Scientific method, in order to disprove the null hypothesis, demands highly controlled experimental conditions, where only a single exposure is altered. While this is feasible when dealing with chemical reactions in a beaker, and even, to a great extent, with bacteria and single cells in a petri dish, the proposition becomes a whole lot more complicated in higher order biology. In this way, the phrase "all things being equal" must really apply to the individuals or groups under study.
We call this formulation "the theory of counterfactual," and it is defined in the following way by the researchers at the University of North Carolina (see slide #3 in the presentation):
So, essentially what it means is figuring out what would have happened to, for example, Uncle Joe if he had not smoked 2 packs of cigarettes per day for 30 years. Now, our complexity as the human organism makes it impossible (so far) to replicate Uncle Joe precisely in the laboratory, so we must settle for individuals or groups of individuals that resemble Uncle Joe in most if not all identifiable ways in order to understand the isolated effect of heavy smoking on his health outcomes.Theory of Counterfactuals–The fact is that some people receive treatment.–The counterfactual question is: “What would have happened to those who, in fact, did receive treatment, if they had not received treatment (or the converse)?”–Counterfactuals cannot be seen or heard—we can only create an estimate of them.–Take care to utilize appropriate counterfactual
So, you see the challenge? This is why we argue about the validity of study designs to answer clinical questions. This is why a randomized controlled trial is viewed as the pinnacle of validity, since in it, just by the sheer force of randomness in the Universe, we expect to get two groups that match in every way except the exposure in question, such as a drug or another therapy. This is why we work so hard statistically in observational studies to assure that the outcome under examination is really due to the exposure of interest (e.g., smoking), "all other things being equal."
But no matter how we slice this pie, this equality can only be approached, but never truly reached. And this asymptotic relationship of our experimental design to reality may be OK in some instances, yet not nearly precise enough in others. We just cannot know the complete picture, since we only have partial information on how the human animal really works. And this is precisely what makes our struggle to infer causality problematic, and precisely what introduces uncertainty into our conclusions.
What is the answer? Is it better to rely on individual experience or group data? As always, I find myself leaning inward toward the middle. Because an individual's experience is prone to many influences, both internal, such as cognitive biases, and external, such as variations in response under different circumstances, it is not valid to extrapolate this experience to a group. In the same vein, because groups represent a conglomeration of individual experiences, smoothing out the inherent variabilities which ultimately determine the individual results, study data are also difficult to apply to individuals. For this reason medicine should be the hybrid of the two: the make-up of the patient can partly fit into the larger set of persons with similar characteristics, yet also jut out into the perilous territory of idiosyncratic individuality. This is precisely what makes medicine so imprecise. This is precisely the tension between the science and the art of medicine. Because "counterfactuals cannot be seen or heard," Uncle Joe!