Search

AimPatients with synchronous colon cancer metastases have highly variable overall survival (OS), making accurate predictive models challenging to build. We aim to use machine learning to more accurately predict OS in these patients and to present this predictive model in the form of nomograms for patients and clinicians.MethodsUsing the National Cancer Database (2010-2014), we identified right colon (RC) and left colon (LC) cancer patients with synchronous metastases. Each primary site was split into training and testing datasets. Nomograms predicting 3- year OS were created for each site using Cox proportional hazard regression with lasso regression. Each model was evaluated by both calibration (comparison of predicted vs observed OS) and validation (degree of concordance as measured by the c-index) methodologies.ResultsA total of 11018 RC and 8346 LC patients were used to construct and validate the nomograms. After stratifying each model into five risk groups, the predicted OS was within the 95% CI of the observed OS in four out of five risk groups for both the RC and LC models. Externally validated c-indexes at 3years for the RC and LC models were 0.794 and 0.761, respectively.ConclusionsUtilization of machine learning can result in more accurate predictive models for patients with metastatic colon cancer. Nomograms built from these models can assist clinicians and patients in the shared decision-making process of their cancer care.

BACKGROUND:Because imaging has a high sensitivity to diagnose hepatocellular carcinoma (HCC) and tissue biopsies carry risks such as bleeding, the latter are often not performed in HCC. Blood-derived circulating tumor DNA (ctDNA) analysis can identify somatic alterations, but its utility has not been characterized in HCC. MATERIALS AND METHODS:We evaluated 14 patients with advanced HCC (digital ctDNA sequencing [68 genes]). Mutant relative to wild-type allele fraction was calculated. RESULTS:All patients (100%) had somatic alterations (median = 3 alterations/patient [range, 1-8]); median mutant allele fraction, 0.29% (range, 0.1%-37.77%). Mutations were identified in several genes: TP53 (57% of patients), CTNNB1 (29%), PTEN (7%), CDKN2A (7%), ARID1A (7%), and MET (7%); amplifications, in CDK6 (14%), EGFR (14%), MYC (14%), BRAF (7%), RAF1 (7%), FGFR1 (7%), CCNE1 (7%), PIK3CA (7%), and ERBB2/HER2 (7%). Eleven patients (79%) had ≥1 theoretically actionable alteration. No two patients had identical genomic portfolios, suggesting the need for customized treatment. A patient with a CDKN2A-inactivating and a CTNNB1-activating mutation received matched treatment: palbociclib (CDK4/6 inhibitor) and celecoxib (COX-2/Wnt inhibitor); des-gamma-carboxy prothrombin level decreased by 84% at 2 months (1,410 to 242 ng/mL [normal: ≤7.4 ng/mL]; alpha fetoprotein [AFP] low at baseline). A patient with a PTEN-inactivating and a MET-activating mutation (an effect suggested by in silico molecular dynamic simulations) received sirolimus (mechanistic target of rapamycin inhibitor) and cabozantinib (MET inhibitor); AFP declined by 63% (8,320 to 3,045 ng/mL [normal: 0-15 ng/mL]). CONCLUSION:ctDNA derived from noninvasive blood tests can provide exploitable genomic profiles in patients with HCC. IMPLICATIONS FOR PRACTICE:This study reports that blood-derived circulating tumor DNA can provide therapeutically exploitable genomic profiles in hepatocellular cancer, a malignancy that is known to be difficult to biopsy.

Copyright © 2015 Radiological Society of North America. Purpose: To determine per-lesion sensitivity and positive predictive value (PPV) of gadoxetic acid-enhanced 3-T magnetic resonance (MR) imaging for the diagnosis of malignant lesions by using matched (spatially correlated) hepatectomy pathologic findings as the reference standard. Materials and Methods: In this prospective, institutional review board-approved, HIPAAcompliant study, 20 patients (nine men, 11 women; mean age, 59 years) with malignant liver lesions who gave written informed consent underwent preoperative gadoxetic acid-enhanced 3-T MR imaging for surgical planning. Two image sets were independently analyzed by three readers to detect liver lesions (set 1 without and set 2 with hepatobiliary phase [HBP] images). Hepatectomy specimen ex vivo MR imaging assisted in matching gadoxetic acid-enhanced 3-T MR imaging findings with pathologic findings. Interreader agreement was assessed by using the Cohen κ coefficient. Per-lesion sensitivity and PPV were calculated. Results: Cohen κ values were 0.64-0.76 and 0.57-0.84, and overall perlesion sensitivity was 45% (42 of 94 lesions) to 56% (53 of 94 lesions) and 58% (55 of 94 lesions) to 64% (60 of 94 lesions) for sets 1 and 2, respectively. The addition of HBP imaging did not affect interreader agreement but significantly improved overall sensitivity for one reader (P < .05) and almost for another (P = .05). Sensitivity for 0.2-0.5-cm lesions was 0% (0 of 26 lesions) to 8% (two of 26 lesions) for set 1 and 4% (one of 26 lesions) to 12% (three of 26 lesions) for set 2. Sensitivity for 0.6-1.0-cm lesions was 28% (nine of 32 lesions) to 59% (19 of 32 lesions) for set 1 and 66% (21 of 32 lesions) to 69% (22 of 32 lesions) for set 2. Sensitivity for lesions at least 1.0 cm in diameter was at least 81% (13 of 16 lesions) for set 1 and was not improved for set 2. PPV was 98% (56 of 57 lesions) to 100% (60 of 60 lesions) for all readers without differences between image sets or lesion size. Conclusion: Gadoxetic acid-enhanced 3-T MR imaging provides high per-lesion sensitivity and PPV for preoperative malignant liver lesion detection overall, although sensitivity for 0.2-0.5-cm malignant lesions is poor.