Use of a stochastic simulation model to assess effects of diagnostic specificity of systems for detecting ovulating cows on herd reproductive performance in year-round calving dairy herds

Many automated systems for detecting ovulating cows in dairy herds require decisions when designing algorithms and selecting cutpoints that require a compromise between diagnostic sensitivity (probability of classifying an ovulating cow as ovulating) and diagnostic specificity [daily probability of not classifying a non-ovulating cow (whether open or pregnant but not yet diagnosed as pregnant) as ovulating]. Because sensitivity must be moderately high, this compromise often results in specificity below 100%. However, little is understood about the effects of reduced specificity on herd reproductive performance. A stochastic model was developed that simulates the reproductive process in a year-round calving dairy herd to assess effects of changes in specificity at various combinations of sensitivity and conception rate (proportion of inseminations resulting in pregnancy) on herd reproductive measures of economic importance. The model included effects of inseminations in pregnant cows on probability of conceptus loss, and variation in the interval from conceptus loss to next ovulation (i.e. the next opportunity to reconceive). moderate assumptions of the probability of conceptus loss following insemination in pregnant cows, reductions in specificity from 99.9 to 99.5, 99, 98 and 97%, resulted in decreases in mean 100 day in-calf rate (100DICR; the proportion of cows with a positive pregnancy diagnosis to an insemination on or before 100 days since calving) of 1.2, 3.3, 6.8 and 9.7 percentage points, respectively. These same reductions in Sp resulted in increases in mean 200 day not in-calf rate (200DNICR; the proportion of cows with negative pregnancy diagnosis results to all inseminations on or before 200 days since calving) of 0.5, 1.6, 3.6 and 6 percentage points, and increases in mean number of inseminations per calving (Insems/Calving; the total number of inseminations in the herd divided by the number of cows that recalved) by factors of 1.2, 1.5, 2.1 and 2.8, respectively. lationship between specificity for detecting ovulating cows and the 100DICR, 200DNICR and Insems/Calving was sensitive to changes in the probability of conceptus loss following inseminations in pregnant cows. However, even with conservative assumptions, specificity still had important effects on 100DICR and 200DNICR. Varying parameters for the interval from conceptus loss to next ovulation had little effect on the relationships between specificity and these measures. results demonstrate that specificity is an important consideration when designing algorithms and selecting cutpoints in automated systems for detecting ovulating cows. Low specificity not only increases Insems/Calving but also prolongs intervals from calving to the establishment of a sustained pregnancy resulting in substantial reductions in 100DICR and increases in 200DNICR. This model could assist when determining economically optimal combinations of ovulation detection sensitivity and specificity when developing automated systems for selecting ovulating cows in commercial herds.