Additional reason to require reporting of underlying rates
The editorial by Gigerenzer al.[1] sensibly recommends that journals
require authors to include baseline risk information in articles
describing relative risks. There exists additional reason for such
requirement, given the pattern, inherent in the shapes of normal risk
distributions, whereby the rarer an outcome, the greater tends to be the
relative difference in experiencing it and the smaller tends to be the
relative difference in avoiding it.[2,3] Because it tends to be
systematically affected by the baseline rate, the relative risk is a
flawed measure of association. One needs the actual rates for an outcome
within the untreated and treated populations in order to effectively
appraise the strength of an association.[4]
The issue is especially pertinent to subgroup analyses, since,
without the actual rates, one cannot identify a meaningful interaction.[5]
But the issues is present regardless of whether authors are comparing
effects sizes. An irony is that those who understand the pattern
described in the second sentence will recognize that where relative
reductions or increases in a base outcome rate are particularly striking,
the percentage point change in an outcome rate - which, as the authors
suggest, is the more useful information - is likely to be small. I add
that the situations where interventions result in very large percentage
reductions in some adverse outcome rate tend also to result in very small
percentage increases in rate of avoiding the outcome - thus, though for
different reasons, resulting in the same sort of contrast as in the
authors' example of mismatched framing.
But there are many presentation issue warranting firm guidelines from
editors, one of which is suggested by usage in editorial itself. The
authors mention a study showing that a factor "increased the risk twofold"
and then refer to a "twofold risk." While everyone might not agree,
technically speaking, 2% versus 1% is a twofold risk, while a twofold
increase to a 1% rate would make it 3%. In any event, there exists a far
more common, and less debatable, misusage of a related nature where a rate
of 2%, for example, is described as two times greater than 1%, as would be
the predominant usage in the great majority of major scientific journals
published in English. But two percent is two times as great as 1%, while
a figure two times greater than 1 % would be 3%. Other problem areas
include the use, say, of " 5%" to mean either 5 percent or 5 percentage
points; the discussion of changes in mortality and survival differences
interchangeably without recognizing that they commonly change in opposite
directions; and the treatment of nonsignificant differences as if they
provided evidence of an absence of association.[6] Ideally editors of the
major scientific journals would address these issues collectively.
References:
1. Gigerenzer G, Wewarth O, Feuful M. Misleading communication of
risk. BMJ 2010; 341:c4830.
4. Scanlan JP. A fundamental problem with case control studies. BMJ
Oct. 2, 2010 (responding to Sedgwick P. Confounding in case control
studies. BMJ 2010; 341:c5136): http://www.bmj.com/content/341/bmj.c5136/reply#bmj_el_242476 (Accessed 16
Oct 2010).
5. Scanlan JP. Problems in identifying interaction where groups have
different base rates. BMJ Sept. 21, 2010 (responding to Altman DG, Bland
JM. Interaction revisited: the difference between two estimates. BMJ
2003;326:219): http://www.bmj.com/content/326/7382/219/reply#bmj_el_241943
(Accessed 16 Oct 2010).
6. Scanlan JP. Additional Issues to be considered by a World Council
of Epidemiology and Causality. Emerging Themes in Epidemiology April 8,
2010 (responding to Bhopal R. Seven mistakes and potential solutions in
epidemiology, including a call for a World Council of Epidemiology and
Causality. Emerging Theme in Epidemiology 2009,6:6): http://www.ete-
online.com/content/6/1/6/comments#396680 (Accessed 16 Oct 2010).
Rapid Response:
Additional reason to require reporting of underlying rates
The editorial by Gigerenzer al.[1] sensibly recommends that journals
require authors to include baseline risk information in articles
describing relative risks. There exists additional reason for such
requirement, given the pattern, inherent in the shapes of normal risk
distributions, whereby the rarer an outcome, the greater tends to be the
relative difference in experiencing it and the smaller tends to be the
relative difference in avoiding it.[2,3] Because it tends to be
systematically affected by the baseline rate, the relative risk is a
flawed measure of association. One needs the actual rates for an outcome
within the untreated and treated populations in order to effectively
appraise the strength of an association.[4]
The issue is especially pertinent to subgroup analyses, since,
without the actual rates, one cannot identify a meaningful interaction.[5]
But the issues is present regardless of whether authors are comparing
effects sizes. An irony is that those who understand the pattern
described in the second sentence will recognize that where relative
reductions or increases in a base outcome rate are particularly striking,
the percentage point change in an outcome rate - which, as the authors
suggest, is the more useful information - is likely to be small. I add
that the situations where interventions result in very large percentage
reductions in some adverse outcome rate tend also to result in very small
percentage increases in rate of avoiding the outcome - thus, though for
different reasons, resulting in the same sort of contrast as in the
authors' example of mismatched framing.
But there are many presentation issue warranting firm guidelines from
editors, one of which is suggested by usage in editorial itself. The
authors mention a study showing that a factor "increased the risk twofold"
and then refer to a "twofold risk." While everyone might not agree,
technically speaking, 2% versus 1% is a twofold risk, while a twofold
increase to a 1% rate would make it 3%. In any event, there exists a far
more common, and less debatable, misusage of a related nature where a rate
of 2%, for example, is described as two times greater than 1%, as would be
the predominant usage in the great majority of major scientific journals
published in English. But two percent is two times as great as 1%, while
a figure two times greater than 1 % would be 3%. Other problem areas
include the use, say, of " 5%" to mean either 5 percent or 5 percentage
points; the discussion of changes in mortality and survival differences
interchangeably without recognizing that they commonly change in opposite
directions; and the treatment of nonsignificant differences as if they
provided evidence of an absence of association.[6] Ideally editors of the
major scientific journals would address these issues collectively.
References:
1. Gigerenzer G, Wewarth O, Feuful M. Misleading communication of
risk. BMJ 2010; 341:c4830.
2. Scanlan JP. Can we actually measure health disparities? Chance
2006:19(2):47-51:
http://www.jpscanlan.com/images/Can_We_Actually_Measure_Health_Dispariti...
(Accessed 16 Oct 2010).
3. Scanlan JP. Race and mortality. Society 2000;37(2):19-35:
http://www.jpscanlan.com/images/Race_and_Mortality.pdf (Accessed 16 Oct
2010).
4. Scanlan JP. A fundamental problem with case control studies. BMJ
Oct. 2, 2010 (responding to Sedgwick P. Confounding in case control
studies. BMJ 2010; 341:c5136):
http://www.bmj.com/content/341/bmj.c5136/reply#bmj_el_242476 (Accessed 16
Oct 2010).
5. Scanlan JP. Problems in identifying interaction where groups have
different base rates. BMJ Sept. 21, 2010 (responding to Altman DG, Bland
JM. Interaction revisited: the difference between two estimates. BMJ
2003;326:219): http://www.bmj.com/content/326/7382/219/reply#bmj_el_241943
(Accessed 16 Oct 2010).
6. Scanlan JP. Additional Issues to be considered by a World Council
of Epidemiology and Causality. Emerging Themes in Epidemiology April 8,
2010 (responding to Bhopal R. Seven mistakes and potential solutions in
epidemiology, including a call for a World Council of Epidemiology and
Causality. Emerging Theme in Epidemiology 2009,6:6): http://www.ete-
online.com/content/6/1/6/comments#396680 (Accessed 16 Oct 2010).
Competing interests: No competing interests