Reproduction in Domestic Animals

Reprod Dom Anim 43 (Suppl. 2), 113–121 (2008); doi: 10.1111/j.1439-0531.2008.01150.xISSN 0936-6768Improved Detection of Reproductive Status in Dairy Cows Using Milk ProgesteroneMeasurementsNC Friggens 1 , M Bjerring 1 , C Ridder 1,2 , S Højsgaard 1 and T Larsen 11 Faculty of Agricultural Sciences, University of Aarhus, Research Centre Foulum, Tjele; 2 Lattec I ⁄ S, Hillerød, DenmarkContentsThis study tested a model for predicting reproductive statusfrom in-line milk progesterone ‘measurements. The model isthat of Friggens and Chagunda [Theriogenology 64 (2005)155]. Milk progesterone measurements (n = 55 036) representing578 lactations from 380 cows were used to test themodel. Two types of known oestrus were identified: (1)confirmed oestrus (at which insemination resulted in aconfirmed pregnancy, n = 121) and (2) ratified oestrus (wherethe shape of the progesterone profile matched that of theaverage progesterone profile of a confirmed oestrus, n = 679).The model detected 99.2% of the confirmed oestruses. Thisincluded a number of cases (n = 16) where the smoothedprogesterone did not decrease below 4 ng ⁄ ml. These cows hadsignificantly greater concentrations of progesterone, bothminimum and average, suggesting that between cow variationexists in the absolute level of the progesterone profile. Usingratified oestruses, model sensitivity was 93.3% and specificitywas 93.7% for detection of oestrus. Examination of falsepositives showed that they were largely associated with lowconcentrations of progesterone, fluctuating around the4ng⁄ ml threshold. The distribution of time from inseminationuntil the model detected pregnancy failure had a median of22 days post-insemination. In this test, the model was runusing limited inputs, the potential benefits of includingadditional non-progesterone information were not evaluated.Despite this, the model performed at least as well as otheroestrus detection systems.IntroductionProgesterone measurements have been used for sometime now, and are accepted as a valid indicator, forassessing the reproductive status of dairy cows (Bulmanand Lamming 1978; Royal et al. 2000). Implicit in thisprocess is a set of biological rules that are used toconvert the progesterone data into different reproductivestatuses (e.g. postpartum anoestrus, luteal andfollicular phases of the oestrous cycle and pregnancy).Traditionally, these rules were expert opinions usuallyapplied manually to progesterone data (e.g. Lammingand Darwash 1998). As measuring technology hasadvanced from the original radioimmunoassays,through laboratory based ELISA methods (Waldmann1993), to on-farm manual tests (Simersky et al. 2007),biosensors (Delwiche et al. 2001), and ultimatelytowards automated in-line systems for measuring milkprogesterone (Lattec 2007), these rules become bothexplicit and fixed. This is because automated systemsnecessarily incorporate a model to process and condensethe raw data, providing the end-user with the relevantbiological interpretation in a palatable form.The fact that these biological rules are codified allowsthem to be tested. Testing the ability of these rules tocorrectly categorize reproductive status becomes increasinglyimportant as automated in-line progesterone monitoringsystems are implemented on commercial farms.Clearly, the value of a progesterone monitoring systemdepends to a large extent upon how good the biologicalmodel is. The objective of this study was to test the abilityof a model based on in-line milk progesterone measurementsto predict reproductive status. The model tested isthat of Friggens and Chagunda (2005).Time-series models exist for detecting oestrus frommilk traits other than progesterone (de Mol et al. 1999;Firk et al. 2003), and decision strategies for interpretingprogesterone data have been suggested (Lamming andDarwash 1998; Opsomer et al. 1998; Prandi et al. 1999;Delwiche et al. 2001). However, to our knowledge, theonly published full biological model to predict reproductivestatus based on a time-series of milk progesteronemeasurements is that of Friggens and Chagunda(2005).The normal procedure for testing a new predictor of agiven state is to compare it with some ‘gold standard’reference measure. However, in the present case themodel is based on the measure, progesterone, which isthe accepted gold standard for assessing the reproductivestatus of dairy cows (Peters and Ball 1995; Cavalieriet al. 2003). Any conventional test of progesteroneagainst another indicator (e.g. visually determinedoestrus or pedometers) rapidly reverses polarity, becominginstead a test of the other indicator againstprogesterone. Thus, testing of such a model requires anon-conventional approach, which to a large extentbecomes an exploration of the shapes of the progesteroneprofiles. Consequently, this study also characterizeskey aspects of progesterone profiles according toreproductive status.Materials and MethodsTest data collectionMilk progesterone measurements were made on proportionalwhole milk samples collected from all milkingcows in one research herd (Danish Cattle ResearchCentre) during the period 12 September 2002 to 30September 2006. Cows were milked in a robotic milkingsystem with free traffic (mean no. visits ⁄ day = 2.4). Oneprogesterone measure was to be made daily during thefirst 120 days from calving. For the remainder oflactation, progesterone measurements were to be madeevery second day. Actual average intervals betweenprogesterone samples before and after 120 days fromcalving were: 1.4 and 2.4 days, respectively. ProgesteroneÓ 2008 The Authors. Journal compilation Ó 2008 Blackwell Verlag