Evidence suggests a sharp increase in stillbirth at the end of gestation. Between 35 and 38 weeks of gestation, cumulative risk of stillbirth in twin pregnancies increases by 60%, from 30 to 50 deliveries per 1000.1 A similar increase occurs after 37 weeks in singleton pregnancies.2 Although this evidence has implications for clinical practice, it is misleading. Increases in the cumulative risk of stillbirth at the end of gestation occur because researchers use the KaplanMeier method, which fails to account for live births as competing events.In studies of stillbirth, women can have three possible outcomes: stillbirth, live birth, or right-censoring. Rightcensoring occurs when a woman is no longer followed in a study (because of dropout or loss-to-follow-up), but is still at risk for the event. When a live birth occurs, the woman is no longer followed in a study because she is no longer at risk for the event. For this reason, live births (whether preterm or term) are competing events because they eliminate the possibility of stillbirth.3 Competing events are recognised in many areas of research. 4,5 In obstetric studies of stillbirth, competing events are arguably more important because of the natural increase in live births near term. However, to our knowledge, only one letter in the obstetrics literature has raised concern over inappropriate treatment of competing events. 6 The purpose of this commentary is to clarify why the Kaplan-Meier approach should not be used when competing events are present. To illustrate, we calculate the incorrect risk of stillbirth in a cohort of roughly two million pregnancies in Quebec, Canada. We describe the degree of error, explain why the error occurs, and show how to obtain correct estimates of cumulative risk when competing events are present. We conclude with a mention of the broader debate on how best to define the cumulative risk of stillbirth.
Kaplan-Meier versus cumulative incidenceThere are several ways to compute cumulative risk. The Kaplan-Meier method 7 is the most common. The KaplanMeier method correctly estimates cumulative risk when there is right-censoring. 8 Women who are right-censored are still at risk of stillbirth even though they are lost to follow up, but the exact timing and outcome of pregnancy is not known.The Kaplan-Meier approach is incorrect when competing events are present. [3][4][5][6][8][9][10][11] To censor live births, the Kaplan-Meier method 'redistributes' the potential stillbirth risk for a given pregnancy to subsequent stillbirths from other pregnancies. In doing so, the Kaplan-Meier method requires the assumption that women with live births are still at risk of stillbirth, which evidently is impossible. As a result, when there is no loss to follow up but competing events are present, the Kaplan-Meier stillbirth risk estimated after all pregnancies have delivered is not identical to the overall (empirical) risk of stillbirth in the cohort.An alternative approach is known as the cumulative incidence method, which can estimate cumulative st...