Curve fitting for repeated measurements made at irregular time-points

In longitudinal studies of human populations, it is often not feasible to measure all subjects at the same time-points. This precludes the use of classical methods of curve fitting for repeated measurements. When the total interval of follow-up is short for all subjects in the study, an intraclass correlation matrix is assumed for the measurements on each subject. An estimation procedure based on iteratively reweighted least squares is described. The model is then generalized to incorporate covariables, with little modification in the estimation procedure. The proposed method is applied to data from a longitudinal study of bone mass in postmenopausal women.