Social Determinants of Long-Term Opioid Use Following Total Knee Arthroplasty

Abstract

Background.

Total Knee Arthroplasty (TKA) risks persistent pain and long-term opioid use (LTO). The role of social determinants of health (SDoH) in LTO is not well established. We hypothesize that SDoH would be associated with postsurgical LTO after controlling for relevant demographic and clinical variables.

Methods.

This study utilized data from the Veterans Affairs Surgical Quality Improvement Program, VA Corporate Data Warehouse, and Centers for Medicare and Medicaid Services, including Veterans aged ≥65 who underwent elective TKA between 2013-2019 with no postsurgical complications or history of significant opioid use. LTO was defined as >90 days of opioid use beginning within 90 days post-surgery. SDOH variables included the Area Deprivation Index, rurality, and housing instability in the last 12 months identified via medical record screener or ICD-10 codes. Multivariable risk adjustment models controlled for demographic and clinical characteristics.

Results.

Of the 9,064 Veterans, 97% were male, 84.2% White, mean age = 70.6 years, 46.3% rural, 11.2% living in highly deprived areas, and 0.9% with a history of homelessness/housing instability. Only 3.7% (n = 336) developed LTO following TKA. In a logistic regression model of only SDoH variables, housing instability (OR = 2.38, 95% CI: 1.09 – 5.22) and rurality conferred significant risk for LTO. After adjusting for demographic and clinical variables, LTO was only associated with increasing days of opioid supply in the year prior to surgery (OR = 1.52, 95% CI: 1.43 – 1.63 per 30 days) and the initial opioid fill (OR = 1.07; 95% CI: 1.06-1.08 per day).

Conclusion.

Our primary hypothesis was not supported; however, our findings do suggest that patients with housing instability may present unique challenges for postoperative pain management and be at higher risk for LTO.

Introduction

Postsurgical long-term opioid use (LTO) occurs in 7% of surgical patients [1]. LTO may reflect prolonged postsurgical pain given that 25% of patients who undergo total knee arthroplasty (TKA) report persistent pain even after successful, complication-free surgery [2]. Postoperative LTO varies significantly in prevalence between those taking opioids preoperatively versus those without previous opioid use [1]. LTO carries myriad risks including bowel dysfunction, endocrine dysfunction, excess sedation, respiratory depression, opioid use disorder, transition to heroin use, and fatal overdoses [3]. Therefore, identifying factors for post-TKA LTO among those with minimal prior opioid use is critical to manage risks.

Multiple studies have identified potential risk factors for developing postsurgical LTO [4-6]. A recent systematic review and meta-analysis identified the following risk factors for persistent postsurgical opioid use: female sex; lower levels of education; preoperative use of opioids, antidepressants, benzodiazepines, alcohol, cocaine, or tobacco; preoperative depressive and anxiety disorders; back pain, and fibromyalgia [1].

While previous data indicate a potential role for Social Determinants of Health (SDoH) in postsurgical LTO (e.g., level of education), the complex interrelations between SDoH, clinical acuity, and other risk factors are not well established. Area deprivation and rurality are associated with higher rates of opioid use in non-surgical populations [7-9], which may be driven by reduced access to non-pharmacological pain management in rural and socioeconomically deprived areas. However, the role of SDoH, such as rural residence, housing instability, and area deprivation, in conveying risk for post-TKA LTO has not been established.

The current study sought to determine whether SDoH were associated with risk for prolonged opioid use after TKA in patients with minimal prior opioid use and a complication-free postoperative course. Further, this study sought to determine whether SDoH were associated with LTO risk beyond the impact of patient demographics and case acuity. Rural-dwelling Veterans, those with housing insecurity/homelessness, and those living in areas of higher deprivation may be at increased risk for LTO following TKA. We hypothesize that SDoH would remain significant risk factors for LTO after controlling for relevant demographic and clinical variables.

Methods

Data Sources.

The primary data source was the Veterans Affairs Surgical Quality Improvement Program (VASQIP) [10]. These data are abstracted by trained surgical quality nurses and address various limitations of claims or electronic health record data [11-13]. Additional data sources included: 1) the Veterans Health Administration (VHA) Corporate Data Warehouse (CDW), used to supplement demographics (e.g., sex, age, race /ethnicity) and assess pre-operative care in VHA facilities, including prescription medications and diagnoses; 2) the VHA Program Integrity Tool (PIT) and VHA Fee-based domain, used to assess care rendered by community providers and reimbursed by VHA; 3) the Centers for Medicare and Medicaid Services (CMS) data on Medicare enrollment and fee-for-service Medicare claims among VHA enrollees, including prescription medications; 4) the VHA Planning Systems Support Group (PSSG) dataset, which provides geocoded patient addresses, including census block group for assigning rurality and neighborhood level deprivation; and 5) the Vital Status File, which provides mortality data from a variety of sources. The study was determined exempt by the VA Central Institutional Review Board

Population.

The cohort for the current study consisted of Veterans aged 65 or older who underwent elective TKA between 2013-2019 within the VHA, identified in VASQIP data using CPT code 27447 (from the primary procedure field). As such, Veterans were excluded based on 1) lack of Medicare enrollment or prior VHA use, because these were both required to assess care fragmentation; 2) postsurgical complications including in-hospital death, 30-day post-discharge readmission to a VHA or community-based hospital, transfer to another facility rather than discharge to home, or any postsurgical complications within 30 days following surgery (supplemental Table 1); or 3) prior significant opioid use, defined below (Figure 1). This study sought to examine SDoH as predictors of LTO among patients who were most likely to have a typical postoperative pain trajectory. As such, we excluded those with complications and significant prior opioid use.

Figure 1. Study exclusions among those receiving TKA from 2013-2019 (N = 16,700).*.

*There were initially 22,308 patients who underwent TKA in the VHA during this time, however, 5,608 were excluded from the parent study due to lack of Medicare enrollment or prior VHA use, resulting in the N = 16,700 TKAs in this study, to which we applied the additional exclusions above.

Opioid Variables.

Opioid medications were assessed via VHA administrative pharmacy data and CMS pharmacy data and included any outpatient prescription of solid oral dosage forms of butorphanol, fentanyl, hydrocodone, hydromorphone, levorphanol, meperidine, methadone, morphine, oxycodone, oxymorphone, pentazocine, tramadol, or codeine. The aim was to identify incident LTO following TKA, so patients were excluded if they had 1) history of long-term prescription opioid use in the year prior to surgery, defined as > 90 days using cabinet supply methodology (which counts the days prescribed and includes consecutive prescriptions as part of the same episode if the number of zero supply days between prescriptions is less than or equal to the number of days’ supply of the prior prescription) [14], or 2) significant opioid use directly prior to surgery, which was defined as ≥30 days in the 90 days prior to surgery (Figure 1). Those with significant prior opioid use were excluded because: 1) such prior use can complicate perioperative pain management, and 2) the primary outcome for this study was incident postoperative LTO, rather than rates of continued/discontinued LTO. The primary outcome for this study, incident postoperative LTO use, was defined as LTO (>90 days using cabinet supply methodology [14]) which began within 90 days post-discharge (i.e., occurring within 180 days post-discharge).

Social Determinants of Health.

The SDoH included rural residence, housing instability, and the Area Deprivation Index. Rural residence was defined using rural-urban commuting area (RUCA) categorized by urban, rural, or highly rural [15]. Homelessness/housing instability at any time in the 12 months prior to surgery was defined via: 1) two electronic health record health factors questions, on surveys administered at least once per year [16], or 2) ICD-10 codes indicating homelessness (supplemental Table 2). The Area Deprivation Index (ADI), which ranks neighborhoods by social disadvantage, was dichotomized as highly deprived (≥85) and less deprived (<85), or missing due to inability to assign a census block group, low population census block group, or group quarters [17].

Demographic and Clinical Variables.

Age at time of surgery, sex, race, and ethnicity were determined from VHA administrative data supplemented with CMS Enrollment information. Length of hospital stay refers to the number of days between admission and discharge, as found in the VASQIP dataset. We used the Risk Analysis Index (RAI) [18] to measure frailty at the time of surgery and the Gagne comorbidity score [19] to account for the impact of common chronic diseases. Care fragmentation was defined as the proportion of days receiving non-VHA care (i.e., Medicare-covered care or care paid by the VHA in the community) during the 12 months prior to surgical admission, relative to total days of care.

Data Analyses.

Descriptive statistics (mean/standard deviation and n/percent) were calculated for SDoH, demographic, and clinical variables to compare the study cohort and the excluded participants (based on either pre-surgical opioid use or postsurgical complexity: Figure 1) to assess the generalizability of study findings.

Logistic regression models assessed the relationship between SDoH and LTO while controlling for clinical and patient risk factors. The first model included only SDoH variables. The second model included SDoH variables and controlled for demographic and clinical variables. Models were estimated with generalized estimating equations to account for facility-level variation in opioid use. Analyses were preformed using SAS Enterprise Version 8.2.

Results

Cohort Characteristics.

Veterans in the study cohort (N=9,064) were 97% male, with 9.1% African American Non-Hispanic, 84.2% White Non-Hispanic, and 6.8% from other racial backgrounds. The mean age was 70.6 years (SD = 4.6). The cohort included 46.3% residing in rural or highly rural areas and 11.2% in highly deprived areas (ADI ≥85). Only 0.9% had a history of homelessness/housing instability in the prior year (Table 1). Compared to those who were excluded based on pre-surgical opioid use or postsurgical complications (Figure 1), the study cohort was slightly younger, had significantly fewer women (3.0% versus 3.8%), African Americans (9.6% versus 10.9%), more rural or highly rural patients (46.3% versus 41.7%), lower rates of homelessness in the prior year (0.9% versus 1.8%), fewer living in deprived areas, shorter length of stay, and less burden with comorbidities on average (Table 1).

Table 1:

Descriptive demographic and clinical variables.

	Study cohort (N=9,064) Mean(SD) or n(%)	Excluded (N=7,636) Mean(SD) or n(%)	p value
Age	70.59 (4.58)	71.49 (5.48)	< 0.01
Women	273 (3.01)	291 (3.81)	< 0.01
Race			< 0.01
African American non-Hispanic	823 (9.08)	835 (10.94)
White non-Hispanic	7,627 (84.15)	6,324 (82.82)
All others	614 (6.77)	477 (6.25)
Rurality			< 0.01
Urban	4,864 (53.66)	4,452 (58.30)
Rural	3,803 (41.96)	2,952 (38.66)
Highly Rural	397 (4.38)	232 (3.04)
Homelessness/housing instability	81 (0.89)	137 (1.79)	< 0.01
ADI ≥ 85%	1,020 (11.25)	1,020 (13.36)	< 0.01
Frailty (RAI_A)	25.35 (3.10)	26.01 (3.70)	< 0.01
Comorbidity (Gagne)	1.04 (1.82)	1.68 (2.17)	< 0.01
Care Fragmentation	0.15 (0.27)	0.16 (0.27)	0.53
Length of Stay (days)	3.86 (3.13)	4.47 (3.38)	<0.01

Rates of Post-TKA Opioid Use.

Overall, 3.7% of the cohort (n = 336) developed LTO following TKA.

Predictors of Post-TKA LTO.

In a logistic regression model that included only the SDoH variables (i.e., rurality, ADI, and homelessness), only homelessness/housing instability in the prior year conferred significant risk for progressing to LTO following TKA (OR = 2.38, 95% CI: 1.09 – 5.22), while rurality was associated with lower risk of LTO (OR = 0.78, 95% CI: 0.62 – 0.98). However, these effects were no longer significant (homelessness: OR = 2.03, 95% CI: 0.88 – 4.68; rurality: OR = 0.79, 95% CI: 0.62 – 1.01) after adjusting for demographic and clinical variables. In this second model, LTO was associated with increasing days of opioid supply in the year prior to surgery (OR = 1.52, 95% CI: 1.43 – 1.63 per 30 days) and days of the initial postsurgical opioid fill (OR = 1.07; 95% CI: 1.06-1.08 per day: Table 2).

Table 2.

Risk for long-term opioid use following TKA.

	Mode1 1: SDOH Variables	Model 2: Controlling for Demographics and Clinical Factors
	OR (95% CI)	OR (95% CI)
Residence (ref: Urban)
Highly Rural	0.80 (0.45, 1.43)	0.91 (0.50, 1.67)
Rural	0.78 (0.62, 0.98)	0.79 (0.62, 1.01)
ADI >85% (ref: <85%)	0.91 (0.63, 1.31)	0.86 (0.59, 1.26)
Missing (ref: <85%)	0.92 (0.45, 1.88)	1.22 (0.58, 2.56)
Homelessness/housing instability	2.38 (1.09, 5.22)	2.03 (0.88, 4.63)
Age		0.97 (0.93, 1.01)
Female Sex		1.33 (0.73, 2.43)
Race (ref: African American non-Hispanic)
White non-Hispanic		1.46 (0.85, 2.49)
All other		1.16 (0.77, 1.73)
Frailty (RAI_A score)		0.94 (0.88, 1.00)
Comorbidity (Gagne score)		1.03 (0.96, 1.10)
Care Fragmentation		0.82 (0.52, 1.30)
Length of Stay		1.02 (0.98, 1.05)
Number of days of opioid supply in the year prior (per 30 days)		1.52 (1.43, 1.63)
Days of initial opioid prescription		1.07 (1.06, 1.08)

OR = Odds Ratio; ADI = Area Deprivation Index

Discussion

Overall, this study found that patients with minimal prior opioid use and a complication-free postoperative course are at a low risk for developing LTO following surgery. Only 3.7% of these patients developed incident LTO following surgery. This is consistent with prior work which has examined rates of post-TKA LTO among those not taking opioids at the time of surgery and identified rates between 1-5% [1, 6, 20-22]. While overall rates of post-TKA LTO may be higher (i.e., 7%), this is due to inclusion of patients taking opioids prior to surgery [1]. Among those with LTO before surgery, over half continue LTO after surgery [6, 23].

Homelessness/housing instability was a risk factor for LTO in the current study. This is particularly notable, given the increased risk for opioid overdose and opioid-related hospital admissions among homeless patients [24]. However, the increased risk for homeless/housing-insecure patients attenuated after adjusting for clinical and demographic variables; the estimate continued to suggest an effect, however the confidence intervals were wide due to low patient numbers, resulting in non-significance. The loss of significance was likely caused by low statistical power, with few patients having both housing instability and receiving an elective TKA. Further work is needed to determine whether there are greater challenges with long-term postoperative pain management among those with housing instability, and to examine the possibility of whether patients with unstable housing may be given longer initial prescriptions due to concerns about accessing refills if needed. Further work is warranted with larger samples of the housing insecure to better understand this finding. In addition, future research targeting data sources with high rates of urgent and emergent surgeries may include larger populations of patients with housing instability compared to our sample of only elective procedures.

Of note, some SDoH that have been identified as risk factors for progressing to LTO use in non-surgical populations were not risk factors for LTO in the current study. Specifically, area deprivation and rurality have been found to predict risk for LTO in population-based studies [7, 8] but were not significant predictors of incident post-TKA LTO in this study. Further, in the current study, rurality was associated with lower risk of LTO in the model that only included SDoH. This suggests that rurality’s effect may vary by study selection criteria. Our population of older, rural adults with minimal prior opioid exposure undergoing elective TKA may be a very specific subset of all rural surgery patients. Further, the difference in findings across studies may be caused by excluding patients with significant prior opioid use. It is possible that some effects of SDoH on LTO risk have already been conferred prior to TKA, i.e., that many of those with higher risk of LTO had developed it prior to TKA and been excluded from the current analyses.

Total days of opioid use in the year prior to surgery and days supplied in the initial opioid fill following surgery were both significant predictors in the full model, which included SDoH, demographics, and clinical variables. Total number of days of opioid use in the prior year also conferred increased risk for progression to LTO. This is consistent with multiple prior studies in which the days of prior opioid use is often the most significant predictor of subsequent LTO [6, 25-29]. These findings underscore the importance of multimodal perioperative pain management and prescribing short timeframes of opioids, adjusting as needed [3, 30].

There are some limitations to this study. The demographics of this sample reflect the VA-served population, including being predominantly male and with longer average postsurgical hospitalizations [31], and thus findings may not generalize to the general population. Findings may be different among younger patients, those with significant prior opioid use, those with postoperative complications, those seeking care in non-VA settings. Further, patients who were excluded from the current study due to prior opioid use or a complicated postoperative course were more likely to be older, frailer, African American, women, urban living, and to have homeless/unstable housing, higher comorbidities, and higher area deprivation. Frailty was also statistically significantly higher among the excluded group; however this was not a clinically meaningful difference. Future work is needed to determine whether SDoH are differentially associated with postoperative LTO among a) those who experience a complicated postoperative trajectory, and b) those who were on LTO prior to surgery. Future work may also benefit from comparing outcomes among patients undergoing TKA who have prior trauma, surgery, or autoimmune disease to those without these histories. Finally, while we did not complete an a priori power analysis, there were clear limitations to power given the relatively low rates of postsurgical LTO and low rates of homelessness or housing insecurity. Larger populations of homeless or housing insecure adults undergoing surgeries would increase statistical power and better capture the risk of homelessness on developing postsurgical LTO.

Conclusions

Number of days of opioid supply in the year prior to surgery and days supplied in the initial postsurgical opioid fill were significant predictors of LTO. Homelessness/housing instability was also a significant predictor of postoperative LTO, however, the small sample size of housing insecure individuals complicated interpretation when demographic and clinical variables were entered into the model, reducing power and rendering the finding non-significant. Larger study populations of homeless or housing insecure patients would improve research in postoperative pain management and risk for incident post-surgery LTO.