Impact of analysis technique on our understanding of the natural history of labour : a simulation study

Objective: To evaluate the discrepancy between historical and more recent descriptions of the first stage of labour by testing whether the statistical techniques used recently (repeated-measures polynomial and interval-censored regression) were appropriate for detection of periods of rapid acceleration of cervical dilatation as might occur at the time of transition from a latent to an active phase of labour. Design and setting: A simulation study using regression techniques. Sample We created a simulated data set for 500 000 labours with clearly defined latent and active phases using the parameters described by Friedman. Additionally, we created a data set comprising 500 000 labours with a progressively increasing rate of cervical dilatation. Methods: Repeated-measures polynomial regression was used to create summary labour curves based on simulated cervical examinations. Interval-censored regression was used to create centimetre-by-centimetre estimates of rates of cervical dilatation and their 95th centiles. Main outcome measures Labour summary curves and rates of cervical dilatation. Results: Repeated-measures polynomial regression did not detect the rapid acceleration in cervical dilatation (i.e. acceleration phase) and overestimated lengths of labour, especially at smaller cervical dilatations. There was a two-fold overestimation in the mean rate of cervical dilatation from 4 to 6 cm. Interval-censored regression overestimated median transit times, at 4- to 5-cm cervical dilatation or when cervical examinations occurred less frequently than 0.5- to 1.5-hourly. Conclusion: Repeated-measures polynomial regression and inter-valcensored regression should not be routinely used to define labour progress because they do not accurately reflect the underlying data.