Research Article: Should Burden of Disease Estimates Include Cannabis Use as a Risk Factor for Psychosis?

Date Published: September 29, 2009

Publisher: Public Library of Science

Author(s): Louisa Degenhardt, Wayne D. Hall, Michael Lynskey, John McGrath, Jennifer McLaren, Bianca Calabria, Harvey Whiteford, Theo Vos

Abstract: Louise Degenhardt and colleagues discuss the evidence and the debate about whether Global Burden of Disease estimates should include cannabis use as a risk factor for psychosis.

Partial Text: Evidence has accumulated suggesting that regular cannabis use is associated with psychotic symptoms and disorders in the general population [1],[2] and elevated among incident cases of psychosis [3],[4]. In this paper, we present the arguments for, and implications of, considering cannabis use as a risk factor for psychosis in the 2005 Global Burden of Disease (GBD) project.

Governments, policymakers, and funders need information on the comparative population health impact of different diseases and risk factors when making decisions about where to focus policy, services, and research. This field was revolutionised when the World Bank provided estimates using the disability-adjusted life year (DALY) [5]. This measure combined measures of premature mortality (years of life lost [YLL]) and morbidity (years lived with disability [YLD]) in order to estimate GBD. Estimates of burden attributable to various risk factors—“comparative risk assessment” (CRA) exercises [6]—are particularly important because they quantify and allow comparison of the extent to which reduction or removal of exposure to risk factors would reduce disease burden by using a measure of estimated Population Attributable Risks (PAR). The GBD uses fairly standard criteria to evaluate “risk factors”, in line with Bradford Hill’s [7] oft-quoted criteria (Box 1).

In the previous global CRA, cannabis use was not included as a risk factor for any disease because of concerns about the quality of the evidence [8]. In the intervening years there has been a steady increase in the number and quality of research studies that have been conducted exploring the links between cannabis use and psychosis. Overall, these studies indicate that chance is an unlikely explanation of their association [9]–[11]. Recent reviews of prospective general population studies of associations between cannabis use and later psychosis (Table 1) [10],[11] concluded that although control for confounding reduced the size of the association, there was an increased risk of psychotic outcomes in individuals who used cannabis, with the greatest risk among those who used cannabis most frequently.

The principal psychoactive ingredient of cannabis is delta-9-tetrahydrocannabinol (THC), which acts upon a specific cannabinoid receptor (CB1) in the brain [19]. Although historically the dopaminergic system has been considered to play an important role in psychotic disorders [20], there is increasing evidence that the cannabinoid system may also be involved [21]. Some studies have used animal models to explore the impact of THC and related compounds on brain function [21]–[24]. These results are also stimulating new preclinical research aimed at describing neurobiological mechanisms of action linking cannabis and outcomes of interest to schizophrenia [22]. Rodent models are being developed to examine the impact of THC exposure on pathways implicated in clinical schizophrenia [21].

There are several major criticisms of the above evidence. The first concerns the varying outcome measures that different studies have used. These include “psychosis,” psychotic symptoms, and schizophreniform disorders diagnosed using psychiatric interviews and psychiatric case registers.

Many prospective studies share the weakness that they cannot precisely specify the timing of first cannabis use and the onset of psychotic symptoms. Participants have usually been assessed once a year or less often and asked to retrospectively report their cannabis use during the past year(s). This assessment has often been in terms of the total number of times cannabis was used, or the number of times on average that cannabis was used each week or month. Nonetheless, there are multiple prospective studies of representative samples of the general population, all of which show that cannabis use at one point in time is associated with psychotic symptoms at a later one, even after using a range of controls for confounding and various statistical approaches to analysis.

Publication bias is a potentially more serious concern: If negative results have been withheld from publication then the consistent positive results would be far less impressive than they seem from the published systematic reviews [52]. This possibility was investigated by the authors of one systematic review who surveyed researchers in the area asking about any studies with negative results that had not been published [10]. They concluded that this was not a serious issue in this instance.

The most difficult task in drawing causal inferences from observational studies is excluding the possibility that the relationship between cannabis use and psychosis is due to other uncontrolled factors (e.g., other drug use, genetic predisposition to develop schizophrenia and use cannabis, or self-medication). This has led some to object to calculation of estimates of population attributable risk (PAR) because the adjusted estimates are modest (typically around 2–3), and so open to the alternative explanation of uncontrolled confounding. For example, some have suggested that the propensity to take risks and engage in socially disapproved behaviour may be a common cause of cannabis use and psychotic symptoms [53]. Fergusson et al. attempted to address these criticisms by using fixed effects regression models to adjust for all unmeasured confounders [31].

Calculation of a PAR is important to place the magnitude of the cannabis and psychosis association in a population health context. Arsenault et al. [11] concluded that elimination of all cannabis use would reduce the incidence of schizophrenia in the United Kingdom by approximately 8%, assuming that the relationship was “causal” in the sense that schizophrenia would not have occurred in the absence of cannabis use; Zammit et al. [35] similarly estimated that 13% of schizophrenia cases in Sweden were attributable to cannabis use.

Some commentators may well argue that it is premature to conclude that the relationships between cannabis use and psychosis are causal, which raises the question of what the standard of proof should be causal inference. Some may argue for “proof beyond reasonable doubt,” the standard implicitly used in the last iteration of the GBD [8]. It is rare, however, to meet this standard of proof for noncommunicable diseases other than smoking-related diseases. What has changed since the last iteration of the GBD? The broad approach to all risk factors has been to set the standard of proof at “more likely than not,” rather than “beyond reasonable doubt.” If the latter was the standard of proof, then no adverse health consequences of cannabis would be considered apart from dependence.

Making estimates of the proportion of psychoses attributable to cannabis will in effect provide worst case estimates of the burden of disease (BoD) attributable to cannabis if the critics are correct that uncontrolled confounding explains the relationships between cannabis use and psychosis. In Australia, for example, cannabis use was included as a risk factor in the Australian BoD study, assuming causal relationships for cannabis dependence, psychosis, suicide, and car crashes [64]. Even after assuming that these relationships were causal, cannabis was not a major contributor to disease burden in Australia, accounting for 0.2% of all disease burden, which amounted to 10% of the total burden attributable to all illicit drugs [65]. These estimates are important for public policy purposes, because failure to make them allows untested estimates to be offered in public policy debate.

Source:

http://doi.org/10.1371/journal.pmed.1000133

 

Leave a Reply

Your email address will not be published.