Bridging the gap between analytical methods and their clinical interpretation

Our team took great interest in the recent paper entitled, “Temporal epilepsy in children: A connectomic analysis in magnetoencephalography” by Martire et al.1 In this study, the authors analyzed the data of 121 children with temporal (TL) or temporal plus epilepsy (TL+) using resting-state connectomes derived from their magnetoencephalograms. Using this connectivity profile, they found that patients with TL+ epilepsy could be distinguished from those with TL epilepsy alone. This is a relevant finding because it can be difficult to distinguish TL+ epilepsy in clinical practice and the presence of epileptogenic networks extending beyond the temporal lobe can be a cause of surgical failure in patients who undergo temporal resection.2 The study was backed by extensive analysis using the partial least squares method (PLS). Although the methodology used for PLS was fairly well explained, the practical interpretation of the same was difficult to fathom for clinicians who are typically not trained in methods of advanced multivariate analysis. Therefore, this posed a challenge to the comprehension as well as critical analysis of the methodology used in this study. As an example, the authors found that a single latent variable explained 66% of the variance in the data and identified significant contributions from the extent of epilepsy (TL vs TL+). However, we found it difficult to gauge the clinical interpretation of this variable and its relevance with respect to the results. On a similar note, we greatly looked forward to findings and discussion related to how extratemporal connectivity differed between TL and TL+ epilepsy. Such investigations can help us understand how resting-state networks differ fundamentally between TL and TL+ epilepsy and explore the relation of these differences with clinical outcomes. Although the authors did present the mean brain connectivity (limited to only the top 1% of the connections) in TL and TL+ epilepsy associated with the latent variable mentioned above (Figure 4), the interpretation of these patterns was not clearly discernible to the reader, and it was not covered comprehensively in the discussion section. Another relevant observation was the absence of measures of diagnostic performance such as sensitivity, specificity, predictive values, and so on. Even though the study concluded that resting-state connectomic analysis can distinguish between TL and TL+ epilepsy, the reader was left wondering about the margin of error and the accuracy of this method. Finally, with the advent of this century, studies featuring “big data,” whole brain connectivity and multivariate analysis have become common in all branches of neurology including epilepsy,3–8 and their applications are only expected to increase in near future. The purpose of this letter therefore is not only to critique this exemplary work undertaken by Martire et al, but also to generate a discussion regarding the up and coming use of advanced methods of analysis such as PLS, connectomics, and so on, and the best way to equip the clinicians with these new tools. To conclude, bridging the gap between the clinician and the analyst is going to be a relevant challenge for modern neurology and the study in question beautifully brings this to notice.