Preoperative Depression, Lumbar Fusion, and Opioid Use

An Assessment of Postoperative Prescription, Quality, and Economic Outcomes

Chloe O'Connell, BS; Tej Deepak Azad, MS; Vaishali Mittal, BS; Daniel Vail, BA; Eli Johnson, BS; Atman Desai, MD; Eric Sun, MD, PhD; John K. Ratliff, MD; Anand Veeravagu, MD


Neurosurg Focus. 2018;44(1):e5 

In This Article


Sample Characteristics

The exclusion criteria yielded a final sample of 60,597 patients who had undergone lumbar fusion (Figure 1), 4985 of whom had a diagnosis of depression within 6 months of the index visit and 21,905 of whom had a diagnosis of spondylolisthesis at the time of surgery (Table 2). In comparisons between the depressed and nondepressed cohorts, there were significant differences in the percentage of women (71.7% vs 55.6%, respectively) and the mean age (53.2 vs 56.4 years, respectively). There were slight variations in the type of insurance plan between controls and depressed patients, some of which were statistically significant, although the effect sizes of these differences were small or negligible according to Hedges' g statistic. Patients with depression were significantly more likely to undergo multilevel fusion procedures (72.3% vs 66.0%) and significantly less likely to undergo laminectomies, both single level (36.1% vs 41.2%) and multilevel (21.7% vs 25.8%). Given the statistically significant differences (yet small effect sizes) between the groups, all of the above variables were included as covariates in the final regression models, as were the other comorbidities and drug-use variables described in Methods. Of the 21,905 patients with a diagnosis of spondylolisthesis at the time of surgery, 1675 had a diagnosis of depression. The difference in the prevalence of spondylolisthesis in the control group versus the depressed group was negligible in magnitude according to Hedges' g statistic (36.3% vs 33.6%).

Figure 1.

Flow diagram of criteria for patient inclusion. Note that many participants fell into more than 1 category per box on the right side of the diagram. For example, some participants were lacking both prescription drug information and mental health information; therefore, they appear in each of those separate categories but were counted only once in the total number excluded.

Postoperatively, the patients with depression received higher doses of opioid medications, and 31.8% of patients with depression met the criteria for having chronic opioid use as compared with 18.4% of controls. A greater proportion of patients without depression received zero opioids 6–12 months postsurgery (57.9% vs 44.1%). Moreover, the patients with depression experienced significantly higher rates of complications, revision fusions, and 30-day readmissions, although the magnitude of the differences was small. There was no significant difference in the likelihood of being discharged home following surgery. The average 1-year per-patient total costs (including index visit) were $66,376 in controls and $73,467 in the depressed patients. At 2 years postfusion, the average total cost increased to $80,257 for controls and $94,150 for depressed patients.

Primary Opioid Outcomes

Opioid use after surgery varied widely depending on depression diagnosis and preoperative opioid use. Depression diagnosis was associated with increased cumulative opioid use (measured in MMEs) 3–12 months after surgery (β = 0.25, p < 0.001; Table 3), controlling for the covariates specified above. The effect of preoperative opioid use was even larger, yet the effect of depression remained after controlling for preoperative opioid use (Figure 2 and Table 4). These relationships are apparent in the graph depicting the predicted total opioid use of a hypothetical patient with different levels of preoperative opioid consumption based on the negative binomial regression model. While a depression diagnosis increased the predicted postsurgical MMEs (months 3–12 postoperatively) of an otherwise healthy opioid-naïve patient from 612.9 to 787.9 (Figure 2), having high preoperative use increased this estimate to 10,224.4 (without depression) or 13,145.9 (with depression).

Figure 2.

Opioid outcomes according to depression diagnosis, stratified by preoperative opioid use. The predicted postoperative opioid use of an otherwise healthy patient with various levels of preoperative opioid use, both with and without depression, was calculated using the final logistic and negative binomial regression models. Error bars indicate the standard error of the overall estimate (estimated probability or cumulative MMEs) using multivariable logistic or negative binomial regression.

Similar patterns were observed with respect to chronic opioid use, defined as ≥ 10 prescriptions or ≥ 120 days' supply in the 3–12 months after surgery. A depression diagnosis was associated with significantly increased rates of chronic postoperative opioid use, even after controlling for preoperative opioid use (adjusted OR 1.28, 95% CI 1.17–1.40; Figure 3). In an otherwise healthy patient with high preoperative use, depression increased the estimated probability of chronic use from 53.2% to 59.2% (Figure 2). In a similar patient with no preoperative usage, depression increased the predicted probability of chronic opioid use from 4.4% to 5.6%.

Figure 3.

Adjusted odds ratios of preoperative depression for binary outcomes. All odds ratios were adjusted for the full set of demographic, comorbidity, and drug-use variables detailed in the text.

Depression was also associated with a decreased probability of opioid cessation during postoperative months 6–12. Again, however, preoperative use was a strong negative driver of this outcome, yet the effect of depression remained after controlling for this variable (OR 0.96, 95% CI 0.95–0.98; Figure 3). To illustrate this, Figure 2 depicts the overall predicted probability of a patient ceasing opiate use. With zero preoperative opioid use, the predicted probability of receiving zero opioid prescriptions from months 6 to 12 postoperatively decreased slightly less than 4% with a diagnosis of depression (from 74.1% without depression to 70.4% with depression). The predicted probability of cessation in a patient with heavy preoperative opiate use was much lower (only 21.0% and 24.7% with and without comorbid depression, respectively).

Secondary Quality Indicator and Cost Outcomes

Overall, a diagnosis of depression within the 6 months before surgery was associated with increased rates of complications (OR 1.14, 95% CI 1.03–1.25), revisions (OR 1.15, 95% CI 1.05–1.26), and 30-day readmissions (OR 1.19, 95% CI 1.04–1.36) after controlling for demographic, comorbidity, and preoperative drug-use variables (Figure 3). The association between depression and complications remained even after controlling for revision surgeries (OR 1.13, 95% CI 1.03–1.26). Preoperative depression did not influence the likelihood of being discharged home after surgery after controlling for these same covariates (OR 0.92, 95% CI 0.84–1.01).

Diagnosis of depression was also correlated with increased 1- and 2-year costs. A depression diagnosis in an otherwise healthy patient was associated with an estimated $3024 increase in total 1-year costs, including the cost of the initial procedure (log-linear regression, β = 0.06, p < 0.001; Table 3 and Figure 4). The estimated cost increase attributable to depression was $5598 at 2 years post–index surgery, again controlling for all previously described demographic, comorbidity, and preoperative drug variables (log-linear regression, β = 0.09, p < 0.001).

Figure 4.

Predicted 1-year and 2-year costs, according to depression diagnosis. Costs were calculated with the log-linear regression model predicting 1- and 2-year costs using a hypothetical patient with an age equal to the mean of the sample and no other comorbidities. Total 1-year and 2-year predicted log costs were exponentiated to calculate the actual predicted cost. Error bars indicate the standard error of the overall estimate using log-linear regression; these overlapped even when the marginal effect of depression was statistically significant due to the combined uncertainty of other coefficients in the model.