Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Open Access

Study Protocol

Evidence- and data-driven classification of low back pain via artificial intelligence: Protocol of the PREDICT-LBP study

  • Daniel L. Belavy ,

    Roles Conceptualization, Funding acquisition, Methodology, Resources, Writing – original draft

    daniel.belavy@hs-gesundheit.de, belavy@gmail.com

    Affiliation Division of Physiotherapy, Department of Applied Health Sciences, Hochschule für Gesundheit (University of Applied Sciences), Bochum, Germany

    ⨯
  • Scott D. Tagliaferri,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – original draft

    Affiliation Institute for Physical Activity and Nutrition, School of Exercise and Nutrition Sciences, Deakin University, Geelong, Victoria, Australia

    ⨯
  • Martin Tegenthoff,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Department of Neurology, BG-University Hospital Bergmannsheil gGmbH, Ruhr-University Bochum, Bochum, Germany

    ⨯
  • Elena Enax-Krumova,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Department of Neurology, BG-University Hospital Bergmannsheil gGmbH, Ruhr-University Bochum, Bochum, Germany

    ⨯
  • Lara Schlaffke,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Department of Neurology, BG-University Hospital Bergmannsheil gGmbH, Ruhr-University Bochum, Bochum, Germany

    ⨯
  • Björn Bühring,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Internistische Rheumatologie, Krankenhaus St. Josef Wuppertal, Wuppertal, Germany

    ⨯
  • Tobias L. Schulte,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Department of Orthopaedics and Trauma Surgery, St. Josef-Hospital Bochum, Ruhr University Bochum, Bochum, Germany

    ⨯
  • Sein Schmidt,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Berlin Institute of Health, Charité Universitätsmedizin Berlin, Berlin, Germany

    ⨯
  • Hans-Joachim Wilke,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Institute of Orthopaedic Research and Biomechanics, Trauma Research Center Ulm, University Hospital Ulm, Ulm, Germany

    ⨯
  • Maia Angelova,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation School of Information Technology, Deakin University, Geelong, Australia

    ⨯
  • Guy Trudel,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Clinical Epidemiology Program, Ottawa Hospital Research Institute, Ottawa, Canada

    ⨯
  • Katja Ehrenbrusthoff,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Division of Physiotherapy, Department of Applied Health Sciences, Hochschule für Gesundheit (University of Applied Sciences), Bochum, Germany

    ⨯
  • Bernadette Fitzgibbon,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliations Monarch Research Institute, Monarch Mental Health Group, Melbourne, Australia, School of Psychology and Medicine, Australian National University, Canberra, Australia, Department of Psychiatry, Monash University, Melbourne, Australia

    ⨯
  • Jessica Van Oosterwijck,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Faculty of Medicine and Health Sciences, Ghent University, Ghent, Belgium

    ⨯
  • Clint T. Miller,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Institute for Physical Activity and Nutrition, School of Exercise and Nutrition Sciences, Deakin University, Geelong, Victoria, Australia

    ⨯
  • Patrick J. Owen,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliation Institute for Physical Activity and Nutrition, School of Exercise and Nutrition Sciences, Deakin University, Geelong, Victoria, Australia

    ⨯
  • Steven Bowe,

    Roles Conceptualization, Funding acquisition, Methodology, Writing – review & editing

    Affiliations Faculty of Health, Deakin University, Geelong, Australia, Te Kura Tātai Hauora-The School of Health, Victoria University of Wellington, Wellington, New Zealand

    ⨯
  • Rebekka Döding,

    Roles Writing – original draft, Writing – review & editing

    Affiliation Division of Physiotherapy, Department of Applied Health Sciences, Hochschule für Gesundheit (University of Applied Sciences), Bochum, Germany

    ⨯
  •  [ ... ],
  • Svenja Kaczorowski

    Roles Writing – original draft, Writing – review & editing

    Affiliation Division of Physiotherapy, Department of Applied Health Sciences, Hochschule für Gesundheit (University of Applied Sciences), Bochum, Germany

    ⨯
  • [ view all ]
  • [ view less ]

Abstract

In patients presenting with low back pain (LBP), once specific causes are excluded (fracture, infection, inflammatory arthritis, cancer, cauda equina and radiculopathy) many clinicians pose a diagnosis of non-specific LBP. Accordingly, current management of non-specific LBP is generic. There is a need for a classification of non-specific LBP that is both data- and evidence-based assessing multi-dimensional pain-related factors in a large sample size. The “PRedictive Evidence Driven Intelligent Classification Tool for Low Back Pain” (PREDICT-LBP) project is a prospective cross-sectional study which will compare 300 women and men with non-specific LBP (aged 18–55 years) with 100 matched referents without a history of LBP. Participants will be recruited from the general public and local medical facilities. Data will be collected on spinal tissue (intervertebral disc composition and morphology, vertebral fat fraction and paraspinal muscle size and composition via magnetic resonance imaging [MRI]), central nervous system adaptation (pain thresholds, temporal summation of pain, brain resting state functional connectivity, structural connectivity and regional volumes via MRI), psychosocial factors (e.g. depression, anxiety) and other musculoskeletal pain symptoms. Dimensionality reduction, cluster validation and fuzzy c-means clustering methods, classification models, and relevant sensitivity analyses, will classify non-specific LBP patients into sub-groups. This project represents a first personalised diagnostic approach to non-specific LBP, with potential for widespread uptake in clinical practice. This project will provide evidence to support clinical trials assessing specific treatments approaches for potential subgroups of patients with non-specific LBP. The classification tool may lead to better patient outcomes and reduction in economic costs.

Citation: Belavy DL, Tagliaferri SD, Tegenthoff M, Enax-Krumova E, Schlaffke L, Bühring B, et al. (2023) Evidence- and data-driven classification of low back pain via artificial intelligence: Protocol of the PREDICT-LBP study. PLoS ONE 18(8): e0282346. https://doi.org/10.1371/journal.pone.0282346

Editor: James Mockridge, PLOS: Public Library of Science, UNITED KINGDOM

Received: December 2, 2022; Accepted: February 10, 2023; Published: August 21, 2023

Copyright: © 2023 Belavy et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Study protocol, no data reported.

Funding: German AeroSpace Center (DLR; www.dlr.de) grant numbers 50WK2273A (to DLB), 50WK2273B (to EE-K), 50WK2273C (to H-JW). The funders had no and will not have a role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: DLB, ST, MT, LS, SeS, HJW, MA, GT, KE, BF, JVO, CTM, PJO, SB, RD, SK declare no conflicts of interest. BB: declares no conflicts of interest relevant to the current work, but has previously received research support, consultancy fees and/or honoraria from AbbVie, Amgen, Biogen, GE/Lunar, Janssen, Galapagos, Gilead, Medimaps, MSD, Sanofi Genzyme, Theramex, UCB. TS: declares prior consulting fees from Johnson & Johnson, Amgen, Kaia Health Software, SpineArt, Implantcast and speaking and travel arrangements from Johnson & Johnson, ICOTEC, Nuvasive, Ulrichmedical, Silony, Amgen, Kaia Health Software. EE-K has received fees from painCert GmbH, Casquar GmbH and Omega Pharma GmbH, outside the submitted work. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Introduction

Back pain is the greatest cause of disability and lost productivity [1] and affects approximately 540 million people worldwide [2]. Reducing the burden of disease associated with low back pain (LBP) is a public health priority. Approximately 90% of LBP have been classified as non-specific [3]. A debate exists as to what is considered non-specific LBP [3–5]. Most authors agree that non-degenerative spinal diseases (e.g.; fracture, cancer, infection and inflammatory spondyloarthropathy) and damage to the spinal nervous system (e.g.; cauda equina, spinal stenosis and radiculopathy) are specific causes of LBP [3–5]. However, degenerative spinal conditions (e.g.; degenerative disc disease, facet arthropathy, spondylolysis) are considered specific by some authors [5]. Others argue they are non-specific [3, 4] based on evidence that clinical tests cannot reliably attribute pain to specific structures and that degenerative spine conditions can be present in people with or without LBP [6–9].

Recommendations for the treatment of non-specific LBP are by necessity generic when it is unclear what is contributing to the pain in each patient. These include exercise, manual therapy, psychological interventions, medication and multidisciplinary interventions [10]. The efficacy of these treatments on non-specific LBP often do not reach the threshold of clinical significance in many patients [3]. Clinical classification approaches exist for the conservative management of LBP, such as McKenzie treatment and STarT Back, but their impact is small and effect size not clinically meaningful versus more generic treatments [11]. A better clinical classification of LBP is a prerequisite to develop targeted treatment approaches.

Multiple domains impact a person experience of pain [12, 13]. In non-specific LBP, damage to the spinal tissues (nociceptive pain drivers), alterations in the processing of the pain signal (nervous system dysfunction), presence of other musculoskeletal diseases (comorbidity drivers), negative mood/affect, poor self-efficacy, and the use of maladaptive coping strategies (cognitive-emotional drivers) can influence the pain experience [6–9, 13]. Further, abnormal central pain processing including central sensitisation [14, 15], reorganisation of (sub)cortical brain pathways [16], and changes in brain connectivity [17–19] can contribute to pain. Beliefs, fear of movement, depressive symptoms, anxiety, past experiences, self-efficacy, presence of other painful conditions may mediate the non-specific LBP experience [20, 21].

Of the known factors which may contribute to an individual’s experience of non-specific LBP, there is not a single one that can differentiate between those who have non-specific LBP and those who are pain free [22]. In a recent systematic review with multivariate meta-analysis [23] we showed that multiple domains are impacted in LBP and that there was only one primary study [24], of small sample size (n = 32) which considered more than two domains of central nervous system adaptation, tissue changes and/or psychological factors.

Artificial intelligence (AI) and machine learning offer automated methods to detect unique sub-groups that ‘explain’ the presence and characteristics of a disease [25, 26]. AI techniques have been used in other diseases (e.g. severe sepsis [27], cardiovascular disease [28], diabetes [29]) to improve clinical outcomes. A systematic review of AI applications in LBP [30] noted that (a) no studies to date considered if possible sub-groups exist, (b) a limited number of body systems were evaluated and (c) sample sizes remained small (ranging from 24 to 171 participants in these studies). There is an evidence gap for a study to use data-science techniques in LBP in an appropriate sample size. Since this review, work from our group [31], using data from the UK Biobank, showed that sub-groups could be derived in (chronic) LBP on the basis of psychosocial variables and then accurately classified, but the data set lacked spinal tissue and pain sensitisation measures and key clinical data (e.g. pain intensity, disability and pain duration).

The key path forward is to prospectively collect data from multiple domains known to contribute to the experience LBP in both a LBP and pain-free collective. The application of AI and data science techniques can then be used to examine for the presence of sub-groups. The aim of the “PRedictive Evidence Driven Intelligent Classification Tool for Low Back Pain” (PREDICT-LBP) project is to develop a prospective, evidence- and data-driven tool for determining sub-groups of patients under the common label of non-specific LBP using clinical measures and machine learning algorithms.

Materials and methods

Study design

This prospective cross-sectional study will compare people with non-specific LBP, as defined in prior work [3], with age-, sex- and BMI-matched referents without a history of LBP. The study was pre-registered at the German Clinical Trial Register (DRKS00030518). The protocol is in accordance with the Helsinki Declaration of ethical principles [32] and has been approved by the local ethical committee (Ethik-Kommission der Hochschule für Gesundheit; reference number 210827_Belavy). Written informed consent will be obtained from all participants. The STROBE guideline [33] and TRIPOD statement [34] for prediction model development were used to report the sections in this protocol (S1 and S2 Tables respectively). We will update our reporting to the TRIPOD-artificial intelligence statement [35] if it is released during our project.

Setting

Participant recruitment will start in the beginning of 2023 and continue for 18 months. It will apply the use of a variety of recruitment procedures (i.e. postal flyers, notice boards, internet approaches, social media) at the Hochschule für Gesundheit Bochum [University of Applied Sciences], in the general public (i.e. local companies, sports centres, ambulatory medical practices), local hospitals (Uniklinik Bergmannsheil Bochum, St. Josef-Hospital Uniklinik Bochum, St. Josef-Hospital Wuppertal) and general practitioners as well as specialist outpatient clinics in the community. Expanded recruitment strategies are effective in obtaining a more balanced sample and increasing the external validity of study results [36, 37]. Recruitment will be stratified equally in each of four age groups (18-25yr, 26-35yr, 36-45yr, 46-55yr), aiming for a 55:45 ratio of women to men due to higher prevalence of LBP in women [2]. Furthermore, the prevalence of acute versus chronic back pain in the wider community is unclear with wide confidence intervals [38]. Cohort studies [39] can inform on prognosis of people with back pain but not the proportion of the population suffering acute versus chronic back pain at any given time. We will therefore pragmatically recruit participants with LBP to target a median duration of pain in the final collective of between three and five months. If eligible and participants consent to participate, study procedures will be completed within six weeks of the date of consent; as some measures may reflect not only traits but also states. Participants will receive a payment of 100 EUR upon completion of measures.

Data collection will occur at the Hochschule für Gesundheit Bochum [University of Applied Sciences] (Tactile spatial acuity assessment, pressure pain thresholds, temporal summation of pain, laser evoked potentials, questionnaires, trunk muscle strength and endurance) and at the Uniklinik Bergmannsheil Bochum (magnetic resonance imaging [MRI] scanning). Personal data of participants will be pseudonymized between the two settings and transmitted via phone call or email.

Participants

In defining the inclusion and exclusion criteria the following underlying principles will be used: firstly, people with specific causes of LBP will be excluded: a) non-degenerative spinal diseases (e.g.; fracture, cancer, infection and inflammatory spondyloarthropathy) and b) damage to the spinal nervous system (e.g.; cauda equina, spinal stenosis and radiculopathy). Secondly, we consider the debate about whether degenerative spinal conditions constitute specific or non-specific LBP as clinical equipoise justifying the need for further research. In this study, we will consider degenerative spinal conditions as non-specific while some severe deformity will be considered specific and excluded. Thirdly, as our aim is to develop a clinical categorisation approach for non-specific LBP that could be of utility in clinical practice, we aim to recruit a collective representative of LBP in the wider population. Fourthly, participants enrolled in the matched control group should have similar characteristics as people with LBP, only without LBP.

Inclusion criteria for both LBP and matched controls: adults aged 18 to 55 years, and being fully conversant in German or English. The age bracket 18 to 55 will be targeted as first incidence of LBP typically occurs before or during mid-life and the peak burden of disease is in the age bracket below 55yrs [2].

Inclusion criterion for LBP: pain localised below the costal margin and above the inferior gluteal folds, with or without leg pain [40].

Exclusion criteria for both LBP and matched controls will be:

  • Known clinical diagnosis of non-degenerative specific spinal conditions: cancer, vertebral fracture, vertebral or intervertebral disc infection or inflammatory spondyloarthropathy.
  • Known clinical diagnosis of spinal nervous system damage: cauda equina, spinal stenosis [41] or diagnosed radiculopathy [42] as defined in prior work. Leg/radiating pain is not an exclusion criterion, nor is spinal stenosis or foraminal stenosis on imaging with no neurological damage.
  • Known clinical diagnosis of severe degenerative spinal conditions: grade IV spondylolisthesis, severe deformity (e.g. hemivertebrae, scoliosis > 50 degrees) requiring more than two prior medical specialist consultations.
  • Previous spine surgery (being on a waiting list for surgery is, of itself, not an exclusion).
  • Prior major head trauma or brain surgery.
  • Known clinical diagnoses of the following neurological conditions: cerebral palsy, dementia, epilepsy, motor neurone disease, multiple sclerosis, Parkinson’s disease, spinal muscular atrophy, stroke, polyneuropathy.
  • Known major psychiatric disorder: mania, bipolar affective disorder, major depressive disorder, severe personality disorder, severe anxiety disorder, severe obsessive-compulsive disorder [43] as these may impact brain MRI measures [44].
  • Current pregnancy or up to 6 months post-pregnancy (pelvic girdle pain associated with pregnancy typically resolves in ~80% of women by 6 months post-partum [45, 46]).
  • Any further absolute contraindications for brain or spine MRI (e.g. cardiac pacemakers, metal implants and foreign bodies, middle ear implants, drug pumps, electronic stimulation devices).

For the pain-free control group, the following additional exclusion criteria will be applied:

  • current spinal (neck, upper back, or low back) pain impacting function (self-report),
  • LBP lasting for more than 24 hours within the last year (except for muscle soreness related to physical activity or exercise),
  • previously, at any point, missed days from work due to LBP,
  • previously, at any time, consulted a health professional for the treatment of LBP (e.g. physiotherapist and general practitioner).

Pain-free controls will be matched to LBP participants on a 1:3 basis (see Sample Size section for rationale) by

  • sex (male or female)
  • age brackets (18-25yr, 26–3, 36-45yr, 46-55yr)
  • height (±5 cm; as body height has an impact on spine morphology)
  • additionally, where possible, a body mass index of ±5 kg/m2 will be preferred.

Method of ascertainment: a screening tool will be used to assess the eligibility of the participants for the study. The screening tool will be implemented by study staff (e.g. study nurse, project manager) as part of verbal screening of the potential candidates for participation.

Quantitative variables and data collection

Each participant will undergo an integrated test battery, which will assess nervous system adaptations, cognitive-emotional factors, co-morbid musculoskeletal disease, and standard outcome measures to describe the collective. Table 1 presents variables that flow into data-analytics versus those that are considered in secondary analyses only as it is less clear whether these latter variables can be “causal” for LBP.

thumbnail
Download:
Table 1. Variables to be assessed in analytical methods.

https://doi.org/10.1371/journal.pone.0282346.t001

Brain MRI.

A 3T MRI scanner with a 32 channel receive-only phased-array head coil will be used. As in preliminary work [31, 47], high resolution magnetisation prepared rapid acquisition gradient echo (MP2RAGE) T1-weighted structural data will be acquired to assess grey matter volumes. Resting state functional connectivity in pain networks will be assessed using an EPI sequence and those data will be pre-processed, denoised and analysed using standardized pipelines. Fieldmaps will also be attained in the scanning pipeline to allow for resting state MRI distortion correction. Diffusion weighted images will be obtained using a EPI-sequences with 60 isotropic oriented gradient orientations and various b-values. These data will be processed to assess regional white matter alterations in pain related brain areas. Furthermore, tractography will be performed to analyse structural connectivity within pain networks and to correlate the structural connectivity to functional connectivity obtained by resting state fMRI.

Contention remains regarding the most important functional networks and brain regions that contribute to non-specific LBP. Therefore, we will assess a wide range of variables (Table 1) and prior to deciding which variables that go to further data processing, dimensionality reduction will performed (see Analytical Methods point “A”).

Spine MRI.

Intervertebral disc composition [48], vertebral fat fraction [49] and paraspinal muscle size and composition [50] will be assessed using a 3T MRI scanner with the following sequences: 1) T2 sag spin-echo multi-echo, 2) Sag DIXON, 3) Axial DIXON, 4) Axial PD and 5) T2-weighted Sag. Radiological assessment will also be performed for the presence of radiological evidence of degenerative changes: facet joint osteoarthritis, pars defects and spondylolisthesis, disc herniation, spinal stenosis, osteophytes, Modic changes and disc degeneration (Pfirrmann grade ≥3; [51]). These outcomes are known to be affected in non-specific LBP [52–54].

Tactile spatial acuity assessment.

Following the MRI examinations, the two-point discrimination threshold (2PDT) at the lumbar region is to be determined with each test person according to established procedures [55, 56]. In this examination, a test instrument with seven tips separated by various defined distances, or a test instrument with only a single tip, is applied to the subject’s skin for approximately 1 second. The test person is instructed to say whether he or she feels "one" or "two" measuring instruments. Each test instrument is tested 8 times in random order. The 2PDT is defined as the distance of the peaks at which participants answered "two" 50% of the time.

Pressure pain thresho1lds.

Pressure-pain thresholds (PPT) will be assessed using a digital algometer at the lumbar paraspinals (local hyperalgesia) and at the elbow and calf (for widespread hyperalgesia) using standard protocols [57, 58]. Changes in PPTs indicates local and/or widespread pain sensitivity in patients with non-specific LBP [59]. Manual pressure will be applied using a 10mm‐diameter rod at a rate of 10 N/cm2. Subjects will be instructed to say ‘pain’ or ‘stop’ at the point pressure turns to pain. The protocol will consist of two trials at each site (left, right, left, right) with 20s of rest between each trial. Order for calves, arms, back will be randomised. The average data of the two trials at each site will be recorded.

Temporal summation of pain.

Temporal summation of pain (TS-Pain) [60] will be assessed as per standard protocol [61, 62]. Ten consecutive pulses of the algometer will be applied to the same locations as PPT testing. Pressure of each pulse will be increased at a rate of 2kg/s and a frequency of 1 Hz, until reaching the previously determined PPT and maintained for one second. Participants will rate their perceived pain intensity during the first, fifth and tenth pulse using a verbal numeric rating scale from zero to 10 [63]. TS-Pain is established when the verbal numeric rating scale scores increase from the first to the fifth, and from the fifth to the tenth pulse. The TS-Pain magnitude will be expressed as the first score subtracted from the last, with higher scores indicating more efficacious transmission of nociception to the brain.

Pain related evoked potentials (PREPs).

Pain related evoked potentials (PREPs) will be assessed using noxious laser heat stimulation of the pain-affected back and a pain free remote area (abdomen) [64, 65]. By means of a neodymium-doped-yttrium aluminium–perovskite (Nd:YAP) laser with a wavelength of 1,340 nm, a short thermal pulse (1–100 ms) will be applied to intact skin, ensuring synchronous activation of Aδ- and C-thermal nociceptors [65, 66]. To obtain reproducible cortical responses, it is important to apply a suprathreshold pain stimulus intensity that is usually less unpleasant than an electrical stimulus used in somatosensory evoked potential (SEP) recordings [65]. Determination of the stimulus intensity will follow standardized approaches reported elsewhere [67, 68]. The stimulus is often likened to a mild burning or stinging sensation but due to its deeper skin penetration with lower superficial skin heating, the Nd:YAP laser causes only marginally visible skin lesions [65]. Cortical responses will be recorded according to the international 10–20 system [69] over the vertex (Cz) and an earlobe reference [64, 70]. To ascertain the integrity of the complete somatosensory lemniscal conduction pathway, the measurement of tibial nerve SEPs will be added to the PREP protocol. As normative data for this sensory nerve of the lower limb already exist [71], data from the current study population will be comparable to already published data [72] With this additional measurement, potentially hidden abnormalities regarding the conduction velocity of the evoked potentials, which could potentially influence the results from the PREP protocol, could be detected.

Cognitive-emotional factors.

We will use the standardised and validated Patient-Reported Outcomes Measurement Information System (PROMIS) [73] questionnaires to measure anxiety, depression, cognitive function, self-efficacy, satisfaction social roles and activities, social isolation and social and instrumental support; all known to be impacted in non-specific LBP [74]. As a German version of the PROMIS questionnaire for emotional and instrumental support is not available, a validated German questionnaire [75] will be used for this outcome.

Additional musculoskeletal pain locations and cardiovascular risk factors.

The presence of other musculoskeletal disease is a known risk factor for non-specific LBP [76] and will be assessed using the Nordic Musculoskeletal Questionnaire [77] which asks for self-report presence of pain in different body regions. Known cardiovascular risk factors for disc herniation (known diabetes, hypertension, high cholesterol and parent who had suffered a myocardial infarction before age 60) will be captured as yes/no response on questionnaires [78].

Central sensitisation inventory questionnaire.

The Central Sensitisation Inventory questionnaire will assess central sensitisation [79] as a measure of central nervous system hypersensitivity. This tool may be a potential surrogate measure of more time and personnel-costly pain threshold and temporal summation of pain measures. A score of >40 on Part A of the inventory indicates central sensitisation.

Demographic and standard outcome measures.

We will collect age, occupation (categories per an established taxonomy [80]), weight, height, BMI, smoking history (per the DKFZ-Lymphomstudie [81]), pain duration, pain intensity (visual analogue scale), Oswestry Disability Index, short form of the international physical activity questionnaire [82], Tampa scale for kinesiophobia [83], fear avoidance belief questionnaire [84], current prescription medication, trunk muscle strength [85] and trunk muscle endurance [86]. These measures will characterize the population but not be included in sub-grouping.

Bias

Project staff conducting the participant testing will be blinded to group status. Should the volunteer reveal their pain status to the operators, the protocol deviation will be noted. Questionnaires will be completed via SoSci (https://www.soscisurvey.de/) and inaccessible to the testers. Data will be encoded by a unique, random, four-digit participant code to blind all steps of analysis.

Study size

For sub-grouping analyses using machine learning methods, the approach of Foreman [87, 88] to calculate the sample size is used: 2k (where k = number of variables) is minimum sample size and 5*2k is the most conservative sample size. The UK Biobank data study [31, 89] provided only a sub-set of the required parameters, thus the impact of the additional measures in our proposed study (e.g. spine MRI) on sub-grouping is unclear. In this preliminary work [31], less than seven variables were required to characterise a LBP collective, with two specific variables being most important for defining sub-groups. Thus, we argue that using up to 8 variables (28 = 256) across spinal tissue, nervous system and psychosocial domains will likely be sufficient to detect the largest and clinically relevant sub-groups. Therefore, we conservatively target a sample of 300 people with non-specific LBP and 100 controls. This will enable valid subgrouping/classification of patients with non-specific LBP with up to 8 variables in the primary analysis, with additional power for sub-domain analyses with less variables as required. The choice of a 3:1 ratio LBP-to-control was pragmatic for project feasibility and because the focus of our project is on sub-grouping in LBP rather than in pain-free controls.

Data handling and processing

To promote data quality, questionnaire data entered by the participant and data entered by study staff (e.g. body mass, endurance times) during testing will be collected via SoSci (https://www.soscisurvey.de/). Monthly back-ups of these data from SoSci will be performed during the data collection process onto the servers of the Hochschule für Gesundheit or the Hochschulnetz NRW (“sciebo”). Any events or protocol deviations during testing will be noted in free-text format in the same system. Where feasible, automatic range checking will be implemented during data entry onto SoSci and will be implemented as part of data processing. Where team staff members are not priorly experienced in the outcome measures, training will be provided by experienced team members and/or collaborators along with standard operating procedure documentation. As part of standard practice, MR image data is checked visually by the MR-operator and investigator team member during scanning and scans repeated once if, for example, movement or other artefacts impact the images collected. MR images are measured manually by trained staff using established protocols [48, 49, 90] to obtain the numerical outcomes of interest. The remaining numerical data (e.g. objective physical activity measures, strength, endurance, body mass) will be collected per established standard protocols. A tabulated list of the outcomes is contained in Table 1.

For the data-science approaches implemented, complete data is required. Therefore, missing data will be imputed. The exact method of data imputation will be determined in consultation with a biostatistician once the degree of missingness is known and whether data are “missing-at-random” or not. We anticipate the missingness will be small, as we are performing a cross-sectional study and not a test-retest design, such as in a randomised controlled trial where dropouts inevitably occur. Additional sensitivity analyses performed to assess impact on the findings.

Analytical and statistical methods

The following approach, based on our preliminary work [31, 47], will be used (see also Fig 1).

thumbnail
Download:
Fig 1. Approach to artificial intelligence tool implementation.

GOÄ: Gebührenordnung für Ärzte (German medical cost database).

https://doi.org/10.1371/journal.pone.0282346.g001

  1. Dimensionality reduction will be performed to identify the most relevant variables (of those listed; Table 1) for subgrouping and classification. Given known challenges with data-analytic variable weighting and multidimensional data (e.g. scaling each variable to 0–1) [31], our first step in dimensionality reduction will focus on traditional statistical methods including paired t-tests, analysis of covariance and multicollinearity assessment through correlation analysis [31]. Following this, pending the number of important variables (e.g. if more than 8 variables remain) we will consider variable weighting using a decision tree classifier and targeted projection pursuit with artificial neural networks [91]. The outputs will be used to select the most relevant variables distinguishing non-specific LBP from controls (2-class model).
  2. Following determination of the most important variables, pain-free controls will be removed from the data for the following clustering steps. Cluster validation methods [92], including Calinski-Harabasz, Davies-Bouldin and Sillouette values, will be used to determine the optimal number of clusters and variables for the remaining non-specific LBP data. Given the potential for high dimensionality introducing noise and influencing clustering, we will evaluate the outputs of clustering using all outcomes and within individual domains (e.g. in preliminary work [31, 47], psychological and brain MRI variables correlated more highly with each other within their domains than between variable domains). The output of this step will allow us to determine the optimal number of clusters and variables.
  3. Based on the information in step B, we will apply fuzzy c-means clustering model [31] of non-specific LBP on either the primary clustering or domain specific clustering. The clusters derived in this step will be used to label classes of non-specific LBP.
  4. Outputs from steps B and C in terms of class labels will be used to evaluate non-specific LBP classification. We will apply Support Vector Machine, Naïve Bayes, k-Nearest Neighbour and Random Forest binary/multiclass classifiers as appropriate for the data. Given we are focusing on internal (i.e. on the current dataset, rather than a second newly collected dataset from a different research centre or project) validation of a classification model, validation of the model will be done via k-fold cross validation, as random data-splitting is not generally recommended for internal model validation [93–96]. The output of this will be a sub-grouping tool for people with non-specific LBP assuming availability of all data and tests and no consideration of test cost.
  5. We will consider the following sensitivity analyses:
  1. we will evaluate random data-splitting and repeat steps B and C to determine if the clustering outputs hold across random samples;
  2. we will apply steps B to E within each domain (e.g. brain MRI, psychosocial variables, spine MRI) as required;
  3. repetition of analysis only with participants without any imputed data; to examine the potential impact of missing data.
  4. consideration of impact of secondary variables (see bottom panel of Table 1) on sub-grouping.
  5. the methods of step (D) will be reapplied under constrained cost assumptions, considering financial-cost, time, and patient burden of obtaining each clinical outcome. Financial costs will be according to current rates according to the Gebührenordnung für Ärzte; time will be according to clinical measure and patient burden will be according to an ordinal scale. The output will be a model for most effective sub-group identification for people with non-specific LBP respecting cost/time/burden constraints.
  6. consider the findings of sub-grouping when patients with or without radiological evidence of degenerative changes (see Quantitative variables and data collection > Spine MRI) are included.

Confidentiality and data management

Data protection regulations for scientific investigations of the state of North Rhine-Westphalia (§28 DSG-NRW) and the General Data Protection Regulation (Art. 89 DSGVO) will be observed. During measurements with participants in the study, image (DICOM format) and tabulated numerical and string data will be generated (“study data”) in electronic form. Personal data (name, date of birth) will be recorded on a written informed consent form. Each participant will be assigned a random code (e.g. PREDICT34621). This code connects the consent form and the study data. We expect to that up a maximum of 30 GB of data will be generated in the study. Personal data on the consent form will not be saved with the data collected during the study. The consent form containing the personal data and the participant code will be stored separately from the study data at the Hochschule für Gesundheit and will be destroyed by the project management after the study has been completed. The study data is thus pseudonymised until the consent form is destroyed. After destroying the consent form, the study data are anonymized. Participation in the study is voluntary and participants can withdraw at any time and request their data be deleted. After anonymization, however, it is no longer possible to revoke participation since the data can then no longer be assigned to a single person. All study data will be stored electronically in a pseudonymised (later anonymous) form, protected by a password, on the servers of the Hochschule für Gesundheit, Universitätsklinikum Bergmannsheil and/or the Hochschulnetz NRW (“sciebo”). Data obtained at Bergmannsheil Bochum will be stored on site on local servers. In the interest of good scientific practice, the study data will be permanently stored in an anonymous form in online register databases after the study has been completed. Anonymized participant-level tabulated data will be made available permanently on a GDPR-conform data repository, such as www.osf.io and/or as supplemental data with published manuscripts. As a rule, anonymized participant-level tabulated data will be transferred to online register databases when the results are published in specialist journals, but at the earliest after data collection for the study has been completed. This will enable free re-use of the collected data. For MR (DICOM format) image data, access to the image data and re-use will be permitted upon written agreement with the Hochschule für Gesundheit.

Dissemination policy

Findings from the outcomes of this trial will be reported in journal articles and potentially PhD theses, which will include results regardless of the direction or magnitude of the effect. The results will also be presented at leading national and international conferences, clinical forums and to other relevant health professionals and stakeholders, as well as to the participants. All participants will be able to elect to receive a brief report of their study results via email at the completion of the study. Participants may also receive a copy of their MRI data. If the participant expresses interest in the study results, they can request to have a copy of published articles sent to them.

Consistent with our current practice, the ICJME authorship guidelines [97] will be followed in defining author and non-authors and author roles will be follow the CRediT taxonomy [98, 99]. We will publish the full protocol in a peer-reviewed journal. Anonymized participant-level data and statistical code will be made available on a data repository, such as www.osf.io and/or as supplemental data with published manuscripts.

Discussion

Based on our recent systematic review [23], our project will be the first larger scale study to assess multiple dimensions of pain contributors with a view to detecting clinically relevant sub-groups in non-specific LBP. We anticipate the production of a comprehensive, multidimensional dataset including imaging of the lumbar spine and the brain, assessment of pain processing, and emotional and psychosocial determinants of pain. This dataset will position us to test the hypothesis that patients with non-specific LBP can be grouped in clusters with common features.

Currently non-specific LBP is viewed as one large all-encompassing diagnostic entity. In our view, the results of the project, if sub-groups do indeed exist, will result in a substantial departure from the current paradigm by taking a fresh, rational perspective to optimise the screening of NSCLBP, which will guide more focused and tailored management. Importantly, should no relevant sub-groups be detected, this will also be a boon for current efforts, in that this will reinforce the need to implement current (generic) approaches, such as reducing medical imaging, for the management of non-specific LBP.

It is relevant to consider some potential limitations of the study. Firstly, we are conducting a prospective cross-sectional study. This means, that if sub-groups are detected, further prospective cohort and/or interventional studies will need to be conducted to, for example, examine the efficacy of interventions targeted at the individual sub-groups. Nonetheless, the current work is a necessary step along this path. Secondly, the necessary sample size to detect a given sub-group is always unknown. As described in the ‘study size’ section, our sample size will enable detection of sub-groups with up to six outcome per sub-group. We argue that 8 outcomes per sub-group of the outcomes collected will be sufficient to detect the largest clinically relevant sub-groups. Nonetheless, there may be smaller sub-groups for which more parameters are needed to detect them in non-specific LBP. However, we question whether an overall reduction in the collective burden of disease is possible by targeting such (small) sub-groups. Thirdly, whilst we aim to assess key known domains in the multidimensional contributors in LBP, we do not assess every known possible variable. We observed in our preliminary work [31, 47, 89] that parameters within a given domain often co-correlate. Further, there is no one single parameter that “explains” non-specific LBP. Therefore, we argue that additional parameters may have further explanatory power, but that it is unlikely that they would make a decisive different in the final clinical approach.

Acknowledgments

We thank our collaborators on prior preliminary work and participants in these prior studies that enabled and lead to the current project.

References

  1. 1. Vos T, Flaxman AD, Naghavi M, Lozano R, Michaud C, Ezzati M, et al. Years lived with disability (YLDs) for 1160 sequelae of 289 diseases and injuries 1990–2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet Lond Engl. 2012;380: 2163–2196. pmid:23245607
  2. 2. Hartvigsen J, Hancock MJ, Kongsted A, Louw Q, Ferreira ML, Genevay S, et al. What low back pain is and why we need to pay attention. The Lancet. 2018;391: 2356–2367. pmid:29573870
  3. 3. Maher C, Underwood M, Buchbinder R. Non-specific low back pain. Lancet Lond Engl. 2017;389: 736–747. pmid:27745712
  4. 4. Chou R, Qaseem A, Snow V, Casey D, Cross JT Jr, Shekelle P, et al. Diagnosis and treatment of low back pain: a joint clinical practice guideline from the American College of Physicians and the American Pain Society. Ann Intern Med. 2007;147: 478–91. pmid:17909209
  5. 5. Kreiner DS, Matz P, Bono CM, Cho CH, Easa JE, Ghiselli G, et al. Guideline summary review: an evidence-based clinical guideline for the diagnosis and treatment of low back pain. Spine J. 2020;20: 998–1024. pmid:32333996
  6. 6. Brinjikji W, Luetmer PH, Comstock B, Bresnahan BW, Chen LE, Deyo RA, et al. Systematic literature review of imaging features of spinal degeneration in asymptomatic populations. AJNR Am J Neuroradiol. 2015;36: 811–816. pmid:25430861
  7. 7. Brinjikji W, Diehn FE, Jarvik JG, Carr CM, Kallmes DF, Murad MH, et al. MRI Findings of Disc Degeneration are More Prevalent in Adults with Low Back Pain than in Asymptomatic Controls: A Systematic Review and Meta-Analysis. AJNR Am J Neuroradiol. 2015;36: 2394–2399. pmid:26359154
  8. 8. de Schepper EIT, Damen J, van Meurs JBJ, Ginai AZ, Popham M, Hofman A, et al. The association between lumbar disc degeneration and low back pain: the influence of age, gender, and individual radiographic features. Spine. 2010;35: 531–536. pmid:20147869
  9. 9. Cheung KM, Karppinen J, Chan D, Ho DW, Song YQ, Sham P, et al. Prevalence and pattern of lumbar magnetic resonance imaging changes in a population study of one thousand forty-three individuals. Spine Phila Pa 1976. 2009;34: 934–40. pmid:19532001
  10. 10. Almeida M, Saragiotto B, Richards B, Maher CG. Primary care management of non-specific low back pain: key messages from recent clinical guidelines. Med J Aust. 2018;208: 272–275. pmid:29614943
  11. 11. Tagliaferri SD, Mitchell UH, Saueressig T, Owen PJ, Miller CT, Belavy DL. Classification Approaches for Treating Low Back Pain Have Small Effects That Are Not Clinically Meaningful: A Systematic Review With Meta-analysis. J Orthop Sports Phys Ther. 2021; 1–49. pmid:34775831
  12. 12. Tagliaferri SD, Miller CT, Owen PJ, Mitchell UH, Brisby H, Fitzgibbon B, et al. Domains of Chronic Low Back Pain and Assessing Treatment Effectiveness: A Clinical Perspective. Pain Pract. 2019;20: 211–225. pmid:31610090
  13. 13. Tousignant-Laflamme Y, Martel MO, Joshi AB, Cook CE. Rehabilitation management of low back pain—it’s time to pull it all together! J Pain Res. 2017;10: 2373–2385. pmid:29042813
  14. 14. Meeus M, Roussel NA, Truijen S, Nijs J. Reduced pressure pain thresholds in response to exercise in chronic fatigue syndrome but not in chronic low back pain: an experimental study. J Rehabil Med. 2010;42: 884–890. pmid:20878051
  15. 15. Giesbrecht RJ, Battié MC. A comparison of pressure pain detection thresholds in people with chronic low back pain and volunteers without pain. Phys Ther. 2005;85: 1085–1092. pmid:16180957
  16. 16. Schmidt-Wilcke T. Neuroimaging of chronic pain. Best Pract Res Clin Rheumatol. 2015;29: 29–41. pmid:26266997
  17. 17. Pfannmöller J, Lotze M. Review on biomarkers in the resting-state networks of chronic pain patients. Brain Cogn. 2018 [cited 8 Mar 2019]. pmid:30879493
  18. 18. Kucyi A, Davis KD. The dynamic pain connectome. Trends Neurosci. 2015;38: 86–95. pmid:25541287
  19. 19. Mano H, Kotecha G, Leibnitz K, Matsubara T, Sprenger C, Nakae A, et al. Classification and characterisation of brain network changes in chronic back pain: A multicenter study. Wellcome Open Res. 2018;3: 19. pmid:29774244
  20. 20. Baliki MN, Apkarian AV. Nociception, Pain, Negative Moods, and Behavior Selection. Neuron. 2015;87: 474–491. pmid:26247858
  21. 21. Smart KM, Blake C, Staines A, Doody C. Self-reported pain severity, quality of life, disability, anxiety and depression in patients classified with “nociceptive”, “peripheral neuropathic” and “central sensitisation” pain. The discriminant validity of mechanisms-based classifications of low back (±leg) pain. Man Ther. 2012;17: 119–125. pmid:22074733
  22. 22. Rabey M, Smith A, Kent P, Beales D, Slater H, O’Sullivan P. Chronic low back pain is highly individualised: patterns of classification across three unidimensional subgrouping analyses. Scand J Pain. 2019. pmid:31256070
  23. 23. Tagliaferri SD, Ng S-K, Fitzgibbon BM, Owen PJ, Miller CT, Bowe SJ, et al. Relative contributions of the nervous system, spinal tissue and psychosocial health to non-specific low back pain: Multivariate meta-analysis. Eur J Pain Lond Engl. 2021. pmid:34748265
  24. 24. Derbyshire SWG, Jones AKP, Creed F, Starz T, Meltzer CC, Townsend DW, et al. Cerebral responses to noxious thermal stimulation in chronic low back pain patients and normal controls. NeuroImage. 2002;16: 158–168. pmid:11969326
  25. 25. Lee W, Alexeyenko A, Pernemalm M, Guegan J, Dessen P, Lazar V, et al. Identifying and Assessing Interesting Subgroups in a Heterogeneous Population. BioMed Res Int. 2015;2015: 462549. pmid:26339613
  26. 26. Lötsch J, Ultsch A. Machine learning in pain research. Pain. 2018;159: 623–630. pmid:29194126
  27. 27. Shimabukuro DW, Barton CW, Feldman MD, Mataraso SJ, Das R. Effect of a machine learning-based severe sepsis prediction algorithm on patient survival and hospital length of stay: a randomised clinical trial. BMJ Open Respir Res. 2017;4: e000234. pmid:29435343
  28. 28. Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N. Can machine-learning improve cardiovascular risk prediction using routine clinical data? PloS One. 2017;12: e0174944. pmid:28376093
  29. 29. Basu S, Sussman JB, Berkowitz SA, Hayward RA, Yudkin JS. Development and validation of Risk Equations for Complications Of type 2 Diabetes (RECODe) using individual participant data from randomised trials. Lancet Diabetes Endocrinol. 2017;5: 788–798. pmid:28803840
  30. 30. Tagliaferri SD, Angelova M, Zhao X, Owen PJ, Miller CT, Wilkin T, et al. Artificial intelligence to improve back pain outcomes and lessons learnt from clinical classification approaches: three systematic reviews. Npj Digit Med. 2020;3: 93. pmid:32665978
  31. 31. Tagliaferri SD, Wilkin T, Angelova M, Fitzgibbon BM, Owen PJ, Miller CT, et al. Chronic back pain sub-grouped via psychosocial, brain and physical factors using machine learning. Sci Rep. 2022;12: 15194. pmid:36071092
  32. 32. World Medical Association. World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA. 2013;310: 2191–2194. pmid:24141714
  33. 33. von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: Guidelines for reporting observational studies. Prev Med. 2007;45: 247–251. pmid:17950122
  34. 34. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Ann Intern Med. 2015;162: 55–63. pmid:25560714
  35. 35. Collins GS, Dhiman P, Andaur Navarro CL, Ma J, Hooft L, Reitsma JB, et al. Protocol for development of a reporting guideline (TRIPOD-AI) and risk of bias tool (PROBAST-AI) for diagnostic and prognostic prediction model studies based on artificial intelligence. BMJ Open. 2021;11: e048008. pmid:34244270
  36. 36. Nijs J, Inghelbrecht E, Daenen L, Hachimi-Idrissi S, Hens L, Willems B, et al. Recruitment bias in chronic pain research: whiplash as a model. Clin Rheumatol. 2011;30: 1481–1489. pmid:21853277
  37. 37. Nederhof E, Jörg F, Raven D, Veenstra R, Verhulst FC, Ormel J, et al. Benefits of extensive recruitment effort persist during follow-ups and are consistent across age group and survey method. The TRAILS study. BMC Med Res Methodol. 2012;12: 93. pmid:22747967
  38. 38. Hoy D, Bain C, Williams G, March L, Brooks P, Blyth F, et al. A systematic review of the global prevalence of low back pain. Arthritis Rheum. 2012;64: 2028–2037. pmid:22231424
  39. 39. Itz CJ, Geurts JW, van Kleef M, Nelemans P. Clinical course of non-specific low back pain: A systematic review of prospective cohort studies set in primary care: Clinical course of non-specific low back pain. Eur J Pain. 2013;17: 5–15. pmid:22641374
  40. 40. van Tulder M, Becker A, Bekkering T, Breen A, del Real MTG, Hutchinson A, et al. Chapter 3. European guidelines for the management of acute nonspecific low back pain in primary care. Eur Spine J. 2006;15 Suppl 2: S169-191. pmid:16550447
  41. 41. Azimi P, Mohammadi HR, Benzel EC, Shahzadi S, Azhari S. Lumbar Spinal Canal Stenosis Classification Criteria: A New Tool. Asian Spine J. 2015;9: 399–406. pmid:26097655
  42. 42. Bardin LD, King P, Maher CG. Diagnostic triage for low back pain: a practical approach for primary care. Med J Aust. 2017;206: 268–273. pmid:28359011
  43. 43. Gühne U, Weinmann S, Becker T, Riedel-Heller SG. [The S3 Guidelines Psychosocial Therapies in Severe Mental Illness—The Update]. Psychiatr Prax. 2018;45: 119–121. pmid:29618140
  44. 44. Opel N, Goltermann J, Hermesdorf M, Berger K, Baune BT, Dannlowski U. Cross-Disorder Analysis of Brain Structural Abnormalities in Six Major Psychiatric Disorders: A Secondary Analysis of Mega- and Meta-analytical Findings From the ENIGMA Consortium. Biol Psychiatry. 2020;88: 678–686. pmid:32646651
  45. 45. Tavares P, Barrett J, Hogg-Johnson S, Ho S, Corso M, Batley S, et al. Prevalence of Low Back Pain, Pelvic Girdle Pain, and Combination Pain in a Postpartum Ontario Population. J Obstet Gynaecol Can. 2020;42: 473–480. pmid:31864910
  46. 46. Bjelland E, Stuge B, Engdahl B, Eberhard-Gran M. The effect of emotional distress on persistent pelvic girdle pain after delivery: a longitudinal population study: Emotional distress and pelvic girdle pain. BJOG Int J Obstet Gynaecol. 2013;120: 32–40. pmid:23107369
  47. 47. Tagliaferri S, Owen P, Miller CT, Angelova M, Fitzgibbon B, Wilkin T, et al. [Preprint] Towards data-driven biopsychosocial classification of chronic low back pain: a pilot study. Open Science Framework; 2022.
  48. 48. Belavý DL, Quittner MJ, Ridgers N, Ling Y, Connell D, Rantalainen T. Running exercise strengthens the intervertebral disc. Sci Rep. 2017;7: 45975. pmid:28422125
  49. 49. Belavy DL, Quittner MJ, Ridgers N, Shiekh A, Rantalainen T, Trudel G. Specific modulation of vertebral marrow adipose tissue by physical activity. J Bone Miner Res. 2018;33: 651–657. pmid:29336053
  50. 50. Belavy DL, Quittner M, Ridgers ND, Ling Y, Connell D, Trudel G, et al. Beneficial intervertebral disc and muscle adaptations in high-volume road cyclists. Med Sci Sports Exerc. 2019;51: 211–217. pmid:30157104
  51. 51. Pfirrmann CW, Metzdorf A, Zanetti M, Hodler J, Boos N. Magnetic resonance classification of lumbar intervertebral disc degeneration. Spine. 2001;26: 1873–8. pmid:11568697
  52. 52. Brinjikji W, Diehn FE, Jarvik JG, Carr CM, Kallmes DF, Murad MH, et al. MRI Findings of Disc Degeneration are More Prevalent in Adults with Low Back Pain than in Asymptomatic Controls: A Systematic Review and Meta-Analysis. AJNR Am J Neuroradiol. 2015;36: 2394–2399. pmid:26359154
  53. 53. Chou D, Samartzis D, Bellabarba C, Patel A, Luk KDK, Kisser JMS, et al. Degenerative magnetic resonance imaging changes in patients with chronic low back pain: a systematic review. Spine. 2011;36: S43–53. pmid:21952189
  54. 54. Ranger TA, Cicuttini FM, Jensen TS, Peiris WL, Hussain SM, Fairley J, et al. Are the size and composition of the paraspinal muscles associated with low back pain? A systematic review. Spine J. 2017;17: 1729–1748. pmid:28756299
  55. 55. Haag LM, Heba S, Lenz M, Glaubitz B, Höffken O, Kalisch T, et al. Resting BOLD fluctuations in the primary somatosensory cortex correlate with tactile acuity. Cortex J Devoted Study Nerv Syst Behav. 2015;64: 20–28. pmid:25461704
  56. 56. Muret D, Daligault S, Dinse HR, Delpuech C, Mattout J, Reilly KT, et al. Neuromagnetic correlates of adaptive plasticity across the hand-face border in human primary somatosensory cortex. J Neurophysiol. 2016;115: 2095–2104. pmid:26888099
  57. 57. Van Oosterwijck J, Nijs J, Meeus M, Van Loo M, Paul L. Lack of endogenous pain inhibition during exercise in people with chronic whiplash associated disorders: an experimental study. J Pain. 2012;13: 242–254. pmid:22277322
  58. 58. Van Oosterwijck J, Nijs J, Meeus M, Lefever I, Huybrechts L, Lambrecht L, et al. Pain inhibition and postexertional malaise in myalgic encephalomyelitis/chronic fatigue syndrome: an experimental study. J Intern Med. 2010;268: 265–278. pmid:20412374
  59. 59. McPhee ME, Vaegter HB, Graven-Nielsen T. Alterations in pro-nociceptive and anti-nociceptive mechanisms in patients with low back pain: a systematic review with meta-analysis. Pain. 2019. pmid:32049888
  60. 60. Corrêa JB, Costa LOP, Oliveira NTB, Lima WP, Sluka KA, Liebano RE. Effects of the carrier frequency of interferential current on pain modulation and central hypersensitivity in people with chronic nonspecific low back pain: A randomized placebo-controlled trial. Eur J Pain. 2016;20: 1653–1666. pmid:27150263
  61. 61. Coppieters I, Cagnie B, Nijs J, van Oosterwijck J, Danneels L, De Pauw R, et al. Effects of Stress and Relaxation on Central Pain Modulation in Chronic Whiplash and Fibromyalgia Patients Compared to Healthy Controls. Pain Physician. 2016;19: 119–130. pmid:27008285
  62. 62. Cathcart S, Winefield AH, Rolan P, Lushington K. Reliability of temporal summation and diffuse noxious inhibitory control. Pain Res Manag. 2009;14: 433–438. pmid:20011713
  63. 63. Mannion AF, Balague F, Pellise F, Cedraschi C. Pain measurement in patients with low back pain. Nat Clin Pr Rheumatol. 2007;3: 610–8.
  64. 64. Franz M, Ritter A, Puta C, Nötzel D, Miltner WHR, Weiss T. Laser heat hyperalgesia is not a feature of non-specific chronic low back pain. Eur J Pain Lond Engl. 2014;18: 1501–1508. pmid:24842773
  65. 65. Lefaucheur J-P. Clinical neurophysiology of pain. Handb Clin Neurol. 2019;161: 121–148. pmid:31307596
  66. 66. Özgül ÖS, Maier C, Enax-Krumova EK, Vollert J, Fischer M, Tegenthoff M, et al. High test-retest-reliability of pain-related evoked potentials (PREP) in healthy subjects. Neurosci Lett. 2017;647: 110–116. pmid:28342940
  67. 67. Ritter A, Franz M, Dietrich C, Miltner W, Weiss T. Human Brain Stem Structures Respond Differentially to Noxious Heat. Front Hum Neurosci. 2013;7. Available: https://www.frontiersin.org/articles/10.3389/fnhum.2013.00530
  68. 68. Ellermeier W, Westphal W, Heidenfelder M. On the “absoluteness” of category and magnitude scales of pain. Percept Psychophys. 1991;49: 159–166. pmid:2017352
  69. 69. Jurcak V, Tsuzuki D, Dan I. 10/20, 10/10, and 10/5 systems revisited: their validity as relative head-surface-based positioning systems. NeuroImage. 2007;34: 1600–1611. pmid:17207640
  70. 70. Hüllemann P, Watfeh R, Shao Y-Q, Nerdal A, Binder A, Baron R. Peripheral sensitization reduces laser-evoked potential habituation. Neurophysiol Clin Clin Neurophysiol. 2015;45: 457–467. pmid:26602971
  71. 71. Tackmann W. Somatosensorisch evozierte Potentiale (SSEP). Evozierte Potentiale Bei Erwachsenen Kindern Thieme Stuttg N Y. 1993; 213–277.
  72. 72. Moncho D, Poca MA, Minoves T, Ferré A, Sahuquillo J. Interside Latency Differences in Brainstem Auditory and Somatosensory Evoked Potentials. Defining Upper Limits to Determine Asymmetry. J Clin Neurophysiol Off Publ Am Electroencephalogr Soc. 2015;32: 424–427. pmid:26061480
  73. 73. Cella D, Yount S, Rothrock N, Gershon R, Cook K, Reeve B, et al. The Patient-Reported Outcomes Measurement Information System (PROMIS): progress of an NIH Roadmap cooperative group during its first two years. Med Care. 2007;45: S3–S11. pmid:17443116
  74. 74. Linton SJ. A review of psychological risk factors in back and neck pain. Spine Phila Pa 1976. 2000;25: 1148–56. pmid:10788861
  75. 75. Lin M, Hirschfeld G, Margraf J. Brief form of the Perceived Social Support Questionnaire (F-SozU K-6): Validation, norms, and cross-cultural measurement invariance in the USA, Germany, Russia, and China. Psychol Assess. 2019;31: 609–621. pmid:30589275
  76. 76. Taylor JB, Goode AP, George SZ, Cook CE. Incidence and risk factors for first-time incident low back pain: a systematic review and meta-analysis. Spine J. 2014;14: 2299–2319. pmid:24462537
  77. 77. Kuorinka I, Jonsson B, Kilbom A, Vinterberg H, Biering-Sørensen F, Andersson G, et al. Standardised Nordic questionnaires for the analysis of musculoskeletal symptoms. Appl Ergon. 1987;18: 233–237. pmid:15676628
  78. 78. Jhawar BS, Fuchs CS, Colditz GA, Stampfer MJ. Cardiovascular risk factors for physician-diagnosed lumbar disc herniation. Spine J. 2006;6: 684–691. pmid:17088199
  79. 79. Neblett R, Cohen H, Choi Y, Hartzell MM, Williams M, Mayer TG, et al. The Central Sensitization Inventory (CSI): establishing clinically significant values for identifying central sensitivity syndromes in an outpatient chronic pain sample. J Pain. 2013;14: 438–445. pmid:23490634
  80. 80. Steeves JA, Tudor-Locke C, Murphy RA, King GA, Fitzhugh EC, Harris TB. Classification of occupational activity categories using accelerometry: NHANES 2003–2004. Int J Behav Nutr Phys Act. 2015;12: 89. pmid:26122724
  81. 81. Latza U, Hoffmann W, Terschüren C, Chang-Claude J, Kreuzer M, Schaffrath Rosario A, et al. Erhebung, Quantifizierung und Analyse der Rauchexposition in epidemiologischen Studien. Robert-Koch-Inst.; 2004.
  82. 82. Mäder U, Martin BW, Schutz Y, Marti B. Validity of Four Short Physical Activity Questionnaires in Middle-Aged Persons. Med Sci Sports Exerc. 2006;38: 1255–1266. pmid:16826022
  83. 83. Rusu AC, Kreddig N, Hallner D, Hülsebusch J, Hasenbring MI. Fear of movement/(Re)injury in low back pain: confirmatory validation of a German version of the Tampa Scale for Kinesiophobia. BMC Musculoskelet Disord. 2014;15: 280. pmid:25138111
  84. 84. Staerkle R, Mannion AnneF, Elfering A, Junge A, Semmer NorbertK, Jacobshagen N, et al. Longitudinal validation of the Fear-Avoidance Beliefs Questionnaire (FABQ) in a Swiss-German sample of low back pain patients. Eur Spine J. 2004;13: 332–340. pmid:14714246
  85. 85. Harding AT, Weeks BK, Horan SA, Little A, Watson SL, Beck BR. Validity and test-retest reliability of a novel simple back extensor muscle strength test. SAGE Open Med. 2017;5: 2050312116688842. pmid:28255442
  86. 86. Ito T, Shirado O, Suzuki H, Takahashi M, Kaneda K, Strax TE. Lumbar trunk muscle endurance testing: An inexpensive alternative to a machine for evaluation. Arch Phys Med Rehabil. 1996;77: 75–79. pmid:8554479
  87. 87. Formann A. [Latent class analysis: Introduction to theory and application]. Weinheim: Beltz; 1984.
    • 88. Dolnicar S. A review of unquestioned standards in using cluster analysis for data-driven market segmentation. Proceedings of ANZMAC 2002. Deakin University, Melbourne; 2002.
    • 89. Tagliaferri SD, Fitzgibbon BM, Owen PJ, Miller CT, Bowe SJ, Belavy DL. Brain structure, psychosocial, and physical health in acute and chronic back pain: a UKBioBank study. Pain. 2022;163: 1277–1290. pmid:34711762
    • 90. Tagliaferri SD, Miller CT, Ford JJ, Hahne AJ, Main LC, Rantalainen T, et al. Randomized trial of general strength and conditioning versus motor control and manual therapy for chronic low back pain on physical and self-report outcomes. J Clin Med. 2020;9: 1726. pmid:32503243
    • 91. Faith J, Mintram R, Angelova M. Targeted projection pursuit for visualizing gene expression data classifications. Bioinforma Oxf Engl. 2006;22: 2667–2673. pmid:16954139
    • 92. Desgraupes B. Clustering indices. Univ Paris Ouest-Lab Modal’X. 2013;1: 34.
    • 93. Riley RD, Ensor J, Snell KIE, Harrell FE, Martin GP, Reitsma JB, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020; m441. pmid:32188600
    • 94. Steyerberg EW, Bleeker SE, Moll HA, Grobbee DE, Moons KGM. Internal and external validation of predictive models: A simulation study of bias and precision in small samples. J Clin Epidemiol. 2003;56: 441–447. pmid:12812818
    • 95. Steyerberg EW, Harrell FE, Borsboom GJJM, Eijkemans MJC, Vergouwe Y, Habbema JDF. Internal validation of predictive models. J Clin Epidemiol. 2001;54: 774–781. pmid:11470385
    • 96. Steyerberg EW, Harrell FE. Prediction models need appropriate internal, internal–external, and external validation. J Clin Epidemiol. 2016;69: 245–247. pmid:25981519
    • 97. Editors IC of MJ. ICMJE. Defining the role of authors and contributors. 2016 [cited 29 Aug 2022]. Available: http://www.icmje.org/recommendations/browse/roles-andresponsibilities/defining-the-role-of-authors-and-contributors
    • 98. Allen L, Scott J, Brand A, Hlava M, Altman M. Publishing: Credit where credit is due. Nature. 2014;508: 312–313. pmid:24745070
    • 99. Brand A, Allen L, Altman M, Hlava M, Scott J. Beyond authorship: attribution, contribution, collaboration, and credit. Learn Publ. 2015;28: 151–155.