Prediction of natural conception in subfertile couples can help to differentiate between couples who should have immediate treatment and couples who can aim for natural conception for some time. Natural conception rates are often estimated using standard techniques such as Kaplan–Meier or Cox proportional hazard models. These estimates can be biased by incorrect handling of data from women who start assisted reproductive technology therapy before the end of the follow-up period. This paper discusses the validity and the impact of the assumption of non-informative censoring as used in the Kaplan–Meier and Cox models.METHODS
In a cohort of 5360 subfertile couples with suspected tubal pathology, the probability of natural conception and the prognostic value of additional tests for tubal pathology were estimated using traditional methods and with a competing risks analysis.RESULTS
The estimated probability of natural conception within 3 years was almost 2-fold higher when assuming non-informative censoring compared with the competing risks model, 41 versus 22%. The prognostic value of tests was more conservative using the competing risks model than with the traditional methods, the fecundity rate ratio for Chlamydia antibody testing was 0.72 versus 0.67, for hysterosalpingography, 0.83 versus 0.71 and for diagnostic laparoscopy, 0.89 versus 0.74.CONCLUSIONS
Given the improbable validity of the non-informative censoring assumption, the predictions of natural conception and of the prognostic value of tests are likely to be overestimated by the traditional analytic methods. We suggest the use of competing risks models as an alternative, more conservative, form of analysis when predicting natural conception and evaluating prognostic fertility tests.