A Statistical Testing Strategy Accounting for Random and Nonrandom (Skewed) X-Chromosome Inactivation Identifies Lung Cancer Susceptibility Loci among Smokers.

INTRODUCTION: Lung cancer is the most common cancer worldwide in mortality and the second in incidence. Epidemiological studies found a higher lung cancer risk for smoking women in comparison to men, but these sex differences, irrespective of smoking habits, remain controversial. One of the hypotheses concerns the genetic contribution of the sex chromosomes. However, while genome-wide association studies identified many lung cancer susceptibility loci, these analyses have excluded X-linked loci. METHODS: To account for nongenetic factors, we first presented an association test based on an additive-multiplicative hazard model accounting for random/nonrandom X-inactivation process. A simulation study was performed to investigate the properties of the proposed test as compared with the Wald test from a Cox model with random X-inactivation process and the partial likelihood ratio test proposed by Xu et al. accounting for nonrandom X-inactivation process. Then, we performed an X chromosome-wide association study on 9,261 individuals from the population-based cohort CARTaGENE to identify susceptibility loci for lung cancer among current and past smokers. We adjusted for the PLCOm2012 lung cancer risk score used in screening programs. RESULTS: Simulation results show the good behavior of the proposed test in terms of power and type I error probability as compared to the Xu et al. and the Wald test. Using the proposed test statistic and adjusting for the PLCOm2012 score, the X chromosome-wide statistical analysis identified two SNPs in low-linkage disequilibrium located in the IL1RAPL1 (IL-1 R accessory protein-like) gene: rs12558491 (p = 2.75×10-9) and rs12835699 (p = 1.26×10-6). For both SNPs, the minor allele was associated with lower lung cancer risk. Adjusting for multiple testing, no signal was detected using the Wald or the Xu et al. likelihood ratio tests. CONCLUSION: By taking into account smoking behavior and the X-inactivation process, the investigation of the X chromosome has shed a new light on the association between X-linked loci and lung cancer. We identified two loci associated with lung cancer located in the IL1RAPL1 gene. This finding would have been overlooked by examining only results from other test statistics.

Provincial variation in colorectal cancer screening adherence in Canada; evidence from the Canadian Partnership for Tomorrow's Health.

Introduction: Although colorectal cancer (CRC) screening program is proven to reduce CRC incidence and mortality, understanding patterns and predictors of suboptimal adherence in screening program requires further investigation in Canada. Methods: We used self-reported data from five regional cohorts of the Canadian Partnership for Tomorrow's Health (CanPath), namely the BC Generations Project (BCGP), Alberta's Tomorrow Project (ATP), the Ontario Health Study (OHS), Quebec's CARTaGENE, and the Atlantic Partnership for Tomorrow's Health Study (Atlantic PATH). We stratified participants into the following four risk categories: 1) age 50-74 years, 2) family history in a first-degree relative, 3) personal history of chronic inflammatory bowel disease and/or polyps, and 4) co-existence of personal risk and family history. Multivariable logistic regression was used to identify predictors of adherence to the screening guidelines. Results: Adherence to CRC screening varied considerably between regions, ranging from 16.6% in CARTaGENE to 47.7% in OHS. Compared to the largest cohort OHS, the likelihood of non-adherence to CRC screening was significantly higher in BCGP (OR 1.15, 95% CI 1.11-1.19), the Atlantic PATH (OR 1.90, 95% CI 1.82-1.99) and CARTaGENE (OR 5.10, 95% CI 4.85-5.36). Low physical activity, current smoking, presence of personal risk, family history of CRC significantly reduced the likelihood of adherence to screening recommendations. Discussion/conclusion: Compared to the national target of ≥ 60% for participation in CRC screening, adherence to regular CRC screening was suboptimal in this cohort of Canadians and varied by region. Further efforts are needed to identify the specific barriers to screening adherence in different provinces and across risk categories.

Response to Biologic Drugs in Patients With Rheumatoid Arthritis and Antidrug Antibodies.

Importance: There are conflicting data on the association of antidrug antibodies with response to biologic disease-modifying antirheumatic drugs (bDMARDs) in rheumatoid arthritis (RA). Objective: To analyze the association of antidrug antibodies with response to treatment for RA. Design, Setting, and Participants: This cohort study analyzed data from the ABI-RA (Anti-Biopharmaceutical Immunization: Prediction and Analysis of Clinical Relevance to Minimize the Risk of Immunization in Rheumatoid Arthritis Patients) multicentric, open, prospective study of patients with RA from 27 recruiting centers in 4 European countries (France, Italy, the Netherlands, and the UK). Eligible patients were 18 years or older, had RA diagnosis, and were initiating a new bDMARD. Recruitment spanned from March 3, 2014, to June 21, 2016. The study was completed in June 2018, and data were analyzed in June 2022. Exposures: Patients were treated with a new bDMARD: adalimumab, infliximab (grouped as anti-tumor necrosis factor [TNF] monoclonal antibodies [mAbs]), etanercept, tocilizumab, and rituximab according to the choice of the treating physician. Main Outcomes and Measures: The primary outcome was the association of antidrug antibody positivity with EULAR (European Alliance of Associations for Rheumatology; formerly, European League Against Rheumatism) response to treatment at month 12 assessed through univariate logistic regression. The secondary end points were the EULAR response at month 6 and at visits from month 6 to months 15 to 18 using generalized estimating equation models. Detection of antidrug antibody serum levels was performed at months 1, 3, 6, 12, and 15 to 18 using electrochemiluminescence (Meso Scale Discovery) and drug concentration for anti-TNF mAbs, and etanercept in the serum was measured using enzyme-linked immunosorbent assay. Results: Of the 254 patients recruited, 230 (mean [SD] age, 54.3 [13.7] years; 177 females [77.0%]) were analyzed. At month 12, antidrug antibody positivity was 38.2% in patients who were treated with anti-TNF mAbs, 6.1% with etanercept, 50.0% with rituximab, and 20.0% with tocilizumab. There was an inverse association between antidrug antibody positivity (odds ratio [OR], 0.19; 95% CI, 0.09-0.38; P < .001) directed against all biologic drugs and EULAR response at month 12. Analyzing all the visits starting at month 6 using generalized estimating equation models confirmed the inverse association between antidrug antibody positivity and EULAR response (OR, 0.35; 95% CI, 0.18-0.65; P < .001). A similar association was found for tocilizumab alone (OR, 0.18; 95% CI, 0.04-0.83; P = .03). In the multivariable analysis, antidrug antibodies, body mass index, and rheumatoid factor were independently inversely associated with response to treatment. There was a significantly higher drug concentration of anti-TNF mAbs in patients with antidrug antibody-negative vs antidrug antibody-positive status (mean difference, -9.6 [95% CI, -12.4 to -6.9] mg/L; P < 001). Drug concentrations of etanercept (mean difference, 0.70 [95% CI, 0.2-1.2] mg/L; P = .005) and adalimumab (mean difference, 1.8 [95% CI, 0.4-3.2] mg/L; P = .01) were lower in nonresponders vs responders. Methotrexate comedication at baseline was inversely associated with antidrug antibodies (OR, 0.50; 95% CI, 0.25-1.00; P = .05). Conclusions and Relevance: Results of this prospective cohort study suggest an association between antidrug antibodies and nonresponse to bDMARDs in patients with RA. Monitoring antidrug antibodies could be considered in the treatment of these patients, particularly nonresponders to biologic RA drugs.

Evaluation of the accuracy of the PLCOm2012 6-year lung cancer risk prediction model among smokers in the CARTaGENE population-based cohort.

BACKGROUND: The PLCOm2012 prediction tool for risk of lung cancer has been proposed for a pilot program for lung cancer screening in Quebec, but has not been validated in this population. We sought to validate PLCOm2012 in a cohort of Quebec residents, and to determine the hypothetical performance of different screening strategies. METHODS: We included smokers without a history of lung cancer from the population-based CARTaGENE cohort. To assess PLCOm2012 calibration and discrimination, we determined the ratio of expected to observed number of cases, as well as the sensitivity, specificity and positive predictive values of different risk thresholds. To assess the performance of screening strategies if applied between Jan. 1, 1998, and Dec. 31, 2015, we tested different thresholds of the PLCOm2012 detection of lung cancer over 6 years (1.51%, 1.70% and 2.00%), the criteria of Quebec's pilot program (for people aged 55-74 yr and 50-74 yr) and recommendations from 2021 United States and 2016 Canada guidelines. We assessed shift and serial scenarios of screening, whereby eligibility was assessed annually or every 6 years, respectively. RESULTS: Among 11 652 participants, 176 (1.51%) lung cancers were diagnosed in 6 years. The PLCOm2012 tool underestimated the number of cases (expected-to-observed ratio 0.68, 95% confidence interval [CI] 0.59-0.79), but the discrimination was good (C-statistic 0.727, 95% CI 0.679-0.770). From a threshold of 1.51% to 2.00%, sensitivities ranged from 52.3% (95% CI 44.6%-59.8%) to 44.9% (95% CI 37.4%-52.6%), specificities ranged from 81.6% (95% CI 80.8%-82.3%) to 87.7% (95% CI 87.0%-88.3%) and positive predictive values ranged from 4.2% (95% CI 3.4%-5.1%) to 5.3% (95% CI 4.2%-6.5%). Overall, 8938 participants had sufficient data to test performance of screening strategies. If eligibility was estimated annually, Quebec pilot criteria would have detected fewer cancers than PLCOm2012 at a 2.00% threshold (48.3% v. 50.2%) for a similar number of scans per detected cancer. If eligibility was estimated every 6 years, up to 26 fewer lung cancers would have been detected; however, this scenario led to higher positive predictive values (highest for PLCOm2012 with a 2.00% threshold at 6.0%, 95% CI 4.8%-7.3%). INTERPRETATION: In a cohort of Quebec smokers, the PLCOm2012 risk prediction tool had good discrimination in detecting lung cancer, but it may be helpful to adjust the intercept to improve calibration. The implementation of risk prediction models in some of the provinces of Canada should be done with caution.

Machine learning predicts response to TNF inhibitors in rheumatoid arthritis: results on the ESPOIR and ABIRISK cohorts.

OBJECTIVES: Around 30% of patients with rheumatoid arthritis (RA) do not respond to tumour necrosis factor inhibitors (TNFi). We aimed to predict patient response to TNFi using machine learning on simple clinical and biological data. METHODS: We used data from the RA ESPOIR cohort to train our models. The endpoints were the EULAR response and the change in Disease Activity Score (DAS28). We compared the performances of multiple models (linear regression, random forest, XGBoost and CatBoost) on the training set and cross-validated them using the area under the receiver operating characteristic curve (AUROC) or the mean squared error. The best model was then evaluated on a replication cohort (ABIRISK). RESULTS: We included 161 patients from ESPOIR and 118 patients from ABIRISK. The key selected features were DAS28, lymphocytes, ALT (aspartate aminotransferase), neutrophils, age, weight, and smoking status. When predicting EULAR response, CatBoost achieved the best performances of the four tested models. It reached an AUROC of 0.72 (0.68-0.73) on the train set (ESPOIR). Better results were obtained on the train set when etanercept and monoclonal antibodies were analysed separately. On the test set (ABIRISK), these models respectively achieved on AUROC of 0.70 (0.57-0.82) and 0.71 (0.55-0.86). Two decision thresholds were tested. The first prioritised a high confidence in identifying responders and yielded a confidence up to 90% for predicting response. The second prioritised a high confidence in identifying inadequate responders and yielded a confidence up to 70% for predicting non-response. The change in DAS28 was predicted with an average error of 1.1 DAS28 points. CONCLUSION: The machine learning models developed allowed predicting patient response to TNFi exclusively using data available in clinical routine.

Analyzing cohort studies with interval-censored data: A new model-based linear rank-type test.

To compare two or more survival distributions with interval-censored data, various nonparametric tests have been proposed. Some are based on the G ρ $$ {G}^{\rho } $$ -family introduced by Harrington and Fleming (1991) that allows flexibility for situations in which the hazard ratio decreases monotonically to unity. However, it is unclear how to choose the appropriate value of the parameter ρ $$ \rho $$ . In this work, we propose a novel linear rank-type test for analyzing interval-censored data that derived from a proportional reversed hazard model. We show its relationship with decreasing hazard ratio. This test statistic provides an alternative to the G ρ $$ {G}^{\rho } $$ -based test statistics by bypassing the choice of the ρ $$ \rho $$ parameter. Simulation results show its good behavior. Two studies on breast cancer and drug users illustrate its practical uses and highlight findings that would have been overlooked if other tests had been used. The test is easy to implement with standard software and can be used for a wide range of situations with interval-censored data to test the equality of survival distributions between two or more independent groups.

Five-year absolute risk estimates of colorectal cancer based on CCRAT model and polygenic risk scores: A validation study using the Quebec population-based cohort CARTaGENE.

The objective was to evaluate the predictive performance of the Colorectal Cancer Risk Assessment Tool (CCRAT) and three polygenic risk scores (Hsu et al., 2015; Law et al., 2019, Archambault et al., 2020) to predict the occurrence of colorectal cancer at five years in a Quebec population-based cohort. By using the CARTaGENE cohort, we computed the absolute risk of colorectal cancer with the CCRAT model, the polygenic risk scores (PRS) and combined clinico-genetic models (CCRAT + PRS). We also tailored the CCRAT model by using the marginal age-specific colorectal incidence rates in Canada and the risk score distribution. We reported the calibration and the discrimination. Performances of the PRSs, combined and tailored CCRAT models were compared to the original CCRAT model. The expected-to-observed ratio of the original CCRAT model was 0.54 [0.43-0.68]. The c-index was 74.79 [68.3-80.5]. The tailored CCRAT model improved the expected-to-observed ratio (0.74 [0.59-0.94]) and c-index (76.39 [69.7-82.1]). All PRS improved the expected-to-observed ratios (around 0.83, confidence intervals including one). PRSs' c-indexes were not significantly different from CCRAT models. Results from the combined models were close to those from the PRS models, Archambault combined model's c-index being significantly higher than the original and tailored CCRAT models (78.67 [70.8-86.5]; p < 0.001 and p = 0.028, respectively). In this Quebec cohort, CCRAT model has a good discrimination with a poor calibration. While the tailored CCRAT provides some gain in calibration, clinico-genetic models improved both calibration and discrimination. However, better calibrations must be obtained before a practical use among the inhabitants of Quebec province.

Integrating NGS-derived mutational profiling in the diagnosis of multiple lung adenocarcinomas.

MICROABSTRACT: Integration of Next Generation Sequencing (NGS) information for use in distinguishing between Multiple Primary Lung Cancer and intrapulmonary metastasis was evaluated. We used a probabilistic model, comprehensive histologic assessment and NGS to classify patients. Integrating NGS data confirmed initial diagnosis (n = 41), revised the diagnosis (n = 12), while resulted in non-informative data (n = 8). Accuracy of diagnosis can be significantly improved with integration of NGS data. BACKGROUND: Distinguishing between multiple primary lung cancers (MPLC) and intrapulmonary metastases (IPM) is challenging. The goal of this study was to evaluate how Next Generation Sequencing (NGS) information may be integrated in the diagnostic strategy. PATIENTS AND METHODS: Patients with multiple lung adenocarcinomas were classified using both the comprehensive histologic assessment and NGS. We computed the joint probability of each pair having independent mutations by chance (thus being classified as MPLC). These probabilities were computed using the marginal mutation rates of each mutation, and the known negative dependencies between driver genes and different gene loci. With these NGS-driven data, cases were re-classified as MPLC or IPM. RESULTS: We analyzed 61 patients with a total of 131 tumors. The most frequent mutation was KRAS (57.3%) which occured at a rate higher than expected (p < 0.001) in lung cancer. No mutation was detected in 25/131 tumors (19.1%). Discordant molecular findings between tumor sites were found in 46 patients (75.4%); 11 patients (18.0%) had concordant molecular findings, and 4 patients (6.6%) had concordant molecular findings at 2 of the 3 sites. After integration of the NGS data, the initial diagnosis was confirmed for 41 patients (67.2%), the diagnosis was revised for 12 patients (19.7%) or was considered as non-informative for 8 patients (13.1%). CONCLUSION: Integrating the information of NGS data may significantly improve accuracy of diagnosis and staging.

Validation of breast cancer risk assessment tools on a French-Canadian population-based cohort.

OBJECTIVES: Evaluate the accuracy of the Breast Cancer Risk Assessment Tool (BCRAT), International Breast Cancer Intervention Study risk evaluation tool (IBIS), Polygenic Risk Scores (PRS) and combined scores (BCRAT+PRS and IBIS +PRS) to predict the occurrence of invasive breast cancers at 5 years in a French-Canadian population. DESIGN: Population-based cohort study. SETTING: We used the population-based cohort CARTaGENE, composed of 43 037 Quebec residents aged between 40 and 69 years and broadly representative of the population recorded on the Quebec administrative health insurance registries. PARTICIPANTS: 10 200 women recruited in 2009-2010 were included for validating BCRAT and IBIS and 4555 with genetic information for validating the PRS and combined scores. OUTCOME MEASURES: We computed the absolute risks of breast cancer at 5 years using BCRAT, IBIS, four published PRS and combined models. We reported the overall calibration performance, goodness-of-fit test and discriminatory accuracy. RESULTS: 131 (1.28%) women developed a breast cancer at 5 years for validating BCRAT and IBIS and 58 (1.27%) for validating PRS and combined scores. Median follow-up was 5 years. BCRAT and IBIS had an overall expected-to-observed ratio of 1.01 (0.85-1.19) and 1.02 (0.86-1.21) but with significant differences when partitioning by risk groups (p<0.05). IBIS' c-index was significantly higher than BCRAT (63.42 (59.35-67.49) vs 58.63 (54.05-63.21), p=0.013). PRS scores had a global calibration around 0.82, with a CI including one, and non-significant goodness-of-fit tests. PRS' c-indexes were non-significantly higher than BCRAT and IBIS, the highest being 64.43 (58.23-70.63). Combined models did not improve the results. CONCLUSIONS: In this French-Canadian population-based cohort, BCRAT and IBIS have good mean calibration that could be improved for risk subgroups, and modest discriminatory accuracy. Despite this modest discriminatory power, these tools can be of interest for primary care physicians for delivering a personalised message to their high-risk patients, regarding screening and lifestyle counselling.

Quantifying the Predictive Accuracy of a Polygenic Risk Score for Predicting Incident Cancer Cases : Application to the CARTaGENE Cohort.

With the increasing use of polygenic risk scores (PRS) there is a need for adapted methods to evaluate the predictivity of these tools. In this work, we propose a new pseudo-R 2 criterion to evaluate PRS predictive accuracy for time-to-event data. This new criterion is related to the score statistic derived under a two-component mixture model. It evaluates the effect of the PRS on both the propensity to experience the event and on the dynamic of the event among the susceptible subjects. Simulation results show that our index has good properties. We compared our index to other implemented pseudo-R 2 for survival data. Along with our index, two other indices have comparable good behavior when the PRS has a non-null propensity effect, and our index is the only one to detect when the PRS has only a dynamic effect. We evaluated the 5-year predictivity of an 18-single-nucleotide-polymorphism PRS for incident breast cancer cases on the CARTaGENE cohort using several pseudo-R 2 indices. We report that our index, which summarizes both a propensity and a dynamic effect, had the highest predictive accuracy. In conclusion, our proposed pseudo-R 2 is easy to implement and well suited to evaluate PRS for predicting incident events in cohort studies.

The establishment of the Household Air Pollution Consortium (HAPCO).

Household air pollution (HAP) is of public health concern with ~3 billion people worldwide (including >15 million in the US) exposed. HAP from coal use is a human lung carcinogen, yet the epidemiological evidence on carcinogenicity of HAP from biomass use, primarily wood, is not conclusive. To robustly assess biomass's carcinogenic potential, prospective studies of individuals experiencing a variety of HAP exposures are needed. We have built a global consortium of 13 prospective cohorts (HAPCO: Household Air Pollution Consortium) that have site- and disease-specific mortality and solid fuel use data, for a combined sample size of 587,257 participants and 57,483 deaths. HAPCO provides a novel opportunity to assess the association of HAP with lung cancer death while controlling for important confounders such as tobacco and outdoor air pollution exposures. HAPCO is also uniquely positioned to determine the risks associated with cancers other than lung as well as non-malignant respiratory and cardiometabolic outcomes, for which prospective epidemiologic research is limited. HAPCO will facilitate research to address public health concerns associated with HAP-attributed exposures by enabling investigators to evaluate sex-specific and smoking status-specific effects under various exposure scenarios.

Incidence and risk factors for adalimumab and infliximab anti-drug antibodies in rheumatoid arthritis: A European retrospective multicohort analysis.

OBJECTIVES: To evaluate the incidence of anti-drug antibody (ADA) occurrences and ADA-related risk factors under adalimumab and infliximab treatment in rheumatoid arthritis (RA) patients. METHODS: The study combined retrospective cohorts from the ABIRISK project totaling 366 RA patients treated with adalimumab (n = 240) or infliximab (n = 126), 92.4% of them anti-TNF naive (n = 328/355) and 96.6% of them co-treated with methotrexate (n = 341/353) with up to 18 months follow-up. ADA positivity was measured by enzyme-linked immunosorbent assay. The cumulative incidence of ADA was estimated, and potential bio-clinical factors were investigated using a Cox regression model on interval-censored data. RESULTS: ADAs were detected within 18 months in 19.2% (n = 46) of the adalimumab-treated patients and 29.4% (n = 37) of the infliximab-treated patients. The cumulative incidence of ADA increased over time. In the adalimumab and infliximab groups, respectively, the incidence was 15.4% (5.2-20.2) and 0% (0-5.9) at 3 months, 17.6% (11.4-26.4) and 0% (0-25.9) at 6 months, 17.7% (12.6-37.5) and 34.1% (11.4-46.3) at 12 months, 50.0% (25.9-87.5) and 37.5% (25.9-77.4) at 15 months and 50.0% (25.9-87.5) and 66.7% (37.7-100) at 18 months. Factors associated with a higher risk of ADA development were: longer disease duration (1-3 vs. < 1 year; adalimumab: HR 3.0, 95% CI 1.0-8.7; infliximab: HR 2.7, 95% CI 1.1-6.8), moderate disease activity (DAS28 3.2-5.1 vs. < 3.2; adalimumab: HR 6.6, 95% CI 1.3-33.7) and lifetime smoking (infliximab: HR 2.7, 95% CI 1.2-6.3). CONCLUSIONS: The current study focusing on patients co-treated with methotrexate for more than 95% of them found a late occurrence of ADAs not previously observed, whereby the risk continued to increase over 18 months. Disease duration, DAS28 and lifetime smoking are clinical predictors of ADA development.

Whole-Genome and Epigenomic Landscapes of Etiologically Distinct Subtypes of Cholangiocarcinoma.

Cholangiocarcinoma (CCA) is a hepatobiliary malignancy exhibiting high incidence in countries with endemic liver-fluke infection. We analyzed 489 CCAs from 10 countries, combining whole-genome (71 cases), targeted/exome, copy-number, gene expression, and DNA methylation information. Integrative clustering defined 4 CCA clusters-fluke-positive CCAs (clusters 1/2) are enriched in ERBB2 amplifications and TP53 mutations; conversely, fluke-negative CCAs (clusters 3/4) exhibit high copy-number alterations and PD-1/PD-L2 expression, or epigenetic mutations (IDH1/2, BAP1) and FGFR/PRKA-related gene rearrangements. Whole-genome analysis highlighted FGFR2 3' untranslated region deletion as a mechanism of FGFR2 upregulation. Integration of noncoding promoter mutations with protein-DNA binding profiles demonstrates pervasive modulation of H3K27me3-associated sites in CCA. Clusters 1 and 4 exhibit distinct DNA hypermethylation patterns targeting either CpG islands or shores-mutation signature and subclonality analysis suggests that these reflect different mutational pathways. Our results exemplify how genetics, epigenetics, and environmental carcinogens can interplay across different geographies to generate distinct molecular subtypes of cancer.Significance: Integrated whole-genome and epigenomic analysis of CCA on an international scale identifies new CCA driver genes, noncoding promoter mutations, and structural variants. CCA molecular landscapes differ radically by etiology, underscoring how distinct cancer subtypes in the same organ may arise through different extrinsic and intrinsic carcinogenic processes. Cancer Discov; 7(10); 1116-35. ©2017 AACR.This article is highlighted in the In This Issue feature, p. 1047.

A two-stage design for phase II trials with time-to-event endpoint using restricted follow-up.

In phase II oncology trials, the use of new cytostatic drugs raises some questions regarding the endpoint. Time-to-event endpoints such as Progression-Free Survival have been recommended and led to new designs. In 2003, Case and Morgan proposed a design based on the comparison of the cumulative hazards at a clinically relevant timepoint. In 2013, Kwak proposed a design based on the one-sample log-rank test. If all the patients are followed from their entry time to the analysis date, the Kwak and Jung's design leads to a smaller sample size as compared to the Case-Morgan's design. However, the Case and Morgan's design requires less information since it only needs to follow every patient during a fixed interval of time. We propose a trade-off between these two approaches that corresponds to an adaptation of Kwak and Jung's design when the follow-up is expected to be restricted. Our proposal is based on the one-sample log-rank test as the Kwak and Jung's design but it uses the same follow-up information as the Case-Morgan's design. Simulation study shows that our proposal allows reducing the sample size as compared to the Case-Morgan's design (median difference of 23% [15%-33%]). Type I and type II error rates are close to their nominal rates planned in the protocol. A real phase II clinical trial in cervical cancer illustrated the interest of this new design. Thus, our proposal can be recommended as an alternative to the Kwak's design when patients' follow-up is restricted.

Bagging survival tree procedure for variable selection and prediction in the presence of nonsusceptible patients.

BACKGROUND: For clinical genomic studies with high-dimensional datasets, tree-based ensemble methods offer a powerful solution for variable selection and prediction taking into account the complex interrelationships between explanatory variables. One of the key component of the tree-building process is the splitting criterion. For survival data, the classical splitting criterion is the Logrank statistic. However, the presence of a fraction of nonsusceptible patients in the studied population advocates for considering a criterion tailored to this peculiar situation. RESULTS: We propose a bagging survival tree procedure for variable selection and prediction where the survival tree-building process relies on a splitting criterion that explicitly focuses on time-to-event survival distribution among susceptible patients. A simulation study shows that our method achieves good performance for the variable selection and prediction. Different criteria for evaluating the importance of the explanatory variables and the prediction performance are reported. Our procedure is illustrated on a genomic dataset with gene expression measurements from early breast cancer patients. CONCLUSIONS: In the presence of nonsusceptible patients among the studied population, our procedure represents an efficient way to select event-related explanatory covariates with potential higher-order interaction and identify homogeneous groups of susceptible patients.

A Rescue Strategy for Handling Unevaluable Patients in Simon's Two Stage Design.

For phase II oncology trials, Simon's two-stage design is the most commonly used strategy. However, when clinically unevaluable patients occur, the total number of patients included at each stage differs from what was initially planned. Such situations raise concerns about the operating characteristics of the trial design. This paper evaluates three classical ad hoc strategies and a novel one proposed in this work for handling unevaluable patients. This latter is called the rescue strategy which adapts the critical stopping rules to the number of unevaluable patients at each stage without modifying the planned sample size. blue Simulations show that none of these strategies perfectly match the original target constraints for type I and II error rates. Our rescue strategy is nevertheless the one which best approaches the target error rates requirement. A re-analysis of one real phase II clinical trials on metastatic cancer illustrates the use of the proposed strategy.

A Bagged, Partially Linear, Tree-Based Regression Procedure for Prediction and Variable Selection.

OBJECTIVES: In genomics, variable selection and prediction accounting for the complex interrelationships between explanatory variables represent major challenges. Tree-based methods are powerful alternatives to classical regression models. We have recently proposed the generalized, partially linear, tree-based regression (GPLTR) procedure that integrates the advantages of generalized linear regression (allowing the incorporation of confounding variables) and of tree-based models. In this work, we use bagging to address a classical concern of tree-based methods: their instability. METHODS: We present a bagged GPLTR procedure and three scores for variable importance. The prediction accuracy and the performance of the scores are assessed by simulation. The use of this procedure is exemplified by the analysis of a lung cancer data set. The aim is to predict the epidermal growth factor receptor (EGFR) mutation based on gene expression measurements, taking into account the ethnicity (confounder variable) and perform variable selection. RESULTS: The procedure performs well in terms of prediction accuracy. The scores differentiate predictive variables from noise variables. Based on a lung adenocarcinoma data set, the procedure achieves good predictive performance for EGFR mutation and selects relevant genes. CONCLUSION: The proposed bagged GPLTR procedure performs well for prediction and variable selection.

Patterns of chromosomal copy-number alterations in intrahepatic cholangiocarcinoma.

BACKGROUND: Intrahepatic cholangiocarcinomas (ICC) are relatively rare malignant tumors associated with a poor prognosis. Recent studies using genome-wide sequencing technologies have mainly focused on identifying new driver mutations. There is nevertheless a need to investigate the spectrum of copy number aberrations in order to identify potential target genes in the altered chromosomal regions. The aim of this study was to characterize the patterns of chromosomal copy-number alterations (CNAs) in ICC. METHODS: 53 patients having ICC with frozen material were selected. In 47 cases, DNA hybridization has been performed on a genomewide SNP array. A procedure with a segmentation step and a calling step classified genomic regions into copy-number aberration states. We identified the exclusively amplified and deleted recurrent genomic areas. These areas are those showing the highest estimated propensity level for copy loss (resp. copy gain) together with the lowest level for copy gain (resp. copy loss). We investigated ICC clustering. We analyzed the relationships between CNAs and clinico-pathological characteristics. RESULTS: The overall genomic profile of ICC showed many alterations with higher rates for the deletions. Exclusively deleted genomic areas were 1p, 3p and 14q. The main exclusively amplified genomic areas were 1q, 7p, 7q and 8q. Based on the exclusively deleted/amplified genomic areas, a clustering analysis identified three tumors groups: the first group characterized by copy loss of 1p and copy gain of 7p, the second group characterized by 1p and 3p copy losses without 7p copy gain, the last group characterized mainly by very few CNAs. From univariate analyses, the number of tumors, the size of the largest tumor and the stage were significantly associated with shorter time recurrence. We found no relationship between the number of altered cytobands or tumor groups and time to recurrence. CONCLUSION: This study describes the spectrum of chromosomal aberrations across the whole genome. Some of the recurrent exclusive CNAs harbor candidate target genes. Despite the absence of correlation between CNAs and clinico-pathological characteristics, the co-occurence of 7p gain and 1p loss in a subgroup of patients may suggest a differential activation of EGFR and its downstream pathways, which may have a potential effect on targeted therapies.

Influence of trial duration on the bias of the estimated treatment effect in clinical trials when individual heterogeneity is ignored.

In randomized clinical trials where the times to event of two treatment groups are compared under a proportional hazards assumption, it has been established that omitting prognostic factors from the model entails an underestimation of the hazards ratio. Heterogeneity due to unobserved covariates in cancer patient populations is a concern since genomic investigations have revealed molecular and clinical heterogeneity in these populations. In HIV prevention trials, heterogeneity is unavoidable and has been shown to decrease the treatment effect over time. This article assesses the influence of trial duration on the bias of the estimated hazards ratio resulting from omitting covariates from the Cox analysis. The true model is defined by including an unobserved random frailty term in the individual hazard that reflects the omitted covariate. Three frailty distributions are investigated: gamma, log-normal, and binary, and the asymptotic bias of the hazards ratio estimator is calculated. We show that the attenuation of the treatment effect resulting from unobserved heterogeneity strongly increases with trial duration, especially for continuous frailties that are likely to reflect omitted covariates, as they are often encountered in practice. The possibility of interpreting the long-term decrease in treatment effects as a bias induced by heterogeneity and trial duration is illustrated by a trial in oncology where adjuvant chemotherapy in stage 1B NSCLC was investigated.

A novel tree-based procedure for deciphering the genomic spectrum of clinical disease entities.

BACKGROUND: Dissecting the genomic spectrum of clinical disease entities is a challenging task. Recursive partitioning (or classification trees) methods provide powerful tools for exploring complex interplay among genomic factors, with respect to a main factor, that can reveal hidden genomic patterns. To take confounding variables into account, the partially linear tree-based regression (PLTR) model has been recently published. It combines regression models and tree-based methodology. It is however computationally burdensome and not well suited for situations for which a large number of exploratory variables is expected. METHODS: We developed a novel procedure that represents an alternative to the original PLTR procedure, and considered different selection criteria. A simulation study with different scenarios has been performed to compare the performances of the proposed procedure to the original PLTR strategy. RESULTS: The proposed procedure with a Bayesian Information Criterion (BIC) achieved good performances to detect the hidden structure as compared to the original procedure. The novel procedure was used for analyzing patterns of copy-number alterations in lung adenocarcinomas, with respect to Kirsten Rat Sarcoma Viral Oncogene Homolog gene (KRAS) mutation status, while controlling for a cohort effect. Results highlight two subgroups of pure or nearly pure wild-type KRAS tumors with particular copy-number alteration patterns. CONCLUSIONS: The proposed procedure with a BIC criterion represents a powerful and practical alternative to the original procedure. Our procedure performs well in a general framework and is simple to implement.