Beyond time alone — interactive exploration of the 7-factor model that reduced prediction errors by >50%
Source: Hamilton EF, Zhoroev T, Warrick PA, Tarca AL, Garite TJ, Caughey AB, et al. New labor curves of dilation and station to improve the accuracy of predicting labor progress. Am J Obstet Gynecol. 2024 Jul;231(1):e1–18. n=8,022 nulliparous vaginal deliveries, 10 US hospitals, external validation (n=527).
The 7 Variables That Predict Dilation Better Than Time Alone
Traditional labor curves (Friedman, Zhang) use only the passage of time. Hamilton et al. demonstrated that incorporating 7 clinically relevant factors reduces prediction error by more than half — from MAE 2.12 cm to 0.83 cm.
From the Gaussian process model kernel analysis. Lower number = more important. Contractions and previous dilation consistently ranked highest.
Interactive Labor Progress Simulator
Adjust the 7 factors to see how the expected dilation curve and prediction interval change. The multifactor model produces a narrower, more individualized prediction band.
Patient Factors
6.0 hrs
3.0 cm
50%
-2
60
Multifactor expected dilation
90% prediction interval (5th–95th)
Time-only model (wide band)
Current observed dilation percentile: —
5th %ile (slow)50th (median)95th %ile (fast)
Head-to-Head: Single Factor vs. Multifactor Models
Prediction error comparison across modeling approaches. Lower is better. All P-values <.0001.
Model
Method
RMSE (cm)
MAE (cm)
Improvement
Single factor (time only)
Polynomial / ML
2.504
2.122
—
Multifactor 7 variables
Mixed-effects
1.172
0.893
53% / 58%
Multifactor 7 variables
Machine Learning
1.126
0.826
55% / 61%
Multifactor without time
Machine Learning
1.126
0.826
55% / 61%
Station (7 variables)
Machine Learning
0.660
0.512
—
External Validation (Independent Dataset, n=527)
Oregon/Washington hospitals — different geography, similar results confirming generalizability.
Model
RMSE (cm)
MAE (cm)
Single factor (time only)
2.661
2.283
Multifactor 7-var (Mixed-effects)
1.378
1.020
Multifactor 7-var (Machine Learning)
1.352
0.946
Key Finding: Time-Only Models Are Equally Poor Regardless of Method
When forward-time is the only variable, both polynomial regression and machine learning produce identical prediction errors (RMSE 2.504, MAE 2.122). Machine learning provides no advantage with a single variable. The improvement comes only when multiple clinically relevant factors are incorporated. This means the problem with traditional labor curves is not the statistical method — it is the underlying assumption that time alone can predict labor progress.
Induced vs. Spontaneous Labor — Model Performs Well for Both
The multifactor model adjusts automatically. Induction is one of the 7 variables.
Group
RMSE (cm)
MAE (cm)
Induced labor (69%)
1.140
0.838
Spontaneous labor (31%)
1.087
0.794
Difference is <1 mm — clinically imperceptible. Traditional labor curves were derived only from spontaneous labor and do not account for the 69% of modern births involving induction.
Why Traditional Labor Curves Fail — And What Replaces Them
📏
The Precision Problem
A time-only model predicting 7 cm with a 90% interval of 4–10 cm covers nearly all possible dilations. This is clinically useless. The multifactor model narrows this to roughly ±1 cm — actionable precision that matches inter-examiner variability.
🔄
Backward vs. Forward Time
Zhang's curves use "negative-time" (counting backward from complete dilation). This creates beautiful retrospective curves but is impossible to use prospectively — you don't know when full dilation will occur during labor. Hamilton uses forward-time: hours since first exam, which is knowable in real time.
💊
The Induction Problem
69% of births in this study involved labor induction — reflecting modern practice. Traditional curves were derived from spontaneous labor only. A model that ignores induction status is ignoring the majority of contemporary labors.
📊
Clinician Non-Compliance
Only 31% of cesareans for dystocia at ≥6 cm followed Zhang-based guidelines in a cluster RCT — and two such RCTs showed no reduction in cesarean rates. Clinicians intuitively consider factors beyond time, which is exactly what the multifactor model formalizes.
🧠
Contractions Are Key
Cumulative contraction count was the most important variable in the smoothness (RBF) kernel. A woman at 4 cm after 200 contractions is fundamentally different from a woman at 4 cm after 50 contractions — yet time-only models treat them identically.
⚡
Real-Time Implementation
The model runs in <1 second on a standard laptop. All inputs can be extracted automatically from EMRs and electronic fetal monitoring systems. An FDA-cleared implementation already exists (PeriWatch Cues). This is not theoretical — it is operational today.
The Bottom Line
The diagnosis of "failure to progress" is the #1 indication for intrapartum cesarean delivery. It is currently based on labor curves that use only time, were derived from spontaneous labors, and have prediction errors exceeding 2 cm. These guidelines have failed to reduce cesarean rates in two large RCTs, and clinicians override them 70% of the time.
Hamilton et al. show that incorporating 7 factors that clinicians already consider — contractions, prior dilation, effacement, station, epidural status, induction status, and time — cuts prediction error by more than half. The result is an individualized, percentile-based assessment of labor progress that is more accurate, more relevant, and more aligned with how clinicians actually think.
Disclaimer: This interactive tool is for educational purposes only, illustrating the concepts from Hamilton et al. 2024. The simulated curves are approximations based on published data and do not replicate the actual Gaussian process or mixed-effects models. Clinical decisions should use validated tools. The actual multifactor model is implemented in PeriWatch Cues (PeriGen, Inc.) and requires real-time EFM and EMR data.