Independent multicentre validation of the ‘six‐point’ model for malignant transformation risk in oral epithelial dysplasia

Background

Histological grading of oral epithelial dysplasia (OED) is used for predicting malignant transformation risk. However, grading is associated with significant subjectivity and observer variability leading to inconsistency in prognosis prediction. Alternate histological feature-specific models (‘six-point’ and ‘two-point’) have been shown to have potentially better prognostic reliability than conventional grading. This study conducts a multicentre validation of these models.

Methods

102 OED cases (dating 2012–2017) were acquired from 4 independent centres (13 (13%) Sheffield; 40 (39%) Belfast; 30 (29%) Birmingham; 19 (19%) Piracicaba, Brazil) were independently scored using the ‘six-point’ and ‘two-point’ models. Feature prevalence, observer agreement and malignant transformation risk were evaluated and compared to conventional grading systems.

Results

The ‘six-point’ system demonstrated superior predictive value (AUROC of 0.81) compared to the ‘two-point’ system (AUROC = 0.73, p = 0.004), WHO grading (AUROC = 0.71, p = 0.03) and binary grading (AUROC = 0.68, p = 0.009). Transformation rate for the ‘six-point’ model was 50% (95% CI 27%–78%) when all 6 features were present compared to 14% (95% CI 5%–32%) when 2–3 features were present.

Conclusions

This study supports the superior performance of the ‘six-point’ system for transformation prediction on a multicentric sample. Findings indicate that feature-specific models may be more reliable than existing histological grading systems for prognosis prediction.