An article by Robert Wagner published over at Christopher Snowdon's Velvet Glove Iron Fist provides a novel and very convincing explanation for why the results from the Pell et al. study published in the New England Journal of Medicine are so different from population-based results for all of Scotland and why the study conclusions are invalid.
Pell et al. compared the number of admissions for acute coronary syndrome in nine hospitals in Scotland (representing 63% of admissions for acute coronary syndrome in the country) during the 10-month period prior to the smoking ban and the corresponding 10-month period the following year. The number of admissions declined from 3235 to 2684, a drop of 17%.
This 17% drop was compared to the trend in overall hospital admissions in all of Scotland for acute coronary syndrome during the preceding 10 years. According to the study, "the trend during the 10 years before legislation was a 3% mean annual reduction, with a maximum reduction of 9% in 2000."
Because the observed 17% reduction in admissions for acute coronary syndrome was much higher than the annual reduction during the 10 previous years and exceeded the highest annual decline between any two years, the study concluded that the observed reduction is attributable to the smoking ban.
Importantly, acute coronary syndrome was defined differently in the two time periods that were compared. In Pell et al.'s study of nine hospitals, acute coronary syndrome was defined as any detectable levels of cardiac troponin in the blood. In the population-based data for all of Scotland, acute coronary syndrome was defined by a physician's diagnosis based on the conclusion that the patient suffered either a heart attack or unstable angina. While troponin levels may have been used in the diagnosis, the troponin levels were not definitive in making the diagnosis.
The Rest of the Story
The key point that Wagner makes is that cardiac troponin is not 100% specific for acute coronary syndrome. There are a number of conditions not related to acute coronary syndrome that can cause low levels of troponin in the blood. But high levels are specific to heart damage. Thus, clinicians usually set a threshold level - a level high enough to rule out non-cardiac causes of troponin in the blood.
As Wagner explains: "The troponin test came into common usage in 2000, replacing earlier tests such as creatinine (CK) which were less specific to the heart muscle. A major complaint about troponin is that it delivers too many false positives at low levels. Many conditions cause small concentrations of troponin in the blood; only heart muscle damage causes large concentrations (with a few rare exceptions). The threshold level for a positive reading must be set high enough to eliminate the false positives. Setting the threshold at zero, as Pell did, is scientific nonsense, because it opens the door to MANY false positives."
"The Joint European Society of Cardiology/ACC Committee for the Redefinition of Myocardial Infarction has recommended that an increased concentration of cTn be defined as a measurement exceeding the 99th percentile of cTn concentrations observed in a healthy reference group."
So basing a diagnosis of acute coronary syndrome on ANY detectable level of cardiac troponin - as Pell et al. did - is going to include not only patients with acute coronary syndrome but also a substantial number of false positives.
But here is the clincher: as the cardiac troponin test is relatively new, it has undergone changes over time. New generations of the test have increased specificity for heart damage. Thus, the proportion of false positive tests is decreasing over time.
What does this mean for the Pell et al. study? What it means is that it is quite possible that a major reason for the 17% decline in admitted patients with suspected heart disease who have detectable troponin levels is that the troponin test became more specific and decreased the number of false positives.
In other words, patients who previously might have had a detectable troponin level because of non-cardiac causes would now (in the next generation of test) have a non-detectable level because of improved test specificity.
Wagner does not posit that the increased specificity of the troponin assay is the sole reason for the 17% decline in acute coronary syndrome cases observed in the Pell et al. study. He argues that the increased specificity of the test explains a portion of the decline. In particular, he argues that the changes in the test explain that portion of the 17% decline that exceeds the observed population-based decline in acute coronary syndrome cases, which was about 7%. Thus, there was an "appearance" of an additional 10% decline in acute coronary syndrome cases due to a decline in false positives.
Now, one might ask whether it is possible that the decline in cases of detectable cardiac troponin was so high that it could not be explained by the combination of declining secular trends in heart attacks and declining false positive tests, and thus implicating a role for the smoking ban. The answer is that the only way to determine this would be to compare the decline in cases of detectable cardiac troponin observed in Scotland to that in some comparison country or location without a smoking ban. However, Pell et al. failed to conduct such a comparison. Thus, their study cannot assess whether to smoking ban played any role at all in the observed 17% decline in cases of detectable cardiac troponin in the nine hospitals they studied.
The comparison of the decline in detectable cardiac troponin levels in the nine Scottish hospitals to declines over time in acute coronary syndrome as diagnosed clinically (not based solely on cardiac troponin and using a reasonable minimum cutoff value when troponin is used) during previous time periods is truly comparing apples to oranges, as I argued previously. Wagner's commentary helps to explain exactly why this is.
I should emphasize that the diagnosis of acute coronary syndrome is based on the totality of clinical information available about a patient, including the history, physical exam, electrocardiogram, echocardiogram or other diagnostic tests, and a variety of laboratory tests which may include cardiac troponin. But in Pell's study, what was being measured was not truly the diagnosis of acute coronary syndrome, but rather - the presence of detectable levels of troponin.
Given the decreases in false positive rates over time using the troponin test, it is unclear why Pell et al. would have chosen to include these false positives in their definition of acute coronary syndrome, rather than set a reasonable cut-off level that would have excluded the false positives.
At any rate, what Pell et al. did was not appropriate to make a determination of whether the observed changes were attributable to the smoking ban and for that reason, the conclusions of the study are not valid.
Population-based data on acute coronary events in Scotland have confirmed that there were no significant changes in the rate of decline from before to after the smoking ban. Thus, non-biased analytic techniques have failed to support the hypothesis that the smoking ban in Scotland has resulted in an immediate and dramatic decline in heart attacks.