Bias Against the Null Hypothesis in Retrospective Registries of Gestational Drug Exposure

Fatma Etwel, Gideon Koren

Research output: Contribution to journalArticlepeer-review

5 Scopus citations


Objective The findings in retrospective pregnancy registries related to prenatal drug exposure (collected after pregnancy outcome is known) are commonly reported in regulatory documents and in the medical literature. However, there is little information about the accuracy of the estimates of risk from such registries. We therefore sought to compare the rates of major congenital malformations reported in retrospective and prospective registries for the same drug to quantify the potential bias of retrospective reports. Methods We searched for all fetal safety reports related to medications for which information from both prospective and retrospective registries was available. These were published either in the peer-reviewed literature or as pharmaceutical company documents between 1984 and 2011. Results For all drugs registries studied, estimates of major congenital malformations from retrospective registries tended to be higher than the rates in prospective registries; median estimates of risk were higher by a factor of 4.18 ± 1.23 (range 2.13–5.97). Conclusions The present study confirms a major and consistent bias against the null hypothesis in studies of teratogenic risk using retrospective registries, and this must be considered when interpreting such data. Spontaneous reporting of outcomes after exposure to a drug is highly selective towards adverse events, which families with normal pregnancy outcomes are less likely to report.

Original languageEnglish
Pages (from-to)1120-1123.e1
JournalJournal of Obstetrics and Gynaecology Canada
Issue number12
StatePublished - 1 Dec 2016
Externally publishedYes


  • Retrospective registries
  • drug safety
  • malformation
  • pregnancy


Dive into the research topics of 'Bias Against the Null Hypothesis in Retrospective Registries of Gestational Drug Exposure'. Together they form a unique fingerprint.

Cite this