Design: Prognostic model development with temporal validation, national registers
Population: 8,909 primiparous women with second- to fourth-degree tears, vaginal delivery at 34 weeks or later
Primary outcome: Anal incontinence (gas and/or faecal) at one year postpartum
Effect: AUC 0.78 (95% CI 0.76 to 0.80); temporal validation AUC 0.78 (0.75 to 0.81)

Researchers used three Swedish registers to build a prediction model in 8,909 first-time mothers with second-, third- or fourth-degree tears after vaginal delivery at 34 weeks or later. The outcome was anal incontinence to gas or stool at one year. They then tested it on a separate later cohort delivered in 2020 to 2022.

Seven variables stayed in the final model. A high Wexner score recorded at the 8-week perineal assessment was the strongest predictor. Discrimination was good for anal incontinence overall (AUC 0.78, 95% CI 0.76 to 0.80), faecal incontinence (0.80) and gas incontinence (0.78), and held in temporal validation (0.78, 0.75 to 0.81). Calibration was best after third-degree tears.

This is a development and temporal validation in one national system. It has not been tested in other populations or shown to change outcomes, so it is a tool for structured follow-up conversations rather than a reason to change management yet.

Assess symptoms with a validated score such as Wexner at the early postnatal perineal review.
Women with high early scores are the group most likely to still have symptoms at one year; arrange closer follow-up.
Offer pelvic floor physiotherapy and referral on symptoms, not on tear grade alone.
Document the grade of tear and early symptoms clearly so later risk can be estimated.
Use the model for counselling only after external validation in your population.

Why it matters

Follow-up after perineal tears is often uniform; this supports targeting it by early symptoms.

The statistics, in plain English

An AUC of 0.78 means that, given one woman who develops incontinence and one who does not, the model ranks the first as higher risk about 78% of the time. That is useful for sorting groups but still misclassifies many individuals.