Frequentist and Bayesian approaches for a joint model for prostate cancer risk and longitudinal prostate-specific antigen data

The paper describes the use of frequentist and Bayesian shared-parameter joint models of longitudinal measurements of prostate-specific antigen (PSA) and the risk of prostate cancer (PCa). The motivating dataset corresponds to the screening arm of the Spanish branch of the European Randomized Screening for Prostate Cancer study. The results show that PSA is highly associated with the risk of being diagnosed with PCa and that there is an age-varying effect of PSA on PCa risk. Both the frequentist and Bayesian paradigms produced very close parameter estimates and subsequent 95% confidence and credibility intervals. Dynamic estimations of disease-free probabilities obtained using Bayesian inference highlight the potential of joint models to guide personalized risk-based screening strategies.     

EPA Underestimates,Oversimplifies, Miscommunicates,and Mismanages Cancer Risks by Ignoring Human Susceptibility

If exposed to an identical concentration of a carcinogen, every human being would face a different level of risk, determined by his or her genetic, environmental, medical, and other uniquely individual characteristics. Various lines of evidence indicate that this susceptibility variable is distributed rather broadly in the human population, with perhaps a factor of 25‐ to 50‐fold between the center of this distribution and either of its tails, but cancer risk assessment at the EPA and elsewhere has always treated every (adult) human as identically susceptible. The National Academy of Sciences “Silver Book” concluded that EPA and the other agencies should fundamentally correct their mis‐computation of carcinogenic risk in two ways: (1) adjust individual risk estimates upward to provide information about the upper tail; and (2) adjust population risk estimates upward (by about sevenfold) to correct an underestimation due to a mathematical property of the interindividual distribution of human susceptibility, in which the susceptibility averaged over the entire (right‐skewed) population exceeds the median value for the typical human. In this issue of Risk Analysis, Kenneth Bogen disputes the second adjustment and endorses the first, though he also relegates the problem of underestimated individual risks to the realm of “equity concerns” that he says should have little if any bearing on risk management policy. In this article, I show why the basis for the population risk adjustment that the NAS recommended is correct—that current population cancer risk estimates, whether they are derived from animal bioassays or from human epidemiologic studies, likely provide estimates of the median with respect to human variation, which in turn must be an underestimate of the mean. If cancer risk estimates have larger “conservative” biases embedded in them, a premise I have disputed in many previous writings, such a defect would not excuse ignoring this additional bias in the direction of underestimation. I also demonstrate that sensible, legally appropriate, and ethical risk policy must not only inform the public when the tail of the individual risk distribution extends into the “high‐risk” range, but must alter benefit‐cost balancing to account for the need to try to reduce these tail risks preferentially.     

Building multivariable prognostic and diagnostic models: transformation of the predictors by using fractional polynomials

To be useful to clinicians, prognostic and diagnostic indices must be derived from accurate models developed by using appropriate data sets. We show that fractional polynomials, which extend ordinary polynomials by including non-positive and fractional powers, may be used as the basis of such models. We describe how to fit fractional polynomials in several continuous covariates simultaneously, and we propose ways of ensuring that the resulting models are parsimonious and consistent with basic medical knowledge. The methods are applied to two breast cancer data sets, one from a prognostic factors study in patients with positive lymph nodes and the other from a study to diagnose malignant or benign tumours by using colour Doppler blood flow mapping. We investigate the problems of biased parameter estimates in the final model and overfitting using cross-validation calibration to estimate shrinkage factors. We adopt bootstrap resampling to assess model stability. We compare our new approach with conventional modelling methods which apply stepwise variables selection to categorized covariates. We conclude that fractional polynomial methodology can be very successful in generating simple and appropriate models.     

Bayesian assessment of times to diagnosis in breast cancer screening

Breast cancer is one of the diseases with the most profound impact on health in developed countries and mammography is the most popular method for detecting breast cancer at a very early stage. This paper focuses on the waiting period from a positive mammogram until a confirmatory diagnosis is carried out in hospital. Generalized linear mixed models are used to perform the statistical analysis, always within the Bayesian reasoning. Markov chain Monte Carlo algorithms are applied for estimation by simulating the posterior distribution of the parameters and hyperparameters of the model through the free software WinBUGS.     

Cancer Dose-Response Modeling of Epidemiological Data on Worker Exposures to Aldrin and Dieldrin

The paper applies classical statistical principles to yield new tools for risk assessment and makes new use of epidemiological data for human risk assessment. An extensive clinical and epidemiological study of workers engaged in the manufacturing and formulation of aldrin and dieldrin provides occupational hygiene and biological monitoring data on individual exposures over the years of employment and provides unusually accurate measures of individual lifetime average daily doses. In the cancer dose-response modeling, each worker is treated as a separate experimental unit with his own unique dose. Maximum likelihood estimates of added cancer risk are calculated for multistage, multistage-Weibull, and proportional hazards models. Distributional characterizations of added cancer risk are based on bootstrap and relative likelihood techniques. The cancer mortality data on these male workers suggest that low-dose exposures to aldrin and dieldrin do not significantly increase human cancer risk and may even decrease the human hazard rate for all types of cancer combined at low doses (e.g., 1 g/kg/day). The apparent hormetic effect in the best fitting dose-response models for this data set is statistically significant. The decrease in cancer risk at low doses of aldrin and dieldrin is in sharp contrast to the U.S. Environmental Protection Agency's upper bound on cancer potency based on mouse liver tumors. The EPA's upper bound implies that lifetime average daily doses of 0.0000625 and 0.00625 g/kg body weight/day would correspond to increased cancer risks of 0.000001 and 0.0001, respectively. However, the best estimate from the Pernis epidemiological data is that there is no increase in cancer risk in these workers at these doses or even at doses as large as 2 g/kg/day.     

Sensitivity of Physiologically Based Pharmacokinetic Models to Variation in Model Parameters: Methylene Chloride

The parameters in a physiologically based pharmacokinetic (PBPK) model of methylene chloride were varied systematically, and the resulting variation in a number of model outputs was determined as a function of time for mice and humans at several exposure concentrations. The importance of the various parameters in the model was highly dependent on the conditions (concentration, species) for which the simulation was performed and the model output (dose surrogate) being considered. Model structure also had a significant impact on the results. For sensitivity analysis, particular attention must be paid to conservation equations to ensure that the variational calculations do not alter mass balance, introducing extraneous effects into the model. All of the normalized sensitivity coefficients calculated in this study ranged between −1.12 and 1, and most were much less than 1 in absolute value, indicating that individual input errors are not greatly amplified in the outputs. In addition to ranking parameters in terms of their impact on model predictions, time-dependent sensitivity analysis can also be used as an aid in the design of experiments to estimate parameters by predicting the experimental conditions and sampling points which will maximize parameter identifiability.     

Probabilistic Risk Analysis for a High-Level Radioactive Waste Repository

A probabilistic risk analysis (PRA) for a high-level radioactive waste repository is very important since it gives an estimate of its health impacts, allowing comparisons to be made with the health impacts of competing technologies. However, it is extremely difficult to develop a credible PRA for a specific repository site because of large uncertainties in future climate, hydrology, geological processes, etc. At best, such a PRA would not be understandable to the public. An alternative proposed here is to develop a PRA for an average U.S. site, taking all properties of the site to be the U.S. average. The results are equivalent to the average results for numerous randomly selected sites. Such a PRA is presented here; it is easy to understand, and it is not susceptible to substantial uncertainty. Applying the results to a specific repository site then requires only a simple, intuitively acceptable "leap of faith" in assuming that with large expenditures of effort and money, experts can select a site that would be at least as secure as a randomly selected site.     

Significance of Exposure Assessment to Analysis of Cancer Risk from Inorganic Arsenic in Drinking Water in Taiwan

The primary source of evidence that inorganic arsenic in drinking water is associated with increased mortality from cancer at internal sites (bladder, liver, lung, and other organs) is a large ecologic study conducted in regions of Southwest Taiwan endemic to Blackfoot disease. The dose-response patterns for lung, liver, and bladder cancers display a nonlinear dose-response relationship with arsenic exposure. The data do not appear suitable, however, for the more refined task of dose-response assessment, particularly for inference of risk at the low arsenic concentrations found in some U.S. water supplies. The problem lies in variable arsenic concentrations between the wells within a village, largely due to a mix of shallow wells and deep artesian wells, and in having only one well test for 24 (40%) of the 60 villages. The current analysis identifies 14 villages where the exposure appears most questionable, based on criteria described in the text. The exposure values were then changed for seven of the villages, from the median well test being used as a default to some other point in the village's range of well tests that would contribute to smoothing the appearance of a dose-response curve. The remaining seven villages, six of which had only one well test, were deleted as outliers. The resultant dose-response patterns showed no evidence of excess risk below arsenic concentrations of 0.1 mg/l. Of course, that outcome is dependent on manipulation of the data, as described. Inclusion of the seven deleted villages would make estimates of risk much higher at low doses. In those seven villages, the cancer mortality rates are significantly high for their exposure levels, suggesting that their exposure values may be too low or that other etiological factors need to be taken into account.     

Detecting multiple change points in piecewise constant hazard functions

The National Cancer Institute (NCI) suggests a sudden reduction in prostate cancer mortality rates, likely due to highly successful treatments and screening methods for early diagnosis. We are interested in understanding the impact of medical breakthroughs, treatments, or interventions, on the survival experience for a population. For this purpose, estimating the underlying hazard function, with possible time change points, would be of substantial interest, as it will provide a general picture of the survival trend and when this trend is disrupted. Increasing attention has been given to testing the assumption of a constant failure rate against a failure rate that changes at a single point in time. We expand the set of alternatives to allow for the consideration of multiple change-points, and propose a model selection algorithm using sequential testing for the piecewise constant hazard model. These methods are data driven and allow us to estimate not only the number of change points in the hazard function but where those changes occur. Such an analysis allows for better understanding of how changing medical practice affects the survival experience for a patient population. We test for change points in prostate cancer mortality rates using the NCI Surveillance, Epidemiology, and End Results dataset.