Re: Cannabis exposure as an interactive cardiovascular risk factor & accelerant of organismal ageing
posted 09/03/2020
publication BMJ Open
Re: Cannabis exposure as an interactive cardiovascular risk factor and accelerant of organismal ageing: A longitudinal study
Ian A Lane, Clinical & Population Health Researcher University of Massachusetts Medical School
I have read with interest Reece and colleagues (2016) paper, but some questions remain.
First, I think the authors have done a commendable job detailing most of their statistical methodology. However, some things are left to be desired, such as a description of the Statistical Power analysis. Using the Benjamini-Hochberg Procedure (a modified Bonferroni Procedure) on the data listed in their Supplementary materials, one can conclude that the authors' results were indeed statistically significant. What remains to be seen, however, is what the magnitude of these effects were, precisely, and what happened to the Power as these multiple comparisons were assessed, since Power decreases with increasing univariate statistical tests. One might assume Power to be sufficient given the N = 1,553, but these data have been parsed in many different ways, and it would be helpful to know the authors’ anticipated effect sizes and any Power analyses for these comparisons that were conducted prior to the start of the study.
Second, due to issues with boundary conditions and computational modeling, the method used in this paper for the mixed-effects linear model may not be quite right [1, 2]. There is often a misapplication of traditional AIC selection criteria in linear mixed effects (LME) modeling, owing to poor justification for use in longitudinal data analysis, due in part to error variance estimates [2], which is partially how this seemed to have been used here, in the Reece et al. paper.
Finally, and most importantly, the authors never detail the algorithm by which they have computed “vascular age” in this study. Cox proportional hazard estimates for vascular age have been proposed from Framingham Heart Study data, as well as the new “SCORE” method [3]. Should the authors have calculated their own VA score algorithm, this should have had some validity and reliability data published prior to the current article and should therefore be cited in the paper. Should they have used one of the other two methods, this should also be referenced, here, as their findings hinge entirely on ratio relations between VA and their other reference points, whether cannabis use or something else, such as BMI. If the VA calculations are inaccurate or the algorithm inappropriately selected for some reason, all the results drawn in this paper effectively become moot.
In sum, although the study was generally well conducted, I do think, taking into account some of the broader limitations (e.g., with respect to limited female participants in the Cannabis-only group, etc.), as well as the aforementioned methodological questions that remain, the authors’ conclusions are too widely generalized from a fairly unique and non-randomized population with no controls. I will look forward to more work from this group in the future.
1. Müller, S., Scealy, J. L., & Welsh, A. H. (2013). Model selection in linear mixed models. Statistical Science, 28(2), 135-167.
2. Liang, H., Wu, H., & Zou, G. (2008). A note on conditional AIC for linear mixed-effects models. Biometrika, 95(3), 773-778.
3. Cuende, J. I., Cuende, N., & Calaveras-Lagartos, J. (2010). How to calculate vascular age with the SCORE project scales: a new method of cardiovascular risk evaluation. European heart journal, 31(19), 2351-2358.
Conflict of Interest:
None declared.
