Effect of MPATH‐Dx 2.0 on Diagnostic Agreement in the Histopathologic Assessment of Melanocytic Skin Lesions

Background The revised MPATH‐Dx V2.0 schema simplified the original 5‐class hierarchy to 4 classes. Lesions previously classified as class II (e.g. moderately dysplastic naevus) in V1.0 are now classified as either class I (e.g., dysplastic naevus, low‐grade) or class II (e.g., dysplastic naevus, high grade; melanoma in situ). Objectives To estimate between pathologist agreement using MPATH‐Dx V1.0 and MPATH‐Dx V2.0 among four dermatopathologists assessing paired clinical and histopathology images of melanocytic lesions, blinded and unblinded to each other's diagnoses. Methods In this cross‐sectional reading study, dermatopathologists assessed cases from a curated set of paired clinical and histopathology digital images of 98 melanocytic skin lesions. We compared diagnostic agreement using the two MPATH‐Dx tools across three scenarios. First, we compared tools for independent review of the 98 cases (blinded review). Second, we compared tools for a subset of 51 cases where there was no consensus using MPATH‐Dx V1.0 (blinded review). Third, we compared tools when the same 51 cases were re‐read with knowledge of the other pathologists' diagnoses (unblinded review). Results In the blinded review of 98 cases, there was a modest increase in overall between‐pathologist agreement from using MPATH‐Dx V1.0 (average κ 0.52) to using MPATH‐Dx V2.0 (average κ 0.58). For MPATH‐Dx V2.0, the pair‐wise agreement for classes I and II were 69% (95% CI 62%–76%) and 78% (72%–83%), respectively. The increase in overall agreement was larger for the subset of 51 more difficult cases (blinded review): MPATH‐Dx V1.0, average κ 0.14; MPATH‐Dx V.2.0, average κ 0.31. When these 51 cases were re‐read in an unblinded setting, MPATH‐Dx V2.0 led to an additional small increase in overall agreement beyond unblinding effects: MPATH‐Dx V1.0, average κ 0.4; MPATH‐Dx V2.0, average κ 0.46. Conclusions The revised MPATH‐Dx V2.0 four‐class schema has the potential to improve diagnostic agreement, especially in difficult borderline cases.