Predictors of Mortality in a Large Cohort of Elders with Essential Tremor

Essential tremor (ET) is a progressive neurological disease [1‒3] that in its advanced stage [4] can result in significant functional impairment [5] and disability [6‒8]. Along with action tremor, patients may experience additional motor and non-motor features [9], including gait abnormalities [10, 11] with attendant lower balance confidence [12], more near falls [13], and lower physical activity [14]. Studies document greater cognitive abnormalities in ET cases than controls [15], a rate of decline in cognitive functioning greater than that observed in controls [16], and increased prevalence [17, 18] and incidence of dementia [17, 18]. Hence, ET is associated with significant morbidity. Conceivably, this increased morbidity is associated with increased mortality, and indeed, prospective data show a 45% increased risk of mortality in ET [19, 20]. Defining contributors to this increased mortality is valuable. Information from these analyses can identify patients at risk for early death and target them for preventive measures [21]. Mortality analyses can also be used to make projections about life expectancies [21]. Two studies [22, 23] identified contributors to increased mortality in ET in samples of 141 and 194, respectively. The current study expands that database to a much larger sample of 347. Given prior results demonstrating a robust association between dementia and mortality risk [22, 23], we had a particular a priori interest in this association [22, 23].

Our initial sample included 379 cases enrolled in an ongoing, prospective, longitudinal study of cognition in elders with ET (Clinical Pathological Study of Cognitive Impairment in Essential Tremor [COGNET]; National Institutes of Health Award #R01 NS086736). Nationwide enrollment began in 2014; to date, cases reside in 39 states. Eligibility requirements were (1) a diagnosis of ET; (2) a minimum baseline age of 55 (although the majority, 95.4%, were aged 65 or older at baseline); and (3) enrollment as an eventual donor in the Essential Tremor Centralized Brain Repository. Our goal was to study older ET cases (65 years and older), although we included a smaller proportion of younger cases (as young as 55 years) to sample and represent such cases. The Yale University, Columbia University, and University of Texas Southwestern Medical Center Institutional Review Boards approved the study protocol. Cases provided written informed consent.

A trained research assistant administered evaluations during three home visits scheduled approximately 18 months apart. Each included demographic and clinical questionnaires, neuropsychological tests, and a videotaped neurological examination.

From our initial sample of 379, we eliminated 32 participants for the following reasons: (1) tremor not severe enough to assign an ET diagnosis or participant had additional movement disorder (e.g., dystonia, Parkinson’s disease; n = 19), (2) incomplete data (n = 10); (3) loss of contact resulting in no information regarding whether termination of participation was due to death or withdrawal (n = 3). This yielded a final sample of 347.

Cases provided information including age, sex, race, education, and age of tremor onset. The difference between baseline and onset age yielded a measure of tremor duration. Cases reported whether they currently used cigarettes, their average weekly number of alcoholic drinks, number of falls during the past year, and number of current medications. The Geriatric Depression Scale (GDS) [24] measured depressive symptoms yielding scores ranging from 0 (none) to 30 (severe) [25]. The Pittsburgh Sleep Quality Index (PSQI) [26] yielded scores ranging from 0 to 21, with higher values indicating more disturbed sleep. The Physical Activity Scale for the Elderly (PASE) assessed physical activity (range = 0–400), with higher scores indicating greater activity [27].

A videotaped neurological examination was performed during evaluations, as detailed elsewhere [28]. Severity of postural tremor (one examination item) and kinetic tremor (five items) was assessed on a 0 to 3 scale in each arm, yielding a Total Tremor Score (TTS). TTS values range from 0 (low severity) to 36 (high severity) [29]. The number of missteps out of ten (range = 0–10) occurring during a tandem gait task measured gait impairment [29]. After reviewing questionnaire and videotaped examination data, an experienced movement disorders neurologist (E.D.L.) assigned clinical diagnoses of ET based on reliable and valid diagnostic criteria from the Washington Heights-Inwood Genetic Study of Essential Tremor (WHIGET) [30].

Cases completed an extensive battery of neuropsychological tests that assessed cognitive abilities across five domains: (1) memory, (2) executive function, (3) attention, (4) language, and (5) visuospatial ability, and globally (mean of the cognitive domains). These required little or no reliance on motor functioning, minimizing any disadvantage to cases with significant tremor. Details of the specific cognitive tests are provided elsewhere [29]. For diagnostic purposes, all neuropsychological test scores were demographically adjusted based on clinically available normative data. Clinical Dementia Rating (CDR) scores [31] were also calculated. CDR scores range from 0 to 3, with 0 corresponding to no cognitive impairment, 0.5 signifying questionable impairment/dementia, and values of 1, 2, and 3 indicating mild, moderate, and severe dementia, respectively [22, 31].

A neuropsychologist and a geriatric psychiatrist took part in a comprehensive review and evaluation of each case, including neuropsychological test performance and CDR scores. They assigned diagnoses of normal cognition (NC), mild cognitive impairment (MCI), or dementia (D) following established protocols; specific details are available elsewhere [29].

Distributions of baseline variables appear in Table 1. We assessed associations between mortality and baseline characteristics via chi-square tests for categorical variables, and Student’s t or Mann-Whitney tests for continuous variables.

Cox proportional hazards regression equations identified predictors of relative risk of mortality. For cases who died during enrollment, the time-to-event variable was time elapsed between baseline evaluation and death. For cases who did not die during enrollment, time-to-event equaled time between baseline evaluation and their most recent observation. We coded mortality status as “1” for deceased cases, and “0” for living cases.

Figure 1 outlines our analytic approach in detail. A series of univariate models assessed associations between risk of mortality and individual explanatory variables (Table 2). We next calculated multivariate models that included age and CDR scores as independent variables (Table 3). Age served as a control variable in these models given its clear conceptual and statistical association with mortality risk. CDR scores were included because (1) we had an a priori interest in dementia as a predictor of mortality [22, 23]; (2) these scores were successful predictors of mortality in previous studies of smaller samples; and (3) these scores provide a more nuanced and fine-grained measure of impairment than do the broader categories of MCI and dementia.

As described, CDR values include 0 (no impairment), 0.5 (questionable impairment/dementia), and 1, 2, and 3 (mild, moderate, and severe dementia, respectively). No cases received a CDR score of 3.0. Therefore, we individually considered the presence of questionable impairment/dementia, mild dementia, moderate dementia, and of dementia, collapsing across severity (i.e., a score of 1.0 or 2.0) as predictors in analyses. Cases receiving a score of 0 (no impairment) served as the reference group.

We further examined the potential contribution of each additional variable significantly associated with mortality in univariate models (Tables 4, 5). Given likely correlations among these variables, we individually added each to age and CDR scores in separate multivariate equations predicting mortality risk. Finally, sensitivity analyses examined whether subtypes of MCI predicted mortality.

Mean length of follow-up was 2.7 ± 0.7 years, median = 2.9, range = 1.3–4.2. Of the original 347 cases, 41 (11.8%) died before completing the three evaluations. Specifically, 27 died after the baseline evaluation and 14 after the second evaluation.

At baseline, deceased cases were older, reported less education, displayed more severe tremor, used more medications, and had more gait impairment, more falls, less physical activity, and more depressive symptoms than did living cases. Finally, deceased cases were less often diagnosed as cognitively normal and more often assigned higher CDR scores (i.e., exhibited greater cognitive impairment) than were living cases (Table 1).

Univariate Cox models evaluated associations of individual explanatory variables to mortality risk (Table 2). Analyses revealed that increased age, greater tremor severity, use of more medications, more impaired gait, more falls, and less physical activity were each significantly associated with mortality risk. In addition, the presence of questionable impairment/dementia, mild dementia, moderate dementia, or dementia collapsing across level of severity was each significantly associated with mortality risk.

Our first set of multivariate models included age and cognitive impairment as defined by CDR score as predictors of mortality risk (Table 3). An equation including both age and presence of questionable impairment as predictors revealed a significant association between mortality and age only. An equation including the predictors of age and the presence of mild dementia revealed that age bore a significant association to mortality risk, whereas mild dementia did not. When age and moderate dementia served as predictors, each bore significant and independent associations with mortality risk. Finally, age and the presence of dementia of any level of severity (i.e., mild or moderate) each were significantly associated with mortality risk.

Next, we conducted a series of multivariate models that examined whether additional variables significantly associated with mortality risk in the univariate models would continue to predict mortality when controlling for age and the presence of dementia, collapsing across severity (Table 4). Given likely correlations among these variables, we conducted individual equations in which we separately added each explanatory variable to age and dementia, collapsing across level of severity. Higher numbers of medications, greater gait impairment, and more falls each bore independent significant associations to mortality risk. Although higher age was also significantly associated with mortality in each equation, dementia was not. Neither tremor severity nor physical activity predicted mortality risk when added to age and dementia in parallel equations. However, age and dementia were each significantly associated with mortality in these models.

Thus, dementia, collapsing across severity, did not emerge as a consistent predictor of mortality risk. This is not surprising as the equations in Table 3 reveal a significant association with mortality that is independent of age for moderate but not mild dementia. Thus, grouping cases with mild and moderate dementia together may mask the relation between moderate dementia and mortality.

To address this, we conducted a parallel series of models examining whether variables significantly associated with mortality risk in the univariate models would be significantly associated with mortality when controlling for age and the presence of moderate dementia alone; i.e., those cases assigned a CDR of 2.0 (Table 5). Use of more medications, greater gait impairment, and more falls each bore significant independent associations to mortality, although both age and the presence of moderate dementia continued to predict risk in each of these equations. Neither tremor severity nor level of physical activity predicted mortality risk when added to age and moderate dementia in parallel equations. However, age and moderate dementia each continued to bear significant associations with mortality in these models.

In sum, dementia and age bear significant and independent positive associations to mortality risk in the current sample of elders with ET. Using more medications and having more falls and more impaired gait are each associated with higher mortality risk as well, even when controlling for the significant effects of age and dementia. The presence of moderate dementia is, in particular, a stronger and more consistent predictor of mortality risk than either the presence of mild dementia alone or the presence of dementia collapsing across severity level. Thus, a clear link between cognitive impairment and mortality risk only appears to emerge once impairment reaches the level of severity associated with a CDR score of 2 (i.e., moderate dementia).

Previous analyses of a subset of the current cases [23] revealed that the presence of the MCI subtype of amnestic MCI [32] predicted mortality in univariate analyses, although it failed to predict mortality risk in multivariate models. We conducted sensitivity analyses in which subtypes of MCI were considered as explanatory variables. Our results paralleled those reported previously: although amnestic MCI did predict mortality risk in a univariate model, it was not significantly associated with mortality in multivariate models that adjusted for age.

We examined predictors of mortality in a cohort of 347 elders with ET, 41 (11.8%) of whom died during the course of the study. Given recent attention to the presence of cognitive features in ET [15], we were primarily interested in the effects of cognitive impairment on mortality in this disease. As dementia in ET is associated with loss of both independent walking and independent living [33], and dementia appears to increase risk of mortality in ET in smaller samples [22, 23], we had a particular a priori interest in this association. Nearly five times more deceased than living individuals were demented (22.0% vs. 4.7%, Table 1), and univariate analyses demonstrated that a CDR score consistent with dementia increased the risk of mortality six-fold (hazard ratio [HR] = 6.33, Table 2). Additional analyses indicated that moderate rather than mild levels of dementia were driving this effect, with multivariate models demonstrating HRs in excess of 7.5 (Table 5). In our prior study [23], a CDR of 2 was associated with a 40.77-fold increase in mortality, although given the smaller sample size and fewer individuals with cognitive impairment in that study, the confidence intervals around this point estimate were very wide (12.46–133.45), making that an unstable estimate. Given the larger size of our current sample, including a greater number of individuals with cognitive impairment, confidence intervals are narrower, allowing for a more stable point estimate and more reliable finding.

These data add another dimension to the importance of identifying dementia in ET patients. Cognitive testing is not a standard component of ET assessments in clinic, and cognitive complaints, when reported, may be incorrectly attributed to advanced age or medication use. This likely leads to an under-recognition of dementia in ET patients. Our data have the potential to raise awareness of dementia in ET.

In the current study, a baseline CDR score of 1 (mild dementia) was associated with increased risk of mortality in a univariate model, with a HR of 3.81. In a model that adjusted for age, the HR was smaller – 1.55, and no longer significant. There are several probable explanations for the results of the multivariate model. First, the number of cases with a baseline CDR = 1 was only 20, with only 5 of these dying. This likely contributed to the wide confidence interval around the HR in the multivariate model. In other words, small sample size limited our ability to precisely quantify the association between CDR = 1 and mortality. Second, mild dementia likely has a more limited impact on daily functioning than moderate dementia. For example, memory impairment in moderate dementia could be present in the form of forgetting physical limitations, resulting in unsafe attempts to stand or ambulate without aids, leading to falls, injuries, and increased mortality risk.

In the current study, greater tremor severity was weakly associated with increased risk of mortality in a univariate model, with a HR = 1.06. However, the association was not retained in a model that adjusted for age and which accounted for the presence of dementia. This is not surprising. Although the severity of action tremor in the arms is functionally important to many ET patients, it would be challenging to imagine a biological mechanism whereby this difficulty would translate into an increased risk of mortality.

Prior studies indicate that the motor features of ET can include gait abnormalities [10, 11] with attendant lower balance confidence [12], more near falls [13], and lower physical activity [14]. Our results reveal that tandem gait impairment and falls were associated with increased mortality risk in ET, independent of level of cognitive impairment. An association between mortality risk and falls was observed in one of our two prior studies; hence, the current data confirm those findings [23]. In the non-ET population, gait and balance problems correlate with deteriorating health, and, ultimately, risk for death [34]. The value of these data in ET is that they add a dimension to the dialogue with ET patients about the presence of reported or observed gait and balance impairment. Recent research has highlighted benefits of emerging technologies for clinical intervention and monitoring of gait impairment. These include daily wearable sensors that can provide information guiding best practice and modifications of traditional rehabilitation techniques, and technological therapies such as transcranial stimulation and transcranial direct current stimulation that could potentially regulate cerebellar activity and improve gait function reducing risks of falls and mortality [35].

It is valuable to compare our data in ET with studies of predictors of mortality in other movement disorders such as Parkinson’s disease. A number of factors have been identified that are associated with increased risk of mortality in Parkinson’s disease, including presence of dementia, degree of motor impairment, and presence of psychotic symptoms [36]. More specifically, with respect to the presence of dementia, studies have demonstrated an approximate 2-fold higher risk of mortality [36‒39], serving to consolidate the association between dementia and poor outcomes.

A limitation of this study is the lack of a control group. Hence, a priori the study was not designed to assess whether the mortality risk observed in those with ET is higher than that in the general population, or whether different variables place individuals with versus without ET at higher risk. This issue has been assessed in prior research [19].

Nonetheless, defining the contributors to increased mortality in ET, the prime goal of these analyses, is a valuable exercise. Data generated from such analyses can be used to identify patients at risk for early death and target them for preventive measures [21]. Also, mortality analyses can also be used to make projections about life expectancies [21]. Second, our follow-up duration was modest; in future research, longer follow-up durations would be of value, as such analyses might reveal different patterns of association.

The study has several strengths including the inclusion of numerous clinical variables as possible predictors, a thorough diagnosis of ET, the exclusion of other diagnoses made by a senior movement disorder neurologist, and a comprehensive cognitive assessment with careful assignment of CDR scores as determined by a neuropsychologist and psychiatrist. Additionally, our large sample allowed our analyses to detect small associations [22].

This study quantifies the association between several independent predictors of mortality risk in individuals diagnosed with ET. Our findings highlight some of the downstream effects of dementia in patients with this disease, primarily their significantly increased risk of mortality. These data provide clinicians with the ability to identify ET patients at heightened risk of mortality and an opportunity to improve health interventions and strategies used with this cohort.

Research complies with internationally accepted standards for research practice and reporting. Research with human subjects was performed with approval of appropriate Ethics Committees and appropriate informed consent. The Yale University, Columbia University, and University of Texas Southwestern Medical Center Institutional Review Boards approved the study protocol, Approval No. STU2020-0564. Cases provided written informed consent.

Elan D. Louis was a member of the journal’s Editorial Board at the time of submission. No other authors have a relationship at the time of submission that could reasonably be perceived as a potential conflict of interest.

This work was supported by the National Institutes of Health Award #R01 NS086736. The NIH played no role in the design, collection, analysis, or interpretation of the data, or in the writing of this manuscript.

E.D.C.: data collection and organization, and writing of first draft; D.S.B.: conceptualization and design, data analysis, and writing of first draft; S.C.: conceptualization and design; review, and critique; E.D.L.: conceptualization and design, and writing of first draft.

The data that support the findings reported here are available from E.D.L.