Every value used by a calculator on this site comes from one of three places. It is taken from a published paper, from a professional guideline, or from a calibration choice recorded in the calculator's evidence register.
This page describes how a published estimate becomes a coefficient in a calculator, why the coefficient is usually smaller than the published estimate, and which inputs are excluded.
Defining the outcome
The first decision for each model is which event it measures. For the labor timing calculator, the event is spontaneous onset of labor: regular contractions beginning on their own, or rupture of the membranes, before any scheduled delivery. Most large datasets record the date of birth rather than the date labor began. A birth following a scheduled induction gives no information about when labor would have started on its own.
The evidence register therefore records which outcome each source measured. Smith (2001) analyzed non-elective deliveries and reported median gestational lengths of 284 days for first births and 282 days for later births. Those medians anchor the calculator's term curve, with delivery timing used as an approximation for onset timing. Marquette and colleagues (2014) measured spontaneous onset directly: among 15,253 uncomplicated pregnancies still ongoing at 41 weeks 0 days, 67.6% began labor spontaneously by 41 weeks 6 days. Both figures are used, and the register identifies them as measurements of different events.
Adapting published estimates
A consumer questionnaire collects less information than a research cohort. A study may adjust for many clinician-measured covariates. The calculator form asks a small number of questions that can be answered from memory. A coefficient transferred from the study to the form is applied to a coarser version of the variable the study measured.
For that reason, each adapted value is set at or below the published estimate. Ananth and colleagues (2006) reported an adjusted odds ratio of 3.6 for recurrent spontaneous preterm birth after a previous spontaneous birth before 35 weeks. The calculator applies 3.0, and applies it to a before-37-weeks outcome rather than the study's before-35-weeks outcome. Both adjustments reduce the claimed effect relative to the source.
The same rule applies to the length of a previous pregnancy. Ng and Steer (2016) found a statistically significant but weak association between successive gestational lengths, with a correlation of 0.10 and regression toward 280 days. The calculator translates this into a shift of about one day for a previous labor at 41 weeks.
Identifying calibration choices
Some model values are not reported in any source. The labor curve's term spread of 9.44 days was selected so that the curve reproduces both Smith's medians and Marquette's late-term figure. The register classifies this value as a calibration choice, separately from values taken directly from a source. The methodology page gives the classification for each model on the site.
The child height predictor illustrates a related point. The corrected Khamis-Roche equation was fitted to 223 boys and 210 girls from the Fels Longitudinal Study in southwest Ohio, all White. The calculator uses the equation because it is the most widely validated method that does not require a bone-age X-ray. It reports the source population on the result page so that a user can judge how well it applies to their child.
Excluded inputs
An input is excluded when its effect cannot be traced to a source. It is also excluded when the questionnaire cannot collect it accurately, or when including it would imply a biological claim that the evidence does not support.
Fetal sex was considered for the labor model and excluded because the reported associations with timing are small and vary with the outcome definition used. Interpregnancy interval was excluded because defining it accurately on a form requires additional questions and the resulting shift would be small. Race was excluded because it is not treated as a biological timing variable. Cervical examination findings and biomarkers were excluded because they cannot be supplied by the user.
For pregnancy dating, the calculator follows the American College of Obstetricians and Gynecologists' Committee Opinion 700 and does not average due dates from different sources. Obstetric dating selects one source by a priority order and reports the others alongside it. A blended date is not supported by either method.
Automated checks
The labor model has automated tests. They fail if the baseline preterm share changes from 5%, if the medians move from 284 and 282 days, if the late-term benchmark is not met, or if any cumulative probability decreases. Page copy has a separate test that fails on phrases the site has decided not to use. Both run on every change to the code.
The evidence registers are re-read at least once a year, and sooner when a guideline that a calculator relies on is revised. A correction received through the contact page is checked against the cited source before any change is made. A confirmed correction is applied and dated on the affected page.
Interpreting a result
A result from any calculator on this site is a population estimate conditioned on the answers entered and scaled to the current date. It describes what occurred across many pregnancies or children with the same answers. It uses no information beyond the form and none of the information available to a clinician. The range shown is the primary output. A single most likely date is shown alongside the range because users ask for one, and it should be read together with the range.