Authors' reply


We have read the letter by Hermans et al.[1] regarding our recently published paper[2] and appreciate their comments and the opportunity to respond. We acknowledge that the current paper does not lend itself well to assessing longitudinal change in the rate of different labour onset types. We are preparing a paper using a counter-factual framework to assess the independent contribution of pre-existing medical conditions and pregnancy complications on preterm rates, which we plan to submit in the coming weeks. This study will probably address their concerns regarding the stability of the odds ratio estimate; we felt that presentation of the odds ratios at separate time-points was not feasible within the current paper. However, additional analyses also showed that there was little change in the calculated odds ratio for pre-existing medical complications when predicting preterm birth at different time periods, indicating that the increase occurred in both term and preterm populations. Although a rate increase in previous medical conditions was observed for medically indicated births overall, the increase was actually greatest in the term as opposed to the preterm population.

The supposition that the increase in medically indicated births is mainly due to the increase in pre-existing medical disease may well be justified. However, an important factor may be changes in clinical practice and perception of risk. Western Australian research has shown that private obstetric practice is responsible for much of the increase in prelabour caesarean section rates over the past few years.[3] Either women of higher risk are self-selecting to attend private obstetricians for reasons of continuity of care, or different thresholds for intervention exist in private and public services. We feel that both changes in risk factors and obstetric practice are contributing, and hope that the aforementioned work in progress will go some way to addressing this.

With regard to our finding that 50% of medically indicated deliveries could be removed by the elimination of six antecedents, this estimate represented the highest probable value and, as stated, made the assumption that all antecedents were causal, which is highly unlikely. These antecedents are neither sufficient nor necessary predictors of preterm birth: preterm births occur in their absence and term births in their presence. Recent research has also shown that early detection and better clinical management can reduce the rates of these problems or at least reduce the risk of a preterm birth given their occurrence.[4] As such, we feel they represent modifiable risk factors rather than being deterministic steps to an inevitable conclusion.

We acknowledge the suggestion of a stepwise model. While we agree with the approach, we wonder how one would approach the problem of interactions? It is likely that the existence of a baseline risk factor may differentially exacerbate the risk of another factor at a subsequent time period. An overall baseline risk would be a composite measure, but any particular information regarding specific antecedents would be lost. An approach such as latent classification or latent class analysis provides the ability to separate out the most influential co-occurrences, but the interpretation can be complex and potentially difficult for a general readership. However, we will explore this approach and welcome the suggestion.


  1. Top of page
  2. References