There is a robust discussion in the recent literature about prediction of mortality in pregnant and postpartum women.
Other researchers and I published an article this year in the American Journal of Obstetric and Gynecology. We validated the Early Warning Score designed by Carle et al. for pregnant and postpartum patients. Despite that study is a very good first attempt to develop a prediction tool to predict mortality in these patients, it feels short in methods, which I will detail below. But before this, I would like to clarify that these are my views alone, and in no way it compromises the view of the other authors of the paper published in the American Journal of Obstetric and Gynecology.
The paper by Carle et al. tries to develop a method to create a tool to predict which obstetric patients are at most risk of maternal death. But this paper does not address several methodological shortcomings, does not adjust for the linearity assumption of logistic regression, and only use the statistical analysis to inform their experts in the development of the Early Warning Score.
Let me give an example for the lack of checking of the linearity assumption of the logistic regression they ran, and why this is important. There are several clinical and laboratory parameters that are not linear in their relationship with the outcome, meaning, some variables have a higher probability of the outcome in higher and lower values of the continuous predictor variables. This is true for blood pressure, respiratory rate, temperature, and many, many more clinical and laboratory parameters. An example we put in the paper, all diastolic blood pressures below 90 mm Hg are normal, most clinicians know this is not the case, as a death person has a blood pressure of 0 mm Hg!! Kidding aside, most people as they approach to death or in very severe disease, have a low diastolic blood pressure, as in the case of obstetric hemorrhage, the most important cause of death in pregnant and postpartum women in the world.
You can see the Carle et al. tool in-action in this link.
I think there is much room for improvement in this analysis. In Colombia with other researchers, we are trying to predict mortality in a large database of obstetric patients, and are going to publish these results soon.
We hope to get the tool right this time…!