Introduction This paper presents the results of a study to evaluate survival analysis the effect of treated diseases on the culling rate (remove from herd) in dairy cows.

Methods Five different models, with and without time-dependent covariates, using Gompertz distribution were studied. Model 1 treated diseases as a binary and time-independent covariates. Models 2 through 5 treated diseases as time-dependent covariates. For each observation, we split follow-up time in intervals each corresponding to a different lactation month. In other words, each observation from study entry until culling or censoring was split into several one-month observations by Lexis expansion of the original dataset. Model 2 assumed an animal experience a certain disease from the beginning of the occurrence of that disease by the end of follow-up period. Model 3 assumed cows are at risk from the begging of the study until the disease occurred (inverse of model 2). In model 4 and 5 an animal was assumed to experience a certain disease for 1 month if the disease occurred during this period. The only difference is in model 4 assumed diseases occurred only one time and in model 5 multiple disease occurrences at different months were considered as different episodes.

Results According to Akaike's Information Criterion (AIC) value and Cox-Snell residuals model 5 was the best model.

Conclusion A comparison of culling models with and without time-dependent covariates found that models without time dependency tended to seriously underestimate the risk of a disease on culling.