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


Data Source

To conduct this study, we queried an observational administrative database of patients in the United States who had undergone lumbar fusion between 2007 and 2014. Utilization data in the inpatient, outpatient, and pharmacy settings were obtained from the MarketScan Commercial Claims and Encounters Database and Medicare Supplemental and Coordination of Benefits Database. These data encompass health care claims submitted on behalf of individuals enrolled in private insurance plans and Medicare through a participating employer, health plan, or government organization. Both inpatient and outpatient data, such as diagnosis, date of service, demographics, and employer information, were reviewed. The linked drug prescription database included information on all prescriptions covered by insurance that were filled by the patient, along with dosage, drug identification number, day supply, and prescription date. The data are frequently used in analyses of health care utilization and spending.[18,31,33,36]

Patient Sample

We obtained our study cohort by utilizing Current Procedural Terminology (CPT) codes to identify patients older than 18 years who had undergone lumbar fusion between 2007 and 2014. For a full list of codes used to define lumbar fusion, see Table 1. For each patient, we defined the "index date" as the date of the first fusion surgery within the coverage period and the "index visit" as the inpatient visit in which that surgery took place. We restricted our analysis to patients who were continuously enrolled in a health insurance plan offering pharmacy and medical benefits for the period 6 months before the index date and at least 2 years after the index date, meaning that they had to have been alive and actively enrolled in a participating plan for a minimum of 2 years following lumbar fusion. This effectively limited our sample to patients who had undergone lumbar fusion between July 1, 2007, and December 31, 2012. To restrict our sample to patients who had undergone elective surgery for degenerative disease and to limit confounding from previous surgery, we excluded those with International Classification of Diseases, Ninth Revision (ICD-9) codes for a preoperative diagnosis of trauma, postlaminectomy syndrome, systemic malignancy, paralysis, schizophrenia, and bipolar disorder.


The primary outcomes used to assess opioid use in the year following surgery were total cumulative dose, chronic use, and cessation of opioids. Total cumulative dose in morphine milligram equivalents (MMEs) 3–12 months post–index visit was calculated by multiplying the total quantity of pills by the strength of the prescription (a number representing the concentration of the active opioid ingredient), and multiplying that product by the MME conversion factor provided by the Centers for Disease Control and Prevention (CDC).[26] This allowed for comparison across opioid types in the form of MMEs (for reference, 50 mg of hydrocodone translates to 50 MMEs, and the CDC recommends avoiding or carefully monitoring a prescription > 90 MMEs per day). "Chronic use" was defined as at least 10 prescriptions (regardless of MMEs) or ≥ 120 days' supply over the 3- to 12-month post–index visit period, an outcome consistent with previous studies examining chronic use as an end point.[28,32] The outcome of "opioid cessation" was defined as filling zero opioid prescriptions in months 6–12 post–index visit. The start point for this outcome was extended from 3 months to 6 months given the clinical judgment that zero opioid medications 6 months postsurgery was a more attainable outcome for lumbar fusion.

We considered 2 sets of secondary outcomes: clinical quality indicators and costs. The assessed quality indicator outcomes were surgical complications, revision fusion, 30-day readmissions, and likelihood of discharge to home. The full list of complications and their corresponding ICD-9 or CPT codes are listed in Table 1. "Thirty-day all-cause readmissions" were defined as any repeat hospitalization, regardless of cause, within 30 days of discharge from the index visit, as is concordant with previous definitions in the literature.[8,11] Revision fusion was a binary variable representing the presence of any subsequent lumbar fusion (using the same ICD-9 and CPT codes used to define the original procedure) within the coverage period. Similarly, discharge home was a binary variable indicating whether the patient was discharged to home after the initial fusion surgery (rather than to a skilled nursing or rehabilitation facility, hospice, and so forth), which is coded as a categorical variable in the MarketScan database representing location of discharge.

The secondary economic outcomes were total costs 1 and 2 years after surgery, after excluding any negative or zero-valued payments. The logarithm of this cost was used in regression modeling to account for the right-skewed nature of the outcome.


Our primary independent variable of interest was a diagnosis of depression prior to surgery. We considered patients to have carried a preoperative diagnosis of depression if they had at least 1 recorded ICD-9 code corresponding to a diagnosis of depression in the 6 months prior to their surgery, a definition that has been shown to have a 99.66% specificity for identifying "true" cases of depression (identified using manual chart review).[12] The high specificity of this particular definition was desirable given the large cohort size and the desire to restrict our depression subgroup to true cases to decrease the likelihood of false-positive associations. A full list of the ICD-9 codes used to define depression diagnosis can be found in Table 1.


To control for differences in demographic variables and comorbidities in patients with and without depression, the full set of covariates listed in Table 1 were included in all models. This list included procedure characteristics (single-level vs multilevel fusion, single-level vs multilevel laminectomy, instrumentation), demographic information (age, sex, geographic region, year of surgery), and comorbidities (anxiety, chronic obstructive pulmonary disease, congestive heart failure, hypertension, ischemic heart disease, uncomplicated diabetes, complicated diabetes, obesity, drug abuse, alcohol abuse, tobacco use, and treatment for tobacco use). Age was defined as a categorical variable roughly corresponding to decade of life (18–34, 35–44, 45–54, 55–64, and ≥ 65 years). All comorbidities were defined as the presence of at least 1 associated ICD-9 code in the patient's chart at any point prior to surgery.

To adequately represent other patient characteristics that have been shown to influence outcomes (particularly prescription drug–related outcomes), such as benzodiazepine use and preoperative opioid use,[32] we also included preoperative opioid, benzodiazepine, and antipsychotic use as covariates in all regression models (coded as binary variables representing the presence or absence of at least 1 prescription for the medication class in the year preceding the index surgery). Given the particularly strong association between preoperative opioid use and adverse surgical outcomes,[19,28] preoperative opioid use was coded as a 3-level variable denoting zero, low, or high usage in the 6 months prior to surgery. "Zero usage" was defined as filling 0 prescriptions for opioid medications in the 6 months prior to the index hospitalization, whereas "low usage" was defined as nonzero usage that, when ranked as a percentile in terms of MME use compared with the rest of the cohort, fell below the top 25% of cumulative MME dose, also in the 6 months preceding surgery. Patients who fell in the top 25% of MMEs received in the 6 months preceding surgery were classified as having "high preoperative usage."

Statistical Analyses

We used logistic regression to examine the relationship between depression status and all binary outcomes (the primary outcomes of chronic opioid use and opioid cessation, as well as the secondary outcomes of complications, 30-day readmissions, revision fusion, and discharge home). Negative binomial regression was performed to predict the counts of cumulative MMEs received months 3–12 postoperation, and a log-linear regression model was generated to predict log total cost at 1 and 2 years after surgery. For all outcomes, the full set of demographic, comorbidity, and drug-related variables described above were included as covariates.

To assist with interpretation of the regression coefficients for cost and opioid use, we present odds ratios and confidence intervals, as well as graphs representing the predicted outcomes of a simulated patient with and without depression according to the appropriate regression models. For these graphs, we chose to simulate a patient with an age equal to the mean of the cohort and without any other comorbidities. For categorical variables (sex, geographic location), the category with the largest number of patients (female sex, residency in the southern geographic region) was selected for both depressed and nondepressed simulated patients, although this did not impact the relative difference in estimates attributable to depression. Standard errors on all plots depict the standard error of the estimate according to the full regression model, whereas differences in point estimates represent the magnitude of the effect on outcome attributable to depression. These graphs are only meant to illustrate the magnitude of the effect of depression; the significance of the marginal effect of depression can be determined using the 95% confidence interval of the odds ratio or the p value of the coefficient for depression in the given tables. A p value of < 0.05 was considered significant for all outcomes. Bonferroni correction was not required, given the existence of the predefined hypothesis that depression would negatively impact all primary and secondary outcomes.