We propose a new model selection method, named the posterior averaging information criterion, for Bayesian model assessment to minimize the risk of predicting independent future observations. The ...theoretical foundation is built on the Kullback-Leibler divergence to quantify the similarity between the proposed candidate model and the underlying true model. From a Bayesian perspective, our method evaluates the candidate models over the entire posterior distribution in terms of predicting a future independent observation. Without assuming that the true distribution is contained in the candidate models, the new criterion is developed by correcting the asymptotic bias of the posterior mean of the in-sample log-likelihood against out-of-sample log-likelihood, and can be generally applied even for Bayesian models with degenerate non-informative priors. Simulations in both normal and binomial settings demonstrate superior small sample performance.
Abstract
Aims
This observational study characterized cardiovascular disease (CVD) mortality risk for multiple cancer sites, with respect to the following: (i) continuous calendar year, (ii) age at ...diagnosis, and (iii) follow-up time after diagnosis.
Methods and results
The Surveillance, Epidemiology, and End Results program was used to compare the US general population to 3 234 256 US cancer survivors (1973–2012). Standardized mortality ratios (SMRs) were calculated using coded cause of death from CVDs (heart disease, hypertension, cerebrovascular disease, atherosclerosis, and aortic aneurysm/dissection). Analyses were adjusted by age, race, and sex. Among 28 cancer types, 1 228 328 patients (38.0%) died from cancer and 365 689 patients (11.3%) died from CVDs. Among CVDs, 76.3% of deaths were due to heart disease. In eight cancer sites, CVD mortality risk surpassed index-cancer mortality risk in at least one calendar year. Cardiovascular disease mortality risk was highest in survivors diagnosed at <35 years of age. Further, CVD mortality risk is highest (SMR 3.93, 95% confidence interval 3.89–3.97) within the first year after cancer diagnosis, and CVD mortality risk remains elevated throughout follow-up compared to the general population.
Conclusion
The majority of deaths from CVD occur in patients diagnosed with breast, prostate, or bladder cancer. We observed that from the point of cancer diagnosis forward into survivorship cancer patients (all sites) are at elevated risk of dying from CVDs compared to the general US population. In endometrial cancer, the first year after diagnosis poses a very high risk of dying from CVDs, supporting early involvement of cardiologists in such patients.
Just Another Gibbs Sampling (JAGS) is a convenient tool to draw posterior samples using Markov Chain Monte Carlo for Bayesian modeling. However, the built-in function dinterval() for censored data ...misspecifies the default computation of deviance function, which limits likelihood-based Bayesian model comparison.
To establish an automatic approach to specifying the correct deviance function in JAGS, we propose a simple and generic alternative modeling strategy for the analysis of censored outcomes. The two illustrative examples demonstrate that the alternative strategy not only properly draws posterior samples in JAGS, but also automatically delivers the correct deviance for model assessment. In the survival data application, our proposed method provides the correct value of mean deviance based on the exact likelihood function. In the drug safety data application, the deviance information criterion and penalized expected deviance for seven Bayesian models of censored data are simultaneously computed by our proposed approach and compared to examine the model performance.
We propose an effective strategy to model censored data in the Bayesian modeling framework in JAGS with the correct deviance specification, which can simplify the calculation of popular Kullback-Leibler based measures for model selection. The proposed approach applies to a broad spectrum of censored data types, such as survival data, and facilitates different censored Bayesian model structures.
To determine whether pretreatment CT texture features can improve patient risk stratification beyond conventional prognostic factors (CPFs) in stage III non-small cell lung cancer (NSCLC).
We ...retrospectively reviewed 91 cases with stage III NSCLC treated with definitive chemoradiation therapy. All patients underwent pretreatment diagnostic contrast enhanced computed tomography (CE-CT) followed by 4-dimensional CT (4D-CT) for treatment simulation. We used the average-CT and expiratory (T50-CT) images from the 4D-CT along with the CE-CT for texture extraction. Histogram, gradient, co-occurrence, gray tone difference, and filtration-based techniques were used for texture feature extraction. Penalized Cox regression implementing cross-validation was used for covariate selection and modeling. Models incorporating texture features from the 33 image types and CPFs were compared to those with models incorporating CPFs alone for overall survival (OS), local-regional control (LRC), and freedom from distant metastases (FFDM). Predictive Kaplan-Meier curves were generated using leave-one-out cross-validation. Patients were stratified based on whether their predicted outcome was above or below the median. Reproducibility of texture features was evaluated using test-retest scans from independent patients and quantified using concordance correlation coefficients (CCC). We compared models incorporating the reproducibility seen on test-retest scans to our original models and determined the classification reproducibility.
Models incorporating both texture features and CPFs demonstrated a significant improvement in risk stratification compared to models using CPFs alone for OS (P=.046), LRC (P=.01), and FFDM (P=.005). The average CCCs were 0.89, 0.91, and 0.67 for texture features extracted from the average-CT, T50-CT, and CE-CT, respectively. Incorporating reproducibility within our models yielded 80.4% (±3.7% SD), 78.3% (±4.0% SD), and 78.8% (±3.9% SD) classification reproducibility in terms of OS, LRC, and FFDM, respectively.
Pretreatment tumor texture may provide prognostic information beyond that obtained from CPFs. Models incorporating feature reproducibility achieved classification rates of ∼80%. External validation would be required to establish texture as a prognostic factor.
Summary Background Neuroendocrine tumours (NETs) can secrete bioactive amines into the bloodstream, causing carcinoid syndrome, with symptoms including flushing and diarrhoea. However, carcinoid ...syndrome frequency in the NET population has never been rigorously assessed, nor has its relationship to presenting clinicopathological characteristics. This analysis assessed the proportion of patients with NETs and carcinoid syndrome in the USA and associated clinical factors. Methods We identified patients (≥65 years of age) from the Surveillance, Epidemiology, and End Results-Medicare database, excluding those with pancreatic tumours or small-cell or large-cell lung cancer, as well as those without complete data. We assessed the incidence of patients with at least two insurance claims of flushing, diarrhoea, or carcinoid syndrome during the 3 months before and after NET diagnosis. We compared demographic and clinical characteristics between patients with and without carcinoid syndrome using χ2 tests. We used the Cochran-Armitage trend test to identify trends in carcinoid syndrome incidence and Cox regression to assess the relationship between carcinoid syndrome and survival. Findings Between April 1, 2000, and Dec 31, 2011, 9512 eligible patients were diagnosed with NETs, of whom 1786 (19%) had carcinoid syndrome. The number of patients with NETs and carcinoid syndrome increased from 50 (11%) of 465 patients in 2000 to 160 (19%) of 854 in 2011 (p<0·0001). The proportion of patients with carcinoid syndrome compared with those without did not differ significantly with respect to age at diagnosis (p=0·65), geographical region (p=0·054), or urban versus rural status (p=0·53). Patients with carcinoid syndrome were more frequently female than male (p=0·0003). Race was associated with a significant difference in the reported incidence of carcinoid syndrome (p<0·0001), as was tumour grade, stage, and primary tumour site (all p<0·0001). Patients with carcinoid syndrome had a shorter overall survival (median 5 years 95% CI 4·5–5·4) than did those without carcinoid syndrome (5·6 years 5·4–5·9; hazard ratio 1·102 1·016–1·194; p=0·019). Use of octreotide (p<0·0001) and chemotherapy (p=0·003) were more common in patients with carcinoid syndrome than in those without it, whereas surgery was used more frequently in patients without carcinoid syndrome (p=0·009); use of radiotherapy was not significantly associated with the presence of carcinoid syndrome at diagnosis (p=0·07). Interpretation This population-based analysis reveals that carcinoid syndrome is significantly associated with tumour grade, stage, and primary tumour site, and leads to shorter survival compared with those patients without carcinoid syndrome. An improved understanding of the heterogeneity of presenting symptoms among patients with NETs might permit more tailored assessment and management than at present and enable future research into the effect of carcinoid syndrome control on patient survival. Funding Ipsen.
Glioblastoma is highly enriched with macrophages, and osteopontin (OPN) expression levels correlate with glioma grade and the degree of macrophage infiltration; thus, we studied whether OPN plays a ...crucial role in immune modulation. Quantitative PCR, immunoblotting, and ELISA were used to determine OPN expression. Knockdown of OPN was achieved using complementary siRNA, shRNA, and CRISPR/Cas9 techniques, followed by a series of in vitro functional migration and immunological assays. OPN gene-deficient mice were used to examine the roles of non-tumor-derived OPN on survival of mice harboring intracranial gliomas. Patients with mesenchymal glioblastoma multiforme (GBM) show high OPN expression, a negative survival prognosticator. OPN is a potent chemokine for macrophages, and its blockade significantly impaired the ability of glioma cells to recruit macrophages. Integrin αvβ5 (ITGαvβ5) is highly expressed on glioblastoma-infiltrating macrophages and constitutes a major OPN receptor. OPN maintains the M2 macrophage gene signature and phenotype. Both tumor-derived and host-derived OPN were critical for glioma development. OPN deficiency in either innate immune or glioma cells resulted in a marked reduction in M2 macrophages and elevated T cell effector activity infiltrating the glioma. Furthermore, OPN deficiency in the glioma cells sensitized them to direct CD8+ T cell cytotoxicity. Systemic administration in mice of 4-1BB-OPN bispecific aptamers was efficacious, increasing median survival time by 68% (P < 0.05). OPN is thus an important chemokine for recruiting macrophages to glioblastoma, mediates crosstalk between tumor cells and the innate immune system, and has the potential to be exploited as a therapeutic target.
Despite a multiplicity of clinical trials testing immune checkpoint inhibitors, the frequency of expression of potential predictive biomarkers is unknown in glioma.
In this study, we profiled the ...frequency of shared biomarker phenotypes. To clarify the relationships among tumor mutational load (TML), mismatch repair (MMR), and immune checkpoint expression, we profiled patients with glioma (n = 327), including glioblastoma (GBM) (n = 198), whose samples had been submitted for analysis from 2009 to 2016. The calculation algorithm for TML included nonsynonymous mutation counts per tumor, with germline mutations filtered out. Immunohistochemical analysis and next-generation sequencing were used to determine tumor-infiltrating lymphocyte expression positive for programmed cell death protein 1 (PD-1), PD ligand 1 (PD-L1) expression on tumor cells, MMR (MLH1, MSH2, MSH6, and PMS2) protein expression and mutations, and DNA polymerase epsilon (POLE) mutations.
High TML was only found in 3.5% of GBM patients (7 of 198) and was associated with the absence of protein expression of mutL homolog 1 (MLH1) (P = .0345), mutS homolog 2 (MSH2) (P = .0099), MSH6 (P = .0022), and postmeiotic segregation increased 2 (PMS2) (P = .0345) and the presence of DNA MMR mutations. High and moderate TML GBMs did not have an enriched influx of CD8+ T cells, PD-1+ T cells, or tumor-expressed PD-L1. IDH1 mutant gliomas were not enriched for high TML, PD-1+ T cells, or PD-L1 expression.
To clarify the relationships among TML, MMR, and immune checkpoint expression, we profiled the frequency of shared biomarker phenotypes. On the basis of a variety of potential biomarkers of response to immune checkpoints, only small subsets of glioma patients are likely to benefit from monotherapy immune checkpoint inhibition.
Therapeutic targeting of the immune checkpoints cytotoxic T-lymphocyte-associated molecule-4 (CTLA-4) and PD-1/PD-L1 has demonstrated tumor regression in clinical trials, and phase 2 trials are ...ongoing in glioblastoma (GBM). Previous reports have suggested that responses are more frequent in patients with tumors that express PD-L1; however, this has been disputed. At issue is the validation of PD-L1 biomarker assays and prognostic impact.
Using immunohistochemical analysis, we measured the incidence of PD-L1 expression in 94 patients with GBM. We categorized our results according to the total number of PD-L1-expressing cells within the GBMs and then validated this finding in ex vivo GBM flow cytometry with further analysis of the T cell populations. We then evaluated the association between PD-L1 expression and median survival time using the protein expression datasets and mRNA from The Cancer Genome Atlas.
The median percentage of PD-L1-expressing cells in GBM by cell surface staining is 2.77% (range: 0%-86.6%; n = 92), which is similar to the percentage found by ex vivo flow cytometry. The majority of GBM patients (61%) had tumors with at least 1% or more PD-L1-positive cells, and 38% had at least 5% or greater PD-L1 expression. PD-L1 is commonly expressed on the GBM-infiltrating T cells. Expression of both PD-L1 and PD-1 are negative prognosticators for GBM outcome.
The incidence of PD-L1 expression in GBM patients is frequent but is confined to a minority subpopulation, similar to other malignancies that have been profiled for PD-L1 expression. Higher expression of PD-L1 is correlated with worse outcome.
Hepatocellular carcinoma (HCC) is the second‐leading cause of cancer‐related deaths worldwide with a poor survival rate. As many as 40% of HCCs are clonal, with alteration of key tumor‐suppressor ...pathways in stem cells as the primary cause of HCC initiation. However, mechanisms that generate metastatic stem cells in preneoplastic liver tissue are not well understood. We hypothesized that chronic inflammation is a major driver of the transformation of genetically defective liver stem cells (LSCs) into highly metastatic liver cancer cells in premalignant liver tissue. We developed models of chronic inflammation in wild‐type (WT) and β2‐spectrin (β2SP)+/− (SPTBN1) mice. CD133+ LSCs derived from preneoplastic livers of β2SP+/− mice treated with interleukin‐6 (pIL6; IL6β2SP+/− LSCs) were highly tumorigenic and metastatic, whereas those derived from WT mice treated with pIL6 (IL6WT LSCs) had significantly less proliferation and no tumorigenic properties. IL6β2SP+/− LSCs not only exhibited nuclear localization of Twist and Slug, markers of epithelial‐mesenchymal transition (EMT), but also constitutive activation of nuclear factor kappa B (NFκB; RELA). Knockdown of NFκB decreased the EMT phenotypes and metastatic capacity of these cells. NFκB in IL6β2SP+/− LSCs was activated by transforming growth factor β (TGFβ)‐activated kinase 1 (TAK1; MAP3K7), which is associated with poor survival in HCC and interleukin‐6 (IL6) expression. The amount of constitutively activated NFκB increased dramatically from normal to cirrhotic to HCC tissues from human patients. Conclusion: IL6‐mediated inflammation programs constitutive activation of the TAK1‐NFκB signaling cascade in CD133+ LSCs, and this program interacts with deficient TGFβ signaling, thereby accelerating the transformation of normal LSCs to metastatic cancer stem cells (mCSCs). Indeed, this study delineates the development of EMT‐positive mCSCs in HCC‐free liver tissue upon chronic inflammation. (Hepatology 2017;65:1222‐1236).