Thank you both very much for the tips, Ramona and David. I'll definitely
look into both!

Yes, that's a very good point about health outcomes. Having worked on child
maltreatment-related diagnoses in electronic health records, I am well
aware of the messiness of health data. For large data sets, I guess we have
to work with what's available, despite limitations, and keep pushing for
better quality data.

As for flu shots, maybe you can look at some large longitudinal data sets?
For example, it looks like IPUMS has information on flu shots
https://healthsurveys.ipums.org/.  It might be easier to get flu shot data
for children. Good luck!


On Sat, Dec 3, 2022 at 8:21 PM Ramona Kyabaggu <[log in to unmask]>
wrote:

> Perhaps you can consider structural equation modeling. SEM allows for
> inclusion of both latent and observed variables in your parameterization
> model.
>
> To evaluate causal mediation (with consideration for temporality) you may
> consider longitudinal SEM, and it sounds like you are also interested in
> multi-level factors, so ML SEM for longitudinal data.
>
> Statistical Horizons in Philadelphia offers several short seminar courses
> in SEM. I really enjoyed their ML SEM course.
>
> https://statisticalhorizons.com/public-seminars/
>
>
>
> Best of luck,
>
> Ramona Kyabaggu, PhD(c), CHIM
> Assistant Professor
> Johnson-Shoyama Graduate School of Public Policy, CB 334.7
> University of Regina
> p: 306.585.4548
> e: [log in to unmask]
>
>
>
> Sent from my Galaxy
>
>
> -------- Original message --------
> From: David Zitner <[log in to unmask]>
> Date: 2022-12-03 7:18 p.m. (GMT-06:00)
> To: [log in to unmask]
> Subject: Re: [SDOH] Data analysis methods
>
> Good question,
> The informatics group at Johns Hopkins is doing interesting work in your
> area.
>
> Another difficult problem, in my opinion, is how to measure health
> outcomes. Diagnostic labels do not tell you how sick someone is. Consider
> diabetes or pneumonia. The label indicates why someone might be sick, but
> not how sick they are. Other proxies also present problems of
> interpretation.
>
> Most health record systems don't systematically capture health status and
> changes in health associated with care. Consider that in Canada after years
> of seasonal flu vaccine no one can answer people who ask "Do people who are
> flu shot acceptors live, overall, longer and healthier lives compared with
> flu shot rejectors (adjusting for other variables that influence health
> status and whether or not you routinely accept seasonal flu vaccine)
>
> P.S. I would be delighted if someone could point me to a study that looks
> at all cause mortality and morbidity in seasonal flu shot acceptors or flu
> shot rejectors.
>
> DZ
> ------------------------------
> *From:* Social Determinants of Health <[log in to unmask]> on behalf of Canan
> Karatekin <[log in to unmask]>
> *Sent:* Saturday, December 3, 2022 2:31 PM
> *To:* [log in to unmask] <[log in to unmask]>
> *Subject:* [SDOH] Data analysis methods
>
> CAUTION: The Sender of this email is not from within Dalhousie.
> Hi,
> Does anyone know what statistical methods would be appropriate for
> examining the effects of structural determinants, through a mediator, on
> health outcomes?  All of these would be different units of analysis. For
> example, a political/structural variable at the state level --> state
> budget --> health outcomes in actual people in that state.  And how do
> researchers deal with the time lag issue, that is, how to decide when to
> measure health outcomes in relation to the upstream structural variables,
> and the issue of people moving in and out of states or other geographical
> units?  We just finished a study using state-and year fixed-effects models,
> and I want to follow it up with a mediational model.
>
> I feel like this is a crucial question for examining social/political
> determinants of health & health inequities, but I have no idea what kind of
> analysis would be appropriate. A lot of the studies I have looked at so far
> are more in the form of predictor --> outcome & not in the form of
> predictor --> mediator --> outcome. Or they follow individual people in
> longitudinal studies, but then the focus is on the people as they move
> through time and space, not as much on the political/structural
> determinants in a specific jurisdiction.
>
> I'm wading into areas that are far, far away from what I was trained in
> decades ago, so I'd appreciate any help or any examples of studies
> that have tried to tackle this question.
> Thanks,
> Canan
> --
> Canan Karatekin, Ph.D.
> Associate Professor | Institute of Child Development, 206C | icd.umn.edu
> University of Minnesota | umn.edu
> http://www.cehd.umn.edu/icd/research/KaratekinLab
> <http://www.cehd.umn.edu/icd/research/KaratekinLab>| 612-626-9891
> <http://www.cehd.umn.edu/icd/research/KaratekinLab>
> To leave, manage or join list:
> https://listserv.yorku.ca/cgi-bin/wa?SUBED1=sdoh&A=1
>
> To leave, manage or join list:
> https://listserv.yorku.ca/cgi-bin/wa?SUBED1=sdoh&A=1
>
> To leave, manage or join list:
> https://listserv.yorku.ca/cgi-bin/wa?SUBED1=sdoh&A=1
>
>

To leave, manage or join list: https://listserv.yorku.ca/cgi-bin/wa?SUBED1=sdoh&A=1