4P-model: A machine learning-driven approach to expand non-invasive prenatal testing for early risk stratification of placenta accreta spectrum via genome-wide cfDNA promoter profiling
In brief
A 15-gene NIPT model stratified placenta accreta risk with AUCs up to 0.97
In a multicenter study of 1,060 people with prior uterine surgery, a model using routine prenatal blood-test data distinguished placenta accreta, more invasive forms, and controls, with AUCs of 0.882 to 0.973 across study cohorts. The results suggest a route to earlier risk assessment, but the model needs further validation before its value in routine care is clear.
- Journal
- Genes & diseases (Q1)
- Published
- 13 June 2026
- Study design
- Case-control study
- Evidence level
- Level 3, Low (CEBM 3b)
- Authors
- Runrun Hao, Beilun Wang, Bin Zhang, Yixuan Li, Qingwen Zhu, Shilin Zhu, et al.
- PMID
- 43061468
- DOI
- 10.1016/j.gendis.2026.102305
Why clinicians should know about it
- Picked for Obstetrics and Gynecology (paper of the day, 22 November 2026): Early prediction of placenta accreta spectrum and obstetric haemorrhage risk
Abstract
Placenta accreta spectrum (PAS) is a leading cause of life-threatening obstetric hemorrhage, yet early diagnosis remains challenging. Traditional non-invasive prenatal testing (NIPT) is limited to fetal aneuploidy screening, while its potential for maternal placental disorders is underexplored. We conducted a multicenter nested case-control study to investigate whether machine learning-driven genome-wide cell-free DNA (cfDNA) promoter profiling, derived from routine NIPT data, could extend the utility of NIPT to enable early prediction and subtype risk stratification of PAS. NIPT data were analyzed from 1060 individuals with a history of uterine surgery (126 increta/percreta, 139 accreta, and 795 controls after propensity score matching). A training set of 280 participants was used for feature selection with recursive feature elimination across five machine learning algorithms, and the best-performing model was validated in an internal cohort, a PLUS cohort with deeper sequencing, and two external cohorts generated on independent platforms. Pairwise differential analyses identified 211 candidate genes with altered promoter coverage. After feature selection, we developed the 4P-model, a 15-gene multiclass risk stratification model, to differentiate increta/percreta, accreta, and controls with a macro-average AUC of 0.973. Robust performance was reproduced across three NIPT platforms and the PLUS cohort, with AUCs of 0.882-0.946. These findings support that machine learning facilitates the translation of genome-wide cfDNA promoter signals embedded in NIPT data into early prediction and risk stratification of PAS. This noninvasive and scalable approach extends the diagnostic window, offering a clinically feasible tool for timely referral, improving perinatal outcomes.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.