The Blunders Of Sammec (1)
400,000 Killed By Smoking!?
by Rosalind B. Marimont
That smoking causes 400,000 deaths annually is now widely promoted as a statistical truth. The recent campaign against teenage smoking asserted that one out of three teenagers who smoked would be killed by his habit. These numbers are a gross misinterpretation of the CDC SAMMEC results, and a gross overestimate of the importance of smoking as a cause of death. Another mantra of the Anti-Smoking Partisans (ASPs) is that smoking kills more people than alcohol and drugs combined. This latter piece of disinformation has been used to justify neglect of the shocking rise in teenage binge drinking and driving. Neither candidate for president has even mentioned teenage drinking, and the Clintons have hardly mentioned drugs until the Republicans made an issue of it.
The 400,000 plus estimate is the result of logical and epidemiological blunders and a lack of scientific integrity, by the fanatic anti-smoking lobby. The CDC estimate is described as the number of deaths ASSOCIATED with smoking, not CAUSED BY it. This is not a semantic distinction, because a death can be associated with many factors.
Among risk factors for heart disease, for example, are hypertension, high serum cholesterol, obesity, sedentary life style, smoking, and genetic factors. If we ran SAMMEC computations for each of these factors, we could estimate the number of heart disease deaths associated with each one. But suppose that John Smith, who died of heart disease, had all of these factors. He would have contributed 6 deaths to the total associated deaths. So that when we sum up these results to arrive at the total deaths, we find that our total is MUCH LARGER THAN THE NUMBER OF PEOPLE WHO ACTUALLY DIED OF HEART DISEASE.
A simple numerical example will demonstrate the SAMMEC method, and its multiple counting error. Let us consider heart disease. A behavior or attribute is said to be a risk factor for death by heart disease(HD) if the population exhibiting that behavior has a higher HD mortality rate than the population which does not exhibit that behavior. If obese people have a higher mortality rate than non-obese, obesity is a risk factor for heart disease death. The ratio of these two mortality rates is unknown as the risk ratio of obesity for heart disease death, and of course, is measured statistically.
How does SAMMEC compute the deaths associated with some risk ratio? Assume that we have measured the risk ratio of obesity for HDD to be 4. Assume that we have a population of 1000 people, of whom 500 are obese and 500 are not. We observe 10 deaths by heart disease. We can then compute that 8 of these deaths would occur among the obese, and 2 among the non- obese, the ratio of 4:1. Let us call this risk ratio r. Then SAMMEC assumes that if the obese people were not obese, they would have the same mortality rate as the non-obese, or only 2 deaths. Therefore 6 deaths among the obese are attributed to obesity, or the fraction (1-1/r) of the deaths of the obese, in this case 3/4. It is easy to compute the fraction of the total deaths, which is called THE SAF, OR STATISTICALLY ATTRIBUTABLE FACTOR. IF WE STOPPED AT THIS POINT, WE WOULD SAY THAT OBESITY CAUSES 3/4 OF THE DEATHS OF ALL OBESE HEART DISEASE PATIENTS.
But is this true? Let us continue our computation, and consider hypertension as a risk factor for HDD. To simplify the calculations let us assume that hypertension also has a risk ratio of 4 and this is the crux of the overcount, assume that the same group of people who are obese are also the hypertensives. Then we find that 6 deaths of our hypertensive group are attributed to hypertension. Similarly we can find that smoking, lack of exercise, and high cholesterol levels each result in 6 deaths. So that we find that our 5 risk factors are associated with 30 deaths by heart disease. BUT ONLY 10 PEOPLE DIED ALTOGETHER, AND ONLY 8 IN THE HIGH RISK GROUP. Only if each person had only 1 risk factor for any cause of death, would the SAMMEC SAF be a true fraction, in the sense that all fractions would add up to 1.
This overcount is not the only problem with the SAMMEC system. In estimating risk ratios, we compared death rates of smokers to those of non-smokers. This ratio would be a true estimate of the effects of smoking only if the two groups were identical in all other respects than smoking. This of course is not true - the measurement is done without controls. For this reason epidemiologists rarely take seriously risk ratios of less than 3.
In the SAMMEC report, of 102 risk ratios of smoking for various diseases, only 40 are greater than 3. IF WE CONSIDER ONLY RISK RATIOS equal to or greater than 3, THE NUMBER OF DEATHS IS CUT IN HALF, TO ABOUT 200,000. Even if we reject only those less than 2, the number is cut by about one third, to about 270,000. And these corrections still leave a number of serious confounders.
One of the most serious confounders in smoking studies is the inverse correlation of smoking with socio-economic status (SES). Low SES is one of the best predictors of disease and early death.
And finally, no attention is paid to the benefits of smoking. For some conditions, such as obesity, the risk ratio of smoking is less than 1, since smokers are less likely than non-smokers to be obese. Also, smokers are Iess likely to have ulcerative colitis. It is of course heresy to suggest that smoking can have any good effects, but like caffeine, nicotine is known to improve alertness, and allay depression and anxiety. There is recent evidence that smoking may provide some protection against Alzheimer's disease and Parkinson's. These good effects are rarely mentioned for fear of being branded a tool of the tobacco companies.
It has been said that truth is the first casualty of war. The deceptions of the war on smoking have done incalculable harm to the nation. The grossly overstated dangers of smoking to health have distorted the nation's health priorities. To equate smoking with alcohol or drugs as teenage dangers is obviously absurd, and would never have happened if the health dangers of smoking had been accurately reported. The war on smoking has become a crusade of good against evil, and logic and science have been prostituted to attain its objective.
(1) Shultz, Novotny, and Rice, "Quantifying the Disease Impact of Cigarette Smoking with SAMMEC II Software", Public Health Reports, May-June 1991, Vol 106, No 3, pp 326-333.
Rosalind B. Marimont
Rosalind B. Marimont is a retired mathematician and scientist, having done research and development for NIST (or the Bureau of Standards (NBS), as it was then) for 18 years, until 1960, and NIH for another 19, until her retirement in 1979. She started in electronics defense work during World War II at NBS, then went on to the logical design of the early digital computers during the fifties. In 1960, she moved to NIH, and there studied and published papers on human vision, speech, and other biomathematical subjects. Since her retirement she has been active in health policy issues - first, the treatment of chronic pain by integrated mind/body methods, and second, the dishonest war on smoking which has corrupted scientific research and gravely disorted the nation's health priorities.
For more than fifty years she has read and evaluated many kinds of scientific studies, and has sometimes served as a reviewer for scientific journals.