Statistical modeling of ruminal pH parameters from dairy cows based on a meta-analysis

Abstract

Adequate feeding of high-performance dairy cows is extremely important to avoid the digestive disorder subacute ruminal acidosis. Subacute ruminal acidosis is defined as a status with a below-average ruminal pH that does not cause direct clinical symptoms at the individual level but is relevant for animal welfare due to a higher risk of secondary health problems at the herd level. The main objective of this study was to apply meta-analytical methods in an exploratory approach to investigate the association between pH parameters of the ventral rumen with milk and diet parameters. Data from 32 studies using continuous pH measurement in the ventral rumen of lactating cows were included in the meta-analysis. Available information extracted from all studies was categorized into parameters associated with management, cow, diet, milk, and pH. The statistical analysis was divided into 4 sections. First, a multiple imputation procedure based on a principal component model was applied, since approximately 19% of the data set consisted of missing values due to heterogeneity in provided information between the studies included in the analysis. In a second step, all potential predictors for the pH parameters, including the daily mean pH, the time with a pH below 5.8, and the pH range, were examined for their prediction suitability using multi-level mixed effects meta-regression models. These analyses were performed on the raw and the imputed data. Because the results of both approaches were consistent, the imputing procedure was considered to be appropriate. Third, automated variable selection was applied to all 3 pH parameters separately for the predictor groups milk and diet using the imputed data set. Thereby, multi-model inference was used to estimate the relative importance of the selected variables. Finally, a functional relationship between the 3 pH parameters was established. The fat to protein ratio of milk, milk fat, and milk protein showed significant associations in meta-regression analysis for all 3 pH parameters when used as a single predictor. Out of the group of diet-specific variables, the acid detergent fiber, neutral detergent fiber, nonfiber carbohydrate, starch content, as well as the forage to concentrate ratio, showed the highest significance in the models. In particular, the multi-model inference showed that the protein, fat, and lactose content of the milk can best quantify the association to the daily mean pH and the time with a pH below 5.8 in a multiple regression model.

Keywords: meta-regression; ruminal pH; statistical modeling; subacute ruminal acidosis.