Estimating fetal age: how accurate is it really?

Estimating fetal age (and predicting calving date) from fetal size is commonly used on dairy farms. However, there is only limited evidence on how effectively fetometry predicts calving date. In their recent paper, Brownlie et al. (2016) use data from the New Zealand National Herd Fertility Study to assess how accurately veterinarians using transrectal ultrasonography predicted calving date, and to identify some of the factors affecting accuracy.

This valuable paper identifies how well the pregnancy diagnosis process predicted calving date. The data came from 12 experienced veterinarians across New Zealand and from 83,104 cows, so are likely to be representative. Overall, 90.3% of cows calved within 10 days of their predicted calving date (correct prediction), with 5.3% of cows calving >10 days later and 4.4% >10 days earlier. These predictions were reassuringly accurate, but the overall result hides important caveats.

Brownlie et al. (2016) found significant effects of region, cow age, and timing of artificial insemination (AI) relative to the end of the AI period, but for these factors the lowest percentage of correct predictions was 87.6%. In contrast, for fetal age, number of AI and number of herd pregnancy tests prior to first estimation of fetal age, the lowest percentages were all <80%. Two key factors drove the accuracy of fetal aging, firstly the likelihood that a specific AI resulted in the fetus now being measured, and secondly fetal age.

Increasing number of AI dates meant more uncertainty, lowering the percentage of correct predictions to 84% in cows with ≥3 AI dates, from 92% with only one AI date. No recorded AI date had even more impact, reducing correct predictions to 79%. Lack of an AI date also drives the poor accuracy of cows diagnosed pregnant later in the process; only 73% of predictions were correct for cows first diagnosed pregnant at the third herd pregnancy test, compared with 92% at the first herd test. In the New Zealand system, fetal aging is really confirmation of an AI date, so to estimate fetal age we just need to be accurate enough to distinguish between two dates which will usually be ∼21 days apart. If there are no AI dates or the pregnancy is likely to be the result of an unrecorded natural mating, then the farmer should be informed that estimates are not very accurate.

The peak in correct predictions (∼93%) was achieved when fetuses were aged as being 12 or 13 weeks. In New Zealand this applies mostly to cows which became pregnant in the first 2–3 weeks of the breeding season and, particularly at the first herd visit, is the predominant fetal age, providing a good opportunity for self-validation. Reducing fetal age moderately decreased accuracy (87% for fetuses ≤5 weeks old), but increasing fetal age to ≥15 weeks markedly decreased correct predictions (72%). Increasing fetal age decreases accuracy because as fetuses grow the variance of the population increases (Kähn 1989). More variance means reduced accuracy. This is often not appreciated by veterinarians who are used to seeing the association between fetometry and fetal age presented as a simple line graph (e.g. Lazim et al. 2016) with no measure of variance included.

This change in variance with time means that standard regression procedures are not suitable for analysing accuracy as they can result in misleading inferences (Davidian and Haaland 1990). Furthermore, simple correlation coefficients do not provide a means of comparing between measurements as they are affected by factors other than how well fetometry predicts age (Goodwin and Leech 2006). Nevertheless, almost all reports of fetometry and fetal age (from White et al. 1985 to Lazim et al. 2016) have focussed on correlation and regression rather than accuracy. One study (Fitzgerald et al. 2015), which claimed to have evaluated ultrasonography as a predictor of gestational age, just reported mean difference between predicted and actual values, with no measure of precision at the individual cow level. Lawrence et al. (2016) were unable to find any studies of fetometry suitable for directly comparing with their data on fetal age and placentome size. It seems remarkable that a commonly used technique has such limited evidence of its predictive ability, but this does seem common in the veterinary field.

This lack of evidence on the accuracy of fetometry may have clinical consequences. Adeyinka et al. (2014) reported that, in contrast to fetal data, the variance of placentome size did not increase markedly during gestation. Thus it is possible that in cattle >14 weeks pregnant, measuring placentomes could more accurately estimate fetal age than fetometry, potentially reducing the effect of increasing fetal age on the accuracy of predicted calving reported by Brownlie et al. (2016). But without comparable accuracy measurements based on fetometry this hypothesis cannot be tested. More method comparison studies are needed.