Cohort and case-control studies

A cohort study assembles a group of patients and follows them over time. An example is the Nurses' Health Study, in which over 20,000 nurses were identified and followed-up annually with tests and surveys for over 25 year (this study is still ongoing). These studies provide very valuable information, but are obviously very expensive and time-consuming.

In most cohort studies, you want to assemble patients without the disease in question, and then follow them until they develop the disease. By comparing the characteristics of patients with and without disease, you can identify risk factors. A study to identify risk factors for breast cancer is diagrammed below:

In a cohort study of prognosis, you would continue to follow patients after they develop the disease in question to see how they do regarding mortality, disease progression, and other important outcomes. Therefore, cohort studies of prognosis don't necessarily begin with a group of disease-free patients. Because they don't, it is important that patients either be at the same stage of the disease at the start of the study, or that you are able to precisely characterize the stage of the disease. A study to determine the prognosis for patients with colon cancer is shown below:

Alternately, you can enroll all patients with colon cancer, being careful to classify them at the beginning of the study:

When it is important to compare the prognosis for patients with disease to those without, a group of patients without the disease in question is sometimes followed in a cohort study. Consider the example of well differentiated prostate cancer described earlier: the best way to know that these patients have the same prognosis as members of the population without prostate cancer is to follow a similar group of patients without prostate cancer for a long period, perhaps 10 or 20 years..

While cohort studies are considered the best way do a study about prognosis, you can also use a case-control design. While cohort studies are prospective (patients are followed forward in time), case-control studies are retrospective (looking back in time). Patients with a disease are identified who have suffered a bad outcome such as death or recurrence, and compared with patients who have the disease but haven't suffered the bad outcome. For example, a researcher might identify a group of breast cancer patients who have died from a cancer registry, and compare them with a similar group of patients with breast cancer who are still living. This is diagrammed below:

Note that in a case-control study there are two major potential biases which don't exist to the same extent in a cohort study. First, you have to pick controls. If the controls are different then the cases (i.e. older, larger, different lifestyle, different habits), that introduces bias. Second, you are looking backwards in time (retrospective design) to determine prognostic factors. While in some cases you may have good records and little bias (for example, descriptions of the surgery performed and the amount of radiation given), other variables may be subject to significant "recall bias". For example, knowing that breast cancer is now thought to be related to fatty food and alcohol could bias womens' recall of their diet.  Another important bias should be obvious from the diagram above:  you will have to rely on the families, friends, and caregivers of women who have died to tell you that patient's diet and habits.