Confounded epidemiological data

Questionnaires are often used to record epidemiological rather than planned experimental data. To appreciate the difference between these, think about a situation where we want to compare some sort of outcome between smokers and non-smokers. 

The planned experimental approach would be to take a group of volunteers, randomly allocate them to smoke or not smoke, leave them to indulge/not indulge 

When we create two groups by dividing subjects on the basis of one demographic characteristic, but find the resulting groups also tend to differ in other ways, there is confounding. 

Our analysis showed that there was an effect of age [see Table 19.1(a)]. Only 26.2 per cent of the older respondents were in favour, vs 74.9 per cent of the youngsters. The results were subjected to a chi-square test and found to be significant. A more conservative view among senior citizens was no great surprise. 

What did initially come as a bit of a surprise [Table 19.1(b)] was an additional divide based on health status, with the sick/disabled being less favourable to the proposal than the healthy (P 0.001). However it seemed unlikely that respondents  

---

When epidemiological data are available, the issues to be dealt with include selection of the appropriate study and control populations, evaluation of exposure levels and tissue doses, determination of the reliability of cancer ascertainment, allowance for the latent period and age distribution of cancers, control of biases and confounding factors, fitting of models to the data to characterize the dose-incidence relationship, and derivation of risk estimates with their associated ranges of uncertainty. 

Knowledge of tumor induction by sex steroids is largely based on interpretation of epidemiological data, with careful exclusion of possible confounding elements. Hepatic tumors have given rise to most concerns, but some evidence also indicates an increased incidence of various other malignancies, including carcinomas of the breast, endometrium, and prostate (71). 

Describe the problem of confounding in epidemiological data and explain how we reduce the risk of false interpretations... 

Unrecognized confounding is always a potential problem with epidemiological data... 

Unfortunately, whenever questionnaires are used to gather epidemiological data (which is much of the time), none of the conclusions ever have the reliability we achieve with planned experiments. In the example we looked at, we had collected data on both of the confounded factors and we could identify and eliminate the problem. However, if we had only gathered data on health status and not recorded respondents ages, we could well have mistakenly concluded that people s health in some way influenced their opinions. 

With epidemiological data, always be cautious about ascribing cause and effect relationships. The factor that appears to be having an influence may in fact be confounded with some other unrecognized factor which is the true cause of the particular outcome. 

In a report the manufacturers have discussed the epidemiological data linking the use of fenoterol to asthma mortahty in New Zealand. They pointed out that asthma mortahty started to fall in 1979 while fenoterol sales were stiU increasing. Sales of fenoterol in Austria, Belgium, and Germany were similar to those in New Zealand at the peak of the New Zealand asthma death epidemic, but asthma mortality in the other countries did not rise. The confounding problem that fenoterol was preferentially prescribed for the more severe cases of asthma was... 

Limited epidemiology data exist for exposure to ethyl acrylate. Mortality from cancer of the colon and rectum was elevated in workers from plants manufacturing and polymerizing ethyl acrylate however, the findings were confounded by coexposure to other chemicals. Currently, there is inadequate evidence to link human exposures to ethyl acrylate with cancer. 

The epidemiological data available suggest some protection from colon cancer in populations subsisting on diets high in fiber, but other confounding variables such as affluence and life style must be considered. The concentration, rather than the amount of bile acids... 

The frustrations expressed by Ward and Dye stem from at least two major difficulties in the interpretation of these epidemiological data. The first is the issue of confounding variables of alcohol and age. Across all studies, in approximately 80% of the cases where THC was detected, alcohol was also found. Also, marijuana is primarily used by young males, who are over-represented in fatal crashes and are associated with socially high risk-taking behaviors,... 

According to EPA (IRIS 1999), the available human epidemiological studies lack quantitative exposure data for lead and for possible confounding exposures (e.g., arsenic, smoking). Cancer excesses in the lung and stomach of lead-exposed workers that are reported are relatively small, dose-response relationships are not demonstrated neither is there consistency in the site of cancers reported. EPA (IRIS 1999) concluded that the human data are inadequate to refute or demonstrate the potential carcinogenicity of lead exposure. 

Limited epidemiological information exists for carcinogenicity in humans following inhalation exposure to kerosene (vapor) (Chan et al. 1979) and other fuel oils such as diesel fuel (vapor) (Partanen et al. 1991). These studies either test kerosene exposure by use of kerosene stoves, and so are limited for the same reasons as the respiratory studies described above, or measure fuel oil exposures according to occupation. In the latter case, confounding from exposure to other chemicals, such as gasoline, exists. Both studies are limited since the duration and level of fuel oil exposure were not identified. Other available data are also reported to be inadequate to assess the carcinogenic potential of fuel oils (lARC 1989 Lam and Du 1988). 

---