Accounting for misclassification error in retrospective smoking data.