Maria Blettner

Controlling for continuous confounders in epidemiologic research

Multiple regression models are commonly used to control for confounding in epidemiologic research. Parametric regression models, such as multiple logistic regression, are powerful tools to control for multiple covariates provided that the covariate‐risk associations are correctly specified. Residual confounding may result, however, from inappropriate specification of the confounder‐risk association. In this paper, we illustrate the order of magnitude of residual confounding that may occur with traditional approaches to control for continuous confounders in multiple logistic regression, such as inclusion of a single linear term or categorization of the confounder, under a variety of assumptions on the confounder‐risk association. We show that inclusion of the confounder as a single linear term often provides satisfactory control for confounding even in situations in which the model assumptions are clearly violated. In contrast, categorization of the confounder may often lead to serious residual confounding if the number of categories is small. Alternative strategies to control for confounding, such as polynomial regression or linear spline regression, are a useful supplement to the more traditional approaches.

Mortality among workers employed in the titanium dioxide production industry in Europe

AbstractObjectives: To assess the risk of lung cancer mortality related to occupational exposure to titanium dioxide (TiO2). Methods: A mortality follow-up study of 15,017 workers (14,331 men) employed in 11 factories producing TiO2 in Europe. Exposure to TiO2 dust was reconstructed for each occupational title; exposure estimates were linked with the occupational history. Observed mortality was compared with national rates, and internal comparisons were based on multivariate Cox regression analysis. Results: The cohort contributed 371,067 person-years of observation (3.3% were lost to follow-up and 0.7% emigrated). 2652 cohort members died during the follow-up, yielding standardized mortality ratios (SMRs) of 0.87 (95% confidence interval [CI] 0.83–0.90) among men and 0.58 (95% CI 0.40–0.82) among women. Among men, the SMR of lung cancer was significantly increased (1.23, 95% CI 1.10–1.38); however, mortality from lung cancer did not increase with duration of employment or estimated cumulative exposure to TiO2 dust. Data on smoking were available for over one third of cohort members. In three countries, the prevalence of smokers was higher among cohort members compared to the national populations. Conclusions: The results of the study do not suggest a carcinogenic effect of TiO2 dust on the human lung.

Medical risk factors and the development of brain tumors

Several diseases and medical treatments are discussed as risk factors for the development of brain tumors. A population‐based case‐control study in the Rhein‐Neckar‐Odenwald area (containing 1.3 million inhabitants) of Germany was established to investigate this question. A total of 226 patients (cases) with primary brain tumors (International Classification of Diseases, ninth edition, classes 191, 191.1, and 192.0) and 418 control subjects (controls) were interviewed using a standardized questionnaire over a period of 2 years. No association was seen for head injuries, hereditary diseases, family history, and radiographic examination of the head and teeth. However, more cases than controls had had meningitis (relative risk [RR], 2.7; 95% confidence interval [CI], 0.9 to 8.6) or epilepsy (RR, 2.6; 95% CI, 0.6 to 11.7). The RR was decreased for those who had allergic diseases (RR, 0.7; 95% CI, 0.5 to 1.0), diabetes (RR, 0.7; 95% CI, 0.3 to 1.8), and infections and colds (RR, 0.3; 95% CI, 0.1 to 0.8).

Sun exposure and number of nevi in 5- to 6-year-old European children

The occurrence and number of melanocytic nevi are among the most important known risk factors for the development of malignant melanoma. Studying the causes of nevi should lead to successful strategies in the prevention of melanoma. Among 11,478 white German children of preschool age the association between benign melanocytic nevi and a number of risk factors for skin cancer was examined. We found that subjects with a reported history of increased sun exposure, for example, painful sunburns, and an increased number of holidays in foreign countries with a sunny climate had significantly higher nevus counts than individuals without these characteristics. Our results provide further evidence that nevus counts may not only be part of a genetic predisposition but also a result of increased exposure to ultraviolet radiation. Together with the fact that a high nevus count is the most relevant risk factor for malignant melanoma, the results strongly indicate a connection between UV-radiation and the development of melanocytic skin cancer. In conclusion, strategies to reduce the incidence of melanoma should begin with young children.

Mortality from cancer and other causes among male airline cockpit crew in Europe

Airline pilots and flight engineers are exposed to ionizing radiation of cosmic origin and other occupational and life‐style factors that may influence their health status and mortality. In a cohort study in 9 European countries we studied the mortality of this occupational group. Cockpit crew cohorts were identified and followed‐up in Denmark, Finland, Germany, Great Britain, Greece, Iceland, Italy, Norway and Sweden, including a total of 28,000 persons. Observed and expected deaths for the period 1960–97 were compared based on national mortality rates. The influence of period and duration of employment was analyzed in stratified and Poisson regression analyses. The study comprised 547,564 person‐years at risk, and 2,244 deaths were recorded in male cockpit crew (standardized mortality ratio [SMR] = 0.64, 95% confidence interval [CI] = 0.61–0.67). Overall cancer mortality was decreased (SMR = 0.68; 95% CI = 0.63–0.74). We found an increased mortality from malignant melanoma (SMR = 1.78, 95% CI = 1.15–2.67) and a reduced mortality from lung cancer (SMR = 0.53, 95% CI = 0.44–0.62). No consistent association between employment period or duration and cancer mortality was observed. A low cardiovascular mortality and an increased mortality caused by aviation accidents were noted. Our study shows that cockpit crew have a low overall mortality. The results are consistent with previous reports of an increased risk of malignant melanoma in airline pilots. Occupational risk factors apart from aircraft accidents seem to be of limited influence with regard to the mortality of cockpit crew in Europe.

Effects of disease-dependent changes of exposure in cross-sectional studies

To determine whether the risk of having an infant with anencephaly or spina bifida is greater among obese women than among average-weight women, we compared 307 Atlanta-area women who gave birth to a liveborn or stillborn infant with anencephaly or spina bifida (case group) with 2,755 Atlanta-area women who gave birth to an infant without birth defects (control group). The infants of control women were randomly selected from birth certificates and frequency-matched to the case group by race, birth hospital, and birth period from 1968 through 1980. After adjusting for maternal age, education, smoking status, alcohol use, chronic illness, and vitamin use, we found that, compared with average-weight women, obese women (pregravid body mass index greater than 29) had almost twice the risk of having an infant with spina bifida or anencephaly (odds ratio = 1.9; 95% confidence limits = 1.1, 3.4). A womans risk increased with her body mass index: adjusted odds ratios ranged from 0.6 (95% confidence limits = 0.3, 2.1) for very underweight women to 1.9 for obese women.&NA; The possibility of disease‐related changes in exposure is a major limitation of many epidemiologic studies. This limitation particularly applies to cross‐sectional studies, in which exposure and presence of disease are measured at the same point of time. In some cross‐sectional studies, attempts are made to collect information not only on current exposure, but also on exposure at some period in the past. Various methods have been proposed to assess the relation of past and current exposure with disease in such situations. Systematic methodologic work is lacking, however, on the performance of these methods in the presence of disease‐related changes of exposure. In this paper, we provide a model to assess the effects of such changes, and we illustrate the effects with a variety of numerical examples. We show that all of the proposed methods can yield seriously biased measures of current exposure effects in the presence of disease‐related changes in exposure, and we illustrate the direction and order of magnitude of the biases. Our results imply that there is usually no alternative to longitudinal approaches to measure exposure in relation to disease onset if disease‐related changes of exposure are of concern.

Measurement error correction using validation data: A review of methods and their applicability in case-control studies

Measurement error is a serious problem in the analysis of epidemiological data. In the past 20 years, a large number of methods for the correction of measurement error have been developed. While at the beginning mostly methods for cohort studies were considered, recently more attention has been paid to case-control studies. Although a variety of methods have been proposed, they are very rarely used in practice. To stimulate their use and further development, this article provides a comprehensive overview on methods developed for multivariable regression analysis of epidemiologic studies with validation data sets. The methods are systematically classified with respect to the underlying theory. An assessment of prerequisites, assumptions and performance of the available methods is given. Particular attention is paid to applicability to case-control studies and need for further research and development is pointed out.

The performance of methods for correcting measurement error in case-control studies

Background. It is generally agreed that adjustment for measurement error (when feasible) can substantially increase the validity of epidemiologic analyses. Although a broad variety of methods for measurement error correction has been developed, application in practice is rare. One reason may be that little is known about the robustness of these methods against violations of their restrictive assumptions. Methods. We carried out a simulation study to assess the performance of two error correction methods (a regression calibration method and a semiparametric approach) as compared with standard analyses without measurement error correction in case-control studies with internal validation data. Performance was assessed over a wide range of model parameters including varying degrees of violations of assumptions. Results. In nearly all the settings assessed, the semiparametric estimate performed better than all alternatives under investigation. The regression calibration method is sensitive to violations of the assumptions of nondifferential error and small error variance. Conclusions. The semiparametric method is a very robust method to correct for measurement error in case-control studies, but lack of functional software hinders widespread use. If the assumptions for the regression calibration method are fulfilled, application of this method, originally developed for cohort studies, in case-control studies may be a useful alternative that is easy to implement.

Identifying Influential Families Using Regression Diagnostics For Generalized Estimating Equations

The Generalized Estimating Equations (GEE) is an approach to analyze correlated data. It is applied here to data from an epidemiological study of oesophageal cancer in a high incidence area in China to investigate familial aggregation. Regression diagnostics for mean structures and association structures are used to identify families that influence estimates of these structures. It is shown that most of the families influencing the mean structure have a low age of disease onset in common. Most families identified by regression diagnostics for the association structure influence the parent correlation. It is concluded that regression diagnostic techniques can be used to identify clusters influencing mean and association structures of the models. Genet. Epidemiol. 15:341–353,1998.

Steigende Inzidenz und Mortalität der Non-Hodgkin-Lymphome Eine epidemiologische Übersicht über neuere Studien zu Risikofaktoren des Non-Hodgkin-Lymphoms

BACKGROUND Non-Hodgkins lymphoma (NHL) are among the small number of malignant tumors with markedly increasing incidence and mortality rates in the recent past. This trend is particularly obvious in industrialized countries. The causes of the observed increase remain unclear. METHODS We conducted a Medline search to identify case control and cohort studies on medical, biological and selected environmental risk factors of non-Hodgkins lymphoma published between 1992 and 1998. Methodological aspects and results of identified studies are presented in tabular form. We furthermore discuss the role of various risk factors for the observed trends in non-Hodgkins lymphoma. RESULTS We identified a total of 64 studies. Iatrogenic immunosuppression as well as numerous diseases associated with an impaired immune system have clearly been recognized as etiological factors for non-Hodgkins lymphoma. However, they can explain only a small percentage of the rate increases of non-Hodgkins lymphoma. Similarly, the HIV/AIDS epidemic in Germany is responsible only for a small proportion of the rising figures of non-Hodgkins lymphoma; other viral agents are currently being associated with a few distinct subtypes of non-Hodgkins lymphoma. Smoking and nutritional factors are weakly or not at all associated with non-Hodgkins lymphoma and can not account for the disease trends. CONCLUSION The observed secular rise in incidence and mortality of non-Hodgkins lymphoma can only partly be explained. New epidemiologic studies should focus on risk factors associated with the function of the immune system and on possible interactions between different etiological factors.ZusammenfassungHintergrund: Die Non-Hodgkin-Lymphome gehören zu den wenigen malignen Tumoren, für die in den letzten Jahrzehnten ein deutlicher Aufwärtstrend in Inzidenz und Mortalität zu verzeichnen ist. Dieser Trend ist insbesondere in Industrieländern zu beobachten. Die Gründe für diesen Anstieg sind noch unklar. Material und Methoden: Basierend auf einer Medline-Recherche identifizierten wir Fallkontroll- und Kohortenstudien zu medizinisch-biologischen und ausgewählten Umweltrisikofaktoren für Non-Hodgkin-Lymphome, die in den Jahren 1992 bis 1998 publiziert wurden. Die wichtigsten methodischen Aspekte und Ergebnisse dieser Arbeiten werden tabellarisch dargestellt, und es wird der Beitrag verschiedener Risikofaktoren zu den ansteigenden Inzidenzraten des Non-Hodgkin-Lymphoms diskutiert. Ergebnisse: Aus den insgesamt 64 identifizierten Studien ergab sich, dass die medikamentöse Immunsuppression sowie verschiedene Erkrankungen, die mit einer gestörten Funktion des Immunsystems einhergehen, als gesicherte Risikofaktoren für Non-Hodgkin-Lymphome gelten können. Sie erklären die ansteigenden Raten jedoch nur zu einem sehr geringen Teil. Der Verlauf der HIV/AIDS-Epidemie in Deutschland trägt ebenfalls wenig zu den Trends bei Non-Hodgkin-Lymphomen bei, andere virale Erreger werden derzeit nur mit bestimmten Non-Hodgkin-Lymphom-Subtypen assoziiert. Rauchen und Ernährungskomponenten sind allenfalls schwache Risikofaktoren mit geringem Einfluss auf die Non-Hodgkin-Lymphom-Morbidität. Schlussfolgerung: Nur ein Teil des beobachteten Anstiegs der Inzidenz und Mortalität der Non-Hodgkin-Lymphome ist derzeit erklärbar. Neue epidemiologische Studien sollten Faktoren mit Einfluss auf das Immunsystem und mögliche Interaktionen zwischen verschiedenen ätiologischen Faktoren untersuchen.AbstractBackground: Non-Hodgkins lymphoma (NHL) are among the smaller number of malignant tumors with markedly increasing incidence and mortality rates in the recent past. This trend is particularly obvious in industrialized countries. The causes of the observed increase remain unclear. Methods: We conducted a Medline search to identify case control and cohort studies on medical, biological and selected environmental risk factors of non-Hodgkins lymphoma published between 1992 and 1998. Methodologial aspects and results of identified studies are presented in tabular form. We furthermore discuss the role of various risk factors for the observed trends in non-Hodgkins lymphoma. Results: We identified a total of 64 studies. Iatrogenic immunosuppression as well as numerous diseases associated with an impaired immune system have clearly been recognized as etiological factors for non-Hodgkins lymphoma. However, they can explain only a small percentage of the rate increases of non-Hodgkins lymphoma. Similarly, the HIV/AIDS epidemic in Germany is responsible only for a small proportion of the rising figures of non-Hodgkins lymphoma; other viral agents are currently being associated with a few distinct subtypes of non-Hodgkins lymphoma. Smoking and nutritional factors are weakly or not at all associated with non-Hodgkins lymphoma and can not account for the disease trends. Conclusion: The observed secular rise in incidence and mortality of non-Hodgkins lymphoma can only partly be explained. New epidemiologic studies should focus on risk factors associated with the function of the immune system and on possible interactions between different etiological factors.