Survival analysis as a statistical methodology for analyzing factors that affect milk coagulation time in Holstein-Friesian and Brown Swiss cows

The aim of this study was to propose the survival analysis technique as a statistical approach for the analysis of rennet coagulation time (RCT) able to make use of coagulating and noncoagulating (NC) milk information in order to estimate potential sources of variations that affects RCT. A total of 1,025 Italian Holstein-Friesian cows (HF; progeny of 54 sires) and 1,234 Brown Swiss cows (BS; progeny of 58 sires) reared in 34 and 38 herds, respectively, were milk-sampled once. Rennet coagulation time was analyzed with a semiparametric proportional hazard model (i.e., a Cox model), with the NC samples considered as censored records. Furthermore, a different censoring scenario, with a new end point at 18 min, was considered after the rearrangement of the time space originally used for the observation of RCT. The percentage of NC samples was almost 10% for HF and 3.5% for BS cows in in the 31 mm set, whereas it increased to 44 and 24.9%, respectively, in the 18 min set. Estimated hazard ratios indicated that the most important factors affecting the coagulation process were herd, days in milk, casein number, and milk acidity (expressed in terms of titratable acidity) for both HF and BS, whereas the SCS was relevant only for BS. The survival model seems to be particularly suitable for this analysis, as it can properly account for censored and uncensored records and appropriately use all available information. Moreover, this methodology allows us to rearrange the time space used for the observation of RCT and to define alternative traits (i.e., RCT with an end point at 18 min). Our restriction of the time space and the increased percentage of censored records did not highlight any substantial differences in terms of the risk of coagulating with respect to the traditional 31 mm testing time. Although further research is needed to investigate the effect of these sources of variation on cheese yield, our results indicate that casein number, acidity, and SCS may be used as indicator traits for enhancing the technological properties of bovine milk.