Introduction
The success of evidence-based praxis depends on the clear, effective communication of statistical information (1). The aim is for the information to be understood, the risks to be correctly estimated and, finally, for informed decisions to be made possible. The development process of the information comprises the selection of the content that should be communicated and the critical appraisal including the decision whether presentation of numerical data is adequate. In order to communicate frequencies correctly it is important to check the existing evidence on various presentation formats with regard to their efficacy and also to check for possible adverse effects.
To present statistical information (probabilities, quality of diagnostic tests as well as benefits, harm and side effects of medicinal measures) verbal descriptors are used. Verbal descriptors are more or less specific linguistic transcriptions of frequencies, e.g. seldom, occasionally, frequent, certain or probable.
However, studies have shown that the interpretations of linguistic descriptions and the resulting perception of risks differ significantly both inter-individually and between (medical) nonprofessionals and professionals (2). Verbal information concerning side effects leads to overestimating the probability of their occurrence (2).
A first attempt to standardize the verbal description of risks was made by the European Commission in 1998 (3). In the guideline on the readability of information on medical products, five verbal descriptors were each allocated with a defined numerical frequency or range of frequency (3). The Federal Institute for Drugs and Medical Devices (BfArM) also demands the use of fixed linguistic descriptions combined with a numerical indicator when making statements about the frequencies of side effects in product information leaflets (4). In a survey carried out in Germany, the participating physicians, pharmacists, and legal practitioners were unable to allocate the verbal probabilities for side effects correctly to the corresponding percentages (5).
There are various formats available for the numerical presentation: natural frequencies, percentages, absolute risk reduction (ARR), relative risk reduction (RRR), number needed to treat/screen/harm (NNT, NNS, NNH). The natural frequencies are given in differing reference parameters (denominators): 1 in 100, 1 in 1000, 1 in 10000. The effects of these formats have been investigated in several systematic reviews (1, 6-8), whereby Akl et al. for the first time included the outcome persuasiveness which is measured by means of hypothetical choices (1).
For a long time the use of natural frequencies was considered superior to percentages and was also strongly advocated in the context of evidence-based medicine (9). Several studies have investigated how often the positive predictive value of a test was correctly estimated when statements were made about the prevalence of a particular illness or about sensitivity and the false-positive rate (10-13). They found out that the number of correct answers was very low, even when the parameters were shown as natural frequencies. That is why positive predictive values and other test rating parameters should be shown directly, without the reader of the information having to carry out the corresponding calculations (11).
A possible adverse effect of presenting risks as natural frequencies can arise through the denominator neglect: Perception is focused on the number of observed incidents (numerators), no matter how small, and not on the reference parameters (denominators) (14). If the risks are compared – perhaps with or without a therapy – the perception is distorted even more if the chosen parameters are different (e.g. 80 of 800 vs. 20 of 100). The disadvantages of this presentation have long since been the subject of discussion. The use of different parameters makes it difficult to compare different statements and to estimate the level of risk correctly (15) which can lead to overestimating the risks (16).