Reader Comments
Post a new comment on this article
Post Your Discussion Comment
Please follow our guidelines for comments and review our competing interests policy. Comments that do not conform to our guidelines will be promptly removed and the user account disabled. The following must be avoided:
- Remarks that could be interpreted as allegations of misconduct
- Unsupported assertions or statements
- Inflammatory or insulting language
Thank You!
Thank you for taking the time to flag this posting; we review flagged postings on a regular basis.
closeInsufficient attention given to personal smoking in analysis and report
Posted by whiteroseupton on 01 Mar 2016 at 15:20 GMT
Dear Editor
I congratulate Julia Dratva and co-authors on their assembly of a 2-cohort study population in which to investigate the effect of early life exposures on FEV1 decline in adults. With more than 12,000 adult participants having 2 FEV1 measurements at least 8 years apart in a broad range of European settings, the study?s power and generalisability are promising. Replication of their findings may take time, however, because studies of adults have generally not incorporated the broad range of early life exposures recorded by SAPALDIA and ECRHS. Nevertheless, all such studies will at least have the opportunity to test an effect of season of birth on FEV1 decline.
I worry, however, that the authors have not addressed personal smoking properly, increasing the possibility that some of their findings may be residually confounded to an important extent by personal smoking. This may be especially true of those attributed to maternal smoking, because maternal and personal smoking are often associated.
The dataset includes 12862 subjects, of whom table 2 reports that 9300 were never smokers (4959 women, 4341 men) and 3562 were current smokers (1690 women, 1872 men). So, if we are to take this classification at face value, none of the participants included in the study were former smokers, which seems implausible. In contrast to table 2, table S3 reports 5323 never smokers and 3527 current smokers, making 8850 in total. Presumably the missing 4012 (12862 ? 8850) were former smokers, inconsistent smokers and/or have missing co-variates?
Please could the authors inform us how they classified smoking status in their study, and how they dealt with smoking status that was inconsistent between surveys. I could not find any information about this in the methods section. Then, having explained this, please could the authors inform us how they adjusted for personal smoking in their analysis. Taking their report at face value, it appears that they adjusted only for pack-years of smoking, without taking account of former or current smoking status. If this is what was done, please could they inform readers what the effect of an additional adjustment for smoking status was on main effects for early life exposures.
Finally, please could the authors clarify the following:
1. Please report heterogeneity p values for a comparison of maternal age and maternal smoking effects on FEV1 decline by personal smoking status. Please inform us whether the personal smoking exposure in this test is never smoking/current smoking, non-smoking/current smoking, or never smoking/former smoking/current smoking.
2. Please clarify whether or not there are typographical errors in table S1 whereby negative signs appear to be missing before the upper bound 95% CIs for maternal age (-0.06 rather than 0.06, since p <0.001) and maternal smoking (-0.04 rather than 0.04, since p=0.02).
Yours faithfully
Mark Upton
General Practitioner, Helmsley Medical Centre, North Yorkshire, UK
RE: Insufficient attention given to personal smoking in analysis and report
JuliaDratva replied to whiteroseupton on 04 Mar 2016 at 09:04 GMT
Dear Mr. Upton
Thank you for your interest in our article. We looked into your questions and answer them in the following.
1. smoking category
Former smokers are a very heterogeneous group and 40% of the former smokers were already former smokers at the time of the first survey. Therefore we decided to consider these and other former smokers as never smokers for the general analyses, e.g. Table 2. Given that smoking is a known risk factor, we would assume the potential misclassification to reduce the power of observing a difference between the categories (smoker/never-smoker), not the other way around.
The potential past and current impact of smoking was addressed by the pack-years smoked both at survey 1 and survey 2.
In our main stratified analyses by smoking status however, we excluded the former smokers altogether to not introduce any missclassification. We did test our models in the former smoker strata. However, they showed no significant or relevant results and did not add to the understanding of the mechanisms or associations.
2. heterogeneity p values for a comparison of maternal age and maternal smoking effects on FEV1 decline by personal smoking status.
I believe you refer to “Separate analysis for synergistic effects between adverse early life factors and adult smoking showed that maternal smoking and higher maternal age were more strongly associated with FEV1 decline among current smokers than among never-smokers (S3-Table).”
In this case we had excluded the former smokers.
Please find below a figure for the stratified analyses by smoking status maternal smoking and maternal age. The first line (1 respectively 3) are the never smokers (former smokers excluded) and the second line (2 respect. 4) the smokers.
3. Please clarify whether or not there are typographical errors in table S1 whereby negative signs appear to be missing before the upper bound 95% CIs for maternal age (-0.06 rather than 0.06, since p <0.001) and maternal smoking (-0.04 rather than 0.04, since p=0.02).
We are sorry for a mistake that introduced itself into this supplemental table. Prior to the publication all main tables and analyses were double checked and rerun to check for mistakes or copy paste issues. This one must have escaped us. We reran the analyses now and mistakes in the p-levels became evident. In fact maternal smoking and maternal age are borderline significant, while the presented coefficients and CI are correct. In the full models mutually adjusting for other early life factors, we find statistically significant results indicating confounding by other early life factors. We will ask the journal to upload the corrected file.
Single exposures and dfev:
outcome exposure N coeff p-value CI_l CI_u
dfevml winter 12753 -2.001 0.002 -3.249 -0.753
dfevml mat.age 12528 -1.123 0.06 -2.307 0.061
dfevml urban 12753 0.607 0.41 -0.853 2.067
dfevml respinf 12753 -0.361 0.70 -2.220 1.498
dfevml daycare 12753 3.918 0.000 2.720 5.115
dfevml sharebr 12753 -0.464 0.40 -1.554 0.625
dfevml animal5 12753 0.616 0.28 -0.498 1.729
dfevml osic2 10054 0.146 0.82 -1.083 1.374
dfevml ysic2 12730 -1.540 0.008 -2.682 -0.397
dfevml fsmo 12753 0.377 0.51 -0.742 1.495
dfevml mat.smok 12753 -1.417 0.06 -2.872 0.037
adjusted for sex, mid age, mid age square, mid BMI, change in BMI (between survey 1 and 2), height, packyears smoked (if smoker), age at highest education, European region (random effect)
Yours faithfully
Julia Dratva