During the second half of the 20th century, there was a sharp increase in hysterectomy in the United States. Concurrent with that trend, there was an epidemic of endometrial cancer that has been attributed to widespread use of replacement estrogens among menopausal women. The epidemic of endometrial cancer was not immediately evident, however, in data on endometrial cancer rates compiled from cancer registries. Devise a hypothesis based on considerations of the population at risk for endometrial cancer that can explain why the epidemic went unnoticed.
What is the purpose of randomization in an experiment? How is the same goal achieved in nonexperimental studies?
When cancer incidence rates are calculated for the population covered by a cancer registry, the usual approach is to take the number of new cases of cancer and divide by the person-time contributed by the population covered by the registry. Person-time is calculated as the size of the population from census data multiplied by the time period. This calculation leads to an underestimate of the incidence rate in the population at risk. Explain.
In the calculations of rates for the data in Figure 5–2, the rate in the exposed group declined after taking the induction period into account. If exposure does cause disease, would you expect that the rate in exposed people would increase, decrease, or stay the same after taking into account an appropriate induction period?
If a person already has disease, can that person be selected as a control in a case-control study of that disease?
If a person has already been selected as a control in a case-control study and then later during the study period develops the disease that is being studied, should the person be kept in the study as (1) a case, (2) a control, (3) both, or (4) neither?
In case-cohort sampling, a single control group can be compared with various case groups in a set of case-control comparisons, because the control sampling depends on the identity of the cohorts and has nothing to do with the cases. Analogously, the denominators of risk for a set of several different diseases occurring in the cohort will be the same. Risk-set sampling, in contrast, requires that the investigator identify the risk set for each case; the sample of controls will be different for each disease studied. If the analogy holds, this observation implies that the denominators of incidence rates will differ when calculating the rates for different diseases in the same cohort. Is that true? If not, why not? If so, why should the denominators for the risks not change no matter what disease is studied, whereas the denominators for the rates change from studying one disease to another?
Explain why it would not be possible to study the effect of cigarette smoking on lung cancer in a case-crossover study but why it would be possible to study the effect of cigarette smoking on sudden death from arrhythmia using that design.
Cumulative case-control studies are conducted by sampling the controls from people who remain free of disease after the period of risk for disease (eg, an epidemic period) has ended. With this sampling strategy, demonstrate why the odds ratio will tend to be an overestimate of the risk ratio.
Often, the time at which disease is diagnosed is used as the time of disease onset. Many diseases, such as cancer, rheumatoid arthritis, and schizophrenia, may be present in an undiagnosed form for a considerable time before the diagnosis. Suppose you are conducting a case-control study of a cancer. If it were possible, would it be preferable to exclude people with undetected cancer from the control series?