Hierarchical clustering analysis identifies metastatic colorectal cancers patients with more aggressive phenotype

Progression Risk Score Estimation Based on Immunostaining Data in Oral Cancer Using Unsupervised Hierarchical Clustering Analysis: A Retrospective Study in Taiwan

This study aimed to investigate whether the progression risk score (PRS) developed from cytoplasmic immunohistochemistry (IHC) biomarkers is available and applicable for assessing risk and prognosis in oral cancer patients. Participants in this retrospective case-control study were diagnosed between 2012 and 2014 and subsequently underwent surgical intervention. The specimens from surgery were stained by IHC for 16 cytoplasmic target markers. We evaluated the results of IHC staining, clinical and pathological features, progression-free survival (PFS), and overall survival (OS) of 102 oral cancer patients using a novel estimation approach with unsupervised hierarchical clustering analysis. Patients were stratified into high-risk (52) and low-risk (50) groups, according to their PRS; a metric consisting of cytoplasmic PLK1, PhosphoMet, SGK2, and SHC1 expression. Moreover, PRS could be extended for use in the Cox proportional hazard regression model to estimate survival outcomes with associated clinical parameters. Our study findings revealed that the high-risk patients had a significantly increased risk in cancer progression compared with low-risk patients (hazard ratio (HR) = 2.20, 95% confidence interval (CI) = 1.10-2.42, p = 0.026). After considering the influences of demographics, risk behaviors, and tumor characteristics, risk estimation with PRS provided distinct PFS groups for patients with oral cancer (p = 0.017, p = 0.019, and p = 0.020). Our findings support that PRS could serve as an ideal biomarker for clinical use in risk stratification and progression assessment in oral cancer.

Identification of Markers Predicting Clinical Course in Patients with IgG4-Related Ophthalmic Disease by Unbiased Clustering Analysis

Purpose: To describe the clinical features of patients with immunoglobulin G4 (IgG4)-related ophthalmic disease (IgG4-ROD) grouped by unbiased cluster analysis using peripheral blood test data and to find novel biomarkers for predicting clinical features. Methods: One hundred and seven patients diagnosed with IgG4-ROD were divided into four groups by unsupervised hierarchical cluster analysis using peripheral blood test data. The clinical features of the four groups were compared and novel markers for prediction of clinical course were explored. Results: Unbiased cluster analysis divided patients into four groups. Group B had a significantly higher frequency of extraocular muscle enlargement (p < 0.001). The frequency of patients with decreased best corrected visual acuity (BCVA) was significantly higher in group D (p = 0.002). Receiver operating characteristic (ROC) curves for the prediction of extraocular muscle enlargement and worsened BCVA using a panel consisting of important blood test data identified by machine learning yielded areas under the curve of 0.78 and 0.86, respectively. Clinical features were compared between patients divided into two groups by the cutoff serum IgE or IgG4 level obtained from ROC curves. Patients with serum IgE above 425 IU/mL had a higher frequency of extraocular muscle enlargement (25% versus 6%, p = 0.004). Patients with serum IgG4 above 712 mg/dL had a higher frequency of decreased BCVA (37% versus 5%, p ≤ 0.001). Conclusion: Unsupervised hierarchical clustering analysis using routine blood test data differentiates four distinct clinical phenotypes of IgG4-ROD, which suggest differences in pathophysiologic mechanisms. High serum IgG4 is a potential predictor of worsened BCVA, and high serum IgE is a potential predictor of extraocular muscle enlargement in IgG4-ROD patients.

Identification of Prognostic Markers in Patients with Primary Vitreoretinal Lymphoma by Clustering Analysis Using Clinical Data

(1) Purpose: Primary vitreoretinal lymphoma (PVRL) is associated with poor prognosis because most of the patients with PVRL develop central nerve system lymphoma. The prognostic biomarker of PVRL is largely unknown. Cluster analysis has been used to identify phenotypic groups within various diseases. In this study, we aimed to describe clinical features of patients with PVRL grouped by clustering analysis and to identify biomarkers for predicting survival prognosis in patients with PVRL. (2) Materials and Methods: Forty patients with PVRL were divided into two groups by clustering analysis using clinical data. Clinical features of the two groups were compared. (3) Result: Clustering analysis classified patients into groups A and B. The survival rate during the follow-up period was significantly lower in group B than in group A (p = 0.03). Serum IgG, serum IgA, vitreous IL-10 and vitreous IL-10 to IL-6 ratio were significantly different between groups A and B (p = 0.03, 0.005, 0.008 and 0.03, respectively). Receiver operating characteristic (ROC) curves generated for the four variables indicated that serum IgA was most suitable for the prediction of prognosis. Patients with serum IgA below 184 mg/dL obtained from the ROC curve had a lower three-year survival rate (p = 0.03) and more episodes of recurrence of lymphoma (3.2 times versus 1.8 times, p = 0.02) compared with patients with serum IgA above 184 mg/dL. (4) Conclusion: The survival rate was significantly different in PVRL patients classified into two groups by clustering analysis. Patients with lower serum IgA had more recurrences and poorer survival than patients with higher IgA.