Search

ABSTRACTVariations in biological behavior suggest that each carcinogenic human papillomavirus (HPV) type should be considered individually in etiologic studies. HPV genotyping assays might have clinical applications if they are approved for use by the FDA. A widely used genotyping assay is the Roche Linear Array HPV genotyping test (LA). We used LA to genotype the HPV isolates from cervical specimens from women with the full spectrum of cervical disease: cervical cancer, cervical intraepithelial neoplasia (CIN), and HPV infections. To explore the feasibility and value of the automated reading of the LA results, we custom-designed novel, optical imaging software that provides optical density measurements of LA bands. We compared unmagnified visual examination with the automated measurements. The two measurements were highly associated. By either method, the threshold between a negative and a positive result was fairly sharp, with a clear bimodal distribution. Visually, most positive results were judged to be strong or medium, with fewer equivocal results categorized as weak (9.5% of positive samples), very weak (6.5% of positive samples), or extremely weak (7.7% of positive samples). The automated measurements of the intensities were significantly associated with the strength of the visual categories (P< 0.001). At the extremes of the automated signal intensities (=20 units or =120 units), the bands were almost always categorized visually as negative and positive, respectively. In the equivocal zone (20 to 119 units), specimens were more increasingly likely to be judged to be visually positive as the number of other, definite infections on the same strip increased (Pfor trend < 0.001). Multiple, concurrent infections comprise =25% of HPV infections; thus, any systematic visual tendency that influences their evaluation when the result is equivocal should be minimized. Therefore, automated reading is probably worth development if easy-to-calibrate hardware and software can be optimized.

AbstractSeveral earlier case‐control studies reported inverse associations of cervical squamous intraepithelial lesions (SIL) with high dietary or biomarker levels of carotenoids, folate, and vitamins C and E. However, most studies did not measure the primary causal factor, cancer‐associated genital human papillomaviruses (HPV), now detected by sensitive viral DNA tests. This nested case‐control study assessed whether high dietary intakes of these nutrients, plus zinc and vitamin A, reduced SIL risk in cancer‐associated HPV DNA‐positive women. Using a 60‐item food‐frequency questionnaire, nutrient estimates were obtained for 33 incident cases with high‐grade lesions, 121 with low‐grade lesions, 97 with equivocal SIL, and 806 cytologically normal controls sampled from a large prospective cohort study. Baseline cervicovaginal lavages were tested for HPV DNA by the polymerase chain reaction. Among DNA‐positive cases (n = 68) and controls (n = 69), age‐adjusted odds ratios (ORs) of SIL in the highest vs. the lowest nutrient quartiles were 1.4 [95% confidence interval (CI) = 0.5–4.2] for vitamin A, 0.6 (CI = 0.2–2.0) for ß‐carotene, 1.3 (CI = 0.4–3.6) for vitamin C, 1.0 (CI = 0.4–3.6) for vitamin E, 0.7 (CI = 0.3–2.1) for folate, and 0.8 (CI= 0.3–2.2) for zinc. ORs in HPV DNA‐negative women approximated 1.0, with the exception of vitamin E (OR = 0.5, CI= 0.3–0.9). These results do not support a protective role for the above nutrients against low‐grade or equivocal SIL, which constituted the majority of diagnoses in this study.

A case-control study of 107 colon cancer cases and 486 controls from an oncological clinic in Istanbul was conducted to examine the association between occupational physical activity and colon cancer in Turkey, where incidence of this disease is low.

Classification of exposure into two levels--one consisting exclusively of unexposed individuals and the other consisting of exposed and perhaps unexposed ones--yields an unbiased estimate of attributable risk when misclassification is nondifferential. The authors advocate, therefore, the use of a broad definition of exposure when estimating attributable risk. Based on this idea, they justify a simple and robust method for estimating the overall attributable risk from several exposures that is based on a division of subjects into two groups, a baseline consisting of those unexposed to all exposures and everyone else.

The authors argue that one can never be certain whether an exposure variable which is measured with error is subject to differential misclassification in either a case-control study or a cohort study. They present hypothetic examples that demonstrate that even when misclassification is nondifferential in a 2 x 3 table, the observed odds ratios in the 2 x 2 table created by collapsing over two exposure levels can be either in the opposite direction from or more extreme than the odds ratio that would be obtained if exposures were classified correctly. The anomalies are explained by the observation that the 2 x 2 tables exhibit differential misclassification. In general, collapsing over categories which have different risks of disease and different probabilities of exposure misclassification can induce differential misclassification and even nonconservative estimates of relative risk. Collapsing of exposure levels can occur in the analysis or at the exposure assessment stage. Since indistinguishable categories can be collapsed implicitly, blind assessment of exposure, i.e., assignment without knowledge of disease status, does not guarantee that misclassification is nondifferential.

Energy adjustment is used in nutritional epidemiology in an attempt to separate specific effects of macronutrients (carbohydrate, fat, and protein) from one another and from the generic effect of the total quantity of energy consumed. However, models in which the risk of disease is allowed to depend simultaneously on daily total energy consumption and separate components of energy that sum to the total are not identifiable: the specific effects of individual macronutrients and the generic effect of energy cannot be disentangled by multivariate analysis. The standard, residual, and partition methods exclude one or more macronutrients from consideration, thereby allowing estimation, but the parameters that are estimated no longer represent specific macronutrient or generic energy effects. Therefore, an interpretation of a regression coefficient from these methods as a specific effect of a macronutrient or as the generic effect of energy requires additional, almost always questionable, assumptions. For example, a conclusion based on data alone that there is a specific fat effect upon the development of breast cancer but no specific effects of other macronutrients and no generic energy effect is not possible. Notwithstanding these serious problems, some useful etiologic inference still can be made.

The authors discuss the interpretation of four alternative energy adjustment methods (Residual, Standard, Partition, and Nutrient Density) that have been proposed for the analysis of nutritional epidemiology studies. These methods have so far been compared under circumstances where intake of the nutrient of interest is measured as a continuous variable. Because it is common practice to categorize nutrient intakes in the analysis, the authors investigate the effect of such categorization on the interpretation of results from the four methods with the use of computer simulations and statistical theory. They consider four cases: where the nutrient intake is either divided into quartiles or ordered so as to investigate trend over the quartile groups, combined with using an adjusting variable that is either continuous or categorized. The results show: 1) the Residual, Standard, and Partition methods are no longer equivalent as they are in the continuous case; 2) compared with the Standard method, the Residual method appears to be more powerful for detecting trends in relative odds, is more robust to residual confounding when the adjustment variable is categorized, and provides more meaningful odds ratios; and 3) the Residual and Nutrient Density methods give closely similar results.

Gender differences in risk for leukemia and other selected and combined disease categories were examined by major occupational category for 74,828 benzeneexposed workers compared to 35,805 unexposed workers from 12 cities in China. No significant differences in the relative risks for total mortality and cancer mortality were found between female and male benzeneexposed workers, although risks tended to be somewhat higher among male than among female employees. Both female and male workers in several occupational categories had notably increased risks for all hematopoietic and lymphoproliferative HLP malignant and nonmalignant disorders combined and for total leukemia. Variation in risk for HLP disorders by occupational category was observed in both genders, with highest risks for male and female chemical manufacturing workers, female nonproduction employees, and male printers. However, the numbers of leukemia and other HLP malignancies in each category were small. The findings suggest that both female and male benzeneexposed workers in several occupational categories experience excess leukemia and other HLP disorders with relatively minor gender differences. Although this population is one of the largest cohorts of benzeneexposed workers studied to date, evaluation of the observed variation in risk for HLP neoplasms among the occupational groups for workers of each gender is limited by the small numbers of these relatively rare malignancies.

Types of control groups are evaluated using the principles described in paper 1 of the series, "Selection of Controls in Case-Control Studies" (S. Wacholder et al. Am J Epidemiol 1992;135:1019-28). Advantages and disadvantages of population controls, neighborhood controls, hospital or registry controls, medical practice controls, friend controls, and relative controls are considered. Problems with the use of decreased controls and proxy respondents are discussed.

A synthesis of classical and recent thinking on the issues involved in selecting controls for case-control studies is presented in this and two companion papers (S. Wacholder et al. Am J Epidemiol 1992;135:1029-50). In this paper, a theoretical framework for selecting controls in case-control studies is developed. Three principles of comparability are described: 1) study base, that all comparisons be made within the study base; 2) deconfounding, that comparisons of the effects of the levels of exposure on disease risk not be distorted by the effects of other factors; and 3) comparable accuracy, that any errors in measurement of exposure be nondifferential between cases and controls. These principles, if adhered to in a study, can reduce selection, confounding, and information bias, respectively. The principles, however, are constrained by an additional efficiency principle regarding resources and time. Most problems and controversies in control selection reflect trade-offs among these four principles.

Several design options available in the planning stage of case-control studies are examined. Topics covered include matching, control/case ratio, choice of nested case-control or case-cohort design, two-stage sampling, and other methods that can be used for control selection. The effect of potential problems in obtaining comparable accuracy of exposure is also examined. A discussion of the difficulty in meeting the principles of study base, deconfounding, and comparable accuracy (S. Wacholder et al. Am J Epidemiol 1992;135:1019-28) in a single study completes this series of papers.

There are disagreements in the literature about the criteria to be used to ascertain whether or not a measure of association is confounded. The authors postulate the general principle that a crude unconfounded measure of association is structured as a weighted average of the stratum-specific values of the measure. They examine the relationships between stratum-specific measures of association, crude overall measures, and weighted averages of stratum-specific measures, and indicate how these relationships may be used to define criteria for the assessment of confounding in cohort studies in which the exposure, disease, and stratification variables are classified dichotomously. The criteria presented differ for the risk ratio and for the disease-odds ratio. In other words, one can reach different conclusions about the confounding effect of a given extraneous variable, depending on which measure of association is chosen. This view differs from that of Miettinen and Cook (Confounding: essence and detection. Am J Epidemiol 1981;114:593-603) who postulated one set of criteria for the assessment of confounding, which was applicable to both measures of association. These different approaches may lead to different conclusions about the presence or absence of confounding.

A key assumption made when using a validation study to correct an estimate of relative risk for bias due to misclassification or measurement error is that the available measure, known to have error but nonetheless used routinely in the main study, is compared to a gold standard measured without error. In most epidemiologic applications, the putative gold standard is in fact measured with error. The effect of the violation of the assumption on the corrected estimate depends on the magnitudes of the errors in the two measures and on their correlation. In particular, when the errors are negatively correlated, independent, or weakly positively correlated, the corrected estimate will tend to overcorrect beyond the true value.

The authors discuss the interpretation of three alternative energy adjustment models for nutritional epidemiology. It is shown that four different effects are addressed by these models: 1) adding nutrient N, 2) substituting nutrient N for "other" nutrients, 3) adding "other" nutrients, and 4) adding both N and "other" nutrients in a specific ratio. Each of these effects may be estimated from any of the three models. The relative standard errors for the four estimated effects are also provided.

Serological assays for measuring antibodies to human papillomavirus type 16 (HPV-16) virus-like particles (VLPs) have become important epidemiologic tools in recent years. However, the interlaboratory replicability of these assays has not been assessed. In this investigation, three laboratories tested a panel of specimens obtained from two different groups: 265 subjects in a vulvar cancer case-control study and 107 healthy volunteer blood donors. Each laboratory used an enzyme-linked immunosorbent assay (ELISA), but no attempt was made to standardize assay procedures among the three laboratories. The data showed good day-to-day intralaboratory replicability in laboratory 1 (correlation coefficient, > or = 0.88) and good intra-assay variability in laboratory 3 (correlation coefficient, > or = 0.93). Interlaboratory correlations, likewise, ranged between 0.61 and 0.80 in both case-control study subjects and healthy blood donors, indicating that ELISA optical density (OD) values between laboratories were linearly related regardless of the population. Kappa coefficients (kappa), based on each laboratory's categorical interpretation of its results (as positive or negative), showed good agreement (kappa, > 0.6) in case-control study subjects and moderate agreement (kappa, > or = 0.4) in blood donors, a population that had few strongly positive sera. When OD values near seropositive cutoffs were treated as indeterminates, there was little discordance between laboratories in either population. The data suggest that each laboratory measured the same humoral immune response and that their HPV-16 VLP ELISAs performed similarly (Pearson correlations). Interlaboratory differences, however, probably due to reagents and procedures, were considerably greater than intralaboratory day-to-day variability. Interlaboratory agreement in determining seropositivity (kappa) could be improved by sharing positive and negative serum controls and by treating marginal results as indeterminate. As part of continuing cooperation to improve interlaboratory agreement, we are preparing bulk serum control specimens to be shared and made available to interested researchers.

Although an estimate of the odds ratio adjusted for other covariates can be obtained by logistic regression, until now there has been no simple way to estimate other interesting parameters such as the risk ratio and risk difference multivariately for prospective binomial data. These parameters can be estimated in the generalized linear model framework by choosing different link functions or transformations of binomial or binary data. Macros for use with the program GLIM provide a simple method to compute parameters other than the odds ratio while adjusting for confounding factors. A data set presented previously is used as an example.

A multi-cancer site, multi-factor case-control study was undertaken to generate hypotheses about possible occupational carcinogens. Probing interviews were carried out with over 2,000 subjects. All incident cases of 19 sites of cancer in males aged 35-70 and resident in Montreal were eligible. The interview was designed to obtain detailed lifetime job histories, and information on potential confounders. Each job history was reviewed by a team of chemists who translated it into a history of occupational exposures. These occupational exposures were then analyzed as potential risk factors in relation to the sites of cancer included. For each site of cancer analyzed as a case series, controls were selected from among the other cancer sites in the study. This report concerns the associations between sites of cancer for which there were over 100 cases processed (stomach; colorectal, also analyzed by subsites; lung; prostate; bladder; kidney; non-Hodgkin's lymphoma) and nine organic dusts (wood; paper; grain; flour; fabrics; cotton; wool; synthetics; fur). All site-exposure combinations were investigated. The ones that provided the most interesting leads were lung-wood dust (odds ratio (OR) = 1.5), stomach-wood dust (OR = 1.5), colorectal-synthetic fiber (OR = 1.5), bladder-synthetic fiber (OR = 1.8), non-Hodgkin's lymphoma-cotton dust (OR = 1.9), colon-grain dust (OR = 2.6), prostate-grain dust (OR = 2.2), and prostate-paper dust (OR = 2.0). Only the associations with wood dust, synthetic fibers and cotton dust showed some evidence of "dose-response" with duration of exposure. Because it is such a common exposure and appears to increase lung and stomach cancer risks, wood dust may be responsible for a great deal of occupational cancer.

The case-cohort design can be an economic alternative to the standard cohort design. Prentice (Biometrika 1986;73:1-11) showed how the case-cohort design can be used to obtain relative risk estimates for comparisons within the cohort being studied. In this paper, the authors consider ways in which the case-cohort design can be used for comparing risk in exposure groups within the cohort to the risk in an external population. The problem reduces to estimating the number of expected cases at each exposure level in the total cohort, when exposure status is available only for members of a subcohort, i.e., a random sample of the total cohort. The authors describe theoretical and empirical properties of several variations of the design and analysis of case-cohort studies. Empirical properties were examined by replicating the selection of the subcohort in a study of second cancer risk after chemotherapy for a first cancer. Use of a case-cohort design in that study would have saved five-sixths of the cost of gathering covariate information at the price of only an 11% loss in efficiency relative to a full cohort study.

The authors present some examples to demonstrate that in certain nondifferential misclassification conditions with polychotomous exposure variables, estimates of odds ratios for categories at intermediate level of risk can be biased away from the null or can change direction. In addition, the authors present two examples to demonstrate that the slope of the dose-response trend for the true distributions can change direction, creating a false inverse trend, even if the misclassification is nondifferential.

Case-control studies have used a variety of measurements to evaluate the relationship of children's exposure to magnetic fields (50 or 60 Hz) with childhood leukaemia and other childhood cancers. In the absence of knowledge about which exposure metrics may be biologically meaningful, studies during the past 10 years have often used time-weighted average (TWA) summaries of home measurements. Recently, other exposure metrics have been suggested, usually based on theoretical considerations or limited laboratory data. In this paper, the rationale and associated preliminary studies undertaken are described as well as feasibility and validity issues governing the choice of the primary magnetic field exposure assessment methods and summary metric used to estimate children's exposure in the National Cancer Institute/Children's Cancer Group (NCI/CCG) case-control study. Also provided are definitions and discussion of the strengths and weaknesses of the various exposure metrics used in exploratory analyses of the NCI/CCG measurement data. Exposure metrics evaluated include measures of central tendency (mean, median, 30th to 70th percentiles), peak exposures (90th and higher percentiles, peak values of the 24 h measurements), and measurements of short-term temporal variability (rate of change). This report describes correlations of the various metrics with the time-weighted average for the 24 h period (TWA-24-h). Most of the metrics were found to be positively and highly correlated with TWA-24-h, but lower correlations of TWA-24-h with peak exposure and with rate of change were observed. To examine further the relation between TWA and alternative metrics, similar exploratory analysis should be considered for existing data sets and for forthcoming measurement investigations of residential magnetic fields and childhood leukaemia.