Variable definitions and categorizations
Sprague Dawley and Wistar were the only strains that were sufficiently well-used (at least 5 group pairs) to deserve separate categories. All other registered strains (Lister hooded rats, Diabetes type 1 rats, Spontaneously hypertensive rats – stroke prone, Spontaneously hypertensive rats and rats of unknown strain) were put in an Other strains category.
The category Males in the Sex variable was let to include both castrated and intact male animals, since castrated males were used in too few studies (n = 3) to deserve a separate category.
Too few studies used diseased animals for the Disease variable to be meaningfully included in the meta-analysis. The variable was therefore omitted from the statistical analysis, but is for the sake of transparency presented in Figure 4.
Elderly rats were defined as being at least 9 months of age at time of ischemic insult.
The continuous variable Number of rats in estrogen treated and control groups was calculated by simply summarizing the number of animals included in the infarct size measurements in each group pair.
Slow-release pellets from the company Innovative Research of America®, various injections and subcutaneous silastic capsules defined the three Estrogen administration mode categories. The Injection category consisted mainly of subcutaneous treatments, but a few group pairs included in this category were treated with intramuscular, intravenous and intraperitoneal injections. Since the other routes than subcutaneous were each too small to define separate categories, they were all included in the injection category. It can be argued that the pharmacokinetic differences between the injection regimens are too large to justify grouping them; however, they do share the important common characteristic of a high plasma concentration peak with short duration, which was why it was still deemed a relevant category . Only one group pair was treated with 17β-estradiol orally, which was far too few for meaningful analysis. However, since no other “superfluous” categories in this variable existed, the single study was put in a separate Oral treatment category.
The variable Estrogen type was omitted since only five groups were treated with other estrogen forms than 17β-estradiol. Premarin was used in two goups, estradiol valerate in one group and estrone in two groups. The variable is nevertheless presented in Figure 4.
The three variables Slow-release pellets: estrogen dose/pellet, Injections: Daily estrogen dose and Silastic capsules: estrogen dose/silastic capsule presented in Table 1 were not included in the main multiple regression model, since each of these variables was only relevant for a limited number of group pairs. These categories were instead used in subsequent simple linear regression models.
To deal with the fact that male rats are not ovariectomized, and that no fields in the multiple regression analysis are allowed to be empty, the variable Washout (Length of time between ovariectomy and estrogen administration) was first separated into two categories, 0–14 days and >14 days, and then combined with the variable Sex, making Males a third category.
In the variable Type of middle cerebral artery occlusion procedure, the category Direct, mechanical refers to all MCAo procedures where the MCA was mechanically occluded from the outside, for example by clips, cauterization or ligation. The categories Emboli and Photothrombosis included only 3 and 1 group pairs, but were treated according to the same principles as the Oral treatment category above.
Regarding the Occlusion duration variable, only methods including actions taken to ensure reperfusion (such as removing the occluding intraluminal filament or arterial clip) were considered transient.
Regarding the category Laser-Doppler flowmetry during surgery, group pairs were put in the [Yes]-category irrespective of whether or not it was explicitly described that the laser-Doppler was used to exclude animals.
Edema correction when calculating infarcts sizes can be performed according to at least two principles. Swanson et al.  described a procedure that focuses on the loss of viable tissue instead of on the infarct area per se:
This method assumes that all edema is in the infarct, and not outside it, and defined the category Correcting for edema in infarct. An alternative is to express the infarction as a percentage of the ipsilateral hemisphere:
The assumption here is instead that the edema is equally distributed in the entire infarcted hemisphere, and group pairs in which this procedure had been adopted were registered in the category Correcting for edema in entire hemisphere.
The outcome variable Infarct size ratio between estrogen treated and control rats (in the text referred to as “EC-ratio”) was calculated by simply dividing the mean infarct volume in the estrogen treated group by the mean infarct volume in the control group, and then multiplying by 100 to yield a percentage. Hence, a percentage above 100% means that the estrogen group on average suffered from larger (nominally; whether significant or not) infarctions than the control group, while numbers below 100% indicates the opposite. For example, a group pair in which estrogen treatment halved the infarct size would obtain an EC-ratio of 50%.
To identify which methodological factors significantly affected the EC-ratio, multiple linear regression analysis was used. As abovementioned, this analysis was the most important improvement from our previous systematic analysis . While the different methodological variables were analyzed separately in the previous article, the current meta-analysis combined them all in a large multiple regression model. In this main model, EC-ratio was the outcome variable, Number of rats in estrogen treated and control groups attributed each group pair a weight in the analysis, and all other extracted variables (except the aforementioned excluded ones, and the administration mode-specific dose variables) were considered independent factors. All category variables were dummy-converted before analysis. The main multiple regression analysis consisted of two steps. First, a multiple regression model with Backward exclusion of variables (p-value set to 0.10 for exclusion) was performed to eliminate independent factors that did not significantly contribute to the model (the variables excluded due to too low explanatory value are listed in the bottom of Table 2). Subsequently, the final multiple regression model, with an Enter procedure, including the significantly affecting variables from the previous model (together with lacking dummy variables so that all categories within a certain variable were included), was run. The results from the final multiple regression analysis is presented in Figure 2 and Table 2.Also, group pairs in which slow-release pellets, injections and silastic capsules had been used, respectively, were separately analyzed for association between dosage and EC-ratio. Even if three further large multiple models, as the main model described above, would theoretically have been preferable to control for confounders, the number of groups pairs pertaining to any of the three administration mode categories was too low (in the range of 30–50) to sustain an adequate statistical power. Instead, the relation between dosage and EC-ratio was analyzed with simple linear regression models. The results from these are presented in Figure 3. Three group pairs received silastic capsules containing crystallized 17β-estradiol, and were excluded from this analysis due to difficulty in translating to dissolved concentrations. In two of these three group pairs, estrogens were protective, while no difference was seen in the third group pair.
All statistical calculations were performed in SPSS (Version 20, IBM Corporation, Armonk, NY, USA). P-values <0.05 were considered statistically significant.