Exploring the Association Between Opioid Use Disorder and Alzheimer’s Disease and Dementia Among a National Sample of the U.S. Population

Abstract

Background:

Past research suggests associations between heavy alcohol use and later life dementia. However, little is known about whether opioid use disorder (OUD) and dementia share this association, especially among age groups younger than 65 years old.

Objective:

Examine the association between OUD and Alzheimer’s disease (AD) and dementia.

Methods:

Electronic health records between 2000 and 2021 for patients age 12 or older were identified in the Cerner Real-World database™. Patients with a prior diagnosis of dementia were excluded. Patients were followed for 1-10 years (grouped by one, three, five, and ten-year follow-up periods) in a matched retrospective cohort study. Cox proportional hazards regressions were used to estimate adjusted hazard ratios (aHRs) of incident AD/dementia stratified by age and follow-up group.

Results:

A sample of 627,810 individuals with OUD were compared to 646,340 without OUD. Individuals with OUD exhibited 88% higher risk for developing AD/dementia compared to those without OUD (aHR = 1.88, 95% CI 1.74, 2.03) within 1 year follow-up and 211% (aHR = 3.11, 95% CI 2.63, 3.69) within 10 years follow-up. When stratifying by age, younger patients (age 12-44) had a greater disparity in odds of AD/dementia between OUD and non-OUD groups compared with patients older than 65 years.

Conclusions:

Additional research is needed to understand why an association exists between OUD and AD/dementia, especially among younger populations. The results suggest that cognitive functioning screening programs for younger people diagnosed with OUD may be useful for targeting early identification and intervention for AD/dementia in particularly high risk and marginalized populations.

Keywords

Alzheimer’s disease cerner real-world data cox proportional hazards dementia opioid use disorder

INTRODUCTION

Studies of some forms of mild to moderate substance use, such as occasional use of alcohol, show limited evidence for increased risk for dementia; however, more intense chronic use of some licit and illicit drugs may increase risk for cognitive impairment, such as dementia, in later life [1 –3]. Most studies of substance use and dementia focus on the link between alcohol and cognitive health and have found that people with alcohol use disorders are at higher risk for dementia [4], and particularly early-onset dementia [5]. Opioid use disorders (OUD) were estimated to affect 6–7 million United States (U.S.) adolescents and adults in 2019 [6], raising questions about whether opioid use disorders may similarly demonstrate an association with increased risk for dementia.

Past research suggests physiological mechanisms that may link OUD and dementia. For example, people with OUD are at increased risk for overdose [7, 8], and several studies in humans have suggested that opioid overdoses can cause delayed onset of damage to the white matter of the central nervous system as well as damage to brain areas most sensitive to hypoxia including the hippocampus, which plays a major role in memory and learning, and the cerebellum [9]. Evidence also suggests that changes due to oxygen deprivation can result in decreased neuronal density, particularly in the global pallidus, which can present as rapidly progressive cognitive impairment [10]. Neuroimaging studies of individuals with OUD suggest atrophy of the frontotemporal region, which can lead to dementia, may be more common individuals diagnosed with OUD as compared with controls [11]. Neuroimaging studies comparing individuals with and without OUD also suggest differences in the brain, particularly the volume of grey and white matter, may be more prevalent among individuals with OUD compared with healthy controls [10 –12]. Lower levels of gray matter and an increase in white matter may be associated with greater dementia risk [13].

Emerging research has begun to estimate relationships between forms of opioid use and cognitive function. A small number of studies examine associations between prescription opioid use and dementia in older adults, with two prospective cohort studies demonstrating slightly higher risk for forms of dementia, including Alzheimer’s disease (AD), in participants who used prescription opioids more frequently than peers with little to no use [14, 15]. Recent research has also begun to examine associations between cognitive function and long-term use of illicit opioids and OUD, which qualitatively differ from prescription opioid use. A study on the prevalence of cognitive impairments among people who use drugs found that individuals with opioid use disorders scored significantly worse on the Montreal Cognitive Assessment domain, a test which includes abstract reasoning, memory and executive functioning, than those with cannabis use disorders or alcohol use disorders [16]. Another study found that patients with OUD had significant impairments in working memory and information-processing speed compared to controls [17]. However, most neuroimaging studies rely on small sample sizes and with limited variation in participant ages making it unclear how the relationship between OUD and dementia may differ across the life course. Cognitive impairments in attention and executive function have also been observed among patients receiving treatment for OUD, but these too are limited by cross-sectional design [18].

To address some of the study design limitations of previous studies that limit inference about the relationship between OUD and dementia, we leverage longitudinal data from a large national database. Our study seeks to provide foundational estimates for differences in dementia risks among people in the U.S. diagnosed with OUD compared with other patients. As many studies focus on older adults or include small sample sizes that make age-related variation difficult to estimate, our study pays particular attention to age differences that may affect risk estimates within these populations. Since roughly 6.7 million people in the U.S. are living with dementia attributable to AD [19], intersections between OUD and dementia constitute an overlap in key public health priorities. A better understanding of this relationship may help identify vulnerable subpopulations at greater risk for developing cognitive conditions that can hinder functioning and overall well-being.

METHODS

The study also followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [20].

Settings and participants

Pilot data analysis

This study was inspired by a small preliminary pilot study with regional data that used machine learning to identify co-occurring conditions among patients diagnosed with OUD. Pilot data were obtained from the Clinical Research Database (CRDB), which is a large scale longitudinal clinical data repository from Loyola University Medical Center in Maywood and 28 outpatient centers covering the western suburbs of Chicago [21]. The application used to leverage data is called the Relationship of Clinical Knowledge to Events Tool (ROCKET) which detects a range of clinical events longitudinally at every patient encounter through electronic health record (EHR) coding and natural language processing techniques. ROCKET allows users to visually display temporarily oriented collections of clinical events, to assess trends of disease.

Patients from these locations represent roughly 20% of the population of Illinois. The CRDB has complied longitudinal data on more than two million unique patients since 2007. The data used in this analysis were from 2007 through 2021. Patients included in the pilot analysis with the CRDB were of any age and had the racial/ethnic backgrounds of white, black, Asian, Hispanic, Native Hawaiian/Pacific Islander, multiracial, other, and unknown.

For the initial regional data analysis, the primary outcome was senile dementia. International Classification of Diseases (ICD)-9 and ICD-10 codes were used to identify patients with a senile dementia diagnosis (defined in Supplementary Table 4). Patients with at least one inpatient, skilled nursing facility, home health aide, or outpatient claim with one of the listed ICD-9 or ICD-10 codes were defined as having senile dementia.

The primary exposure in the pilot analysis was a diagnosis of opioid use disorder using ICD-9 and ICD-10 codes noted in at least one inpatient, skilled nursing facility, home health aide, or outpatient claim (defined in Supplementary Table 4).

Crude odds ratios were used to quantify the association between OUD and senile dementia. The amount of time elapsed between a patient’s OUD diagnosis and their senile dementia diagnosis was also examined. Patients were aligned on their OUD diagnosis date, and temporal data attached to the patient’s senile dementia diagnosis were used to show temporality between the two conditions.

Primary national data analysis

The primary analysis conducted for this study used de-identified longitudinal data from Cerner Real-World Data™ (CRWD) [22]. The April 2022 version of CRWD includes compiled EHR from 122 contributing health systems in the U.S. totaling about 100 million unique patients and over 1.5 billion patient encounters, which include pharmacy, laboratory, admission, and billing information. All admissions, orders for medication or labs, medication dispensing, and specimens are time stamped, providing a temporal data on the relationship between treatment patterns and clinical information [23]. Although data included in CRWD is captured from many U.S. hospitals, health systems do have the choice to participate meaning that it is not a probability-based sample nor fully representative of the entire U.S. EHR from January 2000 to April 2022 were included in these analyses.

Patients were included in the analyses if they were ≥12 years old and had either a diagnosis of OUD (“exposed” group) or a hospital encounter between January 2000 and March 2021 (“unexposed” group). Patients were excluded if they had a diagnosis code reporting history of dementia or had a diagnosis of dementia less than 30 days after their first OUD diagnosis date or visit in an affiliated hospital or clinic (accounting for possibly undiagnosed dementia not being captured until after OUD or hospital encounter date). Patients were also excluded if they had any of the codes used for identification of OUD but did not meet the full OUD inclusion methodology described later.

Study design

This study used a matched retrospective cohort design that compared exposed/OUD patients with unexposed/other patients. The study tested the hypothesis that patients diagnosed with OUD would exhibit significantly greater risk for dementia than those not diagnosed. With over 99 million patients in the database who could be eligible for inclusion in the unexposed group, a subset was randomly drawn for the unexposed group by selecting patients from a randomly selected hospital encounter. A greedy exact matching algorithm was used within this randomly selected group by matching one non-exposed patient to each exposed patient based on both patients sharing the same year of hospital visit, age, sex, andrace.

In this study, we operationalized two analyses with distinct primary outcomes. In our initial regional analysis, we used the term “senile dementia” based on specific ICD-9 and ICD-10 codes. This term was chosen due to its alignment with the CRDB coding system but is not indicative of an expectation of dementia diagnosis in the younger participants. In contrast, for our national analysis, the primary outcome was defined as AD/dementia, based on a broader set of ICD-9, ICD-10, and Systematized Nomenclature of Medicine (SNOMED)codes.

Patients were followed from their OUD diagnosis (for exposed patients) or hospital encounter (for unexposed patients) until they received an AD/dementia diagnosis or until the end of the selected follow-up time. Four set follow-up times were used in this study: one, three, five, and ten years. Within each of these follow-up scenarios, patients could be included from January 2000 up until the day from which there would be at least that many remaining years of follow-up. For example, with one year of follow-up, patients could be included until March 2021, thus allowing at least one year of follow-up by April 2022 (because outcomes of interest were restricted to only be allowed ≥30 days post first OUD diagnosis or hospital visit, an extra month is added to the one-year window to allow a full 12-month period after the inclusion period). Similarly, for three years of follow-up, patients were included up until March 2019. For five years of follow-up, patients were included up until March 2017 and for ten years of follow-up, patients were included up until March2012.

Measures

All variables in this study were identified using EHR codes as well as string matching by searching key terms in the EHR code descriptions. All code descriptions were inspected to ensure relevance to the study variables, and any descriptions that were found to be inappropriate for capturing the variables of interest were removed.

Dependent variables

The primary outcome was diagnosis of incident AD/dementia, identified using ICD-9, ICD-10, and SNOMED codes (defined in Supplementary Table 5). This outcome was initially analyzed as a binary (yes, no) incidence. For survival modeling, the outcome comprised 1) the time (in years) from OUD diagnosis or hospital encounter to AD/dementia incidence and 2) event status, whether a diagnosis of AD/dementia occurred or not. For patients without an AD/dementia diagnosis, the starting point for survival time was the date of OUD diagnosis (exposed) or random hospital encounter date (unexposed). Starting survival time for those that did receive an AD/dementia diagnosis was 30 days post OUD diagnosis/hospital visit, which allowed us to remove patients that were diagnosed with AD/dementia within 30 days of their OUD diagnosis or hospital visit. Patients exiting the hospital system earlier than follow-up without a diagnosis of AD/dementia or patients not experiencing a diagnosis by end of follow-up were right censored. In addition to “time on study,” “age at AD/dementia diagnosis or censoring” was used as an outcome. With age being such a strong predictor of AD/dementia incidence, this version of a time-scale outcome was employed based on its use in numerous studies [24 –26]. These separate outcomes allowed us to examine if bias was present due to the fact that the outcome (AD/dementia) is commonly diagnosed later in a person’s life.

Outcomes for sub-analyses looked at AD alone as well as other dementias (besides AD).

Independent variables

The primary exposure variable in the CRWD data analysis was OUD. Exposed patients were identified according to the qualifications listed by the Centers for Medicare and Medicaid Services chronic condition warehouse OUD algorithm [27]. Specifically, patients were identified as having OUD if they 1) had a qualifying ICD-9/10 or SNOMED diagnosis/procedure code on an inpatient or emergency encounter 2) had at least two diagnosis or procedure codes (or one of each) from any encounter type or 3) had a qualifying Healthcare Common Procedure Coding System (HCPCS) code or National Drug Code (NDC) for opioid medication-assisted treatment. (The full list of ICD-9/10, SNOMED, HCPCS, and NDC codes can be found in Supplementary Table 6). From all these qualifying conditions, the first date was used as the first date of OUD. Patients were removed from the analyses if they had any of the above codes but did not meet the criteria of the OUD inclusion methodology. Non-exposed participants were identified by having none of the ICD-9/10, SNOMED, HCPCS codes or NDC associated with OUD over the follow-up period.

Factors and covariates

The analysis of CRWD data was conducted using independent variables year of hospital visit (2000-2021) and age (at OUD diagnosis or hospital encounter) in years as well as sex (male, female) and race (American Indian/Alaskan Native, Asian/Pacific Islander, Black/African American, Hispanic/Latino, non-Hispanic White, other). Additional independent variables were marital status (married/partnered, single), U.S. one-digit zip code (0, 1, 2, 3, 4, 5, 6, 7, 8, 9), insurance type (private, Medicare, Medicaid, Government/Misc., self-pay, unknown), Elixhauser Agency for Healthcare Research and Quality (AHRQ) weighted comorbidity index (ECI) (< 0, 0, 1–4,> = 5; other neurodegenerative diseases and psychosis removed from calculation, drug use did not include opioid-related indications), and tobacco dependence history (yes, no) [28 –31]. The analysis also included known risk factors for AD/dementia that were not already accounted for through the ECI. Based on similar existing studies, we chose to adjust for stroke status, dyslipidemia, and family history of AD/dementia [32, 33]. To maximize inclusion of as many patients with OUD as possible, patients were not excluded due to lack of demographic information. In cases where demographics were unknown, they were categorized as such and controlled for in analyses. All covariates and additional variables were captured using ICD-9, ICD-10, SNOMED codes (list of unique codes in Supplementary Tables 7 and 8).

Statistical analysis

Demographic characteristics for the CRWD dataset were calculated for the patient sample overall and by OUD status. Standardized mean differences were calculated to indicate balance among variables used in matching and need for adjustment of all other variables [34].

Odds of incident AD/dementia for those with OUD compared to those without were presented overall as well as stratified by age groups (12–44, 45–54, 55–65, and 65+). Results were also presented for each of the four follow-up periods. Frequencies and percentages (representing the risk of AD/dementia for each group) were provided along with CMH odds ratios (ORs) and 95% confidence intervals (CIs). The Cochran-Mantel-Haenszel ORs and 95% CIs accounted for dependence between matched pairs, where each pair was treated as a stratum.

Cox proportional hazards regression was used to estimate the adjusted hazard of incident AD/dementia stratified by age and follow-up groups. Adjusted hazard ratios (aHRs) and 95% Wald CIs compared the instantaneous risk of incident AD/dementia diagnosis for those with OUD to those without. The age groups and follow-up periods were consistent with the strata used in the OR analysis. Two Cox proportional hazards regression models were constructed. The first model (Model 1) used “time on study” as the time scale, and the second (Model 2) used “age at AD/dementia or censoring” as the time scale. Both models controlled for the same confounding variables (marital status, one-digit zip code, insurance, ECI, tobacco history, stroke history, lipidemia history, family history of AD/dementia) while Model 1 additionally adjusted for age at OUD diagnosis or hospital encounter. Models accounted for possible homogeneity within matched pairs by adopting a generalized estimating equation marginal approach which estimated maximum likelihood estimates and standard errors with robust sandwich estimates [35]. Efron’s method was used to handle tied event times [36]. Model goodness-of-fit was assessed by Cox-Snell residuals and the proportional hazards assumption was assessed by scaled Schoenfeld residuals along with a global Schoenfeld test [37, 38]. SupplementaryFigure 1 indicated that the Cox-Snell residuals came from a correctly fitted model, and that the scaled Schoenfeld residuals were independent from time (no clear trend over time and an insignificant p-value). Predictor multicollinearity was assessed with variable inflation factors, which indicated no significant multicollinearity. Model adjusted survival curves were constructed, which compared the survival from AD/dementia diagnosis over time while controlling for covariates. Different lines were drawn for those with OUD and those without OUD. Curves were also stratified by age groups, but only presented for three years offollow-up.

Final sub-analyses looked at the association between OUD and the separate cognitive conditions. The first analysis examined the odds of AD for those with OUD compared to those without OUD, and the second found the odds of non-AD dementia for those with OUD compared to those without. Both were stratified by age groups and follow-up periods. These analyses were performed to determine if the odds of the separate condition outcomes changed significantly compared to the odds of both conditions as one outcome.

Due to testing trends over repeated years of follow-up, all corresponding analyses were adjusted for multiple comparisons with a Bonferroni $\frac{α}{k}$ correction in which α represents the significance level (5%) and k represents the number of comparisons (four) [39, 40]. This was done to reduce any inflation in the false positive error rate. Stratified age group analyses were part of a priori hypotheses, with results among each group of particular interest, and thus no multiple comparison adjustment was applied to those results. All hypothesis tests were two-sided, with a significance level of 5% (in the event of testing across years of follow-up, each year-iteration significance level was adjusted to 1.25%, so that all together a 5% level was maintained). All analyses were performed in R version 4.0.2 (R Foundation for Statistical Computing).

RESULTS

Regional pilot findings

The preliminary analysis performed with the regional Chicago area pilot data examined the temporal relationship between OUD and senile dementia. Among this regional patient sample, we found that more individuals received a senile dementia diagnosis after receiving an OUD diagnosis compared to patients that received a senile dementia diagnosis before an OUD diagnosis or patients that were diagnosed with both conditions in the same visit. Individuals with an OUD diagnosis were 2.62 times more likely to also be diagnosed with senile dementia than those without OUD. This result did not consider the temporality of OUD and senile dementia diagnoses but was rather an estimation of the association of the two conditions. This was an unadjusted value that did not control for any additional factors.

Figure 1 displays the temporal relationship between OUD and senile dementia among patients represented in the regional data. Of the 346 patients with both conditions, 122 were diagnosed with OUD and senile dementia on the same encounter. Of the remaining patients in this study sample, 25.7% (89 patients) were diagnosed with senile dementia before their OUD diagnosis, and 39.0% (135 patients) were diagnosed with senile dementia after their OUD diagnosis. These results suggested the need for deeper investigation into the association between OUD and AD/dementia. Because the small sample size limited the generalizability of the results, a larger national sample was used to build on these pilot findings.

Fig. 1

Temporality of chronic senile dementia diagnosis in reference to chronic opioid use disorder diagnosis with regional pilot data from the Clinical Research Database. The figure displays the temporal relationship between OUD and senile dementia among patients with data in the Clinical Research Database. The dark line down the middle of the figure represents the diagnosis date of the exposure variable, OUD, and is used to align the patient sample at time 0. The small gray bars represent the senile dementia diagnosis date of individual patients. The left side of the figure represents 10.6 years before alignment, and the right side of the figure represents 12.1 years post alignment.

National CRWD findings

Descriptive statistics

Demographic characteristics of patients from the national CRWD data are presented in Table 1. A total of 1,274,150 participants aged 12 years and older with no history of AD/dementia were identified for the CRWD analyses. The study sample consisted of n = 627,810 individuals with OUD and n = 646,340 without OUD. Around 31% of patient visits were from the years 2018–2019 and 26.2% from 2016–2017. Just under half (45.0%) of the patient sample fell into the age category 12-44 years old and 53.1% of the participants were female. Almost three-quarters of the patients were non-Hispanic white (72.1%) and 11.8% were Hispanic/Latino. Over half of the sample was unpartnered/single (55.9%) and 15.4% had an ECI score ≥5. About one third (32.6%) had private insurance, and 20.7% had Medicare, and 19.9% had Medicaid. Thirty percent of patients had a tobacco dependence history, 2.8% had a stroke history, 22.8% had a history of lipid disorders, and 0.7% had a family history of AD/dementia. The matching variables chosen for this study did achieve balance between the exposed and unexposed groups because the SMD values were close to zero and the percentage of characteristics were nearly identical between OUD and non-OUD groups. Those with OUD in the study sample were more commonly single, had a higher percentage of individuals with Medicare or Medicaid insurance, and were more burdened by histories of several chronic conditions than those without OUD.

Table 1

Demographics of patients ≥12 years old, with hospital visits between January 2000 and March 2021, for those with OUD and matched non-OUD patients (1 : 1 ratio, matched with year of hospital visit, age, sex, and race) without any previous diagnosis of AD/dementia or AD/dementia < 30 days post OUD/encounter from Cerner-affiliated hospital systems

	Total	OUD	No OUD	SMD²
		n (% ¹)	n (% ¹)
Total	1,274,150	627,810	646,340
Matching Variables
Year of hospital visit				0.007
< 2010	45121 (3.5)	22431 (3.6)	22690 (3.5)
2010–2011	37025 (2.9)	18269 (2.9)	18756 (2.9)
2012–2013	81599 (6.4)	39850 (6.3)	41749 (6.5)
2014–2015	147035 (11.5)	72200 (11.5)	74835 (11.6)
2016–2017	334462 (26.2)	164457 (26.2)	170005 (26.3)
2018–2019	394390 (31.0)	194682 (31.0)	199708 (30.9)
2020–2021	234518 (18.4)	115921 (18.5)	118597 (18.3)
Age (y)				0.022
12–44	573046 (45.0)	284735 (45.4)	288311 (44.6)
45–54	215132 (16.9)	106218 (16.9)	108914 (16.9)
55–64	227559 (17.9)	112226 (17.9)	115333 (17.8)
> = 65	258413 (20.3)	124631 (19.9)	133782 (20.7)
Sex				0.003
Female	676077 (53.1)	333552 (53.1)	342525 (53.0)
Male	598073 (46.9)	294258 (46.9)	303815 (47.0)
Race				0.004
AI/AN	15669 (1.2)	7681 (1.2)	7988 (1.2)
API	45330 (3.6)	22206 (3.5)	23124 (3.6)
Black/African American	105505 (8.3)	52254 (8.3)	53251 (8.2)
Hispanic/Latino	150129 (11.8)	74135 (11.8)	75994 (11.8)
NH-White	919019 (72.1)	452514 (72.1)	466505 (72.2)
Other	38498 (3.0)	19020 (3.0)	19478 (3.0)
Adjustment Variables
Marital Status				0.319
Single	712339 (55.9)	399930 (63.7)	312409 (48.3)
Married/Partnered	502869 (39.5)	207655 (33.1)	295214 (45.7)
Unknown	58942 (4.6)	20225 (3.2)	38717 (6.0)
U.S. One-digit zip-code				0.500
0	63722 (5.0)	28063 (4.5)	35659 (5.5)
1	248049 (19.5)	127001 (20.2)	121048 (18.7)
2	105762 (8.3)	62984 (10.0)	42778 (6.6)
3	102093 (8.0)	36700 (5.8)	65393 (10.1)
4	53887 (4.2)	23488 (3.7)	30399 (4.7)
5	53929 (4.2)	20240 (3.2)	33689 (5.2)
6	139772 (11.0)	59075 (9.4)	80697 (12.5)
7	64802 (5.1)	23783 (3.8)	41019 (6.3)
8	248255 (19.5)	173108 (27.6)	75147 (11.6)
9	193879 (15.2)	73368 (11.7)	120511 (18.6)
Insurance				0.599
Private	414857 (32.6)	161943 (25.8)	252914 (39.1)
Medicaid	254151 (19.9)	178258 (28.4)	75893 (11.7)
Medicare	264158 (20.7)	158418 (25.2)	105740 (16.4)
Govt/Misc	67250 (5.3)	26850 (4.3)	40400 (6.3)
Self-Pay	151865 (11.9)	71846 (11.4)	80019 (12.4)
Unknown	121869 (9.6)	30495 (4.9)	91374 (14.1)
Elixhauser AHRQ weighted comorbidity index (ECI)³ category				0.754
< 0	174068 (13.7)	121462 (19.3)	52606 (8.1)
0	835426 (65.6)	305885 (48.7)	529541 (81.9)
1–4	67848 (5.3)	46362 (7.4)	21486 (3.3)
> = 5	196808 (15.4)	154101 (24.5)	42707 (6.6)
Tobacco dependence history				0.860
No	892028 (70.0)	323965 (51.6)	568063 (87.9)
Yes	382122 (30.0)	303845 (48.4)	78277 (12.1)
Stroke history				0.168
No	1238884 (97.2)	601640 (95.8)	637244 (98.6)
Yes	35266 (2.8)	26170 (4.2)	9096 (1.4)
History of disorders of lipoprotein metabolism and lipidemias				0.356
No	983829 (77.2)	437828 (69.7)	546001 (84.5)
Yes	290321 (22.8)	189982 (30.3)	100339 (15.5)
Family history of AD/dementia				0.146
No	1264960 (99.3)	619330 (98.6)	645630 (99.9)
Yes	9190 (0.7)	8480 (1.4)	710 (0.1)

¹Denominator is column group; ²standardized mean difference; ³other neurodegenerative diseases and psychosis removed from calculation, drug use did not include opioid-related indications.

Inferential statistics

The unadjusted ORs and 95% CIs for the association between OUD and AD/dementia diagnosis overall for the study sample and stratified by age groups are represented in Table 2. After one year of follow-up, patients with an OUD diagnosis had 4.33 (95% CI: 4.01, 4.68) higher odds of an AD/dementia diagnosis than those without OUD. The odds of AD/dementia for those with OUD increased as the follow-up period increased to 3 years (OR [95% CI]: 4.59 [4.30, 4.89]), 5 years (OR [95% CI]: 5.17 [4.79, 5.59]), and 10 years (OR [95% CI]: 6.66 [5.60, 7.92]).

Table 2

Odds of incident AD/dementia for those with OUD compared to those without OUD (overall and stratified by age)

Follow-Up	Overall
	n (% ¹)	OR² (95% CI)
One year follow-up
No OUD	1,597 (0.25)	1 [REF]
OUD	5,872 (0.94)	4.33 (4.01, 4.68)
3 years follow-up
No OUD	2,412 (0.54)	1 [REF]
OUD	9,239 (2.12)	4.59 (4.30, 4.89)
5 years follow-up
No OUD	1,709 (0.66)	1 [REF]
OUD	7,384 (2.93)	5.17 (4.79, 5.59)
10 years follow-up
No OUD	299 (0.65)	1 [REF]
OUD	1,800 (4.03)	6.66 (5.60, 7.92)
Follow-Up	12–44		45–54		55–64		≥65 years old
	n (% ¹)	OR (95% CI)	n (% ¹)	OR (95% CI)	n (% ¹)	OR (95% CI)	n (% ¹)	OR (95% CI)
One year follow-up
No OUD	51 (0.02)	1 [REF]	51 (0.05)	1 [REF]	150 (0.13)	1 [REF]	1,345 (1.01)	1 [REF]
OUD	702 (0.25)	14.31 (9.89, 20.69)	564 (0.53)	11.26 (7.79, 16.28)	956 (0.85)	6.71 (5.36, 8.40)	3,650 (2.93)	3.22 (2.94, 3.51)
3 years follow-up
No OUD	69 (0.03)	1 [REF]	93 (0.12)	1 [REF]	218 (0.28)	1 [REF]	2,032 (2.25)	1 [REF]
OUD	1,141 (0.57)	17.87 (12.90, 24.76)	955 (1.26)	10.49 (7.94, 13.86)	1,538 (2.01)	7.77 (6.40, 9.43)	5,605 (6.66)	3.27 (3.04, 3.53)
5 years follow-up
No OUD	65 (0.05)	1 [REF]	86 (0.19)	1 [REF]	188 (0.44)	1 [REF]	1,370 (3.04)	1 [REF]
OUD	1,201 (0.97)	19.18 (13.82, 26.63)	988 (2.22)	12.06 (9.00, 16.16)	1,330 (3.27)	7.78 (6.23, 9.48)	3,865 (9.20)	3.30 (3.01, 3.62)
10 years follow-up
No OUD	31 (0.11)	1 [REF]	31 (0.37)	1 [REF]	40 (0.78)	1 [REF]	197 (4.49)	1 [REF]
OUD	623 (2.27)	21.59 (13.21, 35.31)	397 (4.95)	12.82 (7.85, 20.93)	294 (5.90)	6.78 (4.37, 10.53)	486 (11.50)	2.74 (2.15, 3.48)

¹Denominator is row group; ²odds ratio.

When stratifying by age groups, younger patients had a greater disparity in odds of AD/dementia between OUD and non-OUD patients, while older patients exhibited a lesser disparity in AD/dementia odds. However, overall incidence of AD/dementia was much smaller among younger patients relative to older patients. The relative differences in AD/dementia odds for the 12–44 age group increased as the follow-up period increased from 3 years, to 5 years, to 10 years. A similar trend was seen in the age group of 45–54 years where the odds of AD/dementia increased as the follow-up time increased to 3, 5, and 10 years. The age groups of 55–64 years old and 65 + years exhibited a slight increase in the odds of AD/dementia among the exposed for the 3-and 5-year follow-up periods, but a decrease in odds of AD/dementia for the 10-year follow-up period.

Table 3 displays the aHRs of AD/dementia for those with OUD compared to those without for the overall patient sample and for stratified age groups over varying follow-up periods. In Table 3, both Model 1 and Model 2 are Cox regression models. However, the time scale used for Model 1 was “time on study” while “age at dementia/censoring” was the time scale used for Model 2. Throughout all scenarios, Model 1 and Model 2 display significantly higher hazard of AD/dementia for those with OUD compared to those without. The instantaneous risk of AD/dementia for the overall participant sample was very similar for the 1-year and 3-year follow-up periods (Model 1: aHR [95% CI]: 1.88 [1.74, 2.03] and aHR [95% CI]: 1.84 [1.73, 1.96], respectively). As the follow-up time increased from 3 to 5 years (Model 1: aHR [95% CI]: 1.91 [1.78, 2.06]), the aHR of AD/dementia for the overall participant sample slightly increased and greatly increased from 5 to 10 years of follow-up (Model 1: aHR [95% CI]: 3.11 [2.63, 3.69]).

Table 3

Adjusted hazard of incident AD/dementia for those with OUD compared to those without OUD (stratified by age and follow-up groups)

Follow-Up	Overall
	aHR¹ (95% CI)
One year follow-up
Model 1²	1.88 (1.74, 2.03)
Model 2³	1.89 (1.76, 2.02)
3 years follow-up
Model 1	1.84 (1.73, 1.96)
Model 2	1.90 (1.78, 2.02)
5 years follow-up
Model 1	1.91 (1.78, 2.06)
Model 2	1.98 (1.84, 2.13)
10 years follow-up
Model 1	3.11 (2.63, 3.69)
Model 2	3.17 (2.67, 3.75)
Follow-Up	12-44	45-54	55-64	65+
	aHR (95% CI)	aHR (95% CI)	aHR (95% CI)	aHR (95% CI)
One year follow-up
Model 1	5.48 (3.75, 8.00)	4.59 (3.11, 6.75)	2.59 (2.04, 3.29)	1.56 (1.42, 1.70)
Model 2	5.51 (3.83, 7.91)	4.59 (3.18, 6.61)	2.59 (2.04, 3.29)	1.56 (1.44, 1.68)
3 years follow-up
Model 1	5.83 (4.22, 8.05)	3.81 (2.85, 5.10)	2.71 (2.22, 3.29)	1.52 (1.41, 1.63)
Model 2	6.18 (4.47, 8.55)	3.98 (2.97, 5.32)	2.78 (2.28, 3.38)	1.55 (1.44, 1.66)
5 years follow-up
Model 1	6.40 (4.60, 8.91)	4.27 (3.16, 5.76)	2.55 (2.06, 3.15)	1.46 (1.34, 1.59)
Model 2	6.85 (4.92, 9.55)	4.49 (3.32, 6.07)	2.62 (2.12, 3.24)	1.49 (1.37, 1.63)
10 years follow-up
Model 1	8.08 (5.01, 13.02)	6.33 (3.85, 10.39)	3.61 (2.29, 5.69)	1.60 (1.26, 2.02)
Model 2	8.39 (5.20, 13.54)	6.49 (3.95, 10.66)	3.66 (2.32, 5.77)	1.63 (1.29, 2.07)

¹Adjusted hazard ratio; ²Cox proportional hazards regression using “time on study” as time scale, adjusted for age at OUD/encounter, marital status, one-digit zip code, insurance, ECI, tobacco history, stroke history, lipidemia history, and history of AD/dementia. ³Cox proportional hazards regression using “age at dementia/censoring” as time scale, adjusted for same variables except for age at OUD/encounter.

When stratifying by age groups, similar results were seen as in previous age-group trends. Although all age groups exhibited higher hazard of AD/dementia for OUD compared to non-OUD, younger patients had higher aHRs while older patients had lower aHRs. In Table 3, the risk of AD/dementia for patients with OUD compared to those without showed several notable trends. First, both Cox proportional hazard regression models (Models 1 and 2) demonstrated similar results for all comparable age groups and follow-up times. Second, an increase in risk of AD/dementia was observed as the follow-up period increased among those aged 12-44 years old. Among the age group 45–54 years old, the risk of AD/dementia decreased from 1 to 3 years of follow-up and then increased from 5 to 10 years of follow-up. The risk of AD/dementia decreased from the 3-year to 5-year follow-up and increased at the 10-year follow-up for the 55–64-year-old age group. For those 65 years and older, the risk of AD/dementia slightly decreased from 1-year to 3-year to 5-years of follow-up and then increased at 10 years of follow-up. Lastly, consistent with the previous unadjusted ORs, all aHRs were highest among the youngest age group (12-44 years old) and decreased gradually across older age groups. All hazard ratios were significant because their 95% CIs did not overlap with 1.00 (p < 0.05).

Model adjusted survival curves for 3 years of follow-up are presented in Fig. 2. The curves show that those diagnosed with OUD had lower survival from AD/dementia (higher hazard of AD/dementia) than those never diagnosed with OUD. This was consistent across all age groups. The curves also show that overall survival from AD/dementia, for both OUD and non-OUD, decreases as age increases.

Fig. 2

Adjusted survival (from incident AD/dementia) curves for those with OUD and those without OUD with three years of follow-up (top left: 12–44; top right: 45–54, bottom left: 55–64, bottom right: ≥65).

Supplementary Table 1 displays the aHRs calculated from the full regression model (using “time on study” as the time scale) stratified by age groups as well as by 3- and 5-year follow-up group. Compared to private insurance, all other insurance types had higher risk of AD/dementia and those were married/partnered had lower risk of AD/dementia compared to single individuals. Those without a history of tobacco dependence, stroke, lipidemia disorder, or a family history of AD/dementia all had lower risk of AD/dementia than those with such conditions noted in their EHR.

Sub-analyses

Sub-analyses were conducted to examine if the primary results for the association between OUD and AD/dementia would be affected if the outcome were split into the single conditions of AD and non-AD dementia. Supplementary Table 2 displays the odds of AD for those with OUD compared to those without OUD stratified by age and follow-up groups. Supplementary Table 3 displays the odds of non-AD dementia for those with OUD compared to those without OUD stratified by age and follow-up groups. All ORs were computed comparing those with OUD diagnosis to those without. Across all age groups and follow-up times, those with OUD had higher odds of AD and higher odds of non-AD dementia compared to those without OUD. Specifically, the incidence of AD was much lower than the incidence of non-AD dementia, with AD incidence mostly occurring in patients 55 and older. Non-AD dementia incidence was much more common across all age groups. Additionally, the ORs of non-AD dementia were higher than the ORs of AD for all age groups. Consistent with the previous OR and aHR analyses, the disparity in odds of AD/non-AD dementia between OUD and non-OUD patients was much greater among younger patients, while the incidence of the outcomes was higher for older patients.

DISCUSSION

This study suggests that individuals with OUD have greater risk for AD/dementia diagnosis than those who do not have OUD. This association was strongest among younger age groups and gradually decreased when examining older cohorts but remained significant. The length of follow-up period also altered the magnitude of these associations, with significantly higher risk for AD/dementia among those with OUD compared to those without occurring at longer follow-up periods. These findings highlight foundational risk estimates for dementia among people with OUD and confirm our hypothesis that this patient population may be at heightened risk for such cognitive conditions.

While we found a strong association between OUD and AD/dementia, a limited number of studies to date have examined links between forms of opioid use and AD/dementia, and those studies found weaker relationships than our analyses demonstrated. For example, Dublin et al. (2015) examined the association between prescription opioid use and dementia, where opioid use was defined through prescription fills collected by computerized pharmacy data [14]. Their study found a slightly higher risk for AD/dementia among people with the highest cumulative use of prescription opioids compared to people with little to no use (aHR for all cause dementia = 1.29, 95% CI = 1.02–1.62) [14]. However, they found no association with AD or dementia for participants with moderate prescription opioid drug exposure compared to those with little to no use which may have been due to the limitations of using prescription data for participant inclusion [14]. Similarly, a study by Taipale et al. (2017) used a national prescription registry to examine the association between prescription opioid use and risk of AD among Finnish patients and found no significant difference in risk of AD [41]. Since our study examines U.S. patients diagnosed with OUD and does not measure use of prescribed opioids, differences in the sample populations and their typical use of opioids likely drive key differences between our studies. For example, the 2016 Centers for Disease Control (CDC) guideline for prescribing opioids for chronic pain explicitly cautioned against opioid prescribing for patients with chronic pain and OUD, with the exception of methadone and buprenorphine as treatment medications [42]. While early studies of opioid prescribing indicate patients with co-occurring chronic pain and substance use disorders may have received prescription opioids at higher rates than other patients in the late 1990 s and early 2000 s [43, 44], more recent studies suggest many measures of opioid prescribing declined significantly after the CDC guideline was published [45, 46], and this occurred across diverse kinds of pain, including chronic non-cancer, acute, and cancer-related pain [47, 48]. There is limited evidence of the extent to which decreases in prescribing may have disproportionately occurred among patients with OUD, however it is possible that patients with OUD received disproportionately less prescription opioids in the wake of this guideline and rising national attention to OUD. Recent small sample qualitative research suggests patients with substance use disorders often struggle to receive opioid analgesia [49, 50]. Thus, the patients included in the Dublin et al. and Taipale et al. samples are likely to have been exclusively using regulated and relatively safer forms of prescription opioids, while the OUD patients included in our study likely use/used illicitly-manufactured opioids of unknown purity and strength. Future research may examine how different kinds of opioid use, such as use of prescription opioids versus use of illicitly-manufactured opioids, may contribute to varied levels of risk for cognitive conditions.

Further, it is crucial to discuss the potential biological implications of OUD on brain health and the onset of dementia, even though the exact underlying mechanisms remain to be fully elucidated. Directly, chronic opioid use could exert neurotoxic effects on the brain. Opioids can cause hypoxia [51], a condition that arises from the lack of oxygen reaching the brain, which can lead to brain damage. The brain damage incurred from prolonged periods of hypoxia could contribute to cognitive decline and an increased risk of dementia [52, 53]. Additionally, chronic opioid use has been associated with neuroinflammation [54], a state of chronic inflammation within the brain that is recognized as a contributing factor in the pathogenesis of AD and other forms of dementia [55]. It is plausible that the neuroinflammatory response elicited by chronic opioid use could increase the vulnerability of individuals to neurodegenerative disorders. Indirectly, opioid misuse could also lead to lifestyle changes that increase dementia risk. For instance, opioids can disrupt sleep patterns [56], and there is growing evidence to suggest that sleep quality and duration are associated with dementia risk [57]. Opioids may also lead to a decline in physical health and an increase in sedentary behavior, which have both been associated with an increased risk of cognitive decline [58].

Moreover, opioids can increase the risk of trauma and infection, leading to conditions like HIV that are associated with a higher risk of dementia. Specifically, neurocognitive impairment observed in some people living with HIV has been compared to early stages of neurodegenerative diseases, suggesting a potential overlapping pathway [59, 60]. It is also worth noting that the relationship between opioid use and cognitive decline could be bidirectional. While opioid use can lead to conditions that increase the risk of dementia, cognitive impairment may likewise contribute to increased use of opioids. Therefore, the association between OUD and dementia that we observed in our study could be influenced by a combination of these direct and indirect effects. Further research is needed to untangle these complex relationships and to better understand the biological mechanisms that underpin the link between opioid use and dementia.

Past research on the relationship between opioid use and dementia has primarily used samples of older adults (65 + years) [14 , 61] because they have the highest prevalence of such cognitive conditions [62]. But since OUD is most prevalent among young adult age groups (around 20-30 years old) and tends to decline with age [63], inclusion of younger populations can reveal key intersections of these conditions with inverse age trends. Our use of a large national database provided a sufficient sample size for estimating differences in risk for AD/dementia among people with OUD even in younger age groups where dementia is relatively rare. We observed the greatest difference in risk for AD/dementia among the people with OUD in youngest age group (12-44 years). While it is unclear why this trend exists, future research may seek causal explanations that consider the social (e.g., incarceration, employment problems) and physiological effects of OUD, and the accumulation of other dementia risk factors over time that may “crowd out” these effects of OUD over the life course. Factors such as high blood pressure, damaged blood vessels, diabetes, reduced ability to recover from injury, and a weaker immune system can compound over time making an older person more likely to have problems with thinking and memory [64]. Because younger individuals are less likely to be coping with these other conditions associated with dementia risk, our findings suggest that OUD could be a dominant influencer on the early development of AD/dementia even if the likelihood of developing AD/dementia in younger age is quite small. Differences in risk for developing AD/dementia among older people with OUD relative to their peers without OUD may be diminished because additional age-related factors are more likely to be present and contribute to declines in cognitive functioning.

Limitations

There is a significant amount of variability within the different types of dementia that might have influenced our results. However, we did review similar studies and followed published guidelines that explained which codes should be used to receive an accurate capture for the conditions of interest. Because AD usually develops in later life and often takes many years to display symptoms, it is often only diagnosed in older adults (65 years and older). To account for the lack of AD diagnosis in younger patients with OUD, we included other dementia diagnoses in the main outcome. Another limitation relating to the outcome is that we did not consider how rapidly progressive or chronic dementia may affect the results, specifically among non-AD dementias which had a much higher incidence than more chronic AD. Future studies should examine if similar risk estimates are present among rapidly progressive and chronic cognitive decline diagnoses for people with OUD. It is possible that people with an OUD diagnosis earlier in life may recover from rapidly progressive dementia diagnosis where older individuals with OUD may only very rarely have such outcomes.

Conclusion

OUD and dementia are both prevalent conditions of significant public health and medical concern. Our identification of individuals with OUD as a key at-risk population for AD and other forms of dementia can inform targeted prevention and treatment interventions. Research has shown that medication cannot easily reverse or halt the progression of AD/dementia once in the mild-moderate stages. Early detection of these disorders is likely the key to preventing serious cognitive decline. The results of this exploratory study highlight early detection may be strategically deployed through sites and programs serving people with OUD. Future research may build towards a more robust understanding of the association between OUD and AD/dementia to inform the design of programs that may help lessen the AD/dementia disparity among individuals with OUD.

Footnotes

ACKNOWLEDGMENTS

The authors acknowledge Oracle Cerner for the national data access and computation capabilities. We also thank Mr. Steven Birch for completing data queries for the pilot study. We thank Rona Bern for her work on the initial literature review for the study.

FUNDING

FQ, KE, and EFM are partially funded from NIH grants (1) 5R61DA049382: Leveraging CDC opioid overdose surveillance funding from the Albuquerque area southwest tribal epidemiology center to create tribal data and culturally center medications for opioid use disorder; and (2) R01DA057658: Weighting Longitudinal Data to Access Opioid Analgesia Tapering Outcomes among Patients with Co-occurring Chronic Pain and Substance Use Disorder.

CONFLICT OF INTEREST

The authors have no conflict of interest to report.

DATA AVAILABILITY

The datasets generated during and/or analyzed during the current study are not publicly available due to restrictions by Oracle Cerner, the owner of the data. Data could be accessed by signing a data sharing agreement with Oracle Cerner and covering any costs that may be involved (Contact Kendra Stillwell: kendra.stillwell@cernerenviza.com).

The supplementary material is available in the electronic version of this article: .

References

Jeon

, Han

, Jeong

, Park

, Yoo

, Lee

, Kim

, Shin

(2023) Changes in alcohol consumption and risk of dementia in anationwide cohort in South Korea. JAMA Netw Open 6, e2254771.

Hulse

, Lautenschlager

, Tait

, Almeida

(2005) Dementiaassociated with alcohol and other drug use. Int Psychogeriatr 17 Suppl 1, S109–S127.

Rehm

, Hasan

OSM

, Black

, Shield

, Schwarzinger

(2019) Alcohol use and dementia: A systematic scoping review. Alzheimers Res Ther 11, 1.

Prince

, Bryce

, Albanese

, Wimo

, Ribeiro

, Ferri

(2013) The global prevalence of dementia: A systematic review and metaanalysis. Alzheimers Dement 9, 63–75.

Schwarzinger

, Pollock

, Hasan

OSM

, Dufouil

, Rehm

QalyDays Study Group(2018) Contribution of alcohol use disorders to the burden of dementia in France 2008–13: A nation wide retrospective cohort study. Lancet Public Health 3, e124–e132.

Keyes

, Rutherford

, Hamilton

, Barocas

, Gelberg

, Mueller

, Feaster

, El-Bassel

, Cerdá

(2022) What is the prevalence of and trend in opioid use disorder in the United States from 2010 to 2019? Using multiplier approaches to estimate prevalence for an unknown population size. Drug Alcohol Depend Rep 3, 100052.

Webster

, Cochella

, Dasgupta

, Fakata

, Fine

, Fishman

, Grey

, Johnson

, Lee

, Passik

, Peppin

, Porucznik

, Ray

, Schnoll

, Stieg

, Wakeland

(2011) An analysis of the rootcauses for opioid-related overdose deaths in the United States. Pain Med 12 Suppl 2, S26–S35.

Wilder

, Miller

, Tiffany

, Winhusen

, Winstanley

, Stein

(2016) Risk factors for opioid overdose and awareness of overdoserisk among veterans prescribed chronic opioids for addiction orpain. J Addict Dis 35, 42–51.

Milroy

, Parai

(2011) The histopathology of drugs of abuse. Histopathology 59, 579–593.

10.

Herlinger

, Lingford-Hughes

(2022) Opioid use disorder and the brain: A clinical perspective. Addiction 117, 495–505.

11.

Wollman

, Alhassoon

, Hall

, Stern

, Connors

, Kimmel

, Allen

, Stephan

, Radua

(2017) Gray matter abnormalities inopioid-dependent patients: A neuroimaging meta-analysis. Am JDrug Alcohol Abuse 43, 505–517.

12.

Wollman

, Alhassoon

, Stern

, Hall

, Rompogren

, Kimmel

, Perez-Figueroa

(2015) White matter abnormalities in long-termheroin users: A preliminary neuroimaging meta-analysis. Am J Drug Alcohol Abuse 41, 133–138.

13.

Stout

, Jernigan

, Archibald

, Salmon

(1996) Association ofdementia severity with cortical gray matter and abnormal whitematter volumes in dementia of the Alzheimer type. Arch Neurol 53, 742–749.

14.

Dublin

, Walker

, Gray

, Hubbard

, Anderson

, Yu

, Crane

, Larson

(2015) Prescription opioids and risk of dementia orcognitive decline: A prospective cohort study. J Am GeriatrSoc 63, 1519–1526.

15.

Levine

, Rotstein

, Goldberg

, Reichenberg

, Kodesh

(2023) Opioid exposure and the risk of dementia: A national cohort study. Am J Geriatr Psychiatry 31, 315–323.

16.

Bruijnen

, Dijkstra

, Walvoort

, Markus

, VanDerNagel

, Kessels

, De Jong

(2019) Prevalence of cognitive impairment inpatients with substance use disorder. Drug Alcohol Rev 38, 435–442.

17.

Sanjari Moghaddam

, Shadloo

, Shahkhah

, Tafakhori

, Haghshomar

, Meshkat

, Aghamollaii

(2021) Cognitive impairment in opium usedisorder. Behav Neurol 2021, 5548623.

18.

Sanborn

, Gunstad

, Shrestha

, Mistler

, Copenhaver

(2022) Cognitive profiles in persons with opioid use disorder enrolled inmethadone treatment. Appl Neuropsychol Adult 29, 462–468.

19.

(2023) 2023 Alzheimer’s disease facts and figures. AlzheimersDement 19, 1598–1695.

20.

Von Elm

, Altman

, Egger

, Pocock

, Gøtzsche

, Vandenbroucke

(2007) The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: Guidelines for reporting observational studies. Lancet 370, 1453–1457.

21.

Information Technology Services, Loyola University Chicago (2022) The Clinical Research Database (CRDB). Accessed September 2023. Available from: https://www.luc.edu/its/rcs/clinicalresearchresources/theclinicalresear-chdatabasecrdb/.

22.

Cerner (2023) Real-World Data solution. Accessed September 2023. Available from: https://www.cerner.com/solutions/real-world-data.

23.

Ehwerhemuepha

, Carlson

, Moog

, Bondurant

, Akridge

, Moreno

, Gasperino

, Feaster

(2022) Cerner real-world data (CRWD)-Ade-identified multicenter electronic health records database. Data Brief 42, 108120.

24.

Sabia

, Fayosse

, Dumurgier

, Dugravot

, Akbaraly

, Britton

, Kivimäki

, Singh-Manoux

(2018) Alcohol consumption and riskof dementia: 23 year follow-up of Whitehall II cohort study. BMJ 362, k2927.

25.

Mayeda

, Glymour

, Quesenberry

, Whitmer

(2016) Inequalities in dementia incidence between six racial and ethnicgroups over 14 years. Alzheimers Dement 12, 216–224.

26.

Lamarca

, Alonso

, Gomez

, Muñoz

(1998) Left-truncated data with age as time scale: An alternative for survival analysis in the elderly population. J Gerontol A Biol Sci Med Sci 53, M337–M343.

27.

Chronic Conditions Warehouse (2023) Other Chronic Health, Mental Health, and Potentially Disabling Con- ditions Algorithms. Accessed September 2023. Avail- able from: https://www2.ccwdata.org/documents/10280/ 19140001/oth-cond-algo-oud.pdf.

28.

Elixhauser

, Steiner

, Harris

, Coffey

(1998) Comorbidity measures for use with administrative data. Med Care 36, 8–27.

29.

Garland

, Fransoo

, Olafson

, Ramsey

, Yogendren

, Chateau

, McGowan

(2012) The Epidemiology and Outcomes of Critical Illness in Manitoba. Winnipeg, MB: Manitoba Centre for Health Policy. Accessed January 2022. Available from:http://mchp-appserv.cpe.umanitoba.ca/reference/MCHP_ICU_Report_WEB_(20120403).pdf .

30.

Moore

, White

, Washington

, Coenen

, Elixhauser

(2017) Identifying increased risk of readmission and in-hospital mortality using hospital administrative data. Med Care 55, 698–705.

31.

Menendez

, Neuhaus

, van Dijk

, Ring

(2014) The Elixhauser comorbidity method outperforms the Charlson index in predicting in patient death after orthopaedic surgery. Clin Orthop Relat Res 472, 2878–2886.

32.

Kivipelto

, Mangialasche

, Ngandu

(2018) Lifestyle interventions to prevent cognitive impairment, dementia and Alzheimer disease. Nat Rev Neurol 14, 653–666.

33.

Kuźma

, Lourida

, Moore

, Levine

, Ukoumunne

, Llewellyn

(2018) Stroke and dementia risk: A systematic review andmeta-analysis. Alzheimers Dement 14, 1416–1426.

34.

Austin

(2011) An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res 46, 399–424.

35.

Lin

, Wei

(1989) The robust inference for the Cox proportional hazards model. J Am Stat Assoc 84, 1074–1078.

36.

Efron

(1977) The efficiency of Cox’s likelihood function for censored data. J Am Stat Assoc 72, 557–565.

37.

Xue

, Schifano

(2017) Diagnostics for the Cox model. Commun Stat Appl Methods 24, 583–604.

38.

Schoenfeld

(1982) Partial residuals for the proportional hazards regression model. Biometrika 69, 239–241.

39.

Bonferroni CE (1935) Il calcolo delle assicurazioni su gruppi di teste. Studi in Onore del Professore Salvatore Ortu Carboni, pp. 13-60.

40.

Bonferroni

(1936) Teoria statistica delle classi e calcolo delle probabilita. Pubblicazioni del R Istituto Superiore di Scienze Economiche e Commericiali di Firenze 8, 3–62.

41.

Taipale

, Hamina

, Lampela

, Tanskanen

, Tiihonen

, Karttunen

, Tolppanen

, Hartikainen

(2018) Is Alzheimer’s disease associated with previous opioid use? Pain Med 19, 2115–2121.

42.

Dowell

, Haegerich

, Chou

(2016) CDC guideline for prescribing opioids for chronic pain—United States, 2016. JAMA 315, 1624–1645.

43.

Edlund

, Martin

, Devries

, Fan

, Braden

, Sullivan

(2010) Trends in use of opioids for chronic non-cancer pain among individuals with mental health and substance use disorders: TheTROUP study. Clin J Pain 26, 1–8.

44.

Weisner

, Campbell

, Ray

, Saunders

, Merrill

, Banta-Green

, Sullivan

, Silverberg

, Mertens

, Boudreau

, Von Korff

(2009) Trends in prescribed opioid therapy for non-cancer pain for individuals with prior substance use disorders. Pain 145, 287–293.

45.

Dayer

, Breckling

, Kling

, Lakkad

, McDade

, Painter

(2019) Association of the “CDC Guideline for Prescribing Opioids for Chronic Pain” with emergency department opioid prescribing. J Emerg Med 57, 597–602.

46.

Goldstick

, Guy

, Losby

, Baldwin

, Myers

, Bohnert

(2021) Changes in initial opioid prescribing practices after the 2016 release of the CDC guideline for prescribing opioids forchronic pain. JAMA Netw Open 4, e2116860–e2116860.

47.

Encinosa

, Bernard

, Selden

(2022) Opioid and non-opioid analgesic prescribing before and after the CDC’s 2016 opioid guideline. Int J Health Econ Manag 22, 1–52.

48.

Agarwal

, Roberts

, Dusetzina

, Royce

(2020) Changes in opioid prescribing patterns among generalists and oncologists for Medicare part D beneficiaries from 2013 to 2017. JAMA Oncol 6, 1271–1274.

49.

Simon

, Snow

, Wakeman

(2020) Understanding why patients with substance use disorders leave the hospital against medical advice: A qualitative study. Subst Abus 41, 519–525.

50.

Hawk

, McCormack

, Edelman

, Coupet

Jr , Toledo

, Gauthier

, Rotrosen

, Chawarski

, Martel

, Owens

, Pantalon

, O’Connor

, Whiteside

, Cowan

, Richardson

, Lyons

, Rothman

, Marsch

, Fiellin

, D’Onofrio

(2022) Perspectives about emergencydepartment care encounters among adults with opioid use disorder. JAMA Netw Open 5, e2144955.

51.

Kiyatkin

(2019) Respiratory depression and brain hypoxia induced by opioid drugs: Morphine, oxycodone, heroin, and fentanyl. Neuropharmacology 151, 219–226.

52.

Calabrese

, Giordano

, Signorile

, Laura Ontario

, Castorina

, De Pasquale

, Eckert

, Calabrese

(2016) Major pathogenic mechanisms in vascular dementia: Roles of cellular stress response and hormesis in neuroprotection. J Neurosci Res 94, 1588–1603.

53.

Snyder

, Simone

, Giovannetti

, Floyd

(2021) Cerebralhypoxia: Its role in age-related chronic and acute cognitivedys function. Anesth Analg 132, 1502.

54.

Stoicea

, Russell

, Weidner

, Durda

, Joseph

, Yu

, Bergese

(2015) Opioid-induced hyperalgesia in chronic pain patients and the mitigating effects of gabapentin. Front Pharmacol 6, 104.

55.

Heneka

, Carson

, El Khoury

, Landreth

, Brosseron

, Feinstein

, Jacobs

, Wyss-Coray

, Vitorica

, Ransohoff

, Herrup

, Frautschy

, Finsen

, Brown

, Verkhratsky

, Yamanaka

, Koistinaho

, Latz

, Halle

, Petzold

, Town

, Morgan

, Shinohara

, Perry

, Holmes

, Bazan

, Brooks

, Hunot

, Joseph

, Deigendesch

, Garaschuk

, Boddeke

, Dinarello

, Breitner

, Cole

, Golenbock

, Kummer

(2015) Neuroinflammation in Alzheimer’s disease. Lancet Neurol 14, 388–405.

56.

Eacret

, Veasey

, Blendy

(2020) Bidirectional relationship between opioids and disrupted sleep: Putative mechanisms. Mol Pharmacol 98, 445–453.

57.

Spira

, Gamaldo

, An

, Wu

, Simonsick

, Bilgel

, Zhou

, Wong

, Ferrucci

, Resnick

(2013) Self-reported sleep and β-amyloid deposition in community-dwelling older adults. JAMA Neurol 70, 1537–1543.

58.

Santos-Lozano

, Pareja-Galeano

, Sanchis-Gomar

, Quindós-Rubial

, Fiuza-Luces

, Cristi-Montero

, Emanuele

, Garatachea

, Lucia

(2016) Physical activity and Alzheimer disease: A protective association. Mayo Clin Proc 91, 999–1020.

59.

Altice

, Kamarulzaman

, Soriano

, Schechter

, Friedland

(2010) Treatment of medical, psychiatric, and substance-use comorbidities in people infected with HIV who use drugs. Lancet 376, 367–387.

60.

Heaton

, Clifford

, Franklin

Jr , Woods

, Ake

, Vaida

, Ellis

, Letendre

, Marcotte

, Atkinson

, Rivera-Mindt

, Vigil

, Taylor

, Collier

, Marra

, Gelman

, McArthur

, Morgello

, Simpson

, McCutchan

, Abramson

, Gamst

, Fennema-Notestine

, Jernigan

, Wong

, Grant

CHARTER Group.(2010) HIV-associated neurocognitive disorders persist in the era of potent antiretroviral therapy: CHARTER Study. Neurology 75, 2087–2096.

61.

Shen

, Zhao

, Dwibedi

, Wiener

, Findley

, Sambamoorthi

(2018) Opioid use and the presence of Alzheimer’s disease and related dementias among elderly Medicare beneficiaries diagnosed with chronic pain conditions. Alzheimers Dement (N Y) 4, 661–668.

62.

Alzheime’s Society (2023) Who gets Alzheime’s disease. Accessed September 2023. Available from: https://www.alzheimers.org.uk/about-dementia/types-dementia/who-gets-alzheimers-disease.

63.

Vasilenko

, Evans-Polce

, Lanza

(2017) Age trends in rates ofsubstance use disorders across ages 18–90: Differences bygender and race/ethnicity. Drug Alcohol Depend 180, 260–264.

64.

Alzheime’s Society (2021) What can increase a perso’srisk of dementia. Accessed September 2023. Available from: https://www.alzheimers.org.uk/about-dementia/risk-factors-and-prevention/what-can-increase-persons-risk-of- dementia.