MODELLING PEAKFLOW TIME SERIES DATA FROM A CLINICAL

       TRIAL ON PATIENTS WITH NEWLY DETECTED ASTHMA

        Peter Almgren, Matematisk Statistik i Lund

Peakflow is a frequently measured variable in clinical trials
concerning asthma. The aim of this trial was initially to
suggest a simulation model for the peakflow time series based
on data from a clinical trial. A primary question was whether
stationary parts of the time series could be modelled by the
use of time series analysis. A model that roughly describes
the correlation structure for most meanlevel adjusted series
turned out to be an ARMA-model of order (1,1). We have also in
a descriptive manner investigated how much of the variation in
the parameter estimates that may be explained by that the
series are as short as they are. Finally we have worked with
two methods for estimation of the underlying non-observable
distribution for the a-parameter in AR(1)-models. One method
is non-parametric, the other is parametric.