Drop-out behaviour and fertility table analysis of pregnancy rates

Abstract

The life table analysis of infertility study data uses information from patients who drop out and are lost to follow-up. The fact that drop-out patients were not pregnant when last seen has been taken to be useful information that has been entered into the calculation of cumulative pregnancy rates for the population as a whole. However, drop-out patients who subsequently achieve pregnancy have motivations to return to their physician for follow-up, while those who do not become pregnant usually remain lost to follow-up. This can result in a systematic error toward reporting an excess of pregnancies, raising the apparent fecundity rates and plateau values. This study was undertaken to investigate the potential magnitude of these errors. Computer simulations of hypothetical infertility treatments using large cohorts (n = 32 000) were undertaken. Varying assumptions were made regarding post-treatment monthly fecundity rates, ‘cure rates’, dropout rates and the percentage of pregnant drop-outs who would return to their physician after achieving pregnancy. Using a range of clinically reasonable assumptions, very large deviations from ideal behaviour were noted in the direction of elevated cumulative pregnancy rates. The non-ideal behaviour was most apparent with lower fecundity rates, lower ‘cure rates’, higher drop-out rates and higher pregnant drop-out return rates. A change in the method of data collection to minimize the impact of this bias in infertility studies is proposed. This method of active data collection is a departure from that used in classical life table studies and therefore the method is titled the ‘fertility table method’ in order to avoid confusion.