Statistical analysisA total of 65 variables considered a priori as possiblerisk factors for colic were screened for univariable associationwith outcome (colic) using a Chi-squared testfor categorical variables and a univariable logistic regressionmodel for continuous variables (see supplementaryinformation). Two outcomes were investigated: a knownhistory of colic during ownership/care of the horse(colic ever) and a history of colic in the previous 12months. Where variables were highly correlated (Pearsoncorrelation coefficient >0.9) the most statisticallysignificant or biologically plausible variable was selected.The functional form of the relationships between continuousvariables and each outcome were explored usinggeneralised additive models (GAM) [22]. Variables with P<0.25 were considered for inclusion in a multivariablelogistic regression model which was constructed using abackwards stepwise elimination procedure. Variablesremained in the model if they significantly improved thefit (P≤0.05), assessed using the likelihood ratio test statistic.All variables (including those with P>0.25) werethen forced back into the relevant model to ensure nosignificant or confounding variables had been excluded.The fit of each model was also assessed using the Hosmer-Lemeshow goodness of fit test statistic. The criticalprobability for all analyses was set at 0.05. Data analysiswas performed using Stata (Intercooled Stata 9.0, TimberlakeConsultants Ltd, London, UK) and S-plus(Insightful Corp., Seattle, USA).
การแปล กรุณารอสักครู่..
