A recent systematic review found statistically significant reductions in pre-eclampsia and fetal or neonatal death associated with the use of aspirin in pregnancy1,2. However, the reviewers questioned the clinical significance of this finding, stating ‘as the reductions in risk are small to moderate, relatively large numbers of women will need to be treated to prevent a single outcome’1. The average number of women needed to be treated with aspirin to prevent one case of pre-eclampsia was reported as 100 (95% CI 59–167) and the number of women needed to be treated to prevent fetal or neonatal death was 250 (125 to>10,000)1. As pre-eclampsia is a major cause of maternal and perinatal morbidity and mortality3–5, we were concerned that the reviewers' conclusion was rather pessimistic6. Small to moderate risk reduction can be clinically useful, and in this article, we explore the clinical usefulness of aspirin to prevent pre-eclampsia and perinatal death using a decision-making framework7–12.
Pre-eclampsia is a heterogeneous condition with substantially different baseline risks in various subgroups of pregnant women13. In this situation, calculating the average number of women needed to be treated from pooled meta-analysis results can be seriously misleading14. This is because the number of women needed to be treated is sensitive to changing baseline risks15. The lower the risk, the higher the number of women needed to be treated and the lower are our and women's expectations of benefit from aspirin. Conversely, the higher the baseline risk, the lower the number of women needed to be treated, the higher is our expectation of benefit and the more inclined we would be to recommend, and women to accept, aspirin therapy. Therefore, to apply the results of the systematic review, we require information about benefit tailored according to variation in baseline risks7–9,16, not the average numbers of women needed to be treated across all risk groups as reported previously1.
Clinical decision-making frameworks that allow estimation of benefit according to baseline risks have been previously described7–11, but their use in interpretation of results of meta-analyses is not common. Our objective was to explore the clinical application of the effectiveness evidence for aspirin in the prevention of pre-eclampsia and fetal or neonatal death, individualising treatment among women at a defined risk level as identified by uterine artery Doppler test during ultrasound screening in the second trimester of pregnancy, as an example.
Our decision framework for individualising treatment was built in the following steps: 1. Exploration of beneficial and harmful effects of aspirin, compared with no therapy or placebo; 2. Exploration of variation in the effects of aspirin with varying baseline risks; 3. Estimation of the absolute effects of aspirin at various baseline risk levels; 4. Definition of baseline risks of pre-eclampsia and fetal or neonatal deaths according to uterine artery Doppler test results and integration of this diagnostic information into the decision-making framework.
Exploration of beneficial and harmful effects of aspirin in pregnancy
Evidence of benefit was obtained from a Cochrane review of randomised trials1,2. This review used a comprehensive search strategy to locate published and unpublished trials. It synthesised data from 39 relevant randomised trials with over 30,000 women. There was a 15% reduction in pre-eclampsia (32 trials, 29,331 women; relative risk 0.85, 95% CI 0.78–0.92) and a 14% reduction in risk of fetal or neonatal death (30 trials, 30,093 women; 0.86, 95% CI 0.75–0.98). There was also an 8% reduction in the risk of preterm birth (23 trials, 28,268 women; 0.92, 95% CI 0.88–0.97), but there were no differences in other outcome measures such as fetal growth restriction.
There was no evidence of harm from the above systematic review of randomised trials1,2. Caesarean section rate, induction of labour, antenatal admission, placental abruption, admission to special care baby unit, fetal intraventricular haemorrhage and other neonatal bleeding were not increased with the use of aspirin. However, as randomised trials may only study patients for a short period, they are not likely to detect delayed adverse events17. In addition, randomised trials are primarily designed to establish effectiveness and may not therefore collect and report data on harm rigorously17. We, therefore, supplemented the evidence on harm with a systematic search for large observational studies using the following words and their word variants in MEDLINE and EMBASE bibliographic databases: ‘Aspirin [adverse effects]’ combined with ‘pregnancy’. Three large cohort studies18–20 and one large case–control study21 were identified (Table 1). The observational studies18–21 with over 96,000 women between them did not provide any evidence of teratogenicity or long term adverse effects of aspirin use in pregnancy (Table 1).
|Study||Study design, population and exposure||Outcome|
|Slone et al.20||A cohort study of 50,282 pregnancies: 35,418 not exposed to aspirin; 5128 heavily exposed to aspirin; 9736 intermediately exposed to aspirin.||No difference in congenital abnormalities rates between the three groups. Controlling for multiple confounders did not alter the conclusion.|
|Klebanoff and Berendes19||A cohort study of 19,226 pregnancies: 9067 not exposed to aspirin; 10,159 exposed to aspirin.||No difference in mean IQ at four years of age between the two groups. Controlling for multiple confounders did not alter the conclusion.|
|Werler et al.21||A case–control study of 1381 infants with any structural cardiac defects (cases) and 6966 infants with other malformations (controls).||No association between cardiac defects and aspirin.|
|Nielsen et al.18||A cohort study of 18,721 pregnancies: 1462 exposed to non-steroidal anti-inflammatory drugs; 17,259 on no drugs during pregnancy.||No difference in congenital abnormalities between the exposed and non-exposed groups.|
In summary, there is evidence of benefit from the use of aspirin for clinically important outcomes and it seems to be a safe treatment in both short and long term.
As systematic reviews of randomised trials and observational studies did not show any harm with the use of aspirin, we have not further discussed harm in our framework. However, if harm was present, or were to be identified in the future, then it could be integrated into the framework in a similar manner to benefit, as illustrated below.
Exploration of variation in effects according to risk levels
In most circumstances, relative summary measures of benefit (such as relative risk and odds ratio) do not vary greatly across risk levels7,16,22–26. However, as the assumption of constancy of relative risks and odds may on occasions fail9,27 the variation in effect across risk levels should be formally examined.
One approach examines the relationship between relative risk and baseline risk using the control event rate from primary studies as a continuous variable representing risk (low control event rate proxy for low risk and high event rate for high risk). However, this simple approach can introduce serious bias due to the phenomenon of regression to mean28–31. We, therefore, used an alternative, and recommended, approach that examines differences in treatment effects by relating outcome to some underlying patient risk characteristic32. For this, we made use of the Cochrane review's1,2 categorisation of included primary studies into moderate or high risk for developing pre-eclampsia, based on characteristics of the study population. This approach examined whether relative risks varied across these two risk levels. High risk for pre-eclampsia was defined as one or more previous severe pre-eclampsia, diabetes, chronic hypertension, renal disease or autoimmune disease. Moderate risk was defined as any other risk factors. For the outcome of pre-eclampsia, the pooled relative risks for moderate and high risk women do not vary greatly (Fig. 1a). A test for interaction shows that the relative risks are not statistically significantly different (P = 0.72). For the outcome of fetal or neonatal death, relative risks for moderate and high risk women were again not statistically significantly different (Fig. 1b, P = 0.16). We can therefore be confident about our assumption of constancy of relative risk reduction across risk categories.
Estimation of absolute benefit with varying baseline risk
Baseline risk can vary substantially from patient to patient and, in turn, the absolute benefit a patient obtains from treatment also varies7,15. Although relative measures of effectiveness (i.e. relative risk or odds ratio) are useful for assessing the biological strength of effect7, absolute measures of effectiveness (i.e. absolute risk reduction or number of women needed to be treated) are more useful for assessing the clinical worth of an intervention in a patient or a group of patients within a risk category.
We therefore calculated the absolute benefit for various baseline rates and represented these as number of women needed to be treated against baseline risk, as shown in Fig. 2a for pre-eclampsia and Fig. 2b for fetal or neonatal death.
Definition of baseline risks of pre-eclampsia and fetal or neonatal deaths according to uterine artery Doppler test results and integration of this diagnostic information into the decision-making framework
The key issue for clinicians and women evaluating the potential benefit of aspirin is assessment of the baseline risk. Diagnostic or prognostic studies or clinical prediction guidelines33 can provide information on baseline risks for various subgroups of patients. We used a quantitative systematic review of the diagnostic accuracy of uterine artery Doppler velocimetry to predict pre-eclampsia or fetal or neonatal death34. The various pre-test probabilities, likelihood ratios and post-test probabilities for women with a positive and negative Doppler test in clinically low and high risk women are summarised in Table 2. The post-test probabilities provide the baseline risks for various subgroups of women. From these baseline risks, using the graphs in Fig. 2 (please see footnote of Fig. 2 for how the graphs were developed), the number of women needed to be treated for specific risk groups can be determined. The graphs in Fig. 2 also allow us to estimate the confidence intervals around the number of women needed to be treated. In Fig. 3, we summarise the number of women needed to be treated for various groups of women, depending on their clinical risk status and the Doppler test finding.
|Outcome||Doppler test result||Pre-test probability *% (95% CI)||Likelihood ratios† (95% CI)||Post-test probability‡% (95% CI)|
|Clinically low risk women|
|Pre-eclampsia||Positive||3.5 (3.1–3.9)||6.4 (5.7–7.1)||18.8 (16.4–21.5)|
|Negative||3.5 (3.1–3.9)||0.7 (0.6–0.8)||2.5 (2.1–2.9)|
|Fetal or neonatal death||Positive||1.3 (0.9–1.6)||1.8 (1.2–2.9)||2.3 (1.4–3.8)|
|Negative||1.3 (0.9–1.6)||0.9 (0.8–1.1)||1.2 (0.9–1.6)|
|Clinically high risk women|
|Pre-eclampsia||Positive||9.8 (7.9–11.8)||2.8 (2.3–3.5)||23.5 (18.6–29.2)|
|Negative||9.8 (7.9–11.8)||0.8 (0.7–0.9)||7.8 (6.1–10.0)|
|Fetal or neonatal death||Positive||8.9 (5.4–12.3)||4.0 (2.4–6.6)||27.8 (16.4–42.9)|
|Negative||8.9 (5.4–12.3)||0.6 (0.4–0.9)||5.5 (3.1–9.7)|