Case Control Studies

What to do with case study data

2x2 table ab cd OR= ad/bc

Analyzing Data

Determine the magnitude of the association, use measurements of association and impact. Rule out the role of chance in the results by doing confidence intervals and hypothesis testing. Rule out the role of confounding by restriction and statistical adjustments for potential confounders.

Odds ratio as an estimate of RR

If the disease is rare in the population (called the rare disease assumption) then the odds ratio calculated by conducting a good case control study will be a good approximation of the "true" relative risk However, if the disease is not rare, this does not mean that the OR is invalid. It is simply not as approximation of the risk.

Design of a case control study

In a case control study the starting point is cases with the disease in a population. The previous exposure of the cases is analyzed and a representative sample of controls (without the disease being studied) selected from the same population. The direction of inquiry is always backwards in time, but the actual data collection can be carried out in either: -a retrospective manner (backwards in time from the time the study started) or -a prospective manner (forward in time from the start of the study)

Overview of Case Control Study and Design

In a case control study, subjects are recruited on the basis of their disease status (cases) and as a comparison group (controls). The prevalence of exposure is measured retrospectively, which is why it is sometimes called a retrospective case control study

Two types of case control studies

Incidence Density Cumulative Incidence, or Prevalence

The odds ratio (OR)

OR= odds of exposure given disease ----------------------------------------- odds of exposure given no disease (a/c) / (b/d) or ad / bc

Odds Ratio for Case Control

to measure relative risk because you don't have incidents rates, because there are no new cases

Requirements of controls

1. They must not have the disease of interest 2. They should be representative of the population that produced the cases. 3. They can be chosen for their similarity to cases with respect to the confounders. 4. They must have the same probability as being exposed as cases.

4 Requirement of Cases (case study)

1. Cases must have the disease. 2. Cases should be representative of all persons with the disease to whom you want to generalize results. 3. When possible, cases should incident (recruited at the time of diagnosis) 4. Cases are best when they are alive, but they don't have to be alive, as long as you can get accurate exposure information from dead people.

Disadvantages of Case Control Studies

1. Cases presenting treatment may not be representative of all cases. 2. Cases may deferentially recall information about exposure levels. 3. Control groups are difficult to assemble well. 4. Cannot calculate incidence (without incidence, no RR).

4 Basic Steps of Case Control Studies

1. Collect a series of cases of a disease who are representative of all cases to whom you want to generalize results. 2. Collect a series of controls as a comparison for the case group. 3. For each group measure exposure information. 4. Compare the odds of disease between the exposed and unexposed individuals.

Properties of the Odds Ratio

1. Regardless of how the data is sampled, the OR will remain the same. 2. It has similar properties to the RR in the terms of range, magnitude, and statistical significance. 3. Under the conditions of a rare disease, the odds ratio will be decent approximation to the "true" relative risk.

4 Advantages of Case Control Studies

1. Relatively cheap and quick to do 2. Exposure should come before disease if using incidents cases. 3. Smaller sample sizes are needed than other studies. 4. Good for studying rare diseases.

Incidence Density Case Control Study

Are those studies where cases are incident cases. In other words, they are recruited over a period of time as they develop.

Cumulative Incidence or Prevalence Case Control Study

Are those where cases are recruited as they already exist in the population. In other words, they are recruited as prevalent cases.

Measures of Impact

Because you cannot calculate incidence from case control data, not all measures of impact are possible. The AR and the PA cannot be calculated. The AR% can be estimated by 1-1/OR, if the disease is rare the PAR% can be estimated by proportion exposed (OR-1) ----------------------------------- 1+ proportion exposed (OR-1)

When to do a case control study

Case control studies are relatively short studies designed to test etiologic hypothesis. -There is some evidence that certain risk factors are important to the disease -The disease is one for which people usually seek care or are otherwise identified by another source. -The disease should have rapid onset with easily identified features.

Selection of Cases

Cases are usually identified as persons seeking care for a particular disease. Incident cases are preferable to prevalent cases, however in the case of an episodic disease the episode does not necessarily have to be the first in a lifetime. The best way to identify cases is to include all incident cases in a population over a certain time period. If this is not possible, cases are identified through treatment.

Matching

Matching is a method to control confounding. it comes in 2 varieties: 1. Frequency matching The relative proportions of subjects with a confounder is the same between cases and controls 2. 1-to-X matching For everyone case there are chosen X controls with the same value of the confounder.

Target Population

Population that you intended to generalize your findings to. (what you learn from sample you can generalize back to the target population)

Selection of controls

There are a number of sources for controls. They include: -samples of the population that produced the cases -patients at the same medical facility as cases who are being treated for unrelated diseases -neighbors, coworkers, schoolmates -family members Depending on the nature of the disease, the proposed risk factors, and the source of cases, not all of these might be appropriate control groups.