Background The prognosis of seizure recurrence for intractable epileptic patients after anterior temporal lobectomy (ATL) is highly dependent on the frequencies and the duration of seizure, which have been never considered concomitantly. A novel classification using the Shih classification has been proposed recently to describe the severity of seizure recurrence after ATL for patients suffering from intractable epilepsy. However, it is still unclear of which prognostic factors are associated with the severity of seizure using the Shih classification. It is also lacking of statistical methods proposed to model dynamic changes of seizure recurrence. Aims The objectives of this thesis are (1) to identify prognostic factors of seizure remission and recurrence postoperatively, and to evaluate risk factors by using the Shih classification based on at least five-year follow-up. (2) to propose a combined model of higher-order and mover-stayer Markov model to elucidate the dynamic change of the state of attack of intractable epileptic patients. (3) to apply the semi-Markov model to characterize the duration and the frequencies of attack of intractable epileptic patients. Methods Data on 5-year longitudinal follow-up cohort for seizure after ATL from 280 patients between 1987 and 2002 were collected. All patients were categorized into the four Shih types (inactive, delayed, intermittent and intensive), reflecting the severity of seizure from mild state to severe state. Proportional odds logistic regression model was used to calculate the odds ratio (OR) of being different seizure types when the four Shih seizure types were treated as an ordinal property. Defined by the frequencies of monthly episodes, three states are classified, including normal, mild, and severe states. We proposed the three-state-based first-order and high-order Markov model and the mover-stayer model to study the dynamic changes of monthly episodes of seizure and also developed the semi-Markov model to elucidate the sojourn time distributions and the embed transition probabilities. Results The ordinal data property on the severity of seizure based on the Shih classification has been proven by testing the proportional odds assumption (P=0.75). The results show that main seizure type other than complex partial seizure (CPS) (aOR=3.18, 95% CI: 1.22-8.30), perinatal insult (aOR=7.06, 95% CI: 2.25-22.19), the presence of mesial temporal sclerosis (MTS) (aOR=0.50, 95% CI: 0.31-0.83), and multiple medications (aOR=1.79, 95% CI=1.27-2.52) were statistically significant. The results of the first-order Markov chain model shows that the stationary distribution was reached with 89% in the normal state, 7% in the mild state, and 4% in the severe state. Though the mover-stayer model behaved better than the first- and high-order Markov model in dealing with the heterogeneity. The mover-stayer model yielded 62% stayer. Of 38% movers, 89% staying in free of seizure, 7% in the mild state, and 4% in the severe state after long-term follow-up. With the semi-Markov model, we found the risk of leaving for other more severe states for those staying at state 1 decreased with time. In contrast, the risk of not staying at State 2 and State 3 increased with time. The median times of staying in State 1, 2, and 3 were estimates 10.59, 1.34, and 1.74 months, respectively. Perinatal insult was an accelerated factor for the progressive movements from normal to severe state but also a decelerated factor for regressive movements from severe to normal state. Similar findings were noted for main seizure type and sum of medication. On the other hand, MTS was found as a decelerated factor for the progressive movement but an accelerated factor for the regressive returns. Conclusion Beginning with a conventional logistical regression, we found main seizure type, perinatal insult, the presence of mesial temporal sclerosis, and pre-surgical multiple medications played an important role in the prognosis of attack after ATL. The mover-stayer model is feasible for describing the dynamic change of monthly episodes of seizure. The semi-Markov model suggested the longest mean sojourn time in seizure-free state. Perinatal insult, other seizure types, and sum of medication were accelerated factors for the progressive movements from normal to severe state but also a decelerated factor for regressive movements from severe to normal state by using the stochastic process. On the other hand, MTS was found as a decelerated factor for the progressive movement but an accelerated factor for the regressive returns.