| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
|
|
|
|||||||
|
|||||||||
TRACK to:
|
|
Electronic letters published:
![[Read Comment]](/icons/misc/sectionDOWN.gif){kind=icon}
Addressing questions
Data Do Not Support Prevention Paradox
Measuring the population impact
|
|
|||
|
Daniel C. Vinson, Columbia, MO, USA Professor, University of Missouri
Send response to journal:
|
Dr. Heller wonders if we used the correct formula for calculating a polytomous exposure's attributable risk. We've reread the article he cites by Hanley, checked with Dr. Heller, and confirmed that the calculations were correctly done. We greatly appreciate his thoroughness. To calculate the PIN-ER-t for a particular community or practice, one would need to know the distribution of various patterns of drinking and the base rate of injury. Both of those can be learned, though it would take a bit of work. I hope others will take our results and find new ways to put them into practice, both in the work we do as family physicians and in public policies. Dr. Nordstrom's quotation of the other version of the prevention paradox is quite appropriate. Like many of our preventive healthcare interventions (treating hyperlipidemia, for example), advising people to limit alcohol consumption to 1 drink per occasion would benefit each person who drinks 2 to 3 or 4 drinks only a little, but could prevent perhaps a million injuries in the U.S. each year. The version of the prevention paradox that we cited in the paper, which is the version that has been the focus of previous papers looking at alcohol's risks, isn't confirmed by our study. Fewer than half of all alcohol-attributable injuries are attributable to light to moderate drinking. But the population attributable fraction is still sizable. Something for us all to ponder. The lower limit of what is usually considered nonhazardous drinking is 1 drink. (Zero is also a nonhazardous amount, but that's beside the point here.) Dr. Nordstrom points to the last paragraph in our paper, "Consuming 2 or 3 alcoholic drinks for women, or 2 to 4 for men caused about 4% of all emergency department injury visits." Later in that paragraph, we note that those levels are "what has been considered a nonhazardous amount of alcohol." The point in that paragraph is not that the lower limit of nonhazardous drinking is 2 drinks instead of 1, but that 2 drinks has been considered (for most people most of the time) to be within the nonhazardous range. Dr. Nordstrom's question about choice of control group lies at the center of what makes case-control studies difficult. We considered many options. We recruited and interviewed 2103 patients presenting to the emergency departments with medical problems other than injury. We have included them in the analyses for only one paper (J Stud Alcohol 2003; 64:733-740), however, because we found their pattern of alcohol use differed greatly from our population-based control group. Hospital controls for this study just weren't like the population from which the cases came, and that's what an appropriate control group must be. Furthermore, as noted in the paper, we compared the pattern of drinking among our telephone controls with data from Missouri in the Behavioral Risk Factor Surveillance System. On the questions in the BRFSS, our controls were a close match to BRFSS data. We note, however, that both our community controls and BRFSS participants were recruited in the same way, random digit dialing. They may be similar because the samples have similar biases. That's the beauty of comparing injured people with themselves the day before. The case and the control come from the same population; indeed, they are the same persons, perfectly matched. That leaves open other sources of bias, including notably recall bias (it's harder to recall how much you drank yesterday than today). But at least on the question of control group selection, there is no better group (if the research question at hand allows a case-crossover design). We appreciate Dr. Heller's and Dr. Nordstrom's thoughtful comments and questions, and we look forward to hearing from other readers of the Annals. Dan Vinson, MD Competing interests: None declared |
|||
|
|
|||
|
David L Nordstrom, Minneapolis MN, USA Research Associate, University of Minnesota Medical School
Send response to journal:
|
Curiosity about whether it is heavy or moderate drinking that is primarily responsible for alcohol-related injuries led Spurling and Vinson to collect and examine data over 2 years from hospital emergency departments in one Missouri county. Setting aside the question of generalizability, is the evidence from this study for or against the prevention paradox (1)? Does most alcohol-related harm occur in those persons who are not alcoholic -- as claimed by Kreitman two decades ago (2) -- when the harm is restricted to injury? In the study by Spurling and Vinson, approximately equal numbers of people are classified as low (n=180) and high (n=172) risk drinkers, and each group accounts for about half -- 43% vs. 57% -- of the injuries that were estimated to be due to alcohol consumption during the 6 hours before injury. These results are not evidence for the prevention paradox, except in another meaning, also originating with Rose: “a preventive measure that brings large benefits to the community offers little to each participating individual” (1, cited in 3). Two other comments follow. In the paper’s discussion section, the final paragraph gives the minimum of the range of nonhazardous alcohol consumption as 2 drinks, while the methods section gives it as 1 drink. According to the abstract, the study selected control participants from some population by random digit dialing. The choice of case and control participants is crucial to study validity (4). In the Spurling and Vinson study, is the use of population instead of hospital control participants or the use of random digit dialing to select control participants a concern? If so, in which direction would the bias of the odds ratios and population attributable fractions be likely to occur? References 1. Rose G. Strategy of prevention: lessons from cardiovascular disease. Br Med J. 1981;282:1847-1851. 2. Kreitman N. Alcohol consumption and the preventive paradox. Br J Addict. 1986;81:353-363. 3. Rose G. The strategy of preventive medicine. New York: Oxford University Press, 1992, page 12. 4. Lasky T and Stolley PD. Selection of cases and controls. Epidemiologic Reviews. 1994;16:6-17. Competing interests: None declared |
|||
|
|
|||
|
Richard F Heller, Manchester, UK Professor of Public Health, University of Manchester
Send response to journal:
|
The paper by Spurling and Vincent provides a fascinating demonstration of the population impact of alcohol intake on injuries. Taking the prevalence of different levels of alcohol into account offers the chance to examine the population impact of different risk levels derived from the Odds Ratio. The authors use these figures to calculate PAF, which as they say is also called excess fraction, and more commonly called Population Attributable Risk (PAR). It would be important to know if the authors used the formula for polytomous, rather than that for dichotomous, risk levels as discussed by Hanley1. An extension of the approach to estimate population impact is to relate this to a particular population. If you know the prevalence of different levels of alcohol intake in your population, as well as the population size and the incidence of injuries in the population over a particular time period of interest, you can use the PAF (PAR) to calculate the actual number of injuries that different levels of alcohol intake can produce. The statistic is called "the population impact number of eliminating a risk factor (PIN-ER-t)", which can be defined as "the potential number of disease events prevented in your population over the next t years by eliminating a risk factor."2. It may help the policy-maker to apply the results of this demonstration of risk, of even quite low levels of alcohol on injuries, to a particular population. References 1. Hanley JA. A heuristic approach to the formulas for population attributable fraction. J.Epidemiol.Community Health 2001;55:508-14. 2. Heller RF, Buchan I, Edwards R, Lyratzopoulos G, McElduff P, St Leger S. Communicating risks at the population level: application of population impact numbers. BMJ 2003;327:1162-5. Competing interests: None declared |
|||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH |
TRACK: