Letters

Hound of the Baskervilles effect

BMJ 2002; 324 doi: http://dx.doi.org/10.1136/bmj.324.7345.1098 (Published 04 May 2002) Cite this as: BMJ 2002;324:1098

What about the good days?

  1. John H Glaser, data analysis consultant (glaserj{at}alum.mit.edu)
  1. 4 Woodpark Circle, Lexington, MA 02421, USA
  2. University of Wales, College of Medicine, Cardiff CF14 4XN
  3. Sociology Department, University of California at San Diego, La Jolla, CA 92093-0533, USA
  4. Mathematics Department, University of California at San Diego

    EDITOR—Phillips et al have uncovered a fascinating relation between the day of the month and the mortality rate for Chinese Americans and Japanese Americans.1 There clearly is an increase in chronic heart disease deaths of Chinese and Japanese on the fourth day of the month. Phillips et al attribute this to the similarity in the Chinese and Japanese languages of the spoken words “death” and “four.”

    Have Phillips et al considered whether days with a pleasurable association might have a beneficial impact? Their analysis shows a decrease in mortality for Chinese and Japanese on days 20, 26, and possibly 12. Is there any resemblance, either spoken or pictorially, between the words for those days and words evoking feelings of relaxation, wellness, or happiness?

    References

    1. 1.

    Bad4U?

    1. Robert G Newcombe, senior lecturer in medical statistics (Newcombe{at}cf.ac.uk)
    1. 4 Woodpark Circle, Lexington, MA 02421, USA
    2. University of Wales, College of Medicine, Cardiff CF14 4XN
    3. Sociology Department, University of California at San Diego, La Jolla, CA 92093-0533, USA
    4. Mathematics Department, University of California at San Diego

      EDITOR—In a large observational study Phillips et al noted an increase in cardiac mortality on the fourth of the month among Chinese and Japanese Americans that does not occur in white matched controls.1 This increase is particularly marked for inpatient deaths from chronic heart disease in California (figure 3). Phillips et al link this observation to their cultural and linguistic association of the number 4 with death.

      This effect may well be real, but some important issues remain unaddressed. Phillips et al have addressed a hypothesis that to them clearly relates to one specific day. To be really convincing, we need to be sure that the mortality on day four really should be regarded as an outlier, relative to other dates, and this is not the foregone conclusion that might be supposed. Directly from figure 3, the most extreme part of the data, the 95% confidence interval for mortality on the fourth day, overlaps the intervals for most of the other dates. Moreover, the relevant rate ratios, 1.07 for all cardiac deaths, 1.13 for chronic heart diseases, and 1.27 restricted to California, are presented with ordinary 95% confidence intervals, but no P values, in accordance with normal BMJ policy. The one situation, however, in which P values are more directly informative than the corresponding confidence intervals is when we are trying to assess whether a striking observation might be merely a coincidence. For cardiac mortality as a whole, the log rate ratio seems to be 0.068 (SE 0.021, z=3.17, P=0.0015). These figures are reconstructed from the heavily rounded ones given and hence are only a crude approximation. Although a P value of 0.0015 seems fairly extreme, this relies heavily on starting with the hypothesis that it is the fourth of the month that is different.

      A more conservative (more convincing to sceptics) P value involves a Bonferroni correction by a factor of 28 (the number of days studied, which are common to all months), giving 0.043, which, although technically significant, is far from extraordinary. It is true that when a similar process is applied to the more extreme rate ratios of 1.13 and 1.27 above, they remain highly significant. But it is commonly found that by restricting attention to subgroups of the data, one can enhance the nominal statistical significance in this way. It is far from clear whether it was a prior hypothesis that this effect would be much more marked in California than elsewhere. The hypothesis set out by Phillips et al relates exclusively to cardiac mortality, which conveniently squares with both the data presented here and the Baskerville link. But if stress associated with this date is so devastating to Chinese and Japanese people, one would expect to see marked effect on accidental and suicide deaths also, which do not seem to have been examined.

      References

      1. 1.

      Authors' reply

      1. David Phillips, professor,
      2. George Liu, student,
      3. Kennon Kwok, student,
      4. Jason Jarvinen, student,
      5. Wei Zhang, student,
      6. Ian Abramson, professor
      1. 4 Woodpark Circle, Lexington, MA 02421, USA
      2. University of Wales, College of Medicine, Cardiff CF14 4XN
      3. Sociology Department, University of California at San Diego, La Jolla, CA 92093-0533, USA
      4. Mathematics Department, University of California at San Diego

        EDITOR—Glaser thinks that low mortality levels on days 20, 26, and 12 might be linked to positive Chinese and Japanese attitudes. This suggestion is implausible. Firstly, the low mortality values on these days are not statistically significant, because the error bars for these days overlap with the regression line representing the expected mortality levels. (Because of a proofreading error, the top of the error bar for day 26 was cropped short and should have extended beyond the regression line to 67.06. Our original graph had the correct error bars, but we missed the distortion in the page proofs.) Secondly, although Chinese have a general preference for even numbers, the emotions evoked by “20,” “26,” and “12” are no more positive than for other even numbers.

        Nearly all Newcombe's concerns spring from his conjecture that we noticed the fourth day peak but did not predict it, and therefore need to assess the significance of this peak with the Bonferroni correction. Newcombe acknowledges that, even with this correction, the fourth day peak remains significant. We have used and analysed the Bonferroni correction in our previous articles, but this correction is inappropriate for the current investigation because we did predict the fourth day peak. 1 2 The circumstances were as follows.

        The lead author (DPP) and two of his Chinese students (KK and GL) were driving past a Chinese supermarket named “Ranch 99.” DPP asked the others why “9” appeared in this name, and was told that in Mandarin “9” is pronounced identically to the word for “lasting” and is therefore an auspicious part of a business's name. DPP then asked if there were any inauspicious numbers, and was told that, in Mandarin, Cantonese, and Japanese, the number “4” is pronounced almost identically to the word for “death.” At this point, DPP predicted that Chinese and Japanese mortality would peak on the fourth of the month. This prediction sprang to mind because DPP has long used short term mortality fluctuations to seek evidence of psychosomatic processes.3, 4, 5

        Incidentally, “4” is the only number between 1 and 28 with pronounced, negative connotations in Mandarin, Cantonese, and Japanese. Because the fourth day peak was predicted, one can legitimately compare mortality on the fourth (first period) with average mortality on days 1-3, 5-28 (second period). The error bar for the second period is very small and does not overlap with the error bar for the first period. This resolves Newcombe's concern about overlapping error bars.

        Newcombe wonders why we focused on cardiac mortality (and ignored suicides and accidents). We adopted this focus because we wished to test a surprising hypothesis: psychological stress might trigger heart attacks. We were not interested in testing the unsurprising hypothesis that psychological stress might trigger suicide.

        References

        1. 1.
        2. 2.
        3. 3.
        4. 4.
        5. 5.