- 0.1 Key message box
- 0.2 Key message box
- 1 Introduction
- 2 Methods
- 3 Results
- 4 Discussion
- 5 Recommendation
- 6 Acknowledgments
Key message box
What is already known about this subject?
Anecdotal reports have suggested that the use of non-enzyme-inducing antibacterial drugs may reduce the efficacy of hormonal contraceptives, leading to contraceptive failure.
Formal studies have either suggested that there is an interaction of this kind but without conclusive evidence, or have shown no evidence of an interaction.
Most previously negative or inconclusive studies have been too small to rule out an interaction in a small subset of women, and other criticisms of previous studies leave the question of an interaction open.
Key message box
What are the new findings?
We have surveyed Yellow Card reports of unintended pregnancies (the primary endpoint) in women taking non-enzyme-inducing antibacterial drugs, enzyme-inducing drugs (known to impair the efficacy of hormonal contraceptives), and control drugs (not reported to impair the efficacy of hormonal contraceptives).
Compared with the control drugs, the enzyme-inducing drugs were associated with a 13-fold increase in the risk of reports of unintended pregnancies and the antibacterial drugs with a 7-fold increase. The enzyme-inducing drugs were associated with a 7-fold increase in the risk of reports of congenital malformations (another known association), but there was no association with the antibacterial drugs.
The antibacterial drugs were not associated with increases in the secondary endpoints, cardiovascular adverse events or headache (control events). They were not associated with diarrhoea, a potential confounding factor.
How might it impact on clinical practice in the foreseeable future?
This evidence suggests that there is an interaction of antibacterial drugs with hormonal contraceptives, which can potentially impair the effectiveness of the contraceptives.
The precautionary principle dictates that women taking hormonal contraceptives should be advised to take extra contraceptive precautions when a short course of an antibacterial drug is added.
Suspicion that antibiotics might reduce the efficacy of oral contraceptives was first raised in 1973 when a woman taking hormonal contraception also took chloramphenicol and became pregnant.1 In 1975, Dosseter reported that three women had become pregnant while taking ampicillin, despite hormonal contraception, and questioned whether there might be an interaction.2 Later reports documented pregnancies in women who took a variety of non-enzyme-inducing antibiotics, despite hormonal contraception.3
Suspicion of a drug–drug interaction originally focused on the broad-spectrum antibiotics4 including ampicillin, co-trimoxazole, and tetracyclines.5 In 1988 an analysis of reports to the UK’s Committee on Safety of Medicines received between 1968 and 1984 identified 63 pregnancies in women taking oral contraceptives with antibiotics, most commonly tetracyclines and penicillins.6 Many other anecdotal reports and database studies of the (in)effectiveness of oral contraceptives during the use of non-enzyme-inducing antibiotics have appeared (see online supplementary e-appendix).
Nevertheless, results from small studies have been taken as evidence that there is no interaction (see discussion below), a negative that is hard to prove definitively. An interaction of this kind would be of great clinical importance. Given the evidential contradictions, we have taken a different approach to test the null hypothesis that the risk of unintended pregnancies is not altered in women who are exposed to antibiotics.
We searched the list of reports of suspected adverse drug reactions published by the Medicines and Healthcare products Regulatory Agency (MHRA)7 for information on suspected adverse drug reactions between 1963 and July 2018. To be included, reports had to mention exposure to a drug of interest linked to a report of an unintended pregnancy.
We prospectively chose three groups of medications and five types of adverse events to study.
Nine commonly used non-enzyme-inducing antibacterial drugs: amoxicillin, ampicillin, cephalexin, ciprofloxacin, erythromycin, metronidazole, nitrofurantoin, oxytetracycline, trimethoprim.
Twelve enzyme-inducing drugs (positive controls known to reduce the efficacy of oral contraceptives): carbamazepine, eslicarbazepine, griseofulvin, nevirapine, oxcarbazepine, phenobarbital, phenytoin, primidone, rifabutin, rifampicin, ritonavir, topiramate.
Nine control medications (medications commonly used by women of childbearing age), randomly chosen, not expected to alter the efficacy of oral contraceptives: citalopram, ibuprofen, lansoprazole, loperamide, loratadine, paracetamol, propranolol, theophylline, zolpidem.
Fetal malformations (a positive control event, expected to be associated with the group of enzyme inducers, which includes known teratogens, but not with the antibiotics or control medicines).
Cardiac arrhythmias and headache (control events, not expected to be associated with any of the three groups of medicines).
Diarrhoea (a possible confounding factor for an effect of antibiotics).
We extracted the numbers of adverse events in women taking non-enzyme-inducing antibiotics or enzyme inducers and expressed them as rates per 100 000 reports with 95% confidence intervals (CIs). We calculated reporting odds ratios with associated 95% CIs, comparing both the cases (non-enzyme-inducing antibiotics) and the positive controls (enzyme inducers) with the controls (medications commonly used in women of childbearing age). To test for significance we used χ2 tests with Bonferroni correction for multiple comparisons of the data.
The results are shown in table 1. In total we sourced outcomes from 173 073 spontaneous reports: antibiotics (n=74 623), enzyme-inducing medicines (n=32 872), control medicines (n=65 578). We found six reports of unintended pregnancies in the control group (9 per 100 000, 95% CI 2 to 16); 46 (62 per 100 000, 95% CI 44 to 79) in those who took antibiotics; and 39 (119 per 100 000, 95% CI 81 to 156) in those who took enzyme-inducing drugs
Thus, the enzyme inducers increased the frequency of reports of unintended pregnancies 13-fold (OR 13.0, 95% CI 5.5 to 31, p<0.000001) and the antibiotics nearly 7-fold (OR 6.7, 95% CI <2.9 to 16, p<0.00001).
The enzyme inducers (positive cases) increased the reported frequency of congenital abnormalities 7-fold (OR 7.4, 95 % CI 6.4 to 8.6, p<0.00001) whereas the antibiotics did not (OR 0.71, 95% CI 0.59 to 0.86) compared with controls. We found no increase in cardiac arrhythmias, headaches or diarrhoea with either non-enzyme-inducing antibiotics or enzyme inducers.
These spontaneous reports constitute a signal8 of a possible drug–drug interaction, with a 7-fold higher rate of reporting of unintended pregnancies with antibiotics compared with the control medications. The known interaction with enzyme inducers was associated with a confirmatory signal and a 13-fold higher reporting OR. The absolute risks cannot be calculated from these data.
Any mechanism by which an antibiotic reduces the effective circulating concentrations of the contraceptive hormones will increase the risk of pregnancy. However, the risk will vary according to the timing of administration of the antibiotic in relation to the woman’s menstrual cycle. It will also vary according to other factors, including the individual response to a fixed dose of the combined contraceptive hormones.
Whatever the mechanism, it is highly unlikely that it will affect more than a subset of women for four reasons. First, there is large inter-individual variability in oestrogen concentrations after a dose of an oral contraceptive9—a change too small to affect one woman might affect another markedly. Second, in any woman there is large intra-individual variation in oestrogen concentrations from day to day, even at the same phase of the cycle,9 and also between phases of the cycle in which the hormones are being taken (eg, during the pill-free phase). Third, variation in the binding of the highly protein-bound ethinylestradiol (see below) could make some women more susceptible than others. Fourth, the effects of interactions involving bacteria (if that is the mechanism) vary with variations in individual microbiomes; for example, erythromycin alters serum digoxin concentrations by inhibiting digoxin metabolism by a bacterium, Eubacterium lentum; however, not everyone carries this bacterium—only 10% of individuals are affected.10
The risk will therefore vary between women and within an individual according to circumstances. Previous studies have been too small to rule out an interaction in a subset of women.
Three mechanisms have been proposed.
In 1971 an increased incidence of intermenstrual bleeding was reported in women who were reliably taking oral contraceptives and took rifampicin for tuberculosis.11 Enzyme inducers increase hepatic metabolism of the hormones in oral contraceptives, reduce the circulating concentrations of the active drugs and reduce contraceptive efficacy. Enzyme-inducing antimicrobial drugs that do this include rifampicin,12–16griseofulvin,17–19 ritonavir+darunavir20 and nevirapine.21
Altered gastrointestinal function
Antibiotics increase gastric emptying and small intestinal motility, best studied in relation to macrolide antibiotics.22 These effects could theoretically alter gastrointestinal absorption (and reabsorption) of oral contraceptive hormones (probably both estrogens and progestogens).
Diarrhoea and vomiting may also affect absorption.23
Altered enterohepatic recirculation
The absorption and reabsorption of oestrogens and progestogens depend on physicochemical factors in the gut (such as pH, whether compounds are conjugated, and adsorption to microbes and dietary constituents). Conjugated oestrogens (eg, ethinylestradiol glucuronide) are excreted in the bile, deconjugated by gut bacteria, and reabsorbed in the lower small intestine. Microbes affect the enterohepatic circulation of other molecules with related structures, such as androgens24 and bile acids.25
Antibiotics, particularly those that are poorly absorbed, alter the gut microbiome and so may reduce deconjugation and subsequent reabsorption of oestrogens.26–28 For example, ampicillin increased the faecal concentrations of conjugated endogenous oestrogens in pregnant women by up to 70 times within 2–3 days.29 In the last trimester of pregnancy, phenoxymethylpenicillin, ampicillin, and neomycin reduced circulating and/or urinary oestrogens by interfering with their enterohepatic circulation.30–35
Previous studies, current advice, and the evidence on which it is based
Despite the many reports referred to above, in 2009 and again in 2015 the WHO concluded that most broad-spectrum antibiotics do not reduce the contraceptive effectiveness of combined oral contraceptives, patches, or rings; they advised that when antibiotics are used there is no need to take extra precautions.36 37 The US Center for Disease Control and Prevention repeated this advice in 2010,38 as did the Clinical Effectiveness Unit (CEU) of the Faculty of Sexual and Reproductive Healthcare in the UK’s Royal College of Obstetricians and Gynaecologists in 2011 and again in 2017. The CEU concluded that there is no evidence of an interaction,39 40 and advised as follows: “most broad-spectrum antibiotics are non-enzyme-inducing and do not require any special precautions. No additional contraceptive precaution is required unless the antibiotics (and/or illness) cause vomiting or diarrhoea.”
The evidence on which these views are based is shown in the online supplementary table e-Appendix, to which we have added critical comments, noting flaws in the ways in which the evidence has been interpreted to imply that there is no interaction:
It has been assumed that the mechanism would be a direct pharmacokinetic interaction involving enterohepatic recirculation, as described above; other possible mechanisms have not been considered.
Whatever the mechanism, it has been assumed that it operates in all women, which is highly unlikely, as discussed in detail above.
Studies that have been cited as direct evidence against an interaction involved measurements of surrogate markers such as total ethinylestradiol concentrations, not pregnancy rates, the relevant outcome.
Ethinylestradiol is over 98% bound by serum albumin.41 A small increase in binding could produce a relatively large decrease in unbound concentration, with possible profound effects. Thus, a lack of change in total ethinylestradiol concentrations cannot rule out a drug–drug interaction. Unbound ethinylestradiol concentrations have not been studied in this context.42
None of the evidence that has been previously cited against an interaction rules out the possibility of a small risk of an interaction in all women taking oral contraceptives or, more likely, a larger risk in a susceptible subset of women. Some of the evidence was circumstantial or irrelevant, as discussed in detail in the online supplementary e-Appendix. Much of it depended on studies in no more than 24 women, which cannot rule out an effect in a subpopulation of susceptible women in whom a small change in exposure to, for example, a 30 μg/day dose of ethinylestradiol could reduce the effect of the estrogen sufficiently to allow pregnancy to occur. Figure 1 shows the numbers of women one would need to study to establish a significant effect.
We compared antibiotics (cases) with both control medicines and positive control medicines. The control medicines were medicines picked at random from a hospital formulary by a colleague unaware of the purpose of the study. They are medicines that are often taken by women of childbearing age and they have not been implicated in interactions with oral contraceptives. The positive controls were enzyme inducers, which are known to reduce the efficacy of oral contraceptives.
We also included a range of outcomes (adverse events) for comparison with unintended pregnancies, including the positive control event of congenital abnormalities which is known to be associated with some enzyme inducers such as phenytoin, carbamazepine, and topiramate. The enzyme inducers were indeed associated overall with a higher rate of reporting of congenital abnormalities compared with both the antibiotics and the negative controls, neither of which would be expected to be associated with congenital abnormalities.
The finding that the enzyme inducers were associated with signals consistent with the known increased risks of both unintended pregnancies and congenital abnormalities supports the use of this method in seeking other types of drug-induced harms, including drug–drug interactions.
Diarrhoea caused by antibiotics might have reduced the efficacy of the oral contraceptives, but diarrhoea did not appear as a signal; it cannot therefore be regarded as a confounding factor in these data.
A limitation of this study is that the data on which it is based come from spontaneous reports to the MHRA of suspected adverse drug reactions. The MHRA’s database underestimates the true number of suspected reactions43 and neither the numerator nor the denominator is accurately known.
A related limitation is the possibility of reporting bias, since prescribers and patients may believe that antibiotics interact with hormonal contraceptives and therefore be more likely to report unintended pregnancies. However, the control data that we collected are reassuring in this respect, since the results with the controls and the positive controls for both the drug interventions and the adverse events were qualitatively as expected a priori.
There is insufficient information for analysis of the individual medications in each list. It is likely that different medicines carry different risks of unintended pregnancies. This is illustrated by the contrast between the effects of rifampicin and rifabutin; the effects of rifampicin are more marked than those of rifabutin.44
Although we do not know that the reports of unintended pregnancies all concerned women who were taking hormonal contraceptives, it is highly unlikely that anyone would report an unintended pregnancy to the MHRA on a Yellow Card, the purpose of which is to record suspected adverse drug reactions, if the pregnancy was not thought to be due to failure of hormonal contraception.
Given the uncertainties and the possibility of an unintended pregnancy, the concern can be reduced to a practical one: what would women want to be told? Whether or not some antibiotics increase the risk of pregnancy when a woman is taking a hormonal contraceptive containing an oestrogen, instead of dismissing the interaction as being unproven, as the British National Formulary, for example, currently does, it would be wise to advise precaution until other more definitive evidence emerges, as others have suggested.45 Extra precautions can be taken during a course of antibiotics; an unintended pregnancy, whether terminated or taken to term, is a life-changing event.
Excellent recommendations have previously been published in the UK46 and the USA.47 A proposed version is shown in box 1.
Your medicine and the pill
Your doctor has prescribed an antibiotic medicine to treat your infection.
If you are taking the pill or using some other type of hormonal contraceptive, the antibiotic may make it less effective.
This means that you may still get pregnant if you have sex, even though you are taking a contraceptive.
If you are keen not to get pregnant, it is best to avoid sex or to take extra precautions, such as using a condom, while you are taking the antibiotics and for 7 days after completing the course.
We are grateful to Ms Tohfa Ahmad for initial help in scoping the availability of data and to Professor Carl Heneghan for helpful comments on an earlier draft.