The Natural History of Depression up to 15 Years After Stroke
The South London Stroke Register
Background and Purpose—Evidence on the natural history of depression after stroke is still insufficient to inform prognosis and treatment strategies. This study estimates the incidence, cumulative incidence, prevalence, time of onset, duration, and recurrence rate of depression up to 15 years after stroke.
Methods—Data from patients registered in the South London Stroke Register between 1995 and 2009 were used (N=4022 at registration. Maximum number of participants for these analyses n=1233). Depression was assessed in all patients with the Hospital Anxiety and Depression Scale (scores >7=depression) 3 months after stroke, 1 year after stroke, and annually thereafter up to 15 years after stroke. Inverse probability weighting was used to calculate the estimates accounting for missing data.
Results—The poststroke incidence of depression ranged from 7% to 21% in the 15 years after a stroke, with cumulative incidence of 55% and prevalence ranging from 29% to 39%. Most episodes of depression started within a year of stroke, with 33% of the cases starting in the 3 months after a stroke, and none from year 10 onward. Fifty percent of the patients with depression at 3 months had recovered 1 year after stroke. The proportion of recurrent episodes of depression after stroke increased gradually from 38% in year 2 to 100% in years 14 and 15.
Conclusions—The natural history of depression after stroke is dynamic. Depression affects most of the stroke patients with episodes that have a short duration but a high risk of recurrence in the long term.
Although depression is a recognized outcome of stroke,1 most studies investigating depression after stroke have limitations, including selection bias, short follow-up, and small sample size.2,3 The prevalence of depression in the first few years after stroke has been reported in several studies.2 Nonetheless, evidence is poor or lacking on other estimates of the long-term natural history of depression, such as the poststroke incidence, cumulative incidence, time of onset, duration, and recurrence rate.2 Interventions for depression after stroke only show limited effect. Whether these interventions had been started at the right time after stroke and given for an adequate length of time to obtain maximal sustained response has been questioned.4
In this article, the poststroke incidence, cumulative incidence, prevalence, time of onset, duration, and recurrence rate of depression up to 15 years after stroke are estimated in a population-based study.
First in a lifetime stroke patients were recruited from the South London Stroke Register (SLSR), a prospective population-based stroke register covering an inner-city population of 271 817.5 Data from patients, registered in the SLSR between January 1, 1995, and December 31, 2009, and followed up between April 1, 1995 (first 3 months of follow-up assessments), and August 31, 2010, were used (patients at registration, N=4022).
Patients were registered during the acute phase of stroke and were then followed up for 3 months after stroke, 1 year after stroke, and annually thereafter. The World Health Organization definition of stroke was used.6 Follow-up was by postal questionnaire or interview, depending on the capacity of patient to fill in the questionnaire. Such capacity was judged by the patient, the next of kin, or the field worker in a preceding follow-up assessment. Patients unable to complete the follow-up questionnaire, and those not returning them by post, were telephoned to arrange face-to-face interviews or have another follow-up questionnaire posted. Patients who could not be followed up at one time point remained registered and were contacted again for the following annual assessment. At follow-up, patients were assessed for depression using the Hospital Anxiety and Depression Scale (HADS).7 Scores >7 in the HADS depression subscale were considered depression. HADS has been validated in stroke patients showing a good performance both when it is used in a face-to-face interview and when it is self-administered8 (Cronbach’s alpha > 0.80; optimum performance when HADS subscales scores >7 are used to identify depression, sensitivity: 73.1, and specificity: 81.6).7 Despite its good performance, HADS is not a diagnostic scale but a screening tool that indicates risk of depression. However, the term depression will be used in this article for succinctness in patients with scores >7. HADS was routinely collected between 1997 and 2010. Patients registered in 1995 (n=299) and 1996 (n=350) received their first HADS assessment in 1997. Data on HADS were, therefore, not included from these patients in the respective estimates for early rates of depression. Because HADS cannot be answered by proxy, all information was collected directly from patients. Although patients with some degree of cognitive or communication impairment can respond to HADS, no data could be collected from patients with severe cognitive or communication impairment that the field worker, or the patient’s next of kin in case of postal questionnaire, judged would give invalid responses.
The poststroke incidence of depression was calculated among patients assessed at each time point who were also assessed and not depressed in the previous follow-up. Poststroke incidence of depression 3 months after stroke was not calculated because there were no depression assessments before that point. The cumulative incidence of depression was calculated among patients assessed for depression at any time point. The prevalence of depression was calculated among survivors assessed at each time point. The proportion of patients who became depressed for the first time at each assessment, between 3 months and 15 years after stroke, was calculated among patients with complete follow-up until each time point. The proportion of patients depressed at 3 months who recovered each year was calculated among patients with complete follow-up until each time point. Finally, to estimate recurrence rate, the proportion of patients not depressed at one time point, becoming depressed in the following one and having a previous episode of depression reported, was calculated among patients who had ≥3 follow-up assessments. Sociodemographic and clinical characteristics of survivors completing and not completing HADS were compared using χ2 test because these variables were categorical.
As a first step, all estimates were obtained only from patients with complete data (ie, complete case [CC] analysis). However, estimates obtained from CC analysis may be biased if the excluded individuals are systematically different from those included. Therefore, in a second step, estimates were calculated using inverse probability weighting (IPW).9 Using IPW, cases were weighted by the inverse of their probability of being a CC. To weight the probability of being complete, a variable of completeness was created for each estimate. For example, prevalence of depression at 3 months is 1=observed and 0=missing. A logistic regression model was built to identify predictors of completeness. Variables included in the models were those considered to be associated with completeness: age, sex, ethnicity, stroke severity measures (Glasgow Coma Scale, incontinence, and paresis), and disability at baseline. The inverse of the probability of being a CC was calculated and applied to individuals with available data. Finally, estimates were calculated on weighted data. Weighted and CC estimates are presented. For cases with weight >25, only CC estimates are presented because IPW can also introduce error when weights are very large.9 IPW was not used to estimate rate of recurrences because the number of patients available each year was too low, between 1 and 68, to allow for a stable model of completeness to be built.10
Some estimates, particularly those obtained with small number of patients toward the end of the follow-up, had confidence intervals with values >1 or <0. In these cases, the arcsine correction was used.11 When this correction was used, IPW was not possible; therefore, only CC estimates were reported.
Patients or their relatives gave written informed consent. The study was approved by the ethics committees of Guy’s and St Thomas’ Hospital NHS Foundation Trust, King’s College Hospital Foundation, National Hospital for Neurology and Neurosurgery, Queen’s Square Hospital, St George’s Hospital, and Chelsea and Westminster Hospital.
Between 1995 and 2009, the SLSR registered 4022 patients. When the follow-up period finished in August 2010, the follow-up time for survivors ranged from 3 months to 15 years. The number of patients registered in each period, assessed for depression or lost to follow-up, at each time point, is presented in the online-only Data Supplement I. The maximum number of participants available for analysis was 1233, 1 year after stroke. Few differences were observed between sociodemographic characteristics of patients who were and those who were not assessed for depression (online-only Data Supplement II). Up to 10 years after stroke, those who had had more severe strokes were less likely to be assessed (online-only Data Supplement III).
The poststroke incidence of depression after stroke ranged between 7% and 21% per year during the 15-year follow-up (Table 1; online-only Data Supplement III). The proportion of patients depressed at 3 months, first cases after stroke, was 33%. The prevalence of depression ranged from 29% to 39% during the follow-up period (Table 2; online-only Data Supplement IV). Cumulative incidence of depression was 55.4% (53.3%–57.5%) on CC analysis and 58.2% (52.9%–60.5%) using IPW. Thirty-three percent of the assessed patients had their first detected episode of poststroke depression 3 months after the acute event, and this proportion gradually decreased to 4% in year 9. There were no observations of patients having their first episode of depression from year 10 onward (Table 3). Half of the patients who were depressed at 3 months had recovered from depression at 1 year. The other half recovered gradually between years 2 and 9. No cases of depression at 3 months recovering after year 9 were observed (Table 4). The proportion of recurrent cases rose from 38% in year 2 to 100% in years 14 and 15 (Table 5).
Weighted and CC estimates of prevalence, poststroke incidence, cumulative incidence, time of onset, and duration were consistent at all time points.
These analyses and estimates show that the natural history of depression after stroke is dynamic. Depression affects more than half of all stroke patients at some point, with a stable prevalence of ~30% up to 15 years after stroke, with most patients recovering from depression after a few years and having a significant risk of recurrent episodes in the long term.
This study shows a prevalence of depression similar to the one previously reported in other studies, although the follow-up is substantially longer.2,3 Poststroke incidence is an estimate of natural history that has been scarcely investigated.2,12,13 The prevalence and poststroke incidence observed in this study, which are largely stable throughout the follow-up, are estimates suggesting a persisting risk of depression among stroke patients and a dynamic natural history of depression in the long term after stroke. The cumulative incidence of depression in stroke cohorts has been so rarely reported in previous studies that the overall importance of depression among stroke patients has probably been underestimated.2,3 The duration of the episodes of depression is relatively short. Other studies following stroke patients for up to 3 years published similar results.12–14 The increase in recurrent episodes observed during the 15-year follow-up explains why depression starting shortly after stroke and having short duration has a stable prevalence.
The SLSR does not have a control arm, therefore it was not possible to know whether estimates of depression were different from the ones in general population. Two previous studies observed that significantly more stroke survivors were depressed than controls.15,16 Studies observing general population report lower frequency of depression than that observed among stroke patients with a cumulative incidence of depression between 13% and 17% during patient’s lifetime and incidence between 5% and 10%.17–19
It has been reported that medical illness, not only stroke, increases the risk of depression.20 The World Health Organization World Health Survey reported from observations in 60 countries that up to 23% of patients with chronic physical diseases had comorbid depression.21 Life-threatening illness, unpleasant treatments, and drugs causing depression as a side effect may explain this association.20 Most of these apply to stroke patients. However, the increased prevalence of depression specifically among stroke patients may also be attributable to other causes, including the following: (1) depression is a risk factor for stroke, therefore the proportion of patients at risk of depression may be increased among stroke patients; (2) depression and stroke have risk factors in common, such as sedentary lifestyle; (3) depression is a secondary psychological reaction to stroke; (4) depression is secondary to other outcomes of stroke, such as cognitive impairment; and (5) stroke has a direct pathophysiological effect on the brain (eg, increase of cytokine levels).22
As in almost all cohort studies, there are some missing data in this study. This was not only attributable to the difficulty in following up patients for so long but also to the difficulty of some patients in responding to the HADS. The exclusion of patients with cognitive and communication impairment is a limitation affecting most studies of depression in stroke cohorts.2 Nonetheless, missing data were handled using IPW, and to obtain maximum robustness, both the results of IPW and CC analysis are presented. Although IPW adjusts for differences in characteristics of patients with complete and incomplete follow-up, it cannot adjust for unmeasured factors, which may result in some patients being more likely to have incomplete follow-up. However, it should be acknowledged that weighted and CC estimates were always consistent. This suggests that although part of the sociodemographic groups are more likely to be missing than others, this had little impact on the estimates of the natural history of depression after stroke. Therefore, the validity of results from CC analysis should be considered.
To assess the natural history of depression, it would have been ideal to followup patients more frequently because the average duration of episodes of depression is shorter than 1 year.19 It would have also been better to assess depression with a diagnostic tool as well, such as the Diagnostic and Statistical Manual of Mental Disorders-IV criteria.23 However, these limitations are common in large epidemiology studies, such as the SLSR. The HADS shows a good performance detecting depression in patients with no psychiatric conditions according to a systematic review.7 The SLSR is probably the largest population-based cohort of stroke patients followed up for so long. It provides the least biased sampling frame and good statistical power in the analyses of data collected in the long term after stroke, in contrast with previous studies.2 A capture–recapture analysis conducted with the data on incident strokes registered in the SLSR concluded that 88% of the strokes occurring in the study area were being registered.24
Clinicians should acknowledge that depression remains a frequent active problem long after stroke, even when stroke seems to be completely settled and many other medical issues may have presented. With the exception of those patients who do not become depressed shortly after stroke, who seem to be at lower risk, depression requires periodic clinical attention in the long term. The high rate of recurrence of depression should be noted. Assuming that a patient recovering from depression is a closed case could lead to a late diagnosis or an overlooking of a further episode.
Charles D.A. Wolfe is an NIHR Senior Investigator. This article presents independent research commissioned by the National Institute for Health Research (NIHR) under its Program Grants for Applied Research funding scheme (RP-PG-0407-10184). The views expressed in this article are those of the authors and not necessarily those of the National Health Service, the NIHR, or the Department of Health. The other authors have no conflicts to report.
We thank all patients and healthcare professionals involved. Particular thanks to field workers and the team working since 1995 for the South London Stroke Register and the Stroke Research Team at King’s College London.
Sources of Funding
The study was funded by Guy’s and St Thomas’ Hospital Charity, The Stroke Association, Department of Health HQIP grant, UK, National Institute for Health Research Program Grant (RP-PG-0407-10184). Charles D.A. Wolfe acknowledges financial support from the Department of Health via the National Institute for Health Research (NIHR) Biomedical Research Center award to Guy’s and St Thomas’ NHS Foundation Trust in partnership with King’s College London.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.679340/-/DC1.
- Received October 4, 2012.
- Revision received December 21, 2012.
- Accepted December 28, 2012.
- © 2013 American Heart Association, Inc.
- Ayerbe L,
- Ayis S,
- Wolfe C,
- Rudd A
- Hackett ML,
- Yapa C,
- Parag V,
- Anderson CS
- Hackett ML,
- Anderson CS,
- House A,
- Xia J
- Heuschmann PU,
- Grieve AP,
- Toschke AM,
- Rudd AG,
- Wolfe CD
- Seaman SR,
- White IR
- Gleason J
- Wade DT,
- Legh-Smith J,
- Hewer RA
- Aström M,
- Adolfsson R,
- Asplund K
- Kase CS,
- Wolf PA,
- Kelly-Hayes M,
- Kannel WB,
- Beiser A,
- D’Agostino RB
- House A,
- Dennis M,
- Mogridge L,
- Warlow C,
- Hawton K,
- Jones L
- Peveler R,
- Carson A,
- Rodin G
- 23.↵American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders: DSM-IV. Washington DC: American Psychiatric Association; 1994.
- Tilling K,
- Sterne JA,
- Wolfe CD