Estimation of intervention effects using first or multiple episodes in clinical trials: The Andersen-Gill model re-examined