Biomedical subjects

Peter Salzman

Publications and source records attributed to Peter Salzman.

5 recordsLinked to original sources

PubMed2007

A natural history model of stage progression applied to breast cancer.

Invasive breast cancer is commonly staged as local, regional or distant disease. We present a stochastic model of the natural history of invasive breast cancer that quantifies (1) the relative rate that the disease transitions from the local, regional to distant stages, (2) the tumour volume at the stage transitions and (3) the impact of symptom-prompted detection on the tumour size and stage of invasive breast cancer in a population not screened by mammography. By symptom-prompted detection, we refer to tumour detection that results when symptoms appear that prompt the patient to seek clinical care. The model assumes exponential tumour growth and volume-dependent hazard functions for the times to symptomatic detection and stage transitions. Maximum likelihood parameter estimates are obtained based on SEER data on the tumour size and stage of invasive breast cancer from patients who were symptomatically detected in the absence of screening mammography. Our results indicate that the rate of symptom-prompted detection is similar to the rate of transition from the local to regional stage and an order of magnitude larger than the rate of transition from the regional to distant stage. We demonstrate that, in the even absence of screening mammography, symptom-prompted detection has a large effect on reducing the occurrence of distant staged disease at initial diagnosis.

Adult↗

PubMed2006

Using complexity for the estimation of Bayesian networks.

Statistical inference of graphical models has become an important tool in the reconstruction of biological networks of the type which model, for example, gene regulatory interactions. In particular, the construction of a score-based Bayesian posterior density over the space of models provides an intuitive and computationally feasible method of assessing model uncertainty and of assigning statistical confidence to structural features. One problem which frequently occurs with this approach is the tendency to overestimate the degree of model complexity. Spurious graphical features obtained in this way may affect the inference in unpredictable ways, even when using scoring techniques, such as the Bayesian Information Criterion (BIC), that are specifically designed to compensate for overfitting. In this article we propose a simple adjustment to a BIC-based scoring procedure. The method proceeds in two steps. In the first step we derive an independent estimate of the parametric complexity of the model. In the second we modify the BIC score so that the mean parametric complexity of the posterior density is equal to the estimated value. The method is applied to a set of test networks, and to a collection of genes from the yeast genome known to possess regulatory relationships. A Bayesian network model with binary responses is employed. In the examples considered, we find that the number of spurious graph edges inferred is reduced, while the effect on the identification of true edges is minimal.

Algorithms↗

PubMed2006

Utility of correlation measures in analysis of gene expression.

The role of the correlation structure of gene expression data are two-fold: It is a source of complications and useful information at the same time. Ignoring the strong stochastic dependence between gene expression levels in statistical methodologies for microarray data analysis may deteriorate their performance. However, there is a host of valuable information in the correlation structure that deserves a closer look. A proper use of correlation measures can remedy deficiencies of currently practiced methods that are focused too heavily on strong effects in terms of differential expression of genes. The present paper discusses the utility of correlation measures in microarray data analysis and gene regulatory network reconstruction, along with various pitfalls in both research areas that have been uncovered in methodological studies. These issues have broad applicability to all genomic studies examining the biology, diagnosis, and treatment of neurological disorders.

Animals↗

PubMed2006

A stochastic simulation model of U.S. breast cancer mortality trends from 1975 to 2000.

BACKGROUND: We present a simulation model that predicts U.S. breast cancer mortality trends from 1975 to 2000 and quantifies the impact of screening mammography and adjuvant therapy on these trends. This model was developed within the Cancer Intervention and Surveillance Network (CISNET) consortium. METHOD: A Monte Carlo simulation is developed to generate the life history of individual breast cancer patients by using CISNET base case inputs that describe the secular trend in breast cancer risk, dissemination patterns for screening mammography and adjuvant treatment, and death from causes other than breast cancer. The model generates the patient's age, tumor size and stage at detection, mode of detection, age at death, and cause of death (breast cancer versus other) based in part on assumptions on the natural history of breast cancer. Outcomes from multiple birth cohorts are summarized in terms of breast cancer mortality rates by calendar year. RESULT: Predicted breast cancer mortality rates follow the general shape of U.S. breast cancer mortality rates from 1975 to 1995 but level off after 1995 as opposed to following an observed decline. Sensitivity analysis revealed that the impact adjuvant treatment may be underestimated given the lack of data on temporal variation in treatment efficacy. CONCLUSION: We developed a simulation model that uses CISNET base case inputs and closely, but not exactly, reproduces U.S. breast cancer mortality rates. Screening mammography and adjuvant therapy are shown to have both contributed to a decline in U.S. breast cancer mortality.

Age Factors↗

PubMed2004

Simulation-based parameter estimation for complex models: a breast cancer natural history modelling illustration.

Simulation-based parameter estimation offers a powerful means of estimating parameters in complex stochastic models. We illustrate the application of these ideas in the setting of a natural history model for breast cancer. Our model assumes that the tumor growth process follows a geometric Brownian motion; parameters are estimated from the SEER registry. Our discussion focuses on the use of simulation for computing the maximum likelihood estimator for this class of models. The analysis shows that simulation provides a straightforward means of computing such estimators for models of substantial complexity.

Algorithms↗