Back to Calculator

Methodology

How we calculate labor probabilities

Statistical Model

We use a mixture of two skewed normal distributions to model the probability of spontaneous labor onset. This approach captures both preterm deliveries (before 37 weeks) and term deliveries more accurately than a single distribution.

The mixture model combines:

• Preterm component (~11% weight): Models early deliveries with a center around 36 weeks

• Term component (~89% weight): Models the majority of deliveries, peaking just after the due date

Data Sources

Our model parameters are fitted to match official maternity statistics from:

Model Accuracy

The fitted model achieves the following statistics:

10.3%

Preterm rate

Day +0.2

Median delivery

48.6%

Deliver by due date

Conditional Probability

All probabilities shown are conditional on not having delivered yet. This means we recalculate based on the fact that you're still pregnant. For example, if you're at 39 weeks, we only consider the remaining probability distribution from 39 weeks onward.

P(labor on day X | still pregnant) = P(labor on day X) / P(labor after today)

Personalization Factors

Research shows several factors can shift the expected delivery date by a few days. We apply these as location adjustments to the distribution:

Parity

First-time mothers tend to deliver ~2 days later

Bergsjø et al. (2001) - BJOG

Fetal Sex

Boys arrive ~3 days earlier on average

Divon et al. (2002) - Am J Obstet Gynecol

BMI

Higher BMI associated with later delivery (up to +5 days)

Denison et al. (2008) - BJOG

Interpregnancy

Short intervals (<12 months) associated with earlier delivery

Shachar et al. (2017) - Obstet Gynecol

Limitations

These probabilities are for spontaneous labor onset only. They do not account for:

• Scheduled inductions or cesarean sections

• Medical conditions that may affect timing

• Individual variation beyond the modeled factors

• Research shows individual prediction has limited accuracy (c-statistic ~0.56)

References

Key research informing this model:

Built with Reflex