Using graph machine learning to identify functioning in patients with low back pain within the ICF framework

Using graph machine learning to identify functioning in patients with low back pain within the ICF framework | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Using graph machine learning to identify functioning in patients with low back pain within the ICF framework Linda Nieminen, Harri Ketamo, Jari Vuori, Markku Kankaanpää This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5415974/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Jul, 2025 Read the published version in Scientific Reports → Version 1 posted 8 You are reading this latest preprint version Abstract As a comprehensive perspective on functioning is useful when making patient assessments, the WHO has developed the International Classification for Functioning, Disability, and Health (ICF). However, its complex structure poses a problem for implementation as part of clinical practice.The aim of this study was to test a graph machine learning engine, Headai Graphmind, to recognize ICF codes from electronic health records written in Finnish. A dataset of 93 patients aged 18 to 65 years with chronic low back pain was collected. Headai Graphmind was then tested for its ability to match free text with ICF codes on a sample of 20 patients. The results were compared against the findings of a domain expert. Headai Graphmind achieved 0.95 precision, 0.83 recall, and 0.89 F1 score.The application found 112 distinct ICF codes compared to 119 codes found by the domain expert. Headai Graphmind has the capability to recognize ICF codes from the electronic health records of patients with chronic low back pain. The method could be helpful when implementing the ICF classification in clinical practice, and enable retrospective coding of medical information for further use. Health sciences/Health care Physical sciences/Mathematics and computing artificial intelligence functioning graph machine learning ICF low back pain Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 1 Introduction How can we comprehensively treat a patient who has experienced prolonged pain so that the onset of pain chronicity becomes preventable? What are the obstacles that prevent this patient with multiple morbidities from returning to work? The answers to a holistic perspective on functioning and health can usually be found in the health records of individual patients or in group-specific cohorts 1 . Finding answers to these questions is, however, challenging and cannot be achieved by evaluating diagnosis codes 2 alone. For example, two patients with the same low back pain diagnosis can have completely different levels of functioning, with the first patient coping at work, whereas the second has major difficulties coping and a low quality of life 3 . To enhance the understanding of health and health-related states holistically, the World Health Organization (WHO) has developed the International Classification of Functioning, Disability, and Health (ICF) framework 4 . When used as part of functioning assessment, the ICF can create not only a broader picture of health, but can also reflect patients’ self-reported problems more closely than a typical medical assessment 5 . Previously, different approaches have been used to integrate the ICF with electronic health records (EHR) 1 , predominantly in rehabilitation settings, and in some cases where the ICF was embedded in health information systems 6 . However, the optimal implementation of the ICF, including structured documentation conducted by health care professionals, has been slow. This delay has been mainly due to the complex structure of the ICF taxonomy that comprises approximately 1600 codes representing different domains of functioning 1 , 7 , 8 . There is a need, therefore, for efficient and reliable methods to enhance the wider implementation of the ICF as part of the clinical workflow in health care settings. Artificial intelligence (AI) techniques show promise in the implementation of ICF taxonomy as an integral part of medical assessment, since such techniques have the capability to solve complex tasks. Likewise, AI techniques could also serve to promote a broader understanding of those factors concerning the disability of patients 9 , which could, in turn, help to plan more high-quality personalized care with meaningful goals. It should be noted, however, that the selection of a suitable AI or machine learning (ML) algorithm is a complicated process. For example, deep learning algorithms perform well in categorizing tasks when the task is well defined, and the training material is large enough. In contrast, deep learning algorithms do not perform well in cases where the task is ill defined and requires human reasoning when working with unknown factors 10 . In a previous study, Newman-Griffis et al. selected a natural language processing (NLP) technique for linking free text to 29 ICF-based categories 11 . Recently, a semantic network-based (graph) ML engine, Headai Graphmind, which has the capability to imitate human reading and processing of texts, has been applied to forecast future skill needs in the labor market and to optimize learning paths for the future workforce using curriculum gap assessment 12 , 13 . In previous studies, graph-based machine learning or semantic network-based setups have shown promise in different complex settings 14 – 16 . In practice, Headai Graphmind adds, modifies, and reasons natural language according to conceptual learning theories 17 , whereas a semantic network serves as a structure for all the data learned. Furthermore, in contexts where formal procedural rules for matching are unknown, technology that can operate with unstructured data has the potential to fully utilize EHR information and harness data for further analysis conducted by health care professionals 16 . A good example of the applicability of Headai Graphmind is in the recognition of patients at risk for low back pain (LBP). Although vast amounts of research and resources have been devoted to LBP during the last decades 18 , the burden of LBP has increased, making it the most burdensome global health problem affecting years lived with disability (YLD) 19 . Although not all patients with LBP develop a chronic pain problem 20 , the increased use of health care resources and the subsequent rise in costs related to LBP are mainly driven by chronic cases 21 . Therefore, strategies that ensure the early identification of those patients at risk for persistence of pain and disability should be developed and implemented 22 . At present, questionnaires such as the STarT Back Screening Tool (SBT) 23 are used to identify these patients, but they are not comprehensive enough to recognize all the relevant factors (23,24) required to ensure a holistic decision-making process. The findings of a preliminary study 16 suggest that Headai Graphmind could have the potential to make the timing and tailoring of interventions in the LBP patient population easier. In this feasibility study, the EHR of patients with chronic LBP were used to generate natural language definitions of the ICF entities. The main aim of the study is to test the validity, especially criterion validity, of Headai Graphmind applied to determine semantically the best matches between ICF code definitions and the natural language of the EHR. The reliability of Headai Graphmind is tested against the findings of a domain expert who has profound understanding of the EHR in question, the functioning of LBP patients, and the ICF. Our research questions were as follows: (1) What meaningful items of functioning can be found from the electronic health records of patients? (2) Can an AI method perform meaningful item recognition reliably enough to support the health care decision-making process when compared to a domain expert? 2 Methods 2.1 Data architecture and data processing The generic idea behind Headai Graphmind is that it has a pre-trained semantic understanding of language. This understanding is based on gigabytes of generic data collected from research papers (open journals), policy papers (such as the European Union’s archive), labor market information (e.g., job ads), and professional news (technology, business, medical, and so forth). As the data can affect the semantic behavior of the engine, the training data have been selected from sources that are widely used and trusted. Further, when reading this training data, Headai Graphmind learns which words are meaningful, when the words form a compound word (also called n-grams in linguistics), what is the relationship between the words and the n-grams, and eventually the context of the words and n-grams used. This semantic network is applied when starting to perform reasoning between ICF entity descriptions and real-world texts from the EHR. Headai Graphmind turns the EHR texts into a similar type of semantic network as the pre-trained language model (Figure 1). After that, it starts to fit the ICF descriptions to the EHR semantic network and applies the pre-trained language model to understand the small data, such as synonyms, neighboring concepts, and similar meanings, more precisely. This two-layer approach enables matching between two small datasets without getting stacked into a lack of structured data or non-matching words. The data architecture (Figure 2) from the EHR to Headai Graphmind to the research environment was built to meet high data security and privacy requirements. All the data from the EHR were provided as an encrypted pseudonymized csv.-file on a memory stick that was extracted offline on Headai Graphmind’s side. In a study by Nieminen and colleagues 16 , 12 different ICF definition sets were studied to find the most functional setups. The best performing definition sets (called ICF title, ICF real life, ICF real life fuzzy, MesH words, MeSH-ICF title) were also imported to Headai Graphmind before the analysis. However, because these datasets were not GDPR (General Data Protection Regulation) -data, they were imported directly from the csv.-file without encryption. The matching between the EHR data and the ICF definition sets were run in offline mode, and the results with pseudo-identifier were copied into the research environment with authenticated network access. The offline nature of the Headai Graphmind process means that all data and computing are performed outside the network-accessible directories of the Headai Graphmind computing infrastructure, and only Headai Graphmind itself is connected to the network and only few directories are accessible via the network. The Headai Graphmind semantic matching process is described in more detail in Figure 3. The written texts in the EHR are then turned into semantic networks. Each patient’s semantic network is analyzed against each of the more than 1600 ICF definitions in the chosen definition sets (as mentioned above), resulting in approximately 36 000 analyses per patient (MeSH alone contains 30 000 entries). The semantic matching of Headai Graphmind is based on shallow neural networks, and the analysis took between 2 and 10 seconds using one CPU core per patient, depending on data volume and data complexity. In this study, the stored semantic network of the patient (JSON_GRAPH in Figure 3) was only used to support the domain expert’s evaluation. The performance of the algorithm were evaluated with precision, recall, and F1 score. 2.2 Population data We used patient data gathered between October 2019 and February 2021. The natural language data were free text (Finnish language) consisting of the EHR of patients with chronic 25 (duration 3 months or longer) LBP who attended the Department of Physical Rehabilitation and Medicine at Tampere University Hospital, Finland. Additional information was collected in the form of quantitative data, which were retrieved from medical history forms. The quantitative data were used for the data selection process (the fulfillment of the inclusion and exclusion criteria, Table 1). The data were collected retrospectively after the treatment period had ended and patients had returned to primary or occupational care. The search for suitable patients began with medical history forms that were collected from all patients attending the department. Since the study was registry-based and the integrity of the patients was maintained, the Ethics Committee of Tampere University Hospital waived the need for ethical approval and informed patient consent. In addition, Finnish research legislation on the secondary use of health and social data (legislation no. 552/2019) 26 allows the retrieval of retrospective data from the EHR for research purposes without informed consent from the patients. Since the data in the present study were gathered from a single data source, a data transfer contract was drawn up between the investigators and the data controller, Tampere University Hospital, Finland. The study was carried out in accordance with all relevant guidelines and regulations. LBP was defined as pain in the anatomical region between the costal margins and the inferior gluteal folds with or without radicular pain 27 . Multifocal pain was not an exclusion criterion. Inclusion and exclusion criteria are presented in Table 1. The data was processed using a deductive content analysis approach, using both qualitative and quantitative methods. The EHR datasets of 20 patients were randomly selected to form a training set. A satisfactory saturation of the free text was achieved in the training set, since the last 5 patient datasets added did not significantly increase the number of ICF codes. The free text annotation, being the linking of the ICF to the EHR, applied the principals of proposed ICF linking rules 28 . A domain expert searched the EHR texts of the training dataset for suitable words and n-grams that represented the contents of the ICF. These words and n-grams were further annotated with ICF codes using third and fourth level codes wherever possible to produce as specific examples as possible. In addition, the developed vocabulary was enriched by the domain experts’ perspective on the language often used by professionals (“jargon”) in case some words or n-grams were missing from the training dataset. A further 20 randomly selected EHR datasets were used for the quality analysis of Headai Graphmind. The same domain expert who assembled the training data also evaluated the Headai Graphmind results. Both samples of the study population were obtained using computer-based randomization. Quantitative content analysis was used as a method in the following manner: The search of the contents and further annotation with the ICF codes repeated the method used to compile the training data. Thereafter, the disability information gathered from the free texts was synthesized quantitatively to gain an understanding of the ICF categories found (Table 3). The results were then analyzed by the domain expert between the original EHR data and the end results of the Headai Graphmind in the research environment. Only the domain expert who had access to both systems, participated in this part of the study. The analysis dataset was compressed on an individual document level so that recurring codes were condensed into a single finding. The results were interpreted by the domain expert in the following manner. The finding was defined as true positive if the algorithm found the same code as the domain expert from the free text of one patient. False positives were the codes that the domain expert did not find and, after reappraisal, were still regarded as false findings. Codes were defined as false negatives if a code was found by the domain expert but not by the algorithm. Additionally, there were codes that were first found by the algorithm and, after reappraisal, found by the domain expert as well. The population data were analyzed with IBM SPSS Statistics for Windows, Version 28.0, for mean and 95% confidence intervals. 3 Results 3.1 Study population The flow diagram (Figure 4) presents the selection of the patient sample. The EHR of 93 patients with non-specific LBP were collected for the purposes of this and preliminary research. The first round of exclusion was based on medical history forms. Age, main complaint, inadequate information (for example, missing information about duration of symptoms), Visual Analog Scale (VAS), and duration of symptoms were all considered. In total, 335 working-aged patients, whose main complaint was back pain, were identified from the medical history forms. The fulfillment of the inclusion criteria was finally verified from the EHR. In total, the EHR of the selected population (n=93) comprised 312 physicians’ notes, including texts of referrals, physical appointments, and records of contacts by phone and letter. The characteristics of the patient population are presented in Table 2. Table 3 presents the quantitative findings of the domain expert from the evaluation dataset (EHR from 20 patients, 63 EHR notes). The mention of body structures was the highest (n=1,444). The notes also contained versatile information on other domains related to disability, such as neuromusculoskeletal and movement- related functions (n=349), information on joint and bone mobility and muscle endurance, information on products and technology (n=253), information on medication, and the mobility aids the patients used. Additionally, mental functions (n=99) contained information on sleep quality, mental and personality disorders, emotions, and mental energy levels. Social factors were available from community, social, and civic life (n=103) or major life areas (n=89) where work-related factors are described. An extract from a health record note is presented in figure 5 to give an example of the included ICF -related information and annotation. 3.2 Headai Graphmind Headai Graphmind was analyzed for its semantic matching abilities of factor recognition in the four domains of the ICF: body structures (S codes), body functions (B codes), activities and participation (D codes), and environmental factors (E codes) (Table 4). Headai Graphmind performed the matching on the data of the whole study population (93 patients, 312 EHR notes). To obtain an estimate of Headai Graphmind’s reliability, an evaluation of the factor recognition was performed on a random sample of the EHR of 20 patients (63 EHR notes). The sample size was chosen due to the exhaustive nature of the evaluation process. The algorithm reached precision 0.95 (9 5 % CI 0.939-0.961) and recall 0.83 (95% CI 0.823-0.837) , from which F1 score was 0.89 (95% CI 0.884- 0.896) (Table 4). Recall was highest in both the environmental factor and body structure domains (0.85) and lowest in the activity and participation domain (0.78), whereas precision was highest in the body structure domain (0.99) and lowest in the body function domain (0.91). Furthermore, when comparing the content of the codes, the domain expert found 119 distinct codes (30 S codes, 35 B codes, 40 D codes, and 14 E codes) from the evaluation dataset, whereas the algorithm found 112 codes (30 S codes, 35 B codes, 35 D codes, and 12 E codes). The missed codes were d4103 (sitting), d4302 (carrying in the arms), d630 (preparing meals), d6402 (cleaning living area), d825 (vocational training), e1200 (general products and technology for personal indoor and outdoor mobility and transportation), and e1151(assistive products and technology for personal use in daily living). 4 Discussion The main finding of this feasibility study was that Headai Graphmind performed the factor recognition of ICF information from the EHR of patients with LBP with convincing performance when compared to the results of the domain expert (Table 4 ). Regarding our first study question, the EHR notes of individuals with chronic LBP were expressive and contained holistic information about the disability of individuals (Table 3 ). The population data also reflected closely to the WHO ICF core set for LBP 29 . Compared to a previous study on the subject 11 , a wider selection of ICF-based categories was obtained. Gaining this information from the patient population makes it possible to holistically support the decision-making process in the treatment and rehabilitation assessment of patients. Furthermore, these results offer the promise of a new functional application for personalized medicine, where an individualized model of a patient’s history can be used to take preventative actions 30 . The ICF framework perspective identified from the EHR can broaden our understanding of the functioning-related factors affected by the patients’ medical condition. The universal, interdisciplinary language of the ICF can be applied globally in different health care settings and by different health care professions to produce a broader biopsychosocial understanding of health 1 . The biopsychosocial model has been developed to shift the focus from a narrow biomedical model to understanding concepts of health and functioning through biomedical, psychological, and social dimensions 31 . Additionally, the ICF considers environmental and individual factors that could be more easily accessible in future using the AI application tested here. Since Headai Graphmind is based on shallow neural networks, the speed (and resulting low energy consumption) enables large scale analysis in, e.g., monthly analysis. When considering the speed of Headai Graphmind and the complexity of health data, it can be estimated that it would take approximately 15 days to analyze 1 000 000 patient records with 8 core computing setups, which can be regarded as a very low requirement at present. With respect to data safety issues and the embedding of AI architecture to current computing architectures 32 , Headai Graphmind can work as a plug-in and does not require any software integration. The actual tool for decision-making support needs to fit the purposes of end-users and thus, the visual display needs to be designed by professionals. When using AI-derived information for decision-making support, it is important to make sure that noise in the health care data does not drive the decision-making process. Consequently, it is up to the end-user to make sure that the information does not lead to unintended effects, such as discrimination, increased inequities, and decreased inclusion 33 . Several future implications emerged during our study. As part of the study, only one semantic network was performed per patient to study the quality of the process. In future, however, yearly networks can be performed per patient to enable time series analysis. Additionally, we can examine population cohorts to obtain a wider understanding of the functioning and disability of citizens. In future, new studies on the validity and reliability of the developed application must be conducted with texts unrelated to chronic pain. The core architecture of Headai Graphmind as a semantic computing platform is designed to be language agnostic. Therefore, it is used in settings other than health care for tens of real-world customer cases in English, Spanish, French, German, Swedish, Vietnamese, Estonian, Ukrainian, and Finnish. Although different use cases require independent validation, the design of the technology does enable faster development in new languages and cultural environments. Unfortunately, data silos pose a problem for efficient data processing in many health care ecosystems. In Finland, there is a centralized archive of electronic patient data, making data standardization possible 34 . There are also initiatives towards unified health care records in the Nordic countries 35 as well as in the European Union 36 , 37 . The European Health Data Space (EHDS) is one of activities of the European Data Spaces initiative, focused on building strong governance and basic functions to ensure fair data sharing. The method tested in this research could serve as one of the building blocks within the EHDS. 4.1 Limitations The data used in this present study had some limitations. The data consisted of only the medical notes of physicians. Therefore, the results of the present study can only be generalized to physicians’ notes and patients with low back pain. Further analysis of construct and content validity and reliability with new data will be needed to study the applicability of Headai Graphmind with other health care professional and patient groups. The data itself had limitations in terms of providing a biopsychosocial view of patients with chronic LBP. The majority (65%) of ICF-related information (Table 3 ) was related to body structures and functions. The documents provide a good insight into everyday healthcare, where the biomedical model is dominant. If an algorithm, such as presented in this study, were to visualize the ICF-related information for health professionals, it could highlight the need to shift the focus to activities, participation and environmental factors in order to truly support the patients’ wellbeing. The presence of only one domain expert doing the annotation and analysis can be regarded as a major weakness in this study. On the one hand, the annotation and analysis proceeded in a homogenous way, but on the other multiple experts would have brought different interpretations of the text and the results. This would have strengthened the study, especially for future applicability. In the future validation process, multiple experts will be used, and inter-rater reliability will be tested, to ensure the comprehensiveness and integrity of the training data. A semantic network-based ML engine is capable of conceptual reasoning in challenging domains. However, it should be noted that in the present study Headai Graphmind performed best in two cases: with training data based on ICF titles and with training data based on the domain expert’s short explanations of the ICF code written in professional language. When applying Medical Subject Headings (MeSH) vocabulary or definitions that are too generic as training data, the results were imprecise and did not produce accurate matches. Furthermore, where the examples were few and anomalous, the codes were completely missed by the algorithm. There was also difficulty with overlapping terms, such as sitting (d4103) and maintaining a sitting position (d4153). Indeed, this finding agrees with the results of earlier studies on semantic computing, that is, the smaller the training data are, the more critical the quality of the data is. Thus, although semantic computing cannot solve ICF coding alone, it can be applied effectively when there is enough computational, linguistic, and health care expertise involved. 5 Conclusions In conclusion, the findings of this feasibility study suggest that the method developed here with the graph machine learning engine, Headai Graphmind, has the capability to be used as a component of data architecture to build an interface in the current computing architectures of health care facilities. It can facilitate tailoring of holistic treatment decisions based on status of functioning for individuals and improve the biopsychosocial understanding of health of the population. Declarations 6 Conflict of Interest Author H.K. is employed by HeadAI Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. 7 Author Contributions All the authors contributed to designing the study. L.N. worked as the domain expert, collected the data, generated the training and evaluation datasets, annotated the texts, and analyzed the algorithm’s results. H.K. developed the algorithm, the data architecture, and controlled the data procession. L.N. and H.K. were mainly responsible for writing the manuscript, with inputs and critical appraisals from J.V. and M.K. 8 Acknowledgments and funding The authors thank Heidi Parisod from the University of Turku (Finland) for the insightful comments on the manuscript. This study was financially supported by Tampere University Hospital Support Foundation, Tampere University Hospital, Finland (project number MK367) and Finnish State Research Funding (project number 9AC067). 9 Data Availability Statement The original patient data in this study were used under a data transfer contract and are not publicly available due to General Data Protection Regulations. However, anonymized data are available from the corresponding author upon reasonable request and with permission of Tampere University Hospital, Finland. Headai Graphmind is a commercial semantic computing infrastructure. It can be licensed and run in Linux and Azure clouds and servers in isolated mode (as done in this study). Furthermore, Headai Graphmind REST-API is available for cases where data can be transferred to the Internet. References Maritz, R., Aronsky, D. & Prodinger, B. The International Classification of Functioning, Disability and Health (ICF) in Electronic Health Records. A Systematic Literature Review. Appl. Clin. Inf. 8 , 964–980 (2017). World Health Organization. ICD-11: International classification of diseases (11th revision). (2022). https://icd.who.int/ Hartvigsen, J. et al. What low back pain is and why we need to pay attention. Lancet 391 , 2356–2367 (2018). WHO classifications. International Classification of Functioning, Disability and Health (ICF). https://www.who.int/standards/classifications/international-classification-of-functioning-disability-and-health Stallinga, H. A., Roodbol, P. F., Annema, C., Jansen, G. J. & Wynia, K. Functioning assessment vs. conventional medical assessment: a comparative study on health professionals’ clinical decision-making and the fit with patient’s own perspective of health. J. Clin. Nurs. 23 , 1044–1054 (2014). Frattura, L. et al. The FBE development project: toward flexible electronic standards-based bio-psycho-social individual records. Stud. Health Technol. Inf. 180 , 651–655 (2012). World Health Organisation WHO. ICF Beginner’s Guide: Towards a Common Language for Functioning, Disability and Health. (2002). https://www.who.int/publications/m/item/icf-beginner-s-guide-towards-a-common-language-for-functioning-disability-and-health Finnish Institute of Health and Welfare & Copyright Elisa oyj. Koodilistaus luokituksesta ‘THL - ICF Toimintakykykäsitteiden luokitus’. Version 8.7.1 (2024). https://koodistopalvelu.kanta.fi/codeserver/pages/classification-view-page.xhtml (2018). Tagliaferri, S. D. et al. Artificial intelligence to improve back pain outcomes and lessons learnt from clinical classification approaches: three systematic reviews. NPJ Digit. Med. 3 , 93 (2020). Moshayedi, A. J., Roy, A. S., Kolahdooz, A. & Shuxin, Y. Deep Learning Application Pros And Cons Over Algorithm. EAI Endorsed Trans. AI Rob. 1 , e7–e7 (2022). Newman-Griffis, D. et al. Linking Free Text Documentation of Functioning and Disability to the ICF With Natural Language Processing. Front. Rehabilitation Sci. 2 , 742702 (2021). Ketamo, H., Moisio, M., Passi-Rauste, A. & Alamäki, A. Mapping the Future Curriculum: Adopting Artifical Intelligence and Analytics in Forecasting Competence Needs. in Proceedings of the 10th European Conference on Intangibels and Intellectual Capital ECIIC 2019 144–153 (Sargiacomo, M., Chieti-Pescara, Italy, (2019). Headai science. Digital Self – The Core Model Behind Simulations. https://headai.com/science/ Gaudelet, T. et al. Utilizing graph machine learning within drug discovery and development. Brief. Bioinform 22 , (2021). Ketamo, H. & Kiili, K. Conceptual Change Takes Time: Game Based Learning Cannot be Only Supplementary Amusement. J. Educational Multimedia Hypermedia . 19 , 399–419 (2010). Nieminen, L., Vuori, J. & Ketamo, H. & Kankaanpää Markku. Applying semantic computing for health care professionals: the timing of intervention is the key for successful rehabilitation. in Proceedings of 31st conference of open innovations association FRUCT 201–206FRUCT Association, Helsinki, Finland, (2022). Vosniadou, S. Conceptual change approach and its re-framing. in Reframing The Conceptual Change Approach in Learning and Instruction (eds Vosniadou, S., Baltas, A. & Vamvakoussi, X.) 1–15 (Elsevier, Oxford, (2007). Foster, N. E. et al. Prevention and treatment of low back pain: evidence, challenges, and promising directions. Lancet 391 , 2368–2383 (2018). GBD 2016 Disease and Injury Incidence and Prevalence Collaborators. 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. Lancet 390 , 1211–1259 (2017). Meucci, R. D., Fassa, A. G. & Faria, N. M. Prevalence of chronic low back pain: systematic review. Rev. Saude Publica . 49 , 1 (2015). Freburger, J. K. et al. The Rising Prevalence of Chronic Low Back Pain. Arch. Intern. Med. 169 , 251–258 (2009). Buchbinder, R. et al. Low back pain: a call for action. Lancet 391 , 2384–2388 (2018). Hill, J. C. et al. A primary care back pain screening tool: identifying patient subgroups for initial treatment. Arthritis Rheum. 59 , 632–641 (2008). Nieminen, L. K., Pyysalo, L. M. & Kankaanpää, M. J. Prognostic factors for pain chronicity in low back pain: a systematic review. Pain Rep. 6 , e919 (2021). Herman, P. M. et al. Definitions of Chronic Low Back Pain From a Scoping Review, and Analyses of Narratives and Self-Reported Health of Adults With Low Back Pain. J. Pain . 24 , 403–412 (2023). Ministry of Social Affairs and Health. Secondary use of health and social data. (2019). https://stm.fi/en/secondary-use-of-health-and-social-data Hoy, D. et al. The global burden of low back pain: estimates from the Global Burden of Disease 2010 study. Ann. Rheum. Dis. 73 , 968–974 (2014). Cieza, A. et al. ICF linking rules: An update based on lessons learned. J. Rehabil Med. 37 , 212–218 (2005). Cieza, A. et al. ICF Core Sets for low back pain. J. Rehabil Med. 36 , 69–74 (2004). Pajula, J., Viiri, S., Similä, H., Lähteenmäki, J. & Tuomi-Nikula, A. Impacts of the Law on Secondary Use of Health and Social Data on Research and Applications of Data Analytics: Report of Hyteairo Analytics Working Committee. VTT Research Report . VTT Technical Research Centre of Finland vol. No. VTT-R-00118-21 (2021). https://cris.vtt.fi/ws/portalfiles/portal/43960329/VTT_R_00118_21.pdf Mescouto, K., Olson, R. E., Hodges, P. W. & Setchell, J. A critical review of the biopsychosocial model of low back pain care: time for a new approach? Disabil. Rehabil . 44 , 3270–3284 (2022). Iliashenko, O., Bikkulova, Z. & Dubgorn, A. Opportunities and challenges of artificial intelligence in healthcare. in E3S Web of Conferences (ed. Kalinina, O.) vol. 110 02028EDP Sciences, St. Petersburg, (2019). Matheny, M. E., Whicher, D. & Thadaney Israni, S. Artificial Intelligence in Health Care: A Report From the National Academy of Medicine. JAMA 323 , 509–510 (2020). The Social Insurance Institution of Finland & Kanta Services. What are the Kanta Services? - Citizens - Kanta.fi. https://www.kanta.fi/en/what-are-kanta-services Nordic interoperability. https://nordicinteroperability.com/. Horgan, D. et al. European Health Data Space - An Opportunity Now to Grasp the Future of Data-Driven Healthcare. Healthcare 10 , 1629 (2022). The European Health Data Space (EHDS). https://www.european-health-data-space.com/ Tables Table 1 . Inclusion and exclusion criteria. LBP= low back pain, SBT= STarT Back Screening Tool, VAS= Visual Analog Scale. Inclusion criteria Exclusion criteria Aged 18 to 65 years LBP symptoms ≥ 3 months SBT questionnaire fulfilled Pain chart fulfilled Social security number available VAS ≥ 3 Malignancy Recent traumatic fracture to the pain region Osteoporotic fracture Infection (i.e., epidural abscess) Ankylosing spondylitis Modic 1 changes Unstable spondylolisthesis Anomaly of the bone in the pain region Severe scoliosis (>45°) A nerve root disorder with apparent dermatomal and/or myotomal radiculopathy (pain, numbness, paresthesia, tingling, muscle weakness) Any other obvious specific reason for LBP Table 2. Characteristics of the whole study population and population randomized to analysis. BMI= Body Mass Index, LBP= Low back pain, NSAID= non-steroid anti-inflammatory drug, VAS= Visual Analog Scale, SBT= STarT Back Tool. SBT Q3= I have walked only short distances because of my back pain, Q4=In the last two weeks, I have dressed more slowly than usual because of my back pain. Variable Population (n=93) Population randomized to analysis (n=20) Male (n/%) 30/32% 9/45% Age (mean) 45 years (95% CI ±2 years) 43 years (95% CI ±5 years) BMI (mean) 28.3 (95% CI ± 2.7) 28.1 (95% CI ± 1.4) Duration of LBP (n/%) 3-6 months 6/6% 1/5% 6-12 months 14/15% 4/20% 1-2 years 15/16% 7/35% 2-5 years 17/18% 4/20% 5-10 years 8/9% 1/5% >10 years 33/36% 3/15% On pain medication (n/%) 86/92% 19/95% NSAID 69/74% 16/80% Paracetamol 42/45% 8/40% Opiate 30/32% 5/25% Neuropathic pain medication 25/27% 5/25% VAS in motion (mean) 6.3 (95% CI ±0.6) 6.3 (CI ±1.0) VAS in rest (mean) 5.5 (95% CI ±0.5) 5.6 (CI ± 0.3) SBT score total score (mean) 7 (95% CI ± 0.3) 6 (CI ± 0.05) sub score Q5-9 (mean) 4 (95% CI ±0.2) 4 (CI ± 0.005) Yes on Q3 64/69% 12/60% Yes on Q4 51/55% 10/50% On sick leave due to LBP 61/66% 10/50% less than 30 days 11/18% 3/15% 1-3 months 24/39% 3/15% 4-6 months 5/8% 1/5% over 6 months 17/28% 3/15% N/A 4/7% 0 “I can work in the same profession in 2 years’ time despite my health” Most definitely 13/14% 4/20% I’m not sure 42/45% 10/50% Probably not 31/33% 4/20% N/A 7/8% 2/10% Has had physiotherapy 76/82% 14/70% Has been in institutional rehabilitation 15/16% 1/5% Has imaging studies done 83/89% 17/85% Table 3. The quantitative findings of ICF chapters ordered in the ICF components in the EHR of the evaluation dataset (20 patients). The number of findings is presented in brackets, and the percentage calculated from all the findings (n=3601). Chapter domains with only one or no findings are excluded. Body Function, B (n=896/25%) Structure, S (n=1444/40%) Sensory functions and pain, b2 (366) Neuromusculoskeletal and movement related functions, b7 (349) Mental functions, b1 (99) Functions of the digestive, metabolic and endocrine systems, b5 (62) Genitourinary and reproductive functions, b6 (12) Functions of the cardiovascular, hematological, immunological, and respiratory systems, b4 (8) Structures related to movement, s7 (1364) Structures of the nervous system, s1 (73) Structures related to the digestive, metabolic and endocrine systems, s5 (7) Activities and participation, D (n=569/16%) Mobility, d4 (310) Community, social, and civic life, d9 (103) Major life areas, d8 (89) Self-care, d5 (44) Domestic life, d6 (16) Interpersonal interactions and relationships, d7 (7) Environmental factors, E (n=692/19%) Services, systems, and policies, e5 (298) Products and technology, e1 (253) Support and relationship, e3 (137) Natural environment and human-made changes to environment, e2 (4) Table 4. Results of the factor recognition. S= body structures, B= body functions, D= activities and participation, E= environmental factors. S B D E Total Expert found 423 311 226 112 1072 Headai Graphmind found codes in total 371 312 208 100 991 true positives 368 285 195 94 942 false positives 3 27 13 6 49 false negatives 63 53 55 20 191 codes better than expert 4 14 12 1 31 Correct codes in total 427 325 238 113 1103 Precision (95% CI) 0.99 (0.957- 1.00) 0.91 (0.879- 0.941) 0.94 (0.901-0.979) 0.94 (0.917-0.963) 0.95 (0.939-0.961) Recall (95% CI) 0.85 (0.817-0.883) 0.84 (0.809- 0.871) 0.78 (0.735-0.825) 0.83 (0.795-0.865) 0.83 (0.823-0.837) F1 score (95% CI) 0.92 (0.887-0.953) 0.88 (0.849- 0.911) 0.85 (0.79-0.91) 0.88 (0.859- 0.901) 0.89 (0.884- 0.896) Additional Declarations Competing interest reported. Author H.K. is employed by HeadAI Ltd. All other authors have no competing interests. Cite Share Download PDF Status: Published Journal Publication published 02 Jul, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 07 May, 2025 Reviews received at journal 24 Apr, 2025 Reviewers agreed at journal 16 Apr, 2025 Reviews received at journal 15 Apr, 2025 Reviewers agreed at journal 14 Apr, 2025 Reviewers invited by journal 14 Apr, 2025 Submission checks completed at journal 07 Apr, 2025 First submitted to journal 22 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5415974","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":442576559,"identity":"3f5a3743-bf71-40b0-b0bc-ff2cd894ceff","order_by":0,"name":"Linda Nieminen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCklEQVRIiWNgGAWjYBACPmYGBgkGhgQGBiADRDHwMzA2gNj8bDi0sGFokWyAaJFsw6WFAaYFBgwOQLRLNuDSws578MYPhjQGc3bmoxse7rDJM77d3Pa5oIZBgg+nw/iSLXsYchgsm9nSbiSeSSs2u3OwefaMYwwSuP3CYybBw1DBYHCYx+xGYtvhxG03EpuZedgY6vBpkfwD1sL/Dajlf+LmGSAt//DbIs0DdBjQFjaglgOJGySAWnjb8GoxtpYxSOMxOMwGclhy4gygX5h5+yRwauHnP2N4801FspzB+cPPbv5ss0vsn93+mJnnm42EfAMOPWBgwMCD4EggkUQCkhSPglEwCkbBSAAAUgZMAMR4/18AAAAASUVORK5CYII=","orcid":"","institution":"Tampere University","correspondingAuthor":true,"prefix":"","firstName":"Linda","middleName":"","lastName":"Nieminen","suffix":""},{"id":442576560,"identity":"48d16a38-5417-4648-a2cb-c12b7b3d65c2","order_by":1,"name":"Harri Ketamo","email":"","orcid":"","institution":"HeadAI Ltd","correspondingAuthor":false,"prefix":"","firstName":"Harri","middleName":"","lastName":"Ketamo","suffix":""},{"id":442576561,"identity":"aafb145d-ba2a-46d7-afb6-3003d7e09b7c","order_by":2,"name":"Jari Vuori","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Jari","middleName":"","lastName":"Vuori","suffix":""},{"id":442576562,"identity":"ceec3057-a546-4357-afc5-82d067832f3a","order_by":3,"name":"Markku Kankaanpää","email":"","orcid":"","institution":"Tampere University Hospital","correspondingAuthor":false,"prefix":"","firstName":"Markku","middleName":"","lastName":"Kankaanpää","suffix":""}],"badges":[],"createdAt":"2024-11-08 10:53:41","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5415974/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5415974/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-06429-4","type":"published","date":"2025-07-02T15:57:26+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":80783857,"identity":"18affa54-edb7-48e3-801b-3304469c9897","added_by":"auto","created_at":"2025-04-17 05:31:32","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":40448,"visible":true,"origin":"","legend":"\u003cp\u003eSimplified and cleaned example of the semantic network model after Headai Graphmind has read the EHR texts.\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/acef46482b21801c2ff6afd7.jpg"},{"id":80783859,"identity":"51dfb1b6-ec3e-457d-8ba0-386b622ab5b5","added_by":"auto","created_at":"2025-04-17 05:31:32","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":84524,"visible":true,"origin":"","legend":"\u003cp\u003eData architecture and components\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/68170313a8cf4d09054041f2.jpg"},{"id":80783862,"identity":"46378fb6-f4c6-43c2-b5e6-6f6cd0ac291b","added_by":"auto","created_at":"2025-04-17 05:31:33","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":117837,"visible":true,"origin":"","legend":"\u003cp\u003eThe semantic matching process of Headai Graphmind in the current study. MD= Medical doctor.\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/bc03b5c9387785bb54016ca1.jpg"},{"id":80785139,"identity":"b047ef70-b32e-4582-8327-1944a6b818d9","added_by":"auto","created_at":"2025-04-17 05:39:33","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":49908,"visible":true,"origin":"","legend":"\u003cp\u003eThe flow diagram of patient selection.\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/a75b85d702017fcbde6a17f8.jpg"},{"id":80783869,"identity":"4849c100-c689-4c62-82a6-a192ad5dbddf","added_by":"auto","created_at":"2025-04-17 05:31:33","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":64054,"visible":true,"origin":"","legend":"\u003cp\u003eAn extract from a health record with an example of the annotation done by domain expert.\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/fc0e8db0323f598c9a65e0ce.jpg"},{"id":86180503,"identity":"b8f85b37-dfeb-4aed-a07a-20a2a6e983f7","added_by":"auto","created_at":"2025-07-07 16:22:22","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1146036,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5415974/v1/ea44ade2-f162-4031-9cef-1924a8944c63.pdf"}],"financialInterests":"Competing interest reported. Author H.K. is employed by HeadAI Ltd. All other authors have no competing interests.","formattedTitle":"Using graph machine learning to identify functioning in patients with low back pain within the ICF framework","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eHow can we comprehensively treat a patient who has experienced prolonged pain so that the onset of pain chronicity becomes preventable? What are the obstacles that prevent this patient with multiple morbidities from returning to work? The answers to a holistic perspective on functioning and health can usually be found in the health records of individual patients or in group-specific cohorts \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. Finding answers to these questions is, however, challenging and cannot be achieved by evaluating diagnosis codes\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e alone. For example, two patients with the same low back pain diagnosis can have completely different levels of functioning, with the first patient coping at work, whereas the second has major difficulties coping and a low quality of life \u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTo enhance the understanding of health and health-related states holistically, the World Health Organization (WHO) has developed the International Classification of Functioning, Disability, and Health (ICF) framework \u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. When used as part of functioning assessment, the ICF can create not only a broader picture of health, but can also reflect patients\u0026rsquo; self-reported problems more closely than a typical medical assessment \u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Previously, different approaches have been used to integrate the ICF with electronic health records (EHR) \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e, predominantly in rehabilitation settings, and in some cases where the ICF was embedded in health information systems \u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. However, the optimal implementation of the ICF, including structured documentation conducted by health care professionals, has been slow. This delay has been mainly due to the complex structure of the ICF taxonomy that comprises approximately 1600 codes representing different domains of functioning \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. There is a need, therefore, for efficient and reliable methods to enhance the wider implementation of the ICF as part of the clinical workflow in health care settings.\u003c/p\u003e \u003cp\u003eArtificial intelligence (AI) techniques show promise in the implementation of ICF taxonomy as an integral part of medical assessment, since such techniques have the capability to solve complex tasks. Likewise, AI techniques could also serve to promote a broader understanding of those factors concerning the disability of patients \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, which could, in turn, help to plan more high-quality personalized care with meaningful goals. It should be noted, however, that the selection of a suitable AI or machine learning (ML) algorithm is a complicated process. For example, deep learning algorithms perform well in categorizing tasks when the task is well defined, and the training material is large enough. In contrast, deep learning algorithms do not perform well in cases where the task is ill defined and requires human reasoning when working with unknown factors \u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. In a previous study, Newman-Griffis et al. selected a natural language processing (NLP) technique for linking free text to 29 ICF-based categories \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eRecently, a semantic network-based (graph) ML engine, Headai Graphmind, which has the capability to imitate human reading and processing of texts, has been applied to forecast future skill needs in the labor market and to optimize learning paths for the future workforce using curriculum gap assessment \u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. In previous studies, graph-based machine learning or semantic network-based setups have shown promise in different complex settings \u003csup\u003e\u003cspan additionalcitationids=\"CR15\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. In practice, Headai Graphmind adds, modifies, and reasons natural language according to conceptual learning theories \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, whereas a semantic network serves as a structure for all the data learned. Furthermore, in contexts where formal procedural rules for matching are unknown, technology that can operate with unstructured data has the potential to fully utilize EHR information and harness data for further analysis conducted by health care professionals \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eA good example of the applicability of Headai Graphmind is in the recognition of patients at risk for low back pain (LBP). Although vast amounts of research and resources have been devoted to LBP during the last decades \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e, the burden of LBP has increased, making it the most burdensome global health problem affecting years lived with disability (YLD) \u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Although not all patients with LBP develop a chronic pain problem \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e, the increased use of health care resources and the subsequent rise in costs related to LBP are mainly driven by chronic cases \u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Therefore, strategies that ensure the early identification of those patients at risk for persistence of pain and disability should be developed and implemented \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. At present, questionnaires such as the STarT Back Screening Tool (SBT) \u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e are used to identify these patients, but they are not comprehensive enough to recognize all the relevant factors (23,24) required to ensure a holistic decision-making process. The findings of a preliminary study \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e suggest that Headai Graphmind could have the potential to make the timing and tailoring of interventions in the LBP patient population easier.\u003c/p\u003e \u003cp\u003eIn this feasibility study, the EHR of patients with chronic LBP were used to generate natural language definitions of the ICF entities. The main aim of the study is to test the validity, especially criterion validity, of Headai Graphmind applied to determine semantically the best matches between ICF code definitions and the natural language of the EHR. The reliability of Headai Graphmind is tested against the findings of a domain expert who has profound understanding of the EHR in question, the functioning of LBP patients, and the ICF. Our research questions were as follows: (1) What meaningful items of functioning can be found from the electronic health records of patients? (2) Can an AI method perform meaningful item recognition reliably enough to support the health care decision-making process when compared to a domain expert?\u003c/p\u003e"},{"header":"2 Methods","content":"\u003ch2\u003e2.1 Data architecture and data processing\u003c/h2\u003e\n\u003cp\u003eThe generic idea behind Headai Graphmind is that it has a pre-trained semantic understanding of language. This understanding is based on gigabytes of generic data collected from research papers (open journals), policy papers (such as the European Union\u0026rsquo;s archive), labor market information (e.g., job ads), and professional news (technology, business, medical, and so forth). As the data can affect the semantic behavior of the engine, the training data have been selected from sources that are widely used and trusted. Further, when reading this training data, Headai Graphmind learns which words are meaningful, when the words form a compound word (also called n-grams in linguistics), what is the relationship between the words and the n-grams, and eventually the context of the words and n-grams used. This semantic network is applied when starting to perform reasoning between ICF entity descriptions and real-world texts from the EHR. Headai Graphmind turns the EHR texts into a similar type of semantic network as the pre-trained language model (Figure 1). After that, it starts to fit the ICF descriptions to the EHR semantic network and applies the pre-trained language model to understand the small data, such as synonyms, neighboring concepts, and similar meanings, more precisely. This two-layer approach enables matching between two small datasets without getting stacked into a lack of structured data or non-matching words.\u003c/p\u003e\n\u003cp\u003eThe data architecture (Figure 2) from the EHR to Headai Graphmind to the research environment was built to meet high data security and privacy requirements. All the data from the EHR were provided as an encrypted pseudonymized csv.-file on a memory stick that was extracted offline on Headai Graphmind\u0026rsquo;s side. In a study by Nieminen and colleagues \u003csup\u003e16\u003c/sup\u003e, 12 different ICF definition sets were studied to find the most functional setups. The best performing definition sets (called ICF title, ICF real life, ICF real life fuzzy, MesH words, MeSH-ICF title) were also imported to Headai Graphmind before the analysis. However, because these datasets were not GDPR (General Data Protection Regulation) -data, they were imported directly from the csv.-file without encryption.\u003c/p\u003e\n\u003cp\u003eThe matching between the EHR data and the ICF definition sets were run in offline mode, and the results with pseudo-identifier were copied into the research environment with authenticated network access. The offline nature of the Headai Graphmind process means that all data and computing are performed outside the network-accessible directories of the Headai Graphmind computing infrastructure, and only Headai Graphmind itself is connected to the network and only few directories are accessible via the network.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe Headai Graphmind semantic matching process is described in more detail in Figure 3. The written texts in the EHR are then turned into semantic networks. Each patient\u0026rsquo;s semantic network is analyzed against each of the more than 1600 ICF definitions in the chosen definition sets (as mentioned above), resulting in approximately 36 000 analyses per patient (MeSH alone contains 30 000 entries). The semantic matching of Headai Graphmind is based on shallow neural networks, and the analysis took between 2 and 10 seconds using one CPU core per patient, depending on data volume and data complexity. In this study, the stored semantic network of the patient (JSON_GRAPH in Figure 3) was only used to support the domain expert\u0026rsquo;s evaluation. The performance of the algorithm were evaluated with precision, recall, and F1 score.\u003c/p\u003e\n\u003ch2\u003e2.2 Population data\u003c/h2\u003e\n\u003cp\u003eWe used patient data gathered between October 2019 and February 2021. The natural language data were free text (Finnish language) consisting of the EHR of patients with chronic\u003csup\u003e25\u003c/sup\u003e (duration 3 months or longer) LBP who attended the Department of Physical Rehabilitation and Medicine at Tampere University Hospital, Finland. Additional information was collected in the form of quantitative data, which were retrieved from medical history forms. The quantitative data were used for the data selection process (the fulfillment of the inclusion and exclusion criteria, Table 1). The data were collected retrospectively after the treatment period had ended and patients had returned to primary or occupational care. The search for suitable patients began with medical history forms that were collected from all patients attending the department. Since the study was registry-based and the integrity of the patients was maintained, the Ethics Committee of Tampere University Hospital waived the need for ethical approval and informed patient consent. In addition, Finnish research legislation on the secondary use of health and social data (legislation no. 552/2019)\u003csup\u003e26\u003c/sup\u003e allows the retrieval of retrospective data from the EHR for research purposes without informed consent from the patients. Since the data in the present study were gathered from a single data source, a data transfer contract was drawn up between the investigators and the data controller, Tampere University Hospital, Finland. The study was carried out in accordance with all relevant guidelines and regulations.\u003c/p\u003e\n\u003cp\u003eLBP was defined as pain in the anatomical region between the costal margins and the inferior gluteal folds with or without radicular pain\u003csup\u003e27\u003c/sup\u003e. Multifocal pain was not an exclusion criterion. Inclusion and exclusion criteria are presented in Table 1.\u003c/p\u003e\n\u003cp\u003eThe data was processed using a deductive content analysis approach, using both qualitative and quantitative methods. The EHR datasets of 20 patients were randomly selected to form a training set. A satisfactory saturation of the free text was achieved in the training set, since the last 5 patient datasets added did not significantly increase the number of ICF codes. The free text annotation, being the linking of the ICF to the EHR, applied the principals of proposed ICF linking rules \u003csup\u003e28\u003c/sup\u003e. A domain expert searched the EHR texts of the training dataset for suitable words and n-grams that represented the contents of the ICF. These words and n-grams were further annotated with ICF codes using third and fourth level codes wherever possible to produce as specific examples as possible. In addition, the developed vocabulary was enriched by the domain experts\u0026rsquo; perspective on the language often used by professionals (\u0026ldquo;jargon\u0026rdquo;) in case some words or n-grams were missing from the training dataset.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eA further 20 randomly selected EHR datasets were used for the quality analysis of Headai Graphmind. The same domain expert who assembled the training data also evaluated the Headai Graphmind results. Both samples of the study population were obtained using computer-based randomization. Quantitative content analysis was used as a method in the following manner: The search of the contents and further annotation with the ICF codes repeated the method used to compile the training data. Thereafter, the disability information gathered from the free texts was synthesized quantitatively to gain an understanding of the ICF categories found (Table 3). The results were then analyzed by the domain expert between the original EHR data and the end results of the Headai Graphmind in the research environment. Only the domain expert who had access to both systems, participated in this part of the study.\u003c/p\u003e\n\u003cp\u003eThe analysis dataset was compressed on an individual document level so that recurring codes were condensed into a single finding. The results were interpreted by the domain expert in the following manner. The finding was defined as true positive if the algorithm found the same code as the domain expert from the free text of one patient. False positives were the codes that the domain expert did not find and, after reappraisal, were still regarded as false findings. Codes were defined as false negatives if a code was found by the domain expert but not by the algorithm. Additionally, there were codes that were first found by the algorithm and, after reappraisal, found by the domain expert as well.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe population data were analyzed with IBM SPSS Statistics for Windows, Version 28.0, for mean and 95% confidence intervals.\u003c/p\u003e"},{"header":"3 Results","content":"\u003ch2\u003e3.1 Study population\u003c/h2\u003e\n\u003cp\u003eThe flow diagram (Figure 4) presents the selection of the patient sample. The EHR of 93 patients with non-specific LBP were collected for the purposes of this and preliminary research. The first round of exclusion was based on medical history forms. Age, main complaint, inadequate information (for example, missing information about duration of symptoms), Visual Analog Scale (VAS), and duration of symptoms were all considered. In total, 335 working-aged patients, whose main complaint was back pain, were identified from the medical history forms. The fulfillment of the inclusion criteria was finally verified from the EHR. In total, the EHR of the selected population (n=93) comprised 312 physicians\u0026rsquo; notes, including texts of referrals, physical appointments, and records of contacts by phone and letter. The characteristics of the patient population are presented in Table 2.\u003c/p\u003e\n\u003cp\u003eTable 3 presents the quantitative findings of the domain expert from the evaluation dataset (EHR from 20 patients, 63 EHR notes). The mention of body structures was the highest (n=1,444). The notes also contained versatile information on other domains related to disability, such as neuromusculoskeletal and movement-\u003c/p\u003e\n\u003cp\u003erelated functions (n=349), information on joint and bone mobility and muscle endurance, information on products and technology (n=253), information on medication, and the mobility aids the patients used. Additionally, mental functions (n=99) contained information on sleep quality, mental and personality disorders, emotions, and mental energy levels. Social factors were available from community, social, and civic life (n=103) or major life areas (n=89) where work-related factors are described. An extract from a health record note is presented in figure 5 to give an example of the included ICF -related information and annotation.\u003c/p\u003e\n\u003ch2\u003e3.2 Headai Graphmind\u003c/h2\u003e\n\u003cp\u003eHeadai Graphmind was analyzed for its semantic matching abilities of factor recognition in the four domains of the ICF: body structures (S codes), body functions (B codes), activities and participation (D codes), and environmental factors (E codes) (Table 4). Headai Graphmind performed the matching on the data of the whole study population (93 patients, 312 EHR notes). To obtain an estimate of Headai Graphmind\u0026rsquo;s reliability, an evaluation of the factor recognition was performed on a random sample of the EHR of 20 patients (63 EHR notes). The sample size was chosen due to the exhaustive nature of the evaluation process. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe algorithm reached precision 0.95 (9\u003cstrong\u003e5\u003c/strong\u003e% CI 0.939-0.961) and recall 0.83 (95% CI 0.823-0.837) , from which F1 score was 0.89 (95% CI 0.884- 0.896) (Table 4). Recall was highest in both the environmental factor and body structure domains (0.85) and lowest in the activity and participation domain (0.78), whereas precision was highest in the body structure domain (0.99) and lowest in the body function domain (0.91). Furthermore, when comparing the content of the codes, the domain expert found 119 distinct codes (30 S codes, 35 B codes, 40 D codes, and 14 E codes) from the evaluation dataset, whereas the algorithm found 112 codes (30 S codes, 35 B codes, 35 D codes, and 12 E codes). The missed codes were d4103 (sitting), d4302 (carrying in the arms), d630 (preparing meals), d6402 (cleaning living area), d825 (vocational training), e1200 (general products and technology for personal indoor and outdoor mobility and transportation), and e1151(assistive products and technology for personal use in daily living).\u003c/p\u003e"},{"header":"4 Discussion","content":"\u003cp\u003eThe main finding of this feasibility study was that Headai Graphmind performed the factor recognition of ICF information from the EHR of patients with LBP with convincing performance when compared to the results of the domain expert (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). Regarding our first study question, the EHR notes of individuals with chronic LBP were expressive and contained holistic information about the disability of individuals (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e3\u003c/span\u003e). The population data also reflected closely to the WHO ICF core set for LBP \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e. Compared to a previous study on the subject \u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, a wider selection of ICF-based categories was obtained. Gaining this information from the patient population makes it possible to holistically support the decision-making process in the treatment and rehabilitation assessment of patients.\u003c/p\u003e \u003cp\u003eFurthermore, these results offer the promise of a new functional application for personalized medicine, where an individualized model of a patient\u0026rsquo;s history can be used to take preventative actions \u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. The ICF framework perspective identified from the EHR can broaden our understanding of the functioning-related factors affected by the patients\u0026rsquo; medical condition. The universal, interdisciplinary language of the ICF can be applied globally in different health care settings and by different health care professions to produce a broader biopsychosocial understanding of health \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The biopsychosocial model has been developed to shift the focus from a narrow biomedical model to understanding concepts of health and functioning through biomedical, psychological, and social dimensions \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. Additionally, the ICF considers environmental and individual factors that could be more easily accessible in future using the AI application tested here.\u003c/p\u003e \u003cp\u003eSince Headai Graphmind is based on shallow neural networks, the speed (and resulting low energy consumption) enables large scale analysis in, e.g., monthly analysis. When considering the speed of Headai Graphmind and the complexity of health data, it can be estimated that it would take approximately 15 days to analyze 1 000 000 patient records with 8 core computing setups, which can be regarded as a very low requirement at present. With respect to data safety issues and the embedding of AI architecture to current computing architectures \u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e, Headai Graphmind can work as a plug-in and does not require any software integration. The actual tool for decision-making support needs to fit the purposes of end-users and thus, the visual display needs to be designed by professionals.\u003c/p\u003e \u003cp\u003eWhen using AI-derived information for decision-making support, it is important to make sure that noise in the health care data does not drive the decision-making process. Consequently, it is up to the end-user to make sure that the information does not lead to unintended effects, such as discrimination, increased inequities, and decreased inclusion \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eSeveral future implications emerged during our study. As part of the study, only one semantic network was performed per patient to study the quality of the process. In future, however, yearly networks can be performed per patient to enable time series analysis. Additionally, we can examine population cohorts to obtain a wider understanding of the functioning and disability of citizens. In future, new studies on the validity and reliability of the developed application must be conducted with texts unrelated to chronic pain. The core architecture of Headai Graphmind as a semantic computing platform is designed to be language agnostic. Therefore, it is used in settings other than health care for tens of real-world customer cases in English, Spanish, French, German, Swedish, Vietnamese, Estonian, Ukrainian, and Finnish. Although different use cases require independent validation, the design of the technology does enable faster development in new languages and cultural environments. Unfortunately, data silos pose a problem for efficient data processing in many health care ecosystems. In Finland, there is a centralized archive of electronic patient data, making data standardization possible \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. There are also initiatives towards unified health care records in the Nordic countries \u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e as well as in the European Union \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. The European Health Data Space (EHDS) is one of activities of the European Data Spaces initiative, focused on building strong governance and basic functions to ensure fair data sharing. The method tested in this research could serve as one of the building blocks within the EHDS.\u003c/p\u003e \u003cp\u003e4.1 Limitations\u003c/p\u003e \u003cp\u003eThe data used in this present study had some limitations. The data consisted of only the medical notes of physicians. Therefore, the results of the present study can only be generalized to physicians\u0026rsquo; notes and patients with low back pain. Further analysis of construct and content validity and reliability with new data will be needed to study the applicability of Headai Graphmind with other health care professional and patient groups.\u003c/p\u003e \u003cp\u003eThe data itself had limitations in terms of providing a biopsychosocial view of patients with chronic LBP. The majority (65%) of ICF-related information (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e3\u003c/span\u003e) was related to body structures and functions. The documents provide a good insight into everyday healthcare, where the biomedical model is dominant. If an algorithm, such as presented in this study, were to visualize the ICF-related information for health professionals, it could highlight the need to shift the focus to activities, participation and environmental factors in order to truly support the patients\u0026rsquo; wellbeing.\u003c/p\u003e \u003cp\u003eThe presence of only one domain expert doing the annotation and analysis can be regarded as a major weakness in this study. On the one hand, the annotation and analysis proceeded in a homogenous way, but on the other multiple experts would have brought different interpretations of the text and the results. This would have strengthened the study, especially for future applicability. In the future validation process, multiple experts will be used, and inter-rater reliability will be tested, to ensure the comprehensiveness and integrity of the training data.\u003c/p\u003e \u003cp\u003eA semantic network-based ML engine is capable of conceptual reasoning in challenging domains. However, it should be noted that in the present study Headai Graphmind performed best in two cases: with training data based on ICF titles and with training data based on the domain expert\u0026rsquo;s short explanations of the ICF code written in professional language. When applying Medical Subject Headings (MeSH) vocabulary or definitions that are too generic as training data, the results were imprecise and did not produce accurate matches. Furthermore, where the examples were few and anomalous, the codes were completely missed by the algorithm. There was also difficulty with overlapping terms, such as sitting (d4103) and maintaining a sitting position (d4153). Indeed, this finding agrees with the results of earlier studies on semantic computing, that is, the smaller the training data are, the more critical the quality of the data is. Thus, although semantic computing cannot solve ICF coding alone, it can be applied effectively when there is enough computational, linguistic, and health care expertise involved.\u003c/p\u003e"},{"header":"5 Conclusions","content":"\u003cp\u003eIn conclusion, the findings of this feasibility study suggest that the method developed here with the graph machine learning engine, Headai Graphmind, has the capability to be used as a component of data architecture to build an interface in the current computing architectures of health care facilities. It can facilitate tailoring of holistic treatment decisions based on status of functioning for individuals and improve the biopsychosocial understanding of health of the population.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003e6 \u0026nbsp;Conflict of Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAuthor H.K. is employed by HeadAI Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e7 \u0026nbsp; Author Contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll the authors contributed to designing the study. L.N. worked as the domain expert, collected the data, generated the training and evaluation datasets, annotated the texts, and analyzed the algorithm\u0026rsquo;s results. H.K. developed the algorithm, the data architecture, and controlled the data procession. L.N. and H.K. were mainly responsible for writing the manuscript, with inputs and critical appraisals from J.V. and M.K.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e8 Acknowledgments and funding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors thank Heidi Parisod from the University of Turku (Finland) for the insightful comments on the manuscript.\u003c/p\u003e\n\u003cp\u003eThis study was financially supported by Tampere University Hospital Support Foundation, Tampere University Hospital, Finland (project number MK367) and Finnish State Research Funding (project number 9AC067).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e9 \u0026nbsp;Data Availability Statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe original patient data in this study were used under a data transfer contract and are not publicly available due to General Data Protection Regulations. However, anonymized data are available from the corresponding author upon reasonable request and with permission of Tampere University Hospital, Finland. Headai Graphmind is a commercial semantic computing infrastructure. It can be licensed and run in Linux and Azure clouds and servers in isolated mode (as done in this study). Furthermore, Headai Graphmind REST-API is available for cases where data can be transferred to the Internet.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMaritz, R., Aronsky, D. \u0026amp; Prodinger, B. The International Classification of Functioning, Disability and Health (ICF) in Electronic Health Records. A Systematic Literature Review. \u003cem\u003eAppl. Clin. Inf.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e, 964\u0026ndash;980 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organization. ICD-11: International classification of diseases (11th revision). (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://icd.who.int/\u003c/span\u003e\u003cspan address=\"https://icd.who.int/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHartvigsen, J. et al. What low back pain is and why we need to pay attention. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e391\u003c/b\u003e, 2356\u0026ndash;2367 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWHO classifications. International Classification of Functioning, Disability and Health (ICF). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/standards/classifications/international-classification-of-functioning-disability-and-health\u003c/span\u003e\u003cspan address=\"https://www.who.int/standards/classifications/international-classification-of-functioning-disability-and-health\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStallinga, H. A., Roodbol, P. F., Annema, C., Jansen, G. J. \u0026amp; Wynia, K. Functioning assessment vs. conventional medical assessment: a comparative study on health professionals\u0026rsquo; clinical decision-making and the fit with patient\u0026rsquo;s own perspective of health. \u003cem\u003eJ. Clin. Nurs.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 1044\u0026ndash;1054 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFrattura, L. et al. The FBE development project: toward flexible electronic standards-based bio-psycho-social individual records. \u003cem\u003eStud. Health Technol. Inf.\u003c/em\u003e \u003cb\u003e180\u003c/b\u003e, 651\u0026ndash;655 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWorld Health Organisation WHO. ICF Beginner\u0026rsquo;s Guide: Towards a Common Language for Functioning, Disability and Health. (2002). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.who.int/publications/m/item/icf-beginner-s-guide-towards-a-common-language-for-functioning-disability-and-health\u003c/span\u003e\u003cspan address=\"https://www.who.int/publications/m/item/icf-beginner-s-guide-towards-a-common-language-for-functioning-disability-and-health\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFinnish Institute of Health and Welfare \u0026amp; Copyright Elisa oyj. Koodilistaus luokituksesta \u0026lsquo;THL - ICF Toimintakykyk\u0026auml;sitteiden luokitus\u0026rsquo;. \u003cem\u003eVersion 8.7.1\u003c/em\u003e (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://koodistopalvelu.kanta.fi/codeserver/pages/classification-view-page.xhtml\u003c/span\u003e\u003cspan address=\"https://koodistopalvelu.kanta.fi/codeserver/pages/classification-view-page.xhtml\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTagliaferri, S. D. et al. Artificial intelligence to improve back pain outcomes and lessons learnt from clinical classification approaches: three systematic reviews. \u003cem\u003eNPJ Digit. Med.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e, 93 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoshayedi, A. J., Roy, A. S., Kolahdooz, A. \u0026amp; Shuxin, Y. Deep Learning Application Pros And Cons Over Algorithm. \u003cem\u003eEAI Endorsed Trans. AI Rob.\u003c/em\u003e \u003cb\u003e1\u003c/b\u003e, e7\u0026ndash;e7 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNewman-Griffis, D. et al. Linking Free Text Documentation of Functioning and Disability to the ICF With Natural Language Processing. \u003cem\u003eFront. Rehabilitation Sci.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 742702 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKetamo, H., Moisio, M., Passi-Rauste, A. \u0026amp; Alam\u0026auml;ki, A. Mapping the Future Curriculum: Adopting Artifical Intelligence and Analytics in Forecasting Competence Needs. in \u003cem\u003eProceedings of the 10th European Conference on Intangibels and Intellectual Capital ECIIC 2019\u003c/em\u003e 144\u0026ndash;153 (Sargiacomo, M., Chieti-Pescara, Italy, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHeadai science. Digital Self \u0026ndash; The Core Model Behind Simulations. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://headai.com/science/\u003c/span\u003e\u003cspan address=\"https://headai.com/science/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGaudelet, T. et al. Utilizing graph machine learning within drug discovery and development. \u003cem\u003eBrief. Bioinform\u003c/em\u003e \u003cb\u003e22\u003c/b\u003e, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKetamo, H. \u0026amp; Kiili, K. Conceptual Change Takes Time: Game Based Learning Cannot be Only Supplementary Amusement. \u003cem\u003eJ. Educational Multimedia Hypermedia\u003c/em\u003e. \u003cb\u003e19\u003c/b\u003e, 399\u0026ndash;419 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNieminen, L., Vuori, J. \u0026amp; Ketamo, H. \u0026amp; Kankaanp\u0026auml;\u0026auml; Markku. Applying semantic computing for health care professionals: the timing of intervention is the key for successful rehabilitation. in \u003cem\u003eProceedings of 31st conference of open innovations association FRUCT\u003c/em\u003e 201\u0026ndash;206FRUCT Association, Helsinki, Finland, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVosniadou, S. Conceptual change approach and its re-framing. in Reframing The Conceptual Change Approach in Learning and Instruction (eds Vosniadou, S., Baltas, A. \u0026amp; Vamvakoussi, X.) 1\u0026ndash;15 (Elsevier, Oxford, (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFoster, N. E. et al. Prevention and treatment of low back pain: evidence, challenges, and promising directions. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e391\u003c/b\u003e, 2368\u0026ndash;2383 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGBD 2016 Disease and Injury Incidence and Prevalence Collaborators. Global, regional, and national incidence, prevalence, and years lived with disability for 328 diseases and injuries for 195 countries, 1990\u0026ndash;2016: a systematic analysis for the Global Burden of Disease Study 2016. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e390\u003c/b\u003e, 1211\u0026ndash;1259 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeucci, R. D., Fassa, A. G. \u0026amp; Faria, N. M. Prevalence of chronic low back pain: systematic review. \u003cem\u003eRev. Saude Publica\u003c/em\u003e. \u003cb\u003e49\u003c/b\u003e, 1 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFreburger, J. K. et al. The Rising Prevalence of Chronic Low Back Pain. \u003cem\u003eArch. Intern. Med.\u003c/em\u003e \u003cb\u003e169\u003c/b\u003e, 251\u0026ndash;258 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBuchbinder, R. et al. Low back pain: a call for action. \u003cem\u003eLancet\u003c/em\u003e \u003cb\u003e391\u003c/b\u003e, 2384\u0026ndash;2388 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHill, J. C. et al. A primary care back pain screening tool: identifying patient subgroups for initial treatment. \u003cem\u003eArthritis Rheum.\u003c/em\u003e \u003cb\u003e59\u003c/b\u003e, 632\u0026ndash;641 (2008).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNieminen, L. K., Pyysalo, L. M. \u0026amp; Kankaanp\u0026auml;\u0026auml;, M. J. Prognostic factors for pain chronicity in low back pain: a systematic review. \u003cem\u003ePain Rep.\u003c/em\u003e \u003cb\u003e6\u003c/b\u003e, e919 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerman, P. M. et al. Definitions of Chronic Low Back Pain From a Scoping Review, and Analyses of Narratives and Self-Reported Health of Adults With Low Back Pain. \u003cem\u003eJ. Pain\u003c/em\u003e. \u003cb\u003e24\u003c/b\u003e, 403\u0026ndash;412 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMinistry of Social Affairs and Health. Secondary use of health and social data. (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://stm.fi/en/secondary-use-of-health-and-social-data\u003c/span\u003e\u003cspan address=\"https://stm.fi/en/secondary-use-of-health-and-social-data\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoy, D. et al. The global burden of low back pain: estimates from the Global Burden of Disease 2010 study. \u003cem\u003eAnn. Rheum. Dis.\u003c/em\u003e \u003cb\u003e73\u003c/b\u003e, 968\u0026ndash;974 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCieza, A. et al. ICF linking rules: An update based on lessons learned. \u003cem\u003eJ. Rehabil Med.\u003c/em\u003e \u003cb\u003e37\u003c/b\u003e, 212\u0026ndash;218 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCieza, A. et al. ICF Core Sets for low back pain. \u003cem\u003eJ. Rehabil Med.\u003c/em\u003e \u003cb\u003e36\u003c/b\u003e, 69\u0026ndash;74 (2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePajula, J., Viiri, S., Simil\u0026auml;, H., L\u0026auml;hteenm\u0026auml;ki, J. \u0026amp; Tuomi-Nikula, A. \u003cem\u003eImpacts of the Law on Secondary Use of Health and Social Data on Research and Applications of Data Analytics: Report of Hyteairo Analytics Working Committee. VTT Research Report\u003c/em\u003e. \u003cem\u003eVTT Technical Research Centre of Finland\u003c/em\u003e vol. No. VTT-R-00118-21 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cris.vtt.fi/ws/portalfiles/portal/43960329/VTT_R_00118_21.pdf\u003c/span\u003e\u003cspan address=\"https://cris.vtt.fi/ws/portalfiles/portal/43960329/VTT_R_00118_21.pdf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMescouto, K., Olson, R. E., Hodges, P. W. \u0026amp; Setchell, J. A critical review of the biopsychosocial model of low back pain care: time for a new approach? \u003cem\u003eDisabil. Rehabil\u003c/em\u003e. \u003cb\u003e44\u003c/b\u003e, 3270\u0026ndash;3284 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIliashenko, O., Bikkulova, Z. \u0026amp; Dubgorn, A. Opportunities and challenges of artificial intelligence in healthcare. in \u003cem\u003eE3S Web of Conferences\u003c/em\u003e (ed. Kalinina, O.) vol. 110 02028EDP Sciences, St. Petersburg, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMatheny, M. E., Whicher, D. \u0026amp; Thadaney Israni, S. Artificial Intelligence in Health Care: A Report From the National Academy of Medicine. \u003cem\u003eJAMA\u003c/em\u003e \u003cb\u003e323\u003c/b\u003e, 509\u0026ndash;510 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe Social Insurance Institution of Finland \u0026amp; Kanta Services. What are the Kanta Services? - Citizens - Kanta.fi. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.kanta.fi/en/what-are-kanta-services\u003c/span\u003e\u003cspan address=\"https://www.kanta.fi/en/what-are-kanta-services\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNordic interoperability. https://nordicinteroperability.com/.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHorgan, D. et al. European Health Data Space - An Opportunity Now to Grasp the Future of Data-Driven Healthcare. \u003cem\u003eHealthcare\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e, 1629 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThe European Health Data Space (EHDS). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.european-health-data-space.com/\u003c/span\u003e\u003cspan address=\"https://www.european-health-data-space.com/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e. Inclusion and exclusion criteria. LBP= low back pain, SBT= STarT Back Screening Tool, VAS= Visual Analog Scale.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 255px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eInclusion criteria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eExclusion criteria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 255px;\"\u003e\n \u003cp\u003eAged 18 to 65 years\u003c/p\u003e\n \u003cp\u003eLBP symptoms \u0026ge; 3 months\u003c/p\u003e\n \u003cp\u003eSBT questionnaire fulfilled\u003c/p\u003e\n \u003cp\u003ePain chart fulfilled\u003c/p\u003e\n \u003cp\u003eSocial security number available\u003c/p\u003e\n \u003cp\u003eVAS \u0026ge; 3\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 331px;\"\u003e\n \u003cp\u003eMalignancy\u003c/p\u003e\n \u003cp\u003eRecent traumatic fracture to the pain region\u003c/p\u003e\n \u003cp\u003eOsteoporotic fracture\u003c/p\u003e\n \u003cp\u003eInfection (i.e., epidural abscess)\u003c/p\u003e\n \u003cp\u003eAnkylosing spondylitis\u003c/p\u003e\n \u003cp\u003eModic 1 changes\u003c/p\u003e\n \u003cp\u003eUnstable spondylolisthesis\u003c/p\u003e\n \u003cp\u003eAnomaly of the bone in the pain region\u003c/p\u003e\n \u003cp\u003eSevere scoliosis (\u0026gt;45\u0026deg;)\u003c/p\u003e\n \u003cp\u003eA nerve root disorder with apparent dermatomal and/or myotomal radiculopathy (pain, numbness, paresthesia, tingling, muscle weakness)\u003c/p\u003e\n \u003cp\u003eAny other obvious specific reason for LBP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u003c/strong\u003e Characteristics of the whole study population and population randomized to analysis. BMI= Body Mass Index, LBP= Low back pain, NSAID= non-steroid anti-inflammatory drug, VAS= Visual Analog Scale, SBT= STarT Back Tool. SBT Q3= I have walked only short distances because of my back pain, Q4=In the last two weeks, I have dressed more slowly than usual because of my back pain.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eVariable\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003ePopulation (n=93)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003ePopulation randomized to analysis (n=20)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eMale (n/%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e30/32%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e9/45%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eAge (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e45 years (95% CI \u0026plusmn;2 years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e43 years (95% CI \u0026plusmn;5 years)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eBMI (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e28.3 (95% CI \u0026plusmn; 2.7)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e28.1 (95% CI \u0026plusmn; 1.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eDuration of LBP (n/%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e3-6 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e6/6% \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e1/5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e6-12 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e14/15%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4/20%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e1-2 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e15/16% \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e7/35%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e2-5 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e17/18%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4/20%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e5-10 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e8/9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e1/5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026gt;10 years\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e33/36%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e3/15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eOn pain medication (n/%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e86/92%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e19/95%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eNSAID\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e69/74%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e16/80%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eParacetamol\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e42/45%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e8/40%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eOpiate\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e30/32%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e5/25%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eNeuropathic pain medication\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e25/27%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e5/25%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eVAS in motion (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e6.3 (95% CI \u0026plusmn;0.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e6.3 (CI \u0026plusmn;1.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eVAS in rest (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e5.5 (95% CI \u0026plusmn;0.5)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e5.6 (CI \u0026plusmn; 0.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eSBT score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003etotal score (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e7 (95% CI \u0026plusmn; 0.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e6 (CI \u0026plusmn; 0.05)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003esub score Q5-9 (mean)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4 (95% CI \u0026plusmn;0.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4 (CI \u0026plusmn; 0.005)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eYes on Q3\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e64/69%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e12/60%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eYes on Q4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e51/55%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e10/50%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eOn sick leave due to LBP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e61/66%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e10/50%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eless than 30 days\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e11/18%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e3/15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e1-3 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e24/39%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e3/15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e4-6 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e5/8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e1/5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eover 6 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e17/28%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e3/15%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4/7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003e\u0026ldquo;I can work in the same profession in 2 years\u0026rsquo; time despite my health\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eMost definitely\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e13/14%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4/20%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eI\u0026rsquo;m not sure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e42/45%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e10/50%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eProbably not\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e31/33%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e4/20%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eN/A\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e7/8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e2/10%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eHas had physiotherapy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e76/82%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e14/70%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eHas been in institutional rehabilitation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e15/16%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e1/5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 264px;\"\u003e\n \u003cp\u003eHas imaging studies done\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e83/89%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 180px;\"\u003e\n \u003cp\u003e17/85%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u003c/strong\u003e The quantitative findings of ICF chapters ordered in the ICF components in the EHR of the evaluation dataset (20 patients). The number of findings is presented in brackets, and the percentage calculated from all the findings (n=3601). Chapter domains with only one or no findings are excluded.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBody\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 319px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eFunction, B (n=896/25%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eStructure, S (n=1444/40%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"5\" valign=\"top\" style=\"width: 319px;\"\u003e\n \u003cp\u003eSensory functions and pain, b2 (366)\u003c/p\u003e\n \u003cp\u003eNeuromusculoskeletal and movement related functions, b7 (349)\u003c/p\u003e\n \u003cp\u003eMental functions, b1 (99)\u003c/p\u003e\n \u003cp\u003eFunctions of the digestive, metabolic and endocrine systems, b5 (62)\u003c/p\u003e\n \u003cp\u003eGenitourinary and reproductive functions, b6 (12)\u003c/p\u003e\n \u003cp\u003eFunctions of the cardiovascular, hematological, immunological, and respiratory systems, b4 (8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003eStructures related to movement, s7 (1364)\u003c/p\u003e\n \u003cp\u003eStructures of the nervous system, s1 (73)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003eStructures related to the digestive, metabolic and endocrine systems, s5 (7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 323px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eActivities and participation, D (n=569/16%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eMobility, d4 (310)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eCommunity, social, and civic life, d9 (103)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eMajor life areas, d8 (89)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eSelf-care, d5 (44)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eDomestic life, d6 (16)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eInterpersonal interactions and relationships, d7 (7)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEnvironmental factors, E (n=692/19%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 642px;\"\u003e\n \u003cp\u003eServices, systems, and policies, e5 (298)\u003c/p\u003e\n \u003cp\u003eProducts and technology, e1 (253)\u003c/p\u003e\n \u003cp\u003eSupport and relationship, e3 (137)\u003c/p\u003e\n \u003cp\u003eNatural environment and human-made changes to environment, e2 (4)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4.\u003c/strong\u003e Results of the factor recognition. S= body structures, B= body functions, D= activities and participation, E= environmental factors.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 80px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eS\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"top\" style=\"width: 105px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eB\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eE\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003eExpert found\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e423\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e311\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e226\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e112\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1072\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003eHeadai Graphmind found\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003ecodes in total\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e371\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e312\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e208\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e991\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003etrue positives\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e368\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e285\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e195\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e942\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003efalse positives\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e49\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003efalse negatives\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e55\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e191\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003ecodes better than expert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e31\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003eCorrect codes in total\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e427\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e325\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e238\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e113\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1103\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003ePrecision \u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.99\u003c/p\u003e\n \u003cp\u003e(0.957- 1.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003cp\u003e(0.879- 0.941)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003cp\u003e(0.901-0.979)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003cp\u003e(0.917-0.963)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.95\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(0.939-0.961)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 7px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003eRecall\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003cp\u003e(0.817-0.883)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003cp\u003e(0.809- 0.871)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.78\u003c/p\u003e\n \u003cp\u003e(0.735-0.825)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003cp\u003e(0.795-0.865)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.83\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(0.823-0.837)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 159px;\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003cp\u003e(95% CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 90px;\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003cp\u003e(0.887-0.953)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 86px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003cp\u003e(0.849- 0.911)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003cp\u003e(0.79-0.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003cp\u003e(0.859- 0.901)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.89\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003e(0.884- 0.896)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"artificial intelligence, functioning, graph machine learning, ICF, low back pain","lastPublishedDoi":"10.21203/rs.3.rs-5415974/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5415974/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAs a comprehensive perspective on functioning is useful when making patient assessments, the WHO has developed the International Classification for Functioning, Disability, and Health (ICF). However, its complex structure poses a problem for implementation as part of clinical practice.The aim of this study was to test a graph machine learning engine, Headai Graphmind, to recognize ICF codes from electronic health records written in Finnish. A dataset of 93 patients aged 18 to 65 years with chronic low back pain was collected. Headai Graphmind was then tested for its ability to match free text with ICF codes on a sample of 20 patients. The results were compared against the findings of a domain expert. Headai Graphmind achieved 0.95 precision, 0.83 recall, and 0.89 F1 score.The application found 112 distinct ICF codes compared to 119 codes found by the domain expert. Headai Graphmind has the capability to recognize ICF codes from the electronic health records of patients with chronic low back pain. The method could be helpful when implementing the ICF classification in clinical practice, and enable retrospective coding of medical information for further use.\u003c/p\u003e","manuscriptTitle":"Using graph machine learning to identify functioning in patients with low back pain within the ICF framework","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-04-17 05:31:27","doi":"10.21203/rs.3.rs-5415974/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-05-07T12:09:55+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-24T15:52:39+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"171816688496009277652591008112530414952","date":"2025-04-16T08:58:22+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-15T19:33:48+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"9253552467208759272006238727281961299","date":"2025-04-14T08:24:12+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-14T05:26:28+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-07T08:23:52+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-03-22T09:42:51+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d5c76b95-6c30-4d0c-a8a5-479c86607839","owner":[],"postedDate":"April 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":47114271,"name":"Health sciences/Health care"},{"id":47114272,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2025-07-07T16:17:56+00:00","versionOfRecord":{"articleIdentity":"rs-5415974","link":"https://doi.org/10.1038/s41598-025-06429-4","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-07-02 15:57:26","publishedOnDateReadable":"July 2nd, 2025"},"versionCreatedAt":"2025-04-17 05:31:27","video":"","vorDoi":"10.1038/s41598-025-06429-4","vorDoiUrl":"https://doi.org/10.1038/s41598-025-06429-4","workflowStages":[]},"version":"v1","identity":"rs-5415974","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5415974","identity":"rs-5415974","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}