STRENGTHS AND LIMITATIONS OF THIS STUDY

The impact of the partial legalisation of cannabis will be examined using a large sample size and a long observation interval of two years.

Including data from a control group (Austria) will strengthen the internal validity.

Reporting biases in the context of driving under the influence of cannabis (DUIC) will be addressed by using different assessment approaches, including direct and indirect questioning (crosswise model) as well as official statistics on drug-related motor vehicle accidents.

The unknown prevalence of DUIC limits confidence in the sample size calculation and introduces a risk of insufficient sample size in the control group.

Introduction

Germany is set to liberalise its cannabis policy in 2024, allowing adults to possess and cultivate cannabis for recreational purposes.1 As the proposed policy change does not involve commercial distribution of cannabis, it is not comparable to the legalisation models implemented in most US states, Canada or even Uruguay. Technically, the model can be classified as liberal decriminalisation or restricted legalisation. Henceforth, it is termed ‘partial legalisation’ as per the federal government.

According to a 2021 general population survey (Epidemiological Survey on Addiction (ESA)), 8.8% of individuals aged 18–64 years in Germany reported cannabis use within the past 12 months.2 Specifically, 3.5% reported regular (at least monthly) use and 1.5% heavy (daily or near-daily) use.3 With cannabis use prevalence in Germany having nearly doubled since 20124 further increases may be expected and this trend may be facilitated by the partial legalisation of cannabis,5 although probably not as significantly as in countries with less restrictive legalisation models and commercial sales.

In Canada, the most important driver of the cannabis-attributable disease burden is a cannabis use disorder (CUD),6 which affects a sizeable share of people using cannabis.7 Apart from CUD, traffic injuries resulting from ‘driving under the influence of cannabis’ (DUIC) is another major driver of the health burden caused by cannabis use.6 Driving a vehicle immediately after cannabis use is associated with a low to medium increased risk of crashes8 due to impaired cognitive abilities.9–11 However, it should be noted that the evidence is found to be inconsistent,12 possibly due to methodological biases, compensatory mechanisms and development of tolerance among some drivers.12–14 Both CUD and (self-reported) DUIC are especially prevalent among those using cannabis daily or almost daily.15

Liberalising cannabis policies can have negative public health consequences by increasing DUIC and causing traffic events. A systematic review of studies conducted primarily in the USA and Canada suggests that the decriminalisation and legalisation of cannabis for recreational purposes is often associated with higher rates of positive tests for tetrahydrocannabinol (THC) in drivers and increases in injuries and fatalities from motor vehicle accidents (MVAs).16 In line with two additional reviews, results for self-reported DUIC were less consistent compared with toxicological findings,5 16 17 highlighting the possible role of reporting biases in the context of DUIC. Most evidence on the impact of cannabis policy changes on traffic safety comes from North America, particularly the USA and Canada. It remains unclear how applicable these experiences are to other regions. Given the lower reliance on cars for transportation in many countries, where public transport systems play a more significant role, the effects of the partial legalisation of cannabis may differ from those observed in North America.

In Germany, data on MVA with personal injury involving drugs other than alcohol are recorded. Between the years 2011 and 2021, the incidence of such accidents has increased by 73.2%.18 Although the data are not broken down by the specific substances, local studies19–21 lead to the belief that over 50% of these accidents involve a driver who has used cannabis (alone or mixed). However, it is unclear whether those who tested positive for cannabis (or other drugs) were acutely impaired or simply had metabolic residues in their blood that were not relevant to the risk of an MVA. The most recent assessment of self-reported DUIC in the German population was in 2011, with an estimated prevalence of 13.1% among weekly users.22 Like many others European countries, self-reported DUIC is not routinely measured in general population surveys, resulting in a major research gap for the continent. In this study, we aim to evaluate the proposed policy changes to cannabis regulation in Germany. Specifically, we seek to investigate the impact of the partial legalisation of cannabis on (1) Cannabis use prevalence and (2) Traffic safety measured via self-reported DUIC and all-cause MVA data.

Hypotheses

Based on the available literature, we postulate the following hypotheses:

H1: The partial legalisation of cannabis in Germany is followed by a higher increase in cannabis use prevalence compared with the control region Austria. This effect is more pronounced 18 months versus 6 months after the partial legislation came into force.

H2: The partial legalisation of cannabis in Germany is not followed by an increase in (self-reported) DUIC prevalence among at least monthly cannabis users.

H3: The partial legalisation of cannabis in Germany is followed by an increase in all-cause MVA.

Methods and analysis

To examine the impact of partial legalisation of cannabis on cannabis use (H1) and DUIC (H2), we will collect data on both variables before (data collection already completed) and after policy changes in Germany and in an Austrian control group, enabling a pre-post comparison. To test the impact on MVA (H3), we will conduct interrupted time series analyses on nationwide data collected from statistical offices.

Given the sensitivity of DUIC, likely leading to under-reporting on traditional self-report items, we will employ additional assessment methods, including an indirect questioning approach with the crosswise model (CWM;23 see below for explanation) technique and analyse official statistics on drug-related MVA involving personal injuries.

Study design

We will employ a quasi-experimental research design with independent, repeated cross-sectional samples at three measurement points: baseline data have been collected in November/December 2023 (before partial legalisation, t0) and follow-up measurements will be taken 6 months (t1, November/December 2024) and 18 months (t2, November/December 2025) after partial legalisation. Survey data will be collected from residents in Germany (intervention group) as well in Austria (control group: no anticipated change in cannabis-related legislation) at each measurement point. A subset of individuals using cannabis at least monthly (hereafter referred to as ‘regular’ users) will be extracted from the German sample pool to further investigate DUIC.

Adopting a difference-in-difference framework, we will include Austria as a control group to improve internal validity by addressing potential alternative explanations. The Austrian control group will primarily be used to investigate H1 (change of cannabis use) but will also constitute one of several information sources to test H2 (change of DUIC).

Austria is considered an adequate control group as 12-month cannabis use prevalence among those aged 15–64 years has increased in both Austria (2008: 3.5%; 2015: 6.4%; 2020: 6.3%) and Germany (2009: 4.8%; 2012: 4.5%; 2015: 6.1%; 2018: 7.1%; 2021: 8.8%) in recent years.24

Data collection

All samples are sourced from one or multiple actively recruited International Organization for Standardization (ISO)-certified online access panels that recruit participants through both online and offline sources, verify identities with bank details and have measures in place for quality control. The entire data collection will be coordinated by a market research institute, namely INFO GmbH. Specifically, persons registered with existing access panels will be invited by the panel providers to participate in this online retrospective survey via computer-assisted web interviewing (CAWI). Participants will receive reimbursement from the panel provider. In order to guarantee repeated independent cross-sectional samples (as opposed to longitudinal ones), participants are excluded from subsequent waves on completion of the survey, using technical measures.

Data collection procedures will vary slightly between Germany and Austria (figure 1). In Germany, we will screen about 15 000 persons (basic sample, duration of approximately 5 min) at each measurement point for demographic information and substance use. The complete questionnaire (duration of 20–25 additional minutes) will be presented to regular cannabis users only (target sample, expected n=500; see sample size calculations). The final sample size of the basic sample will depend on the target sample, that is, inclusion of at least 500 people using cannabis regularly.

Figure 1

Sample composition at each of the three measurement points. DUIC, driving under the influence of cannabis.

In Austria, a much shorter questionnaire (duration of approximately 10 min) will be applied to 2000 persons at each measurement point, without recruiting a target sample. In both countries, we will (1) Measure the prevalence of substance use in a general population sample and (2) Measure DUIC among regular cannabis users. In the German target sample DUIC will be measured more comprehensively through indirect assessment via CWM,23 as well as potential factors influencing DUIC such as personality, beliefs and social norms. Additionally, more details characterising cannabis use will be collected in the German target sample. The complete questionnaires for the German and Austrian samples are accessible on OSF (https://osf.io/72nyf/) and as online supplemental material S1 and S2, along with an overview of all measured variables and their corresponding subsamples (online supplemental material S3). The market research institute responsible for data collection will use extensive verification procedures (eg, standardised control questions, verification of response times, and variance manually and with AI) to exclude anomalous data sets.

The basic samples will be quota-sampled based on age, gender, education and federal state, reflecting the demographic composition of residents aged 18–64 years in Germany or Austria. During data collection, highly selective sampling can be minimised by continuously monitoring cell coverage. We anticipate that poststratification sample weights will remain below 10. The German target and basic sample will undergo joint weighting, where weights calculated for strata in the basic sample will be applied to corresponding strata in the target sample. A description of the weighting procedure is provided in the online supplemental material S4.

With mandatory answers for our key survey questions on DUIC, cannabis use and sociodemographic variables, we expect no missing data issues. However, participant dropout rates, possibly influenced by stigma and legal consequences surrounding cannabis use and DUIC, especially prior to partial legalisation, will be analysed and reported in subsequent publications.

Sample size calculationsGerman sample

At each measurement point, we aim for a target sample of at least n=500 regular cannabis users. This sample size is derived based on the following rationale: Previous studies investigating self-reported DUIC before and after legalisation reported effect sizes in large ranges of inconsistent direction, but studies comparing toxicological results yielded an average OR of 1.725–27 (adjusted for outliers). The current 12-month prevalence of DUIC in Germany is presently unknown but assumed to be at about 10% among regular users. That estimate is based on a Canadian study reporting a 3-month prevalence of DUIC of 20% among regular users15 and the German study from 2011 reporting a 1-month prevalence in Germany of 13% among weekly users.22 However, we adjust these proportions downward, considering a more widespread cannabis use in Canada, a larger public transportation infrastructure in Germany facilitating easier DUIC avoidance, the likelihood of a higher undisclosed rate in Germany due to its legal status, and the assessment among younger and more frequent cannabis users in the German study associated with a higher risk for DUIC.22 With a power of 80% and error tolerance of alpha=5%, n=4, 80 persons at each measurement point would be required to detect a prevalence increase in a magnitude of OR=1.7. We aim for 500 persons at each measurement point, allowing to detect changes of at least OR=1.62.

In order to achieve the target sample size of 500 persons, a sample of 14 300 individuals is required (estimated regular cannabis use prevalence: 3.5%3). To account for the uncertainty and ensure an adequate target sample size, we aim to screen 15 000 individuals at each measurement timepoint. Assuming an 80% power, an error tolerance of alpha=5%, and a directional hypothesis, this sample size could detect a 10% increase (OR=1.10) in 12-month cannabis use prevalence. This OR is slightly below the typically reported changes associated with legalisation,5 suggesting that the sample size is well suited for detecting even smaller changes.

Austrian sample

In Austria, we will collect a sample size of 2000 individuals at each timepoint. Within this sample, assuming a 6.3% 12-month prevalence,28 around 120 users are expected. Assuming a similar regular cannabis use proportion of approximately 40%3 among these users in Germany and Austria, we anticipate approximately 50 regular users.

Crosswise model

The CWM is an indirect questioning technique designed to address potential under-reporting of sensitive attributes, such as illegal behaviour.23 29 It has been validated in numerous studies (for an overview, see Hoffmann et al30). The CWM ensures that individual responses remain confidential, encouraging respondents to provide honest answers. To achieve this, participants are presented with two questions simultaneously. One question pertains to the sensitive attribute (DUIC) with an unknown prevalence π (‘Have you driven a motor vehicle within 2 hours after consuming cannabis in the past 12 months?’). The other question pertains to a neutral attribute with a known prevalence r, and is used for ‘randomization’ (‘Is your mother’s birthday in May?’). By indicating only whether the two questions have the same answer (‘yes/yes’ or ‘no/no’) or different answers (‘yes/no’ or ‘no/yes’), the individual participants’ answers are concealed. The prevalence of the neutral question r is known (approximately 8% of people are born in May31), allowing for the determination of the prevalence π of DUIC in the sample.23 With r=0.08 and n=500, we can estimate a DUIC prevalence of at least 3.3% (CI 0.06% to 6.54%).

Measurements of outcomesCannabis use

The primary outcome variable is cannabis use in the past 12 months (0=never; 1=less than monthly/monthly/weekly/daily). To assess cannabis use, we will present a set of various consumer goods (fish, fruit, alcohol, tobacco) and ask respondents to specify the frequency of their use (for the complete item, see online supplemental material S1). This approach serves to conceal the study purpose. Thus, we can minimise inaccurate reporting of cannabis use to qualify for the target sample and receive participation incentives.

Additionally, we will examine heavy cannabis use in the past 12 months (0=less frequent than (almost) daily; 1 = (almost) daily).

Driving under the influence of cannabis

To address the potential underestimation of DUIC prevalence in subjective reports influenced by societal stigma or fear of legal consequences, we will use a combination of direct and indirect questioning using the CWM.

Direct measurement

The direct outcome variable is defined as having driven a vehicle within 2 hours after using cannabis in the past 12 months (0 = ‘No, never’ or ‘Yes, more than 12 months ago’; 1 = ‘Yes, within the past 30 days’ or ‘Yes, within the past 12 months’).

Indirect measurement (CWM)

In addition to directly measuring DUIC, we will approach this behaviour using CWM assessing both 30-day DUIC prevalence and 12-month DUIC prevalence in two items. In both items, we will couple a sensitive question with a neutral question (table 1) and participants will be asked to provide simultaneous responses to both questions. Prior to DUIC assessment, we will ask participants whether they know their parents’ birth months.

Table 1

DUIC items using the crosswise model

Demographic and other covariates

Data on sociodemographic characteristics including age, gender (male, female, diverse), and highest level of education will be collected. The education level will be categorised into a three-level variable based on the International Standard Classification of Education (ISCED).32 Our translation of German and Austrian education levels into ISCED categories is described in the online supplemental material S5.

Covariates such as different aspects of stigma surrounding cannabis use, other substance use, personality traits and driving behaviours will be measured in the German target sample.

Additional measures

We will collect more detailed information on various aspects of cannabis use and DUIC, including medical use, specific cannabis products used, typical waiting time between cannabis use and driving, and DUIC in combination with other substances. These variables may contain valuable information for supplementary analyses. For a full overview of all measured variables please refer to the online supplemental material S3.

Data analysis planHypothesis 1: prevalence of cannabis use

To examine differences in the prevalence of cannabis use between timepoints (before partial legalisation (t0), 6 months (t1) and 18 months (t2) after partial legalisation) we will conduct a logistic regression with cannabis use in the past 12 months (yes=1, no=0) as the binary outcome variable. Timepoints, country and their interaction are specified as predictors. If the interaction term t2 × Germany (not necessarily t1 × Germany) is a statistically significant predictor with an exponentiated coefficient (OR) larger than 1, this would support our hypothesis that the prevalence of cannabis use in Germany has increased due to partial legalisation, particularly in the long term (t2). However, if only the interaction term t1 × Germany is significant, this would not support our hypothesis, as it suggests only a short-term increase that diminishes over time, which does not align with the intended implication of our hypothesis.

Hypothesis 2: prevalence of DUICDirect measurement

To analyse the effect of the partial legalisation on DUIC, we ideally aim to include Austria as a control group. However, it is possible that the Austrian sample size might be too small to provide a sufficient number of DUIC cases for a robust comparison with Germany (see sample size calculations). In this case, we will focus solely on temporal changes, comparing t1 vs t0 as well as t2 vs t0 through logistic regression, with DUIC in the past 12 months (yes=1, no=0) as the outcome variable. Timepoints are specified as predictors and the analysis will only include regular cannabis users, excluding pure medical users with a prescription. If both t1 vs t0 and t2 vs t0 predictors fail to demonstrate statistical significance, it would support our hypothesis of no difference in self-reported DUIC prevalence among regular cannabis users after partial legalisation (alpha level of 0.05, two-tailed). We would deem a doubling in self-reported DUIC (ORs ≥2.0 for t1 vs t0 and t2 vs t0) to be a relevant increase with potential implications for traffic safety. This magnitude of change is unlikely solely attributed to alterations in response behaviour after partial legalisation (ie, increased reporting due to reduced social stigma). In acknowledgement of the subjectivity inherent in setting this threshold, and in the absence of relevant non-inferiority margins that can be derived from scientific literature, we refrain from conducting an equivalence test against it to maintain interpretational flexibility. Instead, we add a Bayesian approach of the logistic regression to quantify the strength of evidence the data provides in favour of the no-difference hypothesis. Normal priors were specified for the intercept (β0) and slope parameters (β1 for t1 vs t0 and β2 for t2 vs t0) on a logit scale with large variance to account for vague knowledge. Assuming an estimated prevalence of DUIC of 10% at t0, we set β0 ∼ N (−2.2, 1.12)—thus, the prevalence can range between 1% and 50%. β2 ∼ N (0.53, 0.2652) is assumed on the basis of toxicological results (OR=1.7,25–27 adjusted for outliers, see above), allowing for an increase from 0% to 121%. β1 ∼ N (0.26, 0.132) is assumed, corresponding to an OR=1.3 allowing for an increase of 0%–48%, taking into account potentially weaker and uncertain short-term effects, as prior research has predominantly focused on long-term effects. Posterior distributions will be estimated using Markov Chain Monte Carlo approximation. Sensitivity analyses will be conducted to assess the robustness of results to the choice of priors.

Given a sufficient number of DUIC cases in Austria, we will add country (Austria vs Germany) and its interaction with timepoints as predictor variables in our model, examining the interaction effects t1 × country and t2 × country for statistical significance (as for testing H1).

The potential absence of an Austrian control group may raise concerns about interpreting the impact of the partial legalisation on DUIC, particularly regarding time-varying confounders. Therefore, we will investigate additional models that incorporate these confounders, selected after the t0 survey, based on the strength of correlation with DUIC.

Additionally, we will report data on (risky) alcohol use, driving under the influence of alcohol and DUIC combined with other drugs as cannabis is almost always found in combination with other drugs in driving under the influence.33

Indirect measurement

We will present only descriptive reports along with CIs for DUIC prevalence per timepoint measured indirectly via CWM. This decision is necessitated by the substantial variance in estimates, which precludes reliable statistical inferences. Estimates for DUIC prevalence and the corresponding CIs will be calculated following the approach outlined by Yu et al.23

Hypothesis 3: rate of MVA

An interrupted time series design will be used to evaluate the impact of the partial legalisation on the rate of fatalities from MVA per 100 000 population. The focus on fatalities resulting from MVA is motivated by a large body of observational studies suggestive of a link between cannabis policy liberalisation and increased MVA fatalities.16 Three alternative outcomes will be considered in sensitivity analyses: (1) Rate of persons with non-fatal injuries (light or severe) from MVA per 100 000 population; (2) Rate of MVA crashes with property damages per 100 000; (3) Rate of drug-involved MVA crashes. The required data are available on a monthly basis from the Federal Statistical Office of Germany via its online database GENESIS-Online (Joint New Statistical Information System-Online) using the following codes: for main outcome and for sensitivity analyses (A) 46 241–0008; for sensitivity analyses (B) 46 241–0002; for sensitivity analyses (C) 46 241–0010.

The data are planned to encompass the timeframe between 1 January 2011 and 30 March 2026, comprising 159 months prior to the partial legalisation and 24 months subsequent to the law coming into force. We will consult experts in the field and screen the literature for any possible interventions enacted during the study period that could confound the analyses (eg, increased fines for speeding).

To test the hypothesis, we will perform generalised additive mixed models and the effect of partial legalisation will be assessed via step (binary variable: 0=before, 1=after partial legalisation) and slope (continuous variable: 0=before, 1 … n = time points between implementation of law and end of time series) changes, while accounting for possible seasonality in the data. If other effect types (eg, lagged effects, temporary effects; see Lopez et al34) can be discerned, these will be tested as well within the constraints of the limited number of postintervention observations.

The best-fitting and simplest model will be selected from a range of candidate models. As the COVID-19 pandemic may have affected the outcomes of interest through decreased mobility and increased home office, we will consider including parameters in the model that can describe the variation in the years 2020 to 2022. This could be simple dummy variables or alternatively a measure of public containment measures.35

We seek to perform controlled interrupted time series analyses36 for the primary outcome (MVA fatalities) using comparable data from neighbouring countries if data availability allows. We will ensure that the control group has similar prelaw trends by visual inspection and by testing whether the secular trend varies by country (interaction of time and country). In case of parallel preintervention trends, the generalised additive mixed models will be extended by including data from multiple countries while each is separated by dummy variables. Changes in the outcome will be tested by an interaction between the step/slope intervention variable and the dummy variable indicative of Germany.

Additional analyses

Considering the elevated risk of DUIC15 and other cannabis-related health risks among heavy users, we will examine changes in the frequency of use after partial legalisation among cannabis users. This analysis mirrors the approach outlined for H1, with a binary outcome variable indicating (almost) daily cannabis use (vs less frequent use).

Exploratory subgroup analyses across demographic categories (age groups, genders, education levels) will investigate potential varying effects of partial legalisation, using the same methods outlined above (logistic regressions applied to both past-year use and heavy use prevalence). A false discovery rate37 at level 0.05 will be implemented to correct for multiple testing.

As an additional DUIC indicator, we will explore changes in waiting time between cannabis use and driving. Reporting proportions of <12 hours (risk group) or ≥12 hours (safe group), we will analyse the risk group’s average waiting time using linear regression, modelling the partial legalisation effect.

Patient and public involvement

Neither patients nor the public were involved in the conceptualisation or conduct of the proposed study. Furthermore, there are no plans to disseminate the results to study participants, as we do not have their contact information.

Ethics and disseminationEthical and safety considerations

This study has received approval from the Local Psychological Ethics Committee of the Centre for Psychosocial Medicine (Hamburg, Germany; reference number: 0686).

A market research institute will set up the online survey and prepare the corresponding URLs for CAWI survey participation. These URLs will be sent to potential respondents by the panel providers, not by the market research institute. In this way, all personal information (email, name, bank account) required to register with a panel will remain confidential with the panel provider, while all survey data will be collected by the market research institute. This procedure ensures that personal data from panel registration cannot be linked to survey responses, while still allowing to incentivise participants for study participation (via the panel provider). Prior to participation, all participants will be informed of the voluntary nature of their participation, data protection legislation and the option to request the deletion of their data (online supplemental material S1/S2).

Dissemination plan

We plan to publish the study results in academic peer-reviewed journals and disseminate them by presentations at national as well as international conferences. Additionally, a plain language will be made available to the public and distributed via social media outlets.

The study findings are expected to have significant political implications. To our knowledge, this research project is unique in its aim to investigate the impact of the cannabis law reforms in Germany. Although other routinely conducted repeated cross-sectional general population surveys (such as ESA4 and the German Study on Tobacco Use (DEBRA)38) are also expected to provide valuable insights, this study is specifically designed to evaluate the law reform and will collect data from larger samples of people using cannabis. The study findings will be presented to and discussed with the study funder (Federal Highway Research Institute). This research institute is closely cooperating with the Federal Ministry for Digital and Transport and aims to support the introduction of national and international laws related to traffic safety. Thus, our study will be crucial for a comprehensive evaluation of the cannabis law reform. Our findings are expected to aid in the design and implementation of road safety prevention measures, for example, by customising prevention campaigns to the target population reporting DUIC.

Discussion

This study will provide key data on DUIC in Germany, using a large sample of regular cannabis users recruited from a representative sample of the German population. Additionally, the study will evaluate the impact of the partial legalisation of cannabis, particularly on cannabis use and DUIC. The inclusion of a control group and the long observation interval enable a precise evaluation. Due to possible reporting biases in the measurement of DUIC, direct and indirect (CWM) assessments and (drug-related) MVAs will be analysed, which increases the reliability of data interpretation. The results may contribute to the discussion on drug driving, potentially influencing considerations such as THC limits and prevention campaigns. Also, the findings may help clarify the inconsistent relationship between self-reported DUIC and the partial legalisation of cannabis.5 16 17 Generally, this study will possibly be the first to provide empirical data on the impact of a cannabis law reform in a European country using a controlled study design.

However, there are some risks inherent in the study design and conception. First, there is large uncertainty regarding the (true) prevalence of DUIC. Due to the scarcity of European or even German data, Canadian data were relied on and some adjustments were made based on theoretical considerations. As the prevalence of DUIC is crucial for our sample size calculation, there is some uncertainty that we may not have accounted for. Accordingly, we cannot rule out the possibility that the number of respondents in the Austrian control group is insufficient to be included in testing H2 through a difference-in-difference design. If we need to resort to simple pre-post comparisons, this will weaken any possible conclusions regarding a causal link between DUIC and partial legalisation. Overall, we adopted a conservative approach in the sample size calculation and believe that our study is sufficiently powered to test the two main hypotheses. Second, we cannot fully ascertain the presence of parallel trends in cannabis use prevalence in both Germany and Austria. Increases in cannabis use have been observed in both countries in recent years, but our capacities in directly assessing parallel trends are limited by the lack of comparable survey data from recent years.

Third, it is important to acknowledge that the use of indirect questions (CWM) may introduce some risks, such as misunderstandings due to higher complexity or the exclusion of people who do not know the birth month of their parents (neutral question), which may bias prevalence estimates. However, a previous study conducted in Sweden found this approach to be feasible in a general population survey, producing plausible and consistent estimates of cannabis use prevalence.39 In conclusion, this study will be pivotal to investigate the impact of the cannabis law reform in Germany. The findings are urgently needed to inform the current public debates on cannabis legislation in Germany and other European countries, which predominately relied on North American experiences up until now.

Ethics statementsPatient consent for publication

Not applicable.