Rapid Responses to:
|
|
Rapid Responses: Rapid Responses published:
-
![[Read Rapid Response]](/icons/misc/sectionDOWN.gif)
why not use the NNT?
- Paul Young (20 November 2006)
-
Who will initiate?
- Soubhagya R. Nayak, Ashwin Krishnamurthy, Latha V. Prabhu (6 December 2006)
-
Comment
- Adnan Bhutta (16 December 2006)
-
Absolute risks as opposed to marketing gimmicks…
- James Penston (19 December 2006)
-
Let's be fair to relative risk
- Augusto Pimazoni, MD (20 December 2006)
-
A natural measure!
- Truls W. Gedde-Dahl (21 December 2006)
-
Get your basic statistics and study design right
- Joseph L Onwude (21 December 2006)
-
Re: why not use the NNT?
- David J Plews (22 December 2006)
-
Re: Re: why not use the NNT?
- Eddie Vos (26 December 2006)
-
Pills in the sky
- L Sam Lewis (27 December 2006)
-
CONSORT and Absolute Risk
- Peter Davidson (4 January 2007)
|
|
|||
|
Paul Young, University of Utah 84132
Send response to journal:
|
The authors make an excellent point about requiring the inclusion of absolute risk, but why not go a step farther and suggest that authors include a number needed to treat or number needed to harm? This allows the reader to easily understand the difference in absolute risk between the two treatments or exposures. Competing interests: None declared |
|||
|
|
|||
|
Soubhagya R. Nayak, Anatomist Kasturba Medical College, Bejai, Mangalore575004, India, Ashwin Krishnamurthy, Latha V. Prabhu
Send response to journal:
|
I congratulate Schwartz LM and co-authors of the article; regarding ratio measures in leading medical journals. This study brings out the loop holes in the presentation of the scientific articles. As 68% of articles failed to report the underlying absolute risks for the first ratio measure in the abstract, is a matter of concern. Paul Young’s suggestion is a good one in this regard, but the main concern, who will initiate? The author or editor? Competing interests: None declared |
|||
|
|
|||
|
Adnan Bhutta, Assistant Professor Arkansas Children's Hospital, 72202
Send response to journal:
|
I read the article by Schwartz et al with great interest and agree with the conclusion that the absence of an absoloute risk calculation can easily exaggerate readers' perceptions of benefit or harm. However, I found it quite ironic that the authors themselves have not given the absolute risk for the calculations in their paper inspite of using a relative risk ratio calculation. Competing interests: None declared |
|||
|
|
|||
|
James Penston, Consultant Physician/Gastroenterologist Scunthorpe General Hospital, North Lincolnshire DN15 7BH
Send response to journal:
|
Sir, Schwartz et al.[1] confirm what has hitherto been merely a suspicion of widespread neglect of absolute risks in published studies. They show that these statistics are frequently omitted from abstracts and either secreted away amongst the morass of data in the text and tables of papers or excluded altogether. As the authors remark, the failure to display clearly the absolute risks encourages a falsely inflated view of research results. But it is, of course, no accident that these statistics are eschewed in favour of relative risks; on the contrary, it is a manifestation of the pervasive manipulation of data by those with a vested interest in promoting particular outcomes. It is well recognised that results presented as relative risks are viewed more favourable than those given in terms of absolute risk, and this applies to doctors, other health-care professionals and, crucially, to patients. [2] Indeed, the evidence suggests that many treatments currently recommended in clinical guidelines would not be acceptable to either patients or their doctors if the data were provided in a more honest fashion. Abusing statistics by touting misleading relative risks may increase the profits of pharmaceutical companies and enhance the reputations of individual investigators but it does nothing to improve the understanding of research data. Patients, and doctors caring for them, require relevant and meaningful information in the form of absolute risks rather than a marketing gimmick. References [1] Schwartz LM, Woloshin S, Dvorin EL, Welch HG. Ratio measures in leading medical journals: structured review of accessibility of underlying absolute risks. BMJ 2006;333;1248-50. [2] Penston J. Patients’ preferences shed light on the murky world of guideline-based medicine. J Eval Clin Pract 2007 (In Press). Competing interests: None declared |
|||
|
|
|||
|
Augusto Pimazoni, MD, President and Managing Director MED MARK Medical Marketing Consultants
Send response to journal:
|
Dear Dr. Schwartz: Your paper is a masterpiece in terms of teaching the basics of medical statistics to non-statisticians like myself and I congratulate you for that. After reading your article, I decided to look up my “medical googlepedia” to further explore the technical differences between absolute vs. relative risk and I came up with an excellent text on Effective Clinical Practice under the title “Primer on Absolute vs. Relative Differences”, produced by the American College of Physicians. First of all, the text states that an absolute difference is a subtraction and that a relative difference is a ratio. With these concepts in mind, the example in the text refers to the evaluation of the risk for blindness in a patient with diabetes over a 5-year period. If this risk is 2 in 100 (2%) in a group of patients treated conventionally and 1 in 100 (1%) in patients treated intensively, the absolute difference is derived by simply subtracting the two risks, i.e., 2% minus 1% equals 1%. On the other hand, the relative difference is expressed by the ratio of the two risks, i.e., 1% divided by 2% equals 50%. In this particular example, the relative risk is a much better expression of what happens in the real world, in comparison to absolute risk, for the very simple reason that the relative risk reflects that intensive therapy effectively reduces the risk of blindness by half. OK, let’s recognize the importance of absolute risk but, at the same time, let’s be fair to the relative risk… AUGUSTO PIMAZONI, MD
Reference: “Primer on Absolute vs. Relative Differences” . Effective Clinical Practice. Available at: http://www.acponline.org/journals/ecp/janfeb00/primer.htm. Access on: December 19th, 2006. Competing interests: None declared |
|||
|
|
|||
|
Truls W. Gedde-Dahl, Retired researcher Foundation for Local health trials, N-1259 Oslo
Send response to journal:
|
Thanks to Schwartz et al. for their paper and to others for their responses which all seem valid. Absolute risks often makes research results better understood as the concept accord with the simple and natural way of reasoning made spontaneous by doctors and lay people alike. Though absolute risk of course is no panacea we found differences between absolute risks very useful 16 years ago (1). Yours sincerely
(1) Gedde Dahl TW, Bjark P, Høiby EA, Høst JH, Bruun JN. The severity of meningococcal disease: Assessment by factors and scores and implications for patient management. Rev. Infect. Dis. 1990;12:973-92 Competing interests: None declared |
|||
|
|
|||
|
Joseph L Onwude, Consultant Gynaecologist and Medical Statistician Capio Springfield Hospital, CM1 7GU
Send response to journal:
|
Schwartz et al have highlighted an important extension to the phrase ‘when you see a proportion, look for the numerator and denominator’. Their message is that when you see a relative risk, look for the absolute risks that were used to generate the relative risk. This information should be available in the same line that expresses the relative risk. However this paper has serious deficiencies. Firstly, the analysis of 161 cohort studies suggests that Schwartz et al are not aware that the absolute risk is not a measure of effect of observational studies. They are correct that absolute risk is the proportion of disease in exposed and unexposed groups. Therefore the measurement of absolute risk is only appropriate in prospective studies that start with exposed and unexposed groups. Indeed, the measurement of risk and rate ratios [measure of effect for prospective cohort studies] utilises proportions in a similar way to absolute risk. In retrospective observational studies [retrospective cohort studies, cross sectional and case-control studies] the absolute risk is not applicable because the sample size is often not calculated [you tend to use the dataset you have or have collected over a time period]. The proportions or absolute risks in the comparison groups are replaced by the odds in the groups. Secondly in excluding case-control studies, Schwartz et al recognised that absolute risks are not calculable where a study starts from a disease state and measures the exposure in the past. They are wrong if they did not exclude retrospective cohort studies which measured exposure, having started from the disease state. Thirdly, Schwartz et al are in error in their analysis. They report on absolute risks and relative risks when their study was not a prospective study or randomised controlled trial. Adding 95% confidence intervals is not helpful in this context. They also report ‘mean number of ratio measures 3.6 vs 2.5, p = 0.002’. This has no meaning. The ‘median number of ratio measures’ is more appropriate unless their data was normally distributed. While it is important to report absolute risk as a support for relative risk, absolute risk reduction and number needed to treat, it is more important to aware of the whole issue of intervention versus observational study and prospective versus retrospective study. The absolute risk is a measure of effect of randomised controlled study where there is a requirement that the differences between the comparison groups are quantified. Otherwise why bother with randomisation and controlled treatment. Competing interests: None declared |
|||
|
|
|||
|
David J Plews, GP/PCT GP Medical Adviser Market Surgery, Wath-upon-Dearne, Rotherham, S63 7RA
Send response to journal:
|
I agree with this suggestion. NNT (number needed to treat) and NNH (number needed to harm) is much easier for me, patients and the non-medical press to understand. Competing interests: None declared |
|||
|
|
|||
|
Eddie Vos, Engineer Sutton (Qc) Canada J0E 2K0
Send response to journal:
|
I agree, Numbers Needed to Treat [NNT] or Harm are THE clearest way to report results of trials however the time period has to be clearly stated in the same sentence, and in the abstract. For example, let's start with obtaining the not yet reported NNT's for the 4 groups in the Heart Protection Study (men, women, diabetic and not) regarding all-cause mortality, per year of simvastatin use, in order to get a clear picture what of that statin really does. Competing interests: None declared |
|||
|
|
|||
|
L Sam Lewis, GP Surgery, Newport, Pembrokeshire SA42 0TJ
Send response to journal:
|
HPS treated 10269 with Simvastatin 40mg daily for 5 years, and compared that to 10267 on Placebo. There were 1507 deaths in the Placebo group, and 1328 deaths in the Simvastatin group. See: - http://www.ctsu.ox.ac.uk/~hps/slidesfinal/HPSwebslides_files/frame.htm Thus 179 deaths were avoided in 5 years, by 10269 people being treated. or 1 death avoided in 1 year if 286 people take 40mg simvastatin., ie: NNT= 286 Not a very powerful effect – and even less if you insist, as Eddie Vos regularly does, on subset analysis of the lower risk groups such as women. But that is the nature of the diminishing powers of modern medicines – attested by the fact that we need trials of 10,000 people in order to show the effect ! It may yet be that Lives can be saved cost-effectively , as HPS argued in their recent BMJ paper:- Heart Protection Study Collaborative Lifetime cost effectiveness of simvastatin in a
range of risk groups and age groups derived from a randomised trial of 20 536
people From a patient perspective, I felt that Richard G Rodgers summed it up neatly – http://www.bmj.com/cgi/eletters/333/7579/1145#149825 “At 1% annual risk the annual risk reduction is 0.33%, a NNT of 300: that means there needs to be 300 patient years of treatment for one event to be avoided. For the individual that surely must equate to 300 years of daily pill taking. At 85 (and most, if not all 85 year olds are at 5% 5 year risk) life expectancy is 5-6 years so 300 years does not seem to be meaningful. Even at 35 years (life expectancy about 45 years) there needs to be 6-7 lifetimes of daily pill taking for one event to be avoided. “ I suggest new statistic which answers the question ‘ How many pills do I have to swallow, or for how many years, to be sure of a benefit ? ‘, I propose we call this useful new statistic – ‘pills in the sky’ - - recalling the hymn in which we will all eat ‘pie in the sky’ in the ‘sweet bye-and-bye’. Pass the red wine, and an Aspirin. Yours sincerely, Dr Sam Lewis GP Competing interests: None declared |
|||
|
|
|||
|
Peter Davidson, Public Health Consultant WIHRD, University of Southampton SO16 7PX
Send response to journal:
|
The authors of the paper on ratio measures cite the CONSORT statement and suggest that it recommends the presentation of absolute risks. They are wrong; it calls for the reporting of absolute numbers rather than percentages, but makes no mention of absolute risks. Absolute risks are very valuable to help understand the clinical importance of outcomes in trials for instance. But they are dependent on the underlying risk in the study population, which may be different to patients in clinical practice. It would be very useful to have guidance on the presentation of ratios and differences in the next revision of CONSORT. Competing interests: None declared |
|||