We congratulate Meler et al for their recent work that examined the association between uterine artery Doppler imaging and adverse pregnancy outcomes in patients with severe early-onset preeclampsia. However, although their analyses clearly demonstrated an increased odds of adverse maternal and fetal outcomes among women with an abnormal pulsatility index, further work is needed to achieve the study’s stated objective of the evaluation of the predictive performance of uterine artery Doppler imaging in this situation.

The identification of an association between a potential predictor and an adverse health outcome (ie, a strong and statistically significant odds ratio) does not necessarily mean that the predictor will be clinically useful. Usefulness can vary considerably for predictors with the same odds ratios. For example, the sensitivity and specificity that are associated with a given odds ratio could be anywhere between 0% and 100%. As a result, the crude numbers provided by Meler et al in their Table 2 are more relevant than the odds ratios for the determination of predictive performance. These findings may be summarized as likelihood ratios (LRs): for maternal complications, an abnormal uterine artery Doppler finding had an LR+ of 1.8 (95% confidence interval [CI], 1.4–2.4) and an LR– of 0.27 (95% CI, 0.09–0.77); for neonatal complications, the LR+ was 1.7 (95% CI, 1.3–2.3), and the LR– was 0.42 (95% CI, 0.22–0.80). LRs are helpful because they express the magnitude of the information gained because of the test (ie, change between the pretest and posttest odds). Note that an abnormal Doppler test does little to help determine whether a woman will have complications and that the largest information gain occurs for the prediction of the absence of maternal complications when the test result is normal. However, even this LR– of 0.27 is in the “slightly useful” range and was associated with a fairly wide confidence interval.

Two other points are worthy of consideration. First, a comprehensive assessment of predictive performance must go beyond LRs. Indices such as stratification capacity (proportion in which subjects are assigned to the low, moderate, and high risk categories) are important as well. Finally, assessment of diagnostic and prognostic tools must move past the evaluation of single tests if they are to be clinically relevant. The situation calls for the creation of a single prediction function that will incorporate all available clinical data (eg, gestational age at onset of preeclampsia and blood pressure), which would include Doppler findings (if contributory). Such multivariable prevalence functions and simplified scoring systems are commonplace elsewhere in medicine.