Extracting information from apparent randomness in cardiovascular data

A cosinor analysis of chronobiological data has firmly established the predominance of certain deterministic periods, characterizing life processes, here illustrated by serial blood pressures. The percentage of the variance accounted for by harmonic models, however, is almost invariably much less than 100 for any biologic variables. To deal with such residuals, the authors consider a data-fitting scheme using a scalar summary of related yet distinct vectorial measurements and a hidden Markov model on the residuals from a deterministic, e.g. cosinor fit of harmonics. In applying this model to data from human newborns, after removal of a 24-hour cosine component from the data, a separation of infants with positive versus negative family history of high blood pressure remains possible. This result complements the separation by the deterministic 24-h cosine model, here removed from the time series prior to the use of the hidden Markov model.<>