MULTILEVEL MODELLING IN HEALTH: A STUDY OF MILK PRODUCTION IN PORTUGUESE DAIRY FARMS
Prof. Pedro Oliveira
Instituto de Ciências Biomédicas Abel Salazar, Universidade do Porto
- FCUL - Bloco C/6 Piso 4 Sala: 6.4.30 - (4ª feira) - 14:30
- Quarta-feira, 31 de Outubro de 2018
- Referência Projeto: Project UID/MAT/00006/2013
Paratuberculosis, or Johne's disease, is a chronic granulomatous enteric disease affecting both ruminant and nonruminant animals caused by Mycobacterium avium ssp. paratuberculosis (MAP). No fully effective tools or strategies exist to prevent new infections or disease progression. Diagnosis of MAP infection and measurement of infection effects are difficult because of the long incubation period, lack of diagnostic tests that accurately determine present and future status of the animal, disease dynamics at animal and farm level, and case definition of the positive animal. Multilevel mixed models were used to investigate the association of cow MAP status with variation in milk production and somatic cell count. Corrected milk production, and somatic cell count during 5 lactations lifespan in Portuguese dairy herds, in a total of 191 farms, including 14,829 cows and their respective 36,219 lactations, were considered in the analysis. The data have a longitudinal component which considers the different numbers of observations by cow and a natural hierarchical structure reflecting the "measurement occasion" (first level) nested within cows (second level), and cows nested in farms (third level) Several multivariable multilevel models, having as dependent variable Milk Production or Somatic Cell Count and as explanatory variables Lactation, Cow Status or Farm Status and cross-level interactions were fitted. All models have at least three variance components: a residual variance at level 1, random intercept variances at level 2 and level 3, mimicking the three-level data structure. Models were fitted using Maximum Likelihood Estimation and an unstructured random-effects variance/covariance matrix. The effect of Lactation, with linear and quadratic terms, is significant in all models; these terms account for the curvature associated with Milk Production and Somatic Cell Count along lactations: the production increases from first to third lactation, decreasing afterwards. The variance/covariance analysis confirms the importance of the selected structure. Each level - measurement occasion, Cow and Farm - retains a significant amount of the observed variance in the data. This study provides the first report of MAP effects in Portuguese dairy farms using a Multilevel approach, based on a large size data collection spanning a broad temporal scope, in a three-level structure.