A comparison of Pearl indices and cumulative incidences for use in
meta-analysis of contraception trials
P. O’Brien
Westside Contraceptive Services, London, UK
Objectives: One of the issues encountered when performing
meta-analysis of contraceptive trials is the variable way in which the outcomes
are reported, as cumulative rates (Kaplan Meier curves), single or multiple
decrement life-tables rates, or Pearl indices (events per 100 women-years). The
latter decrease with time, while the others increase. However, it is usually the
ratio of the rates that we use in meta-analysis. In this analysis we compare the
ratio of the Pearl rates to the ratio of the cumulative incidence and their
confidence intervals.
Design and methods: We used data from a large WHO randomised trial of
a frameless IUD and TCu380A, to compared the ratio of Pearl indices to the ratio
of cumulative incidences. We used the method described by Kleinbaum to calculate
the standard error of the ratio of cumulative incidences and the method of
Hasselblad for the standard error of the ratio of Pearl indices, to calculate
their confidence intervals.
Results: For accidental pregnancies the difference in the ratio of
Pearl ratios to ratio of incidence ratios was greatest at year 6 at 13% (Pearl
rate ratio 0.95, 95%CI 0.64 to 1.47; incidence rate ratios 0.85, 0.56 to 1.29).
In 4 of the 6 years of follow-up, the difference in ratios was less than 5%. The
difference was smaller for removals for bleeding and pain, never exceeding 3%
(Pearl ratio 1.10, 0.90 to 1.33; incidence ratio 1.13, 0.94 to 1.36 at 6 years).
For total use-related discontinuations, the difference never exceeded 4% (Pearl
ratio 1.26, 1.10 to 1.45; incidence ratio 1.22, 1.08 to 1.38 at 6 years).
Conclusions: The difference in rate ratios, whether we use the Pearl
indices or cumulative incidence rates, is usually small. The difference in rate
ratios from single and multiple decrement life-tables is likely to be smaller.
The ability to incorporate trials using different reporting methods in
meta-analyses enhances our capacity to systematically review the literature.
Further research is required to understand the determinants of the magnitude and
direction of the differences.