Showing posts with label causation model. Show all posts
Showing posts with label causation model. Show all posts

Friday, February 10, 2017

Cannabis and Causation





Cannabis use is highly politicized in the US at this time largely due to legalization rhetoric that has spilled over into scientific research on the topic.  Despite the broad movement to legalize cannabis across the US, only a minority of the population are regular cannabis users.  More widespread use will undoubtedly lead to increased problems associated with wider exposure, especially wider exposure in populations with vulnerabilities to the toxic effects of cannabis.  The toxic effects of interest include addiction and psychosis.  It is common in clinical practice to encounter daily cannabis smokers who stopped using the substance after several years because they started to get panic attacks, paranoia, or both.  The people I see have all moved on to something else, but there are also a substantial number of chronic smokers who are addicted.  That number is about 9% of users, and that is comparable to the amount of people who have problems from drinking alcohol.  Inpatient psychiatrists commonly see people with florid psychotic episodes from smoking significant quantities of cannabis.  They also see repeat admissions from people who are either detoxified or treated for these psychotic episodes, are discharged and smoke more cannabis to the point of a repeat psychotic episode.  The longstanding controversy among people who are not doing the work and just speculating is whether any good observational studies can be done to show that cannabis does cause psychosis or if this is an artifact of observational methodology.  In other words,  could a reverse causality bias exist that makes people who are prone to schizophrenia or psychotic episodes more likely to smoke cannabis.  In my opinion, there have been excellent observational studies showing the association between cannabis use and psychosis, but as long as that is the technology these studies will always contain the old association is not causation qualifier.

A recent paper (1) in Molecular Psychiatry may have just illustrated the causation that psychiatrists have been experiencing firsthand for decades.  The authors use a novel genetic appraoch to look at the issue of causation.  The main assumption of this study is that using specific genotypes as the independent variable rather than observed individuals gets rid of the confounding demographic and environmental variables that could be casual.  They point out that any actual clinical trial looking at the issue of whether cannabis causes psychosis would be unethical, but that a model that looks at whether causation can be established by looking at single nucleotide polymorphisms (SNPs) from a Genome Wide Association Study (GWAS) looking at the any cannabis use phenotype.  They looked at the top 10 SNPs from that data that were used to calculate gene-exposure (SNP-cannabis) estimates.  SNP-risk of schizophrenia exposure estimates were calculated from available data from the Psychiatric Genomics Consortium.  Instrumental variable estimates were made by dividing the risk of schizophrenia/risk of any cannabis use.  The instrument variable analyses were pooled across SNPs and analyzed with fixed effect meta-analysis.  The authors provide a detailed discussion and rationale for their statistical calculation in the full text of the article and supplementary material.  At this point I am going to post their main graphics.  Click on any graphic to enlarge it.

The first graphic looks at prospective observational studies.  The authors were interested in determining whether their genetically based analysis was in the same direction of this meta-analysis. Ever use cannabis use was associated with a 43% increase in schizophrenia or psychosis.

Figure 1. Meta-analysis of prospective observational studies reporting an association between use of cannabis and risk of schizophrenia or related disorders. Meta-analysis uses a random-effects model. Studies are sorted by type of outcome (schizophrenia only vs schizophrenia and related outcomes). Odds ratios (ORs) and 95% confidence intervals (CIs) express the risk of schizophrenia or psychotic symptoms for ever use of cannabis (compared with never use). For additional information on each study, see Supplementary Table S1. Dunedin, Dunedin Multidisciplinary Health & Development Study; ECA, Epidemiologic Catchment Area; EDSP, Early Developmental Stages of Psychopathology Study; NEMESIS, Netherlands Mental Health Survey and Incidence Study; SC, Swedish Cohort.



Figure 2 looks at the Mendelian Randomization analysis of 34,241 cases of schizophrenia and 45,604 cases of ever use cannabis.  This shows a 37% risk of cannabis users versus non-users for schizophrenia/psychosis risk.  The authors did a sensitivity analysis of this same data by removing each SNP from the analysis to calculate a summary causal effect of 1.33 across all 10 SNPs or 1.88 when restricted to 2 functional SNPs.





Figure 4 is included here to illustrate the authors' sensitivity analysis showing a summary casual effect of about 1.37 (red line).



All things considered this may be a compelling story for causation.  I qualify that of course in a couple of domains.  First. there are a lot of statistical models and calculations operating here.  In my experience mapping complex statistical estimates onto the most complex object in the universe has not worked out very well.  My first hand experience was statistical modeling of quantitative EEG and claims that is was predictive of psychiatric diagnosis.  Those compelling calculations published in Science (4) did not pan out at all in the long run.  It will be interesting to see if the authors applications are more widely applied to other SNPs to determine disease causation from other risk factors.  The second potential problem is a slight variation on that theme and that is the overall imprecision of meta-analysis.  The known  approximate prediction/concordance rates of meta-analyses for clinical trials (2.3) suggests that it may not be good predictor of a reproducible result.  The authors themselves suggest that the potential limitations of their study start with the fact that none of the chosen SNPs met conventional genome wide significance thresholds.  The specific dose effect of cannabis could not be investigated in the study.  The age at exposure is may be a developmental variable of interest and that was unknown.  The Mendelian Randomization techniques may have not been powerful enough to detect pleiotropic (one gene affecting more than one trait) effects, but they discuss how an alternate analysis applies in this situation.

The other question I had was about epigenetic effects on this model.  The authors were certainly aware of smoking as a confounding variable.  The known epigenetic effects of nicotine on brain chromatin would seem to cloud SNPs as pure genetic risk factors.  But this is nonetheless one of the more interesting models and concepts I have seen in a while.

They conclude that their study is "the closest approximation to a randomized trial on the effect of ever use of cannabis and risk of schizophrenia" when such a clinical trial is unethical.   That is an interesting take on their method and causation.  Hopefully it will open up the way for other studies of causation using these techniques.  If that is the case, it is a good idea to study this paper and the supplementary material (26 pages) and have a good idea about its difference from observational/association studies.  The supplementary material is also very useful for the calculations used in the study, a Venn diagram of the overlap between the schizophrenia-GWAS group (N=79,845) and the ever-use cannabis GWAS group (N=37,957), and their review methods of the best observational studies of cannabis use and  schizophrenia/psychosis.  



George Dawson, MD, DFAPA



References:

1: Vaucher J, Keating BJ, Lasserre AM, Gan W, Lyall DM, Ward J, Smith DJ, Pell JP, Sattar N, Paré G, Holmes MV. Cannabis use and risk of schizophrenia: a Mendelian randomization study. Mol Psychiatry. 2017 Jan 24. doi: 10.1038/mp.2016.252. [Epub ahead of print] PubMed PMID: 28115737.

2: LeLorier J, Grégoire G, Benhaddad A, Lapierre J, Derderian F. Discrepancies between meta-analyses and subsequent large randomized, controlled trials. N Engl J Med. 1997 Aug 21;337(8):536-42. PubMed PMID: 9262498.

3: Ioannidis JPA, Cappelleri JC, Lau J. Issues in Comparisons Between Meta-analyses and Large Trials. JAMA. 1998;279(14):1089-1093. doi:10.1001/jama.279.14.1089

4:  John ER, Prichep LS, Fridman J, Easton P. Neurometrics: computer-assisted differential diagnosis of brain dysfunctions. Science. 1988 Jan 8;239(4836):162-9. PubMed PMID: 3336779.

"The standard for psychiatric diagnosis and categorization in the United States and Canada is now DSM-III and soon will be DSMIIIR. The categories defined therein have often been criticized as nothing more than a compilation of symptoms. The results obtained with neurometrics have shown that at least the categories studied are much more than arbitrary groupings of symptoms. ............. Validity-the great deficiency of psychiatric nosology - is beginning to emerge and, thus far, to reveal an impressive concordance with biology." p. 169

5: Smith GD, Ebrahim S. Mendelian Randomization: Genetic Variants as Instruments for Strengthening Causal Inference in Observational Studies. In: National Research Council (US) Committee on Advances in Collecting and Utilizing Biological Indicators and Genetic Information in Social Science Surveys; Weinstein M, Vaupel JW, Wachter KW, editors. Biosocial Surveys. Washington (DC): National Academies Press (US); 2008. 16. Available from: https://www.ncbi.nlm.nih.gov/books/NBK62433/

Selected References on Mendelian Randomization



Attributions:  All graphics except my home-made one at the top are from reference 1 per a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.




Supplementary 1:

With today's publicly available genetic technology it is possible for a person to search their own DNA for the SNPs found in this study.  When I do that using a database where my DNA analysis resides I found the following SNPs from this study from chromosomes 15, 4, and 12 respectively.  I have linked them  to the dbSNP database at NLM:

rs4984460

rs7675351

rs2099149

It is interesting to speculate on what it means to have 3/10 genetic markers for schizophrenia/psychosis susceptibility if any cannabis exposure.


Supplementary 2:  Click on my homemade graphic to see how beautiful it is.  Blogger does not do it justice.