Statistics Don't Lie Part-2

I promised in my earlier article,  that I would write about the various study methodologies and how one could judge the quality of a particular study, and by extension the validity of the results that are achieved from that. Now before we start, a few of you might wonder, who this guy is and what authority does he have to write this? Well I'm a medical student and it is a fact that there are a lot of people who are more knowledgeable and better equipped to write this. But not all of them have the time or inclination; so you have to make do with me. And when they do write, it becomes too technical. Like this. Some may wonder why one should know this. During my third year I had a subject called "Preventive and Social Medicine (SPM)" in my course. It is widely considered to be one of the toughest subjects in medicine. The standard book for SPM is Park. Much of what I write here is from Park. 

1. 

   A very famous meme in Facebook which highlights the difficulty a medico faces before his exams!! Also seen in the picture: Park, Dhingra and Khurana!

   
   
   I can divide my whole life into two unequal halves. 1. Before Park 2. After Park. Despite being a tough book to traverse, I should say, it has had a profound influence on me. It changed the way I approach or react to any information I come across. It made me question the claims made by this toothpaste that said 9 out 10 respondents use their toothpaste. It made me wonder if that battery could last ten times longer or if that drink could make me grow three times faster. I found myself sieving for facts in a haystack of lies.
   After Park, I really feel like a student of science and I believe with a little investment, you too can feel what I feel...

Jumping in, a research work is done to find a solution to a problem or question. We are more concerned about an epiodemiological research in this article. An epidemiological research can be either an Observational study or an Experimental study. An observational study observes events that are occurring with no interference from the observer. For example I observe that there is an over-representation of left handed players in cricket. When such an observation is made using proper statistical methods it becomes a Descriptive study,  which is a type of Observational study. See this article about handedness and sport.

The other types of Observational study are Case control study and a Cohort study. 

A case control study is a backward looking study. For example, I take 50 car accidents in which the drivers died (group A) and 50 more car accidents (group B) in which the drivers survived. I make sure that the magnitude of impact in both the groups are similar and that there is an equal representation of various types of cars in both the groups. The drivers in both the groups have a similar average experience in driving. I also make sure that the drivers of both groups have similar average age, height, weight etc and that both the groups have equal representation of male and female drivers. This process of ensuring comparability of both groups is called matching.

On observation I find that 40 drivers in Group B wore a seat belt while only 7 drivers in Group A wore a seat belt. This is statistically significant. (Click here and here for tests for statistical significance) So I conclude that "Wearing seat belts can save lives". 

This is a case control study because it studies the event (survival after an accident) after it has occurred.

The same kind of study can be done in a forward looking direction ie before the event has occurred; in which case it becomes a Cohort study. For example I take 5 thousand drivers who do not wear seat belts (Group A) and take 5000 drivers who wear seat belts (Group B) I ensure that both the group of drivers are similar in all other aspects like age, sex, experience etc (See confounding variable) Then I follow both the groups for a long time and wait till some of them meet with an accident. After I have obtained sufficient data, I compare the survival rates of both the groups. I find that there has been ten deaths in Group A, while only one guy died in Group B due to an accident. This is again statistically significant and I conclude that "Wearing seat belts can save lives" This is a type of cohort study because it studies the groups before the event has occurred (death after an accident)

A cohort study is better than a case control study but as you may have guessed, it is tougher to perform. It is more expensive and takes a lot of time to complete. 

Finally we come to Experimental study which again has multiple sub-types but we are concerned only about the best type among them "Randomised Controlled Trial (RCT)"
In my opinion RCT is one tool that has revolutionized medicine. RCT is used to know the effectiveness of any intervention. For example consider this sentence from my internal medicine textbook.

Oral acetaminophen has been found to effectively reduce fever

This statement would have been based on the results of several RCTs. Each of the various studies would have taken an experiment group A and a control group B (mandatorily in a random manner) wherein A would receive acetaminophen and B a placebo or some other drug for comparison. Both the groups would then be rigorously followed and the results documented and compared using established methods. The group receiving acetaminophen would have had their fever reduced significantly when compared to group B. (For instance follow this link to read the results of a study done in 2005) So we can conclude safely that "Acetaminophen reduces fever." So when I take acetaminophen, I know it will work. I know there is a lot of evidence to show that it reduces fever. I know....

So that is it... You now know the basics about different research methodologies. In my next article I'll try to explain how to judge the importance of a study result...