Gender differences in post-stroke aphasia rates are caused by age. A meta-analysis and database query

Background Studies have suggested that aphasia rates are different in men and women following stroke. One hypothesis says that men have more lateralized language function than women. Given unilateral stroke, this would lead to a prediction of men having higher aphasia rates than women. Another line of observations suggest that women are more severely affected by stroke, which could lead to a higher aphasia rate among women. An additional potential confounding variable could be age, given that women are typically older at the time of stroke. Methods & procedures This study consists of two parts. First, a meta-analysis of the available reports of aphasia rates in the two sexes was conducted. A comprehensive literature search yielded 25 studies with sufficient information about both aphasia and gender. These studies included a total of 48,362 stroke patients for which aphasia rates were calculated. Second, data were extracted from an American health database (with 1,967,038 stroke patients), in order to include age and stroke severity into a regression analysis of sex differences in aphasia rates. Outcomes & results Both analyses revealed significantly larger aphasia rates in women than in men (1.1–1.14 ratio). This speaks against the idea that men should be more lateralized in their language function. When age and stroke severity were included as covariates, sex failed to explain any aphasia rate sex difference above and beyond that which is explained by age differences at time of stroke.


Introduction
A stroke is a medical condition in which blood flow to the brain is restricted, due to occlusion (ischemic stroke) or haemorrhage (haemorrhagic stroke), resulting in cell death (WHO). In the US alone, approximately 800,000 people experience a stroke each year, according to the American Heart Association (Benjamin, et al., 2017). Stroke is the leading cause of motor and cognitive disability in western countries and aphasia, the inability to comprehend and formulate language because of brain damage, is one of the most common deficits after stroke.
A large variability in the reported frequency of aphasia can be found in the literature (e.g. Ellis, et al., 2018;Godefroy, et al., 2002;Inatomi, et al., 2008;Tsouli, et al., 2009), ranging from 15 to 68 percent of acute patients. A recent meta-analysis, however, concluded that aphasia is present in approximately 30 % of acute patients and 34 % in rehabilitation settings (Flowers, et al., 2016).
The variability of measured prevalence of aphasia has many causes. The method for aphasia identification differs between hospitals and countries and some sub-scores for aphasia in stroke scales have been found to be limited in their accuracy and reliability (Meyer & Lyden, 2009;Thommessen, et al., 2002). The meta-analysis also did not take gender differences into account.
Stroke affects the genders differently. Stroke is more common among men (Appelros, et al., 2009). The symptoms of stroke have also been found to differ somewhat between men and women. Women are often more severely affected overall, more often experience paralysis, impaired consciousness and altered mental status together with a generalized weakness, while men more often experience dysarthria, diplopia, sensory loss, ataxia and balancing problems (Berglund, et al., 2017). An association between pre-stroke dementia, which is more prevalent in women, and stroke severity has also been noted (Gall, et al., 2010).
Aphasia following stroke has also been reported to affect women to a larger degree than men (see Berglund, et al., 2017 for a review), although evidence has been conflicting (e.g. Bersano, et al., 2009;Miceli, et al., 1981;Pedersen, et al., 1995).
Gender differences in certain linguistic domains are also known to exist within the normal population, with differences in first language acquisition speed (Bleses, et al., 2008) and reading and writing abilities (Reilly, et al., in press) being the most consistent, favouring girls/women over boys/men. Differences in word use have also been documented (Schwartz, et al., 2013). The underlying causes for these differences are probably complex and research trying to tie them to brain structure and function has yielded inconsistent results (Wallentin, 2009).
Some studies have argued for the hypothesis that language is more bilaterally organised in the brains of women than in men (e.g. Baron-Cohen, et al., 2005;Hausmann, 2016;Kansaku & Kitazawa, 2001;Shaywitz, et al., 1995), although this is highly controversial (Hirnstein, et al., 2018;Sommer, et al., 2004;Wallentin, 2009). A difference in language lateralization would ultimately lead to a difference in aphasia following unilateral stroke. If men's language is more lateralized in the brain than women's, we would expect them to be more prone to aphasia following unilateral stroke and vice versa, if women have greater language lateralization than men, we would expect their language function to be more vulnerable to stroke.
In this paper I conduct a meta-analysis on aphasia prevalence given stroke for the two genders across published papers and compare it to data from a large American patient database

Meta-analysis methods
A pub-med search including the terms "stroke" AND "aphasia" AND "gender" generated 211 citations. References in review articles on stroke and aphasia were also investigated. A total of 419 titles were considered. 90 papers were selected for further inspection on the basis of their title and abstract.
Given that the analysis is based on fully anonymized and publicly available data, the study poses no ethical concerns.

Results of meta-analysis
The 25 studies included a total of 48,362 stroke patients (23,085 women, 25,297 men). Of these 13,398 (6,828 women, 6,570 men) were diagnosed with aphasia (27.7%). 29.6 % of female stroke patients were diagnosed, while 26 % of males were diagnosed with aphasia (see table 1 and figure 1). This difference was found to be statistically significant using a paired and weighted t-test on the aphasia rates across studies, weighted to add emphasis on studies with more patients, t(27)=6.76, p<0.001, forcing a rejection of the null-hypothesis that there is no difference in aphasia rate between women and men. The overall gender aphasia rate ratio was found to be 1.14 (1.10-1.18 95% CI) with a Cohen's d of 0.37 which is usually considered a small effect (Cohen, 1992).  showing that across studies a small but significant effect of gender exists, indicating that women are more likely to get aphasia from stroke. This effect, however, does not take age or stroke severity into account.

Interim discussion
The aphasia rate across studies (27.7%) was comparable to that reported in a recent metaanalysis (30%) (Flowers, et al., 2016). The slightly lower estimate may in part be related to the inclusion of a study of cases with isolated aphasia (Wasserman, et al., 2015) which had a much smaller aphasia rate than studies with a regular aphasia diagnosis (see table 1). The aphasia rate ratio for the genders in this study, however, was comparable to that of most other studies in the sample.
A higher aphasia rate after stroke for women than for men was found across studies in the meta-analysis. This finding is at odds with the notion that language in men is more lateralized than in women (see introduction). If women have more lateralized language, one would expect their language to be more vulnerable to unilateral stroke than women's language.
But the findings are also at odds with previous critical suggestions that there are no gender differences in language lateralization between women and men (e.g. see Wallentin, 2009;Wallentin, et al., 2014). At face value, the findings would suggest that women in fact have more lateralized language than men. There are, however, reasons to be sceptical about such a conclusion, based on the present analysis.
As mentioned in the introduction, stroke is known to affect the genders differently on a number of accounts, including general severity. The genders also differ on general health levels, meaning that women on average are older when they are hit by stroke (Appelros, et al., 2009). Age has previously been found to be a predictor of experiencing aphasia (Ellis & Urban, 2016). In order to investigate if the gender effects found in the meta-analysis are specific to language or may relate to more general differences that are unlikely to be caused by a gender difference in language, an investigation of aphasia rates that take these considerations into account is needed. Unfortunately, very few studies in the current cohort make detailed, gender-stratified reports of age effects on aphasia. One exception is Bersano et al. (2009) who report aphasia rates for 4 different age groups. Here an interaction between age and gender differences seemingly can be observed. The gender difference is almost nonexisting in the youngest age group (under 64), but gradually grows larger and larger in older age groups. It thus seems that taking age into account is important when trying to understand the gender difference in aphasia rates.
Another possible explanation for the increased aphasia rate is that women are simply affected more severely by stroke in a non-discriminant manner (Berglund, et al., 2017). If aphasia rates can be explained by severity alone, it would again suggest that the gender difference is not restricted to language in any meaningful way. But again, this type of information is not reported in the papers included in the present meta-analysis.
I have therefore added a 2 nd dataset from an American healthcare database (see below) that will allow me to investigate aphasia rates while taking into account age and stroke severity. to study patterns and outcome of disease (O'Malley, et al., 2005). Data from each year for each state, stratified by gender, was used in the analysis. Data from this database has previously been used to study post-stroke aphasia rates (Ellis, et al., 2018), but here we add gender as an explanatory variable and incorporate all available states for all the years in which the ICD-9 diagnoses were used (i.e. 10 times more patients).
Hemiplegia and aphasia are comorbid deficits (Boehme, et al., 2016), but if a gender difference in number of aphasias is accompanied by a similar gender difference in hemiplegia diagnoses, then the difference is likely to be explained by stroke severity rather than being a specific language related phenomenon.
The database allows for two different ways to draw data. Either one can draw "Principal" diagnoses or "all-listed" diagnoses. As aphasia is often unlikely to be the principal diagnosis in a hospital visit, "all-listed" diagnoses were used. However, age information is only available with "principal" diagnoses, and age information was therefore drawn from this data.
The assumption is that age differences in principal diagnosis will be representative for age differences in the "all-listed" diagnoses as well.
To evaluate statistical significance of these findings, a linear mixed-effects regression analysis was conducted, fit by REML, using the lmertest package in R (Kuznetsova, et al., 2017). Pvalues were estimated using Satterthwaite's method. The model incorporated aphasia rate as the dependent variable and gender as the main fixed dependent variable. Age and rate of hemiplegia diagnoses (proxy for stroke severity) were z-score scaled and added as additional covariates. The model also included all possible interactions between the three variables. US state and year for each data-point were included as random effects. The regression was weighted by number of stroke cases in a particular state/year, to put more weight on datapoints from larger states.
Given that the analysis is based on fully anonymized and publicly available data, the study poses no ethical concerns. Using this method, 38.9 % of female stroke patients were diagnosed, while 36.2 % of males were diagnosed with aphasia (see figure 2).

Results of database analysis
The overall gender aphasia rate ratio was found to be 1.073 (1.068-1.079 95% CI) with a Cohen's d effect size across states of 0.63 which is usually considered a medium effect size (Cohen, 1992).
A paired t-test again yielded support to the existence of a gender difference, t(143)=-13.74, p<0.001.  This analysis replicates the findings from the meta-analysis and provides unequivocal evidence for a higher aphasia rate among women compared to men given stroke (see figure 1). However, as figure 3 shows, this effect can be explained completely by the gender difference in age at stroke.
When including age and stroke severity in a regression analysis, however, no significant effect of gender over and above that explained by age and severity could be observed,     (2011)(2012)(2013)(2014) in the HCUP database. The plot illustrates the large age difference between men and women in stroke diagnoses. It also shows a positive correlation between average age and aphasia rate, suggesting that older stroke patients more often get aphasia. When this relationship is taken into account, gender effects are no longer significant in the aphasia rates.

Discussion
On this very large cohort of patients, we replicate the findings from the meta-analysis. Based on raw aphasia rates, women are more likely to get an aphasia diagnosis following stroke than men. The sizes of the effects vary somewhat across the two analyses. This is likely due to the fact that the database data is more homogeneous than the data included in the meta-analysis, where the time of diagnosis post-stroke differs and also to some extent the definition of aphasia (e.g. one study includes dysphasia as well as aphasia and one study only looks at isolated aphasia cases -see table 1 for details).
At the same time, our 2 nd analysis completely revokes the conclusion that one could have been tempted to draw from the meta-analysis. In the database, we find no evidence for any gender difference in aphasia rates over and above that which can be explained by the age differences between the genders when they are affected by stroke. This replicates previous findings that age is a predictor of aphasia following stroke (Ellis & Urban, 2016), and given that age is a more fundamental causal variable than language (i.e. your language cannot change your age, but the opposite may be true), it is likely that most if not all of the gender difference in aphasia rates is caused by the age difference in stroke between women and men.
We also find an independent effect of stroke severity on aphasia rates as measured by diagnoses of hemiplegia. Aphasia and hemiplegia are known to be highly co-morbid. In this study we find that severity effects on aphasia are independent of the gender effects. The gender differences thus do not seem to be related to stroke severity per se. It has to be said, however, that this analysis uses a somewhat crude proxy for stroke severity. Other measures, such as general stroke scale scores (Hantson, et al., 1994;Lindenstrøm, et al., 1991) might interact more with gender.
Bersano et al. (Bersano, et al., 2009) found indications of increasing gender differences in aphasia rates with age. Contrary to this, the database analysis shows no indication of an interaction between gender and age. Bersano and co-workers did not report inferential statistics documenting an actual interaction, but looking at their data, the increasing discrepancy in aphasia between males and females with age is striking. For patients below 64 years the aphasia rate gender ratio is 1.04 and grows to 1.08 in 64-74 year old patients, (http://demo.istat.it/pop2002/index_e.html), one finds that because women live longer than men, the average age of women within the different age bins from middle age and onwards is higher than that of men and that this difference gets larger for the older groups. For the 54 to 64 year old Italians, the mean age difference between men and women is 0.07 years, but for the age group above 84 years, it has grown to 0.52 years. There is a very strong linear correlation between these mean age differences in the Italian population in these age bins and the reported differences in aphasia rate (r= 0.96, data available from author on request), which suggests that at least some of the interaction between gender and age is based on unequal sampling of the different ages. This relationship may explain some of the interaction observed in the Bersano et al. data. This is not to say that there could not be age effects that are not picked up by the current analysis. The data from the database is distributed on a state by year basis and each data-point for age is the result of averaging across many individual patients.
Underneath this gross simplification may be hidden lots of interesting phenomena. Further studies are needed in order to rule out a potential interaction between gender, age and aphasia.
The present analyses are also limited in that they say nothing about the different types of aphasia symptoms that patients may suffer from and the potential interactions that might be found with gender if one looks more carefully at aphasia subtypes.
Taken together, the results are in line with a critical stance towards the brain base of gender differences in language (Wallentin, 2009). This, of course, does not mean that the observed gender differences in language related behavior (see introduction) do not have brain correlates, just that these differences will be dynamic, complex and to a large extent dependent on gender differences in experience and context rather than being tied to genetic sex.

Conclusion
We have found that women more often are diagnosed with aphasia following stroke. This is in direct opposition to the hypothesis that women have less lateralized language function than men. The gender difference is most likely caused by age differences in the two groups at the time of stroke.

Disclosure statement
The author has no conflict of interests to report.