Using alternative statistical formats for presenting risks and risk reductions
Compared to probabilities, natural frequencies (events per 100 or 1000) are likely to be better understood by health-care professionals, policy-makers and consumers. However, compared to absolute risk reductions, relative risk reductions and number needed to treat may be perceived to be greater than natural frequencies and therefore may be more persuasive.
RHL Commentary by Bashour H
1. INTRODUCTION
Communicating evidence concerning risk and risk reduction to patients has practical and ethical goals. Informed patients are more likely to participate actively in their own care and to make wiser decisions. The ability to calculate risk and risk reduction can be useful when communicating with patients if not other health-care professionals or policy-markers.
In this era of evidence-based medicine, physicians are encouraged to appraise the latest and best available evidence and incorporate patient's values in reaching shared clinical decisions. Accurate clinical decision-making requires, among many other factors, an ability to estimate probability (1, 2). The importance of conveying probabilities and risks to patients and policy-makers have been frequently stressed upon. There is an increasing body of evidence supporting the design of effective evidence-based communication tools, but there is variable access to such tools in practice (3, 4, 5). This Cochrane review (6) was conducted to evaluate the effects of using alternative statistical presentations of the same risks and risk reductions on understanding, perception, persuasiveness and behaviour of health-care professionals, policy-makers and consumers.
2. METHODS OF THE REVIEW
Databases – including MEDLINE, EMBASE, and PsycLIT – were searched electronically without any language or date restrictions. The Cochrane Controlled Trials Register was also searched. Two review authors independently screened for relevance the titles and abstracts of identified articles, following which two other review authors independently screened the full text article for inclusion or exclusion of studies. Randomized and non-randomized controlled parallel and cross-over studies were included. The methodological quality of the included studies was evaluated for allocation concealment, randomization design and objective outcome.
Interventions of interest consisted of presentations of risk (frequency and probability) or risk reduction [relative risk reductions (RRR), absolute risk reductions (ARR), and number needed to treat (NNT)] for the same evidence about health care. Participants of interest included health-care professionals, policy-makers, and consumers (including medical students). The outcome effects were standardized using adjusted standardized mean difference (SMD) and the quality of evidence for each outcome was evaluated using the GRADE approach.
3. RESULTS OF THE REVIEW
Out of a total of 93 studies identified, 35 met the inclusion criteria. Most of these studies were from developed countries and had covered chronic diseases such as cancer and cardiovascular, genetic testing, and vaccination. Two of the studies exceeded the date of the search (2007). The methodological quality of the studies was classified as follows: 46% cross-over trials; 17% concealed; and 29% objective outcome studies.
Of the 35 studies, 20 were conducted with health consumers, 14 with health-care professionals, and one with both. No study was conducted with policy-makers. The 35 studies reported 41 comparisons, 21 of which were three-way (i.e. RRR versus ARR versus NNT), making the total number of comparisons 83, including eight comparisons of natural frequencies versus probabilities.
Understanding (measured as correct estimate or interpretation of a risk) as the main outcome with effect size as SMD equals to 0.69 (0.45–0.93) provided moderate quality of evidence suggesting that natural frequencies (prevalence ratios expressed as events per 100 or 1000) (7) may be understood better than probabilities by health-care professional and consumers, based on hypothetical scenarios from seven comparison and 642 participants. As for RRR, compared with ARR for presenting risk reductions, the mean understanding in the intervention groups was higher (SMD 0.02 , 95% confidence interval –0.39 to –0.43), suggesting no difference in understanding with moderate quality of the evidence.
Findings on secondary outcomes including perception (measured as rating on a scale of perceived effectiveness) and persuasiveness (measured as a hypothetical decision or intention or willingness to adopt an intervention) were presented in the review based on a varied number of comparisons and participants. Findings suggested that RRR was more persuasive than ARR and NNT with moderate quality of evidence.
The sensitivity analysis excluding studies of lower methodological quality did not substantially alter the results. In spite of the low number of studies, findings were presented separately for health-care professionals and health consumers with little difference between them.
4. DISCUSSION
4.1. Applicability of the results
Natural frequencies are likely to be better understood by health-care professionals, policy-makers and consumers than probabilities. RRR, compared with ARR and NNT, may be perceived to be greater than natural frequencies and therefore may be more persuasive. However, all this was based on hypothetical scenarios with no actual assessment of behaviour. The review highlighted sources of bias and lack of statistical power justifying the no difference between health-care professionals and health consumers, which is rather serious from a public health point of view. Furthermore, the lack of trials on policy-makers leaves an important gap in the understanding of how best to communicate risk to a key group of users of health knowledge. Even though most of trials were conducted in developed countries, the findings of this review are likely to be applicable to all settings.
4.2. Implementation of the intervention
The findings of this review, although stigmatized by the risk of bias and lack of statistical power, carry potentially important implications for health-care professionals' communication with their patients as well as with policy-makers. Health-care professionals should be encouraged to explain risks associated with interventions in terms of natural frequencies and relative risk. Researchers preparing information material for health-care workers, policy-makers and consumers should be encouraged to express risks in natural frequencies and relative risk.
4.3. Implications for research
Further research into this important statistical issue will be of great value in improving clinical and policy decisions. Alternative methods of estimating risk and risk reduction, including graphical methods, need to be assessed in different settings, especially in settings with economically underprivileged consumers. The context and participants’ expectations should be considered in the design of the studies. The authors of the review have correctly pointed out that language itself could be barrier where appropriate words do not exist to express complex statistical notions. Hence, studies of how best to express risks in different languages could be of interest to some researchers. There is also a need to improve the methodological quality of studies with better study designs. The designs should also take into consideration ethical implications in communicating risks.
References
- Ghosh AK, Ghosh K. Translating evidence-based information into effective risk communication: current challenges and opportunities. Journal of Laboratory and Clinical Medicine 2005;145:171-180.
- Ghosh AK. Clinical applications and update on evidence-based medicine. Journal of Association of Physicians of India 2007 ;55:787-794.
- O'Connor AM, Legare F, Stacey D. Risk communication in practice: the contribution of decision aids. BMJ 2003;327:736-740.
- Trevena LJ, Davey HM, Barratt A, Butow P, Caldwell P. A systematic review on communicating with patients about evidence. Journal of Evaluation in Clinical Practice 2006;12:13–23.
- Visschers VH, Meertens RM, Passchier WW, de Vries NN. Probability information in risk communication: a review of the research literature. Risk Analysis 2009;29:267-287.
- Akl EA, Oxman AD, Herrin J, Vist GE, Terrenato I, Sperati F. et al. Using alternative statistical formats for presenting risks and risk reductions. Cochrane Database of Systematic Reviews 2011;Issue 3. Art. No.: CD006776; DOI: 10.1002/14651858.CD006776.pub2.
- 7Gigerenzer G. What are natural frequencies? BMJ 2011:343:d6386; DOI: 10.1136/bmj. d6386.
This document should be cited as: Bashour H. Using alternative statistical formats for presenting risks and risk reductions: RHL commentary (last revised: 1 February 2012). The WHO Reproductive Health Library; Geneva: World Health Organization.