Methods for analyzing combined data from studies of workers exposed to low doses of radiation

Comparing regression methods for the two-stage clonal expansion model of carcinogenesis

In the statistical analysis of cohort data with risk estimation models, both Poisson and individual likelihood regressions are widely used methods of parameter estimation. In this paper, their performance has been tested with the biologically motivated two-stage clonal expansion (TSCE) model of carcinogenesis. To exclude inevitable uncertainties of existing data, cohorts with simple individual exposure history have been created by Monte Carlo simulation. To generate some similar properties of atomic bomb survivors and radon-exposed mine workers, both acute and protracted exposure patterns have been generated. Then the capacity of the two regression methods has been compared to retrieve a priori known model parameters from the simulated cohort data. For simple models with smooth hazard functions, the parameter estimates from both methods come close to their true values. However, for models with strongly discontinuous functions which are generated by the cell mutation process of transformation, the Poisson regression method fails to produce reliable estimates. This behaviour is explained by the construction of class averages during data stratification. Thereby, some indispensable information on the individual exposure history was destroyed. It could not be repaired by countermeasures such as the refinement of Poisson classes or a more adequate choice of Poisson groups. Although this choice might still exist we were unable to discover it. In contrast to this, the individual likelihood regression technique was found to work reliably for all considered versions of the TSCE model.

Second analysis of mortality of nuclear industry workers in Japan, 1986-1997

IA cohort study of nuclear industry workers was initiated in 1990 to determine the possible health effects of low-level radiation. A total of 5,527 deaths were ascertained among 176,000 male workers who had been retrospectively and/or prospectively followed for an average of 7.9 years during the observation period 1986-1997. Statistical analyses were made mainly on the prospective follow-up outcome of 120,000 workers followed for an average of 4.5 years. The standardized mortality ratio (and its 95% confidence interval) was 0.94 (0.90, 0.97) for 2,934 cases of all causes combined and 0.86 (0.82, 0.91) for 1,305 cases of non-malignant diseases combined, which suggested a healthy worker effect. For 1,191 cases of all cancers combined, it was 0.98 (0.93, 1.04), indicating no difference in mortality from that of the general population. In tests for trend of death rate with increasing radiation dose, no significant correlation was found for all cancers combined. For site-specific cancers, most cancers including leukemia showed no positive correlation with dose, except for cancers of the esophagus, stomach and rectum and multiple myeloma. External causes showed a significant correlation with dose. A separate questionnaire study indicated that these positive findings could be ascribed in part to lifestyle characteristics of the workers. For leukemia only, we attempted to estimate the excess relative risk per unit dose of radiation, which, with reservations because of its wide confidence interval, was within the range of variation of the risks reported in other radiation epidemiological studies. This population must be studied for a longer time and with a consideration of the possible effects of confounding factors.

Survey of methods and statistical models used in the analysis of occupational cohort studies

OBJECTIVES

This survey was conducted to determine the frequency with which different data analysis techniques are being used in occupational cohort studies. Of particular interest was the relative use of external and internal comparison groups, and the choice of multivariable model.

METHODS

Occupational cohort studies published in 1990-91 were located with Medline and Index Medicus, and the contents of several relevant journals were systematically reviewed. Each study was categorised by the methods of external or internal comparisons performed.

RESULTS

Of 200 occupational cohort studies identified, 104 (52%) conducted only external comparisons, 46 (23%) conducted only internal, and 50 (25%) presented both. Of those that used an external referent population, about two thirds used a national standard. 40 of the studies that performed internal comparisons fitted multivariable models, with use divided about equally between proportional hazards regression, Poisson regression, and logistic regression.

DISCUSSION

The finding that logistic regression is used quite commonly, even though it does not directly model time dependent data of the type frequently encountered in occupational cohort studies, was suprising. The reasons why investigators choose from among the available statistical and modelling techniques are likely to include familiarity, ease of use, restrictions in study population characteristics, especially study size, and others. Authors should be encouraged to be more explicit about the statistical methods used in the analysis of occupational cohort studies, as well as whether important assumptions about their data have been evaluated.

Job factors, radiation and cancer mortality at Oak Ridge National Laboratory: follow-up through 1984

A previous study of mortality among white men hired at Oak Ridge National Laboratory between 1943 and 1972 (n = 8,318) revealed an association between low-dose external penetrating ionizing radiation and cancer mortality in follow-up through 1984. The association was not observed in follow-up through 1977. This report considers the role of possible selection and confounding factors not previously studied. Control for hire during the World War II era and employment duration of less than 1 year had little effect on the radiation risk estimates. Risks associated with length of time spent in 15 job categories were considered as proxies for the effects of other occupational carcinogens. Adjustment for employment duration in each job category one at a time produced only small changes in the radiation risk estimate. Adjustment for potential exposures to beryllium, lead, and mercury also had little effect on the radiation risk estimates. These analyses suggest that selection factors and potential for chemical exposure do not account for the previously noted association of external radiation dose with cancer mortality. However, power to detect effects of chemical exposures is limited by a lack of individual exposure measures.

Use of human data in quantitative risk assessment of carcinogens: impact on epidemiologic practice and the regulatory process

Epidemiologic data are increasingly being used to assess cancer risk from chemicals as their value is recognized and as more and better studies become available. Weight-of-evidence approaches are now available for classifying the experimental and epidemiological evidence regarding human carcinogenicity. When the human data are extensive and of good quality, they should be given substantial weight in assessing risk. Both the positive and the negative epidemiologic data should be used in a quantitative risk assessment (QRA), because only then can an unbiased risk assessment be derived. Good-quality epidemiological studies are those with sound methodology, lack of bias, long enough follow-up times to observe a carcinogenic response, adequate exposure information, and dose-response information. Before a lack of carcinogenicity can be inferred, it is essential that the exposures be of substantial duration and intensity, and that the number of exposed persons be reasonably large. Epidemiologists need to give more attention to exposure assessment, because lack of quantitative exposure information is often the limiting factor that prevents the use of epidemiologic data in QRA. Development of methods to estimate historical workplace exposure intensities from surrogate industrial hygiene variables should receive high research priority, since they have the potential to increase the usefulness in QRA of many epidemiologic studies that have limited exposure information. Several frequently used surrogates for exposure measurements have limitations or pitfalls in their use. In particular, the use of "ever/never" exposed has a large potential to produce falsely negative results by means of a "dilution" effect, especially in the common case where the exposure distribution is skewed. Duration of exposure (rather than duration x intensity) may also give misleading results. There is little information to suggest that synergistic exposures to multiple toxicants in an industrial environment are likely to invalidate QRAs, probably because few studies have identified a group of workers with major workplace exposures to multiple carcinogens that cause the same type of cancer. Most of the interactive effects which have been identified to date are between smoking and some occupational carcinogen, so this possibility needs careful evaluation for smoking-related diseases. It is important to evaluate dose-response gradients in a QRA to obtain maximum precision and accuracy in the resulting risk coefficient. The analysis should take into account an appropriate cancer induction period. Various methods to account for cancer induction times are compared; those that incorporate a lag period or model the induction-time distribution are superior to other methods.(ABSTRACT TRUNCATED AT 400 WORDS)

[Meta-analysis in epidemiology (2): quantitative methods]

The main goal of this paper is to offer an introduction to the quantitative methods of meta-analysis used in epidemiology/public health, emphasizing on the most widespread used. We give also a brief reference of the methods to recognize publication bias. The limitations and advantages of the methods commented on are discussed.