PAX1/JAM3 Methylation Predicts Pathological Upgrading Before Conization

1Background & Objective

Problem: colposcopy biopsy vs conization pathology often differ — upgrading 23.1%, downgrading 33.6%.
Consequences: underdiagnosis delays treatment; overdiagnosis causes unnecessary surgery (pregnancy risks).
Objective: use non-invasive PAX1/JAM3 methylation to predict pathological upgrading before conization.

2Study Design & Cohort

549

collected

461

excluded

analyzed

23.1%

upgrade (lit.)

33.6%

downgrade

Dec 2020→ colposcopy + methylation→ Apr 2022 · conization

Colposcopy biopsy+ Methylation (ΔCt PAX1/JAM3)+ Conization pathology

Setting: Zhejiang Provincial People's Hospital, Dec 2020 – Apr 2022; two pathologists.
Upgrading: CIN1→CIN2/3+ · CIN2→CIN3/cancer · CIN3→cancer.
Assay: CISCER® real-time PCR kit (NMPA III), GAPDH internal control.
Analysis: univariate + multivariate logistic; ROC/Youden; calibration + DCA.
Pathology: 2020 WHO; highest grade; two pathologists.

Exclusion criteria
① CDB cancer / AIS / adenocarcinoma
② conization elsewhere / unable
③ lost to follow-up
④ prior cervical treatment / hysterectomy / CRT

CDB CIN1→CIN2 / CIN3 / cancer

CDB CIN2→CIN3 / cancer

CDB CIN3→cervical cancer

Upgrading = conization pathology one grade higher than colposcopy biopsy (2020 WHO).

3Prediction Model — ROC

0.818

AUC (0.720–0.916)

94.4%

Specificity

60%

Sensitivity

ROC curve of the clinical prediction model

Fig. 4 ROC — ΔCt PAX1 + cervical canal lesions predict upgrading; good discrimination.

Fig. 3 Logistic prediction model components.

ΔCt < 4.34

PAX1 cut-off (max Youden)
lower ΔCt PAX1 → higher upgrading risk

Model: ΔCt PAX1 + cervical canal lesions; independent predictors on multivariate analysis.

4Model in Practice

Step 1

Measure ΔCt PAX1 pre-op

→

Step 2

ΔCt < 4.34 → upgrading likely

→

Step 3

Plan wider / deeper conization

0.818

Model discrimination
Se 60% · Sp 94.4% at max Youden · OR 0.035 (P=0.02)

Cut-off ΔCt PAX1 = 4.34 defined by maximizing Youden index.

ΔCt ≥ 4.34: upgrading less likely — may defer or narrow surgery.
Goal: right-size treatment, avoid under- and over-treatment.
Any candidate for conization with pre-op methylation can be stratified.

4Independent Risk Factors

ΔCt PAX1

OR 0.784

95% CI 0.644–0.956 · P=0.016

Cervical canal lesion

OR 3.469

95% CI 1.014–11.870 · P=0.048

ΔCt JAM3 and PAX1 levels of upgrade vs not-upgrade

Fig. 2 ΔCt PAX1/JAM3 — not-upgrade group significantly higher (P<0.01).

5Model Validation

Fig. 5 Calibration — mean abs error 0.031 (bootstrap ×1000).

Fig. 6 DCA — net benefit across 0–0.3 thresholds.

Discrimination: AUC 0.818 — good ability to distinguish upgrading.
Calibration & utility: predicted vs actual agree (err 0.031); DCA favors model over treat-all/none.

Well-calibrated and clinically useful across decision thresholds.