The Longitudinal Course of Post-Stroke Apathy Over Five Years

Consecutive eligible patients hospitalized for ischemic stroke (determined by two neurologists), without a history of dementia as determined by retrospective score of <3.3 on the Informant Questionnaire of Cognitive Decline (IQCODE),¹⁹ and community-dwelling volunteers without a history of stroke or transient ischemic attack who met entry criteria were asked to participate. Criteria for defining an ischemic stroke, dementia (diagnosed by a consensus panel comprising a neuropsychiatrist, psychogeriatrician, neurologist, and neuropsychologist), and study inclusion and exclusion criteria have been published.^3,20 Only patients with ischemic stroke were included in this study. An ischemic stroke was defined as “rapidly developing clinical signs of focal (or global) disturbance of cerebral function, with symptoms lasting 24 hours or longer, with no apparent cause other than of vascular origin” in which a brain CT or MRI scan does not show intracranial hemorrhage.²¹ Other exclusion criteria for stroke patients and controls were inability to give informed consent, insufficient fluency in English to complete testing, history of dementia or other neurological disease known to affect cognition, alcohol or drug abuse, DSM–IV²² diagnosis of mental retardation, severe aphasia (<3 on the Aphasia Severity Rating Scale of the Boston Diagnostic Aphasia Examination),²³ current major psychiatric illness, or lack of an appropriate informant.

From consecutive admissions over a 38-month period between May 1997 and June 2000 to the inpatient stroke units at Prince of Wales and St. George Hospitals in Sydney, Australia, 251 patients met study criteria and gave consent for entry into the study, but 49 were excluded before baseline assessment (Figure 1). After recruitment in the first week after admission for a stroke (baseline), subjects were reassessed at 3–6 months (index) and then 1, 3, and 5 years later. Of the 202 patients assessed at baseline, 129 subjects were lost to follow-up (80 withdrew; 6 moved out of Sydney; and 43 died). Of the 202 patients, 135 completed the Apathy Evaluation Scale (AES)^24,25 at Index, 110 at 1 year, 88 at 3 years, and 70 at 5 years. The 152 patients who completed at least one AES rating comprise the patient sample for this article. Of these 152 patients, 124 (82%) had at least two AES ratings; 82 (54%) had at least three AES ratings; and 45 (30%) had all four AES ratings. Note that follow-ups referred to as 1, 3, and 5 years represent the time after index assessment, which meant approximately 1¼−1½, 3¼–3½, and 5¼–5½ years after baseline admission.

Control subjects were recruited through advertisement and local community groups. Of the 129 volunteers who met eligibility criteria, 23 declined to continue before the baseline assessment, leaving 106 subjects who were assessed at a combined baseline/index assessment, and at 1, 3, and 5 years after their baseline assessment. Between baseline and the 5-year assessment, 31 subjects were lost to follow-up (Figure 1). Of the 106 control subjects, 101 completed the AES at least once, 91 at index, 75 at 1 year, 81 at 3 years, and 75 at 5 years. The 101 controls who completed at least one AES rating comprise the control sample for this paper. Of the 101 control subjects 88 (87%) had at least two AES rating; 78 (77%) had at least three AES ratings; and 55 (55%) had all four AES ratings.

The study had institutional ethics committee approval, and subjects provided written informed consent upon enrollment after receiving a complete description of the study.

The baseline assessment for patients and controls included sociodemographic information, neurological and medical history, the European Stroke Scale (ESS),²⁶ and Activities of Daily Living (ADL)²⁷ and Instrumental Activities of Daily Living (IADL).²⁸

Data were analyzed with SAS, Version 9 (SAS Institute, Inc.; Cary, NC) and SPSS, Version 18 (SPSS, Inc.; Chicago, IL). Independent-sample t-tests were used for between-group comparisons on continuous variables. Between-group comparisons on categorical variables were analyzed with chi-square tests with Yates’ Continuity Correction for 2×2 tables and Fisher’s exact test when expected frequencies were lower than 5 in two or more cells. For all analyses, probability levels reported were two-tailed, and levels of significance were set at 0.05.

Restricted maximum-likelihood (REML)-based linear mixed models (PROC MIXED) were used to estimate change of apathy over time and identify significant baseline predictors of apathy, taking into account within-subject correlation (repeated-measures via random-effect specification). Missing data were imputed using multiple imputations (PROC MI). Full details of statistical analyses are provided in Supplementary Methods (online). Briefly, imputations were based on demographic and clinical variables and variables that were associated with the AES and predicted missing-ness (time-varying: dementia, ADL/IADL, GDS, AES).

Main results are given for imputed data and as additional information about the quality of the data imputation; mean AES scores for actual and imputed data are provided in Table 1.

The model for change in apathy over time included the Intercept and Time (measured as a continuous variable in approximate weeks after stroke for patients and weeks after index/baseline assessment for controls). Because of the small sample size, predictors were tested in two stages. The first stage (“univariate model”) included the Intercept, one Baseline/Index or Time-varying predictor, Time, and an interaction between predictor and Time (Table 2, Table 3). The second stage (“multivariate model,” Table 3) included significant predictors from the univariate analyses, Intercept, Time, and two-way interactions between predictors. Nonsignificant interaction terms were removed from models.

In order to test the trajectories reported by Mayo et al.⁵ and to identify distinctive patterns of individuals with similar trajectories of apathy over the 5 years after stroke, group-based trajectory modeling³² was used and implemented by SAS Proc Traj.³³ Homogeneous latent-trajectory classes were identified, based on this semi-parametric mixed-modeling strategy.³⁴ The censored normal (CNORM) trajectory model is specified to model the conditional distribution of the psychometric scale (Apathy score) data. Trajectory models were fitted on the basis of linear or quadratic trajectories for 3 to 5 groups. The Bayesian Information Criterion (BIC), Akaike Information Criterion (AIC) per subject (smaller in absolute value is better), and the distinctiveness of the trajectories were used to determine the optimal solution for apathy trajectory groups.

At baseline, control and stroke patient groups were similar in age and sex ratio, but controls had completed more years of education (Supplementary Table S1). Of 202 stroke patients recruited at baseline, there were no significant differences between the age and sex of the 152 included and 50 excluded patients. Those included had completed more years of education (Supplementary Table S2).

When missing-ness was tested as a binary outcome variable using logistic regression, more disabled patients were more likely to have missing data AES at 5 years. Of 152 patients in this sample, 70 had AES data at the 5-year assessment. Univariate logistic regression was used to analyze predictors of missing AES data at 5 years. Apathy at index was not a significant predictor (odds ratio [OR]: 1.54; 95% confidence interval [CI]: 0.70–3.37), but Apathy at 1 and 3 years predicted missing AES data at 5 years (OR: 4.53, 95% CI: 1.94–10.57; OR: 5.59, 95% CI: 2.11–14.83, respectively). Dementia at Index, 1, and 3 years (OR: 6.27, 95% CI: 2.04–19.28; OR: 17.68, 95% CI: 3.90 to −80.23; and OR: 4.57, 95% CI: 1.71–12.22, respectively) and lower ADL/IADL scores at Index, 1, and 3 years (OR: 0.72, 95% CI: 0.61 to −0.86; OR: 0.74, 95% CI: 0.64–0.85; and OR: 0.83, 95% CI: 0.75–0.93, respectively) also predicted missing AES data at 5 years. Other significant predictors of missing AES data at 5 years were higher GDS scores at Index, 1-year, and 3-year assessments (OR: 1.18, 95% CI: 1.03–1.36; OR: 1.18, 95% CI: 1.03–1.36; and OR: 1.25, 95% CI: 1.06–1.47, respectively).

Of the 101 control participants, 75 had AES data at the 5-year assessment. Univariate logistic regression was used to analyze predictors of missing AES data at 5 years. Contrary to the systematic attrition in patients, missing 5 year AES data in controls were not predicted by physical or cognitive disability. This indicates that AES levels over time and AES slopes are not comparable between the two groups.

At Index assessment, patients’ mean age was 72.1 (SD: 8.88) years; they had completed 10.2 (SD: 2.75) years of education; 57.9% (88/152) were men. Apathetic patients were older, had lower ADL/IADL scores, and were more likely to have dementia than those without apathy (Supplementary Table S3).

At Index, controls were age 71.1 (SD: 6.1) years and had completed 11.8 years of education; 48.5% (49/101) were men. Four controls (4.4%) were classified as apathetic (4/91). Apathetic control subjects were similar to those without apathy at Index regarding age, education, ADL/IADL scores, and gender ratio (t[89]=1.39, NS; t[89]=0.19, NS; t[88] = −0.49, NS; OR: 2.67, 95% CI: 0.26–26.72, respectively).

In the observed sample data, mean Apathy scores in the Patient sample increased from 32.2 (SD: 10.27) at Index assessment to 33.4 (SD: 10.23) at 1 year, 36.8 (SD: 14.04) at 3 years, and 37.1 (SD: 10.82) at 5 years. This occurred despite greater attrition of patients with more apathy. Similarly, rates of apathy steadily rose from 26.7% (36/135) at Index assessment, 33.6% (37/110) at 1 year, 34.1% (30/88) at 3 years, to 38.6% (27/70) at 5 years.

In the 45 Completers (patients who completed all AES ratings), mean AES scores were similar at Index (29.8; SD: 10.2) and 1 year (29.0; SD: 8.4), but increased to 32.6 (SD: 11.7) at 3 years, and 36.7 (SD: 11.5) at 5 years. However, these scores likely represent the least disabled group among the patients and are therefore not representative of the total sample.

Among the 101 control subjects, observed AES scores increased from a sample mean of 24.0 (SD: 6.08) at Index to 26.3 (SD: 6.90) at 1 year, 26.8 (SD: 7.92) at 3 years, and 29.2 (SD: 7.73) at 5 years. Rates of Apathy in control subjects rose from 4.4% (4/91) at index to 8.0% (6/75) at 1 year, 9.9% (8/81) at 3 years, and 16.0% (12/75) at 5 years.

When using linear mixed models in the observed (not imputed) data to compare AES levels in patients and controls, there were significant differences in AES scores (estimate: 7.47 [standard error {SE}: 1.10], p <0.0001), as well as change in AES over time (estimate: 0.01 [SE: 0.01], p=0.007), between the two groups. Patients had higher AES scores, on average, across the visits, and greater increase over time, as compared with the control group.

Linear mixed models based on the imputed data were used to describe the progression and predictors of apathy over time. AES scores in patients increased significantly over 5 years (parameter estimates with multiple imputation — intercept: 41.10 [SE:3.44], 95% CI: 34.12–48.08; df:36, p <0.0001; slope: 0.02 [SE: 0.004], 95% CI: 0.01–0.03, df: 62, t=4.35, p <0.0001). These results were consistent with parameters estimated without multiple imputation (intercept: 31.79 [SE: 0.83], t[151]=38.22, p <0.0001; slope: 0.03 [SE: 0.004], t[124]=7.80, p <0.0001).

When significant univariate time-varying (dementia, ADL/IADL, GDS) and index predictors (interval CVA) were combined into a multivariate model, dementia, ADL/IADL, and GDS, but not interval CVA, remained significant predictors.

Linear mixed-effect modeling was also applied to the control group. The result of the univariate and multivariate models are summarized in the data supplement (Tables S4, S5). Gender and ADL/IADL were significant predictors for apathy in the control group.

We tested three- to five-group models in linear and quadratic trajectories. The four-group model with linear trajectory was superior to three- and five-group models, based on consideration of both BIC (−1,458.04) and AIC (−1,439.89), as well as good representation of substantively distinct trajectories; hence, the optimal solution to data on Apathy score after stroke. The four-group model indicated three groups showing stable trajectories of Apathy over time, at distinct levels: Low Apathy (48%), Minor Apathy (29%), and High Apathy (12%); and one group showing a linear increase (11%), that is, a Worsening group.

Limitations are the high rate of attrition (significantly through dementia and death) and missing data, with the resulting modest number of subjects and data not missing at random. We tried to account for these limitations by using a large number of imputations (N=20–1,000) and several significant predictors of attrition. Complete case analysis would possibly result in even higher bias; but even in the 45 patients who completed all AES ratings and who probably represent the least disabled patients, AES levels and rates increased over 5 years. Attrition is a major hazard in apathy studies, where motivation is lacking, which may explain the lack of published long-term studies. Thus, our findings are conservative, and rates and levels may be higher, especially given the link between apathy and mortality. Second, and importantly, the statistical analysis does not allow conclusions about causal relationships of the associated factors. Third, we note that use of a hospital-based sample cannot be generalized to community, non-admitted patients with stroke. Fourth, although we did not perform clinical interviews to diagnose cases of apathy, the AES has been validated against clinical diagnosis. Finally, we did not analyze the emotional/ affective, cognitive, and behavioral components of apathy or their associated pathology, which have been linked, respectively, to pathology in orbital–medial prefrontal cortex and related limbic regions within the basal ganglia (e.g., ventral striatum, ventral pallidum), dorsolateral prefrontal cortex, and the related subregions (associative territory) within the basal ganglia (e.g., dorsal caudate nucleus) and the associative and limbic territories of the internal portion of the globus pallidus.³⁷

Implications for clinicians are the importance of recognizing apathy as a major and common aftermath of stroke, which is associated with poor outcomes for the patient. Its occurrence is likely to be associated with dementia and failing functional abilities.¹² Apathy can be readily overlooked because it places few obvious demands on professional caregivers, even though it is burdensome for family caregivers living with the person with apathy; it is associated with disability, and could hamper rehabilitation and reintegration. Importantly, apathy after stroke is more likely to worsen, rather than peak and then improve, yet its management has been little studied. As regards treatment, the statistically significant but modest benefits of some pharmacological strategies for apathy in people with dementia (see review³⁸) need to be weighed against possible adverse effects. Greater benefit may result from nonpharmacological strategies for treating apathy, such as use of therapeutic activities in patients with dementia.³⁹ It is not known whether these findings from studies of people with dementia and apathy are generalizable to post-stroke apathy, and future studies should investigate this.

We conclude that apathy after stroke is common and, contrary to a previous report, becomes more so with time, as cognition and function further decline. Given its reported association with poorer recovery, preventive and management strategies should be investigated.