Under-Reporting of Attributable Risk and Reporting of the Risk Ratio in Epidemiologic Literature

To the Editor:

Population attributable risk percent (attributable or etiological fraction; hereafter referred to as AR) is an integrated index incorporating both frequency of a risk factor and the risk ratio (relative risk or rate ratio: RR) of outcome with and without the risk factor. AR is the proportion of a specific disease in a population attributable to a factor and is essential for appropriate decision making in public health. ^1–3 The frequency with which the AR has been reported, however, was only less than one thirtieth that of RR in several major areas such as cancer and cardiovascular disease. ⁴

I searched the literature using PUBMED and selected 20 English journals whose title included “epidemiology” for review, and then counted the number of papers from 1966 through 1998 reporting RR and/or AR. The terms “relative risk OR odds ratio” and “attributable risk OR attributable fraction OR etiological fraction” were used for the search.

I found 28,848 articles published in the 20 journals. Of these, 3,307 papers reported RR (11.5% of the total articles) and 115 papers reported AR (0.40% of the total articles), giving an AR:RR ratio of 0.035. This ratio was 0.048 (58/1198) for the 1966 to 1992 period and 0.027 (57/2109) for the 1993 to 1998 period. TABLE

Table 1: Number of Articles Reporting Attributable Risk or Risk Ratio (MEDLINE 1966–1998)

The AR:RR ratio for epidemiologic journals (0.035) was similar to the previously reported finding for several medical fields (0.033), and the ratio has also decreased recently. AR is not only under-reported but also is incorrectly used, as has been demonstrated by Rockhill et al. ⁵ For consistency with the previous report, only the terms “relative risk” and “odds ratio” were used to look for papers using RR. An additional 1,280 articles in American Journal of Epidemiology in 1966–1998 were found when the terms “risk ratio,” “rate ratio,” and “hazard ratio” were used, an increase of 8%. The AR:RR ratio would, therefore, have been even smaller if these synonyms were added.

AR assumes the total removal of the risk factor and therefore provides an unrealistically optimistic estimate of the impact of the intervention. ⁶ If the AR of a certain factor is 30% and an intervention trial shows a 50% reduction of the disease, with an intervention targeting this factor, the expected impact of intervening on this factor in this population may be estimated roughly by multiplying the two, resulting in an expected reduction to 15%. AR for a common risk factor can be high even if the RR is relatively low. A high AR identifies those factors for which an effective intervention program can potentially have the greatest public health impact. Taubes has recently suggested that risk factors elucidated by epidemiologic studies should not be considered unless they have a RR of three or greater. ⁷ Although considering potential pitfalls in interpreting epidemiological findings is important, this assertion seems to be too simplified and ignores the significant role that a factor with a moderate RR but a high prevalence can play in promoting disease in the community. ⁸

Takeo Nakayama