Epidemiology Unit, Institute for Medical Informatics, Biometry and Epidemiology, University Hospital of Essen, Hufelandstr. 55 45122 Essen, Germany, firstname.lastname@example.org
To the Editor:
In a recent article in Epidemiology, Hong et al.1 presented results from a case-crossover study that assessed the association between a decrease in ambient temperature and risk of ischemic stroke. They expressed the results as odds ratios for an interquartile change in ambient temperature. The interquartile range was 17.4°C.
I think these odds ratios mislead the reader because the interquartile change is not based on the distribution of observed temperature decreases within 1 day, but rather is based on the distribution of the daily average temperature.
A decline of 17.4°C (approximately 31°F) in mean temperature between 2 days would be very large and occurs only rarely. Therefore, the odds ratio accompanying such a large drop in temperature does not describe what the reader would like to know. It would be more useful, for example, to have an odds ratio for a 5°C temperature decrease.
Furthermore, the authors state that they “found a decreasing effect with time; confidence intervals did not include 1.0 after 2 or more days of exposure.”1(p.474) First, confidence intervals after 2 or more days of exposure do include 1.0 according to Figure 1 of their paper. Second, the authors obviously converted 95% confidence intervals into tests of statistical significance in order to decide whether a trend was present. This is a well-known fallacy in epidemiologic research2 and a practice that has been discouraged since the start of the journal Epidemiology.3,4
Institute for Medical Informatics, Biometry and Epidemiology
University Hospital of Essen
Hufelandstr. 55 45122 Essen, Germany
1.Hong YC, Rha JH, Lee JT, et al. Ischemic stroke associated with decrease in temperature. Epidemiology
2.Stang A, Anastassiou G, Bornfeld N, et al. The misinterpretation of study results based on statistical significance [Letter]. Epidemiology
3.Lang JM, Rothman KJ, Cann CI. That confounded p-value. Epidemiology
4.The editors. The value of p. Epidemiology