Frequency, Management, and Predictors of Abnormal Mood After Stroke
The Auckland Regional Community Stroke (ARCOS) Study, 2002 to 2003
Background and Purpose— Mood disorders are an important consequence of stroke. We aimed to identify significant, clinically useful predictors of abnormal mood after stroke.
Methods— The Auckland Regional Community Stroke (ARCOS) study was a prospective population-based stroke incidence study conducted in Auckland, New Zealand, over a 12-month period from 2002 to 2003. All patients were followed up 6 months after stroke onset and abnormal mood was assessed using the 28-item General Health Questionnaire (GHQ-28) administered as part of a structured telephone interview. Multivariate stepwise logistic regression was used to develop a predictive model for “caseness” (score of ≥5 on the GHQ-28) based on several premorbid patient and clinical variables assessed at baseline and 28 days of follow up.
Results— Of patients available at 6 months (n=1172), complete data on mood was available from 739 (60%) patients and 27% (95% confidence interval, 24 to 30%) were defined as cases. Key baseline predictors of abnormal mood were disability and history of depression after adjustment for sex, age, and comorbidity, but the model failed to predict a large amount of the variation in caseness (C statistic 0.587).
Conclusion— This study emphasizes the complex nature of mood disturbance after stroke and that multiple factors are likely to contribute to mood disorders. A simple, clinically applicable, predictive model in stroke care appears difficult to develop.
Mood disorders are a common complication of stroke that are associated with adverse outcomes, yet they are potentially preventable and responsive to treatment. A recent systematic review indicates that at least one third of patients experience “clinically significant” abnormal mood symptoms after the onset of stroke and that this risk appears consistent over the early and later stages of the recovery process.1 Implementing preventive and therapeutic strategies to reduce the risk of abnormal mood and improve rehabilitation outcomes would appear important in the organization of stroke services. To date, data concerning the predictors of abnormal mood after stroke are conflicting because studies have been complicated by methodological heterogeneity, small sample size, and poor statistical modeling techniques. The most consistent variables identified as being associated with depressive symptoms are stroke severity and the presence of physical disability and cognitive impairment.2
A simple predictive tool that identifies those patients at greatest risk of developing abnormal mood, using information readily available during the acute phase of stroke, could be helpful to clinicians in targeting therapy. We present data on the frequency, management, and predictors of abnormal mood among patients registered in a large population-based stroke incidence study.
The Auckland Regional Community Stroke (ARCOS) study in Auckland, New Zealand (NZ) has been described elsewhere.3 In brief, a prospective, population-based register was used to ascertain all cases of acute stroke that occurred among adults in the usually resident population of Auckland over the 12-month period, March 1, 2002, to February 28, 2003. The Auckland Ethics Committee approved the study and written informed consent was obtained from all patients or proxy.
Each patient, or their proxy when patients were dead or severely disabled, was interviewed as soon as possible after stroke onset (baseline) and for survivors at 28 days and 6 months of follow up. In the main, these interviews were undertaken in person (baseline) and over the telephone (follow up) by trained, supervised study nurses using structured questionnaires. Information obtained at baseline included sociodemographic variables of age, sex, self-defined ethnicity, marital status, place of residence, employment status, premorbid dependence on others for self-care activities of daily living (ADLs), smoking status and alcohol intake, comorbidity (history of hypertension, diabetes, stroke, heart disease, and any significant illness that restricted activity), and concomitant medications. Information on treatments received for depression before stroke was asked only of patients surviving to 28 days. Data collected on the index stroke event included pathologic type, loss of consciousness, and ADL as measured by the Barthel Index (BI) at onset.
At 6 months, abnormal mood was assessed in cognitively competent patients (those scoring ≥7 on the Hodkinson Mental Test4) using the 28-item General Health Questionnaire (GHQ-28).5 The primary outcome was the proportion of patients who scored ≥5 (notated as a 4/5 cut point) according to the standard scoring on the GHQ-28.
All baseline data were summarized and those alive at 6 months with and without a completed GHQ-28 were compared using χ2 tests with Yates continuity correction or two-sample t tests. Sensitivity analyses were undertaken for two higher cut points (5/6 and 8/9) of the GHQ-28 to account for the possibility that older patients with comorbidities may require higher cut points for abnormal mood. Univariate logistic (GHQ-28 4/5, 5/6, 8/9 as dependent variables) models were constructed for all possible covariates and outcomes.
Baseline variables, for which there is some evidence that they might be determinants of abnormal mood after stroke, were considered for possible inclusion as covariates. Multiple dichotomous variables (dummy variables) were used to assess the effect of each categorical variable on the outcome. Only variables that demonstrated a significant association (P<0.05) with the outcome in univariate models were considered for possible inclusion in multivariate models. In addition, the variables sex, age, and comorbidity were forced into each regression model because all have been associated with abnormal mood in the general population.
The initial covariate selection procedure for multiple logistic regression used the automatic variable selection algorithms available in SAS6 using complete participant analysis.7 In addition, all variables found to be significant in the univariate selection process were added to the model and retained if they were significant at P<0.05. When there was high correlation between variables, only one was entered into the model. No significant interactions were identified in any of the models.
A total of 2001 strokes were registered among 1938 patients (mean±standard deviation age, 73±13.8 years; 54% female),3 of whom 90.7% were admitted to the hospital and 87.4% received either computed tomography or magnetic resonance imaging. The Figure shows the progress of patients through the follow-up phases. At 28 days, 467 (24%) patients had died and 130 (7%) either refused or failed to complete an interview. Between 28 days and 6 months, a further 141 (7%) patients had died and 28 (1%) either refused or failed to complete a 6-month interview. Another 26 (1%) patients had their 28-day assessment completed retrospectively at 6 months. Consequently, there were 1172 interviews conducted at 6 months, of which 812 were directly with patients. However, a small number (46) of patients elected to answer only a few questions and another 27 completed most, but not all, of the GHQ-28; these responses were excluded from analyses. Therefore, there were 739 6-month stroke survivors with complete GHQ-28 data, hereafter referred to as the “study group.”
Characteristics of the Auckland Regional Community Stroke Population
Table 1⇓ shows various sociodemographic, medical, and clinical variables of patients at baseline. Over half of the baseline interviews were completed by proxies, and most patients (73.8%) were of New Zealand/European ethnicity, married/partnered (49.6%), living with family (60.4%), and either retired or principally undertaking home duties (72.4%). The most common comorbid conditions were high blood pressure (55.7%) and heart disease (42.7%), and most strokes were the result of cerebral infarction (71.3%). Compared with 6-month survivors without complete GHQ-28 data, the study group (n=739) were more likely to be male, younger, New Zealand/European, partnered, living with family, and functionally independent before the stroke. Otherwise, the two 6-month groups of stroke survivors did not differ substantially with respect to risk factors, psychiatric history, and stroke type.
Abnormal Mood at 6 Months in the Study Group
At 6 months, 198 (26.7%; 95% CI, 24 to 30%) of the study group met criteria for abnormal mood (GHQ-28 4/5; 22% GHQ 5/6; 12% GHQ 8/9). There were 90 (12.5%) patients who had received some form of treatment for depression before their stroke: 60 (8.4%) had “seen a doctor, psychologist, or counselor and received medication,” 17 (2.4%) had “seen a doctor, psychologist, or counselor without receiving medication,” 8 (1.1%) had “been admitted to the hospital for treatment,” and 5 (<1%) had received electroconvulsive therapy (ECT). Over half (52) of the 90 patients with prestroke depression had abnormal mood 6 months after stroke. Eighty-four (11.4%) patients in the study group received some treatment for depression after stroke: 45 (6.1%) had “seen a doctor, psychologist, or counselor and received medication,” 20 (2.7%) had “seen a doctor, psychologist, or counselor without receiving medication,” 16 (2.2%) had received medication from a doctor without psychology or psychiatric assessment, and one patient was admitted to the hospital for treatment. Only 27 (30%) of the patients who received treatment for depression after stroke had also received treatment for depression before their stroke. Approximately half (42) of those with abnormal mood at 6 months had received treatment for depression since their stroke.
Predictors of Abnormal Mood
Premorbid dependence on others for ADL, recent use of psychiatric medication and having received treatment for depression before the stroke, loss of consciousness in the first 24 hours, and requiring “much help” with ADL immediately after stroke were each independently associated with the presence of abnormal mood at 6 months in univariate analyses. Table 2 shows that having prior treatment for depression (odds ratio [OR], 2.14; 95% CI, 1.34 to 3.43) and requiring “much help” with ADLs immediately after stroke (OR, 2.35; 95% CI, 1.33 to 4.14) were significant factors in the multivariate model for abnormal mood. The C statistics were 0.582 for the unadjusted and 0.587 adjusted models, respectively. However, as shown in Table 3, this model only correctly identifies 54% of patients’ mood status at 6 months. Identical predictor variables were significant in models that used the higher GHQ-28 cut points of 5/6 and 8/9, although in the latter, having received prior treatment for depression remained significant (OR, 1.94; 95% CI, 1.07 to 3.54), but loss of consciousness (OR, 2.12; 95% CI, 1.32 to 3.41) replaced requiring help in ADL. Premorbid dependency was also significant (OR, 2.12; 95% CI, 1.01 to 4.43), with a C statistic of 0.638 for the adjusted model.
Approximately one fourth (27%) of patients in the study group had abnormal mood as defined by the GHQ-28 at 6 months after the onset of stroke. This figure falls within the pooled range of estimates found in other population-based stroke incidence studies that included similar measures1 and is over twice the frequency expected in the general population.8,9 Premorbid patient and clinical stroke factors measured during the acute phase did not predict a significant amount of the variation in the presence of abnormal mood. Important baseline predictors of abnormal mood were more disability in ADL after stroke and a history of depression. As found in previous stroke studies,2 but contrary to research in the general population,10,11 no association was found with age and sex.
Some strengths of this study include the use of a large population-based12 stroke incidence study with methodologically ideal criteria13 and the high levels of follow up and outcome assessment. Although this avoided any bias that would have been associated with the inclusion only of stroke patients admitted to hospital or rehabilitation centers, we did experience similar difficulties with the assessment of abnormal mood as experienced in other studies14–19 with the exclusion of patients with cognitive and communication difficulties. Thus, complete information on abnormal mood was only available from 60% of those interviewed, a figure that may appear low as a result of the exclusion of proxy responders and those with incomplete GHQ-28 data. Moreover, for logistic reasons, we used a brief questionnaire of abnormal mood symptoms rather than a detailed psychiatric interview, which would have provided specific Diagnostic and Statistical Manual of Mental Disorders (DSM)20 diagnoses. Even so, the GHQ-28 is a well-validated and widely used questionnaire for the assessment of psychiatric morbidity,21 and the robustness of the primary findings were confirmed in secondary analyses with different cut points for caseness.
Our results both add support to, and contradict aspects of, research on mood disorders in the general population. The relapsing nature of mood disorders22 was confirmed by the finding of previous depression being a strong risk factor for abnormal mood after stroke. The association of nonstroke comorbid disorders and depression in the general population23 was not replicated in these analyses. It is possible that the consequences of a “recent” stroke overshadow the impact of other concomitant illnesses, although our crude composite comorbidity index may have diluted the relative effect of individual illnesses with variable strength of association. On the basis of stroke severity and physical disability being associated with depression after stroke in other studies,2 it has been postulated that severe disability reflects larger strokes and greater involvement of mood processing regions in the brain.24 Alternatively, compared with those with functional independence, physically disabled patients are likely to experience greater changes in social and financial circumstances and may be more susceptible (or less resilient) to such changes. This study adds further support to previous findings that stroke risk factors, namely age and sex, are poor predictors of future abnormal mood.25 It is likely that the impact of stroke risk factors on abnormal mood are diminished by the impact of disability and the recovery phase, including medical interventions at the time of stroke25 and adjustment to the stroke event, and the physical, cognitive, social, and financial consequences. Yet, despite these findings, the ability of premorbid factors and disability to predict abnormal mood after stroke is modest at best.
A major aim of these analyses was to develop a predictive model for abnormal mood after stroke that could be useful to clinicians. Clinical diagnosis ultimately relies on the ability of a clinician to interpret the results of a diagnostic test or to estimate the likelihood of disease on the basis of certain clinical parameters.26 Despite current suboptimal levels of detection and management of mood disorders after stroke in clinical practice,27 the models developed using the ARCOS data do not have sufficient predictive ability to justify their use in practice. For this reason, no attempt was made to validate the models on an external dataset or in clinical practice. However, it would be interesting to know how accurate a model needs to be before clinicians would consider adopting it in clinical practice.
This is the only study of abnormal mood after stroke that has included information regarding the proportion of patients (8.8%) who received some form of “talking therapy” as treatment of depression. However, a sizable proportion (79%) of cases in the study group did not receive any treatment for abnormal mood after stroke and over half of the study group using medication for the treatment of depression were defined as cases on the GHQ-28. This information supports studies showing insufficient access and response to treatments for symptoms of abnormal mood.28,29 This may reflect a failure of health professionals to recognize mood disorders and an unwillingness of patients to report mood symptoms and receive treatment.30 However, the most appropriate management strategies for stroke patients with abnormal mood have yet to be determined.
To optimize the quality of stroke services, reliable methods for identifying those patients at greatest risk of mood disorders after stroke are still required. At present, we only have relatively crude indicators that disability and a history of depression may predict abnormal mood.
Full details of the ARCOS Study Group are available elsewhere.3 We are indebted to the research nurses for their dedication and performance; the support of staff at the Coroner’s Office in Auckland; the assistance of staff of the New Zealand Health Information Service; the help provided by staff at the Auckland office of the New Zealand Stroke Foundation; the support of many doctors, nurses, and administrative staff within and outside Auckland; the ARCOS participants and their families and friends; and Alistair Woodward for his contribution to the ARCOS steering committee.
Sources of Funding
The study was funded by the Health Research Council of New Zealand. During the completion of this work, M.L.H. was in receipt of a University of Auckland Senior Health Research Scholarship. Neither funding body had a role in the conduct or reporting of this report.
- Received March 21, 2006.
- Accepted May 5, 2006.
Hackett ML, Yapa C, Parag V, Anderson CS. The frequency of depression after stroke: a systematic review of observational studies. Stroke. 2005; 36: 1330–1340.
Hackett ML, Anderson CS. Predictors of depression following stroke: a systematic review of observational studies. Stroke. 2005; 36: 2296–2301.
Anderson C, Carter K, Hackett M, Feigin V, Barber PA, Broad JB, Bonita R, on behalf of the Auckland Regional Community Stroke (ARCOS) Study Group. Trends in stroke incidence in Auckland, New Zealand, during 1981 to 2003. Stroke. 2005; 36: 2087–2093.
Hodkinson HM. Evaluation of a mental test score for assessment of mental impairment in the elderly. Age Ageing. 1972; 1: 233–238.
Goldberg DP, Williams P. A User’s Guide to the General Health Questionnaire. Windsor: NFER-Nelson; 1988.
The SAS System for Windows, Release 8.02. Cary, NC: SAS Institute Inc; 2002.
Little RJA, Rubin DB. Statistical Analysis With Missing Data. New York: John Wiley & Sons; 1987.
National Institute for Clinical Excellence. Depression: management of depression in primary and secondary care. 2004. Available at: www. nice.org.uk/CG023.
The World Health Report 2001: Mental Health: New Understanding, New Hope. Geneva: World Health Organization; 2001.
Beekman ATF, Copeland JRM, Prince MJ. Review of community prevalence of depression in later life. Br J Psychiatry. 1999; 174: 307–311.
Weissman MM, Bland RC, Canino GJ, Faravelli C, Greenwald S, Hwu HG, Joyce PR, Karam EG, Lee CK, Lellouch J, Lepine JP, Newman SC, Rubio-Stipec M, Wells JE, Wickramaratne PJ, Wittchen H, Yeh EK. Cross-national epidemiology of major depression and bipolar disorder. JAMA. 1996; 276: 193–199.
Sudlow CLM, Warlow CP. Comparing stroke incidence worldwide. What makes studies comparable? Stroke. 1996; 27: 550–558.
Numminen H, Kotila M, Waltimo O, Aho K, Kaste M. Declining incidence and mortality rates of stroke in Finland from 1972–1991. Stroke. 1996; 27: 1487–1491.
Kotila M, Numminen H, Waltimo O, Kaste M. Depression after stroke: results of the FINNSTROKE study. Stroke. 1998; 29: 368–372.
Appelros P, Viitanen M. Prevalence and predictors of depression at one year in a Swedish population-based cohort with first-ever stroke. Journal of Stroke & Cerebrovascular Diseases. 2004; 13: 52–57.
House A, Dennis M, Mogridge L, Warlow C, Hawton K, Jones L. Mood disorders in the year after first stroke. Br J Psychiatry. 1991; 158: 83–92.
Burvill PW, Johnson GA, Jamrozik KD, Anderson CS, Stewart-Wynne EG, Chakera TMH. Prevalence of depression after stroke: the Perth Community Stroke Study. Br J Psychiatry. 1995; 166: 320–327.
Wilkinson PR, Wolfe CD, Warburton FG, Rudd AG, Howard RS, Ross-Russell RW, Beech R. Longer term quality of life and outcome in stroke patients: is the Barthel index alone an adequate measure of outcome? Quality in Health Care. 1997; 6: 125–130.
Diagnostic and Statistical Manual of Mental Disorders: DSM-IV. Washington, DC: American Psychiatric Association; 1994.
Bowling A. Measuring Disease. Buckingham: Open University Press; 1995.
Fawcett J. Depression. 2003. Available at: http://merck.micromedex.com/index.asp?page=bpm_brief&article_id=BPM01PS19.
Carota A, Berney A, Aybek S, Iaria G, Staub F, Ghika-Schmid F, Annable L, Guex P, Bogousslavsky J. A prospective study of predictors of poststroke depression. Neurology. 2005; 64: 428–433.
Boden-Albala B, Litwak E, Elkind MSV, Rundek T, Sacco RL. Social isolation and outcomes post stroke. Neurology. 2005; 64: 1888–1892.
Gill CJ, Sabin L, Schmid CH. Why clinicians are natural Bayesians. BMJ. 2005; 330: 1080–1083.
House A, Dennis M, Hawton K, Warlow C. Methods of identifying mood disorders in stroke patients: experience in the Oxfordshire Community Stroke Project. Age Ageing. 1989; 18: 371–379.
Peveler R, Carson A, Rodin G. Depression in medical patients. BMJ. 2002; 325: 149–152.
Hickie I, Davenport T, Scott E. Depression: Out of the Shadows. ACP Publishing Pty Ltd & Media 21 Publishing Pty Ltd; 2003.