Search

Exposure to magnetic fields (MFs) is hypothesized to increase the risk of breast cancer by reducing production of melatonin by the pineal gland. A nested case-cohort study was conducted to investigate the association between occupational exposure to MFs and the risk of breast cancer within a cohort of 267,400 female textile workers in Shanghai, China. The study included 1,687 incident breast cancer cases diagnosed from 1989 to 2000 and 4,702 noncases selected from the cohort. Subjects' complete work histories were linked to a job-exposure matrix developed specifically for the present study to estimate cumulative MF exposure. Hazard ratios and 95% confidence intervals were calculated using Cox proportional hazards modeling that was adapted for the case-cohort design. Hazard ratios were estimated in relation to cumulative exposure during a woman's entire working years. No association was observed between cumulative exposure to MFs and overall risk of breast cancer. The hazard ratio for the highest compared with the lowest quartile of cumulative exposure was 1.03 (95% confidence interval: 0.87, 1.21). Similar null findings were observed when exposures were lagged and stratified by age at breast cancer diagnosis. The findings do not support the hypothesis that MF exposure increases the risk of breast cancer.

Associations between alcohol drinking and cardiovascular disease mortality could be confounded by diet if alcohol drinking and diet are related. Depending on the alcohol measure, alcohol-diet relations may or may not be observed. The authors examined associations between alcohol and diet quality (Healthy Eating Index (HEI) scores) using cross-sectional, nationally representative data from the 1999-2000 National Health and Nutrition Examination Survey. Weighted analyses included 3,729 participants agedor =20 years. In adjusted analyses among current alcohol drinkers, as quantity increased from 1 toor =3 drinks/drinking day, the mean HEI score decreased from 65.3 (95% confidence interval (CI): 63.4, 67.1) to 61.9 (95% CI: 60.5, 63.2). As frequency increased from the lowest quartile to the highest, the mean HEI score increased from 60.9 (95% CI: 58.7, 63.2) to 64.9 (95% CI: 63.4, 66.4). As average volume ((quantity x frequency)/365.25) increased from1 drink/day toor =3 drinks/day, the mean HEI score increased from 62.9 (95% CI: 61.2, 64.5) to 65.2 (95% CI: 62.7, 67.8). In stratified analyses, the lowest HEI score, 58.5 (95% CI: 55.5, 61.5), occurred among drinkers who consumed the highest quantity at the lowest frequency. Average volume of alcohol consumed is driven by and masks the contributions of its components. These results suggest the importance of measuring drinking patterns (quantity, frequency, and stratified combinations) in epidemiologic alcohol-diet studies.

Alcohol could contribute to obesity. The authors examined the relation between drinking patterns and body mass index (BMI) (weight (kg)/height (m)(2)) by pooling cross-sectional data from the 1997-2001 National Health Interview Surveys. Weighted analyses included 45,896 adult never smokers who were current alcohol drinkers. Height and weight were self-reported. In adjusted analyses, alcohol quantity and frequency had opposite associations with BMI. As quantity increased from 1 drink/drinking day to > or =4 drinks/drinking day, BMI significantly increased; in men, it increased from 26.5 (95% confidence interval (CI): 26.3, 26.6) to 27.5 (95% CI: 27.4, 27.7), and in women, it increased from 25.1 (95% CI: 25.0, 25.2) to 25.9 (95% CI: 25.5, 26.3). As frequency increased from low quintiles of drinking days/year to high quintiles, BMI significantly decreased; in men, it decreased from 27.4 (95% CI: 27.2, 27.6) to 26.3 (95% CI: 26.2, 26.5), and in women, it decreased from 26.2 (95% CI: 26.0, 26.5) to 24.3 (95% CI: 24.2, 24.5). In stratified analyses of frequency trends within quantity categories, BMI declines were more pronounced in women than in men, but all linear trends were inverse and significant (p trend < 0.001). In all respondents combined, persons who consumed the smallest quantity the most frequently were leanest, and those who consumed the greatest quantity the least frequently were heaviest. Alcohol may contribute to excess body weight among certain drinkers.

While the inverse association between physical activity and coronary heart disease risk is well documented, questions remain regarding the intensity of exercise, the potential for confounding by other risk factors for coronary heart disease, and the role of blood lipids and apolipoproteins. The authors examined these issues in the Boston Area Health Study, a case-control study of 340 patients (266 men, 74 women) who survived a first myocardial infarction between January 1, 1982, and December 31, 1983, and 340 controls matched on sex, age, and residence. The relative risk of myocardial infarction for those in the highest quartile of physical activity, compared with the lowest, was 0.50 (95 percent confidence interval (Cl) 0.31-0.80) for men and 1.00 (95 percent Cl 0.41-2.43) for women. When subjects were categorized by level of energy expenditure on moderate to vigorous sports alone, men in the most active category had 0.39 (95 percent Cl 0.23-0.69) times the risk of those in the least active category, and women, 0.43 (95 percent Cl 0.15-1.26) times the risk. Adjustment for body mass index, smoking, alcohol intake, diet, personal and family medical history, and personality type did not substantially change results nor did further adjustment for blood lipids. This was not surprising as total energy expenditure was uncorrelated with blood lipids or apolipoproteins. Moderate to vigorous sporting activity, however, appeared to be directly related to high density lipoprotein (HDL) cholesterol (p = 0.06), especially the HDL 2 subfraction (p = 0.10). In these data, findings suggest that physical activity is inversely related to myocardial infarction risk, independently of other risk factors for coronary heart disease.

Prospective associations between quantity and frequency of alcohol consumption and cancer-specific mortality were studied using a nationally representative sample with pooled data from the 1988, 1990, 1991, and 1997-2004 administrations of the National Health Interview Survey (n = 323,354). By 2006, 8,362 participants had died of cancer. Cox proportional hazards regression was used to estimate relative risks. Among current alcohol drinkers, for all-site cancer mortality, higher-quantity drinking (≥ 3 drinks on drinking days vs. 1 drink on drinking days) was associated with increased risk among men (relative risk (RR) = 1.24, 95% confidence interval (CI): 1.09, 1.41; P for linear trend = 0.001); higher-frequency drinking (≥ 3 days/week vs.

Power and minimum detectable risk are calculated for seven one-sided tests of standardized mortality ratios with Poisson-distributed events. Each test contrasts the number of observed deaths (D) with the number expected (E). Three tests use exact Poisson probabilities: 1) the exact test, which computes a p value as the probability of equaling or exceeding the number of observed events; 2) the optimal randomized exact test which, although not used in practice, serves as a standard for the other statistics; and 3) the exact "mid-p" procedure, which counts only one-half the probability of the observed event. The remaining four tests use normal approximations to the Poisson ("Z statistics"): 4) Z = magnitude of D-E/square root of E; 5) the Z statistic corrected for continuity, Z = (magnitude of D-E)-0.5)/square root of E; 6) a statistic based on a square root transformation, Z = 2(square root of D-square root of E); and 7) a statistic created by Byar, which, when D is greater than E, is Z = square root of 9D[1-1/(9D)-3 square root of D/E]. Power differences among these procedures with one-sided alpha of 0.05, 0.025, and 0.01 are small as long as four or more events are expected. If fewer than four events are expected, the uncorrected Z has unacceptably high type I error. Simple approximations to the power and detectable risk of these tests are evaluated and prove satisfactory. Differences in minimum detectable risk, actual and approximated, are slight for E of 2.0 or more.

In this cohort study, the authors studied the effect of blood malaria parasite density on fever incidence in children in an endemic area with 9 days' follow-up of 1- to 12-year-old children during two time periods: the end of the dry season (May 1993: n = 783) and the end of the rainy season (October 1993: n = 841) in Bougoula, West Africa (region of Sikasso, Mali). The cumulative incidence of fever (temperature38.0 degrees C) was 2.0% in the dry season and 8.2% in the rainy season (p0.0001). In the rainy season, the risk of fever was increased in children of ages 1-3 years (relative risk (RR) = 2.5, 95% confidence interval (CI) 1.6-4.1); in those with an initial parasitemia15,000/microliter (RR = 2.7, 95% CI 1.4-5.4); in children with an enlarged spleen (RR = 2.0, 95% CI 1.2-3.3); or in those with anemia (hematocrit30%: RR = 1.8, 95% CI 1.1-2.9). In the dry season, anemia was the only predictor of fever incidence. In the rainy season, the best predictors of fever were, in order, age (4 years), enlarged spleen, and high parasite density. Even in the higher risk groups, the cumulative incidence was20%. The authors conclude that most children with high parasite density do not develop fever subsequently. The association between parasite density and fever varies according to age and season. Since even high levels of parasite density do not reliably predict fever incidence, parasite density should be considered as just one of a group of indicators that increase the probability of a fever of malarial origin.In a cohort study, the effect of blood malaria parasite density on fever incidence in children was studied in an endemic area with 9 days' follow-up of children aged 1-12 years during two time periods: the end of the dry season (May 1993: n = 783) and the end of the rainy season (October 1993: n = 841) in Bougoula, West Africa (region of Sikasso, Mali). The number of registered children was 928 in the dry season and 998 in the rainy season. Complete follow-up and information were available for 835 children in the dry season and for 964 children in the rainy season. The 9-day cumulative fever incidence (body temperature above 38.0 degrees Celsius) increased from 2.0% in the dry season to 8.2% in the rainy season (p 0.0001). In the rainy season, the risk of fever increased in children aged 1-3 years (relative risk [RR] = 2.5; 95% confidence interval [CI], 1.6-4.1); in those with an initial parasitemia greater than 15,000/mcl (RR = 2.7; 95% CI, 1.4-5.4); in those with an enlarged spleen (RR = 2.0; 95% CI, 1.2-3.3); or in those with anemia (hematocrit 30%: RR = 1.8; 95% CI, 1.1-2.9). In the dry season, anemia (hematocrit 30%) was the only predictor of fever incidence with a cumulative incidence of 10.0%. In nonanemic children, a parasite count of 2000/mcl was the next best predictor. In the rainy season, the best predictors of fever were age (4 years), enlarged spleen, and high parasite density (1/mcl). Even in the higher risk groups, the cumulative incidence was 20%. Most children with high parasite density do not develop fever subsequently. The association between parasite density and fever varies according to age and season. Since even high levels of parasite density do not reliably predict fever incidence, parasite density should be considered not so much a direct marker of an ongoing attack but as just one indicator of the likelihood of a current or imminent attack or even one just passed.

While multiplicative (log-linear and logistic) models have a firmly established place in epidemiologic methodology, additive and other more general model structures are needed also. The authors propose a parametric family of relative risk functions ranging from subadditive to supramultiplicative that is generated by varying the exponent in a power transform for the log relative risk. The choice of model is facilitated by graphic analysis of goodness-of-fit statistics computed for various values of the exponent. Intermediate quantities available as by-products of the fit are useful for checking the influence of particular observations on the estimated regression coefficients. Three examples illustrate the applications of these methods to random, stratified, and matched samples of cases and controls. Computer software is available for each of these situations. Even though different relative risk models may have markedly different implications for the multifactorial nature of the disease process, it may be difficult to distinguish between them unless the data are quite extensive.

A linear logistic model used to estimate multiple risk functions in both cohort and case-control studies is adapted for sampling plans wherein each case is matched with R controls. The resulting methodology substantially liberalizes current practice by permitting simultaneous analysis of multiple discrete and continuous risk factors. Interactions among risk factors, and between risk factors and matching variables, may be explored. Data from two studies of oesophageal cancer, one conducted among Singapore Chinese and the other on the Caspian littoral of Iran, illustrate the methods.

Nine retrospective mortality studies of workers exposed to vinyl chloride were reviewed to determine whether differences in their hypothesis testing results might be due to differences in statistical power. Where possible, the power of each study was calculated for cancer of the lung, brain and liver. When power was taken into consideration, the results for liver and brain cancer were found to be consistent with an etiologic role for vinyl chloride. For lung cancer, the data were not consistent with an etiologic role in that two studies with very high power yielded negative results.