Health information and health outcomes: An application of the Regression Discontinuity design to the 1995 UK contraceptive pill scare case.

Rigorous causal inference is important and often necessary in order to provide sound policy advice. An evaluation method that is rapidly becoming standard in the economics
literature is the regression discontinuity (RD) design. Here we use the RD method to analyse the eff�ects of the 1995 pill scare in the United Kingdom, showing how this methodology
can be usefully employed in epidemiological studies.
On 19 October 1995 the UK Committee on the Safety of Medicine issued a warning stating that combined oral contraceptives containing either gestodene or desogestrel (the `third generation' pill) were associated with twice the risk of venous thromboembolism as compared to older products. The warning received massive media coverage in the days immediately following the announcement, and this led many women to change contraceptive method or stop contraception altogether. The studies which had motivated the health warning were later disproved, but by that time the consequences of this health scare on pregnancies and abortions had already unfolded.
We use monthly records on conceptions, abortions and births, as well as individual birth records to evaluate the eff�ects of the 1995 UK pill scare on quantitative outcomes, such as number of abortions and births, and qualitative outcomes, such as birth weight. We compare treated with non-treated groups of women, where the treatment is represented by the presence of information on the increased risk associated with the `third generation' oral contraceptives.
Assignment to treatment can be non random and individuals who receive treatment may systematically di�er from those who do not. In contrast to conventional observational studies, within the RD design the analyst has speci�c knowledge of the assignment rule that influences how individuals are assigned to (or selected into) treatment. More specifi�cally, the design requires that there is a known cut-o� point at which there is a discrete jump in the probability of being treated. In our application the cut-o� point is the date on which the health warning was made public (19 October 1995).
We �find that following the health warning on the `third generation' pill, conception rates increased by about 7%, with a 9% increase in abortion rates and a 6-7% rise in birth rates. By contrast, no or only small eff�ects were found on the resulting babies' general health, with the exception of a reduction in the incidence of congenital anomalies. Di�fferences by mother's age and social class were very pronounced, with most of the eff�ects being experienced by women aged less than 25 and of lower socioeconomic status. We also show that standard evaluation methods underestimate the impact of the pill scare by up to 30-40% for some outcomes and detect no signifi�cant e�ffects for others. The empirical assessment of the pill scare impact, therefore, would diff�er substantially depending on the statistical evaluation strategy.