Risk of Dementia in Older Veterans with Multiple Sclerosis

Abstract

Background:

While it is widely accepted that multiple sclerosis (MS) often causes cognitive dysfunction, it is thought that these cognitive symptoms rarely progress to dementia. However, this has not been thoroughly investigated. The objectives of this cohort study are to determine whether people with MS have an increased risk of dementia compared to the general population and to identify factors, such as geographic latitude, which may modify this association.

Methods:

We studied data from a random sample of US veterans aged ≥ 55 years followed at Veterans Affairs Health Care Systems nationwide from 1999–2019. We identified all patients diagnosed with MS using ICD codes over a two-year baseline period. We then identified a comparison cohort of patients without MS matched 1:1 on sex, age, race, and first encounter date. We constructed Cox proportional hazards regression models to determine the association between MS and dementia while controlling for demographic factors and comorbidities, with additional models to examine subgroup effects. We used Fine-Gray subdistribution hazard models accounting for competing risk of death to evaluate the sensitivity of the findings.

Results:

The study included 4084 MS patients and a matched group of 4084 non-MS patients. Overall, patients had mean age 66, were 93.6% male, and 88.1% non-Hispanic White, with mean follow-up time 9.5 years (MS) and 10.8 years (non-MS). In unadjusted models, veterans with MS had greater risk of dementia compared to matched controls (cumulative incidence 16.7% vs 12.4%; Cox HR 1.58, 95% CI 1.41–1.78). The increased risk remained after adjustment for potential confounders (adjusted HR 1.56, 95% CI 1.39–1.76) and when considering death as a competing risk (Fine-Gray HR 1.36, 95% CI 1.21–1.53). The magnitude of the MS-dementia association increased with rising geographic latitude (North HR 1.86, 1.51–2.30; Central HR 1.61, 1.42–1.82; South HR 1.39, 1.18–1.64; interaction p=0.04) and younger baseline age (interaction p<0.001).

Conclusions:

Among older veterans with MS, risk of dementia diagnosis was higher compared to matched controls even after controlling for comorbidities. The risk difference was highest in northern regions and in younger patients. Clinicians caring for older MS patients should be aware of this risk and offer screening and treatment accordingly.

1. Introduction

Multiple sclerosis (MS) is an inflammatory chronic neurological disorder characterized by central nervous system inflammation and demyelination.¹ Among other chronic neurological conditions, MS stands out for its comparatively young average age at diagnosis and the number of life-years affected.^2,3 While disability is common in MS, average life expectancy is only modestly reduced^4,5; thus, most patients live for many years, if not decades, with the disease.

Cognition is one of many neurological functions that is known to be affected in MS,⁶ but the clinical impact of decreased cognitive function in patients with MS, especially with older age, is under-explored.⁷ Data examining dementia as a long-term outcome are even more sparse. The paucity of literature on dementia in MS is possibly due to a longstanding perception that cognitive impairment in MS tends to be mild, rarely causes major disability, and affects distinct subcortical domains (attention, processing speed) in contrast to most neurodegenerative diseases.^6–9 Furthermore, it is challenging to clinically define dementia caused by MS in older patients, because competing pathologies such as Alzheimer’s Disease (AD) are common, and pre-mortem diagnosis has poor concordance with post-mortem pathology.¹⁰

Nevertheless, there are emerging suggestions that MS may confer a greater risk of dementia later in life.^11–13 Indeed, from a pathophysiology perspective, there has been a growing understanding of the long-term effects of MS on decreased gray matter volume and the possibility for MS pathology to mimic other dementia-causing neurodegenerative diseases, challenging the old paradigm of MS as a subcortical disease.^10,14–16

With this context, the objective of our study was to investigate the association between MS and diagnosis of dementia among Veterans who receive care in the Veterans Health Administration (VHA) health care system. In particular, we hypothesized that older patients with MS are at higher risk of dementia compared to matched peers. We also sought to identify potential factors including geographic latitude which might modify the association between MS and dementia risk.

2. Methods

The study population comprised a random sample of all patients aged 55 years or older who received care at any VHA facility from October 1, 1999 through September 30, 2019. This analytic cohort has been previously described.¹⁷ We identified all patients with a diagnosis of multiple sclerosis during the two-year baseline period (starting from the patient’s first encounter date) from the National Patient Care Databases (NPCD), an electronic database that captures information on all in- and outpatient encounters that occur at VHA health-care facilities nationwide. Patients with MS were defined as those with one or more encounters linked to MS ICD codes (340 or G35) during the baseline period; this inclusive case definition was chosen to identify all potential MS cases and reduce the risk of selection bias. Mortality data was obtained from the Vital Status File Database.

To create a comparison cohort, we matched each patient with MS with exactly one patient without MS of the same sex, age in years, race and ethnicity (self-reported and categorized as Asian, Black non-Hispanic, Hispanic, White non-Hispanic, or other/unclassified), and first encounter date. Matching was performed without replacement. Patients in the control group with prevalent diagnoses of demyelinating diseases, optic neuritis, encephalitis, myelitis, or encephalomyelitis during the baseline period were excluded. In addition, veterans with prevalent dementia during the baseline period, those without follow-up visits, and those with missing demographic data were excluded from both groups.

The outcome was incident dementia over follow-up, defined by a pre-specified list of inpatient and outpatient ICD-9 and ICD-10 codes recommended by the VHA Dementia Steering Committee.¹⁸ The complete list is provided in the supplementary material. The use of ICD codes to establish a dementia diagnosis in older adults has been validated against external standards with a positive predictive value estimated from 78.9 to 96.3%.^17,19

We also ascertained demographic information and medical, psychiatric, and substance-related conditions from the NPCD for covariate inclusion and subgroup analysis. Given the small number of patients in the cohort who identified with racial and ethnic minority groups, race and ethnicity were classified as a binary variable to avoid model overfitting and loss of statistical power. Comorbidities were defined by the presence of relevant, pre-specified ICD-9 and ICD-10 codes during the baseline period. Cardiac disease included any of atrial fibrillation, heart failure, or myocardial infarction. Geographic region, defined by the Veterans Integrated Service Networks (VISN), was obtained for each patient. VISNs were organized into three groups according to latitude (North, Central, South), with North corresponding approximately to 42 degrees latitude or higher, and South approximately to 37 degrees latitude or lower (see supplementary material for a map of the VISN groupings). When grouping VISNs which straddle multiple latitudinal regions, preference was given to areas with greater population.

2.1. Statistical Analysis

We compared baseline characteristics between those with and without MS using chi-square tests for categorical variables and the Mann-Whitney test for age. We next constructed unadjusted Kaplan-Meier failure curves according to group, using dementia as the outcome. We then used crude and adjusted Cox proportional hazards regression models to assess the association between MS and risk of dementia diagnosis, with time-on-study as the time scale, and right-censoring at the end of follow-up. The adjusted model included sex, age, race and ethnicity, geographic latitude, depression, tobacco use, substance abuse, stroke, epilepsy, Parkinson’s disease, obesity, diabetes, cardiac disease, and dyslipidemia. Age was modeled as a continuous variable and geographic latitude was modeled as an ordinal variable with equal distance between the three groups; the remainder were binary categorical variables. For subgroup analyses, we re-ran the adjusted Cox model while including interaction terms between the MS variable and pre-specified subgroups (selected based on subject matter knowledge). We tested interaction with baseline age as a continuous variable; in checking the assumption of linearity of the interaction, we used exploratory models with age as a categorical variable using various groupings. We did not consider these exploratory models as formal analysis given their post-hoc nature.

We performed multiple sensitivity analyses to confirm the robustness of our findings to potential sources of bias. First, we repeated our main analyses using a Fine-Gray subdistribution hazard model (adjusted for the same variables listed above). Fine-Gray regression treats death prior to dementia diagnosis as an alternate competing risk, providing a more conservative estimate of the association between MS and dementia than statistical approaches which treat patients who die as censored (which assume they would still be at risk if additional follow-up data had been available). Second, given that age is strongly associated with dementia risk, we repeated the Cox analysis using age (rather than time-on-study) as the time scale, given that this may better account for left-truncated data as well as age effects on dementia incidence.²⁰ Finally, given the potential for misclassification of MS cases when relying on administrative data, we repeated the analysis on the subgroup of MS patients (and their respective matched controls) with a minimum of two separate encounters linked to MS ICD codes, a more conservative case definition of MS.

P-values were two-sided with significance defined as P <.05. No adjustment was made for multiple comparisons. There were no missing data. SAS version 9.4 (Cary, NC) was used for all statistical analyses.

2.2. Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the local Institutional Review Boards at the University of California San Francisco, San Francisco Veterans Affairs Medical Center, and the US Army Medical Research and Material Command, Office of Research Protections, Human Research Protection Office. Written participant consent was not required because data were deidentified administrative data.

3. Results

Our final cohort included 4,084 patients with a diagnosis of MS and a comparison sample of 4,084 patients without MS, one-to-one matched on sex, age (within 0.5 years in either direction), race and ethnicity, and first encounter date. Baseline characteristics are shown in Table 1. Compared to the matched controls, patients with MS were more likely to be in the North region. Prevalent depression and other neurological disease were more common in the MS group, whereas other medical comorbidities, tobacco use, and substance use disorders were more common in the patients without MS.

Table 1.

Baseline characteristics of patients with multiple sclerosis (MS) compared to matched controls

	MS (n=4084)	Control (n=4084)	P-value
Sex, No. (%)			1.00
Female	263 (6.4)	263 (6.4)
Male	3821 (93.6)	3821 (93.6)
Age, mean (SD), years	65.9 (7.0)	66.0 (7.0)	0.61
Race and ethnicity			1.00
Asian	4 (0.1)	4 (0.1)
Black, non-Hispanic	292 (7.2)	292 (7.2)
Hispanic	24 (0.6)	24 (0.6)
White, non-Hispanic	3579 (88.1)	3597 (88.1)
Other or unclassified	167 (4.1)	167 (4.1)
Geographic region			<0.001
North	1377 (33.7)	1101 (27.0)
Central	1192 (29.2)	1078 (26.4)
South	1515 (37.1)	1905 (46.7)
History of stroke	402 (9.8)	284 (7.0)	<0.001
Current tobacco use	516 (12.6)	644 (15.8)	<0.001
Obesity	479 (11.7)	725 (17.8)	<0.001
Diabetes	783 (19.2)	1075 (26.3)	<0.001
Dyslipidemia	1951 (47.8)	2283 (55.9)	<0.001
Epilepsy	176 (4.3)	76 (1.9)	<0.001
Parkinson’s Disease	85 (2.1)	39 (1.0)	<0.001
Cardiac disease*	454 (11.1)	499 (12.2)	0.12
Substance use disorder	152 (3.7)	288 (7.1)	<0.001
Depression	989 (24.2)	621 (15.2)	<0.001

^*History of myocardial infarction, heart failure, or atrial fibrillation

Mean follow-up time was 9.5 years for the MS group and 10.8 years for the comparison group. Overall, 681 (16.7%) MS patients developed dementia during the follow-up period compared to 507 (12.4%) patients in the comparison group (p<0.001, chi-square). Figure 1 illustrates the Kaplan-Meier failure curves for unadjusted dementia incidence for the two groups; dementia-free survival was shorter in the MS group (p<0.001, log rank). In unadjusted Cox models, veterans with MS had greater risk of dementia compared to non-MS veterans (hazard ratio (HR) 1.58, 95% confidence interval (CI) 1.41–1.78). The increased risk remained after adjustment for potential confounders (HR 1.56, 95% CI 1.39–1.76). In the adjusted Fine-Gray subdistribution hazard model, the greater risk of dementia in MS was attenuated but persisted (Fine-Gray HR 1.36, 95% CI 1.21–1.53).

Figure 1:

Unadjusted Kaplan-Meier failure curves for outcome of incident dementia by study group

In subgroup analyses, shown in Table 2, the MS-dementia association was stronger with increasing geographic latitude (North HR 1.86, 95% CI 1.51–2.30; Central HR 1.61, 95% CI 1.42–1.82; South HR 1.39, 95% CI 1.18–1.64; interaction p=0.04). Substance use disorder (SUD) at baseline was an effect modifier: the MS-dementia association was not detected in patients with SUD (SUD+ HR 0.88, 95% CI 0.50–1.54, SUD- HR 1.61, 95% CI 1.42–1.82; interaction p=0.04). There was no effect modification seen according to diagnosis of depression, sex, race and ethnicity, or tobacco use. Finally, increasing baseline age (modeled as a continuous variable) was associated with a decreasing hazard ratio (interaction p<0.001); however, exploratory modeling suggested a non-linear effect with a more pronounced drop-off at age 65.

Table 2.

Effect modification of dementia hazard in MS relative to controls by specified categorical covariates

	Adjusted HR (95% CI)	P-value (chi-square)
Depression		0.76
Present	1.51 (1.18–1.93)
Absent	1.58 (1.38–1.81)
Geography		0.04
North	1.86 (1.51–2.30)
Central	1.61 (1.42–1.82)
South	1.39 (1.18–1.64)
Sex		0.61
Female	1.77 (1.08–2.92)
Male	1.55 (1.37–1.75)
Race/ethnicity		0.35
White, non-Hispanic	1.59 (1.41–1.81)
Asian, Black, Hispanic, other, or unclassified	1.35 (0.96–1.89)
Tobacco use		0.36
Present	1.34 (0.94–1.90)
Absent	1.59 (1.41–1.81)
Substance use disorder		0.04
Present	0.88 (0.50–1.54)
Absent	1.61 (1.42–1.82)

As a sensitivity analysis, the Cox model was re-run using age, rather than time on study, as the time scale. The results were very similar to the primary model, suggesting increased risk of dementia in the MS group (HR 1.55, 95% CI 1.38–1.74). When restricting MS cases to those with a minimum of two or more MS-coded encounters (comprising approximately 85% of the MS cases in the full cohort), the main effect again persisted, with hazard ratios for dementia somewhat higher than in the main cohort (adjusted HR 1.73, 95% CI 1.52–1.98).

4. Discussion

In this study, we examined a large random sample of all older patients in the Veterans Affairs Health Care System nationwide, and we found that patients with a diagnosis of MS have a greater than 50% increased risk of being diagnosed with dementia compared to a matched cohort of patients without MS. The association in our study was robust to adjustment for numerous potential confounders as well as treatment of death as a competing risk. Our findings among an older MS population followed for approximately ten years suggest that MS confers a substantial risk of dementia in older patients who have been living with the disease for decades. This has important potential implications for clinicians treating older patients with MS, suggesting a heightened need for early screening for dementia.

While prior studies have established that cognitive deficits on neuropsychological testing are common in cross-sectional cohorts of MS patients, longitudinal studies finding increased risk of dementia in MS populations have previously been limited to younger and healthier populations.^11–13 Recent administrative claim studies in the United States and Korea both studied populations predominately under age 65, followed patients for a mean of only 4–4.5 years, and did not account for the competing risk of death. Unsurprisingly, both studies reported significantly lower baseline incidences of dementia (3–3.5% cumulative incidences) than we observed in our cohort (12.4%). Our study has now replicated the finding of increased dementia risk in a significantly older cohort followed for a longer period, during which competing causes of dementia are much more prevalent. Consistent with these prior studies, we observed that the risk increase was greatest at younger baseline ages, which may relate to the typically early age of onset of MS and the lower incidence of competing causes of dementia in these younger age groups. Taken together, our study expands the body of evidence suggesting a larger relative increase in dementia risk in middle-aged patients with MS that persists even into older ages.

Our study does not distinguish whether the increased dementia risk that we found is attributable to greater occurrence of Alzheimer’s Disease (AD), vascular dementia, or other dementias among MS patients, or whether it is secondary to MS pathology itself. We controlled for all major traditional vascular risk factors as well as for clinical history of stroke in our study, suggesting that these factors cannot fully explain the dementia risk in MS. AD and MS pathologies can certainly co-exist in individual patients,^10,21 but it is not known whether AD pathology occurs at higher rates in MS patients than would be expected in the general population. Overall, it remains unknown whether MS may confer a risk of accelerated AD and/or vascular pathology in susceptible patients, or whether the disease processes are distinct but leading to overlapping clinical syndromes. Further research will likely need to use biomarkers and/or neuropathological data given the overlapping clinical features of these entities and relatively poor clinicopathological correlation reported in prior case series.^10,14,21

We also detected geographic variability of the strength of the association between MS and dementia in our cohort, with the risk increase being greater in veterans treated in Northern regions of the United States. It has long been recognized that the incidence of MS varies according to latitude, with higher incidences in regions which are farther from the equator; this finding also been demonstrated within the United States.^3,22 It is less clear whether there are differences in the phenotype (relapsing-remitting versus primary progressive), natural course, and prognosis by latitude; however, some studies do support the hypothesis that MS severity is also greater at higher latitudes. For example, studies have found that there are greater oligoclonal bands detected in the cerebrospinal fluid of patients in Northern regions, and that this may correlate with worse prognosis.²³ A large, multicenter, international study also recently found a gradient of increased severity of disability as latitude increases beyond 40 degrees.²⁴ Therefore, it is possible that the geographic trend in our study is indicative of greater severity of MS leading to a higher risk of dementia later in life. However, other explanations such as unmeasured regional variations in access to care, genetic factors, or other environmental factors cannot be excluded.

Surprisingly, we observed that the increased dementia risk was mitigated among MS patients with a diagnosis of substance use disorder. One prior study has found that substance use disorders in MS patients are more prevalent in patients with lower rates of MS-associated disability.²⁵ This finding would therefore be consistent with the hypothesis that the dementia risk in MS is associated with degree of MS severity and/or disability. However, further studies which can directly capture MS-associated disability as a variable would be needed to confirm this hypothesis.

One key strength of this study is its use of longitudinal administrative data within the VHA. Because the VHA data systems are nationalized, this dataset is uniquely able to capture longitudinal diagnoses over a longer follow-up period with minimal patient dropout. The VHA also offers relatively equitable access to care across its eligible population, increasing representativeness of our cohort and possibly mitigating surveillance bias. Another key strength is the robust cohort size which allowed for precise matching to control for key variables. We were also able to control for an extensive range of comorbidities to reduce the chance of confounding, and mortality data allowed us to consider death as a competing risk to mitigate bias.

There are some important limitations to consider when interpreting these results. First, the accuracy of diagnoses ascertained from ICD codes relies on the patient having had a clinical encounter relevant to the diagnosis, and that diagnosis being correctly coded. Regarding potential misclassification of MS cases, our sensitivity analysis restricting the MS group to those with two or more encounters linked to MS codes suggests the robustness of our main finding, finding an even higher hazard ratio with the more conservative case definition. It is possible that the inclusive case definition of the main cohort may have attenuated the size of the effect by capturing “rule-out” or diagnostically uncertain MS cases in the cohort. Alternatively, the more conservative case definition in the sensitivity analysis could have introduced selection bias by systematically excluding patients with milder or clinically stable MS who do not receive frequent follow-up for MS.

Regarding ascertainment of dementia outcomes, patients with MS are more likely to be followed regularly by a neurologist, and it is possible that this increases the likelihood that a dementia diagnosis will be recognized and established (surveillance bias). However, our matched comparison cohort consisted of older patients and the majority had one or more chronic medical comorbidities, thereby increasing the likelihood that they would be followed regularly by a medical provider, potentially mitigating any difference in surveillance.

In terms of generalizability, given that the majority of MS patients are diagnosed between age 20 and 40, most of our MS patient population would have been diagnosed many years before the beginning of the study period. Because the treatment paradigms for MS have changed considerably in recent decades, the experience with newly diagnosed individuals in the modern era may vary from our population. Relatedly, we also did not have data regarding treatments for MS to better understand whether the dementia risk may be modifiable based on MS disease-modifying treatment. Finally, we note that the study population of veterans is predominately male, which may limit the generalizability of the findings. Our subgroup analysis showed that the effect was robust in patients of female sex, albeit with wider confidence intervals given the smaller number of patients.

5. Conclusion

In this nationwide longitudinal cohort of older veterans diagnosed with multiple sclerosis, the risk of incident dementia was significantly increased compared to matched control patients. The risk difference was strongest in higher-latitude regions and at younger baseline ages. Overall, our data suggest that MS is an independent risk factor for dementia later in life. Clinicians caring for older MS patients should be aware of this risk and offer screening and treatment accordingly.