuals. The high representation of supervised algorithms in studies on chronic pain was also highlighted in the first review on the subject published in 2021(33). The scoping review included 53 studies, prioritizing the description of which Machine Learning methods and data types had been used. Unlike this study, the review selected only studies where the origin of chronic pain was unknown, used only three databases, and was limited to the descriptors “machine learning”, “pain”, and “fibromyalgia”, which reduced its scope.
Recently, Zhang et al.(34) (2023) published a scoping review on AI in pain. Unlike our study, the authors selected all pain types. Although the search strategy was more comprehensive and not limited to chronic pain, the researchers selected only intervention studies. From the 30 articles selected for the review, the authors found that most studies focused on pain assessment, followed by AI-based approaches related to pain prediction and clinical decision support, and a minority of studies on AI-based approaches related to pain self-management.
Despite the great potential of using AI in PHC to better manage chronic pain, the automatic use of AI can lead to biased and misleading interpretations of a data set (algorithmic bias). In PHC, where diversity is the norm, algorithms should be developed, trained, and adjusted in a way that avoids discrimination that can lead to unequal management between different economic and social groups in pain treatment. Given health inequalities and inequities, we can postulate that vulnerable population subgroups may have less access to these new technologies. Ensuring universal access to the potential benefits of AI is a huge challenge for governments in the coming years.
Notably, using AI depends on sensitive patient data, usually stored in the Unified Health System’s information system. Thus, it is necessary to ensure that the data will be protected against misuse. The General Personal Data Protection Law (LGPD), Law N° 13.709/2018, was created in Brazil in 2018 to guarantee the ethical use of personal data. This Brazilian legislation regulates the processing of personal online and offline data to protect citizens’ privacy and fundamental rights(52). Inspired by the European Union’s General Data Protection Regulation (GDPR), the LGPD establishes rules for collecting, storing, using, sharing and deleting personal data. This initiative aims to ensure transparency in the use of personal data, protecting the privacy rights of individuals and promoting responsible and ethical practices by companies in processing information(53).
Another point to be discussed and that deserves attention is how much the introduction of this new technology may impact the relationship between health professionals and patients. If technology replaces rather than complements professionals, it may lead to a “dehumanized” health system. For AI to be an ally and not a problem, government decisions must regulate its use so that it benefits everyone without compromising fundamental healthcare values.
The research results should be interpreted in light of its strengths and limitations. In terms of strengths, we should underscore that this review was comprehensive, without time or language filter, in nine databases (MEDLINE; LILACS; SPORTDiscus; SCOPUS; CENTRAL; Science Direct; IEEE Xplore; Association for Computing Machinery Digital Library, and arXiv) from different areas. Two independent authors conducted the search, and a third author was requested for disagreements. Additional studies that were not found in the initial search but included in references of studies found were also included. Another strength was that the structure of this scoping review followed the PRISMA-ScR recommendations, making its methodology more robust and facilitating comparison with international studies.
On the other hand, we must be cautious in interpreting the findings since this scoping review was designed to be a comprehensive search of the scientific literature on the topic and not evaluate the quality of the studies found or to make a quantitative synthesis of the results obtained in the original studies. The lack of methodological detail and a formal assessment of the articles’ quality are limitations of this study. Furthermore, while comprehensive, the literature search is challenging in a new field such as Artificial Intelligence since the nomenclatures are still being consolidated to become keywords and scientific descriptors, and the academic production covers several interconnected areas. Encouraging international standardization in publications in the field can facilitate future reviews.
Regarding this study’s implications, research focusing on clinical interventions should be the focus of new AI studies in chronic pain. Currently, we have a range of observational studies focusing mainly on the prediction and classification of subgroups. The next step is to direct new research to use these predictors in randomized controlled clinical trials to study the effectiveness of the entire process. This will be the question that future chronic pain studies will answer: Does the group using the AI-based strategy have better outcomes than the control group using the current best strategy (without AI)?
The potential of using Artificial Intelligence (AI) in managing chronic pain is promising. AI algorithms have shown great capacity to classify individuals based on chronic pain type and predict risk groups. These technologies offer new possibilities for a more accurate and individualized diagnosis and the development of personalized treatment strategies, which can increase therapeutic efficacy and the quality of life of patients with chronic pain. However, AI implementation in this field also faces significant challenges, mainly the lack of studies focused on PHC. It is necessary to develop robust studies to validate these technologies in different populations and care settings, ensuring that AI solutions are accessible, effective, and safe for all patients.
Finally, the study found results that indicate progress in the field, especially in diagnosing and predicting pain, analyzing brain activity, personalized interventions, and mobility and sensor technology. However, only two studies were randomized controlled clinical trials(47,48) aimed at estimating the effect of an intervention (Cognitive Behavioral Therapy), indicating that it is necessary to encourage research with this focus so that the question can be answered whether the incorporation of AI modifies the most important outcomes related to chronic pain.
CONCLUSION
Our study’s results indicate that the potential of using AI in managing chronic pain is a reality. It is a fertile area for knowledge generation, especially in recent years. Considering the diverse AI applications in different scenarios and populations and the heterogeneity of its uses, we understand that there are several possibilities for improving the management of individuals with chronic pain, whose treatment is currently challenging.
Systematic reviews with specific questions and formal analyses of the quality of the included studies should be encouraged. There is still a significant lack of studies focused on PHC, paradoxically the place with the most significant potential for applying Artificial Intelligence algorithms and the main gateway to health systems, including the SUS, for patients with this condition.
Therefore, research focusing on the use of cutting-edge technology to aid in screening, clinical decision-making, and management of individuals with chronic pain, especially in PHC, should be encouraged. Finally, we emphasize that our study can be considered the most comprehensive ever conducted in the field of chronic pain on Artificial Intelligence, serving as a reference for new research with this focus.

Figures, tables and supplements - Data Repository Scielo Data: https://doi.org/10.48331/scielodata.QFMMDX

REFERENCES
1. Cohen SP, Vase L, Hooten WM. Chronic pain: an update on burden, best practices, and new advances. The Lancet. 2021;397(10289):2082-97.
2. Collaborators G. Global, regional, and national incidence, prevalence, and years lived with disability for 354 diseases and injuries for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. 2018.
3. Vos T, Abajobir AA, Abate KH, Abbafati C, Abbas KM, Abd-Allah F, et al. Global, regional, and national incidence, prevalence, and years lived with disability for 328 diseases and injuries for 195 countries, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. The Lancet. 2017;390(10100):1211-59.
4. Raja S, Carr D, Cohen M, Finnerup N, Flor H, Gibson S, et al. The revised International Association for the Study of Pain definition of pain: concepts, challenges, and compromises. Pain. press doi. 2020;10.
5. Nicholas M, Vlaeyen JW, Rief W, Barke A, Aziz Q, Benoliel R, et al. The IASP classification of chronic pain for ICD-11: chronic primary pain. Pain. 2019;160(1):28-37.
6. Treede R-D, Rief W, Barke A, Aziz Q, Bennett MI, Benoliel R, et al. Chronic pain as a symptom or a disease: the IASP Classification of Chronic Pain for the International Classification of Diseases (ICD-11). Pain. 2019;160(1):19-27.
7. Bennett MI, Kaasa S, Barke A, Korwisi B, Rief W, Treede R-D. The IASP classification of chronic pain for ICD-11: chronic cancer-related pain. Pain. 2019;160(1):38-44.
8. Schug SA, Lavand\'homme P, Barke A, Korwisi B, Rief W, Treede R-D. The IASP classification of chronic pain for ICD-11: chronic postsurgical or posttraumatic pain. Pain. 2019;160(1):45-52.
9. Scholz J, Finnerup NB, Attal N, Aziz Q, Baron R, Bennett MI, et al. The IASP classification of chronic pain for ICD-11: chronic neuropathic pain. Pain. 2019;160(1):53-9.
10. Benoliel R, Svensson P, Evers S, Wang S-J, Barke A, Korwisi B, et al. The IASP classification of chronic pain for ICD-11: chronic secondary headache or orofacial pain. Pain. 2019;160(1):60-8.
11. Aziz Q, Giamberardino MA, Barke A, Korwisi B, Baranowski AP, Wesselmann U, et al. The IASP classification of chronic pain for ICD-11: chronic secondary visceral pain. Pain. 2019;160(1):69-76.
12. Perrot S, Cohen M, Barke A, Korwisi B, Rief W, Treede R-D. The IASP classification of chronic pain for ICD-11: chronic secondary musculoskeletal pain. Pain. 2019;160(1):77-82.
13. Barke A, Korwisi B, Jakob R, Konstanjsek N, Rief W, Treede R-D. Classification of chronic pain for the International Classification of Diseases (ICD-11): results of the 2017 international World Health Organization field testing. Pain. 2022;163(2):e310-e8.
14. Joypaul S, Kelly F, McMillan SS, King MA. Multi-disciplinary interventions for chronic pain involving education: A systematic review. PloS one. 2019;14(10):e0223306.
15. Grossi E. How artificial intelligence tools can be used to assess individual patient risk in cardiovascular disease: problems with the current methods. BMC cardiovascular Disorders. 2006;6(1):1-6.
16. Kueper JK, Terry AL, Zwarenstein M, Lizotte DJ. Artificial intelligence and primary care research: a scoping review. The annals of family medicine. 2020;18(3):250-8.
17. Gunning D, Stefik M, Choi J, Miller T, Stumpf S, Yang G-Z. XAI—Explainable artificial intelligence. Science robotics. 2019;4(37):eaay7120.
18. Piette JD, Newman S, Krein SL, Marinec N, Chen J, Williams DA, et al. Artificial Intelligence (AI) to improve chronic pain care: Evidence of AI learning. Intelligence-Based Medicine. 2022:100064.
19. Mendes EV. As redes de atenção à saúde. Ciência & saúde coletiva. 2010;15:2297-305.
20. Mendes EV. A construção social da atenção primária à saúde. Brasília: Conselho Nacional de Secretários de Saúde. 2015;45.
21. Almeida ERd, Sousa ANAd, Brandão CC, Carvalho FFBd, Tavares G, Silva KC. Política Nacional de Atenção Básica no Brasil: uma análise do processo de revisão (2015–2017). Revista Panamericana de Salud Pública. 2018;42:e180.
22. McGowan J, Straus S, Moher D, Langlois EV, O\'Brien KK, Horsley T, et al. Reporting scoping reviews—PRISMA ScR extension. Journal of clinical epidemiology. 2020;123:177-9.
23. Tricco AC, Lillie E, Zarin W, O\'Brien KK, Colquhoun H, Levac D, et al. PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation. Annals of internal medicine. 2018;169(7):467-73.
24. Aromataris E, Munn Z. JBI manual for evidence synthesis. JBI: Adelaide, Australia. 2020.
25. Nephew BC, Incollingo Rodriguez AC, Melican V, Polcari JJ, Nippert KE, Rashkovskii M, et al. Depression predicts chronic pain interference in racially diverse, income-disadvantaged patients. Pain Medicine. 2022;23(7):1239-48.
26. Tsai P-F, Wang C-H, Zhou Y, Ren J, Jones A, Watts SO, et al. A classification algorithm to predict chronic pain using both regression and machine learning–A stepwise approach. Applied Nursing Research. 2021;62:151504.
27. Liew BX, Hartvigsen J, Scutari M, Kongsted A. Data-driven network analysis identified subgroup-specific low back pain pathways: a cross-sectional GLA: D Back study. Journal of Clinical Epidemiology. 2023;153:66-77.
28. Nijeweme-d’Hollosy WO, van Velsen L, Poel M, Groothuis-Oudshoorn CG, Soer R, Hermens H. Evaluation of three machine learning models for self-referral decision support on low back pain in primary care. International journal of medical informatics. 2018;110:31-41.
29. Fodeh SJ, Finch D, Bouayad L, Luther S, Kerns RD, Brandt C. Classifying Clinical Notes with Pain Assessment. Studies in Health Technology and Informatics. 2017;245:1261-.
30. Haug CJ, Drazen JM. Artificial intelligence and machine learning in clinical medicine, 2023. New England Journal of Medicine. 2023;388(13):1201-8.
31. Sarker M. Revolutionizing healthcare: the role of machine learning in the health sector. Journal of Artificial Intelligence General science (JAIGS) ISSN: 3006-4023. 2024;2(1):36-61.
32. Hasan F, Mudey A, Joshi A. Role of Internet of Things (IoT), Artificial Intelligence and Machine Learning in Musculoskeletal Pain: A Scoping Review. Cureus. 2023;15(4).
33. Jenssen MDK, Bakkevoll PA, Ngo PD, Budrionis A, Fagerlund AJ, Tayefi M, et al. Machine learning in chronic pain research: a scoping review. Applied Sciences. 2021;11(7):3205.
34. Zhang M, Zhu L, Lin S-Y, Herr K, Chi C-L, Demir I, et al. Using artificial intelligence to improve pain assessment and pain management: a scoping review. Journal of the American Medical Informatics Association. 2023;30(3):570-87.
35. Ichesco E, Peltier SJ, Mawla I, Harper DE, Pauer L, Harte SE, et al. Prediction of differential pharmacologic response in chronic pain using functional neuroimaging biomarkers and a support vector machine algorithm: an exploratory study. Arthritis & Rheumatology. 2021;73(11):2127-37.
36. Lee J, Mawla I, Kim J, Loggia ML, Ortiz A, Jung C, et al. Machine learning-based prediction of clinical pain using multimodal neuroimaging and autonomic metrics. pain. 2019;160(3):550.
37. Brown CA, Almarzouki AF, Brown RJ, Jones AK. Neural representations of aversive value encoding in pain catastrophizers. Neuroimage. 2019;184:508-19.
38. Robinson ME, O\'Shea AM, Craggs JG, Price DD, Letzen JE, Staud R. Comparison of machine classification algorithms for fibromyalgia: neuroimages versus self-report. The Journal of Pain. 2015;16(5):472-7.
39. Santana AN, Cifre I, De Santana CN, Montoya P. Using deep learning and resting-state fMRI to classify chronic pain conditions. Frontiers in neuroscience. 2019;13:1313.
40. Rogachov A, Cheng JC, Hemington KS, Bosma RL, Kim JA, Osborne NR, et al. Abnormal low-frequency oscillations reflect trait-like pain ratings in chronic pain patients revealed through a machine learning approach. Journal of Neuroscience. 2018;38(33):7293-302.
41. Tu Y, Ortiz A, Gollub RL, Cao J, Gerber J, Lang C, et al. Multivariate resting-state functional connectivity predicts responses to real and sham acupuncture treatment in chronic low back pain. NeuroImage: Clinical. 2019;23:101885.
42. Lin Y-C, Yu N-Y, Jiang C-F, Chang S-H. Characterizing the SEMG patterns with myofascial pain using a multi-scale wavelet model through machine learning approaches. Journal of Electromyography and Kinesiology. 2018;41:147-53.
43. Golabchi FN, Sapienza S, Severini G, Reaston P, Tomecek F, Demarchi D, et al. Assessing aberrant muscle activity patterns via the analysis of surface EMG data collected during a functional evaluation. BMC musculoskeletal disorders. 2019;20:1-15.
44. Lin H-C, Wang Z, Hu Y-H, Simon K, Buu A. Characteristics of statewide prescription drug monitoring programs and potentially inappropriate opioid prescribing to patients with non-cancer chronic pain: A machine learning application. Preventive Medicine. 2022;161:107116.
45. Klemt C, Harvey MJ, Robinson MG, Esposito JG, Yeo I, Kwon YM. Machine learning algorithms predict extended postoperative opioid use in primary total knee arthroplasty. Knee Surgery, Sports Traumatology, Arthroscopy. 2022;30(8):2573-81.
46. Jones MR, Viswanath O, Peck J, Kaye AD, Gill JS, Simopoulos TT. A brief history of the opioid epidemic and strategies for pain medicine. Pain and therapy. 2018;7:13-21.
47. Piette JD, Newman S, Krein SL, Marinec N, Chen J, Williams DA, et al. Patient-centered pain care using artificial intelligence and mobile health tools: a randomized comparative effectiveness trial. JAMA Internal Medicine. 2022;182(9):975-83.
48. Piette JD, Newman S, Krein SL, Marinec N, Chen J, Williams DA, et al. Artificial Intelligence (AI) to improve chronic pain care: Evidence of AI learning. Intelligence-Based Medicine. 2022;6:100064.
49. Åström M, Thet Lwin ZM, Teni FS, Burström K, Berg J. Use of the visual analogue scale for health state valuation: A scoping review. Quality of Life Research. 2023;32(10):2719-29.
50. Haenlein M, Kaplan A. A brief history of artificial intelligence: On the past, present, and future of artificial intelligence. California management review. 2019;61(4):5-14.
51. de Moraes Batista AF, Chiavegatto Filho ADP. Machine Learning aplicado à Saúde. Sociedade Brasileira de Computação. 2019.
52. Frazão A, Oliva MD, TEPEDINO G. Lei Geral de Proteção de Dados Pessoais e suas repercussões no direito brasileiro: Thomson Reuters Brasil; 2019.
53. Rapôso CFL, de Lima HM, de Oliveira Junior WF, Silva PAF, de Souza Barros EE. Lgpd-lei geral de proteção de dados pessoais em tecnologia da informação: Revisão sistemática. RACE-Revista de Administração do Cesmac. 2019;4:58-67.