Association between preoperative obstructive sleep apnea diagnosis and prolonged postoperative opioid use among patients undergoing common surgical procedures.

This study followed the STROBE reporting guidelines and was included in the umbrella Institutional Review Board protocol for de-identified data managed by the Center for Population Health Sciences at Stanford University (Stanford, California; PHS-40974), which included a waiver of consent. A pre-specified analysis plan defining the study sample, set of associated variables, and data analysis plan was written, date-stamped, and recorded in the investigators’ files before data were accessed. We used data from the Merative ™ MarketScan ® Commercial Claims Database 14 and the Merative ™ MarketScan ® Medicare Supplement , 15 both de-identified databases derived from a large, adjudicated claims data warehouse, which contains claims for over 250 million United States patients enrolled in commercial insurance and Medicare fee-for-service insurance plans. The data include inpatient, outpatient, and pharmaceutical claims. We used these databases to investigate the occurrence of prolonged postoperative opioid use in patients aged ≥18 years with and without OSA who underwent any of 11 common surgeries (total knee arthroplasty; total hip arthroplasty; laparoscopic appendectomy; cholecystectomy [laparoscopic or open]; or operative management of small bowel obstruction, diverticulitis, hernia [inguinal, femoral, or umbilical], testicular or ovarian torsion, ectopic pregnancy, and aortic aneurysm). These surgeries represent a variety of elective and emergent general surgical, urologic, gynecologic, vascular, and orthopedic procedures that are associated with different levels of invasiveness and postoperative pain. 16 First, we constructed an initial sample of patients who underwent the selected surgeries between January 1, 2016 and December 31, 2021, restricted to the first surgery found for each patient and limited to those who did not undergo any additional surgeries in the 365 days following the initial date of surgery and who were continuously enrolled for 730 days before and 365 days after date of surgery (N=411,147). We then excluded patients with a history of opioid use disorder (N=1,014), and patients who filled an opioid prescription in the year prior to the date of surgery (N=102,039). The final dataset included 270,320 opioid-naïve patients (see Figure 1 for a cohort construction flowchart). No patients in this cohort underwent more than one surgery during the study period. Our exposure was OSA, which we defined based on a pre-specified, validated claims-based algorithm 17 with sensitivity of 24.1% and specificity of 67.8%: the low sensitivity implies that it does not capture many patients with OSA. We used a 2-year (730-day) look-back period to isolate patients who had two or more claims with an outpatient or inpatient ICD-10 diagnosis code of G47.3x. Our primary outcome was prolonged postoperative opioid use, defined as having filled ≥10 prescriptions (for any expected days’ supply) or ≥120 days’ supply of any of the included opioid medications within the first year of surgery, excluding the first 90 postoperative days (i.e., postoperative days 91–365). 18 We identified opioid medications by using the MarketScan Redbook 19 to isolate all prescriptions for hydrocodone, oxycodone, morphine, fentanyl, hydromorphone, oxymorphone, and meperidine in capsule, tablet, lozenge, or patch formulations. Our secondary outcome was the number of days of prolonged postoperative opioid use, calculated for the subset of the study population with prolonged postoperative opioid use. We chose a set of covariates a priori to adjust for potential confounders in our statistical analyses based on previous research and/or biological plausibility. First, we included age and sex. Second, we included indicator variables for Elixhauser comorbidities, 20 excluding metastatic cancer and solid tumors without metastases, as patients with cancer were excluded. We identified the presence of a comorbidity using at least 1 claim with relevant ICD-10 codes in the 730 days before their surgery. Third, we included indicator variables for diagnoses of psychiatric disorders (anxiety, attention-deficit/hyperactivity disorder, bipolar disorder, non-tobacco and non-opioid use disorder substance use disorders, and post-traumatic stress disorder) and chronic pain syndromes, using validated sets of codes. 21 , 22 Fourth, we included indicator variables for year fixed effects, which allowed us to assess the relationship between OSA and prolonged postoperative opioid use within individual years to account for changes in clinical knowledge and policy over time. Fifth, we included indicator variables for surgical procedure fixed effects given that patients may have different risks of prolonged postoperative opioid use for each surgery. Finally, we included indicator variables representing insurance status for enrollment in a commercial or Medicare fee-for-service plan. First, we compared demographic variables, comorbidities, type of surgery, year, and insurance status in patients with and without OSA. We calculated absolute standardized mean differences (SMD) to compare covariates between groups; we identified a SMD greater than 0.1 as reflective of meaningful differences between groups. Next, we estimated the association between OSA and prolonged postoperative opioid use, our primary outcome, using a multivariable logistic regression model. Based on previous estimates of prolonged postoperative opioid use, 1 we determined that a clinically meaningful change in the outcome would be greater than 0.3% (relative change of approximately 30% assuming an overall incidence of approximately 1.0%). We converted odds ratios to predicted means and average marginal effect with 95% confidence intervals for ease of interpretation. The average marginal effect shows how much the risk of prolonged postoperative opioid use is expected to change with the exposure to OSA, while holding other factors constant. Finally, we estimated the association between OSA and our secondary outcome, the number of days of prolonged postoperative opioid use, using a multivariable ordinary least squares regression adjusting for the same covariates as those enumerated above. The secondary outcome analysis was restricted to those patients who had at least 1 day of opioid utilization in the 91–365 days following the date-of-surgery. All analyses were performed using STATA 17.0 (College Station, TX) and R version 4.1.1. We conducted several sensitivity analyses to evaluate the robustness of our findings. First, we repeated our baseline analysis using less stringent criteria for assigning OSA to optimize the sensitivity of the definition: rather than assigning patients to the OSA group if they had two relevant claims in the prior 730 days, we included them if they had one claim. Second, we conducted the analysis using a more stringent definition of OSA, requiring patients to have at least 3 relevant claims, and excluded those with only 1 or 2 claims of OSA. Third, we used a more specific definition of OSA, requiring patients to have at least 2 relevant claims and the presence of comorbidities strongly associated with OSA (stroke, 23 myocardial infarction, 23 diabetes, hypertension). 17 Fourth, we respecified the outcome using a more liberal definition of prolonged postoperative opioid use, first redefining the outcome as filling at least one opioid prescription between 91 and 365 days post-surgery 24 , and then redefining the outcome to filling at least one opioid prescription in the 1–90 days and then again in the 91–365 days following discharge. 24 Then, we repeated our analysis restricted to the following smaller cohorts: 1) patients we expected to have more severe postoperative pain (i.e., those who underwent total knee arthroplasty, total hip arthroplasty, cholecystectomy [laparoscopic or open], or operative management of small bowel obstruction, diverticulitis, or hernia), 16 2) patients without pre-existing diagnoses that could confound opioid use, including chronic pain conditions and psychiatric disorders including substance use disorders, and 3) patients with and without insomnia. 25

Our final sample included 270,320 patients who underwent one of the 11 selected surgeries and met the inclusion criteria. Of these, 29,095 (10.8%) met diagnostic criteria for OSA and 241,225 (89.2%) did not. Baseline patient characteristics, including demographics, medical comorbidities, surgical procedure, insurance type, and incidence of persistent postoperative opioid use are listed in Table 1 . Patients with OSA were more likely to have comorbidities that are commonly associated with this condition, including obesity, hypertension, and diabetes. In this cohort, patients with OSA were older and comprised a higher proportion of females compared to those without OSA. The unadjusted incidence of prolonged postoperative opioid use was 0.6% (95% CI 0.5–0.7) for patients with OSA and 0.4% (95% CI 0.4–0.4) for patients without OSA, with an average unadjusted marginal effect of OSA on prolonged postoperative opioid use of 0.2% (95% CI 0.1–0.3). Table 2 presents adjusted estimates of our primary and secondary outcomes. After adjusting for covariates including age, sex, comorbidities, type of surgery, insurance type, year of surgery, we found no difference between the two groups (adjusted incidence for those with OSA 0.4% [95% CI 0.4 to 0.4%] and for those without 0.4% [95% CI 0.3 to 0.4%], with an average marginal effect of 0.0% (95% CI −0.1% to 0.0%; p=0.184). We also found no difference in outcomes after stratifying patients by the presence or absence of insomnia ( Table 3 ). Analysis of the secondary outcome – number of days of prolonged postoperative opioid used – was restricted to patients with prolonged postoperative opioid use. Patients with OSA and prolonged postoperative opioid use had a median 140 days of opioid use in the outcome period (95% CI 84–240), while those without OSA had 154 days (95% CI 98–260). Adjusted analysis found no difference in the number of days of prolonged postoperative utilization for those with versus without OSA (−25.1 days, 95% CI −51.0–0.8, p=0.058). The estimated association between OSA and prolonged postoperative opioid use was upheld in several sensitivity analyses (also in Table 2 ). The findings were replicated using alternate definitions of OSA (Sensitivity Analysis 1 through 3), alternate definitions of prolonged postoperative opioid use (Sensitivity Analysis 4 and 5), and when restricting the cohort to surgeries with higher expected postoperative pain and to patients without diagnoses of conditions known to be associated with prolonged postoperative opioid use and/or OSA (Sensitivity Analysis 6 and 7; insomnia subgroup analyses Table 3 ).

In this study of 270,320 patients who underwent common surgical procedures, we found no additional statistically or clinically relevant risk of prolonged postoperative opioid use associated with having a preoperative diagnosis of OSA. The overall rate of prolonged postoperative opioid use was approximately 0.4% in our study population, in line with other studies estimating this outcome with a similar definition. 26 We also found that – among those who exhibited prolonged postoperative opioid use – those with OSA experienced no statistically significant increase in days of opioid utilization. OSA is a widespread condition that is frequently encountered in surgical populations. While patients are often screened preoperatively for OSA, these screenings are usually meant to identify patients who might be difficult to mask ventilate or intubate, require postoperative treatment with noninvasive positive pressure ventilation, and/or be particularly prone to opioid-induced respiratory depression in the immediate perioperative setting. 27 Despite the evidence linking sleep and prolonged opioid use secondary to poorly controlled pain, our findings imply that this preoperative diagnosis is not associated with a clinically-relevant elevated risk for potentially inappropriate and prolonged opioid utilization stretching for months beyond their surgery. These results diverge from recent evidence mostly from Veterans Administration surgical 28 and non-surgical populations. 29 In these populations, 1) patients with both chronic pain and OSA were twice as likely to receive opioids after adjusting for pain intensity compared to patients without OSA, 29 and 2) a preoperative insomnia diagnosis (a condition highly comorbid with OSA 30 ) predicted greater chronic opioid utilization and health care costs in the two years following arthroscopic hip surgery. 28 This discrepancy in the relevance of OSA as a risk factor for prolonged opioid use implies that the relevance of OSA as a risk factor for prolonged postoperative opioid use may differ based on population. One possible explanation is that clinicians at certain sites may adopt more cautious opioid protocols and enhanced perioperative monitoring for patients with OSA. In addition, the low overall prevalence of OSA may have limited our ability to detect differences between groups. Because of the previous mixed evidence and our findings of a small, non-clinically significant increase in risk of prolonged postoperative opioid use with a diagnosis of OSA with a more liberal definition, the association between OSA and prolonged postoperative opioid use still requires more consideration. Patients with OSA might have enhanced pain sensitivity 31 and higher overall opioid responsiveness 32 which predict opioid misuse in patients with chronic pain. Finally, surgery may further worsen sleep disorders such as OSA and insomnia, 28 , 33 , 34 partly due to postoperative opioid use. Patients with OSA who undergo surgery may be more at risk for deleterious effects of those opioids including the development of opioid dependence and long-term use, and future work may investigate how opioid prescriptions in the immediate postoperative period may modify the relationship between OSA and long-term opioid-related outcomes. Our study should be viewed considering its limitations. First, as a retrospective study based on claims data, it is subject to residual confounding from factors that we were unable to measure. For example, we defined prolonged postoperative opioid use using claims for opioid prescription fills; however, these data may not reflect whether the patient used the medication as prescribed and/or whether patients obtained opioid medications through means outside of legal prescription. Second, our data did not include inpatient postoperative opioid prescribing, which may influence downstream opioid use. Third, our data did not include OSA severity, treatment adherence, utilization of multimodal analgesia, functional status, perioperative pain scores, in-hospital opioid administration, and psychosocial determinants of health. Future prospective studies should examine whether these factors can help identify OSA patients at higher risk for prolonged opioid use. Fourth, our claims-based 17 definition was limited in both sensitivity (approximately 30%) and specificity (approximately 70%), and further was originally validated in Canada, which increases the risk of misclassification bias and could skew the results of our study to no effect. In other words, it is possible that we underestimated the extent of association between OSA and prolonged postoperative opioid use. To test this potential limitation, we conducted additional analyses to increase the sensitivity of this algorithm by modifying the definition of OSA using different thresholds for the number of claims used to define the exposure, and by adding additional relevant comorbidities. These analyses replicated the findings in our baseline analysis. Fifth, the study period ended in 2021: assuming that strategies to minimize long-term opioid use have improved since then, the low absolute values of our risk estimates may still be overestimates when applied to current clinical practice conditions. Sixth, findings from our dataset, representing commercially- and Medicare-insured populations, may not generalize to other populations. Finally, our dataset did not contain the level of clinical detail to account for severity of OSA in its association with prolonged postoperative opioid use. Future research using additional measures from polysomnography may enhance our understanding of the relationship between OSA and postoperative opioid use.

In summary, we found no relative increase in the probability of prolonged postoperative opioid use in patients with OSA who underwent common surgical procedures. These findings suggest that optimization of preoperative OSA – while clinically important for intraoperative and immediate postoperative management – may not be a high-yield target in the effort to reduce the incidence of prolonged postoperative opioid use, although additional research using non-claims-based data is still needed.

In light of the opioid epidemic and increased focus on perioperative opioid use, 1 , 2 recent efforts have prioritized identifying modifiable risk factors for prolonged postoperative opioid use – ongoing opioid use well beyond typical recovery 1 , 3 , 4 – and implementing interventions to improve postoperative pain management. 1 , 3 – 5 However, whether a preoperative diagnosis of obstructive sleep apnea (OSA) is associated with prolonged postoperative opioid use remains unknown. Understanding this potential association is important because OSA is a modifiable, patient-level risk factor that can be optimized preoperatively. Patients with OSA can initiate positive airway pressure therapy prior to elective surgery, which can 6 mitigate the impact of hyperalgesia as a sequela of poor sleep quality. 6 – 7 , 8 9 More than 35% of surgical patients are estimated to be affected by OSA, yet approximately 80% of them lack a formal diagnosis. 9 Plausible biological mechanisms could explain this association between OSA and prolonged postoperative opioid use. 10 Both intermittent hypoxemia and sleep fragmentation, two hallmark phenotypes of OSA, promote hyperalgesia, creating a period of vulnerability to poor pain control and increased opioid usage. 11 – 13 In this study, we examined the association between preoperative OSA and prolonged postoperative opioid use among patients undergoing common surgical procedures in the United States. We used a validated algorithm to identify patients with OSA and then compared their risk of prolonged postoperative opioid use to those without OSA. We also examined the association between OSA and the duration of opioid use among patients who met criteria for prolonged postoperative opioid use. We hypothesized that a preoperative diagnosis of OSA would be associated with an elevated risk of prolonged postoperative opioid use and longer duration of opioid use.