Mindful Medicine: Don't confuse correlation with causation


Susan Fraser, FACP, an infectious disease specialist at Walter Reed Hospital in Washington, D.C. contacted us about the case of a 26-year-old German woman who presented with head congestion and difficulty breathing. She had a prior history of alcohol abuse, beginning at age eight, but she had been abstinent from alcohol for several years. She had undergone a Whipple procedure for chronic pancreatitis. She was previously evaluated at an outside clinic and diagnosed with asthma and allergies. A decongestant and an inhaler were prescribed but symptoms did not improve. She presented again and was briefly hospitalized and treated with theophylline and nebulizers, which temporarily resolved her symptoms. However, after several days her symptoms recurred.

She came to the Army medical clinic where routine chemistries were remarkable for a serum bicarbonate of 10. Hospitalization was recommended but the patient refused. However, several days later she agreed to be admitted for evaluation. Dr. Fraser was called to consult during this hospitalization in an Army hospital in Germany.

Commentary: Medicine cabinet yields clues

Dr. Fraser indicated that the patient was married to an American serviceman and was fluent in English. She was well-groomed, pleasant and cooperative, but seemed wary of medical professionals. She complained of “difficulty breathing” and “head congestion.” On examination she was afebrile and tachypneic and her lungs were clear without wheezes. A chest X-ray was normal. Repeat serum bicarbonate was again low and a blood gas was consistent with both respiratory alkalosis and metabolic acidosis.

The patient denied any recent alcohol use or ingestion of any illicit substances and assured Dr. Fraser that her only medications were oral contraceptive pills and pancreatic enzymes, with no over-the-counter medications. However, a drug screen was positive for a toxic level of salicylates. Upon further questioning, the patient denied ingestion of salicylates.

Suspecting that she was telling the truth, Dr. Fraser asked the patient's husband to bring in “every item in the medicine cabinet.” In this collection, she found large tubs of “Icy Hot.” It turned out that the patient was applying large doses of this substance three or four times per day to treat chronic low-back pain.

Dr. Fraser commented, “I deduced that the patient had developed salicylate toxicity from transdermal absorption of methyl salicylate. Her ‘difficulty breathing’ was tachypnea and her ‘head congestion’ included a sensation of ear fullness and persistent tinnitus which I had not elicited on my review of systems. She had full resolution of her symptoms by discharge.”

Physician plays detective

This is the kind of case that calls for the physician to use detective skills à la Sherlock Holmes. It requires questioning all prior assumptions and tying up loose ends. By focusing on the finding of metabolic acidosis, Dr. Fraser avoided the cognitive error of confirmation bias, which refers to our tendency to ignore or discount data that does not fit with the prevailing diagnosis—in this case, asthma and respiratory alkalosis.

Perhaps more importantly, she questioned whether this woman really had asthma. Asthma is a common cause of intermittent difficulty breathing and often is associated with allergies that could explain the patient's head congestion. Dr. Fraser said she began to develop doubts about the asthma diagnosis when she discovered that the patient had no wheezing on examination while symptomatic. It is not clear that she ever fulfilled the criteria for asthma but subsequent medical personnel automatically accepted it once the label was affixed.

This is a common clinical occurrence termed “diagnosis momentum.” Compounding the difficulty of making an accurate diagnosis in this case was the fact that the patient's symptoms had resolved during a hospitalization where she was treated with asthma medications. We often tend to interpret correlation as causation. If a therapeutic intervention, such as a medication, improves the patient's symptoms, we assume (often correctly) that the improvement is due to our action. But this is not always the case. This patient improved in the hospital due to an environmental change. The withdrawal of salicylate, not the addition of asthma medications, alleviated her symptoms.

Photomicrograph of salicylic acid crystal
Photomicrograph of salicylic acid crystal

This kind of error, confusing correlation and causation, which we term “causal confusion,” can affect not only individual patients, but also groups of patients participating in research studies. Researchers might interpret data from observational studies to show causation when there is merely correlation. In clinical trials, placebo controls are powerful safeguards against this type of error.

A recent case in point is the discrepancy between the conclusions of the observational Nurses' Health Study and the placebo-controlled Women's Health Initiative (WHI) regarding the risks and benefits of hormone replacement in menopausal women. Results from the observational Nurses' Health Study indicated that hormone therapy reduced women's risk of coronary heart disease and led to more widespread use of HRT among post-menopausal women. However, the Estrogen plus Progestin trial of the WHI later suggested an overall increase in the risk of CHD among women randomly assigned to combined hormone therapy as compared with those assigned to placebo. The trial was stopped early because it was found that the health risks associated with estrogen plus progestin outweighed the benefits.

Jerome Groopman, FACP, a hematologist/oncologist and author of the bestselling “How Doctors Think,” and endocrinologist Pamela Hartzband, FACP, are on the Harvard Medical School faculty. They also serve as staff physicians at Boston's Beth Israel Deaconess Medical Center, where Dr. Hartzband co-directs the internal medicine subinternship program.