Doctors, Patients and Causality

Boston.com reported on a study that found a "yawning disconnect between what doctors say and what patients hear." The study, in the Annals of Internal Medicine (here), asked 153 patients who had received angioplasties whether the patients thought that the procedure could prevent death or heart attach. The same question was posed to 27 cardiologists. The patients thought stents would reduce the chances of heart attack while the cardiologists didn't. The study interpreted the finding as the result of misunderstanding in communication (something similar to problems reported for husband-wife communication).

From a theoretical perspective, there are a number of problems with that conclusion. The cardiologists, patients and the researchers are confusing counterfactual, causal, interventional and probability statements. The counterfactual question for patients that had a heart attack: "If you didn't have angioplasty, would you have had a heart attack?" The causal question: "What causes heart attacks?" And, the intervention question: "What diagnoses and treatments might prevent a heart attack?"

The differences in questions can be seen from the causal diagram above. There are lots of "causes" for heart attacks (the "back-door criterion) in addition to coronary blockage (genetics, life style, some kind of event or "shock," etc.). As a result of a heart attack or a diagnosis, either drugs or stents or both are used to treat the blockage. Would a patient with coronary blockage and chest pain and troponin elevation (meaning they had a heart attack) have had the heart attack with an earlier diagnosis and angioplasty? Probably not. But other blockages, not alleviated by stents, can develop over time and other causes, unrelated to blockage, can cause heart attacks.

Maybe too many people are receiving angioplasty when they have stable angina. Maybe drug treatment or prevention treatment or life-style changes would reduce the angina. But, if someone has 95% blockage in an artery, it's probably time for a stent. Essentially, the study poses a trick question for both patients and cardiologists by only asking about the stent -> heart attack link. Although patients and doctors may not communicate clearly, if patients and doctors were oriented to a better causal map (the one above could be improved), they'd probably reach the same conclusions.

PS: There are lots of other problems with the study: (1) small sample size, (2) lack of randomization, (3) single-site bias (all patients were treated at the Baystate Medical center), etc.