Third-generation Pills |
||
|
|
It is educational, with that background, to read a paper from one of the authors of the key study which quietly and thoughtfully examines the possibilities of bias and causality, and comments rationally on how we may evaluate such information in future [1].
For new readers
The original information suggested that venous thromboembolism, which occurs in about 1 in 10,000 women on second-generation pills, happened twice as often with third-generation pills - an excess risk of 1 in 10,000. Most (99%) young women in whom this occurs are treated successfully, so the risk of any additional serious outcomes would be at most 1 in a million if there was an additional risk.
Spitzer looks at the additional risk (an odds ratio, or relative risk of about 2) and asks whether it could be produced by bias, and even if true whether it has credibility.
Bias
A number of different sources of bias could have occurred:
- Confounding: where the effect of exposure risk is distorted because of association of exposure with other factors influencing the outcome.
- Effect modification: a factor modifies the effect of a suspected causal factor.
- Prescribing bias: preferential prescription of drugs depending on characteristics of patient or drug.
- Referral bias: differential hospital referral of patients for diagnosis with similar symptoms but different clinical backgrounds.
- Healthy user effect: occurs when doctors change patients from a well-established product because they are concerned that the patient may be at unnecessary risk or is not doing well.
Spitzer argues that healthy user bias had an effect and that referral bias and prescription bias operated to drive the estimate of risk upwards. While the observed odds ratio was about 1.5, it was about 1.0 for new users of oral contraceptives.
Causality
Austin Bradford-Hill set out guidelines that have been used to elucidate causality from association [2]. The third-generation pill story was one of association, so asking questions about causality is useful in determining the underlying credibility of the association.
As the Table shows, applying the Bradford-Hill criteria gives an underwhelming conclusion as to causality.
Table: Causality and thromboembolism with third-generation oral contraceptives
| Criterion | Comment |
| Experiment | No experiment was done, and no experiment could reasonably be done to test the association |
| Strength | The odds ratios are weak (less than 2) |
| Consistency | Relative risks are consistently weak, and some studies include no additional risk |
| Gradient | Emerging evidence suggests that lower doses of ethinyl oestradiol have higher odds ratios for thromboembolism. The gradient is paradoxical |
| Biological plausibility | There is none |
| Specificity | The outcome is not specific for the intervention., Other factors (obesity, immobility, pregnancy) cause thromboembolism |
| Coherence | There is none. Thromboembolism rates in second generation pills is lower now than 10 years ago |
| Temporality | Not an issue |
| Analogy | There are none |