II-Smoking Cessation: Selection Effects

This is the third in a series of posts looking more closely at the methods used to quantify the benefits of smoking cessation using this paper as an example; earlier posts:

A RCT of smoking cessation couldn’t and wouldn’t be done, so some sort of observational data must be used to quantify the benefits of smoking cessation. The choice of the Cancer Prevention Study II (CPS-II) presented benefits and costs:

  • Biggest benefit: large number of person-years of follow up (7.2 million for females, 4.3 million for males)
  • Biggest cost: selection effects that its use introduced

By selection, I mean the participants in CPS-II differed from the overall population:

  • They were whiter (93% of CPS-II v. 80% 1990 Census) and they had higher levels of education (30% college degree in CPS-II v. 9% U.S. adults).

What does this mean for our estimates of the relative risk of death by smoking status? First, the direction of any bias due to selection is ambiguous, unlike measurement error in smoking status that provides a predictably conservative bias toward identifying no effect. Using racial differences as an example, two stories seem plausible. First, whites could receive more benefit from cessation because they have a longer background life expectancy than minorities. On the other hand, non whites could receive more benefit from cessation at later ages because of healthy survivor effects. I am unsure what direction the net effect that selection bias of this sort may go.

In the end, 11.5 million person years of follow up covers a multitude of sins, and we were able to control for race (white v. non white) and education in the estimation of the relative risk of mortality calculations that underlie our models. However, even though we had large enough cell counts to control for these variables, there is still worry that the underlying population is different from that used to obtain the relative risk estimates.

Other cohort studies available were even whiter and based on one geographic locale (Framingham), or were based on the experience of General Practitioners in the U.K. (British Doctor Study). It would be great to have a long term follow up database that was representative of the overall population in terms of race and education to update this study, particularly given the increase in the Hispanic population of the U.S. The Health and Retirement Study, which is approaching 20 years of follow up could be an option for updating, and the tradeoff would be smaller cell sizes vs. being more representative. Are there other databases that should be considered?

 

 

About Don Taylor
Professor of Public Policy at Duke University (with appointments in Business, Nursing, Community and Family Medicine, and the Duke Clinical Research Institute). I am one of the founding faculty of the Margolis Center for Health Policy, and currently serve as Chair of Duke's University Priorities Committee (UPC). My research focuses on improving care for persons who are dying, and I am co-PI of a CMMI award in Community Based Palliative Care. I teach both undergrads and grad students at Duke. On twitter @donaldhtaylorjr

5 Responses to II-Smoking Cessation: Selection Effects

  1. Jonathan says:

    Don,
    I would tell you to speak with Dean Lillard of Cornell University. Last I spoke with him a while back he said he was working on something like this for the international context. It seemed like a very difficult venture but he was working on it. I’m not sure of the status of it right now. Though he might have a good idea of domestic datasets.

  2. Chris says:

    Why not use propensity scores via matching or weighting to control for the other potential selection biases? See, Schneeweiss et al, http://www.ncbi.nlm.nih.gov/pubmed/19487948 or Lunceford and Davidian, http://onlinelibrary.wiley.com/doi/10.1002/sim.1903/abstract .

    • Don Taylor says:

      @Chris
      I have used propensity score matching to deal with selection around hospice v. normal care to identify potential cost savings. Will check out these papers. Also need to reach the book I posted on After Tobacco, but still have not done so. thx

    • Jonathan says:

      This is just a personal preference but I just don’t like propensity scores. I think that they don’t really tell you anything in terms of where the actual variation comes from. But maybe its because I am behind the times and have never worked with them before. Do you guys feel you capture the actual intended variation. I am not slamming the idea I just want thoughts on it. My most recent Public Finance course slammed them for the most part because they take you one step away from the individual variation we pursue. Thanks and note I am not trying to be malicious here!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: