Transcranial magnetic stimulation as a biomarker for epilepsy

Sir, We read with interest the report by Badawy et al. (2013a) highlighting changes in cortical excitability in subjects with epilepsy and their siblings. It adds to the body of work of this group consistently showing that cortical excitability, measured by transcranial magnetic stimulation (TMS) may have potential as an epilepsy biomarker (Badawy and Jackson, 2012; Badawy et al., 2013b, 2014, 2015). Responses to TMS were shown to have a relatively large interindividual variability (see for example Valls-Solé et al., 1992; Du et al., 2014). Badawy et al.rsquo;s. reports do not provide clear information on interindividual variability. To estimate the variability across these reports, we extracted the data from some of their figures and compared these. Several questions arose: Re-digitalized long-interval intracortical inhibition (LICI) recovery curves. (A) Juvenile myoclonic epilepsy groups from two different studies with different sample sizes (shown in brackets). The curves overlap completely. (B) New onset temporal lobe epilepsy groups from two different studies with different sample sizes (shown in brackets). The curves overlap completely. (C) Idiopathic generalized epilepsy groups from three different studies with different sample sizes (shown in brackets). The curves overlap completely. (D) Non-epilepsy control groups shown in Figure 1 and Figure 2 in Badawy et al. (2013a). The article text indicates that both curves are obtained from the same control group, yet they show a different pattern. Were the results of the epilepsy groups reported by the authors obtained from overlapping subject groups? Given the large interindividual variability, we would not expect such similar curves for groups with such a different sample size (n = 7 and n = 20) and different pathologies (generalized epilepsy with tonic-clonic seizures only, and generalized epilepsy with tonic-clonic, myoclonic and/or absence seizures). (2) Within the report, there is a difference between the LICI curve of non-epilepsy controls shown in Fig. 1 (the red line) and the LICI curve of the non-epilepsy controls shown in the other figures as a grey shaded area (Badawy et al., 2013a). In the middle panel of their Fig. 2, (‘Refractory’), the grey shaded area appears to be shifted down, below the x-axis in the first figure on the left (JME). In their Fig. 1, the response ratio at an inter-stimulus interval of 200 ms appears to be greater than 100%, but in Fig. 2, the curve does not seem to reach 100% (Fig. 1D). Do those figures show data from the same group of control subjects? (3) The effect sizes reported do not seem to fit with the error bars shown in the figures, leading us to wonder what the ‘error bars’ in the figures represent. For example in their Fig. 2, second frame from the top, first on the left (juvenile myoclonic epilepsy drug naïve new-onset) (Badawy et al., 2013a). The accompanying text on p.1182 says: ‘In the drug naïve new-onset groups, cortical excitability was higher in patients compared with their siblings at the 150, 250 and 300 ms interstimulus intervals (P < 0.01, effect sizes ranging 0.5–0.7; maximum in juvenile myoclonic epilepsy). […] (Fig. 2)’. The effect size was calculated as: (mean of epilepsy) – (mean of siblings) / standard deviation of siblings (p.1181). The sample size of the group of siblings of people with new-onset juvenile myoclonic epilepsy was 11. For the 150 ms interstimulus interval, based on the effect size of 0.5–0.7, the standard deviation (SD) should be between ±164 and ±230%, and the standard error of the mean (SEM) between ±49 and ±69%. The ‘error bar’ in the figure shows ±25%; For the 250 ms interstimulus interval, based on the effect size of 0.5–0.7, the SD should be between ±250 and ±350%, and the SEM between ±75 and ±105%. The ‘error bar’ in the figure shows ±30%; For the 300 ms interstimulus interval, based on the effect size of 0.5–0.7, the SD should be between ±192 and ±270%, and the SEM between ±58 and ±81%. The ‘error bar’ in the figure shows ±40% (4) Lastly, the mean resting motor threshold of the control group reported 55.4 ± 5.7% (Badawy et al., 2013a), is similar to the motor threshold of 55.2 ± 5.6% repeatedly reported by the same authors (Table 1, Badawy et al., 2012). Were these results obtained from the same groups of participants? Control group characteristics and resting motor threshold rMT = resting motor threshold. aGroup characteristics and rMT (mean ± SD) are the same as in publications 2015, 2013b and 2013c, while the age range differs slightly. bGroup characteristics and mean rMT are the same as in publications 2015, 2013b and 2013c, while the SD rMT differs slightly. cMean rMT is the same as in all the other publications, while the group characteristics and SD rMT differ. Control group characteristics and resting motor threshold rMT = resting motor threshold. aGroup characteristics and rMT (mean ± SD) are the same as in publications 2015, 2013b and 2013c, while the age range differs slightly. bGroup characteristics and mean rMT are the same as in publications 2015, 2013b and 2013c, while the SD rMT differs slightly. cMean rMT is the same as in all the other publications, while the group characteristics and SD rMT differ. It is essential to clarify these questions, as the promise of any clinical biomarker critically depends on it’s interindividual variability, which ultimately influences its specificity and sensitivity. P.R.B. and J.W.S. are based at NIHR University College London Hospitals Biomedical Research Centre. P.R.B. is supported by the Christelijke Vereniging voor de Verpleging van Lijders aan Epilepsie (Nederland). A.A.dG. received funding from the Dutch TWIN Foundation for Neuromodulation. J.W.S. receives research support from the Dr. Marvin Weil Epilepsy Research Fund, Eisai, NEF, GSK, WHO and EU FP7.