Fig. 3: Critical slowing down and seizure clusters.

a The synchronization indices (SIs) and mean phases at the sample prior to seizures for long (L) and short (Sh) cycles. A strong relationship between the three signals and seizures were observed as depicted by the high SI values. On average, seizures occurred on the rising phase of the signals. b The similarity in autocorrelation (ACFW), variance (Var), and spike rate (Sp) signals across electrodes were compared by computing the mean cross correlation of each signal and comparing across patients. There was a significant effect of signal type (F_2,13 = 58.61, p = 2 × 10⁻¹⁰), and a post-hoc analysis showed that the autocorrelation signal was significantly different to the variance (p = 9 × 10⁻³) and the spike rate signals (p =  1 × 10⁻⁹). In a and b, each dot represents a result for one patient (N = 14 patients); boxes indicate means across patients and lines indicate ±one standard deviation. In b, statistical comparisons were computed using a balanced two-way ANOVA corrected with a Tukey–Kramer multiple comparisons test. c Patient 1 had 69 seizure clusters. Gray lines shows the raw autocorrelation from individual seizures. d, e Autocorrelation d and variance e relative to the time from lead seizures for the remaining patients with seizure clusters. The numbers inset in each subplot denotes the number of lead seizures that occurred in clusters. In c–e, black lines denote the mean autocorrelation with standard error bars. Red lines denotes a moving sum of seizures occurring after the lead seizure computed over a 2 h window.