The Impact of Sparse Follow-up on Marginal Structural Models for Time-to-Event Data