Modeling of long-term screening for lung carcinoma

Association of smoking with tumor size at diagnosis in non-small cell lung cancer

Tumor size at diagnosis (TSD) indirectly reflects tumor growth rate. The relationship between TSD and smoking is poorly understood. The aim of the study was to determine the relationship between smoking and TSD. We reviewed 1712 newly diagnosed and previously untreated non-small cell lung cancer (NSCLC) patients' electronic medical records and collected tumor characteristics. Demographic and epidemiologic characteristics were derived from questionnaires administered during personal interviews. Univariate and multivariate linear regression models were used to evaluate the relationship between TSD and smoking controlling for demographic and clinical factors. We also investigated the relationship between the rs1051730 SNP in an intron of the CHRNA3 gene (the polymorphism most significantly associated with lung cancer risk and smoking behavior) and TSD. We found a strong dose dependent relationship between TSD and smoking. Current smokers had largest and never smokers smallest TSD with former smokers having intermediate TSD. In the multivariate linear regression model, smoking status (never, former, and current), histological type (adenocarcinoma versus SqCC), and gender were significant predictors of TSD. Smoking duration and intensity may explain the gender effect in predicting TSD. We found that the variant allele of rs1051730 in CHRNA3 gene was associated with larger TSD of squamous cell carcinoma. In the multivariate linear regression model, both rs1051730 and smoking were significant predictors for the size of squamous carcinomas. We conclude that smoking is positively associated with lung tumor size at the moment of diagnosis.

Exploring the uncertainties of early detection results: model-based interpretation of mayo lung project

BACKGROUND

The Mayo Lung Project (MLP), a randomized controlled clinical trial of lung cancer screening conducted between 1971 and 1986 among male smokers aged 45 or above, demonstrated an increase in lung cancer survival since the time of diagnosis, but no reduction in lung cancer mortality. Whether this result necessarily indicates a lack of mortality benefit for screening remains controversial. A number of hypotheses have been proposed to explain the observed outcome, including over-diagnosis, screening sensitivity, and population heterogeneity (initial difference in lung cancer risks between the two trial arms). This study is intended to provide model-based testing for some of these important arguments.

METHOD

Using a micro-simulation model, the MISCAN-lung model, we explore the possible influence of screening sensitivity, systematic error, over-diagnosis and population heterogeneity.

RESULTS

Calibrating screening sensitivity, systematic error, or over-diagnosis does not noticeably improve the fit of the model, whereas calibrating population heterogeneity helps the model predict lung cancer incidence better.

CONCLUSIONS

Our conclusion is that the hypothesized imperfection in screening sensitivity, systematic error, and over-diagnosis do not in themselves explain the observed trial results. Model fit improvement achieved by accounting for population heterogeneity suggests a higher risk of cancer incidence in the intervention group as compared with the control group.

Development of a multiplexed tumor-associated autoantibody-based blood test for the detection of non-small cell lung cancer

PURPOSE

Non-small cell lung cancer (NSCLC) has an overall 5-year survival of <15%; however, the 5-year survival for stage I disease is over 50%. Unfortunately, 75% of NSCLC is diagnosed at an advanced stage not amenable to surgery. A convenient serum assay capable of unambiguously identifying patients with NSCLC may provide an ideal diagnostic measure to complement computed tomography-based screening protocols.

EXPERIMENTAL DESIGN

Standard immunoproteomic method was used to assess differences in circulating autoantibodies among lung adenocarcinoma patients relative to cancer-free controls. Candidate autoantibodies identified by these discovery phase studies were translated into Luminex-based "direct-capture" immunobead assays along with 10 autoantigens with previously reported diagnostic value. These assays were then used to evaluate a second patient cohort composed of four discrete populations, including: 117 NSCLC (81 T(1-2)N(0)M(0) and 36 T(1-2)N(1-2)M(0)), 30 chronic obstructive pulmonary disorder (COPD)/asthma, 13 nonmalignant lung nodule, and 31 "normal" controls. Multivariate statistical methods were then used to identify the optimal combination of biomarkers for classifying patient disease status and develop a convenient algorithm for this purpose.

RESULTS

Our immunoproteomic-based biomarker discovery efforts yielded 16 autoantibodies differentially expressed in NSCLC versus control serum. Thirteen of the 25 analytes tested showed statistical significance (Mann-Whitney P < 0.05 and a receiver operator characteristic "area under the curve" over 0.65) when evaluated against a second patient cohort. Multivariate statistical analyses identified a six-biomarker panel with only a 7% misclassification rate.

CONCLUSIONS

We developed a six-autoantibody algorithm for detecting cases of NSCLC among several high-risk populations. Population-based validation studies are now required to assign the true value of this tool for identifying early-stage NSCLC.

Modeling excess lung cancer risk among screened arm participants in the Mayo Lung Project

BACKGROUND

The Mayo Lung Project (MLP) was a randomized clinical trial designed to test whether periodic screening by chest x-ray reduced lung cancer (LC) mortality in men who were high-risk smokers. Among MLP participants, there were more deaths from LC in the screening arm both at the trial's end and after long-term follow-up. Overdiagnosis was cited widely as an explanation for the MLP results, whereas a role for excess LC risk attributable to undergoing numerous chest x-ray screenings largely was unexamined. The authors of this report examined the consistency of the MLP data with a modified 2-stage clonal expansion (TSCE) model of excess LC risk.

METHODS

By using a simulation model calibrated to the initial MLP data, the authors examined the expected statistical variance of LC incidence and mortality between the screening and control arms. A Bayesian estimation framework using a modified version of the TSCE model to evaluate the role of excess LC risk attributable to chest x-ray screening was derived and applied to the MLP data.

RESULTS

Simulation experiments indicated that the overall difference in LC deaths and incidence between the study arm and the control arm was unlikely (P = .0424 and P = .0104, respectively) assuming no excess risk of LC. The authors estimated that the 10-year excess LC risk for a man aged 60 years who smoked and who received 10 chest x-ray screenings was 0.574% (P = .0021).

CONCLUSIONS

The excess LC risk observed among screening arm participants was found to be statistically significant with respect to the TSCE model framework in part because of the incorporation of key risk correlates of age and screen frequency into the estimation framework.

Reliable data on 5- and 10-year survival provide accurate estimates of 15-year survival in estrogen receptor-positive early-stage breast cancer

There are few studies of model-based survival projections using early empirical results for estimating long-term survival. Utilizing Early Breast Cancer Trialists' Collaborative Group (EBCTCG) data, a Markov model was generated to compare empirical results with those modeled beyond the empirical result time horizon in estrogen receptor (ER)-positive early-stage breast cancer (ESBC). Modeling 15-year survival based on 5- and 10-year EBCTCG data resulted in an average error estimate in breast cancer mortality of 0.75% [range -0.83 to 2.19%]. Although modeling life expectancy differences ranged from an underestimate of -7.93% to an overestimate of 12.64%, over the span of 15 years this corresponded to a loss of 18 days or a gain of 40 days of life. Reliable early survival data may be used to generate models that accurately estimate 15-year survival in ER-positive ESBC. Whether early survival data can be employed over the lifetime horizon remains to be demonstrated.

Calibration methods used in cancer simulation models and suggested reporting guidelines

Increasingly, computer simulation models are used for economic and policy evaluation in cancer prevention and control. A model's predictions of key outcomes, such as screening effectiveness, depend on the values of unobservable natural history parameters. Calibration is the process of determining the values of unobservable parameters by constraining model output to replicate observed data. Because there are many approaches for model calibration and little consensus on best practices, we surveyed the literature to catalogue the use and reporting of these methods in cancer simulation models. We conducted a MEDLINE search (1980 through 2006) for articles on cancer-screening models and supplemented search results with articles from our personal reference databases. For each article, two authors independently abstracted pre-determined items using a standard form. Data items included cancer site, model type, methods used for determination of unobservable parameter values and description of any calibration protocol. All authors reached consensus on items of disagreement. Reviews and non-cancer models were excluded. Articles describing analytical models, which estimate parameters with statistical approaches (e.g. maximum likelihood) were catalogued separately. Models that included unobservable parameters were analysed and classified by whether calibration methods were reported and if so, the methods used. The review process yielded 154 articles that met our inclusion criteria and, of these, we concluded that 131 may have used calibration methods to determine model parameters. Although the term 'calibration' was not always used, descriptions of calibration or 'model fitting' were found in 50% (n = 66) of the articles, with an additional 16% (n = 21) providing a reference to methods. Calibration target data were identified in nearly all of these articles. Other methodological details, such as the goodness-of-fit metric, were discussed in 54% (n = 47 of 87) of the articles reporting calibration methods, while few details were provided on the algorithms used to search the parameter space. Our review shows that the use of cancer simulation modelling is increasing, although thorough descriptions of calibration procedures are rare in the published literature for these models. Calibration is a key component of model development and is central to the validity and credibility of subsequent analyses and inferences drawn from model predictions. To aid peer-review and facilitate discussion of modelling methods, we propose a standardized Calibration Reporting Checklist for model documentation.

Differential diagnosis of solitary lung nodules with gene expression ratios

OBJECTIVE

We have developed a new technique that uses the ratios of select gene expression levels to translate complex genomic data into simple clinically relevant tests for the diagnosis and prognosis of cancer. We determined whether select gene pair ratio combinations can be used to detect and diagnose lung cancer with high accuracy and sensitivity.

METHODS

We used gene expression profiling data to train a ratio-based predictor model to discriminate among a set of samples (n = 145 total) composed of normal lung, small cell lung cancer, adenocarcinoma, squamous cell carcinoma, and pulmonary carcinoid (the training set). We then examined the optimal test in an independent set of samples (the test set, n = 122). Finally, we used one aspect of the test to determine whether the gene ratio technique was capable of detecting cancer in specimens from fine-needle aspirations performed ex vivo with normal lung (n = 14) and suspected tumor nodules (n = 15) acquired at our institution.

RESULTS

We found that a ratio-based test with 23 genes could be used to classify training set samples with 90% accuracy. This same test was similarly accurate (88%) when applied to the test set of samples. We also found that this test was 87% and 100% accurate at detecting cancer in normal and tumorous fine-needle aspiration specimens, respectively.

CONCLUSION

The gene expression ratio diagnostic technique is likely to aid in the differential diagnosis of solitary lung nodules in patients with suspected cancer and may also prove useful in developing lung cancer screening strategies that incorporate analysis of fine-needle aspiration specimens.

Prognostic and diagnostic relevance of hnRNP A2/B1, hnRNP B1 and S100 A2 in non-small cell lung cancer

BACKGROUND

S100 A2 and hnRNP A2/B1 (heterogeneous ribonucleoprotein A2/B1) with its splicing variant hnRNP B1 are proteins which are involved in cellular proliferation, differentiation and protein synthesis and are up-regulated in non-small cell lung cancer (NSCLC). Since previous studies using paraffin-embedded tissues indicated a high potential of these markers for diagnosis and screening the present analysis intended to validate these data applying cryostat sections.

METHODS

78 tumor-infiltrated lung cancer specimens and 66 adjacent histologically tumor-free tissues were analyzed; 11 autopsy specimens from patients who did not suffer from a malignant disease served as a control group. Cryostat sections were stained with monoclonal antibodies against hnRNP A2/B1, hnRNP B1 and S100 A2 and were compared with the previously established immunohistochemical profile of the same patients including EGFR, EGFRvIII, pEGFR, c-erbB-2, c-erbB-3, p53, Ki-67, bcl-2, p120 and microvessel density. Furthermore, these results were correlated with clinical parameters.

RESULTS

Expression of hnRNP A2/B1, hnRNP B1 and S100 A2 was increased in the tumor group when compared with the microscopic tumor-free specimens in 10% versus 5% (n.s.), 91% versus 5% and 65% versus 6%, respectively. Increased S100 and A2 hnRNP A2/B1 expressions were negative prognostic factors. With the exception of an increased EGFR expression in hnRNP A2/B1 negative cases the three analyzed markers did not correlate with the immunohistochemical parameters tested previously.

CONCLUSION

Comparison between tumor probes and tumor-free specimens of NSCLC patients failed to approve the diagnostic relevance of hnRNP A2/B1 shown in previous studies, whereas hnRNP B1 revealed a high tumor specificity that could be helpful for tumor cell screening. Moreover, S100 A2 and hnRNP A2/B1 confirmed to be valuable prognostic parameters.

Estimating the growth rates of primary lung tumours from samples with missing measurements

A method to estimate the population variability in tumour growth rate using incomplete data was developed. We assume exponential growth and lognormal distribution for the parameter of the growth curve. Estimates of growth rate obtained based on the cases with two measurements, one of which is obtained retrospectively, are biased towards lower growth rate. For the data sets where two measurements are available for some tumours and only one measurement for others (which means that no tumour was seen in retrospect for those cases), several approaches were developed that can eliminate or substantially reduce the bias. The relative error of the best estimates, as assessed by simulation, rarely exceeds 20 per cent. We found that the results of application of our estimation procedures to chest X-ray screening data agree well with the expectations.

CT Screening for Lung Cancer: Past and Ongoing Studies

It has been widely recognized that the oft-quoted randomized clinical trials (RCTs) of lung cancer screening by chest radiography--studies that were interpreted as showing no benefit--were seriously flawed. We begin by describing the shortcomings of these trials and presenting an analysis of the problems typically encountered in performing RCTs in this area. Screening for lung cancer using computed tomography (CT) has shown that CT offers great superiority over chest radiography in diagnosing small lung cancers in the three studies that performed both CT and chest radiography on all patients. The Early Lung Cancer Action Project (ELCAP), showed that false-positive results can be kept reasonably low and are much less common on repeat screening, and that CT screening can be managed with no notable excess of percutaneous or surgical biopsies when following a well-defined regimen of screening. This regimen details the parameters of the initial CT, the definition of a positive result, and the subsequent work-up of positive results. Following the updated International (I)-ELCAP protocol, it has been further found that (1) the frequency of positive results is low: 15% for the baseline cycle of screening and 6% for the subsequent cycles. (2) The frequency of screen-diagnoses as compared with all diagnoses is 97% or higher. (3) The relative frequency of presurgical Stage I is well over 80%; the median diameter of the screen-diagnosed cases on repeat screening is 8 mm (versus 15 mm at baseline screening). (4) A high percentage of the screen-diagnosed cases were genuine cancers which led to death if not treated. (5) The estimated 8-year cure rate for resected baseline screen-diagnosed lung cancers without evidence of lymph node metastases is 95% and for resected annual repeat cancers is 98%. (6) CT screening appears to be highly cost-effective. These preliminary results of CT screening suggests that the cure rate of screen-diagnosed lung cancer, using the I-ELCAP regimen of screening, may be over 70% as compared with that of usual care of 10% and that of chest radiographic screening of 20%.

Biomarkers in non-small cell lung cancer prevention

Identification of biomarkers is one of the most promising approaches for the detection of early malignant or even premalignant lesions with the chance of diagnosing early stages of non-small cell lung cancer that could be treated curatively. Alterations of chromosomes (3p, 5q, 9p), genes (Rb, C-myc, C-mos, hTERT), proteins (p16, p53, K-ras, hnRNP A2/B1, MCM2, EGFR, erbB-2, erbB-3, erbB-4) and others can be found in lung cancer. Some of these occur at early stages of the disease and few could serve as potential screening markers. The actual literature is reviewed and the relevance of the different biomarkers for early lung cancer detection is discussed.

Genetic susceptibility for lung cancer: interactions with gender and smoking history and impact on early detection policies

OBJECTIVES

To identify a subgroup of former, current and never smokers (males and females) at high risk for developing lung cancer, based on their genetic susceptibility profiles, to estimate their lifetime probabilities of the disease, and to assess the potential mortality reduction that could be achieved by screening the high-risk group.

METHODS

Case-control data (764 cases and 677 matched controls), on two assays of DNA damage and repair (mutagen susceptibility and DNA repair capacity, DRC) in different smoking categories were used to obtain estimates of susceptibility using the formula of total probability. The estimates were inserted into a model of lung cancer natural history and detection by screening to assess mortality reduction due to screening of high-risk individuals.

RESULTS

The high-risk group was defined as that 12.5% of the population who were above the third quartile for bleomycin sensitivity and below the median for DRC. High-risk male current, former, and never smokers had lifetime probabilities for developing lung cancer of 38, 21, and 5%, respectively. Females had lower probabilities to develop lung cancer: 15, 8, and 1.5% for high-risk current, former, and never smokers, respectively. Screening of high-risk smokers (12.5% of all smokers) reduced overall mortality by 7% compared to 30% reduction if all smokers were screened. Analogous results were obtained for former and never smokers.

CONCLUSIONS

Genetic susceptibility constitutes an important factor in the selection of a high-risk group for early lung cancer detection.

CT screening for lung cancer

The Early Lung Cancer Action Project (ELCAP) recently demonstrated that earlier diagnosis of lung cancer can be achieved with CT, and these results have led to considerable demand for CT screening. The advisability of screening seems obvious, as screening has been shown to provide for lung cancer treatment at a relatively early stage, leading to a better chance to avert death from lung cancer than when treatment is prompted by symptoms and/or signs. There are, however, countervailing ideas that CT lung cancer screening has not yet been demonstrated to 'save lives.' Further, it has been stated that CT screening has a notable problem of "overdiagnosis," meaning that screening finds lesions that are not life threatening, leading to unnecessary surgery. These concerns have led to the argument that assessing 'lives saved,' as well as the effects of overdiagnosis, can only be achieved with a randomized, controlled trial comparing CT screening with no screening, using a mortality endpoint. To this end, the National Lung Screening Trial (NLST) has been funded. This randomized, controlled trial is the most expensive screening study ever proposed. It compares CT screening with chest X-ray screening, and its designers envision that it will provide an answer about the benefit of CT screening, or lack thereof, in about 10 years. We do not question the value of the randomized design of 'treatment' trials for comparing competing interventions (treatments), but we have serious concerns about the use of randomization in the evaluation of a diagnostic test, such as CT. We feel that randomization is not necessary for evaluating a diagnostic test and may generate misleading results. Rather, we feel that the desired information is how often and how early is the disease diagnosed using that test. The purpose of this article is to raise the general level of concern about the underpinnings of such randomized 'screening' trials, and to convey some of the evidence that led to our pessimism about the NLST.