Frequency of anxiety after stroke: An updated systematic review and meta-analysis of observational studies

Background Anxiety is a common and distressing problem after stroke. Aims To undertake an updated systematic review and meta-analysis of observational studies of anxiety after stroke and integrate the findings with those reported previously. Summary of review Multiple databases were searched in May 2018 and 53 new studies were included following dual independent sifting and data extraction. These were combined with 44 previous studies to form a combined data set of 97 studies, comprising 22,262 participants. Studies using interview methods were of higher quality. Rates of anxiety by interview were 18.7% (95% confidence interval 12.5, 24.9%) and 24.2% (95% confidence interval 21.5, 26.9%) by rating scale. Rates of anxiety did not lower meaningfully up to 24 months after stroke. Eight different anxiety subtypes were also reported. Conclusions This review confirms that anxiety occurs in around one in four patients (by rating scale) and one in five patients (by interview). More research on anxiety subtypes is needed for an informed understanding of its effects and the development of interventions.


Background
Mood problems are common after stroke with reported rates of depression, apathy, and distress significantly higher than in the general population. 1,2 Anxiety is common in the general population 3 but its presence in stroke patients has been relatively under-recognized both in clinical and research settings. A systematic review of observational studies 4 included 44 studies and reported rates of anxiety as 18.3% when diagnosed by interview and 24.3% by rating scale. The review reported that rates lowered with time after stroke, although they remained higher than in the general population. 3 However the inclusion of relatively small numbers of studies at some time points meant that there was considerable imprecision in rates. Furthermore, studies had also used a number of different scales and cutoff scores to define anxiety, producing considerable uncertainty around the true rate.
More recent research has argued for the importance of subtypes of anxiety (e.g., panic disorder; specific or simple phobias) for understanding its impact and for developing and delivering suitable interventions 5 or adapting those shown to be effective in the general adult population. 6 Our review in 2013 had recorded subtypes when they were reported in primary studies, but this information was available in only three of the eight relevant studies.
Our review of 44 studies had searched databases until March 2011 and we are aware of the publication since then of further, potentially relevant studies. Another recent review in this area 7 was limited to publications over 2011-2017, from a small range of languages, and only those using self-report measures of anxiety. Consequently, updating the Campbell-Burton (2013) review 4 could have several potential benefits, not only making the findings more current but also potentially increasing the sample size and precision, particularly on subgroup analyses. Therefore, the aims of this study were to undertake an updated systematic review of observational studies of anxiety after stroke; to integrate the findings with those previously reported 4 ; and to disaggregate rates of anxiety by subtype, rating scale, and time after stroke.

Method
This review and the original systematic review 4 were both undertaken according to the PRISMA guidelines. 8 The review update protocol was registered on PROSPERO: CRD42018093718.

Inclusion/exclusion criteria
Studies were included if undertaken in populations or groups of patients with a clinical diagnosis of hemorrhagic or ischemic stroke or transient ischemic attack (TIA) and were assessed for symptoms of anxiety on a rating scale such as the Hospital Anxiety and Depression Scale (HADS) 9 or were diagnosed by clinical interview. We translated papers published in languages other than English if the title and abstract indicated potential eligibility. We excluded studies if they: . used proxy measures of anxiety; . were intervention studies; . were limited to patients with subarachnoid hemorrhage or other specific stroke subtypes or demographic characteristics; . were not designed to screen expressly for anxiety, or used nonspecific measures of psychological distress; . used retrospective recruitment or mood reporting; . employed convenience sampling; . reported anxiety as a continuous outcome and we could not derive a categorical assessment.

Study identification and data extraction
We searched the following digital databases: Medline, Embase, CINAHL, PsycINFO, Allied and Complementary Medicine, and Proquest dissertation, using a search strategy developed in Medline (see Supplemental Tables 1 and 2) and adapted to the other databases. We restricted the search to studies published from January 2009 (to ensure relevant studies were not missed) to May 2018 and applied no language restrictions. The search was undertaken by one investigator (ADR) and screening of title and abstract was undertaken by ADR with a second reviewer (NS) and decisions taken against the selection criteria. Independent data extraction was performed by two reviewers (two of: ADR, NS, PK) for all eligible studies.

Quality of evidence
We extracted information on study design, setting, and patient characteristics. Study quality was assessed using the Newcastle-Ottawa Scale (NOS) for cohort studies, 10 see Supplemental Tables 3 and 4, which includes eight criteria. One criterion (comparability of cohorts) was recorded as not applicable because the included studies were all reporting prevalence rates derived from a single cohort. Study quality was not used to determine inclusion. Finally, we assessed the quality of the 44 studies included in the original review using the NOS measure.

Data synthesis
We combined the studies reported in the 2013 review with those identified in the update.
Studies were grouped into two categories based on method of case ascertainment: those using clinical interview for diagnosis; and those using a rating scale. We also extracted data on rates: at five different time points after stroke (up to 1 month; 1-5 months; 6-12 months; 12-24 months; over 24 months) and did this separately for interview and rating scale studies; from different rating scales or different caseness thresholds on the same scale (using whatever had been used in the primary data study); and, for interview-based studies only, rates of anxiety subtypes.
We undertook several meta-analyses. We excluded from pooling one study 11 using the hierarchical diagnostic rule in the Diagnostic and Statistical Manual of Mental Disorders-Third Edition (DSM-III), 12 meaning that anxiety is not diagnosed in the presence of depression, which may falsely deflate the reported rate of anxiety. For studies using rating scales, we used whatever caseness threshold had been used by the primary researchers. When studies reported rates of anxiety at more than one time period, we used the first-reported time period as the primary outcome prevalence rate. The random effects model was used to summarize data. Chi-square was used to test for subgroup differences, and heterogeneity among the studies was assessed by the I 2 statistic. We used Review Manager 5.3 13 for data analysis.
publications. The following results are based on the integrated data set of 97 studies, comprising 44 studies from the original review 11,14-58 and 53 studies from the update   (see supplemental Table 1).

Study characteristics
The 97 studies included 26,262 participants and had been published between 1984 and 2018. Most had recruited patients from hospital (52), while other settings were rehabilitation (19), general population (15), a combination of settings (2) or not reported (8). Most studies were cross-sectional (78) or longitudinal cohort in design (15), although one used a case-control design and the design was not reported in two cases. Cohort studies included a range of data collection time points: 2 time points (n ¼ 4); 3 time points (n ¼ 4); 4 time points (n ¼ 4); 5 time points (n ¼ 2); 13 time points (n ¼ 1). Anxiety was recorded in patients in a very wide range of time periods after stroke (from 2 weeks to 10 years). The studies had been undertaken in 34 different countries: UK (18); Netherlands (5); Norway, Italy, China, and Australia (four each); Sweden, Nigeria, Japan, India, Ireland, New Zealand, and Bosnia and Herzogovina (three each); Thailand, Switzerland, South Korea, USA, Hong Kong, and Croatia (two each); and Benin, Brazil, Spain, Ukraine, Bahrain, Turkey, Tanzania, Finland, Slovakia, Georgia, Russia, France, and Germany (one each). Two studies were undertaken in more than one country; the country of origin was not reported in six studies. Mini-International Neuropsychiatric Interview-Plus (MINI-Plus) 119 ; and the CCND-3. 114 Anxiety prevalence was reported in the interview studies from samples ranging from 50 to 350 participants (total 3109; median 149.5).

Measurement and assessment of anxiety
Nine different standardized scales were used to identify anxiety symptoms and generate caseness rates in 78 studies: the Generalized Anxiety Disorder (GAD) 120 (n ¼ 1); HADS-Anxiety subscale 9 (n ¼ 50); Hamilton Anxiety Rating Scale (HAM-A) 121 (n ¼ 7); Neuropsychiatric Inventory (NPI) 122 (n ¼ 1); Zung Self-rated Anxiety Scale 123 (n ¼ 3); Irritability Depression and Anxiety Scale, Anxiety subscale (IDA-A) 124 (n ¼ 1); Beck Anxiety Inventory (BAI) 125 (n ¼ 2); Adult Manifest Anxiety Scale (AMAS) 126 (n ¼ 1); and the General Health Questionnaire (GHQ-60 anxiety subscale) 127 (n ¼ 1). In addition, one study used a single question measure of anxiety, and another used a series of five researcher-developed questions. Three of these scales (HADS-A; BAI; HAM-A) were used with more than one caseness threshold. In total 20 different combinations of standardized scales and thresholds were used in the included studies. Anxiety prevalence was reported in the rating scale studies from samples ranging from 15 to 4079 participants (total 23,153; median 81).

Anxiety prevalence
The overall prevalence of anxiety when assessed by interview ranged from 0.6% to 33.3% in the primary studies. The updated pooled prevalence derived from the 18 included studies was 18.7% (95% confidence interval 12.5-24.9%), see Figure 2. Heterogeneity among the included studies was very high (97%).
The assessment of anxiety by rating scale produced rates in the range 4.8-63.6% in the 78 included studies. The overall frequency of anxiety ''caseness'' by rating scale was 24.2% (95% confidence interval (CI) 21.5-26.9%), see Figure 3. Heterogeneity among the included studies was very high (95%).
Given the difference in prevalence rates obtained from the interview and rating scale studies, we did not calculate a rate combining data from the two study types.

Pooled anxiety prevalence at different times after stroke
Pooled rates of anxiety in the acute phase (within one month of stroke) were reported as 15.5% (95% CI 6.3-24.7%) in seven studies using interview, and as 25.5% (95% CI 18.6-32.3%) in 19 studies using rating scales.
Between one and five months after stroke rates of anxiety by interview were 21.4% (95% CI 19.2-23.5%) in eight studies using interview methods, and 23.6% (95% CI 18.9-28.2%) in 24 studies using rating scales.
Between 12 and 24 months after stroke, only one study used interview methods to report a rate of 11.0% (95% CI 3.5-18.5%), whereas 11 studies used rating scale methods and found an overall rate of 26.6% (95% CI 16.8-36.3%).

Anxiety subtype caseness
Among the 19 studies that used interview methods to reach a definition of anxiety caseness, 10 also reported the rate of anxiety subtypes.
GAD was reported in eight studies. 43,45,47,59,73,81,86,103 However, a pooled prevalence was not calculated because in some studies it is not clear if GAD had been reported as a subtype of anxiety or as a generic anxiety diagnosis. Similarly, rates were not pooled for phobic disorder, which was reported in three studies, 59,73,101 because it is unclear whether the category ''phobic disorder'' includes all types of phobias or is a distinct phobia subtype.

Quality ratings of studies
Studies were rated on the seven relevant items of the NOS scale, 10 with each item ranked as low or high risk of bias. Among the 97 studies low risk of bias was assigned to scale items ranging from 1 out of 7 to 6 out of 7 items (median 4/7). In studies using interview methods the range was 2/7 to 6/7 (median 4/7), and in studies using rating scale methods low risk of bias ranged from 1/7 to 5/7 items (median 4/7). Studies using interview methods had lower risk of bias than studies using rating scales (Mann-Whitney U ¼ 436.5; z ¼ À2.763; p ¼ .0058). Rates of low risk of bias varied considerably across the seven scored items. All 97 studies had low risk for length of follow-up, 83 for ascertainment of exposure, and 81 for representativeness of the exposed cohort. Low risk was present for 62 studies on adequacy of follow-up. Few studies had low risk of bias for the remaining three items: outcome assessment (n ¼ 20); anxiety shown not to be present at the study start (n ¼ 10); and selection of the non-exposed cohort (n ¼ 4).

Brief summary of the findings
This updated systematic review included 53 studies, which were combined with the 44 studies included in  the 2013 review. 4 The 97 primary data studies included 19 studies using interview methods and 78 studies using rating scales. The pooled prevalence of anxiety after stroke was 18.7% when diagnosed by interview and 24.2% by self-report rating scale, confirming the rates reported in the previous review and also confirming the previously reported pattern of lower rates when using interview. Increasing the number of studies in the data pooling produced increased rate precision, particularly for interview studies. Rates of anxiety were relatively stable in the years after stroke.

Strengths and weaknesses of the study
The updated and combined review used a number of systematic review methods that increase review rigor and tend to reduce bias: searching of multiple databases; dual, independent screening used to determine entry criteria and for extraction; no language or date limits were applied; included studies were assessed for quality; and data pooling was used and reported when appropriate. We searched ProQuest for dissertations, and included conference abstracts, but otherwise did not search for unpublished studies.
The included primary data studies varied in quality, although study quality was not used as an entry criterion to the review. Studies using interview methods tended to be higher quality. Primary studies were included from many countries, although all studies except three were reported in English; this reflects a common finding in systematic reviews, although it is unclear if this would produce a reporting bias similar to that reported in reviews of intervention studies.
Combining the studies found with those reported in the 2013 review allowed further data pooling, although in some cases the pooled estimates were based on small numbers of primary data studies, and levels of heterogeneity were often very high. Rates were reported using a range of different interview methods and ratings scales (and cutoff scores); data pooling for the overall prevalence calculations used whatever cutoff and timing had been reported in the primary study, which inevitably led to the combination of a variety of methods and reported rates. However, it was thought that this potential disadvantage was offset by the advantage gained by increased overall sample size; the rates have now been calculated using aggregate samples of 3109 (in interview studies) and 23,153 (in rating scale studies).
We excluded studies reporting proxy ratings of anxiety as the focus of the review was on self-rating. However, one consequence is the exclusion of studies of patients with strokes causing severe cognitive or language impairment, limiting the review's external validity.

What this review adds
Updating the review led to the addition of a large number of studies published up to 2018, allowing rates to be estimated from 19 studies (for interview) and 78 studies (for rating scale), resulting in increased precision in estimates. Caseness rates generated by interview are confirmed as meaningfully lower than those generated by rating scale (on average anxiety is shown to occur in one in five patients rather than one in four), a direction of difference replicating that seen in depression after stroke. 1,2 The update confirmed that anxiety continues to be prevalent many years after stroke onset. The review update also allowed the calculation of rates for some anxiety subtypes such as panic disorder and phobias, which were shown to vary considerably, supporting the view 5 that this diagnostic detail is essential for an informed understanding of the phenomenon and development of effective interventions. However, it is notable that only small numbers of studies reported subtypes; for example, rates of social phobia and OCD were based on just two studies with a combined sample size of 293. In some studies, it was not clear whether subtypes were differentiated from a generic anxiety diagnosis.

Implications for research
This updated review has included almost 100 studies and 26,262 participants, reporting the rate of anxiety after stroke, although in the case of some primary studies, this was not their main objective. Almost 80 studies reported the rate of anxiety by rating scale and there seems little value in further new studies adding to this total. However, there remains little evidence on rates of anxiety more than 12 and 24 months after stroke. A crucial advantage in future research would be gained by greater consensus on the rating scale (and its threshold for caseness) providing the most robust indication of anxiety after stroke: for example, receiveroperated characteristic (ROC) analysis of studies using interviews and rating scales could provide this. Further studies into anxiety subtypes (diagnosed by interview) would provide a useful addition to the published research. Similarly further studies assessing which factors tend to be associated with the onset and/or persistence of anxiety after stroke are warranted; 28 quantitative and qualitative research could both make contributions to answering this important question.

Implications for practice
The updated review has confirmed the high rate of prevalence of anxiety after stroke and also confirmed that rates International Journal of Stroke, 15 (3) are sustained beyond the early months after stroke; that is, beyond what could be termed the initial reaction to stroke onset and discharge home after hospital admission. This suggests it is important to continue to assess or screen for anxiety 12 months or more after stroke onset, although the continued lack of evidence for interventions in this patient group does preclude evidencebased decisions about treatments if anxiety is identified. 129 Anxiety continues to be a problem for many patients, which also has implications for the mood and quality of life of unpaid carers, 130 and its rate is similar to that of depression after stroke. Anxiety subtypes reported in this review tend to have a relatively low prevalence but their presence confirms the impact of mental health problems, which may compound any physical and cognitive effects of the stroke as well as cause distress.

Declaration of conflicting interests
The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Peter Knapp is an author on one study included in this review. Otherwise the authors have no conflicting interests to declare.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work received no specific grant funding. Much of the work for the review update was undertaken by Alexander Dunn-Roberts in fulfilling the requirements for a Master's in Public Health degree at the University of York.