Advertisement
Research Article

Impact of Scotland's Smoke-Free Legislation on Pregnancy Complications: Retrospective Cohort Study

  • Daniel F. Mackay,

    Affiliation: Centre for Population and Health Sciences, University of Glasgow, Glasgow, United Kingdom

    X
  • Scott M. Nelson,

    Affiliation: Centre for Population and Health Sciences, University of Glasgow, Glasgow, United Kingdom

    X
  • Sally J. Haw,

    Affiliation: Scottish Collaboration on Public Health Research Policy, Western General Hospital, Edinburgh, United Kingdom

    X
  • Jill P. Pell mail

    j.pell@clinmed.gla.ac.uk

    Affiliation: Centre for Population and Health Sciences, University of Glasgow, Glasgow, United Kingdom

    X
  • Published: March 06, 2012
  • DOI: 10.1371/journal.pmed.1001175

Reader Comments (2)

Post a new comment on this article

Adjusting for seasonal patterns

Posted by agbarnett on 18 Mar 2012 at 03:50 GMT

The rates over time in all three birth outcomes are clearly seasonal (Figures 1 to 3). This is not surprising, as strong seasonal patterns have been found in preterm birth and birth weight in many parts of the world (www.ncbi.nlm.nih.gov/pubm...). A seasonal pattern creates a potential issue for this study because the change-point of 1 January 2006 appears to coincide with the regular seasonal decline in preterm birth. This means some of the decline in preterm birth attributed to the smoking policy change may be due to season. For the small for gestational age and low birth weight outcomes the 1 January 2006 change-point possibly coincides with the seasonal increase. This means the decline attributed to the smoking policy may increase after adjusting for season. Adjusting for season could be achieved by including a categorical variable for month of birth, or a cosinor for month of birth. A residual check of the rates over time using the autocorrelation function or cumulative periodogram would test whether there was any remaining seasonal pattern.

No competing interests declared.

RE: Adjusting for seasonal patterns

DFM_glasgow replied to agbarnett on 19 Mar 2012 at 10:15 GMT

The model contained a control for week of conception which will easily take care of any seasonal variation. The model is not a time series model so correlograms or their spectral equivalent can't be used.

No competing interests declared.

RE: RE: Adjusting for seasonal patterns

agbarnett replied to DFM_glasgow on 21 Mar 2012 at 05:05 GMT

Week of conception would only adjust for season if it was fitted as a categorical variable. I presumed that week of conception was fitted as a linear variable to model the long-term trend ("We were able to account for underlying trends", page 7). If week of conception were fitted as a categorical variable it would have around 51 degrees of freedom per year. This is quite a lot for modelling a seasonal pattern. Also a unique term for each week could remove the interesting variation around the time of the policy change. Week of conception could be modelled using a spline with between 4 and 6 degrees of freedom per year to remove any trend and seasonal patterns.

No competing interests declared.