Background

Substantial increases in cannabis use prevalence among women of reproductive age over the past decade is a global public health concern [1,2,3,4]. Of particular concern is rising use of cannabis during pregnancy – a critical period for both women and their offspring [5,6,7]. These stark upticks in prenatal cannabis use can be seen in North America, particularly the United States (US) and Canada [2, 8, 9]. In the US, estimates of prenatal cannabis use more than doubled from 3.4% in 2002–2003 to 7.0% in 2016–2017 [10]. Similarly, Canada saw a relative increase of 61% in prenatal cannabis use prevalence from 2012 to 2017 [2].

Emerging evidence indicates prenatal cannabis use and exposure is not without consequence. Indeed, a recent systematic review and meta-analysis by Marchand et al. found that in-utero cannabis exposure was associated with an increased risk of several adverse neonatal health outcomes compared with infants not exposed [11]. Neonates with in utero cannabis exposure had higher rates of preterm birth, low birth weight, small for gestational age, admission to the neonatal intensive care unit, and smaller head circumference [11]. Importantly, THC can persist in breast milk up to 6 weeks after prenatal cannabis use cessation [12], which has large implications for pregnant women who use cannabis and intend to breastfeed. A recent call to action highlighted the growing body of evidence supporting risk of adverse neonatal health outcomes associated with in utero cannabis exposure [13]. Collectively, prenatal care clinicians are integral to preventing these adverse health outcomes via enhanced screening for and clear communication about risks of cannabis use and exposure.

Effective screening for cannabis use during pregnancy is essential for prevention of adverse perinatal consequences associated with in-utero cannabis exposure. However, most studies reporting on prenatal cannabis use rely on maternal self-report [8,9,10, 14,15,16]. Additionally, these self-report measures are relatively quick and easy to administer in clinical settings [17]. Self-report also allows for patient contextual elaboration regarding prenatal cannabis use, including frequency and mode of administration. However, there are many limitations of self-reported measures, including stigma and fear of punitive consequences, especially in high-risk populations, such as pregnant women [18, 19]. Additionally, the setting, interviewer, and population have also been shown to influence the validity of self-report [20]. Thus, examining the validity of self-reported measures of prenatal cannabis use across diverse populations with different administrators in different settings is of clinical importance. Accurate detection of prenatal cannabis use is also an important methodological issue for studies that aim to examine the effect of cannabis use and exposure on perinatal health outcomes. To date, several studies have assessed the validity of self-reported measures of prenatal cannabis use in comparison to estimates from biochemical samples, such as urine, hair, umbilical cord, or meconium samples [17, 21,22,23,24]. However, this evidence has yet to be reviewed and synthesized.

To fill this evidence gap, we performed a scoping review that systematically identified and synthesized existing evidence on the validity of self-reported measures of prenatal cannabis use in comparison to estimates from biochemical samples. In conducting this scoping review, we also aimed to determine if there was enough available evidence on this topic to perform a future systematic review and identify potential questions that could be answered with existing evidence.

Methods

We aligned this scoping review with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) Checklist (Supplementary Table 1) [25].

Protocol and registration

We followed the classic framework for scoping reviews by Arksey and O’Malley, as well as recent guidance to increase rigor and reporting of scoping reviews [25,26,27]. We developed our a priori protocol using the PRSIMA extension for Scoping Reviews as a guide [25]. Given the rapid timeframe of this review, we opted not to publish the protocol for this review, but it is available upon request.

Eligibility criteria

We included studies that examined the validity of self-reported measures of cannabis use in pregnant women. More specifically, studies needed to compare estimates of self-reported prenatal cannabis use to a biochemical measure of cannabis use, including but not limited to hair, urine, and meconium sampling. We included studies from any geographical location if they were published in 2010 onward and written in English. We included only peer-reviewed articles regardless of study design, so long as they met other criteria, which included systematic reviews, with or without meta-analysis, and reviews of the literature. We excluded studies that did not include pregnant women only, were published before 2010, were published as conference abstracts or book chapters, or were not published in English. We also excluded studies reporting solely on the prevalence of prenatal cannabis use via self-report (e.g., national level surveillance data) or estimates of prevalence from biochemical samples, as comparison between self-report and a biochemical measure was not possible.

Information sources

We systematically searched PubMed, PyschINFO, CINAHL, and Cochrane/CENTRAL from January 2010 to June 2021. We also included the first 200 results from Google Scholar, when sorted via relevance ranking. Given the shifting landscape of prenatal cannabis use, we limited our search from 2010 onward to identify contemporary measures of self-reported cannabis use, as opposed to dated measures that included cannabis in a category with other illicit substances (e.g., cocaine, heroin). We developed unique search strategies for each database, which we then piloted. After the initial pilot searches, we adapted the initial search terms to exclude those that did not yield relevant results, which included the following search terms: “survey”, “weed”, and “CBD”. We developed the final search strategy (Supplementary Table 2) using terms specific to our population (e.g., “pregnant”, “prenatal”, “pregnancy”) and topic (e.g., “cannabis” and “marijuana”) [28, 29], including terms to capture validation (“validity”, “evaluation”, “validation”, “agreement”) [25].

Selection of sources of evidence

We used an online systematic review management software, Covidence, to streamline the review process (Covidence Systematic Review Software, Veritas Health Innovation, Melbourne, Australia). As the first step of our review process, we exported all search results from each database into EndNote (Clarivate Analytics, Philadelphia, USA). Next, we imported citations from EndNote (Clarivate Analytics, Philadelphia, USA) into Covidence. As part of this import process, Covidence automatically de-duplicated citations based on a match of the citation title, author, and date.

After search results were imported into Covidence, we performed a two-stage review process in which we screened references for inclusion based on eligibility criteria. To ensure reviewer agreement, two reviewers (KS and ED) piloted the screening process with 25 citations. Inter-rater agreement was high (95%) and thus formal screening began. Then, two members of the research team performed title and abstract screening independently. Upon completion of title and abstract screening, two reviewers screened remaining studies in their full-text, PDF form. Only articles meeting all inclusion criteria moved forward for data extraction. We resolved disagreements between reviewers at any stage using consensus and discussion. Lastly, we performed forward and backward citation searches for the final list of included studies.

Data extraction

Two reviewers independently performed data extraction for each included study using modifiable templates in Covidence. We abstracted the following datapoints from each study: study details (e.g., author, setting, dates, purpose, funding), population and sample size, study design and methods, details about measures used (both self-reported and biochemical), outcomes, limitations, recommendations (both for practice and research), and conclusions. Specific outcomes of interest included negative predictive value (NPV), positive predictive value (PPV), sensitivity, specificity, and percent agreement between maternal self-report and biochemical sampling. However, due to expected variation in reporting of outcomes, we included studies that reported other outcomes measuring the relation between maternal self-report and biochemical tests.

Synthesis of results

We performed a narrative synthesis, mapping existing evidence across key categories, including type of biochemical sample used for comparison of self-report and type of self-report used (e.g., health care provider screening, written screener, etc.). We also present data in tabular form by country and a separate table reporting recommendations for both future research and practice.

Results

After de-duplication, we screened a total of 927 unique articles, resulting in 12 articles included in this scoping review. We detail the study screening and selection process in accordance with PRISMA guidelines in Fig. 1.

Fig. 1
figure 1

PRISMA Flow Diagram

Full size image

Study characteristics

We report key characteristics of included studies in Table 1. Of the 12 studies included, 7 were conducted in the US [17, 21, 23, 30,31,32,33] and 4 in other countries, including Brazil [34], the Netherlands [24], France [35], and South Africa [36]. The included systematic review contained studies conducted in an array of countries from across the globe [37]. Studies were heterogenous in overall study design, population, sample size, and measures used. Studies included pregnant women at different stages of pregnancy, ranging from the first prenatal visit to delivery. Sample sizes ranged from 83 [17] to 281,025 pregnant women [23].

Table 1 Key characteristics of included studies (n = 12)
Full size table

Validity outcomes

Urine

Most included studies (n = 8) compared maternal self-reported prenatal cannabis use prevalence to urine [17, 23, 24, 30,31,32,33, 36]. Overall agreement between maternal self-report and urine was poor to moderate, ranging from 34% [23] to 64% [33]. Young-Wolff et al. (2020) reported the lowest agreement, with self-report screening identifying only 34% of those testing positive via urine [23]. Similarly, El Marroun et al. (2011) found poor agreement between self-report and urine, with only 35% of the 92 women reporting cannabis use had positive urine screens [24]. Similarly, Chang et al. (2017) found that only 36% of women with a positive urine screen disclosed use to a health care provider [31]. Garg et al. (2016) and Yonkers et al. (2011) found the highest level of agreement between self-report and urine toxicology, with 60 and 64% agreement, respectively [17, 33].

Hair

Hair analysis was used in 2 of the included studies [30, 34], with both studies reporting poor agreement of self-reported prenatal cannabis use. One study conducted in Brazil found no agreement with hair samples due to a 0% disclosure rate for cannabis [34]. Another study conducted in the US found overall prevalence of cannabis use via hair sampling was 28% (compared to 11% via self-report only); 6 participants who reported cannabis use had a negative hair sample [30].

Umbilical cord

A single study compared maternal self-reported cannabis use to umbilical cord homogenate assays, comparing biochemical results to both survey and medical record review [21]. Metz et al. (2019) found moderate agreement between 30-day use via survey and umbilical cord homogenate (kappa = 0.52) and slight agreeance between medical record review and umbilical cord homogenate (kappa = 0.17) [21].

Meconium

Lamy et al. (2017) was the only included study that examined maternal self-report prevalence to meconium samples [35]. In this study, overall concordance between self-report of 3rd trimester cannabis use and cannabinoid metabolites in meconium samples was low (Kappa = 0.30). In fact, 2 women who reported daily use during the 3rd trimester of pregnancy were negative for meconium cannabinoid metabolites.

Type of self-report

Health care provider screening

Two studies relied on health care provider verbal screening for self-reported cannabis use. Chang et al. (2017) recorded first obstetric visits for assessment of disclosure of cannabis use via health care provider verbal screening and found that 74% of patients who tested positive for cannabis did not disclose use [31]. Another study assessed disclosure of cannabis use to a health care provider and found fair agreement between self-report and umbilical cord homogenate (kappa 0.27, 95% CI 0.02–0.51) [21].

Interview (structured or semi-structured)

A total of 5 studies utilized structured or semi-structured interviews to assess self-reported cannabis use [17, 33,34,35,36] and found poor to moderate agreement with estimates via biochemical sampling. One study assessed self-report using a semi-structured interview conducted by trained midwifery students via the French version of the 5th Edition of the Addiction Severity Index and found a low level of agreement between self-report and meconium sampling (Kappa = 0.30) [35]. Another study by Bessa et al. (2010) found that of pregnant adolescents testing positive for cannabis, none disclosed use [34]. Williams et al. (2020) utilized the Alcohol, Smoking and Substance Involvement Screening Test (ASSIST) for assessment of self-reported cannabis use and found the positive predictive value to be only 37.6%. Garg et al. (2016) found that nearly 60% of participants disclosed cannabis use [17]. Importantly, in this study, trained researchers with no clinic affiliation interviewed patients at a clinic serving patients with a current or past history of substance use [17]. Yonkers et al. (2011) had researchers perform intake assessments in a sample of pregnant women who reported substance use and found that the agreement between self-report and urine toxicology was moderate (Kappa = 0.74).

Self-administered questionnaires

Three studies used a written survey to assess self-reported cannabis use [21, 23, 32]. Young-Wolff et al. (2020) found that self-reported screening correctly identified only 34% of those who had a positive urine toxicology test [23]. Klawans gave participants a written survey that assessed cannabis use and found that although 27 women (11.6%) tested positive for cannabis via universal screening, only 10 women (4.6%) reported current cannabis use [32]. Beatty et al. (2012) used audio-enhanced computer-assisted self-interview (ACASI) technology to screen for self-reported use, in which biological measures of cannabis use (both hair and urine) revealed actual prevalence of use to be 3 times higher than self-report [30].

Research and practice recommendations

Included studies had numerous recommendations for both future research and practice on this topic (Table 2). Several studies called for more research on the validity of maternal self-reported prenatal cannabis use specifically in larger samples that are more diverse to improve generalizability of findings [17, 23, 33]. Most recommendations were focused on integrating study findings into clinical practice. The most cited recommendation for clinical practice was utilization of both self-report and biochemical estimates of use to improve overall identification of cannabis use [17, 23, 24, 36, 37]. A common area of future research recommendations included identify factors associated with perinatal illicit drug disclosure and how these factors impact sensitivity and accuracy of self-report [17, 31, 38]. Several studies also recommended further research on maternal self-report using more representative samples [17, 23, 33].

Table 2 Research and practice recommendations of included studies
Full size table

Discussion

In this scoping review, we identified and synthesized contemporary evidence on the validity of maternal self-reported cannabis use during pregnancy. We found 12 studies that examined the validity of self-reported prenatal cannabis use in comparison to a biochemical estimate. Most studies were conducted in the US and conducted in either a hospital or clinical setting. The most commonly used biochemical measure used was urine testing, which leaves substantial gaps relating to the evidence on validity of self-report compared to other biochemical measures, such as hair, meconium, or umbilical cord sampling. Given the potential adverse maternal and child health effects of prenatal cannabis exposure, our findings necessitate additional research examining validity of self-reported prenatal cannabis use.

Accurate identification of women who use cannabis during pregnancy is imperative for prenatal care providers so that discussions about use and referral to treatment, if necessary, can occur. Undoubtedly, this cannot occur without utilization of valid measures of prenatal cannabis use. However, in our review, we found that self-report of prenatal cannabis use was largely unreliable. Consistent with prior studies, we found that biometric estimates found higher prevalence of prenatal cannabis use compared to self-report. Although biometric estimates of prenatal cannabis use are more resource and time-intensive compared to self-report measures [39], several included studies recommended that a combination of self-report and biochemical screening should be employed by clinicians to improve accuracy of identifying women who use or are exposed to cannabis during pregnancy [17, 23, 24, 36, 37]. Indeed, evidence supports that indirect cannabis exposure can lead to positive biochemical samples for metabolites of the drug [40].

Prior research has shown that ACASI approaches have been associated with increased disclosure of substance use [41, 42]. However, we did not find this to be true; one included study using ACASI found that biochemical estimates revealed nearly three times the amount of cannabis users as self-report [30]. Interestingly, Yonkers et al. (2011), who used interviews to assess self-reported cannabis use, had the highest agreement between self-report and urine toxicology (kappa = 0.74), which perhaps was due to their population of pregnant women from an integrated obstetrical/substance use treatment program [33]. The second highest congruence between self-report and biochemical estimates were reported in another study utilizing interviews for self-report in a clinic serving patients with a current or history of substance use and found approximately 60% disclosed use. The high level of agreement in Garg et al. (2016) is likely attributable to the absence of punitive consequences for participants in their study and perhaps the population as well [17]. Agreement between self-reported cannabis use and biochemical estimates were lowest in studies utilizing health care provider screening [21, 31]. Importantly, in several studies, women knew they would be subsequently tested for cannabis after self-reporting use. In turn, these studies may report agreement levels that are higher than typical agreement.

We found several evidence gaps, which future research should work to address. As there was only one review that examined maternal self-report to meconium samples [35], we found insufficient evidence to comprehensively examine the validity of self-report in comparison to this type of biochemical measure. We found that self-report was more reliable in populations with a current or prior history of drug use and when assessed via interviews compared to health care provider screenings and self-administered surveys. As there are many factors influencing the agreement between self-report and biochemical estimates of cannabis use, such as social norms, fear of punitive action, and metabolite detection methods, future research should aim to better understand these factors. Beyond standardized clinic protocols for screenings and discussions of prenatal cannabis use, another important point of consideration for future studies is to examine the extent to which setting, population, and health care provider characteristics are associated with the validity of self-reported prenatal cannabis use, as we did not find a single study examining this relation.

To meet the aims of this study, we determined a scoping review, as opposed to a systematic review, was the best approach for several reasons. First, scoping reviews are used to determine the breadth and depth of existing evidence on a topic through systematic identification and mapping of available evidence [26, 27]. Secondly, scoping reviews are ideal to identify any knowledge gaps as well as to pinpoint specific research questions that could be answered via a more precise systematic review [27, 43]. Accordingly, an aim of this scoping review was to determine if there was enough evidence for, and to specify the research questions of, a systematic review on this topic. Indeed, our review suggests enough evidence for a systematic review. A systematic review on this topic would be able to provide a meticulous summary of available primary research that clinicians can use to develop prenatal cannabis use screening guidelines and policies. Until such a review is undertaken, prenatal health care providers are left to navigate the inherent complexities of shifting cannabis policies and increases in prenatal cannabis use in murky waters. Specifically, our findings support a systematic review that aims to answer the following research questions:

  1. 1)

    What is the validity and reliability of self-reported cannabis use during pregnancy?;

  2. 2)

    How is the accuracy of biochemical estimates of cannabis use impacted by cannabinoid pharmacokinetics variability and metabolite detection methods (e.g., point of care testing, mass spectrometry)?;

  3. 3)

    How does accuracy of self-reported cannabis use during pregnancy vary across environmental factors (e.g., cannabis legalization, setting, health care provider traits)?;

  4. 4)

    What is the extent to which validity of self-reported cannabis use varies as a function of time between collection of self-reported and biochemical samples?;

  5. 5)

    What potential harms or adverse outcomes exist for screening of prenatal cannabis use (both self-report and biochemical estimates)? How do these harms or adverse outcomes vary across cannabis policies?

Limitations

There are a few limitations of this scoping review. First, we excluded studies not published in English, which likely resulted in failure to identify all potentially relevant studies. Second, we utilized date restrictions to capture recent studies with contemporary relevancy (e.g., delineate cannabis from other illicit substances, use non-stigmatizing language). However, by using date restrictions in the search, we may have missed in-press or recently published articles yet to be indexed. Among included studies, there was inconsistency in the reported measure for agreement, with some studies reporting agreement in the form of Cohen’s kappa and others reporting sensitivity or specificity. A future systematic review can aim to address this limitation by calculating a consistent measure of agreement across studies for comparison. Another important piece to consider when comparing biochemical estimates in comparison to self-report is the window of time between collection of the two measures. This was beyond the scope of this review but is an important question to answer in a systematic review with possible meta-analysis. Lastly, the small number of studies that used meconium or umbilical cord sampling precluded a proper synthesis of studies for those measures.

Conclusion

We conducted a scoping review to identify and map available evidence on the validity of self-reported prenatal cannabis use. We found validity of self-report was poor in comparison to biochemical estimates. Further research is urgently needed to understand and examine factors associated with the validity of self-reported prenatal cannabis use, as well as to develop valid measures of self-reported use. Additionally, a systematic review is urgently needed to guide clinical practice and policy. Until this necessary research can be conducted, clinicians should use the recommendations of prior studies as outlined above.