Predicting the evolution of neck pain episodes in routine clinical practice

Forty-seven health care centers were selected by the Spanish Back Pain Research Network to be invited to participate in this study, based on their past involvement in research on neck and low back pain. The centers were located across 11 out of the 17 Administrative regions in the country (Andalucía, Aragón, Asturias, Baleares, Castilla-León, Cataluña, Extremadura, Galicia, Madrid, Murcia, Vascongadas). The population of these regions is 35,776,167, approximately 77% of the total population of the country [7].

Fifteen centers belonged to the Spanish National Health Service (SNHS), 6 to not-for-profit institutions working for the SNHS, and 26 were private. They included 8 primary care centers, 18 physical therapy practices, and 21 specialty Services in rheumatology (5), rehabilitation (6), neuroreflexotherapy (4), and orthopedic surgery (6).

The inclusion and exclusion criteria were as follows.

Inclusion criteria: a) Seeking care for NP in a participating unit, b) suffering from neck pain, with or without pain in the arm, unrelated to trauma or systemic disease, and c) proficiency in Spanish.

Pain unrelated to systemic disease was defined as pain unrelated to cancer or inflammatory disease (e.g., rheumatoid arthritis), in patients who did not present signs suggesting fibromyalgia (defined as diffuse pain with unexplained fatigue or sleep disturbances) or “red flags” for latent systemic diseases. “Red flags” were defined as “oncologic disease during the previous 5 years, constitutional symptoms –unexplained weight loss, fever, chills-, history of intravenous drug use, or immunocompromised host” [8‐11]. Patients with “red flags” could be included in the study if the appropriate diagnostic test had ruled out the presence of systemic diseases.

Patients who were included were asked to sign an informed consent, authorizing the use of demographic and clinical data related to their care for the purpose of this study.

Exclusion criteria were: central nervous system disorders (treated or untreated), other causes of referred or radiated pain in the arm (e.g., peripheral nerve damage) and not having signed the informed consent.

In order to analyze the influence of up to 40 variables, the sample had to include at least 400 subjects who would not experience improvement [12]. Approximately 80–85% of patients with spinal pain, experience a clinically relevant improvement in pain, referred pain and disability, at 3 months, while losses to follow-up at that period range between 5 and 10% [13‐16]. Therefore, sample size was established at 2934 subjects. There were no concerns about the sample size being too large, due to the observational nature of the study.

Since this study did not require any changes to standard clinical practice, according to the Spanish law it was not subject to approval by an Institutional Review Board. All procedures followed were in accordance with the ethical standards of the Helsinki Declaration of 1975, as revised in 1983.

As per standard practice in Spain, patients and clinicians received no compensation for participating in this study.

Patients were recruited consecutively at the participating centers. All patients seeking care for NP were screened for inclusion and exclusion criteria.

All patients complying with inclusion criteria were invited to participate, and all those who accepted to sign the informed consent were included. Recruiting clinicians explained to eligible patients the importance of answering fully and accurately a series of questionnaires assessing their clinical status, and complying with the follow-up visit for an assessment of their evolution.

Patients were assessed upon recruitment and at follow-up. The follow-up assessment was planned at 3 months because: a) this study sought to analyze the outcome of a single episode of neck pain rather than relapses, b) this timeframe implies that all patients who are symptomatic at follow-up, would be chronic [17]; c) existing studies conducted in the environment where this study took place, have shown that losses to follow-up remain minimal for periods of up to 3 months [15, 18, 19], rise at 6 months [20‐22], and become increasingly significant thereafter [13, 23].

Patients were asked to complete the self-administered questionnaires at both assessments. Questionnaires were completed in private, with no interference from health care personnel or any other actors. Data from the questionnaires were inserted into a database by a team of auxiliary personnel with no connection to the treating physician. In order to make clinical decisions, the treating clinicians had access to that information, but were not able to alter the data. Participating in this study did not imply any changes to patients’ clinical management, and clinicians were instructed to manage their patients as usual.

This registry used the same variables and measuring instruments as the only registry of neck and back pain patients available in Spain. The latter was originally developed for post-marketing surveillance of a minimally invasive technology (“neuro-reflexotherapy”), and has shown to be reliable and lead to low proportions of missing data and losses to follow-up [14, 16, 24].

The registry gathered data from patients and from clinicians. Data requested from patients at the first assessment, were: gender, age (date of birth), duration of the current pain episode (date of pain appearance), time elapsed since the first episode (years), and employment status (classified as working, on sick leave, receiving disability compensation, student, housewife, unemployed, retired, or other; at the analysis phase, these categories were collapsed into: “working”, “receiving financial compensation for NP” -on sick leave or disabled for that reason-, or “non worker” –any other status-).

At both assessments, patients were asked to report pain and disability, which are considered two of the main outcome measures for patients with spinal pain [25]. To this end, they completed two separate 10-cm visual analog scales for NP and AP (−VAS-, for which 0 = no pain and 10 = worst imaginable pain) [26], and a validated Spanish version of the Neck Disability Index (−NDI-, for which 0 = no disability and 100 = worst possible disability) [18].

Data requested from recruiting clinicians were: diagnostic procedures prescribed for the current episode (X-Rays, CT scan, MRI, EMG, other -blood analyses, scintigraphy, etc.-), patients’ radiological findings on imaging procedures performed for the current or previous episodes, as reported by radiologists (disc degeneration, facet joint degeneration, scoliosis, difference in leg length, spondylolisis, spinal stenosis, annular tear, disc protrusion, disc herniation, other radiological findings, no findings), diagnosis (pain caused by disc herniation, spinal stenosis or “common non-specific NP”), and treatments undergone by the patient throughout the study (drugs –analgesics, NSAIDs, steroids, muscle relaxants, opioids, other drugs-, physiotherapy and rehabilitation -which were collapsed into a single category at the analysis phase, and were defined as any form of exercise, heat, cold, electrotherapy or hands-on techniques, such as massage or mobilization-, neuroreflexotherapy intervention -defined as the implantation of surgical material in specific areas of the skin, for up to 90 days- [27], surgery, other treatments -e.g., spinal injections-).

Pain caused by disc herniation or spinal stenosis, was diagnosed if radicular pain or neurologic signs existed, and were consistent with the location in which disc protrusion/herniation or spinal stenosis had been documented on MRI. Patients not fitting into these definitions, were classified as suffering from “common, non-specific NP”.

All the analyses were undertaken by a team of independent biostatisticians who had no contact or communication with the clinicians involved in this study.

For categorical variables, absolute and relative frequencies were calculated. When distribution was normal, values for continuous variables were described through their mean and standard deviation (SD). When their distribution departed from normality, median and percentiles 25 and 75 were used.

Reductions in VAS or NDI scores between the baseline and follow-up assessments, were considered to reflect improvement only if they were greater than the minimal clinically important change (MCIC). The MCIC for pain and disability has been established as 30% of their baseline scores, with a minimum value of 1.5 for VAS and 7 NDI points for neck pain-related disability [19, 28]. This implied that it was impossible for patients with baseline scores below these values, to show a clinically relevant improvement at follow-up.

For instance, it was impossible for a patient with a disability score smaller than the MCIC at baseline to experience improvement in disability, and including this patient’s data into the model exploring factors associated with a clinically relevant improvement in disability could skew results. In the case of patients with baseline scores below the MCIC, lack of improvement could only be spotted if they experienced a worsening of pain or disability (i.e., an increase in the score at follow-up, as compared to baseline). Therefore, in the case of patients with a baseline score which was smaller than the MCIC for a given variable, only those who at discharge had shown to have experienced a worsening were included in the analyses.

Outcome measures were neck pain, arm pain and disability. In order to quantify the likelihood that a given patient would experience a clinically relevant improvement in these variables, three separate multivariable predictive logistic regression models were developed. Improvement in NP, AP or disability were the dependent variables, and the maximal models included: gender; age (in years); baseline intensity of NP (VAS points); baseline intensity of AP (VAS points); NP-related disability at baseline (NDI percentage); duration of the current episode in days (number of days); duration of the current episode (classified as acute or chronic, with a cut-off limit at 90 days) [17]; time elapsed since the first episode (years), employment status (“working”, which was the reference category, “non worker” or “receiving financial compensation for NP”); diagnostic tests undertaken at any moment during the study period (X-Rays, MRI, other); findings in imaging procedures undertaken during the study period or previous episodes (disc degeneration, facet joint degeneration, scoliosis, spondylolisis, spinal stenosis, disc protrusion, disc herniation, other findings, no findings); clinical diagnosis (pain caused by disc protrusion/hernia, pain caused by spinal stenosis, or non-specific, common neck pain); and treatments used during the study period (drugs -analgesics, NSAIDs, steroids, muscle relaxants, opioids, other-, physiotherapy, rehabilitation, neuro-reflexotherapy, surgery).

The value p > 0.05 was used to eliminate variables from the model, following a non-automatic backward strategy. Nomograms were developed to illustrate the results of the models [12].

The area under the ROC curve (AUC) was used to asess discrimination of the final models, and calibration was assessed using the Hosmer-Lemeshow test [12].

The selection of variables was validated by using bootstrapping: 100 bootstrap samples were drawn [29]. Sample size was estimated using the number of observations that contained no missing values [29, 30]. The variables selected were displayed for each sample drawn, and the total number of times each variable was selected was counted [29].

Both the apparent performance of each bootstrap sample (“bootstrap performance”) and the performance of the bootstrap model in the original sample (“test performance”), were determined. Overfitting was defined as the average difference between bootstrap performance and test performance [31].

The Stata/IC v15.1(StataCorp, 2017) was used for statistical analysis.