Non-Cancer Chronic Pain Conditions and Risk for Incident Alzheimer’s Disease and Related Dementias in Community-Dwelling Older Adults: A Population-Based Retrospective Cohort Study of United States Medicare Beneficiaries, 2001–2013

Accumulating evidence suggests that certain chronic pain conditions may increase risk for incident Alzheimer’s disease and related dementias (ADRD). Rigorous longitudinal research remains relatively sparse, and the relation of overall chronic pain condition burden to ADRD risk remains little studied, as has the potential mediating role of sleep and mood disorders. In this retrospective cohort study, we investigated the association of common non-cancer chronic pain conditions (NCPC) at baseline to subsequent risk for incident ADRD, and assessed the potential mediating effects of mood and sleep disorders, using baseline and 2-year follow-up data using 11 pooled cohorts (2001–2013) drawn from the U.S. Medicare Current Beneficiaries Survey (MCBS). The study sample comprised 16,934 community-dwelling adults aged ≥65 and ADRD-free at baseline. NCPC included: headache, osteoarthritis, joint pain, back or neck pain, and neuropathic pain, ascertained using claims data; incident ADRD (N = 1149) was identified using claims and survey data. NCPC at baseline remained associated with incident ADRD after adjustment for sociodemographics, lifestyle characteristics, medical history, medications, and other factors (adjusted odds ratio (AOR) for any vs. no NCPC = 1.21, 95% confidence interval (CI) = 1.04–1.40; p = 0.003); the strength and magnitude of this association rose significantly with increasing number of diagnosed NCPCs (AOR for 4+ vs. 0 conditions = 1.91, CI = 1.31–2.80, p-trend < 0.00001). Inclusion of sleep disorders and/or depression/anxiety modestly reduced these risk estimates. Sensitivity analyses yielded similar findings. NCPC was significantly and positively associated with incident ADRD; this association may be partially mediated by mood and sleep disorders. Additional prospective studies with longer-term follow-up are warranted to confirm and extend our findings.


Introduction
Alzheimer's disease and related dementia (ADRD) is a group of serious neurodegenerative disorders characterized by progressive decline in cognitive and psychomotor function [1,2]. Prevalence of ADRD continues to increase worldwide, exacting enormous social, economic and healthcare costs. For example, in the United States, at least 5.8 million adults are living with Alzheimer's disease (AD), the most common form of dementia, with attributable Medicare costs alone exceeding $146 billion participants' health status, health services utilization, prescription medications, and payment sources using a combination of survey and administrative records. Our study used data from MCBS Cost and Use files linked with Medicare fee-for-service claims to ascertain demographics, access to care, lifestyle factors, medical history and medication. Based on the recommendations by MCBS investigators [36][37][38] and sampling strategies documented in prior published studies using MCBS [36,38,39], we combined 11 MCBS cohorts in order to maximize reliability and precision of our study estimates; these included the following cohorts: 2001-2003; 2002-2004; 2003-2005; 2004-2006; 2005-2007; 2006-2008; 2007-2009; 2008-2010; 2009-2011; 2010-2012 and 2011-2013.

Study Sample
The study sample comprised continuously fee-for-service enrolled community-dwelling Medicare beneficiaries, aged 65 years or over, who had complete information on NCPC status and were still alive at the end of follow-up. Institutionalized participants were excluded, as were participants with diagnosed ADRD at baseline. As depicted in the sample selection flow chart (see supplementary Figure S1), application of all a priori exclusion criteria yielded a final sample size of 16,934  IRB/ethics approval: The present study was approved as exempt protocol by the WVU Institutional Review Board (IRB), due to the deidentified nature of the data used in the study.

Dependent Variable: Incident Alzheimer's Disease and Related Dementia (ADRD) at Follow-Up-Yes/No
The Medicare fee-for-service (FFS) claims for inpatient (IP), skilled nursing facility (SNF), outpatient (OT), home health agency (HHA), and physician office (PO) visits for years 2001-2013 as well as MCBS self-reported Health Status and Functioning files were used to ascertain the presence of ADRD at baseline and follow-up. The presence of ADRD at both baseline (year 1) and follow-up (years 2 and 3) was ascertained using a validated CMS algorithm (Centers for Medicare and Medicaid Services (CMS) Chronic Condition Algorithms) [40] of at least one fee-for-service claim with any of these International Classification of Diseases, ninth Edition, clinical modification (ICD-9-CM) diagnostic codes: 331.0, 331.11, 331.19, 331.2, 331.7, 290.0, 290. 10, 290.11, 290.12, 290.13, 290.20, 290.21, 290.3, 290.40, 290.41, 290.42, 290.43, 294.0, 294.10, 294.11, 294.20, 294.21, 294.8, and 797, or an affirmative response to the self-reported Health Status question "Has a doctor ever told you that you had Alzheimer's?" [41]. Using a combination of claims and survey data to ascertain ADRD has been recommended by MCBS investigators to increase capture of ADRD and has been shown to yield results similar to those of expert in-person-assessment [42].

Key Independent Variable: Non-Cancer Chronic Pain Condition (NCPC) and Number of NCPCs
Baseline NCPC were identified using Medicare fee-for-service claims. NCPC included five common NCPC (back or neck pain, headache, joint pain, neuropathic pain, and osteoarthritis). The presence of any NCPC was identified using either two outpatient claims (90 days apart) or one inpatient claim using ICD9-CM codes as recommended by the Centers for Medicare and Medicaid Services [40] and consistent with prior studies of NCPC [43,44]. Any NCPC was assessed as a binary variable (yes/no) during baseline. Relative NCPC burden was ascertained with a count variable (0-5 NCPCs).
Also included as covariates were other chronic physical health conditions reported at baseline (yes/no). These included hypertension, diabetes, heart disease, respiratory illness, and history of stroke, traumatic brain injury (TBI), and cancer (with the exception of non-melanotic skin cancer), as well as specific auto-immune conditions associated with chronic pain, including rheumatoid arthritis (RA), and systemic lupus erythematosus (SLE). In addition, certain commonly used medications that have been linked to ADRD risk and/or used in chronic pain management were also evaluated as covariates. These included non-steroidal anti-inflammatory drugs (NSAIDS) [46], opioid analgesics [47], benzodiazepines [48] and certain other psychotropic drugs (e.g., selective serotonin reuptake inhibitors (SSRIs), serotonin norepinephrine reuptake inhibitors (SNRIs), monoamino oxidase inhibitors (MAOIs), tricyclic antidepressants (TCAs), and atypical antidepressants) [49]. Benzodiazepine use was evaluated as a separate medication category given that: (1) a significant number of study participants (10%) were prescribed these medications; and (2) prior research has suggested that benzodiazepines not only has multidimensional neurological effects [50,51], but is also associated with elevated risk for ADRD [48].
Mediators: To examine the potential mediating influence of mood and sleep disorders on the relation of NCPC to incident ADRD, we evaluated the effects of diagnosed depression, anxiety and insomnia-related sleep disorders both separately and in combination (see supplementary Table S1 for ICD-9-CM diagnostic codes).

Statistical Analysis
Potential differences in baseline characteristics by NCPC and ADRD status were determined using Rao-Scott chi-square tests (categorical variables). Logistic regressions were used to assess the unadjusted and independent association of NCPC and NCPC burden to incident ADRD. In these multivariable logistic regressions, we carried out block-wise adjustment for demographic and socioeconomic factors; lifestyle factors and chronic physical health conditions; and analgesics and psychotropic medications to assess the adjusted associations of NCPC to ADRD risk. Similar nested multivariable logistic regression models were used to evaluate potential mediating effects of baseline diagnoses of depression, anxiety, and sleep disorders both individually and in combination. To assess the linear relationship of NCPC to risk of ADRD, regression models with polynomial contrast were used. For our sensitivity analyses, we used multinomial logistic regression to build competing risk of death models with the following four outcomes: ADRD-free and alive at the end of follow-up (N = 15,785) (referent category); ADRD-free and not alive at the end of follow-up (N = 1220); ADRD positive and Alive at the end of follow-up (N = 1149); ADRD positive and not alive at the end of follow-up (N = 259).
All bivariate and multivariable analyses were carried out using MCBS complex survey design sampling weights. MCBS data-cycles are released with cross-sectional as well as longitudinal weights. We used 3-year backward longitudinal weights for analysis of our pooled cohorts to yield up to two years of continuous follow-up. All analyses were performed using SAS survey procedure (SAS version 9.4, SAS Institute, Inc.).

Results
Our analytical sample comprised 16,934 eligible participants with a mean age at baseline of 74 ± 0.07 years. The majority of the cohort were female (57%), white (83%), married (56.5%) and reported a family income at or above 200% of the FPL. Study participants were predominantly from metropolitan areas (71%) and most had private insurance (79%).
Overall baseline prevalence of any NCPC in this study was 36% (weighted), with significant differences by cohort (range 31% to 39%, p < 0.0001). Pain burden in this population was high, with 37.5% reporting at least two co-existing NCPCs and 15% indicating three or more NCPCs. A significantly higher percentage of participants with baseline NCPC reported a history of chronic physical health conditions associated with ADRD, including hypertension (82% vs. 68.7%), heart disease (31.5% vs. 19.9%), stroke (14% vs. 10%), diabetes (23.5% vs. 19%) (p's < 0.0001). Participants with baseline NCPC were also significantly more likely to report mood disorders, including depression (11% vs. 4.5%) and anxiety (7.9% vs. 3.5%), as well as sleep disorders (25.5% vs. 10.2%) when compared to participants without baseline NCPC (Table 1).   Table 2 summarizes the baseline characteristics of the study sample by incident ADRD status. A total of 1149 participants were diagnosed with incident ADRD during the 2-year follow-up period (overall incidence rate = 6.8 per 100 participants). Incident ADRD rates did not differ by study cohorts (range 4.9 to 6.2 per 100, p = 0.9), and were significantly higher in women (6.2% vs. 5.0% in men), and in participants who were black (7.4% vs. 5.5% in Non-Hispanic whites), widowed (8.2% vs. 4.5% in married), and more poorly educated (p's < 0.0001). The proportion of participants diagnosed with incident ADRD was also significantly higher in those who were <200% of the federal poverty level (7.4% vs. 4.3%), were on Medicaid (9.1% vs. 5.2%) and who lacked private insurance (7.2% vs. 5.3%) (p's < 0.0001). In addition, the percentage diagnosed with ADRD was significantly greater in those who indicated a history of specific physical health conditions at baseline, including hypertension, heart disease, stroke, and TBI, as well as in those who reported use of opioid analgesics (6.9% vs. 5.4%), benzodiazepines (8.5% vs. 5.4%), or other psychotropics (9.8% vs. 4.8%) (p's < 0.0001). The proportion of participants with incident ADRD increased significantly with rising number of comorbid physical health conditions (range 4.6% to 10.6%, p < 0.0001). Participants who reported sleep, depression or anxiety disorders at baseline also included a significantly higher proportion of ADRD cases (p's < 0.0001). Notably, as detailed in Table 2, incident ADRD rates were significantly higher in those who reported any NCPC at baseline (6.6 vs. 4.3 per 100), and rose with increasing number of NCPC conditions, from 6.2 in those with 1 NCPC condition to 17.3 per 100 participants in those with 4 or more NCPC conditions (p's < 0.0001).
The magnitude of this association was modestly reduced after additional adjustment for lifestyle factors and comorbid physical health conditions (AOR = 1.28, CI 1.10-1.48, p = 0.001), and further attenuated by inclusion of medications in the model (AOR adjusted for NSAIDs and opioid analgesics (AOR = 1.25, CI 1.08-1.45, p = 0.003) ( Table 3). Further adjustment for benzodiazepines and other psychotropics only slightly reduced the magnitude of this estimate (AOR = 1.21 (1.04-1.40, p = 0.01)). As detailed in Table 3, the association of NCPC to incident ADRD rose significantly in both strength and magnitude with increasing number of pain conditions. Relative to beneficiaries with no NCPC at baseline, those with ≥4 NCPCs were twice as likely to be subsequently diagnosed with ADRD after adjustment for demographics, socioeconomic status, lifestyle factors, and medical conditions (AOR = 1.98, CI = 1.36-2.89, p-trend <0.00001). Further adjustment for analgesic medications only slightly attenuated these associations (AOR for ≥4 vs. no NCPC = 1.91, CI 1.31-2.80, p-trend = 0.0008). When number of NCPCs at baseline were assessed as a continuous variable, risk for incident ADRD increased 12% for each additional NCPC in the fully adjusted model (AOR = 1.12, CI = 1.06-1.20, p = 0.0007).

Sensitivity Analyses
Analyses using competing risk of death models yielded findings consistent with those from our primary analyses (Table 4). Relative to participants without NCPC at baseline, those with any NCPC and still alive at follow-up were significantly more likely to be diagnosed with incident ADRD after adjustment for demographics, lifestyle characteristics, physical health conditions, analgesics, and other factors (AOR for any NCPC = 1.26, 95% CI = 1.08-1.46, p < 0.003). In contrast, baseline NCPC status was unrelated to death during the follow-up period, either with or without a diagnosis of incident ADRD (AORs, respectively = 0.99 and 0.97, p's ≥ 0.6). Further adjustment for psychotropics did not appreciably change these estimates. These findings suggest survival bias is unlikely to explain the positive association observed between baseline NCPC and incident ADRD.

Sensitivity Analyses
Analyses using competing risk of death models yielded findings consistent with those from our primary analyses (Table 4). Relative to participants without NCPC at baseline, those with any NCPC and still alive at follow-up were significantly more likely to be diagnosed with incident ADRD after adjustment for demographics, lifestyle characteristics, physical health conditions, analgesics, and other factors (AOR for any NCPC = 1.26, 95% CI = 1.08-1.46, p < 0.003). In contrast, baseline NCPC status was unrelated to death during the follow-up period, either with or without a diagnosis of incident ADRD (AORs, respectively = 0.99 and 0.97, p's ≥ 0.6). Further adjustment for psychotropics did not appreciably change these estimates. These findings suggest survival bias is unlikely to explain the positive association observed between baseline NCPC and incident ADRD.

Sensitivity Analyses
Analyses using competing risk of death models yielded findings consistent with those without NCPC at baseline, those with any NCPC and still alive at follow-up were signific adjustment for demographics, lifestyle characteristics, physical health conditions, analgesics 1.46, p < 0.003). In contrast, baseline NCPC status was unrelated to death during the follow-u (AORs, respectively = 0.99 and 0.97, p's ≥ 0.6). Further adjustment for psychotropics did no survival bias is unlikely to explain the positive association observed between baseline NCPC Table 4. Association of baseline non-cancer chronic pain conditions to incident Alzheimer's di competing risk analysis using linked Medicare Current Beneficiary Survey (MCBS) and Medicare c (AOR) with 95% confidence intervals (CI) calculated from multinomial logistic regression).
Analgesics: nonsteroidal anti-inflammatory drugs and opioid analgesics. ++ Regression results from polynomial contrast for linear relation indicate a strong linear effect of NCPC on risk of ADRD. Table 4. Association of baseline non-cancer chronic pain conditions to incident Alzheimer's disease and related dementias in older Medicare beneficiaries *: competing risk analysis using linked Medicare Current Beneficiary Survey (MCBS) and Medicare claims data, 2001-2013 (odds ratios (OR) and adjusted odds ratios (AOR) with 95% confidence intervals (CI) calculated from multinomial logistic regression).

Discussion
In this retrospective cohort study of older community-dwelling Medicare fee-for-service enrollees, diagnosed NCPC at baseline remained significantly and positively associated with risk for incident ADRD after adjustment for demographics, socioeconomic factors, medical history, medications, and other factors. The strength and magnitude of this association rose with increasing number of NCPCs, indicating increasing risk for incident ADRD with rising chronic pain condition burden. The relation between NCPC and risk of ADRD appeared to be mediated in part by the presence of sleep and mood disorders. To our knowledge, this investigation is the first large retrospective cohort study to assess the collective and incremental association of NCPC to incident ADRD, and to explore the potential mediating role of mood and sleep disorders in this association.
The significant positive associations NCPC to ADRD risk observed in this study are consistent with those of prior longitudinal studies regarding the relation of specific chronic pain conditions to risk for dementia. Recent large retrospective matched cohort studies in Taiwanese nationals [26,28,29,34,52], prospective cohort investigations in Norwegian adults [25,27] and a small retrospective cohort study of Canadian elders [53] all indicated significantly increased risk for incident all-cause dementia in those diagnosed with headache [25,27,29,34,52,53], OA/knee pain [26,33], and fibromyalgia [28] after adjusting for demographics, comorbid conditions, and other potential confounders. Although the few longitudinal studies investigating the association of non-specific chronic pain to subsequent cognitive deterioration or incident ADRD, all in British [31] and U.S. adults [24,30,32], have varied widely in study population, design, and methodology, results have likewise indicated that severe chronic pain [30,31], persistent pain [32], and/or reported pain interference [24] may predict subsequent worsening in memory [30,31], accelerated cognitive decline [32], and dementia [24,32].
While some longitudinal studies investigating the association between pain and dementia risk have included depression and/or anxiety as covariates in their adjusted models [24,[28][29][30]32,34], no studies to our knowledge have explored the potential mediating role of mood disorders or evaluated the influence of sleep impairment on the association of chronic pain to incident ADRD. Sleep and mood disorders are common in older adults, have been strongly and reciprocally associated with chronic pain [54,55], and have been shown to be independent risk factors for ADRD [13][14][15]56,57], suggesting that sleep and mood impairment may mediate the observed associations between pain and ADRD risk. The role of sleep and mood neuromodulators, such as serotonin [58,59], dopamine and histamine [60,61] has been well documented in pain expression, likely contributing to the documented bidirectional relationships of sleep [54] and mood disorders [55,62] to pain conditions. In the current study, inclusion of depression, anxiety, and insomnia-related sleep disorders in the model weakened but did not eliminate the observed associations between NCPC and incident ADRD risk, suggesting a partial mediating role. In agreement with prior research, sleep [15,54] and mood disorders [13,14,57] remained strongly and positively associated with both baseline NCPC and with risk for incident ADRD in this cohort after adjustment for other factors.
NCPC and increased ADRD risk; possible pathways: although the mechanisms underlying the observed association of NCPC to incident ADRD remain speculative, chronic pain may operate via several pathways to increase risk for dementia [21,23,[63][64][65]. Adults experiencing chronic pain have demonstrated diminished attention, impaired learning and memory, altered processing speed, reduced psychomotor efficiency, and compromised executive function [17][18][19]22,23], hallmarks of cognitive decline that may ultimately presage the development of cognitive impairment and dementia. Contributing to the documented decline in cognitive function that accompanies chronic pain, NCPC may promote specific adverse neurostructural and neurofunctional changes. For example, experimental and neuroimaging studies have demonstrated neurodegenerative changes in subjects with chronic pain that parallel those observed in ADRD [21,65,66], including reduction in grey matter volume in the amygdala, hippocampus and frontal cortices, the brain regions integrally involved in cognitive and behavioral functioning [66,67]. The increases in both peripheral and systemic inflammation that have been linked to chronic pain [68,69] may contribute to these neurodegenerative changes by contributing to neuroinflammation [63]. For example, chronic pain-induced microglial neuroinflammation has been directly implicated in Alzheimer's disease pathogenesis via production of amyloid beta plaques and neurofibrillary tangles [21]; persistent inflammation negatively affects neuroplasticity and synaptic performance via reduction in brain-derived neurotrophic factors [66,68,69]. Neuronal receptors in the brain are neither infinite nor mutually exclusive and serve a range of neurologic functions under limited resources. During persistent pain, nerve endings fire rapid pain impulses to the brain for remedial actions, which in turn, exhausts the neuronal resources that are also involved in cognitive functions [69][70][71]. In addition, the presence of chronic pain conditions has been correlated with dysregulation of noradrenergic-modulated endogenous pain autoinhibition [69], which has, in turn, been linked to negative cognitive outcomes, including decline in working and long-term memory [70,71].
In agreement with previously published research [3,7,45], risk for ADRD increased strongly with age in this study, and was elevated among African Americans and in adults who had less education or lower family income, were on Medicaid, or had a history of diabetes, stroke, or TBI. Similarly, the significant positive relationships between baseline sleep and mood disorders and likelihood of incident ADRD observed in this sample of US Medicare beneficiaries are consistent with the findings of numerous prior investigations [14,15]. Both psychotropics and analgesics are frequently employed to manage chronic mood disorders, insomnia, and pain conditions in the older population. NSAIDs have been reported to reduce risk of ADRD in most but not all studies [46], whereas prior research has demonstrated modest or no association of opioid analgesics to ADRD risk [47]. In this study, use of neither NSAIDs nor opioid analgesics was significantly related to risk for ADRD. In agreement with some [49] but not all previous studies [48,72], use of psychotropics, but not benzodiazepines, remained significantly associated with incident ADRD after adjustment for multiple confounders. However, as noted above, adjustment for these medications only slightly attenuated the observed relationships of NCPCs to incident ADRD.

Strengths and Limitations
This study has several strengths, including the population-based design, the use of longitudinal data from multiple cohorts, and the large, nationally representative sample of U.S. community-dwelling elders enrolled in FFS Medicare plans. Comprehensive information was available on a broad range of demographic and lifestyle characteristics, as well as on medical history, medication use, and other factors, allowing us to assess the potential confounding influence of these factors. Furthermore, NCPC's, medication use, and history of other health conditions were ascertained using claims data and established algorithms. ADRD was identified using a combination of Medicare claims and survey data, likely leading to greater capture of this often under-reported and underdiagnosed outcome [41,73,74]. Notably, previous studies have indicated high specificity (89-95%) and acceptable sensitivity (64-85%) for the ascertainment of ADRD using multiple years of Medicare claims data [41,73,74]. We used a 3-year backward cohort design, with a two-year continuous follow-up and incident ADRD measured at two time points.
Our study also has a number of limitations, including the relatively short follow-up period and lack of information on NCPC duration or on chronic pain symptoms, precluding a more comprehensive assessment of the potential role of chronic pain and chronic pain conditions in the development of ADRD. Given that ADRD is often underdiagnosed and progression is generally slow and insidious, and that the study follow-up period was short, under-ascertainment of ADRD is likely in this study, potentially biasing our risk estimates toward the null. In addition, given that we used a conservative method [75] for ascertaining chronic pain (i.e., ≥1 inpatient visit or two outpatient visits for any chronic pain conditions 90 days apart), and that NCPC is typically under-reported in medical claims data [76][77][78], NCPC may have been under-ascertained in this study, again potentially biasing risk estimates towards the null. Moreover, as sleep and mood disorders often accompany the development of ADRD, these disorders may have reflected prodromal ADRD in some who were diagnosed with incident ADRD. While we were able to adjust for a wide range of potential confounders, including smoking and BMI, we lacked information on certain lifestyle-related and other factors strongly linked to ADRD risk, including alcohol consumption, physical activity, genetic and familial predisposition, and social isolation [7,9]. Due to small cell sizes, we were also unable to adjust for Parkinson's disease and related movement disorders, conditions which have been linked to chronic pain, as well as to mood and sleep disorders and cognitive impairment [79,80]. However, small cell sizes would also suggest that any residual confounding from these movement disorders is unlikely to explain our findings. Both ADRD and chronic pain have been associated with increased mortality [81][82][83], introducing potential survival bias. However, competing risk of death analyses yielded findings similar to those of our primary analyses, arguing against a substantive effect of survival bias on our study results. Finally, definitive conclusions regarding causality are not possible due to the short follow-up period in this study and the insidious nature of ADRD development and progression. However, while reverse causality cannot be ruled out, a growing literature suggests chronic pain can disrupt neurocognitive function and may increase risk for cognitive decline and incident dementia [18,65,84], whereas evidence for an inverse relationship remains sparse [22].

Conclusions
In this large, population-based study in a nationally representative sample of US community-dwelling elders enrolled in FFS Medicare, NCPC at baseline remained significantly and positively associated with risk for incident ADRD after adjustment for demographics, lifestyle factors, medical history, medications, and other factors. This association increased in magnitude with increasing NCPC burden and appeared to be partially mediated by the presence of mood and sleep disorders. Additional large, prospective studies with longer term follow-up are warranted to confirm our findings; to further elucidate the potentially important contribution of chronic pain to accelerated cognitive decline, new onset cognitive impairment and the development of ADRD; to clarify the potential mediating role of sleep and mood disorders; and to explore possible underlying mechanisms.  Table S1: ICD-9_Clinial Modification (ICD-9-CM) Diagnostic Codes.
Author Contributions: Conceptualization, design, and methodology, and to the interpretation and presentation of results, S.K., U.S. and K.E.I.; statistical analyses, S.K. and U.S.; manuscript draft, S.K.; subsequent iterations, S.K., U.S. and K.E.I.; critical review of the final draft, U.S. and K.E.I. All authors have read and agreed to the published version of the manuscript.