Does Alcohol Increase the Risk of Preterm Delivery?

The authors respond: Kilduff et al suggest that underreporting is more likely in a questionnaire for the medical record than in a research questionnaire. For the period 1 February 2000 through 31 December 2000 we have collected both types of information: In the questionnaire for the medical record (QMR) we asked a single question comparable with that that used for the analyses of preterm delivery 1 (“How many drinks do you approximately drink per week now that you are pregnant (one drink is the equivalent of one [bottle of] beer, one glass of wine, or one schnapps)?”). The question did not specify subcategories of alcohol, and possible answers were 0; <1 drink/week; any whole number of drinks/week: 1, 2 etc. In the research questionnaire (QRES) we asked about average weekly intake of beer, wine, fortified wine and spirits, including strength of beer, and alcohol free beer and wine (subsequently coded as 0). Possible answers for each subtype of alcohol were as above. Intake of <1 drink/week was coded as a quarter of a drink/week. A total of 4,546 women returned QMR, of whom 4,411 had answered the question on alcohol intake, and 4,030 had filled in QRES. For 3969 women information was available for both instruments. Mean difference between the two measures (QRES − QMR) was 0.1 drinks/week (standard deviation, SD = 0.4). Eighty-six percent of women reported the same intake in both questionnaires, 5% underreported, and 10% overreported intake in QRES compared with QMR (Table 1). Interestingly, the tendency toward underreporting in QMR compared with QRES was most evident at the lowest intake levels, and might be explained by the more detailed questioning in QRES. Further, women who had not filled in QRES were more likely to be abstainers (61% versus 46% as measured in QMR), and smokers (21% versus 13%) compared with women who had filled in QRES. So, in this case, one would have to weigh what little may possibly be gained by using information from QRES against this selection bias. Table 1: Agreement Between Two Measures of Alcohol Intake During Pregnancy (Drinks/Week): Questionnaire for the Medical Record (QMR) Versus Research Questionnaire (QRES)Comparing the data from the questionnaire for the medical record with information from a more extensive interview, where the same precategorized answers were used as those reported earlier, 1 69% of women reported the same intake, 23% underreported their intake in the questionnaire compared with the interview (95% within one category), and 8% overreported (86% within one category). 2 In a later study we found that mean intake was 0.4 (SD = 1.2) drinks/week lower in the questionnaire compared with a two-week diary, and 0.3 (0.9) drinks/week lower compared with an average measure from an interview. 3 With respect to information on smoking habits, measurement error of potential confounders may distort the results. 4 We have previously compared the prospectively collected information on smoking habits with retrospectively collected information from questionnaires and found no noteworthy differences. 5 Differences were independent of recall time and pregnancy outcome, including preterm delivery (mean difference between methods (current − retrospective): 0.17 cigarettes/day (−0.32, 0.65) for preterm versus term deliveries). 5 Interestingly, recall diminished with increasing alcohol intake, particularly for women smoking ≥10 cigarettes/day. 5 It may be that both measures were underreported compared with interviews. We have recently collected data that may shed light on this point (data not yet available for analyses). Alternatively, measurements of cotinine in saliva, 6,7 serum, 8 or urine, 9 or of carbon monoxide in expired air 10 may be used as measures of smoking habits. It seems, however, that pregnant women claiming to be non-smokers may have high cotinine levels in serum and urine 8,11 (possibly because of exposure to passive smoking or denial of smoking status), and vice versa. 8,11 The findings of smokers with low cotinine levels suggest that because of intraindividual differences in cotinine concentrations in body fluids, a combination of self-reports and biological markers would be preferable. We take this opportunity to note that there was a minor error on page 513, left column, last paragraph, third sentence in the original article. 1 The definition of a drink is the equivalent to 4 cL (centiliters) of spirits, not 4 mL as stated in the original. Ulrik Kesmodel Niels Jørgen Secher Sjúrđur Fróđi Olsen