Prescription Opioids and Longitudinal Changes in Cognitive Function in Older Adults: A Population-Based Observational Study

Abstract

Background:

Opioids are frequently prescribed to alleviate pain in older adults, yet the relationships between prescription opioids and long-term cognitive function are unclear.

Methods:

In this analysis of the Mayo Clinic Study of Aging, a longitudinal population-based cohort study of older adults with formal neuropsychological testing and cognitive evaluations performed every 15 months, the associations between prescription opioids, global and domain-specific cognitive function, and mild cognitive impairment were evaluated through time-dependent linear mixed effects and Cox proportional hazards models.

Results:

4,218 participants (51% male) were included with enrollment between 11/1/2004 and 4/1/2019 and median age of 76 (interquartile range 72, 82) years. 2,977 subjects (71%) received at least 1 opioid prescription during a median follow-up of 7.5 (5.0, 10.7) years. Overall, there was an estimated 0.096 reduction in the global cognitive z-score per year, including decreases of 0.050 in memory, 0.080 in language, 0.044 in visual-spatial cognition, and 0.112 in attention. In multivariable analyses, each receipt of an opioid prescription resulted in an additional −0.007 (95% CI: −0.009, −0.005) change in global cognitive z-score (p<0.001), with significant effects seen in the domains of memory (−0.005, 95% CI: −0.007, −0.003; p<0.001), language (−0.002, 95% CI −0.003, 0.000; p=0.024) and attention (−0.004, 95% CI: −0.006, −0.002; p<0.001) but not visual-spatial function (0.000, 95% CI: −0.001, 0.001; p=0.897). Opioid prescriptions were associated with incident mild cognitive impairment (MCI) in adjusted analysis (hazard ratio 1.21, 95% CI: 1.04, 1.42; p=0.014).

Conclusion:

Prescription opioids are associated with small but statistically significant declines in long-term cognitive function in older adults, which may represent effects of opioids or other related factors.

Introduction

Chronic pain is common in older adults, with a prevalence of bothersome pain for those ≥ 65 years exceeding 50%.^1–3 Pain in this population is associated with decreased quality of life,^4–6 inferior perception of health,⁷ and increased healthcare resource utilization.⁵ Persistent pain symptoms may also be associated with accelerated cognitive decline.^8–11 Thus, it is critically important to optimize analgesic outcomes for older adults.

Opioids are frequently prescribed in older adults to treat pain symptoms. Approximately 10% of US adults >60 years of age are using opioids at any given time.¹² Despite it being known that a single therapeutic dose of opioid causes transient declines in attention, working memory, and verbal memory in older adults,¹³ the long-term effects of opioids on cognitive performance, including the development of mild cognitive impairment (MCI) or dementia, is unclear. The limited evidence suggests that opioid use may be a risk factor for long-term cognitive impairment in older adults.^14–16 However, most data are based on the utilization of cognitive screening tools rather than longitudinal assessment of global and domain-specific cognitive function and with limited adjustment for important covariates that may influence propensity for opioid use and/or cognitive function.

In this study, we determined the associations between opioid availability, longitudinal changes in global and domain-specific cognitive function, and incident MCI utilizing data from a longitudinal population-based cohort study, the Mayo Clinic Study of Aging (MCSA). We hypothesized that opioid use in older adults would be associated with accelerated cognitive decline and an increased risk of incident MCI. In addition, we describe patterns of opioid prescribing in this population-based cohort to provide insights into current prescribing practices in older adults.

Methods

This is a secondary analysis of a population-based prospective cohort study conducted under approvals from the local institutional review boards (IRBs) and in concordance with the Strengthening the Reporting for Observational Studies (STROBE) guidelines.¹⁷

The MCSA is a prospective, population-based study examining the epidemiology of cognitive decline and risk of MCI among residents living in Olmsted County, Minnesota, USA.¹⁸ In 2004, Olmsted County residents between the ages of 70 and 89 were enumerated using the Rochester Epidemiology Project (REP) medical records-linkage system in an age- and sex- stratified random sampling design.¹⁹ The study was extended to include those aged 50 and older in 2012. The only exclusions for MCSA enrollment are a terminal illness or hospice. The present analysis includes all participants who were aged 65 years or older and without dementia at enrollment.

The MCSA performs detailed clinical evaluations at 15-month intervals and has been described in detail previously.¹⁸ Clinical evaluations include:1) study coordinator evaluation and risk factor assessment, including assessment of demographics, medications (including non-opioid analgesics and over-the-counter medications), comorbidities (e.g. hypertension, diabetes, heart failure, coronary disease, hyperlipidemia, stroke), family history, activities of daily living, questions about memory to both the participant and informant using the Clinical Dementia Rating scale,³⁸ and neuropsychiatric symptoms (i.e., Beck Depression Inventory, Beck Anxiety Inventory); 2) neurologic evaluation by a neurologist, including an interview, neurological examination, and the Short Test of Mental Status; and 3) neuropsychometric testing, including nine tests covering four domains: Memory [Auditory Verbal Learning Test Delayed Recall Trial,²⁰ Wechsler Memory Scale-Revised Logical Memory-II & Visual Reproduction-II,²¹] language [Boston naming test,²² category fluency,²³], visuospatial skills [WAIS-R Picture Completion and Block Design subtests,²⁴] and executive function [Trailmaking Test B,^23,25 WAIS-R Digit Symbol subtest²¹]. Using the mean and standard deviation (SD), test scores were converted to z-scores and tests within each domain were averaged for a domain-specific score. Global cognition was calculated using the z-transformed averages of the four cognitive domains.

Clinical diagnoses were determined by a consensus committee of those who evaluated each participant without knowledge of previous diagnoses at MCSA visits. Cognitive performance was compared with the age-adjusted scores of cognitively unimpaired individuals previously obtained using Mayo’s Older American Normative Studies.²⁶ Participants with scores 1.0 SD below the age-specific mean in the general population were considered for possible cognitive impairment. The operational definition of MCI was based on clinical evaluations including a history from the patient and informant. Diagnoses of MCI were made using published criteria after considering education, occupation, visual or hearing deficits, and reviewing all other participant information.²⁷ Diagnoses of dementia were based on published criteria.²⁸ Participants who performed in the normal range and did not meet criteria for MCI or dementia were deemed cognitively unimpaired.

The primary exposure of interest in this study was receipt of any post-enrollment opioid prescription, which was defined by the presence of active opioid prescriptions obtained during study enrollment but prior to measurement of outcomes of interest (i.e. time-dependent). Opioid prescriptions were obtained through the REP prescription database.²⁹ Opioid medications included: oxycodone, hydrocodone, codeine, fentanyl, hydromorphone, oxymorphone, morphine, methadone, tramadol, tapentadol, propoxyphene, and buprenorphine, and meperidine. Prescriptions were characterized by medication, form, route, frequency, duration of availability, and total prescribed daily dose in oral milligram morphine equivalents (MME), using the Centers for Disease Control and Prevention opioid conversion tool.³⁰ Opioid availability was assumed to occur over the entirety of the prescription (e.g., a 5 day supply of 5 mg oxycodone tablets written every 6 hours as needed was counted as 5 days of duration). Prescription refills were included as part of the original opioid prescription rather than counted as unique prescriptions. Prescriptions written in the period extending from 4 days before through 14 days after a hospital discharge date or date of surgery were considered opioids prescriptions related to hospitalization or surgery, respectively.

Additional exposure variables of interest at the time of MCSA enrollment, which were selected a priori prior to data collection given known relationships with cognitive decline and/or prescription opioid availability, included: demographic features (i.e., age, gender, educational status, marital status), smoking status (i.e., never, former, current), Charlson comorbidity index, presence of an APOE- ε4 allele, diagnosis of MCI, chronic pain diagnoses within the 12 months prior to enrollment using validated International Classification of Disease, revision 9 and 10 (ICD-9, ICD-10) codes as previously described (complete list provided in Supplemental Table 1),³¹ outpatient prescriptions for antidepressants and benzodiazepines (Supplemental Table 2), prescription opioid availability within the 12 months prior to enrollment, and neuropsychiatric symptoms as assessed with Beck Depression Inventory and Beck Anxiety Inventory.

Statistical analysis:

Baseline subject demographics and characteristics, as well as opioid prescription information was presented as median (25^th percentile, 75^th percentile) for continuous and nearly continuous variables and frequency (percentage) for categorical variables. Demographics and other subject characteristics were presented overall and according to receipt of any opioid prescription during study period with absolute standardized differences for comparison. Study period is defined as the time period beginning at enrollment and ending at death or the date that data for this study were retrieved (04/01/2019). All opioid prescriptions occurring during the study period were summarized. In addition, the prescription data was summarized on a per-participant basis with frequencies and percentages corresponding to the number of participants receiving a prescription with the given characteristic at any time during the study period.

A swimmer plot was used to visually display opioid availability and intensity in participants from 1 year prior to enrollment to end of follow-up. To reduce the size of the plot, only participants who averaged 7 or more days of opioid availability per year were included. To determine opioid availability for any given day during follow-up, the maximum prescribed dose, frequency, and refill allowance was used to estimate the number of days that each prescription would provide opioids to the participant (i.e. participants were assumed to consume opioids at the maximum prescribed rate). In the case of missing data, the order text was reviewed to determine prescription duration. For any given day, a participant was considered to have opioid availability if they had any prescription providing opioids for that day. If the participant had multiple prescriptions providing opioids on a given day, the daily MMEs for each prescription were summed to estimate the total MME available.

Missing data was handled using the method of multiple imputation. Ten multiply-imputed datasets were generated with imputed values for missing datapoints based on covariates including age, visit date, gender, prior prescription, education, chronic pain diagnosis, anti-depressant/psychotic medication, benzodiazepines, Charlson index, global and component z-scores, marital status, smoking status, baseline Beck anxiety and depression indexes, and presence of an APOE ε−4 allele.

To assess time-dependent associations between post-enrollment opioid prescriptions and cognitive trajectory, participants with one or more global cognitive z-scores were included. Multivariable linear mixed-effects models with random intercepts and slopes were used to model the z-scores. Models were adjusted for baseline covariates and covariate-by-time interaction terms (Supplement). Current opioid availability (i.e., an active opioid prescription on the day of cognitive evaluation) was included as a covariate. The effect of daily MME on cognitive trajectory was also assessed in interaction analysis. Estimates for the change in per-year decline in cognition were presented along with 95% confidence intervals and p-values. A similar analysis was conducted using each of the raw scores as outcomes. Mean (SD) baseline and subsequent visit raw scores were summarized in those having and not having opioid prescriptions. Estimates for the fixed effect terms from the longitudinal model were used to estimate cognitive trajectories for two hypothetical 75 year old male participants with differing covariates. Trajectories were presented graphically for the hypothetical participants assuming 0, 2, and 4 prescriptions occurring in short succession at 2 years post enrollment.

Univariable and multivariable Cox proportional hazards models with time-dependent covariates were fit to the data to assess the associations between opioid prescriptions and hazard for diagnosis of MCI in participants who were cognitively unimpaired at enrollment. Only participants without prevalent MCI at enrollment and with more than 1 MCSA visit were included. Age was used as the time scale, and the multivariable model was adjusted for gender, education, marital status, smoking status, presence of an APOE ε−4 allele, and time-dependent Charlson index, Beck anxiety and depression indexes, anti-depression medications, and benzodiazepines. The time-dependent exposure of interest was any post-enrollment opioid prescription such that all participants began follow-up defined as unexposed and participants with a new opioid prescription during the study period were considered unexposed prior to and exposed subsequent to that prescription. Estimates for the hazard ratio, 95% confidence intervals, and p-values were presented.

For all linear mixed effects models and Cox proportional hazards models, the 10 multiply imputed datasets were analyzed separately, and estimates were combined using Rubin’s rules. All analyses were done using SAS version 9.4 (SAS Institute Inc. Cary, NC, USA). P-values <0.05 were considered statistically significant.

Results

Opioid prescriptions

A total of 4,218 participants (51% male) were included over a median (IQR) follow-up of 7.5 (5.0, 10.7) years (Supplemental Figure 1), with median (IQR) age at enrollment of 76 (72, 82) years. Of these, 2,977 (71%) received 1 or more opioid prescription during study period. Compared to those who were never prescribed an opioid during study period, those who were had a higher prevalence of chronic pain diagnoses (34% vs. 18%), higher Charlson comorbidity indices (3 [1, 5] vs. 2 [1, 5]), fewer years of educational training (32% vs. 40% with 16+ years), higher rates of opioid prescriptions in the year preceding enrollment (25% vs. 13%), and higher Beck depression (4 [1, 8] vs. 3 [1, 7]) and anxiety scores (2 [0, 5] vs. 1 [0, 4]; Table 1).

Table 1 –

Participant demographics at time of enrollment (N=4218)^*

	Overall (N=4218)	No opioid Rx during follow-up (N=1241)	One or more opioid Rx during follow-up (N=2977)	Absolute Std. Diff.
Age, years	76 (72, 82)	76 (71, 81)	77 (72, 82)	0.171
Gender
Male	2140 (51%)	658 (53%)	1482 (50%)	0.065
Female	2078 (49%)	583 (47%)	1495 (50%)	0.065
Education (n=4211)
<12 years	305 (7%)	58 (5%)	247 (8%)	0.147
12 years	1374 (33%)	362 (29%)	1012 (34%)	0.103
13–15 years	1101 (26%)	324 (26%)	777 (26%)	0.002
16+ years	1431 (34%)	493 (40%)	938 (32%)	0.174
Marital status (n=4212)
Married/living together	2835 (67%)	856 (69%)	1979 (67%)	0.052
Widowed	934 (22%)	236 (19%)	698 (23%)	0.109
Separated/divorced	273 (6%)	98 (8%)	175 (6%)	0.080
Single, never married	170 (4%)	50 (4%)	120 (4%)	0.000
Smoking status (n=4209)
Never	2141 (51%)	667 (54%)	1474 (50%)	0.085
Former	1893 (45%)	531 (43%)	1362 (46%)	0.059
Current	175 (4%)	40 (3%)	135 (5%)	0.068
Charlson index	3 (1, 5)	2 (1, 5)	3 (1, 5)	0.150
APOE-E4 positive (n=4069)	1070 (26%)	304 (26%)	766 (26%)	0.009
Mild cognitive impairment	635 (15%)	177 (14%)	458 (15%)	0.032
Chronic pain diagnosis in prior year	1238 (29%)	227 (18%)	1011 (34%)	0.362
Antidepressants	749 (18%)	192 (15%)	557 (19%)	0.086
Benzodiazepines	148 (4%)	37 (3%)	111 (4%)	0.041
Opioid Rx in prior 6 months	536 (13%)	70 (6%)	466 (16%)	0.329
Opioid Rx in prior 12 months	901 (21%)	156 (13%)	745 (25%)	0.323
BDI-II grand total (n=4106)	4 (1, 7)	3 (1, 7)	4 (1, 8)	0.143
BAI total (n=4193)	1 (0, 4)	1 (0, 4)	2 (0, 5)	0.157

Rx – prescription; BDI-II – Beck Depression inventory; BAI – Beck Anxiety Inventory.

^*Data are summarized as number (percentage) for categorical variables and median (25^th, 75^th percentile) for continuous variables. When not all data are available, number of observations with complete data is reported. Groups are compared using absolute standardized differences.

These 2,977 participants received 21,949 opioid prescriptions (Table 2), with a median of 3 (2, 8) prescriptions per participant during study period. Oxycodone and tramadol were the most prescribed opioids (each represented 30% of all prescriptions) followed by hydrocodone (13% of prescriptions); 1,886 individuals (63%) received at least one prescription for oxycodone only, 1,486 (50%) received at least one prescription for tramadol only, and 440 (15%) received concurrent prescriptions for both oxycodone and tramadol. Twenty three percent of opioid prescriptions (5,096/21,949) were associated with hospitalization, and 10% (2,299/21,949) were associated with surgery. For those participants receiving opioid prescriptions, the median maximum prescribed dose (IQR) in oral milligram morphine equivalents (MME) was 10 (8, 15), with a median maximum daily dose of 60 (45, 90), median maximum prescription duration of 15 (7, 40) days (inclusive of refills), and median maximum total MME per participant of 475 (225, 950). Forty four percent of participants (1,301/2,977) received a new opioid prescription (i.e. not a refill) while they had opioid medications remaining on a previous prescription based on estimated opioid availability. Longitudinal opioid prescriptions for individuals with at least 7 days of opioid availability per year of follow-up (n=391) are displayed graphically (Supplemental Figure 2), with substantial inter-participant variability in prescribing frequency, length, and intensity.

Table 2 –

Prescriptions occurring during follow-up^*

	Summary of all prescriptions (N=21949)	Per-participant summary (N=2977)†
Medication
Oxycodone	6686 (30%)	1886 (63%)
Tramadol	6660 (30%)	1486 (50%)
Hydrocodone	2830 (13%)	992 (33%)
Codeine	857 (4%)	378 (13%)
Propoxyphene	383 (2%)	183 (6%)
Fentanyl	946 (4%)	147 (5%)
Hydromorphone	1155 (5%)	326 (11%)
Morphine	1102 (5%)	507 (17%)
Oxycodone and tramadol	631 (3%)	440 (15%)
Other/Other combination	699 (3%)	338 (11%)
Medication form
Tablet	19357 (88%)	2828 (95%)
Patch	949 (4%)	149 (5%)
Solution	1595 (7%)	731 (25%)
Other	48 (<1%)	13 (<1%)
Hospitalization	5096 (23%)	1741 (58%)
Surgical episode	2299 (10%)	1045 (35%)
ICU admission	863 (4%)	495 (17%)
Dose (MME, n=21887)§	5 (5, 9)	10 (8, 15)
Daily dose (MME, n=21887)	30 (15, 45)	60 (45, 90)
Total MME, (n=21887)	300 (150, 600)	475 (225, 950)
Prescription length (days)	13 (5, 30)	15 (7, 40)
New prescription obtained with meds remaining‡	6315 (29%)	1301 (44%)

MME – oral morphine milligram equivalents

^*Data are from 21949 prescriptions in 2977 participants written between enrollment and end of follow-up. Median (25^th, 75^th) number of prescriptions 3 (2, 8). Values are number (percent) for categorical variables and median (25^th, 75^th) for continuous variables. Prescriptions for the same medication and form were combined (n=279).

^†In the per-participant summary, categorical variables reflect meeting the definition for the characteristic at any time during follow-up. For example, out of 2977 participants receiving any opioid prescription during follow-up, 1886 (63%) participants had at least one prescription for oxycodone. Percentages will no longer sum to 100. Values summarized for continuous variables are the maximum observed for each participant during follow-up.

^§When multiple prescriptions were written on the same date, the dose OME was summed regardless of whether it was intended for the medications to be taken at the same time or not.

^‡Based on usage of maximum frequency and dosage, was there a new prescription obtained prior to the estimated date at which subject would exhaust the supply from this prescription.

Cognitive changes

There were 3,982 participants included in analyses of longitudinal changes in global and domain-specific cognitive function. The global cognitive z-score of participants decreased over time, with an estimated slope (referred to hereafter as global cognitive trajectory) of −0.096 SD per year (Table 3). Each of the four cognitive domains comprising the global score also decreased with time. The global cognitive trajectory declined further with each opioid prescription: i.e., each opioid prescription was associated with an additional −0.007 (95% CI: −0.009, −0.005) SD acceleration in the per-year decline of global cognitive trajectory (p<0.001; Figure 1). Thus, receipt of 3 opioid prescriptions was associated with an estimated annual accelerated decline of −0.021 (−95% CI: −0.027, −0.015) SD (i.e., −0.007 × 3 prescriptions) compared to no prescriptions. Opioid prescriptions were associated with accelerated declines in the z-scores for the separate cognitive domains of memory, language, and attention, but not visual-spatial skills (Table 3). The proportional acceleration was greatest for memory (10% change in annual cognitive trajectory for each prescription). Declines in raw test score performance were greater in opioid-exposed compared to unexposed participants (Table 4).

Table 3 –

Associations between opioid prescriptions and changes in cognitive Z-scores from linear mixed effects models^*

	Slope prior Estimate	Change in slope following each Rx Estimate (95% CI)	p-value
Global	−0.096	−0.007 (−0.009, −0.005)	<.001
Memory	−0.050	−0.005 (−0.007, −0.003)	<.001
Language	−0.080	−0.002 (−0.003, 0.000)	0.024
Visual-spatial	−0.044	0.000 (−0.001, 0.001)	0.897
Attention	−0.112	−0.004 (−0.006, −0.002)	<.001

^*Models adjusted for baseline age, gender, prescription in the prior 6 and 12 months, baseline Charlson index, education, marital status, smoking status, APOE-ε4 allele, Beck depression, Beck anxiety, pain diagnosis, antidepression medication, benzodiazepine medication, test naïvety, and current opioid availability.

Figure 1.

Estimated global cognitive Z-score trajectory for two hypothetical 75 year old male participants based upon prescription opioid availability, including no prescriptions, 2 prescriptions, and 4 prescriptions occurring 2 years after study enrollment.

Participant 1 is a college educated, married, never smoker with no opioid prescriptions in the 1 year prior to enrollment, low Charlson score (2), low Beck depression and anxiety indexes (2 and 0 respectively), and no chronic pain, antidepressant, or benzodiazepine availability at baseline. Participant 2 is a high school educated, single, former smoker with no opioid prescriptions in the 1 year prior to enrollment, moderate Charlson score (5), moderate Beck depression index (12) and low Beck anxiety index (0), chronic pain at baseline, and no antidepressants or benzodiazepines at baseline.

Table 4 –

Summary of estimated association between receipt of opioid prescription and change in cognitive decline measured on raw scales of cognitive tests^*

	Opioid Unexposed			Opioid Exposed
Test (valid range)	Baseline Mean (SD)	Following visit Mean (SD)	Difference Mean (SD)	Baseline Mean (SD)	Following visit Mean (SD)	Difference Mean (SD)	Estimated slope prior to Rx	Estimated Change in slope following each Rx Estimate (95% CI)	P-value
Memory
WMSR Log Mem II Total (0–43)	17.7 (8.1)	17.9 (8.6)	0.1 (5.2)	19.1 (8.3)	18.2 (9.8)	−1.0 (6.0)	−0.126	−0.027 (−0.039, −0.014)	<.001
WMSR Vis Rep II Total (0–41)	21.5 (9.0)	21.8 (9.6)	0.3 (6.5)	23.7 (8.6)	20.1 (9.5)	−3.7 (7.4)	−0.396	−0.020 (−0.032, −0.007)	0.002
ALVT half hour delay (0–15)	7.3 (3.8)	7.3 (3.8)	0.0 (2.4)	7.8 (3.6)	6.4 (4.1)	−1.4 (2.9)	−0.175	−0.007 (−0.013, −0.001)	0.015
Language
Boston naming score (0–60)	54.6 (5.0)	54.5 (5.4)	−0.1 (2.8)	54.9 (5.0)	53.9 (6.2)	−1.0 (4.0)	−0.238	−0.009 (−0.016, −0.001)	0.029
Category fluency (0–74)	42.9 (10.1)	42.1 (10.4)	−0.8 (6.7)	42.2 (9.6)	37.9 (11.0)	−4.3 (7.6)	−0.824	−0.011 (−0.024, 0.003)	0.127
Visual-spatial
WAISR PIC Comp Total (0–20)	13.5 (3.3)	13.5 (3.4)	0.1 (2.6)	13.9 (3.2)	13.7 (3.6)	−0.2 (2.4)	−0.013	−0.001 (−0.005, 0.004)	0.745
WAISR Block Des Total (0–51)	24.0 (8.6)	23.3 (8.8)	−0.7 (5.8)	24.3 (8.4)	20.9 (8.6)	−3.4 (5.8)	−0.597	0.000 (−0.009, 0.009)	0.933
Attention
WAISR Digit Sym Total (0–93)	43.8 (11.2)	43.0 (11.6)	−0.9 (5.9)	44.3 (10.4)	38.4 (11.4)	−5.9 (7.5)	−0.969	−0.021 (−0.035, −0.007)	0.004
TMTB Total (0–300)#	104.9 (54.0)	110.3 (60.1)	5.4 (38.5)	102.3 (49.3)	139.4 (78.7)	37.1 (58.4)	6.24	0.25 (0.12, 0.38)	<.001

^*Baseline and following visit test scores are presented as mean (SD) for patients prior to any opioid exposure (unexposed) and following exposure (exposed). Baseline values for exposed are from the last visit prior to exposure.

^#Higher scores on TMBT represent worse performance.

CI = confidence interval, SD = standard deviation, Rx = prescription, WMSR = Wechsler Memory Scale-Revised, Log Mem II = Logical Memory-II, Vis Rep II = Visual Reproduction II, ALVT = Auditory Verbal Learning Test, WAISR = Wechsler Adult Intelligence Scale-Revised, PIC Comp = Picture Completion, Block Des = Block Design, Digit Sym = Digit Symbol, TMTB = Trailmaking Test B.

In interaction analysis, the daily MME prescribed moderated the change in global cognitive trajectory. Each 15 MME per day increase in opioid availability was associated with declines in global cognitive trajectory (i.e., a single 30 MME per day prescription was associated with a change in the slope of global cognitive Z-score of −0.006 units, 95% CI: −0.008, −0.004, and each additional 15 MME per day increase was associated with an additional change of −0.001, 95% CI: −0.003, 0.000; p=0.026; Supplemental Table 3).

Incident MCI

A total of 2,963 cognitively unimpaired individuals were included in time-to-event analyses for incident MCI with 769 (26%) developing MCI during follow up. In univariate analysis, prior receipt of an opioid prescription during follow-up was associated with an increased hazard for incident MCI (HR 1.29, 95% CI: 1.10, 1.50, p=0.001). In multivariable analysis, opioid prescriptions were also associated with incident MCI (HR 1.21, 95% CI: 1.04, 1.42, p=0.014, Supplemental Table 4). The presence of a chronic pain diagnosis at enrollment was not associated with incident MCI. Other factors potentially associated with pain that were associated with incident MCI included Charlson comorbidity index, the Beck depression index, and a current antidepressant medication prescription, but not the Beck anxiety index or a current benzodiazepine prescription (Supplemental Table 4).

Discussion

In this population-based longitudinal analysis of community-dwelling older adults, prescription opioid availability was associated with accelerated cognitive decline and risk of incident MCI. Selectivity in this association among separate cognitive domains was present, with memory being the most affected and visual-spatial skills unaffected. This accelerated decline was accompanied by a higher incidence of MCI in those receiving opioid prescriptions.

More than two-thirds of study participants received at least one opioid prescription during follow-up. Nearly half of participants, including more than 60% of those receiving any opioid prescription, were prescribed oxycodone, and more than one in three study participants were prescribed tramadol. Less than a quarter of opioid prescriptions were associated with recent hospitalization and only one in ten were associated with surgery. This suggests that prescription opioids are indeed widely available in community dwelling older adults, largely for conditions not linked to acute hospitalization or surgical care. Beyond this study, the United States Centers for Disease Control and Prevention (CDC) reported that 25% of adults greater than or equal to 65 years of age filled an opioid prescription in 2018,³² and survey data obtained in 2013 through 2016 from the National Health and Nutrition Examination Survey (NHANES) reported prescription opioid use prevalence of 10% in those 60 years and older.¹² As trends in opioid prescribing for US adults have changed considerably in recent years,³³ future longitudinal studies are necessary to assess the generalizability of study findings, evaluate geographic and socioeconomic disparities in prescription opioid availability, and further assess the associations between opioid availability and long-term cognitive function in older adults.

A recent systematic review evaluated associations between opioids and cognition in older adults with cancer and chronic non-cancer pain.³⁴ The review identified 10 studies that were generally limited by small sample sizes, lack of formal neuropsychological assessments, unclear methodologies, and poor reporting. Some studies showed improvement while others showed impairments in cognition with opioid medications, leading the authors to conclude that assessment of distinct cognitive domains, rather than relying on cognitive screening tests, may be beneficial in detecting changes in cognitive function in older adults with chronic pain receiving opioid therapy. Two extant larger studies analyze prospective longitudinal data. In a longitudinal study of more than 3,000 community-dwelling older adults assessed every 2 years with the Cognitive Abilities Screening Instrument, those with the highest cumulative exposure to prescription opioids when compared to those minimal or no opioid exposure had an increased risk of all-cause dementia, but there were no significant differences in longitudinal trajectories of the screening score in analyses not adjusted for other factors.¹⁶ Patients with the heaviest consumption of non-steroidal anti-inflammatory drugs also had modest increases in incident dementia, suggesting that chronic pain, which was not assessed in the investigation, may be important. In another study of more than 500 older adults who had Mini-Mental State Examinations (MMSE) performed at baseline and approximately 7 years later, opioid use at enrollment, either alone or in combination with another centrally acting medication, was associated with greater declines in MMSE scores than in those without centrally acting mediation use; however, only 9 subjects were using opioids at enrollment, which substantially limits generalizability.¹⁵

In this investigation, we overcome the limitations of previous investigations by employing longitudinal neuropsychological testing and multidisciplinary cognitive evaluations in a large population-based cohort with adjustment for key factors that may confound opioid and cognition relationships, including chronic pain diagnoses, concomitant use of antidepressant and benzodiazepine medications, and mental health symptoms, amongst others. While a single therapeutic dose of opioid is known to cause acute and transient declines in attention, working memory, and verbal memory in older adults,¹³ this study provides evidence for small but statistically significant incremental impairments in long-term cognitive function associated with prescription opioid availability. For any individual participant, each unique opioid prescription was associated with a more than 7% increase in the rate of global cognitive decline over time, with the greatest changes seen in the domain of memory, findings which are supported by previously described impairments in working memory and attention in those with opioid use disorder.^35–37 While observed accelerations in cognitive decline may seem modest in size, opioid prescription and cognition relationships are additive such that 3 opioid prescriptions would result in a more than 20% increase in the rate of global cognitive decline. Most importantly, accelerated cognitive decline as ascertained through neuropsychological testing was accompanied by a higher incidence of clinically diagnosed MCI in those receiving opioid prescriptions.

The key question is whether observed associations between prescription opioids and cognitive decline represent causal relationships or whether prescription opioids serve as a marker for other conditions associated with cognitive dysfunction. For example, previous investigations have noted associations between chronic pain, cognitive dysfunction, and even dementia in adults, but have not adjusted for opioids.^9,10 Conversely, associations have been described between opioid exposures and long-term cognitive dysfunction, but have not included adjustment for chronic pain.¹⁶ In the current investigation, we included longitudinal assessments of opioid exposures while also accounting for the presence of chronic pain at enrollment. In multivariable models, prescription opioid availability was associated with increased risk of incident MCI, but the presence of chronic pain at enrollment was not. While it is important to recognize limitations in our approach, including most notably that we included chronic pain diagnoses present at study enrollment rather than actual quantitative or qualitative measures of pain over time, these findings highlight that both pain and opioid exposures should be considered in studies evaluating associations with cognitive function. There remain several mechanistic possibilities for the relationships described in this investigation and in previous work: 1) opioid exposures, independent of chronic pain, are causally linked to worsening cognitive function; 2) opioid exposures mediate the relationships between chronic pain and cognitive dysfunction; 3) chronic pain, independent of opioid exposures, is causally linked to worsening cognitive dysfunction; or 4) opioid exposures and/or chronic pain serve as a surrogate marker for underlying conditions (i.e. chronic disease states, mental health disorders) associated with cognitive dysfunction. Well-designed prospective investigations with detailed assessments of pain and neuropsychological performance will be necessary to further differentiate these possibilities. For now, geriatricians and chronic pain practitioners should not change their clinical practices in response to this data. Prescription opioid use in older adults should be tailored to each individual patient with shared decision making and thorough assessment of risks and benefits. In addition to the numerous well-known side effects of prescription opioids, this study provides evidence for a potential additional side effect in that prescription opioids may be associated with long-term impairments in cognitive function. Future research will be essential to further understand the mechanisms behind prescription opioids and cognition relationships.

Limitations of the study warrant consideration. First, despite pre-specified adjustment for factors likely to confound exposure-outcome relationships, the potential for residual confounding remains. For example, it is possible that patients with early cognitive impairment may be more likely to complain of pain, which may prompt opioid prescriptions; this possibility has not been evaluated. Second, although we adjusted for chronic pain diagnoses, we were unable to perform more detailed assessments of pain etiologies, characteristics, severity, or functional limitations, as described previously. These features may have implications on cognition and other aspects of daily functioning. Third, while we obtained detailed information on prescription opioid availability, it is important to recognize that opioid availability does not imply opioid use. In some cases, opioid availability may overestimate use; alternatively, in those with opioid misuse, availability may underestimate opioid use. This remains a ubiquitous limitation in observational research exploring opioid availability in community-dwelling persons. Fourth, it is possible that the associations between opioid availability and cognitive function may not be consistent across all patients, such that some patients may experience paradoxical improvements in cognitive performance with opioid availability (e.g., secondary to improvement in pain symptoms). Future studies are necessary to further assess heterogeneity in opioids and cognition relationships. Finally, these results are derived from a relatively homogeneous group of older adults in the Midwest of the United States with high educational status (i.e., more than 1/3 with at least 16 years of education). Future studies will be necessary to evaluate differences in these relationships across racial groups, educational level, and socioeconomic status.

In summary, prescription opioid availability in a longitudinal evaluation of more than 4,000 community-dwelling older adults aged 65 years and older was associated with small but statistically significant impairments in cognitive function in multiple domains and an increased risk of MCI. These findings suggest that prescription opioid availability may have long-term effects on cognitive function or serve as a marker for related processes that affect cognitive function. Future longitudinal studies with detailed assessments of opioid availability, pain features, and neuropsychological function are necessary to further define these relationships.

Key Points:

• In this longitudinal evaluation of more than 4,000 older adults enrolled in the Mayo Clinic Study of Aging, seven out of 10 participants received at least 1 opioid prescription during a median (interquartile range) follow-up of 7.5 (5.0, 10.7) years.

• In multivariable analyses, each receipt of an opioid prescription was associated with small but statistically significant accelerations in longitudinal declines in global cognitive performance, including in the domains of memory, language, and attention but not visual-spatial function.

• Opioid prescriptions were associated with a 20% increase in the hazard for incident mild cognitive impairment in adjusted analysis.

Why Does This Matter?

These findings suggest that prescription opioids, though frequently utilized for pain symptoms in older adults, are associated with long-term impairments in cognitive function or serve as a marker for processes that affect cognitive function.