Digital image analysis of tumour pattern and histological models for prognostic evaluation of invasive non-mucinous adenocarcinoma of the lung

Abstract

The 2021 World Health Organisation classification of lung adenocarcinoma is based on the predominance and percentage of high-grade histological patterns, e.g. solid and micropapillary patterns, determined by semiquantitative estimation. Digital pathology can be used to evaluate the area of each pattern and calculate the exact percentage. To evaluate the prognostic predictive ability of a histological model for invasive non-mucinous adenocarcinoma using digital pathology. This retrospective cohort study included 76 patients with invasive non-mucinous lung adenocarcinoma who underwent lung resection at Songklanagarind Hospital between January 2010 and December 2016. The histological pattern area was measured on a digital slide using the QuPath Open software version 0.3.2. Clinical and pathological data, including the presence of tumour spread through airspaces, tumour necrosis, tumour-infiltrating lymphocytes, and lymphovascular invasion, were collected. The primary outcome was 5-year overall survival. The best model was provided by the Akaike information criterion, and the prognostic discrimination ability was compared with that of other models from previous studies by identifying the area under the curve (AUC) in the receiver operating characteristic analysis. The best model was validated using bootstrapping. The best model was a combination of stage and an 82 % cut-off high-grade pattern (AUC = 0.776). Tumours with ≥82 % high-grade pattern resulted in significantly worse prognoses (p = 0.001) than those with <82 % high-grade pattern. Our model had the highest AUC among all models from previous studies. This was validated using bootstrapping, with an AUC of 0.708. The best model for survival prediction was a combination of stage and an 82 % cut-off high-grade pattern.