134 Chapter 5 Power across the brain The power-problem in LSM studies is a heavily discussed topic.10,70 Although our patient samples of stroke (n = 147) and tumor (n = 196) meet the general recommendations of 100-130 for LSM analyses, power does not only depend on the number of participants. Lesion distribution and volume also have a large effect upon the overall power of LSM studies. A sample with an average lesion volume of 3 cm3 will cover substantially less voxels than a sample with an average lesion volume of 10 cm3. Hence, the number of voxels with sufficient power for lesion-symptom analysis will be lower in study samples with smaller lesion sizes. As a result, the number of areas in which we had sufficient overlap to directly compare the LSM between the study samples was only 38 out of the 124 atlas areas. Unfortunately, post-hoc power analyses for multivariate inference cannot be computed, complicating the investigation of the exact effect of power on the multivariate LSM results. However, power analyses for the univariate LSM suggested that also in these areas of overlap, the number of voxels with adequate power to allow for replication was only a quarter to one-third. Overall, the number of voxels with adequate power was higher in tumor than in stroke, especially at a whole-brain level. These power issues are not unique to our sample. The topography of damage and lesion volume of our stroke sample are comparable to the sample described by Corbetta and colleagues (n = 132),71–73 although they included both hemorrhagic and ischemic strokes and relatively small lesions compared to our study. The authors mention the relatively small number of overlapping cortical lesions as one of the main limitations of the study. Even in a large-scale, multi-cohort LSM study (n = 2950) with high lesion coverage (average 2.7 cm3), some brain areas still achieved a maximum lesion overlap of five patients74, demonstrating the uneven lesion distribution that is inherent to the stroke population. Clinical representativeness Our tumor sample included treatment-naïve diffuse glioma patients who underwent awake brain surgery. Consequently, the tumor population consisted of patients with different tumor grades and molecular markers. As edema, plasticity, lesion momentum and molecular markers are related to tumor grading, different tumor grades could be viewed as distinct (though related) etiologies. Additionally, previous research has indicated that certain molecular markers are independent determinants of cognitive functioning.28 Subgroup analysis stratified by tumor grade or IDH mutation, according to WHO 2021 classification75, would be an interesting follow-up study. Stroke populations could also be further subdivided according to time since stroke onset.76 Current subgroup sample sizes were too
RkJQdWJsaXNoZXIy MTk4NDMw