Words versus Strands: Reliability and Stability of Concordance Rates of Self-Reported and Hair-Analyzed Substance Use of Young Adults over Time

Substance use is a global concern [1], being associated with physical and mental health problems [2]. Adolescents and young adults are at increased risk of substance use and substance use disorders [3]. However, measuring substance use poses challenges. Substance use can be captured either by self-reports that are prone to reporting biases or by biological markers, such as hair analyses or urine tests, which seem to be more objective data but come with limitations, such as sensitivity.

The accuracy and reliability of self-reports and hair analyses have been investigated cross-sectionally but not yet at two time points for a community-based cohort of young adults [4‒7]. However, measuring substance use across two time points can indicate stability and changes in substance use behavior over time. Consequently, a longitudinal perspective is crucial to identifying trends and patterns, providing a better understanding of these (intra-individual) dynamics to inform prevention and intervention strategies. The present study thus reports substance use prevalence rates (12-month and 3-month) among females and males aged 24 in a cohort in Switzerland (i) to draw comparisons to prevalence rates at age 20 [8], (ii) to assess the accuracy and reliability of self-reports and hair analyses at two measurement points (ages 20 and 24), and (iii) to investigate who underreports substance use across both time points [5].

Substance use is particularly prevalent among young adults. Cross-sectional survey data reveal that by age 21, a significant majority of young adults in countries like New Zealand (62%) and the USA (54%) have used cannabis [9]. UN agencies have reported increased use between 2011 and 2021 not only of alcohol, tobacco, and other illegal substances (e.g., cannabis, hallucinogens, stimulants, opioids) but also of non-medical prescription medications (e.g., opioid painkillers, tranquilizers), with global substance use rising by 23% over the past decade, resulting in one in every 17 people worldwide having used a substance in 2021 [1]. However, cannabis remains the most widely (illegally) used substance globally, followed by non-medical use of prescribed medication [1].

Several data sources provide an overview of substance use in Switzerland [8, 10‒15]. These studies, particularly the z-proso project (on which the present study is based), reveal frequent substance use among young adults (e.g., 57% use of cannabinoids in the past year) [8]. High prevalence rates in studies are supported by wastewater analyses, which indicate that Swiss cities have among the highest wastewater concentrations of cocaine and 3,4-methylenedioxymethamphetamine (MDMA, “Ecstasy”) in Europe [16, 17].

Additionally, sex differences have been extensively documented. While males tend to exhibit higher rates of illegal substance use, females are more likely to engage in the non-medical use of prescription medications [1, 8]. But even though males use more substances overall, females progress more quickly from substance use to addiction, making them more vulnerable to substance use disorders [18].

Although studies on substance use prevalence rates to date have been crucial in advancing successful prevention and intervention strategies, they have limitations. First, policy and intervention planning require up-to-date information about dynamically changing substance use prevalence, which quickly renders older data obsolete. Second, studies have primarily relied on self-report questionnaires. These are essential for gathering information on substance use patterns, frequency, and intensity. Still, they rely on individuals’ willingness and ability to recall and report substance use behaviors accurately. Self-reports may be especially prone to recall bias due to fear of stigmatization or consequences associated with illegal substance use [19] but also possible intoxication effects [20], depending on the setting of the study.

Hair data can address these limitations as it offers a biological marker for substances and their metabolites, which are incorporated into the hair following consumption and can be detected in the hair roots and shafts as hair grows. Hair analysis can thus provide a retrospective timeline of substance use over recent months [21] and might be a more reliable and accurate method less susceptible to biases. Unlike urine or blood testing, hair samples can be gathered quickly and in most settings [22, 23]. While highly reliable for detecting cocaine, MDMA, and opioid use, the sensitivity of hair analysis is lower for cannabis, where only weekly to daily use can be reliably detected [24]. Additionally, tetrahydrocannabinol (THC) hair concentrations can also be influenced by external contamination [25, 26]. Several studies have investigated the concordance of self-reports and biological markers, including HAs. A recent systematic review found generally high agreement between self-reports and biological markers in clinical study settings. However, others have reported notable discrepancies between self-reports and hair analyses, with the use of certain substances underestimated. For example, at the community-representative level, anonymized [5], who lay the foundation for this study with repeated measures, reported underestimated use of illegal substances in self-reports compared to hair analyses among young adults in the z-proso cohort; MDMA, cocaine, ketamine, amphetamines, codeine, and opioid painkiller use was underreported by 30%–60%.

Self-reported behavioral and psychological characteristics may influence the accuracy of substance use self-reports [5]. For example, for codeine and/or cocaine use, higher levels of self-reported delinquency, attention-deficit hyperactivity disorder (ADHD) symptoms, internalizing symptoms, and lower levels of self-control are associated with less underreporting. One study using four repeated measures over 1 year reported findings of specific characteristics influencing underreporting (i.e., sex, age, ethnicity) but also reported overreporting of the substance used most often in recent months or that caused the most difficulties [27].

Studies investigating the concordance of self-reports and biological markers have used cross-sectional designs; many focused on clinical or high-risk samples or relied on urine or blood tests. However, these approaches might not represent substance use in a more general population and do not account for intra-individual variability across time. Knowing about factors influencing reporting accuracy, such as memory recall, social desirability, or changes in usage patterns, can improve the accuracy of substance use assessments and inform the design of effective preventions and interventions. Therefore, this study uses data from two time points at ages 20 and 24 from a community-based sample to gain insights into substance use (12-month and 3-month prevalence rates) and their patterns’ temporal dynamics and assess the concordance of self-reports and hair analyses at these two time points.

We examined the following research questions (RQ) and hypotheses (H): RQ1.1: What are the 12-month self-report prevalence rates of cannabinoids (THC, cannabidiol, cannabis substitutes/synthetic cannabinoids), cocaine, amphetamine, substituted amphetamines (i.e., methamphetamine, MDMA), methylphenidate, hallucinogens (lysergic acid diethylamide (LSD)/psilocybin, 2C substances, ketamine), medical and non-medical opioids (codeine, opiate painkillers, heroin), dextromethorphan, Z-drugs, and tranquilizers (benzodiazepines) among 24-year-olds in the z-proso sample? H1.1: We expected to find high self-report prevalence rates (similar to age 20; [8]) among 24-year-olds, particularly for cannabis, cocaine, Ecstasy, and codeine [8, 28]. RQ1.2: How do prevalence rates of substance use among 24-year-olds differ between sexes? H1.2: We expected a higher rate of illegal substance use among men and a higher use of non-medical use of prescribed drugs among women [8, 28]. RQ2.1: How does the test-retest reliability of accuracy metrics (positivity and agreement rates, sensitivity, specificity, kappa values) between self-report substance use and hair analysis change over time, and how does this vary among substances? Since this is exploratory, we only assume the following hypothesis: H2.1: We expected consistent underreporting for most substances [4‒6]. RQ2.2: How does the test-retest reliability for the answer patterns (agreement, false negatives, false positives) between self-report substance use and hair analysis change over time, and how does this vary among substances? H2.2: Again, exploratory; we expected a certain consistency among the answer patterns. RQ3: What predicts (longitudinally) underreporting of substance use among the 24-year-olds? H3: We expected self-reported delinquency, ADHD, low self-control, and internalizing symptoms to be associated with less underreporting [5].

Data came from the Zurich Project on Social Development from Childhood to Adulthood (z-proso) [29]. In 2004, 56 out of 90 public elementary schools with 1,675 children were randomly selected in the city of Zurich. The selection process included stratification based on school size and districts’ socio-economic backgrounds, ensuring the sample closely mirrored the demographic structure of first-graders at that time in Zurich’s public schools. The first assessment included 1,360 pupils at age 7. Since then, eight additional assessment waves have been conducted, with the last two carried out in 2018 (wave 8; age 20; n = 1,180) and 2022 (wave 9; age 24; n = 1,160). In addition to completing the anonymous survey, participants were invited to provide 3 cm of hair to detect psychoactive substances stored in it for the previous 3 months at waves 8 and 9. If scalp hair was shorter than 1 cm, arm, leg, or chest hair was provided. A total of 1,002 hair samples were obtained at wave 8, 887 hair samples were obtained at wave 9, and 762 participants donated hair samples at both waves 8 and 9. However, one participant was excluded because of missing data on all relevant variables, resulting in an overlapping sample at waves 8 and 9 of 761 participants. The number of participants most likely decreased due to the online participation option that more participants took advantage of after COVID-19. The online participants did not donate a hair sample. Online supplementary Tables S1 and S2 (for all online suppl. material, see https://doi.org/10.1159/000541713) in the supplement show that both wave 8 and 9 samples (all vs. only both time points) are similar in their demographics.

Table 1 shows that the sample is evenly split in terms of sex-assigned-at-birth and consistent with Switzerland’s immigration policies. It is a multi-ethnic sample with a high percentage of participants having a migration background. The study was conducted in accordance with the Declaration of Helsinki. The Cantonal Ethics Committee Zurich (BASEC #2017-02021) and the Ethics Committee of the Faculty of Arts and Social Sciences, University of Zurich, approved the study. Participants provided written informed consent and received compensation for participation in the survey and the hair donation.

An extensive list of substances was given to the participants, including cannabinoids, stimulants, hallucinogens, and non-medical and medical opioids (complete list in online suppl. Table S3). Participants indicated how often they had consumed each substance in their life as well as in the past year (1 = never; 2 = once, 3 = 2–5 times [quarterly]; 4 = 6–12 times [monthly]; 5 = 13–52 [weekly] 6 = 53–365 times [daily]), and within the past 3 months (0 = never; 1 = once; 2 = 2–5 times [quarterly], 3 = weekly, 4 = [almost] daily). Regarding prescription drugs, participants were asked to report only non-medical use (i.e., more frequently or at higher doses than prescribed).

The substances and their metabolites were quantified using liquid chromatography-tandem mass spectrometry, which has been described using different data [30, 31] and in the supplement of the wave 8 cross-sectional investigation [5], on which this study is based. Participants completed an additional questionnaire on over-the-counter medications and variables that could affect hair toxicology analyses (e.g., hair bleaching [0 = no, 1 = yes], hair color [light, brown, dark]).

The following subscales from the Social Behavior Questionnaire (SBQ) [32, 33] were included: physical aggression, ADHD, and internalizing symptoms. On a 5-point Likert scale, ranging from 1 = never to 5 = very often, participants reported how frequently they had experienced or engaged in these behaviors within the last year for physical aggression and ADHD symptoms and within the previous month for internalizing symptoms. Delinquency was assessed using a 24-item dichotomous checklist [34, 35], including minor and major delinquent acts, such as graffitiing and assault. For this scale, a sum score was computed. Finally, self-control was measured using a 10-item adaptation of the Grasmick Self-Control Scale [35, 36]. On a 4-point scale ranging from 1 = fully untrue to 4 = fully true, participants reported on items such as Sometimes I do dangerous things just for the fun of it.

Sex-assigned-at-birth and parental migration background were measured as binary variables (0 = female, 1 = male; 0 = at least one parent born in Switzerland; 1 = both parents born abroad). Information was gathered in W4 to W6 for the migration background variable. International Socio-Economic Index of Occupational Status scores were derived from parental occupation and ranged from 16 (e.g., unskilled worker) to 90 (e.g., judge) [37]. The regression models used participants’ highest educational degree as a dummy variable, including compulsory schooling and preparatory vocational education, academic education, and vocational degrees (as reference category).

All analyses were conducted in SPSS 29 [38] and R [39]. For data visualization, ggplot2 [40] was used. Prevalence estimates were computed separately for females and males for the entire sample. Group differences were tested using a χ²-test. Dummy variables were created to compare the reliability of self-reported substance use and its detection in hair analyses to assess the reliability of the two methods. These variables were coded as 0 for no self-reported use and no substance detection in hair and 1 for self-reported use and substance detection in hair, based on the concentration of the substance or its metabolites exceeding the limit of quantification. This coding was applied independently to both self-report data and hair analysis results. However, due to a programming error in the filter structure of the questionnaire, the 3-month self-reports for MDMA were not assessed at wave 8 and could thus only be reported at wave 9. In the cross-sectional investigation [5], a “by proxy” prevalence estimation of MDMA was calculated by using self-reported ratios from 3- and 12-month prevalence rates of cocaine and amphetamine use, which often occur in similar contexts such as parties.

McNemar tests were used to investigate reliability and compare self-report and hair analysis prevalence estimates based on the dummy variables. We calculated the following diagnostic accuracy metrics: positivity rate, under-/overreporting rate, specificity, sensitivity, balanced accuracy, detection ratio, agreement, and kappa for the reported substances at both time points [21] (Table 2). An explanation of these statistics can be found in wave 8 [5] (online suppl. Table S4). The metrics for the time points were compared using z-scores, except for the detection ratio, which cannot be used for proportion tests. For the accuracy metrics, we used (1) the entire sample participating at both time points (n = 761) and (2) an optimal sample of participants who provided ≥3 cm of scalp hair that weighed ≥5 mg (n = 613) (online suppl. Tables S5–S7). Similarly, χ² tests were conducted to analyze the test-retest reliability of the self-reports, hair analyses, agreement (identical coding for both methods, 0/0 and 1/1 vs. 0/1 and 1/0), false positives (negative hair analysis but positive self-report), and false negatives (positive hair analysis but negative self-report). Online supplementary Table S8 provides an overview of the test-retest variables and answer patterns.

To assess the (longitudinal) influence of response biases on underreporting, binary logistic regressions were conducted, focusing on behavioral and psychological predictors of underreporting of cocaine and codeine. These two substances were considered representations of illegal and medical substances with notable baseline prevalence. The regression models were adjusted for the concentrations of the substances and their metabolites found in hair, allowing for unintentional substance exposure of exposure from over 3 months prior.

The 12-month self-reported prevalence rates (Fig. 1, online suppl. Table S9) of substance use at age 24 in the total sample (n_Wave9 = 987) are high. Cannabis is the most-used substance (47.4%), followed by cannabidiol products (24.9%). Substances with rates above 10% are codeine, MDMA, cocaine, opioid painkillers, and LSD/psilocybin (only in males). LSD/psilocybin, medical stimulants, benzodiazepines, amphetamines, z-drugs (only in females), and cannabis substitutes (only in males) have been non-medically used by more than 5% in the past year.

In general, more males reported illegal substance use compared to females. In contrast, females tended to report more non-medical use of prescribed medications than males.

The appendix contains 3-month prevalence rates of hair analysis, which confirm highly prevalent substance use (n = 761). It also includes past-year prevalence rates for the self-report/hair analysis sample and statistical testing of sex differences (online suppl. Fig. S1, S2; Tables S10, S11).

Changes in 3-month prevalence rates between ages 20 and 24 have been tested for the use of cannabis, amphetamines, MDMA/Ecstasy, cocaine, ketamine, codeine, and opioid painkillers (Table 2). Self-report-derived prevalence rates significantly increased for cocaine and decreased for codeine. Hair analysis-derived prevalence rates of cocaine and ketamine significantly increased over time, while it was decreasing for codeine.

Comparing 3-month prevalence rates of self-report and hair analysis over time (Table 2), consistent underreporting (similar underreporting rates between time points, measured through z-tests, and investigation of false negatives through χ² tests) was observed for daily cannabis, Ecstasy/MDMA, cocaine, ketamine, and codeine use. Underreporting of Ecstasy/MDMA, cocaine, ketamine, and opioid painkillers significantly increased from age 20 to 24. Contrarily, weekly to daily cannabis, amphetamine, and opioid painkiller use was consistently overreported. The detection ratio increased for all substances from wave 8 to wave 9, except for opioid painkillers. The agreement rate decreased for cocaine, ketamine, and opioid painkillers but increased for codeine. Kappa values also decreased for amphetamine, cocaine, ketamine, and opioid painkillers. Like in wave 8, most hair analyses demonstrated higher sensitivity but lower specificity than self-reports, suggesting that hair analysis more often confirmed positive results from self-reports than vice versa. However, significant differences in kappa values were found again for amphetamines, cocaine, ketamine, and opioid painkillers. Except for opioid painkillers, hair sensitivity increased significantly at wave 9 for all substances. Self-report specificity did not change for most substances across time, except for a decrease in ketamine and opioid painkiller use. Furthermore, self-report sensitivity was similar to hair analysis sensitivity, which, in most cases, was lower than the specificity. It also decreased for frequent cannabis, amphetamines, and cocaine use but increased for ketamine use. Finally, balanced accuracy was higher for most substances in the hair analyses than the self-reports. Similar results were found when running the analyses with the optimal sample only. The most substantial difference was found for self-report sensitivity for ketamine, which was lower in the optimal sample (online suppl. Table S5).

Each method’s test-retest reliability revealed statistically significant kappa values for all substances, indicating specific measurement stability across time for each method separately (Table 3). Participants had similar answer patterns across time (Table 4). When investigating the agreement rates (self-report and hair analysis were either positive or negative) across time, a high concordance rate for all substances was found. Self-reports and hair analysis aligned at waves 8 and 9 for most participants, reinforcing the credibility of both methods for capturing substance use. However, several participants moved from the no-agreement group to the agreement group and vice versa between the time points, meaning initial agreement was not always maintained. The highest rates of no agreement at both time points were found for cannabis and codeine.

False negatives (positive hair analysis but negative self-report) were consistent across time for most substances, except amphetamine and opioid painkillers; false positives (negative hair analysis but positive self-report) were consistent across time for cannabis and amphetamines (Table 5). The significant relationship between false negatives for cannabis, cocaine, ketamine, and codeine at both time points indicates that the discrepancy between self-reports and hair analysis does not occur by chance. A certain consistency in incorrect self-reporting can be found. The same applies to false positives for cannabis and amphetamines.

Consistent with findings from wave 8 [5], regression analyses revealed that sociodemographic variables were not associated with the underreporting of cocaine and codeine use in wave 9 (online suppl. Tables S12, S13). Still, cross-sectionally higher delinquency and ADHD symptoms were associated with lower odds of underreporting cocaine use. Higher levels of internalizing symptoms were found to decrease the odds of underreporting codeine use. Further analysis, including only individuals who did not report medical use of codeine, was not possible due to the reduced sample size. Longitudinally, only higher levels of internalizing symptoms were associated with lower odds of underreporting codeine (online suppl. Table S13).

This study examined the 12-month prevalence of self-report, the sex differences in prevalence rates, the concordance between 3-month prevalence rates of self-report and hair analysis, and predictors of underreporting using two waves of a community-representative sample of substance use among young adults in Switzerland aged 20 at wave 8 and 24 at wave 9.

In line with our H1.1, we found higher prevalence rates across the last 12 and last 3 months for the 24-year-olds compared to 4 years earlier for many substances, with a particular rise in cocaine and opioid painkiller use. Additionally, in line with H1.2, males reported higher illegal substance use, whereas females reported higher non-medical prescription medication use. Hair analysis corroborated these results, which align with previous findings [1].

We found strikingly high prevalence rates for cocaine use (3 months: 9.4% in self-report and 22.8% in hair analysis), which increased more than 2-fold in hair analysis between ages 20 and 24. Also, ketamine use showed a 3-fold increase in hair. These numbers support previous findings indicating either a later age of onset for cocaine and ketamine use compared to substances such as cannabis or MDMA [9] or a general increase in use in the population. Nevertheless, the sample is Zurich-based, implying cocaine, ketamine, MDMA, methamphetamine, and cannabis use is high in the city, as confirmed by wastewater analysis [41].

Across time, underreporting is highly consistent but also significantly increased for MDMA (caution: underreporting was only estimated for age 20), cocaine, and ketamine (Table 2). Also, overreporting was consistent, which decreased for weekly to daily cannabis use and amphetamines but increased for opioid painkillers. The increase in underreporting of cocaine use might be explained by the stigma of the substance, while ketamine might be ingested unintentionally in the context of cocaine or MDMA use. The changes in overreporting of weekly to daily cannabis use and amphetamines might be explained by measurement problems of hair analysis specifically for these two substance groups [5].

The observed increase in the detection ratio between waves 8 and 9 for most substances indicates a shift in substance use responses as the detection methods did not change over the studied period, supporting H2.1. Declining agreement rates for several substances further imply a divergence in self-reported substance use compared to hair analysis. Only codeine stands out with an increased agreement rate and a decrease in use, which might reflect stronger controlled codeine access. As of January 1, 2019, access to codeine and codeine- or dihydrocodeine-containing drugs in Swiss pharmacies has been more restricted as the acquisition must be discussed with and documented by a pharmacist [42]. Therefore, participants using codeine in 2022 may have had a better understanding of the substance they used than in 2018.

The decline in kappa values from wave 8 to wave 9 for most substances underscores the growing discrepancy between self-reported use and hair analysis. This trend is particularly noticeable with substances such as amphetamines and cocaine. Both substances are linked to social stigma and can thus affect the accuracy of self-reports [19, 43]. The substantial decrease in self-report sensitivity combined with increased hair analysis sensitivity might further support participants’ reluctance to disclose specific substance use. However, the consistent specificity of self-report (except for ketamine and opioid painkillers) demonstrates that even though numerous participants underreport substance use, when participants do report use, it tends to be reliable.

Both methods demonstrate notable measurement stability, supporting H2.1. The consistency in participants’ response patterns reinforces this conclusion. However, that some participants changed their answers over time and shifted between agreement and disagreement groups suggests that initial agreement cannot absolutely predict future concordance or vice versa.

Moreover, stability across false negative and false positive patterns underscores the self-report and hair analysis data discrepancies, confirming H2.2. The consistent patterns of false negatives for codeine, cannabis, ketamine, and cocaine indicate that the use of hair analysis is highly relevant. Furthermore, the consistent patterns of false positives for amphetamines and cannabis might point to limitations in hair analysis. THC, in particular, is more difficult to detect than other illegal substances [24]. These patterns of false negatives and positives suggest a systematic bias rather than a random error in reporting, potentially driven by external factors like stigma and recall bias. However, it is also important to consider that the underreporting of substance use may also be influenced by the adulteration of specific street substances with another undeclared psychotropic agent [44], which users cannot recognize or report accurately. For example, ketamine is known to be an adulterant of Ecstasy pills or an admixture of cocaine (called “CK”) [45‒47]. This adds another layer of complexity, emphasizing the need for methods that can detect or inquire about adulterants to ensure more reliable substance use assessments.

It is worth noting that the accuracy of substance use reporting might differ between population-based studies and substance-specific studies. Broader samples may yield less specific self-reports compared to focused studies on particular substances or populations. These specialized studies often employ more elaborate measures, including face-to-face interviews, possibly improving self-report validity [4]. Indeed, clinical case-control studies have shown high correlations between self-reported and hair-tested cocaine use measures [48].

Therefore, predicting response patterns with certainty is not feasible, and these findings highlight the necessity of using multiple methods to capture substance use accurately. After all, even though longitudinal investigations allow for a nuanced understanding of prevalence rates and reliabilities, prevalence rates still vary with age, suggesting consistency across measurements might change in line with changes in prevalence across age groups. Further analysis of how substance use patterns change within specific age groups and the ability to track them accurately would be valuable.

Under- and overreporting are driven by different mechanisms, such as hair testing not being able to detect substances or metabolites (especially difficult for THC and amphetamine), misremembering the time frame of substance use, lack of knowledge about the substances (e.g., exposure to contaminated substances or receiving other substances than expected) and medications (e.g., not being able to distinguish opioid from non-opioid painkillers), or social desirability, stigma, and fear of consequences. Studies have demonstrated that, specifically, people from minority ethnicities [49, 50] and those having low antisocial behavior levels [49, 51] tend to underreport their substance use. Contrary to studies finding ethnicity effects for underreporting [27, 49], at age 24, we replicated the age 20 findings [5], supporting H3, showing no migration background effects in the models. Our longitudinal analyses also reinforce associations between underreporting and behavioral and psychological characteristics. The observed relationship between internalizing symptoms, delinquency, ADHD symptoms, and low self-control with underreporting of cocaine and codeine use may reflect a more consistent self-image that is in line with actual behavior, including substance use, or a general pattern of non-social behavior spanning from cocaine use to lower susceptibility to social desirability bias. This suggests that stable psychological and personal traits may impact the dynamics of substance use and its self-reporting, emphasizing the need to consider these characteristics in future studies. However, longitudinal investigations only show a stable pattern for higher delinquency and underreporting.

Several limitations must be addressed. First, the binary classification of substance use might fail to capture possible nuances between occasional and regular consumption patterns. Second, the absence of MDMA at wave 8 for the 3-month self-reported use prevalence did not allow for direct comparisons. Third, direct comparisons were conducted using multiple tests for each substance, possibly inflating type I errors. Fourth, even though hair analysis is considered a “gold standard,” it still presents challenges, such as inaccurate detection of non-regular cannabis use and passive exposure to THC. Finally, demographic characteristics, such as gender and ethnicity, can influence the likelihood of agreement to provide hair samples [52].

Substance use prevalence among young adults in the Zurich region of Switzerland is high, and ideally, hair analyses and self-reports should be employed together as a multimethod approach to capture substance use. Although certain illegal substance use is often underreported, our findings indicate self-reported substance use is, for the most part, a dependable indicator of actual substance use. The consistent yet variable underreporting nevertheless highlights the importance of objectively assessing substance use, particularly for studies investigating the use of cocaine, Ecstasy/MDMA, ketamine, and codeine. Combining both methods is desirable for low- and high-risk samples to accurately determine various substances independent of age range.

The authors thank all participants, field managers, and the toxicology laboratory team for their invaluable contributions to the study’s success.

The study was reviewed and approved by the Cantonal Ethics Committee Zurich (BASEC #2017-02021) and the Ethics Committee of the Faculty of Arts and Social Sciences, University of Zurich, approval No. 2018.2.12 (phase 5) and approval No. 21.12.13 (phase 6). Written informed consent was obtained from participants to participate in the study.

The authors have no conflicts of interest to declare.

The Swiss National Science Foundation (SNSF) funded the study; grant #10531C_189008 was awarded to L.S. and grant #105314_214979 to B.B.Q. Furthermore, the z-proso study has been financially supported by the SNSF as a research infrastructure, grant #10F114_170409, awarded to Michael Shanahan, grant #10F114_198052, awarded to D.R. and L.S., by the Jacobs Center, and by the Jacobs Foundation (JF). In earlier stages of z-proso (2003–2016), the SNSF, JF, Swiss Federal Office of Public Health, Department of Education of the Canton of Zurich, Swiss State Secretariat of Migration and its predecessors, Julius Bär Foundation, and Visana Plus Foundation supported the study. Funding was provided for independent fundamental research.

C.J., L.E., A.S., L.S., and B.B.Q. initially conceptualized the study. C.J., L.E., A.S., L.B., D.R., M.R.B., and T.M.B. handled data curation. C.J., A.S., and L.B. took on the formal analysis. D.R., M.E., L.S., and B.B.Q. secured funding. D.R., M.E., M.R.B., T.M.B., L.S., and B.B.Q. were involved in the investigation. C.J., L.E., A.S., L.B., L.S., and B.B.Q. developed the research approach for methodology. L.S. and B.B.Q. managed project administration. C.J. and L.E. were responsible for data visualization. C.J. wrote the original draft and the manuscript’s review and editing saw contributions from L.E., A.S., L.J.F., L.B., M.L., D.R., M.E., M.R.B., T.M.B., L.S., and B.B.Q. All authors approved the final manuscript.

Lilly Shanahan and Boris B. Quednow contributed equally to this work.

The data that support the findings of this study are not publicly available due to sensitive personal information. However, researchers can request the data from the principal investigators (D.R.; [email protected]) of the z-proso project with a brief sketch of its planned use.