Incident functional limitations among community-dwelling adults using opioids: A retrospective cohort study using a propensity analysis with the Health and Retirement Study

Abstract

Background:

Opioid analgesics are commonly used to manage pain; however, it is unclear how they affect patient function. This study examines the association between opioid analgesics and incident limitations in activities of daily living (ADL), instrumental activities of daily living (IADL), and cognitive functioning among community-dwelling older adults.

Methods:

Data included 10,003 participants of the 2016 and 2018 waves of the Health and Retirement Study, which sampled United States adults aged 51–98 years old. The primary exposure was self-reported opioid pain medication use in 2016. Outcomes included incident limitations in ADL, IADL, and cognitive functioning in 2018. Statistical methods adjusted for confounding using multivariable logistic regressions, inverse probability of treatment weighting, and propensity scores.

Results:

Opioid use (Adjusted Odds Ratio [aOR]: 1.34, 95% Confidence Interval [CI]: 1.07–1.68) was associated with a statistically significant higher odds of incident ADL limitation in multivariable regression and in propensity score adjustment (aOR: 1.41, 95% CI, 1.13–1.76). The association between opioid use and ADL and IADL limitations was modified by age. Adults < 65 years old had higher odds of incident ADL (aOR: 1.83, 95% CI, 1.38–2.42) and IADL (aOR: 1.42, 95% CI, 1.06–1.90) limitations compared to those ≥65 years old.

Conclusions:

Community-dwelling adults using opioid analgesics to manage pain may be at risk for incident ADL limitations. Middle-aged adults, compared to those older than 65 years of age, experienced the greatest odds for incident ADL and IADL limitations following opioid use. According to sensitivity analyses, our findings were robust to unmeasured confounding.

1. INTRODUCTION

Pain is highly prevalent among adults in the United States. Findings from the 2016 National Health Interview Survey, a nationally representative survey of adults in the United States, indicated ≥25% of adults forty-five years or older report experiencing pain on most days [1]. The prevalence of chronic pain increases with age [2]. The public health impact of pain management is salient because the U.S. population is aging. Chronic pain, like aging, is associated with higher healthcare spending, reduced functional independence, and a greater reliance on medications [3]. Estimates for the annual cost of pain in community-dwelling adults range between $560–635 billion per year in America [3].

Opioids were prescribed to 22.1% of United States adults with chronic pain in 2019 [4]. The use of prescription opioids can be appropriate in some circumstances [5], but the potential benefits must be carefully weighed against the possibility for adverse events. Opioid-related adverse events including falls, hospitalizations, and misuse or abuse can lead to new functional limitations [6–8]. Similarly, polypharmacy and co-prescribing of opioids and benzodiazepine substitutes can lead to falls and fractures that disrupt participation in valued activities [8]. The overuse of prescription opioids to manage pain is a major concern—particularly for community-dwelling older adults—in the context of the United States opioid epidemic. Despite policy [9] and practice [5] initiatives to reduce prolonged opioid prescribing, 18.4% of all Medicare Part D beneficiaries [10] and 27.7% of beneficiaries with disabilities [11] continue to receive prescription opioids. From 2016 to 2017, the rate of United States prescription opioid-related overdose deaths decreased or remained stagnate in all age groups except those ≥65 years who experienced a 10.5% increase [12].

The frequent use of prescription opioids among adults has increased interest in the functional and health outcomes of adults who have been prescribed opioids. The Agency for Healthcare Research and Quality conducted a rigorous comparative effectiveness review, which found opioid analgesics were associated with small short-term (≤6 months) improvements in function compared to placebo [13], Short-term functional improvements may be a result of better pain management as population level research shows prescription opioid use does not have an appreciable benefit on disability, quality of life, or mental health [14]. Results from randomized controlled trials studying the likelihood of short-term functional improvement following opioid analgesics remain equivocal [13, 15, 16]. Long-term (>1 year) outcomes of upper body and lower body physical impairment have not improved in patients using opioids for chronic non-cancer pain [17]. These studies have contributed to our understanding of functional outcomes among adults using prescription opioids; however, none have simultaneously analyzed large nationally representative samples, outcomes longer than one-year, and measured activity limitations in activities of daily living (ADL), instrumental activities of daily living (IADL), and cognitive function. Our study addresses these knowledge gaps by incorporating a large nationally representative sample and a follow-up period of 2-years. We also measure specific ADL, IADL, and cognitive function outcomes, which are critical for meaningful and independent participation within the community.

The purpose of this study was to investigate the association between any opioid use and incident functional limitations among community-dwelling middle-aged and older adults. We hypothesized the odds of incident limitations in ADL, IADL, and cognition were greater in community dwelling older adults who self-reported any opioid use to manage pain compared to those who were naïve to opioids. To mitigate the risk for bias from confounding and to demonstrate the robustness of our findings, we report our findings with multivariable adjustment, inverse probability of treatment weighting, propensity score adjustment, and propensity score matching.

2. METHODS

2a. Data Sources

Data for this retrospective cohort study came from the 2016 and 2018 observation waves of the Health and Retirement Study (HRS), which is publicly available https://hrs.isr.umich.edu/data-products [18]. The HRS is an ongoing, nationally representative longitudinal study of adults aged > 50 years in the United States. Age is truncated at 98 years to protect anonymity for adults exceeding this age. The HRS began in 1992 and observation waves are completed every two years. As of 1998, the HRS has replenished the sample every 6 years to maintain the national representativeness of the sample. The 2016 wave served as our baseline data because this was the first observation wave in which participants were asked about the use of opioid analgesics for pain management. The 2016 observation wave also included a replenishing cohort of participants born between 1960–1965 (Late Baby Boomers). A follow-up observation was completed in 2018 to establish temporality between our primary exposure and outcomes. The HRS (Health and Retirement Study) is sponsored by the National Institute on Aging (grant number NIA U01AG009740) and is conducted by the University of Michigan. The University of Michigan’s Institutional Review Board approved the HRS. This secondary analysis of de-identified public data does not qualify as human subjects research.

2b. Participant Cohort

The selection of the participant cohort is presented in Figure 1. Inclusion was restricted to community-dwelling self-respondents who were functionally independent in 2016. Participants were excluded from the final sample if they resided in a nursing home, required a proxy interview, had missing covariate data, were deceased or unable to respond in 2018, or had baseline limitations in ADL, IADL, or cognition. We excluded those with baseline ADL, IADL, or cognitive limitations to measure incident functional limitations in a subgroup of functionally independent adults. A total of 10,003 community-dwelling self-respondents who were independent in ADL, IADL, and cognitive function were included in our analysis.

Figure 1.

Cohort flow diagram

Note: Unweighted raw frequencies depicted in Figure 1.

2c. Outcomes

The primary outcomes of interest were incident functional limitations in the follow-up 2018 wave. Functional limitations were operationalized as any incident ADL, IADL, or cognitive limitation. Since the 2016 baseline cohort reported no limitations in these areas, any new report of ADL, IADL, or cognitive limitation in 2018 was defined as an incident functional limitation. These HRS outcomes have established validity and are generalizable to clinical practice [19, 20]. Furthermore, the HRS relied on seminal rehabilitation science theory to develop measures of ADL [21] and IADL [22] limitations. The ADL limitations included at least one self-reported difficulty in: dressing, bathing, eating, bed mobility, toileting, or household mobility. The IADL limitations included at least one self-reported difficulty in: meal preparation, grocery shopping, making phone calls, taking medications, or managing finances. These outcomes were dichotomized (no, yes).

Cognitive functioning was measured using a modified 27-point version of the Telephone Interview for Cognitive Status (m-TICS) [20]. The m-TICS includes the immediate and delayed recall of a 10-word list (memory; 0–20 points); serial 7 subtractions (working memory; 0–5 points), and correctly counting backwards from 20 to 10 as quickly as possible (attention and processing; 0 points: incorrect on two attempts, 1 point: correct on second attempt, 2 points: correct on first attempt). As in previous research, participants scoring 12 or higher on the m-TICS were dichotomized as having no cognitive limitations and those who scored below 12 were identified as having cognitive limitations [23].

2d. Primary Exposures

The primary exposure was recent use, within 3 months of the interview, of opioid pain medications during the baseline 2016 survey wave. Opioids were classified as a schedule II narcotic by the U.S. Drug Enforcement Agency in 2014 and require a prescription [9]. History of opioid use was determined by asking participants if they had taken any opioid medications for pain in the past 3 months. The HRS does not have data on dosage, duration, or pro re nata use. Compared to medical record data, survey-based self-reported prescription narcotic use among community-dwelling adults elicited moderate agreement (kappa=0.40) with a sensitivity of 0.51 and specificity of 0.90 [24]. This variable was dichotomized (no, yes) by merging non-affirmative categories with “No”. Participants were prompted to recall opioid use by medication names such as: Vicodin, OxyContin, codeine, morphine, or similar medications. History of opioid use included: “Naive” (N=8,932, 89.29%) and “Opioid use” (N=1,071, 10.71%).

2e. Covariates

Confounding variables included demographic characteristics and self-reported health status. These were defined a priori using prior research and clinical expertise [10, 11]. Demographic characteristics included age, sex (female, male), race/ethnicity (Hispanic, non-Hispanic Black, non-Hispanic White), education (less than high school, General Educational Development or high school degree, some college, or college and above), Census region (Northeast, Midwest, South, or West), and healthcare payor (Private insurance, Medicaid, or Public). Self-reported health status included use of over-the-counter pain medication such as Advil, Aleve, Tylenol, and aspirin in the past three months (no, yes), an overnight hospital stay (no, yes) or major surgery (no, yes) in the past two years. Baseline comorbidities (0, 1, 2, 3+) included diabetes, cancer, lung disease, heart disease, stroke, emotional, nervous, or psychiatric condition, arthritis, osteoporosis, hip fracture, or dementia. Smoking status and alcohol use were categorized as never, former, and current. Depressive symptoms (no, yes) were measured using an eight-item version of the Center for Epidemiologic Studies Depression Scale (CESD). Participants who scored three points or higher were classified as having depressive symptoms [25]. To measure pain severity, participants were asked, “How bad is the pain most of the time: mild, moderate, or severe”.

2f. Analysis

Univariate and bivariate analyses were completed with raw frequencies and percentages, chi-square for categorical variables, and t-test for continuous variables to assess the association between the primary exposure of opioid use and all variables. Due to the observational nature of this study, we addressed confounding using multivariable adjustment, a marginal structural model estimate of the inverse probability of treatment, and propensity score models. We used all of these analyses because each has different assumptions for the homogeneity of respondents between levels (yes, no) of opioid exposure.

Three multivariable logistic regressions were used to separately assess the adjusted odds of incident ADL, IADL, or cognitive limitations following opioid use. A collinearity assessment via correlation matrix led to the decision to collapse all health condition variables into a single “Number of Comorbidities” variable, which was broken into quartiles (0, 1, 2, 3+) based on the data distribution. Model fit was assessed for each of the three models with AIC, c-statistics, and a likelihood ratio test comparing the unadjusted model to the final model. The functional form of age was kept as continuous because the addition of higher order terms did not improve model fit. Due to the large number of variables and sample size, model diagnostics were assessed graphically. Inferential tests were conducted as two-sided analyses using Wald statistics with 95% profile likelihood confidence intervals.

In addition to multivariable adjustment, we addressed confounding using a marginal structural model estimated by inverse probability of treatment weight (IPTW). The IPTW estimated each respondent’s probability of using opioids, given their covariate distribution. A logistic regression was used to regress opioid use on all confounders to obtain the denominator of our weight. Then, we regressed opioid use on an empty model to obtain the numerator of our weight. This weight was stabilized to avoid overweighting individual cases. Finally, we ran a logistic regression using the stabilized weight to obtain an estimate for the effect of opioid use on each outcome.

Propensity score analyses were used to balance the measured confounding variables across levels of the opioid use exposure. All measured covariates were included in a stepwise logistic regression model to calculate the propensity that a survey respondent would report using opioid analgesics. This propensity score was used in three analyses. First, we used subclassification by stratifying our outcome analyses on propensity score quintile. Second, we adjusted for the propensity score as a covariate in our outcome analyses. Third, we matched respondents who reported opioid use to those who did not use opioids. A greedy nearest neighbor matching algorithm was used with a ratio of 1:1 without replacement. The logit of the propensity score was used as the distance metric with a narrow caliper width equal to 0.20 standard deviations.

The assumptions for IPTW and propensity score analyses were assessed using balance, comparability of estimates across quintiles, and overlap. IPTW balance was achieved as the mean for stabilized weights equaled 1.0 (range: 0.14–5.5). For the subclassification model, Cochran-Mantel-Haenszel tests were used to assess estimates of the association between opioid use and each confounder across propensity score quintiles. For the covariate adjustment with propensity score model, type 3 sum of squares assessed estimates controlling for propensity score. In subclassification and covariate adjustment, we found a residual association (p<0.05) between the opioid use exposure and three confounders: pain severity, overnight hospital stay, and major surgery. Therefore, the subclassification and covariate adjustment with propensity score outcome models further adjusted for pain severity, overnight hospital stay, and major surgery.

All respondents were included in IPTW analysis. Of the 10,003 respondents, 98.3% were included in the subclassification and propensity score covariate adjustment models as 9,832 were found to have overlapping propensity scores after trimming below the 1st and above the 99th percentiles (Supplemental eFigure 1 Panel A). Supplemental eFigure 1 Panel B shows the overlap among the 2,136 matched respondents. Among the 1,071 respondents who used opioid analgesics, 99.7% were matched to individuals who did not use opioids. Supplemental eTables 1–2 show the standardized differences [26] and balance for the association between covariates and the opioid use exposure. After matching, an interpretation of our maximum standardized difference (0.12) for the variable “Race and Ethnicity” using the Pearson correlation coefficient shows that membership to the opioid use versus opioid naïve group explains less than 0.36% of the variance in the Race and Ethnicity variable, which is well beneath the maximum target of 10% [26]. Therefore, it is unlikely assignment to the opioid vs. naïve group is explained by any of these measured confounders. Because unmeasured confounding remained a possibility, we conducted a sensitivity analysis to determine how strong an unmeasured confounder must be to change our estimates for the effect of opioid use on functional limitations.

Data management and analyses were conducted using SAS statistical software version 9.4 (SAS Institute, Cary, NC). Multivariable logistic regression visualizations were produced using RStudio version 1.3.1093 [27]. Complex survey design weights were not used due to the risk of bias when using them with some propensity score analyses [28].

3. RESULTS

3a. Cohort Characteristics

Table 1 shows that most of the overall cohort was naïve to opioids (89.3%), had no incident limitations in ADL (93.9%), IADL (93.4%) or cognition (75.7%), were on average 63.4 (Standard deviation: 10.2) years old, female (56.1%), and identified as non-Hispanic white (70.9%). Approximately 33% of participants received a college degree or above. Most (42.2%) resided in the South census region and had private insurance (55.0%). The overall sample reported no pain most of the time (68.0%) but used over-the-counter pain medicine in the past three months (73.2%), had no hospitalization (82.1%), and no major surgery (91.5%) within the prior two-year period. With respect to self-reported health comorbidities, most reported one comorbidity (32.8%), current alcohol use (67.1%), never smoking (48.9%) and were without depressive symptoms (87.0%). Among those using opioids, 77.1% reported using over-the-counter pain medications. Among respondents who reported opioid use in 2016, 40.9% reported opioid use in 2018. Percent agreement for 2016 and 2018 opioid use was 88.5%, indicating good reliability.

Table 1.

Unweighted association between opioid use group, primary outcomes, and baseline confounders

Dependent: History of Opioid Use for Pain		Naive to Opioids N=8932	Used Opioids N=1071	Total	p
New ADL Limitation	No	8460 (94.7)	932 (87.0)	9392 (93.9)	<0.001
	Yes	472 (5.3)	139 (13.0)	611 (6.1)
New IADL Limitation	No	8380 (93.8)	963 (89.9)	9343 (93.4)	<0.001
	Yes	552 (6.2)	108 (10.1)	660 (6.6)
New Cognitive Limitation	No	6773 (75.8)	796 (74.3)	7569 (75.7)	0.295
	Yes	2159 (24.2)	275 (25.7)	2434 (24.3)
Age	Mean (SD)	63.5 (10.3)	62.6 (10.0)	63.4 (10.2)	0.007
Sex	Male	3918 (43.9)	474 (44.3)	4392 (43.9)	0.832
	Female	5014 (56.1)	597 (55.7)	5611 (56.1)
Race and Ethnicity	White	6333 (70.9)	756 (70.6)	7089 (70.9)	0.043
	Black	1629 (18.2)	220 (20.5)	1849 (18.5)
	Hispanic	970 (10.9)	95 (8.9)	1065 (10.6)
Education	Less Than High School	761 (8.5)	85 (7.9)	846 (8.5)	<0.001
	GED or High School	2641 (29.6)	333 (31.1)	2974 (29.7)
	Some College	2546 (28.5)	367 (34.3)	2913 (29.1)
	College and Above	2984 (33.4)	286 (26.7)	3270 (32.7)
Census Region	Northeast	1261 (14.1)	109 (10.2)	1370 (13.7)	0.003
	Midwest	1900 (21.3)	236 (22.0)	2136 (21.4)
	South	3760 (42.1)	458 (42.8)	4218 (42.2)
	West	2011 (22.5)	268 (25.0)	2279 (22.8)
Healthcare Payor	Private	4914 (55.0)	590 (55.1)	5504 (55.0)	0.971
	Medicaid	2107 (23.6)	255 (23.8)	2362 (23.6)
	Public	1911 (21.4)	226 (21.1)	2137 (21.4)
Degree of Pain Most of the Time	No pain	6411 (71.8)	393 (36.7)	6804 (68.0)	<0.001
	Mild pain	992 (11.1)	149 (13.9)	1141 (11.4)
	Moderate pain	1319 (14.8)	403 (37.6)	1722 (17.2)
	Severe pain	210 (2.4)	126 (11.8)	336 (3.4)
Used Over-the-counter Pain Medicine	No	2433 (27.2)	245 (22.9)	2678 (26.8)	0.003
	Yes	6499 (72.8)	826 (77.1)	7325 (73.2)
Overnight Hospital Stay	No	7552 (84.5)	661 (61.7)	8213 (82.1)	<0.001
	Yes	1380 (15.5)	410 (38.3)	1790 (17.9)
Major Surgery	No	8282 (92.7)	872 (81.4)	9154 (91.5)	<0.001
	Yes	650 (7.3)	199 (18.6)	849 (8.5)
Baseline comorbidities	0	2783 (31.2)	152 (14.2)	2935 (29.3)	<0.001
	1	2983 (33.4)	299 (27.9)	3282 (32.8)
	2	1958 (21.9)	332 (31.0)	2290 (22.9)
	3+	1208 (13.5)	288 (26.9)	1496 (15.0)
Alcohol Use	Never	878 (9.8)	72 (6.7)	950 (9.5)	<0.001
	Former	2051 (23.0)	294 (27.5)	2345 (23.4)
	Current	6003 (67.2)	705 (65.8)	6708 (67.1)
Smoking History	Never	4479 (50.1)	412 (38.5)	4891 (48.9)	<0.001
	Former	3427 (38.4)	479 (44.7)	3906 (39.0)
	Current	1026 (11.5)	180 (16.8)	1206 (12.1)
Depressive Symptoms	No	7862 (88.0)	837 (78.2)	8699 (87.0)	<0.001
	Yes	1070 (12.0)	234 (21.8)	1304 (13.0)

Note: Depressive symptoms (yes) were classified if respondents scored three points or higher on an eight-item version of the Center for Epidemiologic Studies Depression Scale (CESD). Number in dataframe = 10,003

Table 1 also presents the cohort characteristics according to history of opioid use for pain. History of opioid use was significantly associated with age, race and ethnicity, education, census region, pain severity, use of over-the-counter pain medications, recent hospitalization, major surgery, number of comorbidities, alcohol use, smoking history, and depressive symptoms. These findings indicate that confounding factors, especially pain severity, recent hospitalization, major surgery, and number of comorbidities, influence assignment to the opioid use exposure. Below, we present multiple analytic methods to adjust for these confounding factors.

3b. Adjusted Analyses

Table 2 presents the unadjusted and adjusted odds of incident ADL, IADL, and cognitive limitations. In unadjusted analyses, opioid use is associated with roughly a 2-fold increase in the odds of incident ADL and IADL limitations. This association was weakened after adjusting for measured confounders. Adjusted analyses included multivariable logistic regression, IPTW, propensity score subclassification, covariate adjustment with propensity scores, and propensity score matching. All analyses are compared in Table 2.

Table 2.

Unweighted Odds ratio of incident ADL, IADL, and Cognitive limitations for community-dwelling adults using opioids for pain

		ADL	IADL	Cognition
Cohort	Analysis	Odds Ratio (95% Confidence Intervail, p value)
Full cohort1	Unadjusted	2.67 (2.18–3.26, p<0.001)	1.70 (1.36–2.11, p<0.001)	1.08 (0.94–1.25, p=0.278)
Full cohort	Multivariable4	1.34 (1.07–1.68, p=0.011)	1.13 (0.89–1.43, p=0.332)	1.06 (0.90–1.24, p=0.489)
Full cohort	IPTW	1.28 (0.98–1.66, p=0.067)	0.94 (0.72–1.22, p=0.635)	1.11 (0.91–1.35, p=0.287)
Overlap region cohort2	Propensity score subclassification5	1.41 (1.13–1.76, p=0.002)	1.18 (0.94–1.50, p=0.159)	1.03 (0.88–1.20, p=0.731)
Overlap region cohort	Propensity score covariate adjustment5	1.41 (1.13–1.76, p=0.002)	1.18 (0.93–1.50, p=0.159)	1.03 (0.88–1.20, p=0.730)
1:1 Matched cohort3	Conditional logistic regression6	1.29 (0.99–1.69, p=0.059)	1.03 (0.78–1.37, p=0.828)	1.02 (0.84–1.24, p=0.843)

ADL=Activities of daily living; IADL=Instrumental activities of daily living; IPTW=Inverse probability of treatment weighting with propensity score.

¹N=10,003, all respondents.

²N=9,832, 98.3% of all respondents with propensity scores in the overlap region.

³N=2,136, 99.7% (N=1,068) of the 1,071 respondents who used opioids matched to a control who was naive to opioids in 2016.

⁴The multivariable analyses adjusted for age, sex, race and ethnicity, education, census region, insurance type, pain severity, use of over-the-counter pain medications, hospital admission, major surgery, comorbidities, alcohol use, smoking history, and depressive symptoms.

⁵Propensity score quintiles were used to adjust for baseline confounders in the covariate adjustment model. Cochran-Mantel-Haenszel test and Type 3 sum of squares showed residual associations (p<0.05) between opioid use and the following variables: use of over-the-counter pain medications, hospital admission, and major surgery. Therefore, these models were also adjusted for over-the-counter pain medications, hospital admission, and major surgery in addition to propensity score quintile. Propensity score covariate adjustment fit statistics are as follows: ADL Model: AIC = 4,213, C-statistic = 0.72; IADL AIC = 4,712, C-statistic = 0.61; Cognition Model AIC = 10,895, C-statistic = 0.53. Propensity score subclassification conditional models fit statistics exclude concordance: ADL Model: AIC = 4,171; IADL AIC = 4,669; Cognition Model AIC = 10,847.

⁶Conditional logistic regression from 1:1 matched cohort fit statistics include: ADL Model: Quasilikelihood under the Independence model Criterion (QIC) = 4,536; IADL QIC = 4,826; Cognition Model QIC = 11,122.

Findings from our multivariable logistic regression are shown in Figure 2. The odds of incident ADL limitations were significantly greater among those who reported using opioid analgesics to manage pain (adjusted Odds Ratio [aOR]:1.34, 95% Confidence Interval [CI]: 1.07–1.68). In contrast, the odds of incident IADL, and cognitive limitations were not significantly associated with opioid use. Table 2 shows these findings were consistent with the analyses using propensity score subclassification and covariate adjustment with propensity score.

Figure 2.

Unweighted Multivariable logistic regression: Adjusted association between opioid use and incident functional limitations

ADL Model: Number in dataframe = 10,003, Missing = 0, AIC = 4,122, C-statistic = 0.76

IADL Model: Number in dataframe = 10,003, Missing = 0, AIC = 4,653, C-statistic = 0.68

Cognitive Model: Number in dataframe = 10,003, Missing = 0, AIC = 10,738, C-statistic = 0.63

We conducted an exploratory analysis on whether pain severity, overnight hospitalization, and age modified the association between opioid use and functional limitations. Using multivariable logistic regression, we found no statistically significant interaction between pain severity and opioid use on incident ADL, IADL, or cognitive limitations (p>0.05). Likewise, association between opioid use and ADL, IADL, and cognitive limitations did not depend on overnight hospitalization (p>0.05). However, we did find a statistically significant association between opioid use and age category (younger than 65 years versus 65+ years old) for ADL (p<0.001) and IADL (p=0.006) limitations, but not for cognitive (p>0.05) limitations. Among 4,158 adults 65 years or older (median=73, IQR=10 years), stratified analyses show no statistically significant association between opioid use and incident limitations in ADL, IADL, or cognition (p>0.05). Surprisingly, among the 5,845 adults younger than 65 years old (median=56, IQR=7 years), opioid use was significantly associated with higher odds of incident ADL limitations (aOR 1.83, 95% CI 1.38–2.42, p<0.001) and incident IADL limitations (aOR 1.42, 95% CI 1.06–1.90, p=0.018). However, this was non-significant for cognitive limitations (p>0.05). These findings indicate that those under 65 years old, compared to those 65 years and older, may be particularly vulnerable to incident functional deficits following the use of opioid analgesics to manage pain.

3c. Sensitivity Analysis

Due to the potential risk for residual unmeasured confounding, we conducted a sensitivity analysis to understand how our findings could be influenced by an unmeasured confounder [29]. Specifically, in 2016 the Centers for Disease Control and Prevention (CDC) issued opioid prescribing guidelines that encourage the complementary use of evidence-based non-pharmacologic pain interventions, such as those delivered by licensed occupational or physical therapists [30]. Therefore, it is possible that many respondents who used opioids to manage pain also received non-pharmacologic interventions from licensed and skilled rehabilitation providers. Receiving these additional interventions could have improved the respondents’ function or prevented incident functional limitations. In this case, skilled rehabilitation would act as a negative confounder because it is associated with an increased likelihood of receiving opioids [31] and a decreased likelihood of having pain or functional limitations among community-dwelling adults [32]. In other words, our original estimates for the association between opioid use and incident functional limitations may be biased toward the null. This raises a particularly interesting question of whether the null findings for the association between opioid use functional limitations in our most homogeneous analysis (i.e., propensity score matching) could be explained away by unmeasured confounding. It is possible those using opioids to manage pain were more likely to receive skilled rehabilitation services that could reduce their chances of developing incident functional limitations.

We assumed the prevalence of skilled rehabilitation was 25% among those in the opioid use group and 12% among those who were naïve to opioids based on prior literature [31]. Supplemental eTable S3 shows our sensitivity analysis results. The null association between opioid use and incident ADL limitations in our 1:1 matched cohort would become statistically significant if we had controlled for skilled rehabilitation under our assumptions for the prevalence of rehabilitation (e.g., 25% in the opioid use group and 12% among those naïve to opioids) and its effect on reducing the odds of incident ADL limitations by 20%. To reverse the non-significant findings for IADL and cognitive limitations, the association between skilled rehabilitation and functional limitations would need to exceed an odds ratio of 0.50 and coincide with a 50% higher rate of skilled rehabilitation among those using opioids (i.e., prevalence of 37.5%).

4. DISCUSSION

Using a large sample of community-dwelling middle-aged and older adults from the United States, we found the use of opioid analgesics to manage pain was associated with increased odds of incident ADL limitations among those who were functionally independent at baseline. After controlling for demographic factors and health status, use of opioid analgesics in 2016 increased the estimated odds of incident ADL limitations in 2018 by as much as 41% in propensity score adjusted and propensity score stratified analyses. In contrast, use of opioid analgesics was not associated with incident IADL or cognitive limitations in adjusted analyses. This finding was surprising because IADL and cognitive limitations often precede impairment in ADLs in the disablement model. The disablement process may not directly apply to this scenario because we are not observing an aging-related functional decline. Instead, we are observing pain limiting physical function rather than executive or cognitive function. Further research may want to consider the impact of long-term opioid use on emotional well-being, mental health, and social participation.

Notably, stratified analysis showed adults younger than 65 years—compared to those 65 years and older--were particularly at risk for incident ADL and IADL limitations following opioid use. One explanation is that younger adults may be more mobile, predisposing them to a higher risk for falls and fall-related injuries that result in incident functional limitations. However, fall risk increases with age [33], so although we cannot measure number of falls in this dataset, we believe it is unlikely the younger cohort is experiencing more falls. A second explanation is that younger adults have higher rates of opioid-related substance use disorders [12], which may increase their risk for overdose and hospitalizations leading to functional limitations. Third, it is possible that adults younger than 65 have less health insurance coverage, are burdened by time-constraints, lack paid-time-off, or face additional barriers [34] to accessing skilled non-pharmacologic pain interventions [30]. Future research may consider examining the opportunity costs associated with healthcare utilization in the context of non-pharmacologic pain treatment among middle-aged adults.

We used multivariable logistic regression, IPTW, propensity score adjustment, and propensity score matching. This strategy was beneficial because it demonstrates the robustness of findings across different assumptions for the homogeneity of our respondents’ who used opioids vs. those who were opioid naïve. Specifically, the homogeneity of adults in our opioid use vs. naïve groups increases as we move from an overall cohort, toward an overlap region cohort and matched cohort. Our results were robust across multivariable logistic regression, propensity score adjustment, and stratification, which rely on overall cohorts and overlap region cohorts. However, the 1:1propensity score matching cohort produced less precise estimates. Therefore, we considered whether this null finding could be explained by unmeasured confounding vis-à-vis the use of skilled rehabilitation (i.e., occupational and/or physical therapy) since this information is not available in the HRS database. Our sensitivity analysis [29] found the estimates for the association between opioid use and incident ADL limitations in our 1:1 matched cohort would have been similar to the multivariable logistic regression and other propensity score methods had we adjusted for skilled rehabilitation. In contrast, it is unlikely that controlling for skilled rehabilitation could overturn the non-significant findings for the association between opioids and IADL or cognitive limitations. For these null findings to achieve statistical significance, 37.5% of the cohort using opioids would have needed to access a skilled rehabilitation intervention capable of reducing the odds of IADL or cognitive limitation by 50%.

To our knowledge, the present study is the first to examine the association between opioid analgesics and ADL, IADL, and cognition among community-dwelling U.S. adults. Previous population-level research using 2000–2010 data from the Medical Expenditure Panel Survey—a similarly large, nationally representative survey—found prescription opioid use was not associated with meaningful changes in physical or cognitive function [14]. A meta-analysis of randomized control trials with follow-up periods ≤6 months found small to no benefits in function or quality of life (e.g., 2 points on 0–100 point Short-Form 36) among those using opioid analgesics for pain [13]. An international multi-center cohort study found no statistically significant improvements in upper body or lower body physical function during a 2-year follow-up among those using opioid analgesics for pain.

Our present findings expand on these earlier works by examining a large, community-dwelling cohort of functionally independent adults over a 2-year period. Most importantly, we contribute new findings leveraging the HRS’s measures of ADL, IADL, and cognitive functioning that were designed using theoretical models seminal in the field of rehabilitation science [21, 22]. These outcomes offer additional generalizability to clinical practice and the HRS International Family of Studies. Incident limitations in ADL, IADL, or cognitive functioning may serve as barriers to chronic disease self-management, navigating the community to access care, and safely managing medications. Increased attention to outcomes of ADL, IADL, and cognitive functioning is necessary to maintain community living in older adults and to reduce healthcare expenditures on long-term care settings. Although the use of opioid analgesics may alleviate painful conditions, it appears that opioids alone are inadequate in preventing functional decline secondary to pain. Pain treatments should prioritize managing pain and minimizing dysfunction in ADL because these deficits may lead to increased healthcare expenditures [35] or greater caregiver burden [36].

With respect to our null findings in the association between opioid analgesics and cognitive functioning, clinical practice patterns [5] may explain the lack of association between opioid analgesics and incident cognitive limitations. Physicians may be reluctant to prescribe depressants of the central nervous system in patients at risk for cognitive impairments [5]. Nonetheless, research on the association between opioid use and cognitive functioning has produced mixed results. Patients can experience delirium [37], as well impairments in memory and attention [38] after taking an opioid. Although the relationship between pain and cognition remains important [39], to date, there is limited evidence that long-term opioid use negatively affects cognitive functioning [40, 41]. Similarly, it is possible we found no association between opioid use and incident IADL limitations because the IADL we measured are all cognitively demanding. Clinical assessment for IADL performance is complexly intertwined with functional cognition [42], which may indicate both outcomes were at risk for selection bias because health care providers may be less likely to prescribe narcotics to those who are unable to manage them.

4a. Generalizability and Strengths

Internal validity is the primary strength of this study because we rigorously considered measured and unmeasured confounding. The HRS offers comprehensive measures of sociodemographic factors and health status like pain severity or hospitalization. Confounding variables are typically a major limitation of observational research; however, we rigorously controlled these measured confounders and demonstrated robust findings across multivariable logistic regression, subclassification with propensity scores, and covariate adjustment with propensity scores. Finally, we studied the degree to which skilled rehabilitation—a variable that cannot be measured in this database—may have influenced our findings given the CDC’s 2016 opioid prescribing guidelines that recommend using non-pharmacologic pain interventions [30] alongside prescription opioids [5]. The longitudinal nature of the HRS ensured opioid use preceded the functional outcomes and allowed us to follow functionally independent community-dwelling adults and older adults for two years until they developed incident ADL, IADL, or cognitive limitations. We also used measures of ADL, IADL, and cognitive functioning that generalize to clinical rehabilitation practice, are rooted in seminal theoretical models, have established construct validity, and are used globally in some HRS International Family of Studies [19, 20].

4b. Limitations

There are a few limitations that should be considered when interpreting our findings. We were unable to implement complex survey design weights given known biases when using them with some propensity score adjustment methods [28]. Therefore, our findings do not generalize to the U.S. population of middle-aged and older adults. Instead, our findings generalize to a subgroup of community-dwelling, functionally independent adults sampled from the HRS. Second, recall bias is an inherent limitation in self-reported data. It is possible that opioid analgesic use for pain was underreported due to a low sensitivity (0.51) as mentioned in our methods; however, prior research indicates that self-reported prescription narcotic use has moderate agreement when compared to medical record data [24]. The high specificity (0.90) and moderate agreement increase our confidence that the respondents who reported opioid use for pain were indeed taking opioids. Third, this database lacks information on opioid dose and Anatomical Therapeutic Chemical codes, so we could not calculate milligram morphine equivalents or generalize our opioid exposure to other countries with different underlying prescribing patterns. We also lack information related to polypharmacy. While a majority of those reporting opioid use also used over-the-counter pain medications, it is unclear whether respondents also used other high-risk drugs. Co-prescribing drugs like benzodiazepines and gabapentinoids with opioid narcotics is growing in prevalence and is associated with an increased risk for incident falls and fractures among older adults, which may influence our findings and warrants future study [8, 43]. Although administrative claims data could provide more information on opioid dose, duration and polypharmacy, these databases would lack the generalizable measures of function and long-term follow-up found in the HRS. Future research using the HRS to study opioids should consider using a time-varying exposure; however, this is not currently possible due to data limitations. Specifically, 2016 was the first year asking about opioid utilization and 2018 is the most recent data available. A time-varying exposure could introduce a risk for reverse causality when considering opioid use and function in overlapping periods of 2018 data. Confounding by indication is a chief limitation in any pharmacoepidemiology study relying on retrospective data. We accounted for this issue to the best of our ability by rigorously matching the opioid use and non-use groups and demonstrating balance across all measured confounders.

4c. Implications

The present study has important implications for the policy and procedures of pain management. Opioids alone are not a sufficient treatment for the complex functional, psychological, and social deficits that coincide with persistent pain in older adults [44]. The 2016 CDC guidelines for prescribing opioids in those with chronic pain underscore the importance of non-pharmacologic pain treatment [5]. Similarly, opioid prescribing guidelines for patients with cancer have also encouraged non-pharmacologic treatments [45]. Non-pharmacologic treatments may be particularly valuable in both populations because patients with cancer have a similar likelihood for opioid-related emergency department visits or hospitalizations compared to those without cancer [46]. Local post-surgical opioid prescribing guidelines recommend using patient education, non-opioid pharmacologic treatments, and specify how many opioids should be prescribed following certain surgeries but do not prioritize non-pharmacologic treatments [47]. A recent policy brief [30] details the licensed professionals trained to deliver non-pharmacologic pain interventions. Among these professionals are occupational therapy and physical therapy practitioners, whose evidence-based pain treatments [48] reduce healthcare expenditures [49], decrease duration of prescription opioid use [50], and most importantly prioritize ADL, IADL, and cognitive functioning. Patients’ pain severity should be monitored along with their ability to independently perform ADL and IADL that are critical for successfully managing their daily health, navigating the community to access health services, and participating as an active and social member within their environment.

4d. Conclusions and future research

This retrospective cohort study offers many strengths including rigorous adjustment for measured and unmeasured confounding, a 2-year follow-up, and generalizable outcome measures for ADL, IADL, and cognitive functioning that are found in the HRS International Family of Studies. We found opioid use in 2016 was associated with incident ADL limitations in 2018, which demonstrates a need for dually treating pain and function. Opioids alone may be insufficient to overcome ADL limitations that coincide with painful conditions. Compared to adults 65 years and older, middle-aged adults were particularly vulnerable to incident ADL and IADL limitations following opioid use. This finding may, in-part, be explained by the differences in healthcare access between the two age groups.

More research is needed to identify the effect of pain management strategies on functional independence in community-dwelling older adults, particularly among international communities. Our present findings are likely not generalizable to other countries because opioid availability, prescribing patterns, and overall health care provision may vary significantly between countries. While granular data on prescription drugs is limited in the HRS, its International Family of Studies offers researchers an opportunity to prioritize future cross-national comparisons between pain severity, pain treatments, and ADL, IADL, or cognitive functioning. Finally, the increased rates of comorbid illness, alcohol use, depression, and functional deficits in persons using opioids remain concerning and are important areas for future research.

Key Findings:

After controlling for sociodemographic factors and health status, use of opioid analgesics in 2016 increased the estimated odds of incident activities of daily living limitations in 2018 by as much as 41% in a U.S. sample of community-dwelling adults without baseline limitations.

Community-dwelling middle-aged adults, compared to those 65 years and older, were particularly vulnerable to experiencing incident limitations in activities of daily living and instrumental activities of daily living following opioid use.

Data availability:

Health and Retirement Study, [RAND HRS Longitudinal File]* public use dataset. Produced and distributed by the University of Michigan with funding from the National Institute on Aging (grant number NIA U01AG009740). Ann Arbor, MI, (2021).