Spinal diseases are a group of conditions that affect the spinal column, leading to various symptoms ranging from pain to paralysis. The types of conditions may include spinal stenosis, herniated disc, scoliosis, osteoporosis, and degenerative disc disease, each with a unique etiology [1]. These conditions can be caused by various factors, such as genetic predisposition; age-related degeneration; trauma; infections; autoimmune and metabolic disorders; and lifestyle choices, including posture, exercise, and weight management [2]. Low back pain (LBP) is a significant health problem highly associated with spinal disorders [2], which affected an estimated 7.5% of the world’s population in 2017, with approximately 568.4 million cases reported worldwide in 2019 [3]. It has prevailed as the leading cause of disability worldwide, contributing to 63.7 million years lived with disability as of 2019, influencing people of working age (from 20 to 65 years) and beyond [4]. In 2017, the cost of LBP topped the health care spending in the United States, estimated at US $134.5 billion [5]. Furthermore, LBP leads to wage and productivity losses, reflecting high costs to society [6-8]. Consequently, significant research efforts have been placed on spinal disorders, including technological patient management.

Presently, physicians are encouraged to deploy an evidence-based approach toward diagnosis and treatment by considering the best scientific (ie, matching symptoms and signs with relevant investigations and ensuring that the radiological features are concordant with the observed symptoms and signs) or research evidence and clinical experience while considering patients’ values and preferences [9]. However, the overwhelming number of scientific publications makes it challenging for physicians to stay updated with the latest evidence. To address this issue, computer-based tools, such as clinical decision support systems (CDSSs), can be used.

CDSSs are computerized tools used in health care to provide personalized treatment recommendations, aid in clinical diagnosis, and predict patient-specific outcomes and prognoses [10]. These tools significantly enhance disease management in health care by improving diagnostic accuracy through timely information and narrowing down potential conditions [10]. It ensures that evidence-based treatment recommendations align with current medical guidelines, aiding medication management with alerts for interactions and allergies [11]. In personalized medicine, CDSSs use genetic data for tailored treatment plans [10]. They allow the optimization of health care workflows, reduces errors, and improves communication among professionals, thereby enhancing patient outcomes and efficient health care delivery [11]. The CDSSs can be broadly classified into knowledge-based and non–knowledge-based systems. Knowledge-based CDSSs use rules to match patient data with preset knowledge domains based on up-to-date, evidence-based clinical information, from which the best recommendations can be derived [11]. In contrast, non–knowledge-based systems use data-driven methods such as artificial intelligence (AI) or machine learning to make predictions or decisions. Although limited by their lack of transparency and auditing capability, non–knowledge-based systems can provide alternative perspectives and highlight potentially overlooked factors [10]. Recently, newer methods have been developed to interpret some AI findings, offering the possibility of greater acceptance of the non–knowledge-based methodology [12,13].

A systematic review and meta-analysis reported a 10% to 20% decrease in morbidity when CDSSs were used in patient care [14]. Physicians using CDSSs are more likely to order appropriate treatment or therapy and make fewer medication errors, thereby improving overall patient safety [10,15]. Despite these successes, research regarding the use of CDSSs in spinal disorders is still in its infancy, with much to be explored.

Previous reviews have investigated the diagnostic and predictive performances of AI and machine learning [16-26]. However, no systematic or scoping review on the use of CDSSs in patients with spinal disorders has been identified. Therefore, this scoping review aimed to assess the extent of the literature in which CDSSs were implemented in clinical practice to assist health care professionals in offering personalized and meaningful care for patients with spinal disorders. The following review questions were answered: (1) Which CDSS tools can be identified in the current literature on spinal disorders? (2) What are the different purposes that the CDSS tools serve for spinal disorders? (3) How are these CDSS tools developed and assessed for effectiveness? and (4) What are the user’s perceptions and experiences regarding the use of CDSS tools?

This review was conducted using the Joanna Briggs Institute (JBI) methodological guidance for scoping review and reported according to the PRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) statement [27,28]. The protocol for this review was registered in the Open Science Framework.

The following inclusion criteria were used to determine study inclusion: (1) the study examined the CDSS use in patients with spinal disorders affecting the spinal column, cord, nerves, discs, or vertebrae in the cervical, thoracic, lumbar, or sacral regions of the spine and those with back pain, neuropathic pain, numbness, abnormal sensation, or tension caused by spinal issues; (2) all types of CDSS were considered, including integrated or independent systems, with purposes including diagnosis, disease or treatment prognosis, and treatment management of spinal disorders; (3) all participants were considered, with no restrictions placed on their cultural or racial background, geographic location, sex, or clinical management setting (acute or community); and (4) there were no restrictions placed on the study type, design, or source. The studies were excluded if they did not involve human participants, did not use a digitalized solution for ease of accessibility and use, were not applied in a clinical setting, or were reviews.

Both published and unpublished studies were located through PubMed, Embase, Cochrane, CINAHL, Web of Science, Scopus, ProQuest, and PsycINFO databases from inception until October 11, 2022. A limited initial search of PubMed was conducted to identify related articles and gather relevant keywords to develop a complete search strategy. The search strategy (Multimedia Appendix 1) was formed using the main concepts, including clinical decision support system and spinal disorders, combined with Boolean operators of AND and OR. The keywords and index terms were adapted for each database, and the reference lists of the included sources were screened for additional relevant studies. No limitations were placed on the sources’ language or date of publication to ensure that all relevant information on the topic was captured. In addition, sources of unpublished studies or gray literature, such as ClinicalTrials.gov, the International Standard Randomized Controlled Trial Number Register, the World Health Organization International Clinical Trials Registry Platform, and the Directory of Open Access Journals, were also searched.

Potential records were collated and uploaded to EndNote 20 (Clarivate), with duplicates removed [29]. Two independent reviewers (ZAT and QYCH) screened the titles and abstracts based on the eligibility criteria. The full text of potentially relevant studies was retrieved and further assessed for eligibility by both reviewers. The studies that did not meet the inclusion criteria were recorded and reported in the scoping review. Any disagreements between the 2 reviewers at each stage of the selection process were resolved through discussion or involving an additional reviewer (BB).

Data were extracted from the studies by 2 independent reviewers (ZAT and QYCH) using a data charting form adapted from the standardized data extraction tool of the JBI [27]. The extracted data included details about the participants, concept, context, study methods, and key findings relevant to the review questions. Iterative updates to the charting table allowed for the addition of valid unforeseen data [27]. We organized the research according to the applications examined and summarized the characteristics of each group, including the settings, participants, study designs, performance measures, and overall conclusions.

A total of 26,828 records were identified from PubMed, Embase, Cochrane, CINAHL, Web of Science, Scopus, ProQuest, and PsycINFO databases. Of these, 73 (0.27%) full-text papers were retrieved after screening titles and abstracts and assessed against predetermined eligibility criteria (Figure 1); eventually, 31 (0.16%) studies were included for synthesis in this review, as summarized in Table 1. The studies were conducted in the United States (13/31, 42%), Australia (4/31, 13%), the Netherlands (3/31, 10%), Switzerland (2/31, 7%), Germany (3/31, 10%), Canada (1/31, 3%), Russia (1/31, 3%), Sweden (1/31, 3%), Ireland (1/31, 3%), South Korea (1/31, 3%), and the United Kingdom (1/31, 3%).

The use of CDSSs in spinal disorders is summarized into 4 major categories based on their primary purpose and application, as presented in Table 1: of 31 CDSSs, 8 (26%) were for the prevention of unnecessary imaging, 6 (19%) were for diagnostic applications, 11 (35%) were for prognostic applications, and 6 (19%) were for treatment recommendations. Only 5 (16%) of the 31 studies investigated user perceptions and experiences concerning the use of CDSSs [31,41,51,56,57].

Of the 31 CDSS studies reviewed, the implementation and results of the 8 (26%) CDSSs used to determine if radiologic imaging was necessary for patients with lower back pathologies [30,33-35,37], patients with cervical spine trauma [31,32], and patients in general [36] are presented in Table 2. The CDSSs were mainly embedded into the electronic health record system or the computerized physician order entry, apart from the guidelines proposed by Goergen et al [31], which used a physical report card and independent software. These CDSSs were often implemented in health care settings, such as the emergency departments, where patients with back pain or cervical spine trauma were first seen by the physicians. They functioned as alerts to remind physicians to consider whether spinal imaging is necessary and can take different forms, including hard-stop, soft-stop, and passive alerts. Hard-stop alerts aim to prevent the physician from proceeding with imaging orders that do not meet the guideline requirements. In contrast, soft-stop alerts may allow the physician to continue with the ordered imaging but require them to provide a reason. Passive alerts only require acknowledgment and do not require further user interactions. Although some studies did not specify the type of alert used, the information provided in the studies allowed for inference that all studies used a soft-stop alert, excluding 1 study that used a passive alert function [37].

The included studies reported ≥1 of the following outcomes: change in the frequency of imaging order, change in the frequency of imaging order 1 to 30 days after LBP presentation, and adherence to order guidelines. All studies reported a decrease in imaging ordered on the initial presentation of LBP after the implementation of a CDSS, although the decrease was not clinically relevant in some studies [30,34]. Ip et al [33] reported a notable increase (22.7%; P=.03) from 2.2% (188/8437) to 2.7% (352/13,008) in the lumbar spine–magnetic resonance imaging (LS-MRI) ordered by outpatient specialists within 30 days of the patient’s primary care visit. This increase may be explained by the fact that the CDSS intervention was implemented in the primary care setting but not in the outpatient setting. However, when considering the total percentage of the LS-MRI orders for LBP visits before and after CDSS implementation, there was a statistically significant decline (12%; P=.002) from 8.9% (753/8437) to 7.8% (1009/13,008) in imaging orders after adjusting for outpatient specialist orders.

Zafar et al [37] compared the outcomes of different CDSS deliveries for LS-MRI orders [37]. The CDSS report cards that were generated every 4 to 6 months led to fewer magnetic resonance imaging (MRI) orders (50/1739, 2.9%) for cases compared with immediate CDSS alerts (94/2021, 4.7%).

Furthermore, CDSSs, generally, were reported to improve adherence to imaging guidelines. For example, Hynes et al [32] reported a 99.2% adherence rate to the established imaging guidelines after CDSS implementation (125 indicated imaging out of 126 total imaging), an increase of 22.5% (76.7 to 99.2%) from preimplementation [32]. Similarly, Solberg et al [36] discovered a reduction of 20% in the volume of MRI spine orders and an increase in the appropriateness of MRI spine orders based on health impacts [36].

Of the 31 studies reviewed, 6 (19%) explored diagnostic CDSSs (Table 3) and 3 (10%) examined the accuracy of CDSS compared with expert or gold standard diagnoses [38,39,42]. A moderate agreement was found between the CDSS and expert diagnoses for back pain (Cramer V=0.424) [38]. A higher agreement of 67% (58/86) of the cases between the CDSS and expert diagnosis (Cramer V=0.711) was found for patients with spinal disorders in general [39]. Another study by Lin et al [40] found that a CDSS performed a diagnosis comparable to that of experts and correctly recommended 75.82% of diagnoses based on gold-standard criteria [40]. In a recent study by Kim et al [42], the CDSS diagnosis demonstrated a 94% agreement with the gold-standard radiographic assessment for scoliosis, with higher agreement reported for patients within the normal and mild postural deformation range [42].

Of the 31 CDSS studies reviewed, 11 (35%) prognostic CDSS studies (Table 4) were knowledge based [44,45,47-54,59], with regression-based predictive algorithms. White-box models were used across all studies; most CDSSs were presented as web-based calculators, whereas others were presented as independent software. Prognostic CDSSs are used for various purposes, most commonly to predict the likelihood of complications, functional outcomes, pain, and quality of life following spinal surgery (8/11, 73%). Other purposes included predicting the outcome of brace treatment for adolescent idiopathic scoliosis (1/11, 9%), the risk of back pain chronicity (1/11, 9%), and treatment outcomes between surgical and nonsurgical options for spinal disorders (1/11, 9%). Regarding rigor, external validation was only available for 3 (27%) CDSS models (FUSE-ML, Surgical Care and Outcomes Assessment Programme-Comparative Effectiveness Translational Network Tool, and STarTBack), and an impact study was only performed for the StarTBack model.

A total of 2 key aspects, namely discrimination and calibration, are often measured to evaluate the performance of a model. Discrimination can be assessed using various measures such as area under the receiver operating characteristics, accuracy, sensitivity, positive predictive values, negative predictive values, R² measure or value, or any specific statistic measure, such as Nagelkerke, c-index, mean absolute error, and root mean square error. In contrast, calibration can be evaluated using techniques such as calibration plot, calibration intercept and slope, and the Hosmer-Lemeshow chi-square statistic.

The impact study was the only study that conducted a clinical impact testing follow-up, as reported by Foster et al [59]. This study developed an innovative web-based calculator that assesses patients’ risk of developing chronic LBP and offers tailored treatment options for each risk stratum. Results from the impact study revealed small but significant improvements (P=.03) in Roland-Morris disability scores, with a mean difference of 0.71 (95% CI 0.06-1.36) compared with usual care after 6 months of implementation. Furthermore, the group with a higher risk of developing chronic LBP experienced a large and clinically significant improvement. Work absence was also reduced by 50% (4 days instead of 8 days; P=.03), and there was a 30% decrease in prescriptions for sickness certificates (45/368, 12.2% vs 40/554, 7.2% cases; P=.03).

Of the 31 CDSS studies reviewed, studies exploring the use of CDSS for treatment recommendations for spinal disorders were divided into 2 categories based on their focus: 2 (6%) CDSSs for recommendations for spinal surgery [55,58] and 4 (13%) CDSSs for treatment of LBP [38,41,56,57] (Table 5). All CDSSs were knowledge based, except for 1, which was structured on medical ontology and fuzzy logic principles [55]. The system inputs required to generate personalized treatment recommendations include symptoms, clinical findings, and instrumental findings.

Byvaltsev and Kalinin [55] studied using a CDSS to recommend total disc replacement, minimally invasive rigid stabilization, and open rigid stabilization [55]. The researchers observed lower pain levels and improved functional status 6 months after surgery among those who received treatment recommendations using the CDSS [55]. Those who underwent minimally invasive rigid stabilization had better outcomes 3 months after surgery [55]. In the work of Benditz et al [38], although 49.6% (55/111 cases) of the treatment recommendations made by the CDSS were consistent with those of spinal surgeons, 36% (40/111) were overestimated and 14.4% (16/111) were underestimated [38]. In contrast, a study by Downie et al [56] revealed that CDSS recommendations were highly concordant with those made by pharmacists for cases involving self-care (18/20, 90%), medications (25/25, 100%), and referral advice (22/25, 88% [56]).

Of the 31 CDSS studies reviewed, 5 (16%) studies examined the user acceptability of CDSS use and gathered feedback for improvement [31,41,56,57]. User perceptions were mixed, with the most receptive toward CDSS use [41,56,57] because it provides evidence-based content to support patient care and empowerment by involving patients in decision-making. Some perceived the use of CDSS as additional work [31], while others doubted the tool’s accuracy owing to the complexity of LBP [41]. However, in cases where physicians felt that complex treatment or imaging was not recommended, CDSSs were found helpful in supporting their recommendations and reassuring patients about the decision [41]. Furthermore, the physicians were more likely to use CDSS if it lightened their workload or improved their efficiency [57].

We identified 4 major applications of the CDSS: preventing unnecessary imaging, aiding diagnosis, aiding prognosis, and suggesting treatment options. Only 2 studies used non–knowledge-based algorithms for diagnosis and treatment recommendations, whereas knowledge-based algorithms were the most commonly applied approach. Common input variables included age, gender, height, smoking status, education level, employment status, race or ethnicity, medical comorbidities, preoperative pain and disability, previous spinal surgery, symptom duration, surgical approach and intervention, BMI, American Society of Anesthesiologists score, and surgical diagnosis.

MRI detects soft tissue abnormalities [60], but the increased cost, time, and logistical demands compared with other imaging techniques make its use inconsistent with value-based care for nonspecific indications [61]. The National Emergency X-Radiography Utilization Study criteria, Canadian Cervical Spine Rule, and American College of Physicians and American Pain Society guidelines were created to direct the diagnosis and treatment of back pain and suspected spinal injury [62,63]. However, adherence to these guidelines is poor owing to defensive medicine, the continued use of unnecessary imaging to avoid missing serious pathologies [35,64].

Integrated CDSSs offer 2 benefits. First, they act as gatekeepers by adding an extra step before imaging is ordered [35]. Second, they educate or remind physicians of the existing guidelines, reducing the need to memorize multiple protocols [35]. Most studies have found that using the CDSS decreases the number of imaging tests ordered both at the time of the initial LBP visit and up to 30 days after the visit. However, other studies have not found a significant decrease in imaging orders, suggesting a potential mistrust of the system or a lack of awareness of imaging guidelines [34]. Furthermore, an insignificant decrease in imaging order may arise from the decision to use computerized tomography or x-ray instead of MRI, which could be more appropriate for some patients [30].

The use of alert-based CDSS raised concerns about alert fatigue, where repeated alerts may lead to physicians ignoring system prompts. Unnecessary imaging frequency was reduced when CDSS-generated report cards were distributed to physicians every 4 to 6 months compared with real-time alerts [37]. Furthermore, the ease of use of CDSS can hinder proper imaging if separate software is required, requiring the physician to toggle between the ordering and the CDSS system. In addition, the lack of real-time consequences for ignoring prompts may contribute to the continuation of unnecessary imaging practices.

In general, diagnostic CDSSs operate through questionnaires that generate probable diagnoses. CDSS-generated diagnoses were found to be primarily concordant with expert or gold-standard recommendations, indicating potential feasible use. Despite its ability to provide reliable diagnoses, most studies still recommend using the diagnostic CDSS as an aid instead of a replacement for the expertise and judgment of trained and experienced health care professionals [38,39,43]. In addition, patient-provider interactions are essential, and a human connection is a part of building a healing and therapeutic relationship [65]. Health care providers can assess a patient’s physical and emotional well-being better than a machine, which is only as good as its algorithm. As an aid, diagnostic CDSS could allow a brief initial assessment of the patient’s condition and assist in triaging, allowing patients with critical spinal disorders to receive early attention [38,39].

To ensure generalizability and continued validity of the CDSS, it is crucial that regular updates with the latest evidence-based information be made available to the system [40]. Meanwhile, given the lack of non–knowledge-based CDSS for spinal diagnostic purposes, AI or machine learning algorithms should be explored. The potential of AI in the field of diagnosis remains to be fully tapped, especially in the areas of computer vision and image recognition. There are promising signs of the increased prominence of diagnostic CDSSs and their ability to produce faster and more accurate findings [66].

All the included prognostic CDSS studies used white-box models. This model allows for the adaptation and modification of variables to identify areas for optimization to improve the outcomes [67]. Traditional statistical methods for prognostic modeling use simpler computation methods that allow insight into causal effects [67]. In contrast, machine learning methods are often referred to as black-box models owing to the computational complexity that allows for fast and accurate predictions but at the cost of transparency. Previous research has shown that machine learning models may perform poorer than traditional statistical methods, suggesting that this tradeoff is not justified [68]. The poorer performance may have resulted from using low-dimensional data; however, with the increasing availability of high-dimensional data and repositories of large data sets, such as biomarkers and imaging techniques, machine learning could have a competitive advantage over traditional statistics [54].

The prognostic CDSS systems are currently available as independent programs, as most are in the process of development or testing, and specialized sets of algorithms and flexibility for adjustments are required. Such an implementation could also be intentional to ease access for the users of a different electronic system, reduce the cost of integration, and ensure the confidentiality of data [10].

The prognostic CDSSs reviewed in our study were fragmented in their methodology, and none were ready for clinical implementation. The emergence of prognostic models employing AI and big data has been on the rise. However, reviews have identified poor standardization and quality of their development [69,70]. Previous reviews found that most prognostic model research ends with model development, with only a small number of studies performing external validation and even fewer conducting impact studies [70]. This aligns with the findings of our review, in which the included studies were found not to adhere well to standards, limiting the model’s validity, generalizability, and application in real-world clinical settings. Only 2 (6%) of the 31 included studies [52,54] used a reporting guideline, namely the Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis statement [71]. Future developments should adhere to the Prognosis Research Strategy prognostic model research framework, which emphasizes model development, external validation, impact testing [72], and reporting guidelines to ensure standardization and generalizability of the models.

The predictive ability of prognostic models is expected to weaken with time owing to changes in diagnostic and treatment approaches [72]. Therefore, it may be more beneficial to improve and recalibrate existing models instead of developing new models. In addition, including biomarkers and imagery data may improve model performance, but caution should be taken to address issues such as class imbalances, missing data, and the need for adequate validation [54]. Although adding more variables to a model can increase its predictive power, it can also make the model less user-friendly. To balance the tradeoff between accuracy and user-friendliness, parsimonious models that include only the most important or highly correlated predictors of the outcome are preferred. Techniques such as recursive feature elimination, principal component analysis, factor analysis, and multidimensional scaling can be used to identify key predictors [73].

According to Benditz et al [38], only 49.6% of the treatment recommendations made by the CDSS agreed with those of the physicians [38]. Although this low level of concordance may be seen as a problem and may affect confidence in the use of the CDSS, it is important to note that concordance is not necessarily the best indicator of performance; instead, testing the clinical effects of treatment options recommended by the CDSS may be a more accurate method to assess its performance.

Suggestions to improve the acceptance and usability of CDSSs include integrating them into the existing workflow and clinical decision-making processes [41]. This integration eases access to evidence-based information, encouraging use and adherence to the best practice guidelines [58].

Although the CDSS has been widely accepted for recommending treatment or management of spinal disorders, concerns and suggestions have been raised. The top barrier to CDSS use is interference with physician autonomy [57]. The physicians may feel threatened by CDSS recommendations and worry that they may eventually diminish their role in the care process [74], leading to questions about their competence [41]. In addition, ease of use is a common barrier; some physicians have negative sentiments toward the simplicity of their CDSS [56]. Furthermore, physicians are unwilling to use CDSS if it increases the time and cost [57]. Involving clinicians in the development of CDSS can improve system acceptance and adoption by ensuring that it meets the needs and preferences of users.

This review was conducted rigorously and adhered to established guidelines, including the JBI methodological guidance for scoping reviews and the PRISMA-ScR statement, ensuring transparency and credibility of the review [27,28]. In addition, 2 independent reviewers (ZAT and CQYH) were involved in the complete review process, which reduced potential biases. Furthermore, a systematic search was used to ensure a comprehensive coverage of the available literature.

Owing to the heterogeneous nature of the data included in this review, statistical analysis was not feasible, even among studies with similar objectives. Therefore, a rigorous and transparent scoping review was conducted to elucidate the mechanisms of action, effectiveness, and user acceptance of the CDSS for spinal disorders, with the hope of fostering interdisciplinary understanding and collaboration.

The methodology of this scoping review did not require a formal quality assessment of the included studies, and consequently, such an evaluation was not conducted. We recognize that the quality of the literature incorporated is crucial in shaping the outcomes of this study, thus constituting a limitation to the findings. During the screening process for study inclusion, interrater reliability was not systematically evaluated, representing another acknowledged limitation of this study. However, to address potential inconsistencies in judgment, we actively engaged in discussions and sought the input of a third reviewer (BB) to reach a consensus.

The current implementation of CDSSs for spinal disorders is fragmented and inconsistent, which poses a challenge to comprehending and advancing this field. The lack of a standardized reporting structure in the reviewed studies presents a limitation in quantifying the effectiveness of the CDSS. To better understand the impact of CDSS on health care delivery and optimize its use in clinical practice, further research with standardized reporting methods is needed.

Our recommendation for future work is to focus on assessing the quality of prediction models while adhering to transparent reporting guidelines, such as the Transparent Reporting of Multivariable Prediction Models for Individual Prognosis or Diagnosis—Systematic Reviews and Meta-Analyses [75]. Specifically, we suggest systematically evaluating models using validated tools, such as the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies to extract prognostic model studies and the Prediction Model Study Risk of Bias Assessment Tool to assess the quality of these models [76,77]. It is important to prioritize these efforts to ensure that the models are thoroughly evaluated and that their quality is properly assessed before application.

Previous studies assessing CDSS effectiveness typically focused on the concordance between CDSS recommendations and health care providers’ decisions. A more favorable approach involves directly comparing CDSS suggestions with real clinical outcomes. To enhance CDSS development, future research should prioritize seamless system integration, considering end users’ requirements. In addition, investigations into external validation and impact studies are essential for a thorough evaluation of the system’s effectiveness across diverse health care settings. Emphasizing these factors will contribute to a more robust understanding of CDSS performance and its potential for broader implementation in the clinical practice for spinal disorders.