The current study examined whether past-month cannabis use has a positive effect on the average amount of leisure time physical activity in a large and representative sample of the Belgian population. Additionally, our descriptive analyses provide insights into the defining characteristics of both the current users and non-users. Two complementary analyses revealed no evidence of a positive relation between past-month cannabis use and leisure-time physical activity after adjusting for potential confounders and the complex survey design.

The observation of no effect of past-month cannabis use on physical activity challenges a set of previously discussed recent studies reporting potential benefits (YorkWilliams 2019; Gillman et. al 2015; Ong et. al 2021; Gibson et al. 2023; Korn et al. 2018). The current study therefore serves as a cautionary note against premature conclusions, and underscores the need for further investigation of this important association. Moreover, the absence of an effect in the present study suggests that current cannabis users are not necessarily at a disadvantage in terms of physical activity levels (but note that absence of evidence does not equal evidence of absence). Furthermore, the suggested lack of association between cannabis use and physical activity can also be informative for other factors previously linked to cannabis use (e.g. obesity and cannabis consumption; Fearby et al. 2022). Finally, the propensity score model revealed that the probability of reporting past-month cannabis use decreased with age, was higher for men than women, and was higher for people with a generalized anxiety disorder, in line with previous findings (Trost et al. 2002; Crippa et al. 2009; Hayatbakhsh et al. 2007).

Our sample also better covers the effects of cannabis as it is used in daily life, compared with studies focusing on unrealistically smaller doses or recruiting subjects with a likely positive attitude towards cannabis use. Moreover, the use of the HIS dataset enabled us to easily adjust for other important health-related variables. Finally, by combining information from different HIS editions, we were able to study a relatively large number of cannabis users (a group that is usually relatively small in single samples).

At the methodological level, in the context of regression adjustment, the use of propensity scores (Hernán and Robins 2020) adds several important advantages: firstly, it allows for evaluation of the similarity of users and non-users in baseline covariates, a task that would otherwise be virtually impossible given the dimensionality of the data. Secondly, matching on the propensity scores provided a simpler interpretation in terms of the effect of cannabis in the users, while preventing extrapolation when no similar users and non-users (in terms of their propensity to use cannabis) can be found. Finally, the use of propensity scores, together with baseline covariates has been shown to make tests of the causal null-hypothesis more robust against potential misspecification of the regression model, by still delivering valid results in that case so long as the propensity score model is correctly specified (Vansteelandt and Daniel 2014).

Besides our large and representative sample and informative analyses, there are some considerations to keep in mind when interpreting the current findings. Firstly, although we specifically selected the variable of past-month cannabis use instead of past-year or even lifetime-cannabis use in order to better target regular users, this variable does not perfectly capture this target group. One can imagine that some of the respondents reporting past-month use were in fact first time users and might have not used thereafter, or that some users might have provided false information regarding their use, due to the illicit nature of cannabis in Belgium (though measures were taken to limit this influence, e.g., by having participants provide this information anonymously in a sealed envelope). Furthermore, no information was available on how cannabis was consumed in relation with leisure-time physical activity (before, during or after exercise), as was an important aspect in previous studies (YorkWilliams 2019), nor were we able to account for frequency/intensity of cannabis use or the consumption of other drugs. Finally, the current dataset did not have information regarding the precise cannabinoids that were used in the past month.

Similarly, while the variable leisure time physical activity provides a proxy for an individual’s general level of physical exercise, the absence of an association with past-month cannabis use may also be due to this variable not adequately capturing the type, intensity, or amount of participants’ physical activity. Akin to the potential effect of social desirability in reporting past-month cannabis use, respondents may have overreported the amount of physical exercise they got on a weekly basis. Alternative variables in the HIS dataset were considered (e.g., the binary indicators of meeting the World Health Organization (WHO) recommendations on health enhancing physical exercise), but were not selected due to this information being recorded for our selected HIS editions.

Finally, even though we selected important control variables, the possibility remains that some potential confounders were not available or included in the models. Importantly, the cross-sectional nature of the study additionally complicates a valid confounding adjustment as it is difficult to disentangle causes from effects of cannabis use. Moreover, we note the possible bias resulting from a complete case analysis (i.e., omitting missing data, see Supplementary Material 3 for a discussion) and the added value of the more efficient (but more complex) method of multiple imputation (Li et al. 2015). Finally, though we conducted a sensitivity analysis, it remains hard to formally verify the adequacy of logistic regression models, especially with complex survey designs (as these designs often violate core assumptions of standard logistic regression).

It is equally important to evaluate our findings in light of both the time and location in which these data were collected. Data collection for the included HIS editions dates back to as early as 2001. Since then, societal attitudes and actual use of cannabis have potentially changed. A recent study investigating the attitudes towards medicinal cannabis use in the general population of Belgium, for example, observed that a large portion of the respondents would be open to trying medical cannabis should this be needed (Pav et al. 2024). These generally high levels of acceptance and use, specifically in the Flemish region, were also reflected in the relatively large and increasing number of positive cannabis samples examined by the doping control laboratory in Ghent (Van Eenoo and Delbeke 2003). Taken together with the increasing trends in average potency and consumption frequency (Gisle and Drieskens 2018), it seems reasonable to expect differences in the association between cannabis use and physical activity over time. While data from the 2023 HIS edition were not available yet, it would be an interesting follow-up to repeat a similar analysis with these most recent data.

Another important contrast with previous studies is that our sample targeted the Belgian population as opposed to the majority of studies on the current effect taking place in the United States of America. As opposed to Belgium, a large number of states have legalized recreational and/or medical cannabis use (Cheng et al. 2003). Consequently, cannabis users in studies conducted in these states may show important differences with the user population queried in Belgium, both in frequency and accurate reporting of cannabis consumption.

Considering that cannabis use in the current and previous studies was mainly considered for recreational purposes, an interesting outstanding question is whether the association between cannabis use and physical exercise depends on the motivation for consumption (i.e., recreational versus medical). In the current study, no distinction could be made between these two types of users. Potentially, cannabis use for medical purposes could show a stronger (albeit negative) association with physical exercise, considering the likely presence of other health-related issues in this population. Follow-up studies with more information on the type of cannabis use will be important to differentiate the effect of cannabis use and the effect of general health status.

Additionally, despite the lack of evidence that people who use cannabis are at a disadvantage regarding physical activity levels, we argue that it is still crucial for public health agencies to continue promoting adequate levels of physical exercise, especially in the population of cannabis users. Despite the fact that the effects of cannabis use on mental and physical health are not yet fully-understood, increased levels of physical activity could potentially offset some of the commonly suggested adverse effects (predominantly of THC), such as on respiratory health (Pinckard et al. 2019), cardiovascular disease (Duncan et al. 2021), or mental health (Smith and Merwin 2020).

Taken together, our findings highlight possible important directions for future research. Firstly, using variables that better capture the constructs of interest (e.g., better/objective measures of cannabis use or physical activity) could help to draw more firm conclusions on the proposed association. A worthwhile follow-up study in this regard would include data from fewer waves for a more sensitive measure of physical activity and/or cannabis use. Secondly, as the possibility of unmeasured confounders remains an issue for observational studies, it will be of interest to revisit the considered scientific question based on longitudinal data and to explore how other factors may contribute to this important association (e.g., the role of exercise enjoyment, sleep quality, bodily pain, or the presence of cannabis use disorder). Finally, future studies, both on the current topic and more generally, could benefit from the incorporation of propensity based methods as applied here.