Novel Model Predicts Patients With Epilepsy Who Fail to Achieve Remission Post-Breakthrough

March 27, 2019

The novel longitudinal model utilized 4 variables—the presence of seizures, the number of seizures, the number of adverse events, and if treatment was altered—and was ultimately more predictive than a standard Cox model.

Marta Garcia-Finana, PhD

Longitudinal data may be able to provide more accurate predictions in identifying patients with epilepsy who are most likely to achieve remission after a breakthrough seizure compared to the baseline covariates in a standard Cox model.1

New study data has revealed that by using a novel, longitudinal discriminant approach to the data from the SANAD study,2 73% of patients who failed to achieve a second remission were identified correctly (95% CI, 58 to 88) and 84% of patients who did achieve a second remission (95% CI, 69 to 96) after a breakthrough seizure were correctly identified.

Notably, patients who did not achieve a second remission were correctly identified on average after 10 months of observation post-breakthrough seizure.

In total, 536 patients experienced a breakthrough seizure. A number of patients were excluded for various reasons, and of the remaining 300 patients who experienced a breakthrough seizure, 185 patients (62%) went on to achieve a further period of 12‐month seizure remission within 2 years of experiencing their breakthrough seizure and 115 patients (38%) were observed for 2 years following breakthrough without experiencing a year-long remission.

“The model developed in this paper is a useful first step in developing a tool for identifying patients who develop drug resistance after an initial remission,” Marta García‐Fiñana, PhD, a professor of biostatistics at the University of Liverpool’s Institute of Translational Medicine, and colleagues, wrote. They noted that “the presence and number of post-breakthrough seizures are the most informative variables for predicting poor outcome following first breakthrough.”

García‐Fiñana told NeurologyLive that she and colleagues see their study as a good starting point for aiding clinical decision making. "However, we acknowledge that further research, including a larger dataset with longer follow up and external validation, needs to be conducted before our proposed model can be used to influence patient counseling and management decision," she said.

"The results were somehow surprising to us," García‐Fiñana continued. "Although we expected that a model based on longitudinal data would perform better than a model based on baseline covariates alone, we did not anticipate that the improvement would be so remarkable."

The investigators acknowledged that 4 variables were considered for the use in their model, recorded at each follow-up visit. The group defined a change in treatment as increases in the dose of a drug or the addition or removal of a drug. The variables chosen were as follows:

  1. Whether a patient had experienced seizures since their previous clinic visit.
  2. How many seizures were experienced since the previous clinic visit.
  3. The number of patient‐reported adverse events (AEs) experienced since the previous clinic visit (depression, dizziness, allergic reactions, headaches, and tiredness, among others).
  4. Whether a patient's treatment was changed at the last clinic visit.

Roughly 17% of patients (95% CI, 16 to 18) were left unclassified by the novel model, “as there was considerable uncertainty about their status and longer follow‐up would have been required,” Hughes and coauthors wrote. “A key point of this approach is that by leaving a relatively small proportion of patients unclassified, much greater predictive accuracy is obtained for patients who are classified.”

They additionally noted that if unclassified patients were considered as incorrectly classified, the predictive accuracy of the model indicates a sensitivity of 57% (95% CI, 41 to 71), specificity of 72% (95% CI, 59 to 8), and probability of correct classification of 66% (95% CI, 57 to 74).

The sensitivity, specificity, and probability of correct classification of the longitudinal analysis in comparison to the Cox model was 73% vs. 62%; 84% vs. 66%; and 80% vs. 64%, respectively. The area under the curve for the novel model was 87% compared to 66% with the cox model.

“Because we required patients to have experienced a 12‐month remission and a breakthrough seizure with an additional 2 years of follow‐up post-seizure, the sample size for our analysis is relatively small,” the authors detailed. “These factors potentially limit the power of our analysis.”

They concluded that this study and its predecessor3 provides “a useful tool to identify patients who are not likely to achieve remission from seizures early in their clinical follow‐up, both after diagnosis and after a breakthrough seizure following remission,” and that “incorporating these models into an easy to use calculator (possibly as a webtool or app) would be a necessary next step in making these models clinically useful.”

REFERENCE

1. Hughes DM, Bonnett LJ, Marson AG, García-Fiñana M. Identifying patients who will not reachieve remission after breakthrough seizures. Epilepsia. ePub March 22, 2019. doi: 10.1111/epi.14697.

2. Marson AG, Al-Kharusi AM, Alwaidh M. The SANAD study of effectiveness of valproate, lamotrigine, or topiramate for generalised and unclassifiable epilepsy: an unblinded randomised controlled trial. Lancet. 2007;369(9566):1016-26. doi: 10.1016/S0140-6736(07)60461-9.

3. Hughes DM, Bonnett LB, Czanner G, Komárek A, Marson AG, García-Fiñana M. Identification of patients who will not achieve seizure remission within 5 years on AEDs. Neurology. 2018;91(22):e2035-e2044.