Preterm delivery in patients admitted with preterm labor: a prediction study

To develop a model for prediction of preterm delivery in patients treated with parenteral tocolysis using combinations of maternal demographic and clinical factors.

We performed a retrospective cohort study using a perinatal database to identify women admitted with preterm labor and treated with parenteral tocolysis from 1980 to 1994. We developed an explanatory model using multiple logistic regression to determine the effect of four variables (prior preterm delivery, substance abuse, maternal complications and third-trimester care) on the likelihood of preterm delivery. For the prediction model, we initially included these four variables and then removed them in a stepwise fashion to determine the combination of the variables that offered the greatest model sensitivity and specificity.

A total of 900 women were identified for the study and 247 (27%) had a preterm delivery. In the explanatory model, prior preterm delivery (OR 2.4; 95% CI 1.5-3.6), substance abuse (OR 2.2; 95% CI 1.2-5.1), initiation of care in the third trimester (OR 2.0; 95% CI 1.3-2.8) and medical complications of pregnancy (OR 1.8; 95% CI 1.2-2.6) increased the likelihood of preterm delivery. For the prediction tool, a three-variable model (prior preterm delivery, substance abuse and initiation of care in the third trimester) had high specificity (98%) and modest negative predictive value (73%).

A simple three-variable model can correctly identify 98% of women with preterm labor treated with parenteral tocolysis who will not deliver preterm. Patients with no prior history of preterm delivery, no substance abuse and initiation of prenatal care before the third trimester have a 73% probability of not delivering preterm.