Diagnosing the pregnancy status of dairy cows: How useful is milk mid-infrared spectroscopy?

Accurate and timely detection of pregnancy is vital in commercial milk production enterprises. There are several methods to do so including observation of non-return to oestrus, trans-rectal palpation, trans-rectal or trans-cutaneous ultrasonography, and analysis of progesterone and pregnancy-associated glycoproteins in milk or blood. However, these methods have an associated cost, are not all efficient and some require animal handling, which might limit their practical implementation. Fourier-transform mid-infrared (MIR) spectroscopy is already routinely used in the dairy industry worldwide to analyze major milk components (e.g. fat, protein, lactose and urea) for payment, herd management, quality control, or genetic evaluation programs. Additionally, MIR can be used to predict other parameters associated with milk composition in dairy cows with reasonable accuracy, such as fatty acids, ketone bodies, methane emissions, energy intake and feed efficiency. Because the establishment of pregnancy affects milk composition through altering nutrient partitioning between physiological functions, it might be hypothesized that MIR could be used to detect the pregnancy of a dairy cow. Thus, in the study of Dr P. Delhez and associates it was aimed to investigate the potential of milk MIR to predict the pregnancy status of dairy cows. Their results were published in the Journal of Dairy Science, Volume 103 of 2020, page 3264 to 3274. The title of the paper is: Diagnosing the pregnancy status of dairy cows: How useful is milk mid-infrared spectroscopy?   

For the study, MIR spectra and insemination records were available from 8064 Holstein cows of 19 commercial dairy farms. Three options were studied in the model to classify cows as open or pregnant. In the first option, a novel residual MIR spectrum was used, measuring the difference between the spectrum after an insemination and a spectrum before this insemination at a specific stage during the same lactation. By doing this, it was expected that the MIR signal would be simplified after insemination by removing the effects specific to each cow when being open while preserving the potential pregnancy signal. The second option was exploration of predictions at different stages after insemination, because the stage of gestation is known to influence milk composition. The third option was the use of cow-independent validation because the commonly applied method of random cross-validation has recently been reported to produce overoptimistic results. 

For the first option, 6754 MIR spectra were considered. The results, however, showed little ability to detect the pregnancy status, as the area under the receiver operating curve was 0.63 and 0.65 for cross-validation and testing, respectively. The second option, involving 1664 records, aimed to reduce noise in the MIR spectra used as predictors by subtracting a spectrum before insemination (i.e., open spectrum) from the spectrum after insemination. The accuracy was comparable with the first approach, showing no superiority of the method. Given the limited results for these models when using combined data from all stages after insemination, the third option explored separate models at seven stages after insemination comprising 348 to 1566 records each (i.e., progressively greater gestation), with single MIR spectra after insemination as predictors. The models developed here, using data recorded after 150 days of pregnancy, showed promising prediction accuracy with the average value of area under the receiver operating curve of 0.78 and 0.76 obtained through cross-validation and testing, respectively.  

In conclusion, the results of the study showed that milk MIR spectral data collected at different stages after insemination, when used directly or taking a spectral difference, were not sufficient to detect the pregnancy status of dairy cows. However, the models developed using data recorded after 150 days of pregnancy showed promising prediction accuracy. If this can be confirmed using a larger data set and can be done a little earlier, the models could be used as a complementary tool to also detect foetal abortion.