The Incidence of Persistent Postoperative Opioid Use Among U.S. Veterans: a national study to identify risk factors

Abstract

Objective:

To calculate the incidence and identify the predictors of persistent postoperative opioid use at different postoperative days.

Background Data:

A subset of surgical patients continues to use long-term opioids. The importance of the risk factors at different postoperative days is not known.

Design:

A historical cohort

Setting:

postoperative period

Patients:

opioid-naive U.S. veterans

Interventions:

The surgical group had any one of 19 common invasive procedures. The control group is a 10% random sample. Each control was randomly assigned a surgery date.

Measurements:

The outcomes were the presence of persistent opioid use as determined by continued filling of prescriptions for opioids on postoperative day 90, 180, 270, and 365.

Main Results:

A total of 183,430 distinct surgical cases and 1,318,894 controls were identified. 1.0% of the surgical patient were using opioids at 90 days, 0.6% at 180 days, 0.4% at 270 days, and 0.1% at 365 days after the surgery. Surgery was strongly associated with postoperative persistent opioid use at day 90 (OR 3.67, 95% CI, 3.43-3.94, p<0.001), at day 180 (OR 2.85, 2.67-3.12, p<0.001), at day 270 (OR 2.63, 2.38-2.91, p<0.001) and at day 365 (OR 2.11, 1.77-2.51, p<0.001) compared to non-surgical controls. In risk factor analysis, being male and single were associated with persistent opioid use at earlier time points (90 and 180 days), while hepatitis C and preoperative benzodiazepine use were associated with persistent opioid use at later time points (270 and 365 days).

Conclusions:

Many surgeries or invasive procedures are associated with an increased risk of persistent postoperative opioid use. The postoperative period is dynamic and the risk factors change with time.

1. INTRODUCTION

Opioids are frequently used for postoperative pain management; however, the optimal length of postoperative opioid use is not known. Prescription opioids are rarely needed for longer than 15 days for postoperative pain [1, 2], and it is generally agreed that the postoperative pain usually resolves in 2-6 months [3, 4]. However, a subset of surgical patients continues to use opioids for longer than six months postoperatively [5–7].

Persistent postoperative opioid use can be considered a surgical adverse event. Opioid overdose is a known danger of postoperative opioid use [8–10], and the risk of opioid overdose or dependence is strongly associated with the duration of postoperative opioid prescription [11]. Since the long-term opioid therapy has poor efficacy [12, 13]; continued post-surgical opioid use is considered an unfavorable outcome for most surgeries with the primary indication of pain relief [14, 15]. Prolonged opioid use is also associated with other postoperative complications, such as increased rate of knee revision after knee arthroplasty [15], or worse physiologic and psychologic outcomes after bariatric surgery [16].

Rates of persistent postoperative opioid use differ between studies owing to use of different criteria for diagnosis. One year after surgery, an estimated 0.4-0.7% of patients still have ongoing prescriptions for opioids [5–7]. When persistent use was defined as any opioid prescription between 90 and 180 days after the surgery, the reported incidence was 3-6% [17]. Using an opioid-naive population and a more stringent criterion of having filled 10 or more prescription or more than 120-day supply of an opioid in the postoperative days 90-365, the incidence for chronic opioid use was reported to be 0.6-1.4% among surgical patients [18].

This study used a simpler definition for persistent opioid use as continued use of opioids at postoperative time points 90, 180, 270, and 365 days. While some comorbidities, such as mood disorders, smoking, alcohol and substance use disorders, and pain disorders have been identified as risk factors for persistent postoperative opioid use [17, 18], the association of these risk factors at different postoperative time points is not known. We hypothesized that different predictors of ongoing opioid use are important at different postoperative days. In this retrospective analysis, we aimed to calculate the association of predictors of the persistent postoperative opioid use at the several time points following some common surgeries and invasive procedures.

2. METHODS

Study Design:

We performed a historical (retrospective) cohort study of opioid-naive U.S. veterans between the ages of 18 and 85 who received medical care from any one of the 170 VA medical centers or 1,063 clinics between 1/1/2008 and 12/31/2015. The University of Maryland Institutional Review Board, and the VA Maryland Health Care System Research and Development Committee approved this study and waived the need for Health Insurance Portability and Accountability Act (HIPAA) authorization or informed consent for this retrospective study.

Study Population:

We compared two groups: surgical and control. The surgical group had any one of 19 commonly performed surgeries. These 19 surgeries or invasive procedures, identified by CPT (Current Procedural Terminology® by American Medical Association) codes (Supplemental Table 1), were: (1) coronary artery bypass graft (CABG), (2) thoracotomy, (3) thoracoscopy (generally known as video-assisted thoracoscopic surgery or VATS), (4) anterior cervical discectomy and fusion (ACDF), (5) lumbar laminectomy, (6) total knee arthroplasty (TKA), (7) total hip arthroplasty (THA), (8) inguinal hernia repair, (9) laparoscopic and (10) open appendectomy, (11) laparoscopic and (12) open cholecystectomy, (13) laparoscopic and (14) open colectomy, (15) laparoscopic and (16) open nephrectomy, (17) transurethral procedures, (18) cataract extraction, and (19) colonoscopy. These procedures are commonly performed and cover a spectrum of minimally painful procedures (cataract surgery) to invasive procedures associated with significant postoperative pain (thoracotomy). A common theme among these surgeries is the exposure to the perioperative period, including mental and physical preparation for the surgery as well as perioperative (or surgical) stress. Some of these surgeries were mainly done to relieve pain (total knee arthroplasty and total hip arthroplasty). We also included both available laparoscopic and open approaches for some thoracic and abdominal surgeries. Colonoscopy was included in this study since it is an invasive procedure and many colonoscopies are done with sedation. Patients were excluded if they had a second procedure within the first postoperative year.

In order to evaluate whether the exposure to surgery modifies the chance of opioid use, we calculated the baseline rate of prescription opioid use among a reference or control group [17, 18]. The control (reference) group was derived from a 10% random sample of veterans who did not have any surgical procedures between 1/1/2008 and 12/31/2015. Each member of the control group was assigned a sham surgery date chosen at random between 1/1/2008 or their enrollment date (whichever is later) and 12/31/2015. We excluded the controls who did not have any health encounters (at least one vital sign reading or a medical diagnosis) in the first year after their sham surgery date. Patients younger than 18 years and older than 85 years at the surgery date were excluded.

We aimed to study only patients without ongoing opioid use at the time of surgery (opioid-naive), and therefore, patients were excluded if they filled a prescription for the opioids within a 90-day period preceding their surgery date (real or sham). Patients with substance use disorders were not excluded since many of these patients are in remission and are not taking opioids. For these patients, surgery may precipitate relapse or opioid misuse, leading to persistent opioid use [19, 20].

Outcomes:

The duration of opioid use was defined as the length of opioid prescriptions (in days) from the surgery date to the first day of cessation. The cessation of opioid use was defined as once the patients did not fill any opioid prescription for 90 consecutive days, and the first day of this period is considered the cessation date [5, 21, 22]. The opioid medications were hydrocodone, oxycodone, hydromorphone, morphine, methadone, oxymorphone, codeine, and transdermal fentanyl (Supplemental Table 2). Only opioid agonists were included and the partial agonists (such as buprenorphine) and antagonists were not included. Tramadol was not included as it was not investigated in the previous studies [17, 18, 23]. During most time of the study (2008-2014), tramadol was not considered a controlled substance. The opioid doses were converted to morphine milligram equivalent (MME) [23]. The daily averages of opioid use for every postoperative month were calculated and recorded. Each patient was followed for the first 365 postoperative days.

The primary outcomes were the presence of ongoing postoperative opioid use at 90, 180, 270, and 365 days after the surgery. Patients were considered to be using opioids at any specific postoperative day if they continued filling the prescription opioids at that specified postoperative day and beyond (day 90, 180, 270, or 365). For example, patients were considered negative for persistent postoperative opioid use at 90 days if the duration of postoperative opioid use was 90 days and they did not receive any more prescriptions. The patients were considered positive for persistent postoperative opioid use at 90 days if the length of the duration of postoperative opioid use was 91 days or more.

Patient variables:

Data collected for each patient included demographic data (age at the surgery date, sex, race, and marital status), and presence of selected medical and mental health conditions (including substance use disorders), and preoperative use of selected medications. Race was recorded as white, African American, or others. Marital status was categorized as either married or single (including single, widowed, divorced, separated, and never married).

Medical and mental health conditions were identified by ICD-9 and ICD-10 codes (International Classification of Diseases, ninth and tenth revision, Supplement Table 3). These conditions were hypertension, diabetes, cancer, osteoarthritis, chronic obstructive pulmonary disease (COPD), hepatitis C, human immunodeficiency virus (HIV), depression, post-traumatic stress disorder (PTSD), bipolar disease, and substance use disorders (dependence or abuse of opioids, smoking/nicotine, alcohol, amphetamines, cocaine, cannabis, and sedatives), and opioid overdose. These comorbidities were assigned as being present if the corresponding diagnosis codes were documented in the medical records within the three years preceding the real or sham surgery date.

Preoperative use of some medications (opioids, antidepressants, antipsychotics, benzodiazepines, or gabapentinoids, Supplement Table 2) were also recorded due to the importance of their interaction with opioids. A patient was considered to use a medication if they filled a prescription for that medication within a 90-day period preceding their surgery date (real or sham).

Data Source:

We obtained data from the VA corporate data warehouse (CDW). This database houses the electronic medical records of more than 22 million veterans across the United States including diagnoses, inpatient and outpatient procedures, prescribed and dispensed medications, and demographic data. CDW was accessed securely through the VA Informatics and Computing Infrastructure (VINCI), with the approval from the appropriate VA authorities.

Identifying the association of surgery and persistent opioid use and identify the predictors:

We used multivariable logistic regression to examine the relationship between surgery and outcomes. We studied four distinct outcomes: persistent postoperative opioid use at postoperative days 90, 180, 270, and 365. For each outcome we ran two models, differing only in the way in surgery was entered into the model. In the first model, surgery was entered as a binary variable indicating whether the patient received any surgery or not. In the second model, surgery was entered as a categorical variable indicating which, if any of the 19 surgical interventions, the patient received. We therefore ran a total of eight analyses modeling opioid use at four time points, expressing surgery as a binary (yes no) variable or a categorial variable (representing 19 surgical interventions), 4x2=8. In all models, the reference group was the non-surgical controls. The logistic regressions included demographic data (age at the surgery date, sex, race, and marital status), surgical year, medical comorbidities (hypertension, diabetes, cancer, osteoarthritis, COPD, hepatitis C, HIV), mental health comorbidities (depression, PTSD, bipolar disease), substance use disorders (dependence or abuse of opioids, smoking/nicotine, alcohol, amphetamines, cocaine, cannabis, and sedatives), preoperative history of opioid overdose, and preoperative medications use (gabapentinoids, antidepressants, antipsychotics, benzodiazepines). These factors were included in the analysis based on plausible links to pain perception (such as pro-inflammatory conditions) [24–27] as well as longer opioid use (mental health conditions) [28–30]. The amount of the immediate postoperative opioid prescription was not included in the analysis, as it was felt that the type of surgery (e.g. cataract removal versus lumbar laminectomy) was a surrogate for the quantity of immediate postoperative pain and the associated dose and length of opioid prescription.

Sensitivity analysis:

Sensitivity analysis was performed to estimate the robustness of the association of the surgery and postoperative persistent opioid use to the potential unmeasured confounding factor. For this purpose, E-value for the association was calculated. E-value is the minimum strength of association that an unmeasured confounder should have with both the exposure and the outcome to fully explain away that specific exposure-outcome association [31, 32]. Greater E-values therefore indicate lower likelihood of an unknown confounder. The median E-value was reported to be about 2 [33].

Further Statistical Analysis:

The odds ratios and their 95% confidence intervals (95% CI) were reported. The incidences and their 95% CI were calculated with Wilson’s method [34]. For difference between the two groups, a Student’s t-test was used for continuous variable and chi-square was used for the binary or categorical variable. Standardized differences of means were calculated to demonstrate the effect size between the two groups. Using a Sidak correction for the eight comparisons, we accepted two-tailed p values <0.006 as statistically significant [35]. We used Stata version 15 (StataCorp, College Station, TX) and SAS version 9.4 (SAS Institute Inc., Cary, NC) for analysis.

3. RESULTS

A total of 183,430 distinct opioid-naive surgical patients and 1,318,894 distinct opioid-naive control patients met inclusion criteria (Supplemental Figure 1). The surgical group was older and had higher prevalence of medical conditions, mental health comorbidities, and substance use disorders compared with controls (Table 1). Demographics (age, gender [18]), medical comorbidities (hypertension [24, 25], diabetes [22], cancer[36], osteoarthritis [27–29]) and psychiatric comorbidities (mood disorders [17, 37], substance use disorders[38]) are known to affect pain perception. Therefore, these variables were included in the logistic regression when generating odds ratios.

Table 1.

Characteristics of the general population.

	surgical	control	P Value	Standardized differences of means **
Total	183,430	1,318,849
Age (years)	63.8 (0.026)	58.5 (0.015)	<0.001	−0.377
Male gender	95.4 (0.018)	92.7 (0.023)	<0.001	−0.114
Married	47.5 (0.043)	54.4 (0.043)	<0.001	0.139
Race			<0.001	−0.118
White	73.8 (0.038)	69.1 (0.040)
Black	16.2 (0.032)	15.4 (0.031)
Other	10.0 (0.026)	15.5 (0.032)
Surgical Year			<0.001	0.153
2008	9.7 (0.026)	9.9 (0.026)
2009	12.4 (0.029)	10.4 (0.027)
2010	15.1 (0.031)	11.2 (0.028)
2011	14.4 (0.031)	12.1 (0.028)
2012	13.7 (0.030)	13.1 (0.029)
2013	12.3 (0.029)	14.2 (0.030)
2014	11.4 (0.028)	14.6 (0.031)
2015	11.0 (0.027)	14.5 (0.031)
Hypertension	68.7 (0.040)	37.9 (0.042)	<0.001	−0.650
Diabetes	32.5 (0.041)	19.7 (0.035)	<0.001	−0.295
Hepatitis C	5.5 (0.020)	1.9 (0.012)	<0.001	−0.192
Cancer	35.4 (0.042)	12.8 (0.029)	<0.001	−0.548
Osteoarthritis	32.1 (0.041)	10.4 (0.027)	<0.001	−0.550
COPD	17.7 (0.033)	6.2 (0.021)	<0.001	−0.360
HIV	1.1 (0.009)	0.6 (0.007)	<0.001	−0.054
PTSD	13.6 (0.030)	10.6 (0.027)	<0.001	−0.092
Depression	25.8 (0.038)	15.8 (0.032)	<0.001	−0.248
Bipolar Disease	3.2 (0.015)	2.2 (0.013)	<0.001	−0.062
Smoking	29.5 (0.040)	13.4 (0.030)	<0.001	−0.401
Alcohol use disorder	9.1 (0.025)	4.0 (0.017)	<0.001	−0.207
Amphetamine use disorder	0.4 (0.005)	0.2 (0.004)	<0.001	−0.037
Cannabis use disorder	2.9 (0.015)	1.6 (0.011)	<0.001	−0.088
Cocaine use disorder	3.5 (0.016)	1.5 (0.011)	<0.001	−0.128
Sedative use disorder	0.4 (0.005)	0.2 (0.004)	<0.001	−0.037
Preoperative Opioid use disorder	1.7 (0.011)	0.9 (0.008)	<0.001	−0.071
Preoperative opioid overdose	0.2 (0.003), n=286*	0.0 (0.002), n=431*	<0.001	−0.042
Preoperative Medication use
Benzodiazepines	7.5 (0.023)	4.0 (0.017)	<0.001	−0.151
Gabapentinoids	9.9 (0.026)	3.9 (0.017)	<0.001	−0.238
Antidepressants	19.3 (0.034)	11.0 (0.027)	<0.001	−0.233
Antipsychotics	5.1 (0.019)	3.0 (0.015)	<0.001	−0.107

Data was presented as percentage (standard errors). The mean for age was reported.

^*For the rare occurrence of opioid overdoses, the exact number were also reported.

^**Standardized differences of means show the effect size between the two groups.

Duration and dose of postoperative opioid prescription:

Most surgical patients received prescriptions for opioid medications for one month or less (Supplemental table 4). Among surgical patients who received opioid prescriptions, patients who underwent cataract extraction and colonoscopies had the longest period of prescription refills. Patients after thoracotomy, ACDF, and lumbar laminectomy received longer periods of opioid prescriptions compared to other procedures, such as laparoscopic and open abdominal surgeries. 11.8 % of the patients with opioid prescriptions received doses larger than 50 MME. The prescribed doses remained unchanged from month to month (supplemental table 5).

Incidence of persistent postoperative opioid use:

In the postoperative period, 1.0 % of all the surgical patients continued to use opioids past the postoperative day 90, 0.6% continued to use it past the day 180, 0.4 % past the day 270, and 0.1 % past the postoperative day 365. After postoperative day 90, 1,474 surgical patients continued using opioids. Of these 1,474 patients, about 79% (1,168 patients) continued using opioids after postoperative day 180, 52% (774 patients) continued to use after day 270, and 23% of them (347 patients) continued to use opioid after the first postoperative year.

Association of the surgery with persistent opioid use:

Surgery was associated with 2-3 times higher odds of persistent postoperative opioid use compared to the controls at day 90, 180, 270, and 365 after adjusting for the patient variables (Figure 1 and supplemental table 6). Different surgeries had different incidences of postoperative opioid use, with the highest rates among post-thoracotomy and post-laminectomy patients. Even the minor procedures, such as cataract extraction or colonoscopy, had comparable rates of persistent postoperative opioid use to CABG or laparotomy cases (Table 2).

Figure 1.

The adjusted odds ratios (95% confidence intervals) of different variables associated with continued opioid prescription at various postoperative timepoints (days 90, 180, 270, and 365).

Table 2.

The incidence (in percentage, with 95% confidence intervals) of the persistent postoperative opioid use at day 90, 180, 270, and 365 after the surgical procedure.

Postoperative months	No.	30 days	90 days	180 days	270 days	365 days
Controls	1,318,849	0.9 (0.9 - 0.9)	0.2 (0.2 - 0.2)	0.1 (0.1 - 0.1)	0.1 (0.1 - 0.1)	0.03 (0.03 - 0.04)
All the surgical cases	183,430	16.0 (15.8 - 16.1)	1.0 (1.0 - 1.1)	0.6 (0.5 - 0.6)	0.4 (0.4 - 0.5)	0.1 (0.1 - 0.1)
CABG	23,002	21.4 (20.9 - 22.0)	0.7 (0.6 - 0.9)	0.4 (0.3 - 0.5)	0.3 (0.2 - 0.4)	0.1 (0.1 - 0.1)
Thoracotomy	1,019	28.0 (25.2 - 30.9)	2.9 (2.0 - 4.1)	1.9 (1.2 - 2.9)	1.7 (1.0 - 2.7)	0.3 (0.1 - 0.9)
VATS	8,667	21.6 (20.8 - 22.5)	1.8 (1.6 - 2.1)	1.0 (0.8 - 1.3)	0.6 (0.5 - 0.8)	0.2 (0.1 - 0.3)
ACDF	6,185	17.1 (16.2 - 18.1)	1.6 (1.3 - 2.0)	1.1 (0.8 - 1.4)	0.8 (0.6 - 1.1)	0.3 (0.2 - 0.4)
Lumbar Laminectomy	7,591	23.4 (22.5 - 24.4)	2.7 (2.4 - 3.1)	1.5 (1.2 - 1.8)	1.2 (1.0 - 1.4)	0.4 (0.3 - 0.6)
THA	9,935	22.3 (21.5 - 23.1)	1.1 (0.9 - 1.4)	0.6 (0.5 - 0.8)	0.4 (0.3 - 0.6)	0.1 (0.1 - 0.2)
TKA	19,812	36.9 (36.2 - 37.5)	1.8 (1.6 - 2.0)	0.8 (0.7 - 0.9)	0.5 (0.4 - 0.6)	0.2 (0.1 - 0.2)
Inguinal Hernia	6,605	7.3 (6.7 - 8.0)	0.4 (0.3 - 0.6)	0.3 (0.2 - 0.5)	0.2 (0.1 - 0.4)	0.1 (0.02 - 0.2)
Lap Appendectomy	6,442	7.5 (6.8 - 8.1)	0.3 (0.2 - 0.5)	0.2 (0.1 - 0.4)	0.2 (0.1 - 0.3)	0.1 (0.03 - 0.2)
Open Appendectomy	3,158	11.8 (10.7 - 13.0)	0.7 (0.5 - 1.1)	0.4 (0.2 - 0.7)	0.3 (0.2 - 0.6)	0.1 (0.02 - 0.2)
Lap Cholecystectomy	13,639	7.5 (7.1 - 8.0)	0.6 (0.5 - 0.8)	0.4 (0.3 - 0.5)	0.3 (0.2 - 0.4)	0.1 (0.03 - 0.1)
Open Cholecystectomy	3,962	16.1 (14.9 - 17.3)	0.9 (0.6 - 1.2)	0.6 (0.4 - 0.9)	0.5 (0.3 - 0.8)	0.2 (0.1 - 0.4)
Lap Colectomy	5,464	16.9 (15.9 - 17.9)	0.7 (0.5 - 0.9)	0.5 (0.3 - 0.7)	0.3 (0.2 - 0.5)	0.07 (0.01 - 0.14)
Open Colectomy	12,534	16.9 (16.2 - 17.5)	0.7 (0.6 - 0.9)	0.4 (0.3 - 0.5)	0.3 (0.2 - 0.4)	0.1 (0.05 - 0.2)
Lap Nephrectomy	2,063	13.1 (11.7 - 14.7)	0.4 (0.2 - 0.8)	0.2 (0.1 - 0.6)	0.2 (0.1 - 0.6)	0.1 (0.01 - 0.3)
Open Nephrectomy	4,066	17.0 (15.8 - 18.2)	0.8 (0.5 - 1.1)	0.4 (0.2 - 0.6)	0.3 (0.2 - 0.5)	0.2 (0.04 - 0.3)
Transurethral procedures	15,944	3.9 (3.1 - 4.9)	0.6 (0.5 - 0.7)	0.3 (0.3 - 0.5)	0.2 (0.2 - 0.3)	0.1 (0.1 - 0.2)
cataract extraction	2,048	5.9 (5.6 - 6.2)	0.6 (0.4 - 1.1)	0.5 (0.3 - 0.9)	0.3 (0.2 - 0.7)	0.1 (0.01 - 0.3)
Colonoscopy	31,293	5.7 (5.4 - 6.1)	1.0 (0.9 - 1.2)	0.6 (0.5 - 0.7)	0.4 (0.4 - 0.5)	0.2 (0.1 - 0.2)

Sensitivity analysis:

The E-values for the associations of the surgery and opioid use were calculated as 6.80 (95% CI 6.32-7.34) for postoperative opioid use at 90 days. In other words, an unmeasured confounding factor that is associated with surgery with an OR of 5.38 and with the opioid use at OR of 5.38 can explain away this association [31, 32]. E-values for postoperative opioid use were calculated as 5.15 (95% CI 4.66 – 5.69) at 180 days, 4.70 (95% CI 4.19-5.27) at 270 days, and 3.64 (95% CI 2.94-4.46) at 365 days.

Surgery-specific odds of persistent opioid use disorder:

Most surgeries/procedures were associated with higher odds of persistent postoperative opioid use at the studied time points (Table 3). The highest association with persistent postoperative opioid use were observed with thoracotomy and lumbar laminectomy. The minor surgeries (colonoscopy and transurethral procedures) were still associated with close to twice the odds of persistent postoperative opioid use compared to controls.

Table 3.

The surgery-specific adjusted odds ratio (95% confidence interval) of the opioid use at postoperative days 90, 180, 270, and 365. The reference is the controls. The logistic regression included all the patient variables in Table 1.

	90 days		180 days		270 days		365 days
Surgery / invasive procedure (compared to controls)	Odds ratio	P-Value	Odds ratio	P-Value	Odds ratio	P-Value	Odds ratio	P-Value
CABG	2.70 (2.52 - 2.89)	<0.001	2.03 (1.90 - 2.18)	<0.001	1.84 (1.72 - 1.96)	<0.001	1.71 (1.56 - 1.88)	<0.001
Thoracotomy	5.48 (4.45 - 6.73)	<0.001	4.14 (3.34 - 5.12)	<0.001	3.54 (2.85 - 4.40)	<0.001	3.21 (2.39 - 4.30)	<0.001
VATS	3.96 (3.64 - 4.32)	<0.001	3.31 (3.04 - 3.61)	<0.001	2.69 (2.46 - 2.93)	<0.001	2.68 (2.39 - 3.01)	<0.001
ACDF	3.71 (3.36 - 4.10)	<0.001	3.07 (2.79 - 3.38)	<0.001	2.81 (2.55 - 3.09)	<0.001	2.68 (2.37 - 3.04)	<0.001
Lumbar Laminectomy	4.98 (4.59 - 5.40)	<0.001	3.73 (3.43 - 4.05)	<0.001	3.53 (3.26 - 3.82)	<0.001	3.49 (3.15 - 3.88)	<0.001
THA	2.36 (2.14 - 2.60)	<0.001	1.92 (1.75 - 2.12)	<0.001	1.67 (1.52 - 1.84)	<0.001	1.76 (1.55 - 1.99)	<0.001
TKA	3.08 (2.89 - 3.29)	<0.001	2.22 (2.07 - 2.37)	<0.001	2.12 (1.99 - 2.25)	<0.001	1.92 (1.76 - 2.09)	<0.001
Inguinal Hernia	2.16 (1.87 - 2.49)	<0.001	1.76 (1.53 - 2.04)	<0.001	1.66 (1.45 - 1.91)	<0.001	1.84 (1.54 - 2.20)	<0.001
Lap Appendectomy	1.85 (1.60 - 2.15)	<0.001	1.79 (1.56 - 2.06)	<0.001	1.49 (1.29 - 1.71)	<0.001	1.63 (1.36 - 1.96)	<0.001
Open Appendectomy	2.64 (2.20 - 3.17)	<0.001	2.05 (1.70 - 2.47)	<0.001	1.89 (1.57 - 2.27)	<0.001	1.61 (1.24 - 2.09)	<0.001
Lap Cholecystectomy	2.27 (2.08 - 2.49)	<0.001	2.15 (1.97 - 2.34)	<0.001	1.91 (1.76 - 2.08)	<0.001	1.83 (1.63 - 2.05)	<0.001
Open Cholecystectomy	2.38 (2.03 - 2.80)	<0.001	2.47 (2.14 - 2.86)	<0.001	2.34 (2.03 - 2.69)	<0.001	2.07 (1.71 - 2.52)	<0.001
Lap Colectomy	2.76 (2.42 - 3.15)	<0.001	2.40 (2.11 - 2.74)	<0.001	2.13 (1.87 - 2.43)	<0.001	2.27 (1.91 - 2.69)	<0.001
Open Colectomy	3.64 (3.35 - 3.95)	<0.001	3.06 (2.82 - 3.32)	<0.001	2.69 (2.48 - 2.92)	<0.001	2.90 (2.61 - 3.23)	<0.001
Lap Nephrectomy	2.34 (1.88 - 2.92)	<0.001	2.27 (1.85 - 2.79)	<0.001	2.12 (1.73 - 2.59)	<0.001	2.22 (1.70 - 2.88)	<0.001
Open Nephrectomy	2.75 (2.38 - 3.18)	<0.001	2.20 (1.90 - 2.55)	<0.001	2.12 (1.83 - 2.44)	<0.001	1.92 (1.57 - 2.33)	<0.001
Transurethral procedures	3.15 (2.91 - 3.41)	<0.001	2.31 (2.12 - 2.50)	<0.001	2.32 (2.14 - 2.50)	<0.001	2.07 (1.86 - 2.31)	<0.001
cataract extraction	1.69 (1.31 - 2.19)	<0.001	1.60 (1.25 - 2.03)	<0.001	1.50 (1.18 - 1.89)	0.001	1.74 (1.30 - 2.34)	<0.001
Colonoscopy	2.67 (2.53 - 2.83)	<0.001	2.23 (2.11 - 2.36)	<0.001	1.95 (1.85 - 2.06)	<0.001	1.96 (1.83 - 2.11)	<0.001

Predictors of persistent postoperative opioid use:

Some variables, such as male gender, single marital status, or absence of preoperative cocaine use disorder, were associated with persistent opioid use mainly at early time points (90 and/or 180 days).

Preoperative opioid use has stronger trend of association at earlier time and preoperative opioid use disorder has stronger trend of association with the persistent opioid use at later time (270 and 365 days). Most other variables, such as osteoarthritis, hypertension, depression, hepatitis C, smoking/nicotine use, and preoperative use of gabapentinoids, antidepressants, and benzodiazepines, were associated with persistent use at all studied postoperative times. Hepatitis C and preoperative benzodiazepine use showed some trends with stronger association at later time points (Figure 1). The OR for preoperative opioid overdose was not calculated since none of the patients with preoperative OD received opioid prescription at 365 days after the surgery.

In response to the opioid epidemic, Veterans Affairs Medical Centers have implemented many policies for opioid management [39]. Therefore, the effect of the year of the surgery date on the rates of persistent postoperative opioid use was evaluated. Year 2013 was coincided with the highest rate of persistent postoperative opioid use (Supplemental Table 7).

4. DISCUSSION

In this historical cohort, we found about 1% of patients had ongoing opioid use 90 days after a surgery. About half of these patients continued to use opioids chronically at least for the next six months. Surgery was found to increase the odds of persistent opioid use by 2-3 times compared to nonsurgical controls. Both extensive and painful surgeries (such as lumbar laminectomy) and minimally invasive ones (colonoscopy, cataract removal) were associated with higher odds of persistent opioid use compared to controls. Different patient-related factors are associated with the postoperative opioids at different postoperative times and the strength of this association changes with time.

This analysis had several unique features and interesting findings. First, our clinical definition for persistent postoperative opioid use was based on whether the patient was still receiving opioid prescriptions at specific postoperative time points. This clinical definition may identify high-risk patients at an earlier time, beginning with day 90. Half of the patients who are using opioids at 90 days continue using the opioids past 270 days [5, 21].

Second, even the minor procedures, such as colonoscopy and transurethral procedures, were significantly associated with the elevated odds of persistent opioid use compared to controls. This may suggest that even small procedures confer a risk of developing persistent opioid use due to factors in the perioperative window (e.g. encounters with the health care system and perioperative exposure to opioids). In this dataset, the patients who received opioids with little justification for severe acute postoperative pain (reference group, colonoscopy and cataract extraction) actually had the longest lengths of opioid prescriptions. This emphasizes the importance of realistic goal setting, including the duration of opioid treatment, before initiating opioid treatment [40]. We need to discuss one possibility that some of these opioid-naïve patients, who had minor surgeries, might have started opioid use in the close proximity to their surgery date and were therefore captured in our analysis. This scenario by itself cannot explain the persistent opioid use among the patients with minor surgeries since the risk is still much higher compared to the controls. Therefore, some aspect of perioperative care and surgical exposure should be in motion to increase the risk of persistent opioid use.

Third, the predictors for persistent opioid use change throughout the postoperative course. Male gender and single status were associated with persistent opioid use at the two early timepoints, and hepatitis C and preoperative benzodiazepine use were more strongly associated at later timepoints. Therefore, the postoperative period is dynamic rather than homogenous, and different factors are important at different times. It is important to understand the trajectory of postoperative pain and the transition of acute postoperative pain to chronic postsurgical pain and differentiate the patients with persistent pain from the ones who may develop dependence on opioid medications [41–44].

As the main limitation, this study was not able to recognize any possible mechanism for persistent postoperative opioid use. Many patients may have persistent postoperative pain [43]. This study included medical conditions associated with pain, such as arthritis or cancer, but we did not find the “chronic pain” as reliable and objective diagnosis to be included in this analysis. Another reason for continuing opioid use may be the development of dependence or tolerance to opioids. These patients may continue opioid use long after surgery not to treat pain, but to avoid withdrawal symptoms. Unfortunately differentiating opioid dependence/tolerance from the postoperative pain is extremely challenging [45], and we did not have any method to determine the underlying cause of persistent opioid use in our study.

The study has other limitations. We only analyzed filled opioid prescriptions, not the actual opioid use. It is known that a significant portion of prescribed opioid medications remains unused after surgery [46]. Also, this database provided us with an enormous sample size. This huge sample size may produce statistically significant findings out of clinically insignificant factors. We used Sidak correction for multiple comparisons. Despite this, the many statistically significant results of this study should be clinically verified. We also studied a population of US Veterans, who are different from general population in the terms of medical access, as well as the impact from opioid crisis and mental health comorbidities [47, 48]. This might have limited the generalizability of this study. However, there are also advantages to using Veterans Affairs database. The follow-up time for most studies using administrative databases (using claim data from insurance carriers) is usually limited by the lack of continual insurance coverage and pharmacy benefit due to changes in insurance carriers [18]. Fortunately, the continued enrollment of the veterans in VA allowed us to follow patients long-term, including past age 65, without loss of data.

In conclusion, surgical and invasive procedures are significantly associated with persistent postoperative opioid use, and persistent postoperative opioid use occurs in about 1% of surgical patients. Even minor surgical procedures are associated with significantly elevated odds of persistent postoperative opioid use. Further, the postoperative period is a dynamic period and different factors are important at different time points. Given these data, prolonged opioid use may be considered a surgical risk and is worth discussing in preoperative visits with patients.

Highlights.

• About 1% of patients continue using opioids past the 90^th postoperative day. From these chronic postoperative opioid users, half continues to use opioids for the next six months.

• For the persistent postoperative opioid use, postoperative period is dynamic and the risk factors change with time.

• Being male and single are the risk factors for persistent opioid use at earlier time points (90 and 180 days), while the association of hepatitis C and preoperative benzodiazepine use with persistent opioid use are more pronounced at later time points.