Examining the relationship between fatigue and cognition after stroke: A systematic review

ABSTRACT Many stroke survivors experience fatigue, which is associated with a variety of factors including cognitive impairment. A few studies have examined the relationship between fatigue and cognition and have obtained conflicting results. The aim of the current study was to review the literature on the relationship between fatigue and cognition post-stroke. The following databases were searched: EMBASE (1980–February, 2014), PsycInfo (1806–February, 2014), CINAHL (1937–February, 2014), MEDLINE (1946–February, 2014), Ethos (1600–February, 2014) and DART (1999–February, 2014). Reference lists of relevant papers were screened and the citation indices of the included papers were searched using Web of Science. Studies were considered if they were on adult stroke patients and assessed the following: fatigue with quantitative measurements (≥ 3 response categories), cognition using objective measurements, and the relationship between fatigue and cognition. Overall, 413 papers were identified, of which 11 were included. Four studies found significant correlations between fatigue and memory, attention, speed of information processing and reading speed (r = −.36 to .46) whereas seven studies did not. Most studies had limitations; quality scores ranged from 9 to 14 on the Critical Appraisal Skills Programme Checklists. There was insufficient evidence to support or refute a relationship between fatigue and cognition post-stroke. More robust studies are needed.


Introduction
Fatigue is frequently reported after stroke (Barbour & Mead, 2012). It is conceptualised as disproportionate mental or physical exhaustion and lack of energy triggered by simple activities that do not ameliorate with ordinary rest (De Groot, Phillips, & Eskes, 2003;Staub & Bogousslavsky, 2001). However, there is no universally agreed-upon definition, despite the consensus that fatigue is clinically significant, which makes it difficult to diagnose and impedes its assessment (De Groot et al., 2003). The reason for such lack of definition stems from the fact that fatigue is complex.
While the exact aetiology of fatigue is unclear, it is argued that post-stroke fatigue is multifaceted and studies have revealed associations between fatigue and several commonly reported and can hinder full recovery . However, the relationship between post-stroke fatigue and cognitive impairment has rarely been examined. Some studies that address this question have not found any significant relationship between the two (Schepers et al., 2006;Van Eijsden, van de Port, Visser-Meily, & Kwakkel, 2012), whereas others have found that attention and speed of information processing were associated with fatigue after stroke (Appelros, 2006;Hubacher et al., 2012;. Systematic reviews are currently the most efficient method of reviewing the existing literature and justifying further research (Moher, Liberati, Tetzlaff, & Altman, 2009). There is a systematic review (Ponchel, Bombois, Bordet, & Hénon, 2015) that investigated factors associated with fatigue, including cognition, however there was no previous review examining the relationship between cognition and fatigue after stroke.

Objective
The aim was to review the correlation between the severity of fatigue, as assessed on questionnaire measures, and cognitive abilities after stroke in adults.

Methods
The review was based on the PRISMA Statement for systematic reviews (Moher et al., 2009) and followed the guidelines provided by the Centre for Reviews and Dissemination (CRD; Centre for Review and Dissemination, 2009).

Type of participants
Studies were included if: (1) their sample included stroke patients, or at least 75% of participants were stroke patients. If stroke patients comprised less than 75%, studies were included if they reported or provided separate data for the stroke patients. Stroke was defined as a clinical syndrome of presumed vascular origin, typified by rapidly developed clinical signs of focal or global disturbance of cerebral functions, lasting more than 24 hours with no apparent cause other than vascular origin, as provided by the World Health Organisation (Aho et al., 1980). Subarachnoid haemorrhage (SAH) was not included because it requires different management from stroke (i.e., surgical operation) (Bederson et al., 2009;Sacco et al., 2013) and if a study included both stroke and SAH patients, it was included only if 75% of them were stroke patients or if separate data were available; (2) the sample included adults aged 18 years or over, or in the case that children were included only if separate results were available for those aged 18 years and over.

Type of studies
Studies were included if they: (3) used the term "fatigue". Studies that assessed concepts related to fatigue, such as exhaustion, lack of energy, vitality or tiredness, were included if they reported these to be aspects of fatigue. Frailty was not considered as fatigue in this review because the term is highly associated with old age and natural ageing processes rather than as an explicit disease symptom. Despite the fact that fatigue might be a symptom of frailty, these two are distinctive (Avlund, 2013); (4) included an assessment of fatigue that provided a quantitative score of at least three response categories. The reason is that a yes/no or agree/disagree format would only give information regarding the presence of fatigue and not allow correlations with severity of fatigue to be examined. If the study evaluated tiredness, for example, but assessed it with any assessment other than a fatigue scale, it was not included. These fatigue assessments included questionnaires, rating scales, visual analogue and ranking scales.
Furthermore, the SF-36 (Ware & Sherbourne, 1992) vitality subscale was considered as it is described as a measure of energy/fatigue. Studies that involved interviews on fatigue or ratings of fatigue as present or absent were not included; (5) reported on any aspect of cognition. This included any reference to memory, attention, spatial abilities, visual neglect, speed of information processing, mental flexibility, executive function, mental slowness, orientation, concentration, cognitive control, decision making, problem solving, ataxia, apraxia, mental speed, reasoning and learning. Studies of language impairment were not included. People with language problems have difficulty completing cognitive assessments and are usually excluded from such studies (El Hachioui et al., 2013). Therefore, it was decided to not include papers that reported only patients with language impairment; (6) assessed cognition using an objective standardised quantitative measure. Studies using self-report assessments of cognition were not included as these have been shown to be more closely correlated with mood and confidence than cognitive function (Payne and Schnapp, 2014); (7) examined the relation between fatigue as defined and specified in (3) and (4) above, and cognition, as defined and specified in (5) and (6) above. There were no restrictions regarding publication time or language. Translations were obtained for studies in languages other than English.

Exclusion criteria
Systematic, narrative and literature reviews were not included.

Data extraction and quality assessment
For each of the studies meeting the above criteria, one of the authors (CL) conducted the data extraction and the quality assessment, while the other (NL) checked the details. The extraction was concerned with: (1) participants' characteristics (age, gender, type of stroke and time since stroke onset); (2) fatigue assessment methods; (3) cognitive assessment methods; (4) study design (settings and procedure); and (5) results (any association reported between fatigue and cognitive impairment). Quality assessment was conducted with the Critical Appraisal Skills Programme Checklists (CASP) quality assessment tool for cohort studies (CASP, 2014). CASP consists of 12 items that examine the context of the study results in relation to their validity, content and scientific contribution. It was chosen because it is the only tool available in a version for reviewing cohort studies as opposed to other commonly used lists. Each question was answered "yes", "no" or "can't tell". In order to provide a quality index, a score of 1 was given for every "yes" answer and 0 for every "no" or "can't tell" answer. For the items that are open-ended (7,8,12) a point was given if the answer was in favour of the study. The results of the CASP scoring were categorised as following: 0-5 = poor quality; 6-10 = average quality; 11-16 = high quality.

Data synthesis and correlation classification
The correlation coefficients between fatigue and cognition were summarised. Correlations were considered statistically significant at the 5% level of significance. The classification of the strength of correlation varied. It is generally suggested that correlations from .1 to .3 are considered weak, from .4 to .6 moderate, .7 to .9 strong and 1 is a perfect correlation (Hatcher, 2003).     another (Kutlubaev et al., 2013) was identified in the reference list of a narrative review; they were both included. These papers are summarised in Table 2.
Conducting a search on Web of Science of the citation indexes of these 11 papers identified a further 149 papers which were also assessed against the inclusion criteria. There were no duplicates among these 149 papers, however there were some duplicates of papers already identified in the initial search (n = 57), which were removed. Of the remaining 92 papers, none met the criteria and were excluded for either not having stroke patients (n = 12), not using the term fatigue (n = 16), not assessing fatigue (n = 25), not including any cognitive domain (n = 26), not assessing cognition (n = 0) or not assessing the relationship between fatigue and cognition   14 Naess, Beiske, and Myhr (2008) ✓  (n = 13). Table 3 summarises all papers and the reason of their exclusion. Figure 1 outlines the study selection process. In total, 11 studies fulfilled the inclusion criteria. Tables 4-7 summarise these studies.

Demographic characteristics
Overall, the studies provided data on 1597 participants post-stroke, 982 men (62%) and 615 women (38%). Generally, the studies had a small sample and only two (Tang, Lu et al., 2010;Van Eijsden et al., 2012)    reported missing data for 852 participants (57%). Two participants had suffered an SAH (0.1%). Five hundred and thirty eight (36%) of the participants had a major comorbidity, as reported by five studies Naess et al., 2005;Naess & Nyland, 2013;Park, Chun et al., 2009;Schepers et al., 2006). Four studies mentioned hypertension, hypertonia, diabetes mellitus and cardiovascular diseases, migraine and depression, whereas one did not specify these. Seven studies did not report comorbidities. Sample sizes ranged from 24 to 458. The mean age in studies ranged from 47.2 (SD = 8.3) to 66.2 (SD = 11.7) years (median values ranged from 47.8 to 70.5). The time since stroke onset at recruitment ranged from 7 days to 18 years, with one study (Hubacher et al., 2012) not reporting this information.
One study (Tang, Lu et al., 2010) reported a median of 5 years (range 0-25) and another ) a mean of 14.8 years (SD = 2.8) spent in education, and information from three other studies (Hubacher et al., 2012;Naess & Nyland, 2013;) indicated that 150 participants had attended higher, 16 secondary and 15 primary education. Five studies did not provide relevant information on education. The proportion of married participants was reported in five studies and ranged from 27 to 83%. Six studies did not report

Settings, design and quality assessment
Time since stroke onset at baseline assessment ranged from 7 days to 18 years. Six studies had a follow-up in their design and two did not (Park, Chun et al., 2009;Tang, Lu et al., 2010), however, one of the longitudinal studies (Naess & Nyland, 2013) only recorded mortality at follow-up, with no data available for fatigue at that time. Time of follow-up ranged from 2 weeks to 1 year post-stroke. Seven studies were single centre and four studies were multicentre. Two studies were conducted in Asia (Park, Chun et al., 2009;Tang, Lu et al., 2010) and the rest were conducted in Western Europe. Ten studies were considered to be of high quality according to the CASP list (Table 3).

Fatigue
Seven studies used one assessment of fatigue and three used more than one instrument (Hubacher et al., 2012;Tang, Lu et al., 2010;. Eight studies used the Fatigue Severity Scale (fatigue severity; FSS) (Krupp, LaRocca, Muir-Nash, & Steinberg, 1989); the remaining studies each used a different scale (see Table 6).
In the eight studies using the FSS, the mean fatigue score ranged from 3.1 (SD = 1.   ) measured cognitive fatigue whereas eight only reported overall fatigue. Three of the six longitudinal studies reported that fatigue remained stable over time whereas three studies (Hubacher et al., 2012;Schepers et al., 2006;) reported that fatigue increased over time (Table 6). According to cut-off scores, 900 (57%) participants were not fatigued and 697 (44%) participants were fatigued.

Cognitive impairment
Seven studies used the Mini Mental State Examination (MMSE; Folstein, Folstein, & McHugh, 1975) (Naess & Nyland, 2013;Park, Chun et al., 2009;Tang, Lu et al., 2010;Van Eijsden et al., 2012). Three studies used the Trail Making Test-Parts A and B (mental flexibility, visual search, speed of information processing and executive functions) (Army Individual Test Battery, 1944) Schepers et al., 2006; to assess executive function. The rest are summarised in Table 7. The mean score on the MMSE across three studies (Naess & Nyland, 2013;Park, Chun et al., 2009;Van Eijsden et al., 2012) ranged from 28.0 (SD = 1.7) to 28.2 (SD = 2.1). Three studies (Kutlubaev et al., 2013;Schepers et al., 2006;Tang, Lu et al., 2010) reported that median on the MMSE ranged from 27 to 28. One study (Naess et al., 2005) reported that the majority of their participants scored ≥ 28 on the MMSE. The mean scores in the studies that used Trail Making Test ranged from 40.4 to 62.8 for Task A and from 83.7 to 153.9 for Task B with median of 123 for task B. All scores are presented in Table 7.
Correlations between fatigue and cognition after stroke Table 8 summarises the correlations between measures of fatigue and measures of cognitive function. Seven studies (Kutlubaev et al., 2013;Naess et al., 2005;Naess & Nyland, 2013;Park, Chun et al., 2009;Schepers et al., 2006;Van Eijsden et al., 2012; found no significant correlation (p < .05) between cognitive variables and fatigue whereas four found significant correlations (Hubacher et al., 2012;Radman et al., 2012;Tang, Lu et al., 2010). The correlation between cognition and fatigue after stroke ranged from r = −.36 to .54 (Table 8) Of the 77 correlation coefficients calculated, 21 were significant and 56 were not. Overall, the above coefficients suggest that there is an association between concentration, sustained attention, speed of information processing, memory retrieval and verbal fluency and fatigue after stroke, but not with global cognitive impairment.

Discussion
The majority of the correlations between cognition and fatigue after stroke were not significant (p > .05). Only four of the 11 studies revealed significant correlations between fatigue and divided attention, sustained attention, speed of information processing, long-term memory and concentration. The studies revealed either non-significant or weak to moderate correlations. Non-significant or weak significant correlations between cognition and fatigue could potentially be attributed to participants in the sample having low levels of fatigue or minimal cognitive impairment.
With regard to fatigue, the majority of the studies used the FSS to measure fatigue. The mean fatigue score across the studies was 4.0 on the FSS, indicating that fatigue was low in their samples. Four out of eight studies reported a mean FSS of 4.0 or slightly above, all of which were only just above the cut-off (e.g., 4.1). The Fatigue Assessment Instrument (FAI) classifies severe fatigue as any score above 4 (Radman et al., 2012). The studies that used FAI reported a mean of 3.2, also a low level of fatigue.
The majority of the studies used the MMSE (Folstein et al., 1975). Of these, three reported a mean score of around 28, and the rest (three) reported a median of 27 or 28, which shows that the majority of the participants did not have dementia. That indicates that the majority of participants (88%) were not considered to be cognitively impaired on the MMSE. The MMSE is a screening assessment of global cognitive status (Folstein et al., 1975) and does not assess adequately executive functions, visuospatial functions and attention (Radman et al., 2012;Woodford & George, 2007). It is therefore not suitable for assessing cognitive impairment after stroke, which frequently affects these functions (Cumming, Marshall, & Lazar, 2013). It is also well recognised that the MMSE is not sensitive to cognitive impairment after stroke (Blake, McKinney, Treece, Lee, & Lincoln, 2002;Nys et al., 2005) and many of the participants in the studies that used the MMSE may have been misclassified as not impaired. Therefore the majority of the studies used an insensitive test of cognition and this may compromise the assessment of the relationship between cognition and fatigue. The results are similar to these of another systematic review (Ponchel et al., 2015) which examined the effect of cognitive disorders on post-stroke fatigue.
Of the 21 significant correlation coefficients (plus unspecified number of correlations from Radman et al., 2012), 18 came from the same study (Hubacher et al., 2012) which used the FSMC and the MFIS that consider cognitive manifestations of fatigue. These scales may be better at assessing cognitive aspects of fatigue because they were developed with this aim. Therefore such scales are more likely to correlate significantly with cognitive impairment. For instance, the MFIS includes items such as: "I have been forgetful" (targeting memory), "I had trouble concentrating" (which targets attention), and "My thinking has been slowed down" (speed of information processing), and the FSMC items such as: "My powers of concentration decrease considerably when I'm under stress" (attention) and "During episodes of exhaustion, I am noticeably more forgetful" (memory). When the scores are separated for cognitive and motor sub-scales within the scales (Table 8) of the 18 significant correlations, 8 are attributed to the cognitive components of fatigue being correlated with cognitive impairment. More general fatigue scales such as the FSS, did not reveal such relationships. This could be due either to the cognitive measures used (mostly the MMSE in these studies) or to the fact that the FSS does not measure cognitive components of fatigue. Fatigue scales that consider cognitive symptoms may reflect subjective cognitive complaints rather than fatigue. Only two studies (Radman et al., 2012;Tang, Lu et al., 2010) revealed a significant association between fatigue and cognition without measuring the cognitive aspects of fatigue (the former used the FAI and the latter the SF-36: vitality).
The results seem to be in accordance with studies in other medical conditions. For instance, studies in cancer patients undergoing chemotherapy (Castellon et al., 2004;Tchen et al., 2003;Vardy, 2008), HIV/AIDS patients (Millikin, Rourke, Halman, & Power, 2002), multiple sclerosis patients (Jougleux-Vie et al., 2014;Kinsinger, Lattie, & Mohr, 2010;Middleton, Denney, Lynch, & Parmenter, 2006), and patients with traumatic brain injury, (Johansson, Berglund, & Rönnbäck, 2009) have reported that subjective mental fatigue was associated with subjective cognitive performance but not with objective cognitive performance. The majority of the significant correlations in this review were from a study that assessed mental fatigue as a cognitive complaint (Hubacher et al., 2012). All these indicate that there may be a significant relationship between subjective cognitive impairment and fatigue, but not with cognitive ability. Therefore, it is essential to assess general fatigue as well as perceived cognitive fatigue and to compare their relationship with cognitive impairment.
Most studies used more than one measures of fatigue or cognition and produced inconsistent results. For example, Winkens, Van Heugten, Fasotti, et al. (2009) used both TMT-A and B and the Paced Auditory Serial Addition Test (PASAT) to measure speed of information processing, and both PASAT and the Symbol Digits Modalities Test (SDMT) to measure working memory, speed of information processing and sustained attention. None of the above tests was significantly correlated with the Mental Slowness Questionnaire (MSQ), as a measure of fatigue.
The inconsistency of the measures makes the interpretation of results difficult. Hubacher et al. (2012) used three scales of fatigue and correlated each of them with all eight subtests of the Brief Repeatable Battery of Neuropsychological Tests (BRB-N) battery. Some of the correlations were significant while others were not. Sustained attention was measured by SDMT and PASAT. The latter was significantly correlated with the FSMC, however, when sustained attention was measured with SDMT, it was not significantly associated with any of the fatigue scales. The authors acknowledged that the PASAT is also a measure of working memory and SDMT a measure of mental speed. It is therefore unclear which components were, or were not, associated in each correlation. It is difficult to conclude whether cognitive domains were significantly associated with fatigue, or whether the lack of significant relationship was due to the measures used. There is a need for more studies with more appropriate measures for fatigue and cognition.
It appears that fatigue scales that measure general fatigue symptoms rather than cognitive subcomponents, would be more appropriate in measuring fatigue to assess evidence of an association with cognitive impairment. The majority of the studies did not include cognitive tests because their objective was not to measure and associate cognitive impairment with fatigue. They mostly used global cognitive status assessments such as the MMSE as screening measures for participant inclusion/exclusion to the study. Therefore it would appear that tests that are designed to address specific cognitive impairments would be more appropriate. A combination of a general fatigue scale with a domain-specific cognitive measure is more likely to assess accurately the relationship between fatigue and cognitive impairment after stroke.
Most of the studies were of good quality according to the CASP guidelines, however, one did not provide correlation coefficient values (Radman et al., 2012) and another (Van Eijsden et al., 2012) was considered of moderate quality because there was not sufficient information nor justification as to why they excluded individuals with limited mobility, and the significance level was .2, which is higher than the conventionally used level of .05. Some studies did not report some information (see Table 4). After establishing contact with authors of the two papers (Radman et al., 2012;Van Eijsden et al., 2012), they were unable to provide the correlation coefficients.
It is possible that no significant associations were found because there is no significant relationship between cognition and fatigue post-stroke. However, there are other possible explanations for why the studies did not reveal significant associations between fatigue and cognitive impairment after stroke. For instance, 538 (36%) of the participants had major comorbidities (such as diabetes mellitus, cardiac disease and hypertension), which could significantly affect the experience of fatigue. Fatigue is very common in cardiac disease (Casillas, Damak, Chauvet-Gelinier, Deley, & Ornetti, 2006), but cognitive impairment is not. If, for instance, participants' fatigue was due to the comorbid condition and not the stroke, yet their cognitive impairment was attributed to their stroke, then this would mask any association between cognitive impairment and fatigue.
Another consideration is that some studies assessed fatigue within a month of the acute phase (Radman et al., 2012;Schepers et al., 2006), whereas others assessed it within the chronic phase Naess et al., 2005;Naess & Nyland, 2013;. However, there is conflicting evidence with regard to time of fatigue onset. Some studies reported fatigue being related to the acute phase (Choi-Kwon et al., 2005;Christensen et al., 2008), while other studies have argued that fatigue is a long-term issue. For instance, Schepers et al. (2006) reported that fatigue tends to increase over time, while Van de Port Kwakkel, Schepers, et al. (2007) found that fatigue peaks at approximately 12 months post-stroke. The results of the systematic review did not reveal any specific pattern. Fatigue was similar across studies and scores did not differ significantly according to time of assessment. The three highest scores according to FSS (4.7; 4.1 and 4.0) were spread out across the acute (4.7), middle (4.1) and chronic phase of stroke (4.0). Lower scores were also found in the acute, middle and chronic phases. When comparing data between baseline and follow-up assessments within the same longitudinal studies (Hubacher et al., 2012;Radman et al., 2012;Schepers et al., 2006;Van Eijsden et al., 2012;), a pattern emerges suggesting that fatigue tends to increase to some extent over time. Cognitive impairment tends to decrease over time (Danovska, Stamenov, Alexandrova, & Peychinska, 2012) and so any relationship between the two would be expected to be negative. Given that the time of onset and peak of fatigue is controversial, the fact that the above studies considered different time frames does not allow conclusive interpretations.
Overall, the papers had limitations. Most of them did not make it clear which type of correlation they used and therefore it is not possible to assess whether the statistical analyses followed were appropriate (parametric vs. non-parametric). The majority of the studies had very small samples with only two having more than 200 individuals (Tang, Lu et al., 2010;Van Eijsden et al., 2012).
Despite the exclusion of papers on people with SAH, one paper (Park, Chun et al., 2009) was included because it indicated there were only a few participants with the condition. The authors were contacted regarding the percentage of the participants with SAH but there was no response. The paper was included on the basis that SAH is far less common than other stroke types and on the assumption that the proportion of people with SAH would be low. Two studies (Naess et al., 2005;Naess & Nyland, 2013) mentioned an age range between 15 and 49 years. The review was focused on papers reporting on adult samples. After contacting the authors, only one participant was 17 years old at the baseline and 30 at the follow-up, and therefore these papers were included. This systematic review had limitations. It did not include studies that analysed their data with logistic regression. The inclusion criterion for the fatigue scales was that the scale had at least three response categories, which means that studies that used no/yes for the presence of fatigue were excluded. This means that the review may have missed studies with information on the presence of fatigue rather than severity and its association to cognitive impairment. However, only one study with such analysis was detected (Appelros, 2006) and the relationship between fatigue (yes/no) and cognition (MMSE) was not significant (p = .16) which is consistent with the results the majority of the studies included.
Another potential limitation is that more gerontological, medical and social sciences orientated databases, such as AgeLine and Science Direct, would have identified more papers. However, it is likely that the databases searched identified the majority of studies on the topic because they provided results drawn from medical, psychological and nursing journals. Another limitation is that, due to the nature of their search engines, some dissertation databases could not be searched in a systematic way identical to the one used in this review. The searching strategy included two thesis databases; but these would not include all dissertations (one is only for the UK and the other only for Europe). Two studies were identified either by chance  or through the reference list of other papers (Naess et al., 2005). Therefore it is possible that other suitable studies were not included.
Despite no language restriction, the use of English terms in the search strategies limited the results to papers that had an abstract and key words in English. Therefore, some papers in other languages may have been missed.

Author's conclusion
The findings of this review demonstrated that there was no evidence of a significant association between fatigue and global cognitive status after stroke. However, there was some evidence to suggest that attention, memory and speed of information processing may be significantly associated with fatigue.
Future studies should incorporate the use of both domain-specific and global cognitive tests, and investigate the association with general fatigue and cognitive fatigue scales. Furthermore, future studies could also include both subjective ratings of cognition and objective cognitive tests and examine their relationship to both general and cognitive fatigue scales. In summary, more studies are needed with measures that are sensitive to cognitive impairment after stroke and with fatigue scales that do not address cognitive components of fatigue. This will enable a more accurate investigation of the relationship between fatigue and cognitive impairment after stroke.