Electrographic predictors of successful weaning from anaesthetics in refractory status epilepticus

Abstract

Intravenous third-line anaesthetic agents are typically titrated in refractory status epilepticus to achieve either seizure suppression or burst suppression on continuous EEG. However, the optimum treatment paradigm is unknown and little data exist to guide the withdrawal of anaesthetics in refractory status epilepticus. Premature withdrawal of anaesthetics risks the recurrence of seizures, whereas the prolonged use of anaesthetics increases the risk of treatment-associated adverse effects. This study sought to measure the accuracy of features of EEG activity during anaesthetic weaning in refractory status epilepticus as predictors of successful weaning from intravenous anaesthetics. We prespecified a successful anaesthetic wean as the discontinuation of intravenous anaesthesia without developing recurrent status epilepticus, and a wean failure as either recurrent status epilepticus or the resumption of anaesthesia for the purpose of treating an EEG pattern concerning for incipient status epilepticus. We evaluated two types of features as predictors of successful weaning: spectral components of the EEG signal, and spatial-correlation-based measures of functional connectivity. The results of these analyses were used to train a classifier to predict wean outcome. Forty-seven consecutive anaesthetic weans (23 successes, 24 failures) were identified from a single-centre cohort of patients admitted with refractory status epilepticus from 2016 to 2019. Spectral components of the EEG revealed no significant differences between successful and unsuccessful weans. Analysis of functional connectivity measures revealed that successful anaesthetic weans were characterized by the emergence of larger, more densely connected, and more highly clustered spatial functional networks, yielding 75.5% (95% confidence interval: 73.1-77.8%) testing accuracy in a bootstrap analysis using a hold-out sample of 20% of data for testing and 74.6% (95% confidence interval 73.2-75.9%) testing accuracy in a secondary external validation cohort, with an area under the curve of 83.3%. Distinct signatures in the spatial networks of functional connectivity emerge during successful anaesthetic liberation in status epilepticus; these findings are absent in patients with anaesthetic wean failure. Identifying features that emerge during successful anaesthetic weaning may allow faster and more successful anaesthetic liberation after refractory status epilepticus.

Keywords: EEG; functional connectivity; status epilepticus.

Figure 1

Schematic depicting the calculation of functional connectivity from cEEG. Top: An example of 11 s of cEEG data from a patient undergoing an anaesthetic wean. Middle: Two pairs of 1-s long EEG tracings are highlighted. The cross correlations between pairs are shown. Bottom: Two examples of the functional connectivity maps are shown, with circles (the ‘nodes’ of the functional connectivity maps) representing each EEG lead and lines (the ‘edges’ of the functional connectivity maps) representing connections defined by a statistically significant peak in the cross-correlogram between the signals recorded from each lead.

Figure 2

Representative examples of the raw data from patients undergoing successful (left column, outcome = 1) and unsuccessful (right column, outcome = 0) anaesthetic weans. Top row: Intravenous anaesthetic infusion rate versus time. Time = 0 indicates the time of cessation of intravenous anaesthesia. Rows 2–5: Frequency-based quantitative metrics (relative alpha power, relative delta power, relative theta power, and alpha/delta ratio) versus time. Rows 6–13: Spatial-correlation-based quantitative metrics (network density, number of independent components, mean clustering coefficient, characteristic path length, number of non-trivial components, size of the largest independent component, clustering coefficient of the largest component, and characteristic path length of the largest component) versus time. In both the successful and unsuccessful cases, the alpha/delta ratio rises as intravenous anaesthesia is discontinued. In the successful but not the unsuccessful case, as intravenous anaesthesia is withdrawn, there is a gradual rise in network density, clustering coefficient, characteristic path length, and size of the largest component and fall in the number of independent components as network connectivity rises.

Figure 3

Comparison of the quantitative metrics between successful and unsuccessful anaesthetic weans. Values for each wean are calculated by averaging over the 30-min epoch ending at the time of anaesthesia cessation. Top row: There are no significant differences between the two groups in the any of the spectral power metrics. [Alpha: t(45) = 0.6619, P = 0.51144; Delta: t(45) = 0.2370, P = 0.8137; Theta: t(45) = 0.8312, P = 0.41026; ADR: t(45) = 0.3669, P = 0.71543; two-tailed Student’s t-test]. Bottom row: Compared to the unsuccessful anaesthetic weans, successful anaesthetic weans had significantly greater network density [t(45) = 3.2549, P = 0.0021576], characteristic path length [t(45) = 3.1617, P = 0.0028063], clustering coefficient [t(45) = 2.9903, P = 0.0045068], size of largest component [t(45) = 4.2437, P = 0.00010838], and characteristic path length of the largest component [t(45) = 3.0867, P = 0.0034591], and significantly fewer independent components [t(45) = 3.8467, P = 0.0003745]. There was no significant difference in the number of non-trivial components [t(45) = 0.4939, P = 0.62376] or the clustering coefficient of the largest component [t(45) = 1.2350, P = 0.22325]. Within each box plot, the horizontal red line indicates the median value, blue boxes indicate the interquartile range (IQR), and black bars indicate minimum and maximum values (excluding outliers, defined as either >1.5 times the IQR above the third quartile or 1.5 times the IQR below the first quartile, which are indicated by red plus symbol). All P-values are calculated by a two-tailed Student’s t-test; *statistical significance after accounting for multiple comparisons with α  =  0.05. ADR = alpha/delta ratio; CC = clustering coefficient; CPL = characteristic path length; CC_L and CPL_L = clustering coefficient and characteristic path length of the largest component, respectively; SCL = size of the largest component.

Figure 4

Network connectivity parameters versus time. The four spectral power and eight network connectivity parameters (network density, number of independent components, clustering coefficient, characteristic path length, number of non-trivial components, size of the largest component, clustering coefficient of the largest component, and characteristic path length of the largest component) are plotted for the successful (blue) and unsuccessful (red) weans. Lines indicate mean values, and shaded areas denote the standard error of the mean. The dashed vertical line at time = 0 indicates the time of anaesthesia cessation. EEG data were not available for every anaesthetic wean at all times (in a few cases cEEG was only started 2–3 h prior to anaesthetic weaning); the number of weans included versus time is shown in the inset. ADR = alpha/delta ratio; CC = clustering coefficient; CPL = characteristic path length; CC_L and CPL_L = clustering coefficient and characteristic path length of the largest component, respectively; SCL = size of the largest component.

Figure 5

Test of oversampling bias. The comparisons for each quantitative metric was repeated using a single wean from each of the 34 patients; 1152 unique combinations were considered. A histogram of the P-values of the differences between the two groups (using a two-sided Student’s t-test) is shown for each of the four spectral power metrics and each of the eight functional connectivity metrics. Orange bars denote P-values that are statistically significant (α  =  0.05) after accounting for multiple comparisons; blue bars tally the P-values that do not reach statistical significance. For the spectral power metrics, in none of the 1152 combinations are any statistically significant differences between the two groups observed. For the six functional connectivity metrics in which a statistically significant difference was observed in the initial analysis (Fig. 3), the majority of the 1152 comparisons yield statistically significant P-values. These six histograms skew strongly to the left, indicating that in cases that did not reach statistical significance after correcting for multiple comparisons there was still a trend towards a difference between the groups. ADR = alpha/delta ratio; CC = clustering coefficient; CPL = characteristic path length; CC_L and CPL_L = clustering coefficient and characteristic path length of the largest component, respectively; SCL = size of the largest component.

Figure 6

Accuracy of classifier tested by 100-fold cross validation. (A) For each of 100 iterations, the classifier was trained on data from a randomly selected 80% of patients, and then tested on the remaining 20% of the data. Data used for training and testing were calculated by averaging the quantitative metrics for each wean over the 30-min period ending at the time of anaesthesia cessation. Results from the true classifier model are shown on the left, and from a control model trained on randomly shuffled outcome labels on the right. Median accuracy from 100 iterations is indicated by the horizontal red line, blue boxes indicate the IQR of the distribution, and black bars indicate the minimum and maximum values (excluding outliers, defined as either >1.5 times the IQR above the third quartile or 1.5 times the IQR below the first quartile, which are indicated by red plus symbol). The classifier model trained on true outcome labels accurately predicted wean outcome 75.5% of the time, significantly greater than the control model [χ²(1) = 94.5114, P = 2.44 × 10⁻²², Kruskal-Wallis H-test]. (B) Again, for each of 100 iterations, the classifier was trained on weans from a randomly selected 80% of patients, and then tested on the remaining 20% of the data. Here the model was then applied to the full time series for the withheld 20% of data, yielding a prediction (success versus failure) as a function of time. The accuracy of the model at each time point was calculated for each iteration, and the mean ± standard error of the mean (SEM) of the accuracy is plotted as a function of time. The true classifier model is plotted in blue, and the control model is plotted in red. The two models begin to diverge at ∼16 h prior to wean cessation. The classifier is more accurate than the control up to 12 h prior to wean cessation and increases in accuracy over time.

Figure 7

Examples of the classifier applied to time series. For each of six example anaesthetic weans in patients with successful (wean outcome: 1) and unsuccessful anaesthetic liberation (wean outcome: 0): the top parameter is the anaesthetic dose over time. The second and third parameters display the alpha delta ratio and network density over time. The fourth parameter displays the prediction ‘score’, demonstrating the sign (positive, predictive of success, and negative, predictive of failure) and the confidence of the prediction proportionate to the magnitude of the prediction score. The final parameter demonstrates the time-varying prediction of the wean outcome [success (positive) versus failure (negative) based on the sign of the prediction score].