ABSTRACT
BACKGROUND: Fine-needle aspiration (FNA) is a robust diagnostic technique often used for tissue diagnosis of metastatic carcinoma. For interpretation of FNA cytology, cell block immunohistochemistry (IHC) and clinicocytologic parameters are indispensable. In this review of a large cohort, the current report: 1) describes clinicocytologic parameters and immunoprofiles of aspirates of metastatic carcinoma, 2) compares the predictivity of immunostains and classical approaches for IHC interpretation, and 3) describes machine learning-based algorithms for IHC interpretation. METHODS: Aspirates of metastatic carcinoma that had IHC performed were retrieved. Clinicocytologic parameters, IHC results, the corresponding primary site, and histologic diagnoses were recorded. By using machine learning, decision trees for predicting the primary site were generated, their performance was compared with 2 human-designed algorithms, and the primary site was suggested in the historical diagnosis. RESULTS: In total, 1145 cases were identified. The 6 most populated groups were selected for machine learning and predictive analysis. With IHC input, the decision tree achieved a concordance rate of 94.5% and overall accuracy of 83.6%, which improved to 95.3% and 85.8%, respectively, when clinical data were incorporated and exceeded the human-designed IHC algorithms (P < .001). The historical diagnosis was more accurate unless indeterminate diagnoses were regarded as discordant (P < .001). CDX2 and TTF-1 immunostains had the highest weight in model accuracy, occupied the root of the decision trees, scored higher as features of importance, and outperformed the predictive power of cytokeratins 7 and 20. CONCLUSIONS: Cytokeratins 7 and 20 may be superseded in immunostaining panels, including organ-specific immunostains such as CDX2 and TTF-1. Machine learning generates algorithms that surpasses human-designed algorithms but is inferior to expert assessment integrating clinical and cytologic assessment.
Subject(s)
Carcinoma , Algorithms , Carcinoma/pathology , Humans , Immunohistochemistry , Keratin-7 , Machine LearningABSTRACT
Adequate empirical antimicrobial coverage is instrumental in clinical management of community-onset Enterobacteriaceae bacteraemia in areas with high ESBL prevalence, while balancing the risk of carbapenem overuse and emergence of carbapenem-resistant organisms. It is unknown whether machine learning offers additional advantages to conventional statistical methods in prediction of ESBL production. To develop a validated model to predict ESBL production in Enterobacteriaceae causing community-onset bacteraemia. 5625 patients with community-onset bacteraemia caused by Escherichia coli, Klebsiella species and Proteus mirabilis during 1 January 2015-31 December 2019 from three regional hospitals in Hong Kong were included in the analysis, after exclusion of blood cultures obtained beyond 48 h of admission. The prevalence of ESBL-producing Enterobacteriaceae was 23.7% (1335/5625). Deep neural network and other machine learning algorithms were compared against conventional statistical model via multivariable logistic regression. Primary outcomes compared consisted of predictive model area under curve of receiver-operator characteristic curve (AUC), and macro-averaged F1 score. Secondary outcomes included sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Deep neural network yielded an AUC of 0.761 (95% CI 0.725-0.797) and F1 score of 0.661 (95% CI 0.633-0.689), which was superior to logistic regression (AUC 0.667 (95% CI 0.627-0.707), F1 score 0.596 (95% CI 0.567-0.625)). Deep neural network had a specificity of 91.5%, sensitivity of 37.5%, NPV of 82.5%, and PPV of 57.9%. Deep neural network is superior to logistic regression in predicting ESBL production in Enterobacteriaceae causing community-onset bacteraemia in high-ESBL prevalence area. Machine learning offers clinical utility in guiding judicious empirical antibiotics use.
