Use of a non-linear spline regression to model time-varying fluctuations in mammary-secretion element concentrations of periparturient mares in Michigan, USA

A study was designed to model the fluctuations of nine specific element con centrations in mammary secretions from periparturient mares over time. Duri ng the 1992 foaling season, serial samples of mammary secretions were colle cted from all 18 pregnant Arabian mares at the Michigan State University eq uine teaching and research center. Non-linear regression techniques were us ed to model the relationship between element concentration in mammary secre tions and days from foaling (which connected two separate sigmoid curves wi th a spline function); indicator variables were included for mare and mare parity. Element concentrations in mammary secretions varied significantly d uring the periparturient period in mares. Both time trends and individual v ariability explained a significant portion of the variation in these elemen t concentrations. Multiparous mares had lower concentrations of K and Zn, b ut higher concentrations of Na. Substantial serial and spatial correlation were detected in spite of modeling efforts to avoid the problem. As a resul t, p-values obtained for parameter estimates were likely biased toward zero . Nonetheless, results of this analysis indicate that monitoring changes in mammary-secretion element concentrations might reasonably be used as a pre dictor of impending parturition in the mare. In addition, these results sug gest that element concentrations warrant attention in the development of ne onatal milk-replacement therapies. This study demonstrates that non-linear regression can be used successfully to model time-series data in animal-hea lth management. This approach should be considered by investigators facing similar analytical challenges. (C) 2000 Elsevier Science B.V. All rights re served.