Introduction

Neck disorders are a leading source of disability worldwide1 and the second most common reason for primary care musculoskeletal consultations2. Cervical spondylotic radiculopathy is a common neck disorder with an estimated annual incidence of 83.2 cases per 100,000 persons3. Characterized by pain, numbness, and sensory and motor deficits in the affected arm, cervical radiculopathy can adversely impact quality of life and functional ability if not treated effectively4,5,6.

Several treatment options exist for cervical spondylotic radiculopathy, including nonoperative therapies and surgery7. Approximately 1/4 to 1/3 of patients require operative treatment3,8, usually in the form of anterior cervical discectomy and fusion (ACDF)9,10. While most patients benefit from this surgery, recent evidence shows that clinical outcomes are heterogeneous, and psychological factors such as depression and depression risk can predict some of the variability11. This accords with other studies showing that psychological health is associated with the clinical outcomes of surgery for lumbar spinal stenosis surgery12, and the pain experience in general13,14. However, it is unknown whether the relationships between psychology and surgical outcomes are associational (i.e., predictive) or if psychological factors affect the levels of pain and disability experienced by patients following cervical spine surgery.

Randomized trials are widely regarded as the benchmark method for determining the effect of a treatment or exposure due to their ability to balance confounding and prognostic variables within exposure groups15. However, it can be challenging, unethical, or impossible to assign participants to some exposures (e.g., depression). Advances in epidemiological methods have improved the accuracy of estimating effects from observational data and can yield results consistent with those from randomized trials16.

Considering the potential impact of psychological health on patients with spinal disorders, it is crucial to determine whether psychological features are simply associated with postoperative pain and disability or if they directly impact these surgical outcomes. Therefore, we analyzed data from a national spine surgery registry to evaluate the effects of preoperative psychological health factors on neck surgery outcomes. Specifically, we estimated the average treatment effects of depression and severe psychological symptomology on the two-year trajectories of arm pain, neck pain, and disability in patients undergoing ACDF for cervical spondylotic radiculopathy.

Methods

Study design and participants

We undertook a retrospective cohort study with data from the Canadian Spine Outcomes and Research Network, a national network of surgical spine centres across Canada. Psychological health variables were measured at the preoperative baseline, and pain and disability outcomes were measured preoperatively and then 3, 12, and 24 months after surgery.

In July 2021, we accessed data from patients with spine surgeon-diagnosed cervical radiculopathy who had undergone single or multi-level ACDF from January 2015 to March 2021. Data from patients diagnosed with tumours, infections, inflammatory arthritis, fractures, or myelopathy were excluded. Local ethics boards approved the data collection at each spine centre, and all patients gave written informed consent. Research Ethics Boards approved the current study protocol at the University of New Brunswick (2019-161) and Horizon Health Network (2019-2797). The research was conducted in accordance with the Declaration of Helsinki.

Psychological health exposures

We examined three evaluations of preoperative psychological health that comprised two measures of depression and one measure of severe psychological symptomology. Depression was measured prior to surgery using the Patient Health Questionnaire-8 (PHQ-8). The PHQ-8 has previously been validated for use in the general17 and spine surgery populations18. Scores ≥ 10 on the PHQ-8 indicate patients at moderate-to-severe risk of depression, a classification with 88% sensitivity and specificity for major depression19.

Health-related quality of life was measured using the Short-Form 12-Item Health Status Survey version 2 (SF-12), which is reliable and valid in people living with non-cancer-related pain20,21. This scale yields a mental component summary (MCS) score ranging from 0 to 100, with higher scores indicating better mental health. Patients with MCS scores less than 46 were categorized as experiencing depression, while those with MCS scores less than 37 were classified as having severe psychological symptomology22. The validity of these MCS cutpoints was established through comparisons to the outcomes of the structured composite international diagnostic interview, version 2.122.

Clinical outcomes

Pain and disability scores were recorded at the preoperative baseline and 3, 12, and 24 months postoperatively. Neck pain and arm pain intensity were assessed using an 11-point numeric pain rating scale (NPRS) with scores ranging from zero “no pain” to 10 “worst pain imaginable”. The NPRS has previously been validated for use in patients with various pain sources23,24 and is reliable for use in patients with cervical radiculopathy25,26. Estimated NPRS minimal important change values range from 2.2 points for patients with cervical radiculopathy25 to 2.5 points for patients undergoing cervical spine fusion27.

Disability was quantified using the neck disability index. Scored on a 100-point scale28, this index is the most widely used and validated tool to assess neck-related disability29, with an intra-class correlation of 0.88 in patients with cervical radiculopathy or myelopathy30. Studies of patients undergoing ACDF for cervical radiculopathy estimate minimal important changes to be 16.0–27.6%31.

Final pain and disability outcomes were represented by established trajectories of arm pain, neck pain, and neck pain-related disability from the preoperative baseline to 24 months after surgery. The full details of this analysis have been reported previously11. Briefly, we constructed latent class growth models that identified three trajectory subgroups for each outcome (Fig. 1). To be included in the model, patients were required to have valid outcome data reported for at least two of the four designated measurement points (i.e., baseline, 3 months, 12 months, 24 months). Depending on the outcome, 8.5% to 9.0% of patients were excluded for missing data. Average follow-up at each point was: 99.9% at baseline, 92.7% at 3 months, 84.7% at 12 months, and 75.6% at 24 months.

Fig. 1
Fig. 1
Full size image

Estimated perioperative trajectories and prevalence estimates of (A) disability, (B) neck pain, and (C) arm pain outcomes from preoperative baseline to 24 months post-surgery. Shaded areas represent 95% confidence intervals.

Each model identified a “poor” outcome subgroup with a 15.5% to 23.5% prevalence. For the current analysis, we collapsed the “good” and “excellent” subgroups for arm and neck pain and the “fair” and “excellent” subgroups for disability to create contrasts with patients assigned to the “poor” outcome subgroup.

Potential confounders and additional variables

Baseline demographic, health, and clinical data were collected on all patients as part of a standardized assessment, and all data were maintained in the central CSORN registry. Self-reported variables comprised age, sex, highest level of education attained, and current smoking status. Surgical wait time was calculated as the days from surgical consultation to the date of surgery. Physical health-related quality of life was estimated with the physical component score of the SF-1232.

Data analysis

All statistical analyses were performed using STATA/SE version 18 (StataCorp, College Station, TX). To estimate the average treatment effects of depression and severe psychological symptomatology on the trajectories of arm pain, neck pain, and disability, we combined exposure (propensity score) and outcome (regression) modelling to construct doubly robust effect estimates33.

First, we estimated covariate balancing propensity scores to balance the exposure groups on key confounders34. We examined for covariate balance by ensuring that standardized differences for all confounders were less than 0.1 between the exposure groups and variance ratios were between 0.5 and 2.035,36. We visually inspected the distribution of propensity scores with overlap plots35. Propensity scores were converted to inverse probability of treatment weights for application in subsequent regression models.

Average treatment effects were estimated with weighted robust Poisson Regression models. Results were reported with adjusted risk ratios (RR) and marginal risk estimates showing the probability of poor outcomes for each exposure group. When binary outcomes are common, robust Poisson models are preferred as they yield more precise and less inflated estimates than logistic regression37,38.

Owing to the uncertainty of the causal path between the psychological variables and postoperative pain and disability39, we applied two adjustment sets. First, we conditioned on a limited number of confounders to minimize potential overadjustment bias40: age, sex, education, and smoking status. Second, we constructed more inclusive models that additionally controlled for the physical component of health-related quality of life, surgical wait time, the number of spinal levels operated on, and the preoperative scores of neck pain intensity (arm pain models), arm pain intensity (neck pain models), or neck and arm pain intensity (disability models).

Sensitivity analyses

We conducted sensitivity analyses to investigate potential bias owing to unmeasured confounding and extreme inverse probability of treatment weights. The potential for unmeasured confounding in all significant models was estimated with E-value methodology41. E-values quantify the impact, on the risk ratio scale, that an unmeasured confounder would need to fully explain away the estimated effect, conditional on the measured covariates. To evaluate the effect of extreme weights on variance inflation, we truncated the inverse probability weights at 5% and 95% and repeated all models after applying the alternate weights42.

Results

We included data from 352 patients (43.8% female; mean [SD] age = 50.9 [9.5] years) treated at one of 12 participating spine centres. Propensity score model diagnostics were acceptable; standardized differences between patients with and without the psychological health characteristics of interest were < 0.01, and variance ratios ranged from 0.78 to 1.65, indicating balance on all confounders. Overlap plots showed an adequate balance of propensity scores between the exposure groups. The preoperative characteristics of the study sample are reported in Table 1. Risk ratios of the effect of the psychological health variables on disability, neck pain, and arm pain for limited and fully adjusted models are reported in Fig. 2, with corresponding poor outcome probabilities reported in Figs. 3 and 4.

Table 1 Preoperative demographic, health, and clinical data.
Full size table
Fig. 2
Fig. 2
Full size image

Effects of psychological health states on the risk of poor outcome for disability, neck pain, and arm pain. (2A) Models adjusted for age, sex, education level, and smoking status. (2B) Models adjusted for age, sex, education level, smoking status, physical component score, surgical wait time, number of spinal levels operated on, and baseline pain. PHQ-8 = Patient Health Questionnaire; MCS = Mental Component Score; RR = Risk ratio; CI = confidence interval; E-valueEST = E-value for the estimate; E-valueCI = E-value for the confidence interval.

Fig. 3
Fig. 3
Full size image

Probability of poor outcomes in control and at-risk groups undergoing ACDF for cervical radiculopathy from models adjusted for age, sex, education level, and smoking status. At-risk groups include moderate-severe risk of depression (PHQ-8), severe psychological symptomatology (MCS), and depression (MCS). PHQ-8 = Patient Health Questionnaire; MCS = Mental Component Score.

Fig. 4
Fig. 4
Full size image

Probability of poor outcomes in control and at-risk groups undergoing ACDF for cervical radiculopathy from models adjusted for age, sex, education level, smoking status, physical component score, surgical wait time, number of spinal levels operated on, and baseline pain. At-risk groups include moderate-severe risk of depression (PHQ-8), severe psychological symptomatology (MCS), and depression (MCS). PHQ-8 = Patient Health Questionnaire; MCS = Mental Component Score.

Effects of psychological health

Depression (PHQ-8)

Approximately half (52.1%) of the patients were identified as being at moderate-to-severe risk of depression. The limited adjustment models showed that moderate-to-severe risk of depression greatly increased the risk of poor outcome for disability (RR [95% CI] = 14.07 [4.42 to 44.85]) and approximately doubled the risk of poor outcome for neck pain (RR [95% CI] = 2.35 [1.42 to 3.89]) and arm pain (RR [95% CI] = 1.91 [1.22 to 2.00]) (Figs. 2, 3). The fully adjusted models yielded similar estimates for disability (RR [95% CI] = 6.73 [1.85 to 24.45]) and neck pain (RR [95% CI] = 1.90 [1.09 to 3.32]), while the arm pain results shifted to include the null (Figs. 2, 4).

Depression (MCS)

Nearly 2 in 3 patients (61.4%) reported MCS scores consistent with a depressive state. Depression increased the risk of poor disability outcome in the limited (RR [95% CI] = 3.46 [1.59 to 7.50]) and fully-adjusted models (RR [95% CI] = 2.77 [1.30 to 5.90]). The remaining model results were non-significant (Figs. 2, 3, 4).

Severe psychological symptomatology (MCS)

Approximately 1 in 3 patients (32.6%) reported MCS scores consistent with severe psychological symptomatology. Severe symptomatology increased the risk of poor disability (RR [95% CI] = 3.91 [2.25 to 6.62]), neck pain (RR [95% CI] = 2.35 [1.54 to 3.59]), and arm pain (RR [95% CI] = 1.92 [1.28 to 2.87]) in limited adjustment models (Figs. 2, 3). Similar results were identified in the fully adjusted models for disability (RR [95% CI] = 2.84 [1.58 to 5.09]), neck pain (RR [95% CI] = 2.08 [1.38 to 3.13]) and arm pain (RR [95% CI] = 1.82 [1.17 to 2.82]) (Figs. 2, 4).

Sensitivity analyses

For the limited adjusted models, the E-values for the estimates ranged from 3.23 to 27.63, and E-values for the confidence interval ranged from 1.74 to 8.31 (Fig. 2). In fully adjusted models, E-values ranged from 3.04 to 12.94 for the estimates and from 1.40 to 3.10 for the confidence intervals (Fig. 2). This indicates that substantial unmeasured confounding (RR ≥ 3.04), above and beyond the impact of measured confounders, would be needed to explain away the observed effects, but weaker confounding could not. Lower levels of unmeasured confounding (RR ≥ 1.40), above and beyond the measured confounders, would be required to shift the confidence interval to include the null.

There were minimal differences between effect estimates using the standard and truncated inverse probability weights, indicating no apparent weight-induced bias on model results. Consequently, we elected to report the results from models applying the standard weights.

Discussion

This study aimed to estimate the average treatment effects of preoperative psychological health factors on postoperative trajectories of pain and disability among patients undergoing ACDF for cervical radiculopathy. We found that patients identified with PHQ-8-measured depression and MCS-measured severe psychological symptomatology were at increased risk of poor outcomes for postoperative disability, arm pain, and neck pain. Similarly, patients with MCS-measured depression were at increased risk of following a poor disability trajectory. Although the effects of MCS-measured depression on postoperative neck and arm pain were non-significant, the effects were similar in magnitude to the other psychological health exposures. Compared to the MCS cut-point for depression, MCS scores indicating severe psychological symptomatology yielded larger and more consistent effect estimates, suggesting a potential dose–response relationship between psychological health and poor outcome risk. In other words, patients with worse psychological health were at greater risk of experiencing poor outcomes. These results were confirmed by sensitivity analyses demonstrating the models’ robustness to bias from extreme weights and residual confounding.

The relationships between pain, disability, and depression are complex43. The current findings support previous research indicating that depression is strongly associated with perceived disability44, and worse disability outcomes following other spinal surgeries12,45,46,47,48. Better mental health status has been similarly linked to better clinical outcomes following ACDF and other spine surgeries.12,46,48,49,50,51,52 We have advanced knowledge gained by studies to date through more robust control of confounding and less biased estimates of effect.

The current results and prior evidence support the routine measurement of psychological health factors in patients who are potential candidates for ACDF. The information gained may help inform surgical decision-making and identify patients who could benefit from preoperative interventions to improve psychological health. However, collecting health information in clinical practice requires a balanced approach that does not overly burden patients or exceed available resources. In our experience, the psychological health screening tools used in the current study (PHQ-8 and SF-12) can be efficiently implemented, even in busy surgical practices.

Future research is needed to better define the effect of psychological health on spinal surgery outcomes. Although randomized trials are traditionally viewed as the optimal design to establish effects, it is not possible to assign patients to different levels of psychological health. However, it will be possible to design and undertake trials to evaluate the potential impact of modifying psychological health before spine surgery, and this will be an important priority for future research efforts. Additionally, it remains uncertain whether the findings of this study are unique to patients undergoing neck surgery or whether they may generalize to patients experiencing other spinal disorders or pain conditions. Future studies evaluating the impacts of psychological health on both surgical and non-surgical outcomes will be essential in answering this question.

The current study had several limitations to consider when interpreting these results. Confounding is a primary concern in all observational studies aimed at estimating the effects of exposures on outcomes of interest53. To address confounding, we applied doubly robust methods incorporating propensity scores with inverse probability weights with regression adjustment. Since there is always uncertainty around the true causal structure, we considered two adjustment strategies: a limited and a more inclusive set of potential confounders, each with advantages and shortcomings. For example, if the limited adjustment set excluded key covariates, the model results would be impacted by confounding bias. Conversely, if the more inclusive adjustment set included potential mediators on the causal pathway between psychological health and the outcomes of interest (e.g., preoperative pain intensity, surgical wait time) or colliders, the results could be susceptible to overadjustment bias40. However, each strategy produced consistent results; 8 of 9 models yielded the same statistical outcome, and there were no substantive shifts in the magnitude of the effect estimates. We also performed sensitivity analyses investigating the potential impact of bias owing to the weighting strategy and the impact of unmeasured confounding, thus enhancing confidence in model outcomes.

Additional study limitations relate to information bias. Although our exposure measures employed standard screening approaches to estimating psychological health, we did not employ criterion diagnostic methods (e.g., diagnostic interviews). Screening tools such as those used in the current study are often used in epidemiological research when more time- and resource-intensive measures are not possible, and they likely represent a more feasible option for application in surgical practice. Nevertheless, when compared to a criterion, screening tools are a potential source of misclassification owing to measurement error. Similarly, we relied on patient-reported measures for covariates (e.g., smoking history, level of education) that may be prone to recall or social-desirability bias.

Conclusions

Most patients with cervical radiculopathy exhibited signs of depression or severe psychological symptomology prior to surgery. Patients with lower levels of preoperative psychological health were at increased risk of experiencing a poor outcome following ACDF. These results highlight the need to routinely collect psychological health information and potentially intervene to improve psychological health prior to spinal surgery.