Data source

We used data from the IBM MarketScan Commercial Claims and Encounters Database, a validated national commercial insurance repository gathering healthcare information for patients enrolled in private health plans. It includes >263 million individuals and inpatient and outpatient services and pharmaceutical claims across the United States. This database has been previously used to study opioid use in gynecologic settings. The study was exempt by the Johns Hopkins University Institutional Review Board as the database contains only deidentified data.

Study design and patient cohort

In this retrospective cohort study, we included opioid-naïve women aged 18 to 64 years undergoing benign hysterectomy from January 2011 to December 2016 and having at least 1 perioperative opioid prescription ( Figure 1 ). We used the Current Procedural Terminology (CPT) codes to identify hysterectomies ( Appendix , Supplemental Table 1 ), and the index date was the date of hysterectomy. To verify opioid use preoperatively and NPOU postoperatively, patients must have prescription coverage and be continuously enrolled in the database for at least 1 year both before and after the index date. We defined perioperative opioid prescription as at least 1 prescription between 30 days before and 14 days after hysterectomy. We excluded patients with >1 opioid prescription from 365 to 31 days preoperatively (nonnaïve) to better understand the independent association of perioperative prescription on NPOU. We excluded patients who underwent additional procedures or received anesthesia within 1 year after the index date to avoid erroneous overestimation of NPOU caused by unrelated opioid prescriptions. Lastly, we excluded patients with radical hysterectomy, gynecologic malignancy within 1 year before or after hysterectomy, and a delivery within 12 weeks before hysterectomy and those with >1 procedure type on the index date. As opposed to other studies, we sought to examine NPOU in women with exclusively benign pathologies to better delineate patterns of opioid use without the unique, complex implications of a cancer diagnosis.

Flowchart of the study population selection process

The number of observations is included in the flowchart.

AlAshqar et al. Persistent opioid use after benign hysterectomy. Am J Obstet Gynecol 2022.

Opioid use

Opioid use was determined using pharmacy dispensing codes in the MarketScan database. To identify opioid prescriptions, therapeutic classes and subclasses of generic drug identifiers for opioids were used ( Appendix , Supplemental Table 1 ). Metric Quantity (METQTY) is the number of pills dispensed per prescription, which alongside opioid type and number of days’ supply was identified using the National Drug Code in prescription claims data. For each prescription, morphine milligram equivalent (MME) was calculated by multiplying METQTY by the strength per pill by the MME conversion factor provided by the CDC. Invalid opioid pharmacy claims (days’ supply of ≤0 or >365 or pill quantity of ≤0) were excluded as were outliers (pill quantity ≥99th percentile). For women with multiple perioperative opioid prescriptions, only the first prescription was included in the analysis as a potential predictor of persistent opioid use postoperatively.

Primary outcome

A priori, our primary outcome was NPOU after hysterectomy ( Figure 2 ), which we defined as the fill of at least 1 eligible opioid between postoperative day 15 to 90 and postoperative day 91 to 365, in addition to at least 1 perioperative prescription between 30 days before to 14 days after hysterectomy. This definition captures patients filling additional prescriptions beyond the resolution of postsurgical pain on the basis of commonly accepted periods that align with the typical resolution of postsurgical pain.

Schematic presentation of the study timeline and new persistent opioid use criteria

AlAshqar et al. Persistent opioid use after benign hysterectomy. Am J Obstet Gynecol 2022.

Covariates

To evaluate NPOU-associated factors, we included sociodemographic, preoperative clinical, operative, and opioid prescription-related factors as covariates. Sociodemographic factors included patient’s age based on categories from the MarketScan database (18–34, 35–44, 45–54, or 55–64 years), procedure year (2011–2016), geographic region, union status, wage type, and admission status. Clinical factors included preoperative psychiatric and pain diagnoses, mood disorder medication use, tobacco use, and Charlson Comorbidity Index, whereas operative factors included hysterectomy approach, surgical indication, length of stay, and surgical complications. Patient demographic and clinical characteristics and the MarketScan, CPT, International Classification of Diseases, and clinical classification system diagnosis codes by which they were obtained are in the Appendix .

We included 4 variables to characterize perioperative opioid use: (1) total MME of the first prescription, (2) perioperative days’ supply (number of days covered by an opioid prescription), (3) having dispensed >1 prescription, and (4) first prescription timing (pre- or postoperatively), with the former 2 as continuous variables and the latter 2 as binary variables.

Statistical analysis

We analyzed the data using SAS (version 9.4; SAS Institute, Cary, NC). We conducted bivariate analyses for baseline (sociodemographic and preoperative clinical), operative, and prescription-related covariates for women with and without perioperative prescription and NPOU using the chi-squared test and Mann-Whitney U test for categorical and continuous covariates, respectively, and unadjusted odds ratios (ORs) with 95% confidence intervals (CIs) for developing NPOU were calculated. We used multivariate logistic regression analysis to estimate the adjusted ORs (aORs) and 95% CIs for developing NPOU and identify its independent predictors. Variables found to have statistical significance ( P <.05) in unadjusted analysis with all cell sizes >10 and variables with known associations with NPOU were included in the multivariate analysis. We used the median values of the continuous variables as dichotomous cut points, except for the total MME of the first perioperative prescription, wherein the 75th percentile served that purpose and days’ supply, which was analyzed as a continuous variable. We assigned the groups below the cutoff values as the referent. Statistical significance was set at P <.01 with 2-sided tests for all analyses. We used a P value cutoff of .01 rather than the usual .05 to decrease the possibility of type 1 error attributed to multiple comparisons and a large sample size.