Brain meeting: Prof. Howard Bowman

This talk will report on two lines of work with Cathy Price’s PLORAS group. I will present simulations and some very simple maths that suggest that two existing approaches to inference and prediction after stroke are not as effective as might be believed.

1) Small Samples: the availability of a large data set, such as PLORAS, enables one to subsample to simulate small data sets and thereby, assess the variability in inference associated with such small samples. We have done this and the results suggest that in lesion-deficit mapping, the combination of small samples and the file drawer effect could have substantially inflated published effect sizes [Lorca-Puls et al; 2018].

2) Proportional Recovery: A number of recent papers have claimed that recovery from stroke is proportional to initial severity and can be predicted accurately from this initial behavioural measurement. We show that these high effect sizes are in most cases a mathematical quirk of correlating two coupled variables [Hope et al; 2018].

Overall, then, these two findings provide support for the large sample size, lesion-site approach employed in the PLORAS project. That is, the findings suggest that (1) large samples brings crucial benefits to the reliability of lesion-deficit mapping, and (2) simple routes to high predictive power (e.g. purely on the basis of initial severity) are unlikely to be true panaceas, indicating the importance of lesion site. 

There will be coffee, tea and cake in the conservatory directly after the talk.