Natural Language Processing of Referral Letters for Machine Learning-Based Triaging of Patients With Low Back Pain to the Most Appropriate Intervention: Retrospective Study.

BACKGROUND: Decision support systems (DSSs) for suggesting optimal treatments for individual patients with low back pain (LBP) are currently insufficiently accurate for clinical application. Most of the input provided to train these systems is based on patient-reported outcome measures. However, with the appearance of electronic health records (EHRs), additional qualitative data on reasons for referrals and patients' goals become available for DSSs. Currently, no decision support tools cover a wide range of biopsychosocial factors, including referral letter information to help clinicians triage patients to the optimal LBP treatment. OBJECTIVE: The objective of this study was to investigate the added value of including qualitative data from EHRs and referral letters to the accuracy of a quantitative DSS for patients with LBP. METHODS: A retrospective study was conducted in a clinical cohort of Dutch patients with LBP. Patients filled out a baseline questionnaire about demographics, pain, disability, work status, quality of life, medication, psychosocial functioning, comorbidity, history, and duration of pain. Referral reasons and patient requests for help (patient goals) were extracted via natural language processing (NLP) and enriched in the data set. For decision support, these data were considered independent factors for triage to neurosurgery, anesthesiology, rehabilitation, or minimal intervention. Support vector machine, k-nearest neighbor, and multilayer perceptron models were trained for 2 conditions: with and without consideration of the referral letter content. The models' accuracies were evaluated via F1-scores, and confusion matrices were used to predict the treatment path (out of 4 paths) with and without additional referral parameters. RESULTS: Data from 1608 patients were evaluated. The evaluation indicated that 2 referral reasons from the referral letters (for anesthesiology and rehabilitation intervention) increased the F1-score accuracy by up to 19.5% for triaging. The confusion matrices confirmed the results. CONCLUSIONS: This study indicates that data enriching by adding NLP-based extraction of the content of referral letters increases the model accuracy of DSSs in suggesting optimal treatments for individual patients with LBP. Overall model accuracies were considered low and insufficient for clinical application.

Can We Change Health Care Costs in Patients With Complex Back Pain?: Results From a 5-year Before and After Study.

STUDY DESIGN: A before and after study cohort study. OBJECTIVES: The aim of this study was to examine changes in health care costs after multidisciplinary spine care in patients with complex chronic back pain (CBP), to analyze the predictive value of patient and disease characteristics on health care costs, and to study the potential impact of biases concerning the use of real world data. SUMMARY OF BACKGROUND DATA: Due to high direct and indirect societal costs of back pain there is a need for interventions that can assist in reducing the economic burden on patients and society. METHODS: All patients referred to a university-based spine center insured at a major health care insurer in the Netherlands were invited. Personal and disease-related data were collected at baseline. Health care costs were retrieved from the health care insurer from 2 years before to 2 years after intervention. Repeated measures analysis of variances were calculated to study changes in health care costs after intervention. Multivariable regression analyses and cluster robust fixed effect models were applied to predict characteristics on health care costs. To study regression to the mean, a fixed effect model was calculated comparing 2 years before and 2 years post-intervention. RESULTS: In total 428,158 declarations during 4.6 years were filed by 997 participants (128,666 considered CBP-related). CBP-related costs significantly increased during the intervention period and reduced 2 years after the intervention. Total health care costs kept rising. The intervention was associated with a 21% to 34% (P < 0.01) reduction in costs depending on the model used. Reduction in costs was related to being male and lower body mass index. CONCLUSION: This study suggests that reduction in CBP-related health care utilization in patients with complex CBP can be achieved after a multidisciplinary spine intervention. The results are robust to controlling for background characteristics and are unlikely to be fully driven by regression to the mean. LEVEL OF EVIDENCE: 4.

Personal and Societal Impact of Low Back Pain: The Groningen Spine Cohort.

STUDY DESIGN: Cross-sectional study. OBJECTIVE: The aim of this study was to study the personal and societal impact of low back pain (LBP) in patients admitted to a multidisciplinary spine center. SUMMARY OF BACKGROUND DATA: The socioeconomic burden of LBP is very high. A minority of patients visit secondary or tertiary care because of severe and long-lasting complaints. This subgroup may account for a major part of disability and costs, yet could potentially gain most from treatment. Currently, little is known about the personal and societal burden in patients with chronic complex LBP visiting secondary/tertiary care. METHODS: Baseline data were acquired through patient-reported questionnaires and health insurance claims. Primary outcomes were LBP impact (Impact Stratification, range 8-50), functioning (Pain Disability Index, PDI; 0-70), quality of life (EuroQol-5D, EQ5D; -0.33 to 1.00), work ability (Work Ability Score, WAS; 0-10), work participation, productivity costs (Productivity Cost Questionnaire), and healthcare costs 1 year before baseline. Healthcare costs were compared with matched primary and secondary care LBP samples. Descriptive and inferential statistics were applied. RESULTS: In total, 1502 patients (age 46.3 ±â€Š12.8 years, 57% female) were included. Impact Stratification was 35.2 ±â€Š7.5 with severe impact (≥35) for 58% of patients. PDI was 38.2 ±â€Š14.1, EQ5D 0.39 (interquartile range, IQR: 0.17-0.72); WAS 4.0 (IQR: 1.0-6.0) and 17% were permanently work-disabled. Mean total health care costs (&OV0556;4875, 95% confidence interval [CI]: 4309-5498) were higher compared to the matched primary care sample (n = 4995) (&OV0556;2365, 95% CI: 2219-2526, P < 0.001), and similar to the matched secondary care sample (n = 4993) (&OV0556;4379, 95% CI: 4180-4590). Productivity loss was estimated at &OV0556;4315 per patient (95% CI: 3898-4688) during 6 months. CONCLUSION: In patients seeking multidisciplinary spine care, the personal and societal impact of LBP is very high. Specifically, quality of life and work ability are poor and health care costs are twice as high compared to patients seeking primary LBP care. LEVEL OF EVIDENCE: 3.

Factor analyses for the Örebro Musculoskeletal Pain Questionnaire for working and nonworking patients with chronic low back pain.

BACKGROUND CONTEXT: The Örebro Musculoskeletal Pain Questionnaire (ÖMPQ) has good psychometric properties to predict return to work in patients with acute low back pain. Although it is used in patients with chronic back pain and nonworkers, there is no evidence on the factor structure of the ÖMPQ in these populations. This is deemed an important prerequisite for future prediction studies. PURPOSE: This study aimed to analyze the factor structure of the ÖMPQ in working and nonworking patients with chronic back pain. STUDY DESIGN/SETTING: This is a cross-sectional study in a university-based spine center. PATIENT SAMPLE: The patient sample consists two cohorts of working and nonworking adult patients (>18 years) with specific and nonspecific chronic back pain. OUTCOME MEASURES: The Örebro Musculoskeletal Pain Questionnaire. METHODS: Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were performed in working (N=557) and nonworking (N=266) patients for three, four, five, and six factors identified in literature. A goodness of fit index was calculated by a chi-square. Root mean square error of approximation (RMSEA) was calculated, and the number of factors identified was based on RMSEA values <.05. A Tucker-Lewis index (TLI) and a normed fit index (NFI) >0.90 are considered to indicate acceptable fit. RESULTS: In working patients, a five-factor solution had the best fit (RMSEA<0.05; NFI and TLI >0.90), but substantial adaptations should be made to get proper fit (removal of the work-related items). In nonworking patients, a four-factor analysis had the best fit (RMSEA<0.05). For both samples, items related to duration could not fit in the overall model. CONCLUSIONS: Factor structure of the ÖMPQ was not confirmed in working and nonworking patients with chronic back pain. Substantial adaptations should be made to obtain a factor structure with acceptable fit.

Can Patients With Low Back Pain Be Satisfied With Less Than Expected?

STUDY DESIGN: A prospective cohort study within care as usual. OBJECTIVE: (1) To explore the psychometric properties of a baseline disability questionnaire designed to collect patients' expectation. (2) To analyze relations between satisfaction with care and treatment success in patients with chronic low back pain (CLBP). (3) To determine the chances of being satisfied with the received care in absence of treatment success. SUMMARY OF BACKGROUND DATA: There is a lack of evidence on determinants of treatment satisfaction in patients with CLBP, specifically the role of patient's expectation of disability reduction after treatment. METHODS: Treatment expectation was measured with questions inspired by the Pain Disability Index (PDI) at baseline. Treatment success was considered if disability at the end of therapy was lower than, or equal to pretreatment expectation. An exploratory factor analysis was performed on the new questionnaire. Binary logistic regression models were used to analyze how much variance of satisfaction with care was explained by treatment success, pain disability at baseline, sex, age, duration of complaints, and pain intensity. The odds ratio of being satisfied when treatment was successful was calculated. RESULTS: Six hundred nine patients were included. The factor structure of the PDI-expectancy had optimal fit with a one factor structure. There were low correlations between the expected and baseline disability, pain intensity, and duration of pain. Correlation between treatment success and satisfaction with care was low (χ = 0.13; P < 0.01). Treatment success had a low contribution to satisfaction with care. Of all participating patients, 51.4% were satisfied with care even when treatment was not successful. The odds ratio for being satisfied was 2.42 when treatment was successful compared to when treatment was not successful. CONCLUSION: The PDI-expectancy is internally consistent. Pretreatment expectation contributes uniquely but slightly to satisfaction with care; patients whose treatment was considered successful have 1.38 to 4.24 times higher chance of being satisfied at the end of treatment. Even when treatment was not successful, 51.4% of the patients with CLBP are satisfied with care. LEVEL OF EVIDENCE: 2.

Extensive validation of the pain disability index in 3 groups of patients with musculoskeletal pain.

STUDY DESIGN: A cross-sectional study design was performed. OBJECTIVE: To validate the pain disability index (PDI) extensively in 3 groups of patients with musculoskeletal pain. SUMMARY OF BACKGROUND DATA: The PDI is a widely used and studied instrument for disability related to various pain syndromes, although there is conflicting evidence concerning factor structure, test-retest reliability, and missing items. Additionally, an official translation of the Dutch language version has never been performed. METHODS: For reliability, internal consistency, factor structure, test-retest reliability and measurement error were calculated. Validity was tested with hypothesized correlations with pain intensity, kinesiophobia, Rand-36 subscales, Depression, Roland-Morris Disability Questionnaire, Quality of Life, and Work Status. Structural validity was tested with independent backward translation and approval from the original authors. RESULTS: One hundred seventy-eight patients with acute back pain, 425 patients with chronic low back pain and 365 with widespread pain were included. Internal consistency of the PDI was good. One factor was identified with factor analyses. Test-retest reliability was good for the PDI (intraclass correlation coefficient, 0.76). Standard error of measurement was 6.5 points and smallest detectable change was 17.9 points. Little correlations between the PDI were observed with kinesiophobia and depression, fair correlations with pain intensity, work status, and vitality and moderate correlations with the Rand-36 subscales and the Roland-Morris Disability Questionnaire. CONCLUSION: The PDI-Dutch language version is internally consistent as a 1-factor structure, and test-retest reliable. Missing items seem high in sexual and professional items. Using the PDI as a 2-factor questionnaire has no additional value and is unreliable.

Responsiveness and minimal clinically important change of the Pain Disability Index in patients with chronic back pain.

STUDY DESIGN: Prospective cohort study. OBJECTIVE: The objective of this study was to test the responsiveness and minimal clinically important change (MCIC) of the Pain Disability Index (PDI) in patients with chronic back pain (CBP). SUMMARY OF BACKGROUND DATA: Treatment of patients with CBP is primarily focused on reduction of disability. For disability measurement, the PDI is a widely used questionnaire. There are, however, no data available on responsiveness and MCIC. METHODS: Two hundred forty-two patients with CBP were included in this study. Patients filled in the PDI at baseline and at discharge. The PDI consists of 2 subscales: 1 measuring voluntary activities and 1 measuring obligatory activities. PDI was anchored at 2 self-reported global perceived effect (GPE) scales for complaints and self-care, respectively. Responsiveness was considered sufficient when Area Under the Receiver Operating Characteristics (ROC) Curve (AUC) was higher than 0.70. To test interpretability, change scores and MCIC were calculated. MCIC was tested by determination of optimal cut-off point of the ROC curve and determination of specificity and sensitivity of the optimal cut-off point. RESULTS: AUCs were 0.76 and 0.77 depending on the external criterion. The subscale obligatory activities did not meet the criteria for responsiveness (AUC: 0.63-0.69). MCIC of the PDI was 9.5 points for GPE "complaints" and 8.5 for GPE "self-care." CONCLUSION: The total score of the PDI as well as the subscale of voluntary activities is responsive. Partly because of floor effects, the subscale obligatory activities are not sufficiently responsive in patients with CBP. However, the responsiveness of this subscale in other patient groups should be further tested. In patients with CBP, change can be considered clinically important when PDI score has decreased 8.5 to 9.5 points.