About Case-Control Studies

One step up from cross sectional studies, in terms of value of evidence in hypothesis testing, would be case control studies.  These are studies where the comparison groups (case and control) are chosen for the outcome under study.  Choosing where to find the subjects for both the group with the condition under study and that without the condition is critical.  Sometimes the disease status is established with the use of laboratory tests or determination of a panel of experts.  Many case control studies will take a sample from hospitalized patients, choosing those with the study condition and the non-case group from hospitalized subjects.  These studies are retrospective with a backward direction to ascertain a related attribute or past event that is correlated with the disease under study.  The research may have more than one comparison group.

The purpose of these studies is to discover a risk factor that might be correlated with the condition under study.  One case control study will not prove that something is a risk factor.  Care must be taken to identify any confounding factors that might be in the causal pathway.  For example, there is a correlation between the use of alcohol and the use of tobacco.  Depending on whether or not smoking is being investigated as a risk factor, it will be necessary to understand the use of alcohol in the two groups to determine if it is a risk factor or a confounding factor.  This can be very difficult and statistics alone may not accomplish this.  The analysis must involve critical thinking skills about the known or suspected risk factors and confounding factors.

Sometimes there are data sets out there that can be used to run analyses that have not been gathered for that purpose.  If that is the case, the conclusions can be wildly inappropriate.  When reliable investigators conduct case control studies they design them to obtain good information that will either provide a new avenue for hypothesis generation or that will confirm previous research.  It is important to consider the statistical properties of the analysis that will be conducted with the gathered data. Did the study authors gather the data for the purpose identified in the publication?  The use of sample size calculation will ensure the study is sized appropriately to detect a correlation for the risk factors with the condition under study.

To judge how much weight to put on information published from a case control study, it is important to critically analyze how the populations were identified.  If it was self-reported by the subjects were there qualifying questions to determine the accuracy?  If it was based on diagnosis by medical professionals, was this verified. Is it possible that some in the control group also had the condition but had not been diagnosed?  Also, some thought should be given to the determination of the risk factors being studied.  Would a reasonable person answer the questions truthfully?  Can we believe self-report for these conditions?  Was the control population biased? If hospital-based controls were used, were they appropriate?  If determination of the disease status involved x-rays was the radiologist blinded?

It is not always easy to determine the answers to these questions when reading the results of a case control study analysis.  This is where your critical thinking skills will help you out.  Is it reasonable to believe that these determinations are unbiased?  Are there other studies with similar results so the mass of evidence is in agreement?  Many of the published research articles today are from case control studies.  These are the appropriate design for many outcomes, especially those that are rare.  These designs are also used where there is a long latent period or duration of expression.  These studies are usually less expensive than cohort or cross-sectional studies and are subsequently the largest portion of published research, as a rule.  As both the occurrence of disease and the risk factors occur in the past, it is difficult to determine which came first, which can make the observed association stronger than the actual association.  Carefully done case control studies still contribute a lot of information in the study of diseases.

What Can We Learn From Cross-Sectional Studies?

I have become a little frustrated lately with the barrage of inaccurate statements that are posted all over the Internet. The statements that I find the most frustrating are those that purport to cite medical research and then have attention grabbing headlines that are phrased to excite peoples' fears about their environment.  Many times they will quote an article from a respected journal but will somehow misconstrue the actual intent of that article.  In any case, it has driven me to this.  I want to begin a discussion here of the types of epidemiological evidence and help you learn how to judge how much belief you could reasonably attach to a study that might be cited in one of the articles.  I thought I would begin at the bottom of the ladder, so to speak, of weight of evidence.  Today's posting will discuss cross-sectional studies.

It is wise to keep in mind that correlation is not causation, but as Edward Tufte has said, 'it sure is a hint'.    

First, as you probably know, a cross-sectional study is like taking a snapshot on a particular day.  Our national census is cross-sectional in that you report your address on the first day of April of census year.  It doesn't matter where you resided the day before or where you are moving to the next day.  It is only for that day.  When a cross-sectional analysis is performed you are only looking at the characteristics that apply for the date and time the evidence is gathered.  This means that both the outcome and the exposure are from the same time period.  

There are a lot of questions you should ask yourself about this type of analysis.  For example, if you were trying to make an inference about a long-term exposure, you would need to have a window of exposure to justify the connection between exposure and outcome. It might also help if you could quantify the exposure. The classic example is the father of modern epidemiology John Snow. He deduced the cause of the Broad St. cholera epidemic was water from a particular well.  He removed the pump handle and the epidemic subsided.

 If you are looking at a highly stable population where nobody moves in or out the error induced by people moving around is minimal.  If your population is going through an upheaval of any kind in which there is migration of populations, from the rural areas to the cities, for example, any exposure that is focused on urban population might be biased due to what is basically an inaccurately measured exposure.  So you can ask yourself questions about the population stability where the cross-sectional analysis subjects are situated.  Also, the outcome might be associated with the place the subjects live.  If you are doing a cross sectional study in Rochester, MN, where the Mayo Clinic is located, there are some outcomes that will show up differently here if people can move to Rochester, MN, to be treated for the particular disease you are researching.

I know this can sound very complicated and when I first began my graduate program I almost threw my hands up in the air when these questions arose because they just seemed insurmountable.  That is why these types of studies should be carefully done and all of these questions addressed.  So it would be expected any issue that would make cross-sectional studies invalid or biased should be addressed in the article publishing these results.

If you doubt these biases were adequately addressed, you can see what direction the analysis results would be biased in the worst case.  If you are satisfied the investigators dealt adequately with these possible biases, the next question would be, what does it all mean?  

The evidence gathered in cross-sectional studies is of value to generate hypotheses for studies to be conducted that would provide better evidence.  What those studies could be is the subject for another day.

Borgen-- Is it the Best TV Show about politics, ever?

Well if it isn't, please let me know what is.  I just started watching the third and final season and only now realize how deep my mourning period since I watched the last episode of Season 2 was.  I am like a five  year old on  Xmas morning.  I want to keep watching, but I want to have more episodes to anticipate.

When the West Wing was at its best there was a similar feeling.  Only then I wished very hard that Bartlett was our real president.  Birgitte Nyborg is a Danish politician and the Danish system of government is not like the American system in many key concepts.  Birgitte Nyborg was Prime Minister for Seasons 1 and 2 and we were able to see the many ways consensus  was reached in  the Danish system.  We also were witness to how real compromise could work and everyone could still go home and get a good night's sleep.

But it is not just the idealism that deserves high praise, it is the detail with which these characters deal with their humanity in all of its many facets.  The actors perform with the highest quality and demonstrate humanity and complexity. Yet they find ways to lose and come back and try again.  It is breathtakingly beautiful in the execution.