Open Source Steps Towards Predicting Preterm Birth

I always get excited about open source projects and OBGYN-related predictions. So, it is rather predictable that I am excited to feature the QUiPP app in this blog post.

The researchers, led by Ms. Kuhrt and Dr. Shennan of King’s College London, collected data on 1,249 women at high risk of spontaneous preterm birth from 2010 to 2014. Study link here. All of the pregnancies were singletons from 22 to 30 weeks gestation. The women were a part of the EQUIPP study, one of the largest efforts to quantify risk of preterm birth to date. What makes a pregnancy high risk for preterm birth? The authors included women who had a preterm birth in the past, preterm rupture of membranes in the past, a past miscarriage from 16 to 24 weeks of pregnancy, prior surgery on the cervix, a short cervix <25mm, and women presenting to the hospital with symptoms of preterm labor.

The researchers collected cervical length measurements from ultrasounds and fetal fibronectin (fFN) measurements. FFN is a common test on Labor and Delivery units that helps to predict risk of preterm delivery. If fFN is positive, it doesn’t tell us much, but if it is negative, it is very unlikely the woman will deliver in the next two weeks (high negative predictive value).

The authors divided the data from the participants in half. They used the first half of cases to create a model to predict spontaneous preterm birth. They then used the second half of the cases to test out whether their model worked. This is an excellent way to validate a proposed predictive model.  The authors predicted 5 outcomes: spontaneous preterm birth before 30, 34, and 37 weeks, and birth within 2 and 4 weeks of obtaining the cervical length and fFN. They found that their predictive model worked pretty well, and had a great negative predictive value of >90%, similar to the tests its uses for inputs.

The authors then took their research to the next level: they made it available to clinicians to use in their everyday practices. Doctors can download the QUiPP app on their phone or learn more at www.quipp.org. This is an important next step because the app now allows clinicians to give patients personalized risk scores, instead of just wide ranges from the literature.

For asymptomatic women, clinicians input the following information: fFN, cervical length, and whether she has had a spontaneous preterm birth in the past. For women with symptoms of preterm labor, clinicians input: fFN and whether the patient has a history of a spontaneous preterm birth in the past.

One small critique is that the app does not account for the impact of progesterone use on risk. Some of the women in the study were exposed to progesterone as part of another research study; others were not. We know that progesterone helps to prevent recurrent spontaneous preterm birth. Therefore, whether or not a woman receives progesterone would be a helpful factor to use in predicting her risk of spontaneous preterm birth. I hope that future iterations of the app will allow clinicians to input progesterone use in calculating the ultimate risk score.

One of the greatest challenges researchers face is turning all the predictive models they generate into usable tools for clinicians. In fact, this very problem was highlighted in a January 2016 Gray Journal article by Dr. Kleinrouweler, found here.  This group of researchers in the UK fulfills this need by creating the QUiPP app, so that clinicians everywhere can easily access their predictive model and patients can benefit directly from their research efforts.