It is likely, that the biggest Impact & influence in current medical research may not come from the IQ of our scientists, their concepts, or the sophistication of the laboratory. Then what? Can you guess? It is the man-made mathematic sub-specality called statistics. We are going to either ratify or reject any research work ( on which we toil) based on the quality of numbers we generate. Such is the critical value of this specialty. Just pause a moment, and think over. How much importance do we give to the credibility and “quality of the interpretation” of any study? We have conveniently left it to our esteemed mathematical colleagues and some other invisible forces for a proxy inference.
I don’t think we will ever find an answer for this. Whether facts are made by statistics or statistics are made by facts ?
In recent times, one technique called propensity matching and scoring is used to conduct medical research where multiple covariables and confounders play.
What does the word propensity mean?
What is propensity matching in medical research?
In simple terms, it is doing a study without a true control group. It is a statistical gimmick where in we create an Imaginary or virtual patient arm What a way to conduct a scientific study? Those days, if someone suggests a study without a true control arm, it will go straight to the dustbin. (Of course, the concept came into vogue because we can’t have controls for ethical reasons or the rarity of the condition ) We do have many other conventional covariable analytical methods available A well-written reference (Ellicott C. Matthay, SSM – Population Health, 2020,)
Who created this propensity score?
I thought It was a new concept.No, it was proposed by Rosenbaum et al in 1983. (Ref 1) The extreme popularity it enjoys today is unexplainable. I think it is the simplicity, joy of doing a study without a troublesome control population, and the subsequent herd behavior of medical researchers.
Read here the pros and cons
Final message
Only two questions need to be answered before crowning the “propensity score” to glory in the statistical world. 1. Who has the final authority to define, what amounts to a confounding effect? 2, What are the statistical chances of missing an important confounder in toto due to baseline ignorance?
Most statistical methodologies are like Holywood movies, some strike gold for no reason in spite of a lot of flaws. A few examples are meta-analysis and non-inferiorly trials. Propensity matching with a synthetic control arm could be a useful methodology in very selected situations. It is unfortunate it has become a fancy tool and doesn’t deserve the wholesome approval for doing away with the true control arm.
Statistics may be great science, but it seems to work fine, only in the absence of continuous, unpredictable biological interference with mathematics.
Lastly, can propensity score take into account of confounding effects of the non-academic mindset of many researchers in senior positions? What shall we do with many important therapeutic guidelines created apparently based on solid evidence created with poorly created virtual (propensity) matches?
Postamble
Experience-based medicine, wild logical guesses, empiricism, and trial of error methods, all these are unavoidable in medical care and research. We have to move ahead with all the uncertainties in-situ and take our patients to a positive destination.
Reference
Propensity score in cardiology research
Dr.S.Venkatesan MD 