Statistical analysis
A total of 65 variables considered a priori as possible
risk factors for colic were screened for univariable association
with outcome (colic) using a Chi-squared test
for categorical variables and a univariable logistic regression
model for continuous variables (see supplementary
information). Two outcomes were investigated: a known
history of colic during ownership/care of the horse
(colic ever) and a history of colic in the previous 12
months. Where variables were highly correlated (Pearson
correlation coefficient >0.9) the most statistically
significant or biologically plausible variable was selected.
The functional form of the relationships between continuous
variables and each outcome were explored using
generalised additive models (GAM) [22]. Variables with P0.25) were
then forced back into the relevant model to ensure no
significant or confounding variables had been excluded.
The fit of each model was also assessed using the Hosmer-
Lemeshow goodness of fit test statistic. The critical
probability for all analyses was set at 0.05. Data analysis
was performed using Stata (Intercooled Stata 9.0, Timberlake
Consultants Ltd, London, UK) and S-plus
(Insightful Corp., Seattle, USA).