Abstracts

Rationale: In observational studies which objective is to compare therapeutic strategies, investigators have no control over the treatment assignment. Therefore, large differences could lead to biases estimates of treatment effects. Objective: In a multicenter cohort of intractable adult epileptic patients treated surgically or medically, we compared the seizure-free patients after surgery with those patients who were medically treated, using the propensity score method to reduce the potantial confounding effects caused by unbalanced baseline characteristics. Methods: Adult patients with partial medically intractable epilepsy, potentially operable were eligible and followed every year over five years. The clinical outcome was the proportion of seizure-free patients (1a, 1b of Engel's classification). A propensity score estimate was performed using a multivariate logistic regression. The data were stratified on sample quintiles of the propensity score distribution. To compare surgical versus medical groups, we presented the odds ratio and its 95% confidence interval (CI). Results: Among the 279 included, 120 were operated and 159 continued their medical treatment. Patients with a propensity score smaller than 0.05 were excluded. Initially, 11 covariates differed significantly between the two groups. After stratification on the propensity score, no variables remain statistically significant. At two years after the surgery, the proportion of seizure-free patients is 28% in the surgical group and 5.5% in the medical group. The crude odds ratio of seizure-free outcome for operated patients compared to medical patients was 15.7 (95% CI: 8.2, 30). Adjusting for the propensity score yielded estimated odds ratio of 16.7 (95% CI: 7.2, 37.8). Conclusions: Stratification on the estimated propensity score reduced the systematic baseline differences between the surgical and medical patients with partial medically intractable epilepsy. It results in a less biased estimate of the surgery effect. Restriction to persons whose propensity score are greater than or equal to 0.05 focuses on individuals who would have been eligible to participate in clinical trials. The odd ratio is slightly increased by the adjustment.