English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Since the use of multivariable survival analysis for pregnancy outcome studies was first introduced into the reproductive epidemiology literature (1), few researchers examining fetal loss have applied these methods, which include Cox proportional hazards and other statistical models. Because logistic regression can yield misleading results, Li et al. (2) are to be commended for electing to make use of these methods. It might appear, therefore, that their study was methodologically strong. Unfortunately, several key limitations to the study (2) and some of the authors’ decisions limit the credibility of their findings. A synopsis of the concerns follows. First, the response rate was very low. Second, data for early gestational weeks are sparse and may not be representative. Third, most of the cases were interviewed after they miscarried, and the association hinges on the subgroup of these women who both miscarried early and were interviewed after they miscarried. Fourth, the subset analyses are flawed. Fifth, the lack of association with fever undermines the central hypothesis. Sixth, residual confounding from measured risk factors was not excluded. 1. Participation was 39 percent. The authors (2) acknowledge that bias could have occurred, but they do not give much weight to this possibility. With a response rate this low, substantial biases can be present even when the differences between participants and nonparticipants are modest. This bias can occur because the sample could easily be unrepresentative with respect to the joint distribution of miscarriage risk and use of hot tubs or Jacuzzis (any type of whirlpool bath; Jacuzzi Brands, Inc., West Palm Beach, Florida). In table 1, we provide two possible scenarios in which, conditional on outcome, the proportion of participants exposed differs little (i.e., by 3 percent in scenario 1 and by 4–5 percent in scenario 2) from that of nonparticipants. Yet the association is essentially wiped out in the total population. Although these examples use two-by-two tables (applicable in the absence of variable left truncation and confounding from other factors), similar examples could be shown for more complex situations by using survival analysis. The simple point is that small differences in the prevalence of exposure will change the measures of association substantially when the nonparticipants represent a high fraction of the cohort. A hint about the difference between those who did and did not participate is provided by the reasons given for nonparticipation. “Too busy” was the most common one. Hot tub users who participated were more educated and more likely to exercise and use vitamins than were nonusers who participated; if, in addition, having a miscarriage made an otherwise busy but health-conscious woman even more motivated to be interviewed, bias would be likely. Another clue about the potential for biased participation might have been the difference between gestational age at the time of the positive pregnancy test (this was the definition for entry) for participants versus nonparticipants. This comparative information apparently was not available, however. 2. Li et al. (2) are to be commended for the remarkable early recruitment of pregnant women. Approximately half of the women were recruited 12 days or fewer after missing their first menstrual period. Thirty women were recruited before 28 days had passed from their last menstrual period, that is, on average, before missing their period. The crucial questions are, How representative of the cohort are those who entered the study at an extremely early point in pregnancy? And, how accurate are their gestational dates? Epidemiologists studying miscarriage are faced with the difficult challenge of trying to ascertain, as early as possible, pregnancy losses. Paradoxically, success in doing so may result in a highly select sample and, if not analyzed carefully, may give rise to biased estimates of effects. The problems of sparse data and selection in the early weeks of pregnancy have been discussed for several decades by authors studying large cohorts (3–7). Given this conundrum, we must be attentive to how the data in the early weeks are handled, when differential entry by underlying risk is possible and, in several studies, has been observed (3, 5–7). Moreover, errors in gestational dating may be more prevalent among those

Invited Commentary: Hot Tubs and Miscarriage--Methodological and Substantive Reasons Why the Case Is Weak