Confusing Association with Causation:

What 10,675 Residents of London in 1853 and

23 Million U.S. Adults with Type 2 Diabetes May Have in Common

A tidbit of health news[1] that garnered considerable media attention recently highlights a pattern of events that occurs all too often when health care research is presented in the popular press. First comes a peer-reviewed, scientific report in which the investigators describe—in an appropriately measured, accurate way—an association between a risk factor and an outcome, sometimes hinting, while not openly declaring, a causal relationship. (Often, this initial cautious interpretation is used because journal peer reviewers and editors insist on it.) Second, the principal investigator’s sponsoring institution issues a press release that more directly refers to a possibility that the risk factor caused the outcome and calls for public action. Third, the media kicks the whole thing up a notch by overtly asserting the existence of an important public health risk that should be addressed by patients, physicians, or the government. Rounding out the “informational” picture, interested blogs pick up the news, often removing even the slightest hint of the investigator’s original (appropriate) caution in interpreting the study findings. The results for public health are misinformation, confusion, and sometimes decision making that does more harm than good.

Among the latest examples is a study of the association between the use of insulin and at least one negative health outcome (major cardiac event, cancer, or death) in patients treated for type 2 diabetes in the United Kingdom from 2000-2010, conducted by Craig Currie and colleagues.[2] Patients were not randomized (assigned by the investigators) to treatment, as in an experiment. Instead, the study was observational, meaning that the investigators measured (observed) the outcomes of patients whose doctors chose treatments—including insulin and/or oral drugs, such as metformin or sulfonylureas. Patients were classified into cohorts (groups) based on the treatment they received at any given time. For example, a patient initially taking metformin (the most commonly recommended first-line treatment) who was then switched to insulin because the metformin didn’t control his/her blood sugar was classified into two cohorts: the metformin cohort while the disease was being controlled by metformin, then the insulin cohort after the disease progressed to the point that insulin was necessary. Thus, by design, the insulin cohort would be expected to be “sicker” (have less well controlled disease and more negative outcomes) than the metformin cohort.

As my explanation above hints, the problem with observational designs of this type is that doctors choose treatments based on patient characteristics, usually including disease severity, comorbidities (other serious medical conditions), and age. So, there is often no way to tell whether the patients’ outcomes are due to the treatment or to the patient characteristics, a problem technically known as “confounding by indication.” Appropriately, Currie et al. acknowledged this possibility in discussing their findings:

“Because diabetes is a progressive disease, therapy choice will reflect this progression such that patients who are well controlled on first-line therapy, usually metformin, will remain on this therapy, whereas those who fail in the sense of increased levels of glycemia or development of complications will be treated more aggressively with the addition of other treatment options.”[2]

The study abstract also appropriately noted that “differences in baseline characteristics between treatment groups should be considered when interpreting these results.”[2]

But, a press release from Currie’s sponsoring institution was a little less cautious, with a headline that declared: “Taking insulin to control type 2 diabetes could expose patients to ‘greater risk of health complications’.”[3] Although noting for the press release that “patients currently being treated with insulin should not, under any circumstances, stop taking their medications,” Currie added that “this study shows that we need to investigate this matter urgently and the drug regulatory authorities should take interest in this issue.”

Predictably, media headlines went a little further than the press release in describing the hazards of insulin treatment, pronouncing insulin “risky for type 2 diabetics.”[1] And blog postings proclaimed the “risk” even more boldly. One blogger stated that “insulin doubles death rate in type 2 diabetics,” that insulin monotherapy (treatment with insulin alone) “resulted in” a wide variety of negative health outcomes, including heart attacks and kidney complications, and that “the general public [has become] a mass of guinea pigs for medical experimentation.”[4]

The effects of all this misinterpretation on public health, especially the decisions made by the approximately 23-24 million U.S. adults with type 2 diabetes,[5] may not be known for some time. But, for a sense of the potential damage, we can look to 19th century London, where residents were understandably terrified of cholera—a virulent disease of the intestinal tract that killed 50,000 British residents in a two-year period beginning in 1848,[6] and 10,675 Londoners in 1853 alone.[7] At the time, prevailing wisdom, shared by the vast majority of physicians and, unfortunately, politicians, was that cholera was caused by miasma (bad air). “Science” contributed to the erroneous view with an analysis of the association between elevation above the Thames River and cholera deaths—the higher above the Thames, where the air smelled better, the lower the mortality rate.[6] The association was so strong that one could predict cholera mortality rates based on elevation with nearly perfect accuracy.[8]

The only problem with this theory was, of course, that the nearly perfect association was entirely coincidental and had nothing whatsoever to do with the actual cause of cholera. At higher elevations above the Thames, residents were less likely to be exposed to water contaminated with Vibrio cholerae, the true causal agent of the disease. Unfortunately, as Steven Johnson notes in his fascinating book, The Ghost Map, zealous politicians were erroneously convinced by association that “all smell is disease,” and legislated a public works system to abolish (smelly) cesspools by dumping sewage into the Thames River. As a result, Johnson notes: “In the space of about thirty-five years, the Thames had been transformed from a fishing ground teeming with salmon to one of the most polluted waterways in the world—all in the name of public health,” thus “[delivering] the cholera bacteria directly to the mouths of Londoners.”[6]

In coming posts, I’ll be providing more examples of the potential harm caused by the “association versus causation” phenomenon because it is both frequent and much more important than is often acknowledged in research reporting. I’ll also be examining the old, ubiquitous enemy of public policy driven by the best of intentions—the unintended consequence. But meanwhile, perhaps the most important moral of the cholera and diabetes stories is that we shouldn’t be getting our health news from either blogs or politicians. We need to be able to use research sources, as much as possible, to read and interpret information on our own. It’s my hope that this blog will move us farther on that path. What do you think? Write to me or leave a comment below.

Sources:

[1] Tate N. Insulin risky for type 2 diabetics: study. NewsmaxHealth. February 6, 2013; and BBC News. Type 2 diabetes: insulin greater risk, finds Cardiff study. February 4, 2013.
[2] Currie CJ, Poole CD, Evans M, Peters JR, Morgan CL. Mortality and other important diabetes-related outcomes with insulin vs other antihyperglycemic therapies in type 2 diabetes. J Clin Endocrinol Metab. 2013;98(2):668-677.
[3] Cardiff University News Centre. Taking insulin to control type 2 diabetes could expose patients to ‘greater risk of health complications’. February 4, 2013.
[4] Stevenson H. Insulin doubles death rate in type 2 diabetics: study. GreenMedInfo.com. February 19, 2013.
[5] Of about 25.6 million U.S. adults with diagnosed or undiagnosed diabetes in 2011, approximately 90%-95% had type 2 diabetes. http://www.cdc.gov/diabetes/pubs/pdf/ndfs_2011.pdf
[6] Johnson S. The Ghost Map. New York: Riverhead Books; 2006.
[7] UCLA Department of Epidemiology, School of Public Health. Broad street pump outbreak.
[8] See: UCLA Department of Epidemiology, School of Public Health. Cholera mortality—London, 1849.