Ecological versus case-control studies for testing a linear-no threshold dose-response relationship

On ecological studies: a short communication

The typical objective of research is to try and identify cause-and-effect relationships. As with any research design, there are strengths and weaknesses involved in trying to achieve this objective. Some study designs are stronger than others in attempting to establish cause-and-effect associations. The task of establishing cause-and-effect relationships is challenging (Hill, 1965) and any study that does not include experimentation, that is by manipulating a variable's exposure, is inhibited from drawing causal inferences (Heath, 1995). Similarly, statistical significance is also based on probability rather than certainty. This article focuses on a particular research design, namely, the ecological study, and attempts to serve as a reminder that the design has its place in the realm of various research designs.

Ecologic versus individual-level sources of bias in ecologic estimates of contextual health effects

A number of authors have attempted to defend ecologic (aggregate) studies by claiming that the goal of those studies is estimation of ecologic (contextual or group-level) effects rather than individual-level effects. Critics of these attempts point out that ecologic effect estimates are inevitably used as estimates of individual effects, despite disclaimers. A more subtle problem is that ecologic variation in the distribution of individual effects can bias ecologic estimates of contextual effects. The conditions leading to this bias are plausible and perhaps even common in studies of ecosocial factors and health outcomes because social context is not randomized across typical analysis units (administrative regions). By definition, ecologic data contain only marginal observations on the joint distribution of individually defined confounders and outcomes, and so identify neither contextual nor individual-level effects. While ecologic studies can still be useful given appropriate caveats, their problems are better addressed by multilevel study designs, which obtain and use individual as well as group-level data. Nonetheless, such studies often share certain special problems with ecologic studies, including problems due to inappropriate aggregation and problems due to temporal changes in covariate distributions.

Introduction to the post-Human Genome Project era, a target for interactions between polygenic and/or multiphenotypical components in cancer control in South America

Epidemiological studies have suggested that the propensity to develop malignancy involves a complex mix of genetic and environmental determinants, however both older and innovative techniques display unresolved fundamental questions regarding etiology. Current barriers to achieving the potential benefit from this understanding are: 1) incomplete background on the various environmental and genetic factors involved in the carcinogenesis mechanism; 2) difficulties in accurately differentiating specific molecular subtypes and measuring the effective cellular exposure dose; and 3) difficulties in determining the multifactorial interaction between genetic and environmental factors. To extrapolate Human Genome Project research findings to the Post-Human Genome Project era, South America provides a large population and large-pedigree families, thus including genetically heterogeneous and less heterogeneous groups. An initial strategy might be to trace high risk populations and the respective exposures to which they are susceptible, such as: 1) migration, identifying rural migrant populations; 2) inherent susceptibility, studying "long term homogeneous populations" or large families living in similar rural environments; and 3) dissection of gene-environmental interaction

The ecological effects of individual exposures and nonlinear disease dynamics in populations

To describe causally predictive relationships, model parameters and the data used to estimate them must correspond to the social context of causal actions. Causes may act directly upon the individual, during a contact between individuals, or upon a group dynamic. Assuming that outcomes in different individuals are independent puts the causal action directly upon individuals. Analyses making this assumption are thus inappropriate for infectious diseases, for which risk factors alter the outcome of contacts between individuals. Transmission during contact generates nonlinear infection dynamics. These dynamics can so attenuate exposure-infection relationships at the individual level that even risk factors causing the vast majority of infections can be missed by individual-level analyses. On the other hand, these dynamics amplify causal associations between exposure and infection at the ecological level. The amplification and attenuation derive from chains of transmission initiated by exposed individuals but involving unexposed individuals. A study of household exposure to the only vector of dengue in Mexico illustrates the phenomenon. An individual-level analysis demonstrated almost no association between exposure and infection. Ecological analysis, in contrast, demonstrated a strong association. Transmission models that are devoid of any sources of the ecological fallacy are used to illustrate how nonlinear dynamics generate such results.

A cross-national investigation of diet and bladder cancer

The existence of a large unexplained portion of attributable risk, and the marked variation in bladder cancer rates globally, have stimulated an interest in the role of nutrition in cancer of the urinary bladder. For this cross-national comparison study, we had complete data available for 50 countries. Using stepwise regression followed by general linear modelling, age-truncated (45-74 years), world-standardised, sex-specific bladder cancer mortality rates were regressed on an array of nutritional and socioeconomic independent variables in an effort to identify important predictors of bladder cancer mortality. Separate principal components analyses were used to summarise the nutritional and the socioeconomic (SES) variables. In the stepwise analyses, using food scores expressed in kcal/day per capita (as opposed to the nutritional components), total fat consistently entered the model first, and explained the greatest share of variability (R2) for both males and females. General linear models were fitted that included total fat, tobacco, alcohol, the three SES components (comprising seven socioeconomic predictors) and two food categories found significant in stepwise modelling, roots/tubers and vegetable oil. The R2 values were 0.84 for male rates and 0.77 for female rates, meaning that these study factors account for 84% of bladder cancer mortality in men and 77% in women. Substitution of the nutritional components for the foods resulted in general linear models with slightly higher R2 values (0.85 for males, 0.77 for females), but with attenuated fat effects. Results are discussed in light of biological plausibility.

Consumption of meat and fruit in relation to oral and esophageal cancer: a cross-national study

Using data from 59 countries, we conducted an international comparison study to identify nutritional predictors of age-adjusted oral and esophageal cancer mortality rates. Statistical models accounted for per capita tobacco disappearance data, alcohol consumption, and various measures of socioeconomic status. For oral cancer, stepwise regression results showed protective effects for milk/dairy products (B = -0.030, p < 0.0001) and cabbage consumption (B = -0.391, p = 0.01) and increased risk from vegetable oil (B = 0.072, p = 0.04) and excess animal fat calories (B = 0.344, p < 0.0001) as well as marginally increased risk from cereals (B = 0.008, p = 0.08). Results were obtained after accounting for all background factors, including sex (model R2 = 0.52). For esophageal cancer, stepwise results indicated protective effects for fruit (B = -0.046, p = 0.0006) and total caloric intake (B = -0.013, p < 0.0001) and increased risk from vegetable oil (B = 0.061, p = 0.04) and meat (B = 0.031, p < 0.0001) consumption (model R2 = 0.55). When analyzed separately by sex, results were similar, indicating that the risk factors are probably the same in both sexes, even though women consistently have fewer deaths, on average, from these cancers. On the basis of results from stepwise regression models, we also fitted general linear models for mortality rates of each cancer site, and results were similar in terms of magnitude and direction of effects. Although the evidence provided by this type of analysis using data aggregated by country is limited in terms of control for potential confounding effects and modeling of possible effect modification, an effect of high meat, animal product, or vegetable oil and low fruit and cabbage consumption is consistent with the known biology of these tumors.