354 THE CHINA STUDY
or associate with each other. For example, how does dietary fat relate to
breast cancer rates? Or how does blood cholesterol relate to coronary
heart disease? How does a certain kind of fatty acid in red blood cells
relate to rice consumption? We could also compare blood testosterone
levels or estrogen levels with breast cancer risk. We did thousands of
different comparisons of this type.
In a study of this kind, it is important to note that only the average
values for county populations are being compared. Individuals are not
being compared with individuals (in reality, neither does any other epi-
demiological study design). As ecological studies go, this study, with its
sixy-five counties, was unusually large. Most such studies only have ten
to twenty such population units, at most.
Each of the Sixty-five counties provided 100 adults for the survey.
One-half were male and one-half female, all aged thirty-five to sixty-
four years. The data were collected in the following manner:
- each person volunteered a blood sample and completed a diet and
- one-half of the people provided a urine sample;
- the survey teams went to 30% of the homes to carefully measure
food consumed by the family over a three-day period;
- samples of food representing the typical diets at each survey site
were collected at the local marketplace and were later analyzed for
dietary and nutritional factors.
One of the more important questions during the early planning
stages was how to survey for diet and nutrition information. Estimating
consumption of food and nutrients from memory is a common method,
but this is very imprecise, especially when mixed dishes are consumed.
Can you remember what foods you ate last week, or even yesterday?
Can you remember how much? Another even more crude method of
estimating food intake is to see how much of each food is sold in the
marketplace. These findings can give reasonable estimates of diet trends
over time for whole populations, but they do not account for food waste
or measure individual amounts of consumption.
Although each of these relatively crude methods can be useful for
certain purposes, they still are subject to considerable technical error
and personal bias. And the bigger the technical error, the more difficult
it is to detect significant cause-effect associations.
We wanted to do better than crudely measure which foods and how