Incremental validity of estimated cannabis grams as a predictor of problems and cannabinoid biomarkers: Evidence from a clinical trial

Abstract

Background

Quantifying cannabis use is complex due to a lack of a standardized packaging system that contains specified amounts of constituents. A laboratory procedure has been developed for estimating physical quantity of cannabis use by utilizing a surrogate substance to represent cannabis, and weighing the amount of the surrogate to determine typical use in grams.

Method

This secondary analysis utilized data from a multi-site, randomized, controlled pharmacological trial for adult cannabis use disorder (N=300), sponsored by the National Drug Abuse Treatment Clinical Trials Network, to test the incremental validity of this procedure. In conjunction with the Timeline Followback, this physical scale-based procedure was used to determine whether average grams per cannabis administration predicted urine cannabinoid levels (11-nor-9-carboxy-Δ9-tetrahydrocannabinol) and problems due to use, after accounting for self-reported number of days used (in the past 30 days) and number of administrations per day in a 12-week clinical trial for cannabis use disorder.

Results

Likelihood ratio tests suggest that model fit was significantly improved when grams per administration and relevant interactions were included in the model predicting urine cannabinoid level (X²=98.3; p<0.05) and in the model predicting problems due to cannabis use (X²=6.4; p<0.05), relative to a model that contained only simpler measures of quantity and frequency.

Conclusions

This study provides support for the use of a scale-based method for quantifying cannabis use in grams. This methodology may be useful when precise quantification is necessary (e.g., measuring reduction in use in a clinical trial).

1. Introduction

To advance our understanding of the precipitants and effects of cannabis use, define excessive or problematic use, and develop effective interventions for problematic use, we must first establish a reliable system for measuring quantity and frequency of cannabis use. The most commonly used quantification method is a calendar-based tool designed to enhance recall, known as the Timeline Followback (TLFB) method (Sobell and Sobell, 1992; Fals-Stewart et al., 2000). The TLFB probes individuals to report on substance use (yes/no), amount of the substance (e.g., in grams or joints), and in some instances, the number of times used per day in the designated assessment time frame (i.e., past 30 days, past 90 days). Asking individuals to report on their own quantity and frequency of cannabis use is perhaps the easiest and most cost-effective method; however, it is limited by an individual's ability to recall specifics about his/her use (e.g., Schwarz, 2007) and difficulty estimating the amount of cannabis physically used (Gray et al., 2009). While the TLFB uses memory aids to enhance retrospective recall, this does not circumvent quantity estimation errors. Researchers have developed standardization systems for estimating quantity of use for some substances. For example, cocaine and heroin quantity are often estimated by having participants report the amount of money spent on the substance per day (Ehrman and Robbins, 1994). Alcohol use is often reported with reference to a pre-defined “standard drink” based on the approximate ethanol content (Kalinowski and Humphreys, 2016).

Reliable quantification of cannabis use is particularly difficult because of multiple modes of preparations, variations in the amount used for each preparation, strength (i.e., amount of THC/psychoactive constituents, often referred to colloquially as “potency”), and the number of others sharing for a particular administration. Additionally, cannabis is not obtained in a standardized amount as are alcohol and cigarettes (Gray et al., 2009). To overcome this, Mariani and colleagues (2011) used a surrogate substance (oregano) in concert with a traditional 30-day TLFB to estimate how much cannabis individuals used during each episode of use. Individuals placed an amount of oregano that represented their typical quantity of use into a pipe or rolling paper/leaf cigar wrappers, depending on their typical methods of use. The oregano was then weighed on a scale to obtain typical quantity in grams. Given the added time and cost associated with this detailed quantification procedure, it is important to determine whether quantity (gram estimation) provides incremental predictive validity above and beyond frequency of cannabis use and simpler methods for assessing quantity of cannabis use.

Prior work has demonstrated that quantity of cannabis used significantly predicts cannabis problems and dependence, even after accounting for frequency of use (Norberg et al., 2012; Walden and Earleywine, 2008; Grant and Pickering, 1999; Zeisser et al., 2012); however, the detailed quantification procedure did not significantly add incremental validity to a single-item measure of quantity when predicting problems due to cannabis use (ΔR²=0.05; Norberg et al., 2012). To our knowledge, whether quantity (in grams) or Mariani and colleagues' (2011) estimation procedure incrementally predicts quantitative urine cannabinoid levels has not been examined. If self-reported quantity of use reflects the degree of physiological exposure to cannabis use, it can be used to understand dose-specific effects on health and neurocognitive functioning. The association between self-reported cannabis use and biomarkers is limited due to variation in the bioavailability of cannabis. However, we can conclude that improvement in prediction of a cannabis biomarker (i.e., quantitative urine cannabinoid level) means that the added specificity in self-reported cannabis use results in a more accurate indicator of physiological exposure to cannabis. Thus, the goal of this study is to replicate and extend work by Norberg et al. (2012) and determine whether self-reported quantity (grams) of cannabis use, facilitated by use of the surrogate substance, predicts 1) quantitative urine cannabinoid levels, and 2) problems due to cannabis use, after accounting for self-reported frequency (number of days) of cannabis use and number of joints/blunts/other methods of administration per day. It is hypothesized that use of the surrogate substance to facilitate report of cannabis use quantity will provide incremental validity beyond number of days used and number of joints/blunts when predicting urine cannabinoid level, but not problems due to use, consistent with findings from Norberg and colleagues (2012).

2. Method

2.1 Participants

Adults (N=302) ages 18 to 50 who were seeking treatment for cannabis use disorder (CUD) were recruited for a multisite clinical trial sponsored by the National Drug Abuse Treatment Clinical Trials Network (NIDA CTN) using community/media advertisements (Clinicaltrials.gov: NCT01675661). Applicants were eligible if they provided a positive urine cannabinoid test at screening, endorsed criteria for Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR) cannabis dependence, were interested in treatment for CUD, and, if female, agreed to use birth control. Applicants were excluded if they met DSM-IV-TR substance dependence other than cannabis or tobacco, provided a urine drug screen positive for non-cannabinoid substances, recently used synthetic cannabinoids, were currently using or allergic to N-acetylcysteine (due to aims of the clinical trial), were in treatment for substance use, had asthma, were pregnant or breastfeeding, or had any uncontrolled medical or psychiatric illness. Two participants were excluded from analyses due to missing scale data (for computation of grams), resulting in an analytic sample of N=300. The average age of participants was 30.3 (SD=9.0) and the sample was 71.7% male, 58.7% White, 27.3% Black or African American, and 30.3% unemployed and looking for work. Additional characteristics of the full sample are provided in Gray et al. (2017).

2.2 Procedures and Measures

Study procedures and results from the primary clinical trial have been described in detail elsewhere (McClure et al., 2014; Gray et al., 2017). The multi-site study was approved by Institutional Review Boards at each study site prior to data collection.

Briefly, participants were randomized to receive N-acetylcysteine (2400 mg/day) or matched placebo for 12 weeks. All participants received abstinence-based contingency management in addition to medication or placebo. At an initial screening visit, pre-treatment visit, weekly study visits, and at one month follow-up, participants provided urine samples for quantitative cannabinoid testing. Urine drug screens were collected twice per week, but quantitative testing was conducted only on the first sample obtained per week. They also reported on cannabis use in the past 30 days (at the screening visit) and in between study visits via the TLFB. Data from the initial screening visit, pre-treatment visit, and weekly study visits are included in the current analysis.

2.2.1 Self-Report of Cannabis Use/Gram Estimation

At the initial screening visit, participants were asked to complete a Timeline Follow-Back (TLFB; Sobell and Sobell, 1992) to assess frequency and quantity of past 30 day cannabis use prior to study initiation. For each day, participants reported whether they had used cannabis (yes/no) and the number of joints, blunts, pipes, bowls, vaporizers, spliffs, edibles, or other methods used. If participants shared a joint/blunt/etc. or otherwise did not use a full joint/blunt/etc., partial numbers were reported. Participants were then provided with rolling papers and dried motherwort. For each method of cannabis use (e.g., joints, blunts) that the participant reported in the previous 30 days, they were asked to place an amount of motherwort (in place of the oregano used by Mariani et al., 2011) that represented their typical quantity of use into rolling papers or directly on the scale, depending on their typical methods of use. The motherwort was weighed on a scale to obtain typical quantity in grams.

At subsequent visits, the scale estimation procedure was only repeated if a participant reported a new mode of use (i.e., used blunts since last visit, but had not reported any blunt use at initial screening visit). Otherwise, participants reported only their daily cannabis use in between visits (yes/no) and the number of joints/blunts/etc. used. See Figure 1 for an illustrative example of daily gram calculations for a hypothetical participant.

Figure 1. Example Computation of Daily Grams for a Single Participant.

Note. Number of units used is multiplied by the number of grams endorsed for that modality at screening.

2.2.2 Quantitative Urine Cannabinoid Level

Urine cannabinoid samples were collected at the screening visit, pre-treatment/randomization visit, and weekly during treatment. Urine cannabinoid (11-nor-9-carboxy-Δ9-tetrahydrocannabinol) was batch assayed in thawed frozen (-80C) samples using an enzyme immunoassay (Abbott Laboratories) on an Architect Autoanalyzer (Abbot Labs) in the Clinical Neurobiology Labs at the Medical University of South Carolina. The lowest quantifiable amount was 10 ng/mL and levels 20 ng/mL or above were reported, while values above 200 ng/mL were diluted to provide a quantifiable amount. The inter-assay coefficient of variability (CV) for two controls run with each assay were 11.2% (low) and 5.9% (high) respectively. Urine creatinine was also measured to provide an estimate of sample dilution, as previous research has shown that failure to account for sample dilution may lead to misinterpretation (Huestis and Cone, 1998; Lafolie et al., 1991). Because there is some deliberation regarding whether normalizing cannabinoid level through the use of a cannabinoid-creatinine ratio is the optimal way of accounting for dilution (Mikulich-Gilbertson, 2016), we instead used creatinine as a covariate in relevant models in which unadjusted cannabinoid level was the outcome variable. Creatinine-normalized cannabinoid values are presented for descriptive purposes only in Tables 1-2.

Table 1. Baseline Quantitative Cannabinoid Level and Self-Report of Past 30-Day Cannabis Use (N=299).

	M	SD
Unadjusted Cannabinoid	1060.0	1367.0
Creatinine-Normalized Cannabinoid	927.1	1130.0
Cannabis Use Days	26.0	6.2
Jointsa Per Day	3.4	2.8
Estimated Grams Per Jointa	0.7	0.6
Estimated Grams Per Day	2.5	3.4

Note: Baseline measures were unavailable for one participant; Creatinine-Normalized Cannabinoid (ng/mg; metabolite/creatinine) =quantitative urine cannabinoid level (ng/mL) divided by creatinine (mg/mL) (Huestis and Cone, 1998);

^aIncludes all methods of administration (e.g., joints, blunts, bowls, bongs).

Table 2. Pearson Correlations Between Past 30-Day Cannabis Use (TLFB) and Normalized and Unadjusted Urine Cannabinoid Measures at Baseline (N=299).

	Cannabis Use Days	Joints Per Daya	Estimated Grams Per Jointa	Estimated Grams Per Day
Unadjusted Cannabinoid	0.15*	0.22*	0.22*	0.26*
Creatinine-Normalized Cannabinoid	0.17*	0.37*	0.02	0.20*
Jointsa Per Day	0.45*	--	--	--
Estimated Grams Per Jointa	0.09	0.07	--	--
Estimated Grams Per Day	0.29*	0.62*	0.68*	--

Note:

^*p<0.05; Baseline measures were unavailable for one participant; Unadjusted Cannabinoid= quantitative urine cannabinoid levels unadjusted for creatinine levels; Creatinine-Normalized Cannabinoid (ng/mg; metabolite/creatinine)=quantitative urine cannabinoid level (ng/mL) divided by creatinine (mg/mL) (Huestis and Cone, 1998);

^aIncludes all methods of administration (e.g., joints, blunts, bowls, bongs).

2.2.3 Cannabis-Related Problems

The Marijuana Problems Scale (MPS; Stephens et al., 2000) is a 19-item self-report measure designed to assess problems related to cannabis use in the past 30 days. Participants responded to each item on a 3-point scale with “0” indicating no problems, “1” indicating a minor problem, and “2” indicating a serious problem. Consistent with the scoring procedure used by Stephens et al. (2000), any items scored as a “1” or “2” were tallied to create a total problem count, irrespective of the severity of endorsement. The range of possible scores was 0-19. The MPS was administered to participants at screening, weeks 4 and 8 of treatment, and at the end of treatment (week 12). If a study visit was missed, the MPS was completed at the next visit.

2.3 Data Analytic Procedure

Multilevel modeling (MLM) was used (2-level random effects model) to examine whether average grams of cannabis used per administration during the 30 days prior to collection of the urine cannabinoid test explained additional variance in unadjusted quantitative cannabinoid level when age, gender, body mass index (BMI), urine creatinine level, number of days since treatment start, number of cannabis use days, and average number of cannabis administrations per day were included in the model. These models are robust to missing data and account for non-independence of observations (since observations are nested within individual). Lower level predictors varied within-subject over time and included urine creatinine, days since treatment start, and days with cannabis use in the 30 days prior to the urine cannabinoid test. Interactions of interest between clinical variables and gram estimates were assessed. Upper level (participant level) predictors varied only between-participants and included gender, age, and BMI at study start. Residual Maximum Likelihood (REML; Patterson and Thompson, 1971) was used to produce unbiased variance estimation. A random intercept was estimated in all models and a random slope included creatinine and an unstructured covariance pattern was used.

Unadjusted quantitative cannabinoid level values (y variable) were natural log-transformed to correct for non-linearity between the predictors (x variables) and y variable. The resulting mean of residuals was approximately 0. Because there was significant inflation of endorsement of no problems on the Marijuana Problem Scale (MPS) as the trial progressed, multilevel zero-inflated Poisson models were used to estimate the probability of endorsing zero problems (logistic part) and the number of problems when endorsed (Poisson part). Lower level predictors varied within-subject over time and included the days since treatment start and frequency of cannabis use in the prior 30 days. The 30 day time frame was chosen because it is an interval frequently used in TLFB studies. Upper level (participant level) predictors again included gender, age, and BMI.

In all models, predictors were not grand or group-mean centered in order to maintain a meaningful zero point for interpretation of parameter estimates. All available data were incorporated, with a maximum of 14 urine quantitative cannabinoid assessments per participant and a maximum of 4 time points per participant with completion of the MPS. Statistical analyses were conducted using SAS version 9.4 (SAS Institute Inc., 2012) and no corrections for multiple testing were conducted.

Incremental validity was established by comparing model fit (-2 log likelihood) of nested models using the likelihood ratio test.

3. Results

3.1. Missing and Censored Data/Outliers

Mean quantitative cannabinoid levels and self-reported cannabis use at baseline are presented in Table 1. Though a baseline quantitative urine cannabinoid was unavailable for one participant (not included in Tables 1 and 2), all available data from this participant was used in subsequent analyses. Urine cannabinoid levels greater than 4.0 standard deviations above the mean of all samples at all time points (M=863.70; SD= 1612.70)¹ occurred in 29 instances (0.87% of UDS values) and were replaced with this upper limit for all formal analyses (new value=7314.50) as extreme outliers likely contain more measurement error. Urine cannabinoid levels and creatinine levels below the lower limit of detection were replaced with a value equal to the lower limit of detection divided by the square root of 2, as this simple replacement method has been shown to produce the lowest error rate (Croghan and Egeghy, 2003). Across the entire trial, 547, or 16.45% of values were below 10 ng/mL. In instances where the values were below 10 ng/mL, the majority of participants reported no cannabis use (86.1%) or 1-2 days of cannabis use (9.1%) in the two weeks prior to the sample. An additional small percentage of samples (N=20, 0.60% of UDS values) were below 30 ng/mlL, but precise quantitative estimates could not be verified and 30 ng/mL was treated as the lower limit of detection.

For a small number of participants (n=4), it was not possible to estimate grams for cannabis ingested via edibles (cannabis-containing brownie, cookie, etc.) via the scale. These four participants reported a cumulative 10 days (Range=1-6 days per participant) of edible consumption on the TLFB. Edible use was not included in the total gram count for these days; however, the majority of these instances (9 out of 10) were accompanied by other methods of use. Pearson correlations between quantitative cannabinoid level and self-reported cannabis use are presented in Table 2.

Missing items on the MPS were replaced with the mean of all other items, rounded to the nearest whole number. The maximum number of items imputed per assessment was 3 (15.8%) An average of 0.01 items were missing per MPS assessment (SD=0.12, Range=0-3).

3.2. Quantity of Cannabis Use Predicting Urine Cannabinoid Level

Average grams per administration 30 days prior to the urine cannabinoid test showed incremental validity, as evidenced by a significant likelihood ratio test indicating improvement in model fit when average grams per administration and a relevant interaction was added to the model (X² = 98.3). Model fit (-2 log likelihood) improved when grams per joint/etc. and the interaction between grams per joint/etc. and days since treatment start were included in the model (Model 3A Table 3). A parallel improvement in model fit was not evident when joints/etc. per day were added to the model (Model 2A in Table 3).²

Table 3. Frequency of Cannabis Use (Past 30 Days), Joints/Blunts/Etc. per Day (Past 30 Day Average), and Average Grams per Method of Administration Predicting Unadjusted Quantitative Cannabinoid Level.

	Frequency Only Model		Frequency + Joints/Daya		Frequency, Joints/Daya, + Grams/Jointa
	Model 1A		Model 2A		Model 3A
	b	SE	b	SE	b	SE
Fixed effects
Intercept	3.39*	0.42	3.41*	0.41	3.36*	0.40
Tx Daysb	<0.01	<0.01	<.01	<.01	<-.01	<0.01
# of Days Used MJ	0.07*	<0.01	0.07*	<.01	0.06*	<0.01
Avg # of Jointsa/Day			0.03	0.02	0.05*	0.02
Avg # of Grams/Jointa					0.25*	0.07
Tx Daysb*Grams/Jointa					0.01*	<0.01
-2 Log Likelihood	9811.7		9815.1		9716.8
Likelihood Ratio Test			X2 = 3.40; ns (vs. Model 1A)		X2 = 98.3* (vs. Model 2A)

Note.

^*p<0.05; Beta estimates (b) are unstandardized. All models included creatinine, age, gender, and body mass index as covariates (not shown).

^aJoints refers to joints, blunts, pipes, and all other methods of administration.

^bTx days=the number of days since treatment began; ns= nonsignificant decrement in model fit.

See Table 3 for parameter estimates and results of the full model. Models are first presented with only frequency as a predictor, then quantity (number of joints/etc. per day) and frequency as predictors, and finally, with estimated grams per joint/etc. in the model to show the incremental validity of adding estimated grams per joint/etc. and relevant interactions. After considering variance explained by age, gender, BMI, and creatinine level, results showed that the number of days used (frequency) and average number of joints/blunts/etc. per day significantly predicted urine cannabinoid levels 30 days prior to collection of the urine cannabinoid test. Additionally, an interaction was present between estimated grams per administration and days since treatment start, suggesting that the more time that had elapsed since randomization, the stronger the relationship between grams per administration and urine cannabinoid level. Urine creatinine was a significant predictor of urine cannabinoid level. Age, gender, and BMI did not show significant associations with urine cannabinoid level in any model.

3.3. Quantity of Cannabis Use Predicting Cannabis Use Problems

The addition of joints/etc. per day to the model did not improve model fit (Model 2B in Table 4) when predicting problems due to use, but the inclusion of average grams per administration did significantly improve model fit (Model 3B in Table 4) per the likelihood ratio test (X² = 6.4).

Table 4. Frequency of Cannabis Use (Past 30 Days), Joints/Blunts/Etc. per Day (Past 30 Day Average), and Average Grams per Method of Administration Predicting Number of Cannabis Problems (Zero-Inflated Poisson Multilevel Model).

	Frequency Only Model		Frequency + Joints/Daya		Frequency, Joints/Daya, + Grams/Jointa
	Model 1B		Model 2B		Model 3B
	b	SE	b	SE	b	SE
Logistic Part- Predicting No Problems
Intercept	0.34	0.90	0.33	0.90	0.40	0.91
Tx Daysb	0.01*	0.01	0.01*	0.01	0.01*	0.01
# of Days Used MJ	-0.25*	0.07	-0.27*	0.08	-0.23*	0.07
Avg # of Jointsa/Day			0.22	0.36	0.17	0.34
Avg # of Grams/Jointa					-0.40	0.34
Poisson Part- Predicting Number of Problems
Intercept	0.76*	0.25	0.75*	0.25	0.70*	0.25
Tx Daysb	-0.01*	<0.01	-0.01*	<0.01	-0.01*	<0.01
# of Days Used MJ	0.02*	<0.01	0.02*	<0.01	0.02*	<0.01
Avg # of Jointsa/Day			<-0.01	0.01	<-0.01	0.01
Avg # of Grams/Jointa					0.12*	0.06
-2 Log Likelihood	4402.8		4402.4		4396.0
Likelihood Ratio Test			X2 = 0.4; ns (vs. Model 1B)		X2 = 6.4* (vs. Model 2B)

Note.

^*denotes p-values less than 0.05; Beta estimates (b) are unstandardized. Only intercept modeled as random. Logistic portion is predicting zero problems, with higher values indicating a greater likelihood of endorsing zero problems. The Poisson portion predicts number of problems, with higher values indicating more problems.

^aJoints refers to joints, blunts, pipes, and all other methods of administration.

^bTx days=the number of days since treatment began; ns=non-significant improvement in model fit.

The average problem count for the MPS at baseline was 6.96 (SD=4.16)³. As shown in Table 4, frequency (number of days) of cannabis use was a significant predictor of the presence of problems due to cannabis use (logistic) and the number of problems endorsed (Poisson), but the number of joints/blunts/etc. per day did not significantly predict problems (logistic or Poisson). Average grams per joint/blunt/etc. was a significant predictor of more problems due to use (Poisson), but not the presence or absence of problems (logistic). Age, gender, and BMI did not show significant associations with presence of problems or number of problems endorsed in the full models.

4. Discussion

This study examined whether estimation of grams per administration of cannabis using a surrogate substance (Mariani et al., 2011) predicts quantitative cannabinoid level and cannabis use problems, after accounting for frequency of cannabis use and average number of administrations (joints/blunts/etc) per day. We found that a model that included estimated grams per method of administration (e.g., joint, blunt) fit the data significantly better than a model including only self-reported number of days used and number of administrations per day when predicting both outcomes: quantitative urine cannabinoid level and problems due to cannabis use.

Estimated grams per administration significantly predicted increases in quantitative cannabinoid level, with number of days used and number of joints/blunts/etc. per day in the model. The significant interaction between days since start of treatment and estimation of grams per administration may suggest that at lower levels of use, estimated grams per administration becomes more important in predicting cannabinoid biomarkers. Though there was not a significant effect of treatment in the primary study (Gray et al., 2017), decreased frequency of use was evident in the full sample (Hser et al., 2017); therefore, later in the trial, participants used cannabis less frequently. Although speculative, it is possible that at near daily use, urine cannabinoid levels will be high regardless of the grams smoked per administration, but that this measure has more value at lower and more moderate levels of use. In sum, because quantitative urine cannabinoid testing cannot provide specific information regarding the timing, quantity, and frequency of substance use, enhancing the accuracy of self-reported use is necessary to complement use of biomarker data in clinical trials. Mariani and colleagues' (2011) scale method for gram estimation may serve as one way to partially improve the accuracy of estimates of cannabis quantity obtained via self-report.

Results related to cannabis problems are consistent with previous literature suggesting that frequency of cannabis use is an important predictor in the development of problems due to use (Norberg et al., 2012). Inconsistent with other findings (Norberg et al., 2012; Walden and Earleywine, 2008; Grant and Pickering, 1999; Zeisser et al., 2012), a simple quantity measure (number of joints/blunts/etc. per day) did not statistically predict cannabis use problems in our sample. Likewise, estimated grams per administration was also not a consistent predictor of problems. Grams per administration did not predict the presence or absence of problems, but the number of problems endorsed, when they were endorsed. The finding that average grams per administration in the 30 days prior predicts the number of problems endorsed suggests that more precise information about quantity of use, including an estimate of grams per joint/blunt/etc., may have some utility, however. Further, the incremental validity of including grams per joint/blunt/etc. was suggested by improvement in model fit.

Grams per administration is an estimate of amount of cannabis used per episode, which may be related to increased tolerance or heightened risk for intoxication, resulting in more problems due to use. Perhaps problems resulting from cannabis use are slower to change following reductions in use. Replication in treatment and non-treatment seeking samples using various measures of problematic use will be necessary to better understand the relationship between grams per administration and problematic use.

Results must be interpreted with consideration of several limitations. First, the sample was restricted to treatment-seeking adults who met criteria for CUD. Therefore, our sample does not represent the entire possible continuum of problematic cannabis use. Nevertheless, the fact that quantity of cannabis used predicted outcomes within this more homogeneous sample provides evidence of a relationship between gram estimation and biochemical measurement and, to a lesser extent, clinical outcomes. Further, comparing self-report and biomarkers across the duration of a clinical trial provides more variability in cannabis use behavior, as many participants reduced their use over the course of the trial (Hser et al., 2017).

Second, because TLFB assessments were completed at each visit to account for days between visits, the time period over which participants were asked to recall their use was reduced (except at baseline). Thus, it is possible that estimated grams only adds predictive value when individuals are asked to recall amount of use over shorter periods of time.

Third, estimation of grams per joint, blunt, and other methods of administration was conducted using the scale only at the initial screening visit (baseline). In subsequent reports, participants referenced their “typical” joint/blunt/etc. weight, as determined at this initial time point. However, if throughout the course of treatment, participants reduced their typical grams per administration method or otherwise deviated from the initial measurement, this was not captured. Differences between density of individual participants' cannabis products and our surrogate substance, motherwort, may have also impacted our estimation of grams.

Fourth, we were not able to quantify grams using the scale for some instances of edible use, resulting in a slight underestimation of grams for a small percentage of days. Additionally, gram estimation cannot fully account for variation in inhalation and strength (Gray et al., 2009; Freeman, Morgan, et al., 2014; van der Pol et al., 2014), and the ratio of delta-9-tetrahydrocannabinol (THC) to cannabidiol, which may be an important contributing factor in the development of problems due to cannabis use (Morgan et al., 2014; Freeman et al., 2014). To date, there is not a gold standard for quantifying cannabis use beyond analyzing THC content from cannabis samples. However, legal and ethical issues preclude participants from providing researchers with samples of their own cannabis in locations in which cannabis is illegal. Additionally, variation in sample THC content used by a single individual across episodes of use would still be likely. Researchers have asked individuals to provide the cost or dollar amount associated with cannabis (e.g., Mariani et al., 2011). Assuming that higher strength variations are more expensive, this may be a way to account for strength. However, other factors are also related to cost, such as local regulations (e.g., Pacula et al., 2010). Others have proposed methods that simply ask individuals to rate the strength of each cannabis administration (Gray et al., 2009; Freeman et al., 2014; Morgan, et al., 2014), and for more regular users, strength ratings are significantly associated with actual THC content (Freeman et al., 2014; Morgan, et al., 2014). Future research should combine gram estimation with measures of cannabis strength to determine if this provides more accurate estimation of cannabinoid biomarkers and/or problems due to use.

Finally, our measures used to validate the gram estimation procedure are also imperfect measures. Self-report of problems is subject to inherent biases. Additionally, the urine cannabinoid marker is not a direct measure of circulating levels of THC, but is rather a secondary metabolite of THC as it is excreted in the urine. Factors that impact metabolism of THC (e.g., liver disease) will impact the urine cannabinoid measure. A number of other factors also impact how much THC circulates in the blood during acute administration, including the cannabidiol/THC ratio (Nadulski et al., 2005).

5. Conclusion

Despite the limitations discussed prior, this study suggests that precise estimation of grams of cannabis per administration by weighing a surrogate substance in the laboratory adds to the predictive validity of simpler measures of quantity and frequency of cannabis use in predicting biochemical outcomes, but inconsistently predicts problems due to use. For researchers to begin to establish meaningful cut-offs for high-risk cannabis use, more precise estimation may be beneficial to the extent that it more accurately represents cannabinoid levels. Researchers may use grams per episode to determine clinical cut-offs for high-risk episodic use in terms of “standard joints” (Zeisser et al., 2012), similar to cut-offs developed in the alcohol literature suggesting that 4 or more standard drinks for a female or 5 or more standard drinks for a male during one alcohol use episode may confer greater health risk (Wechsler et al., 1995). Lastly, precise quantification of cannabis use offers some relative advantages over urine cannabinoid biomarker data. It can be adapted for remote data collection, and it is better suited to detect variability in use patterns. To optimally understand use patterns and adapt interventions accordingly, we need to be able to quantify how much cannabis is being used and when. Future research may also explore options for quanitifying cannabis use using remote, real-time assessments. For example, individuals may use a Wi-Fi based scale to measure their own cannabis use in grams and report on their use in real-time via a mobile device to reduce error in retrospective recall. This may also help circumvent the issue of varying grams per administration as individuals attempt to reduce their cannabis use.

Highlights.

• Cannabis use is difficult to quantify due to variability in products/modes of use

• A method for estimating grams of cannabis use has been previously developed.

• The method predicts outcomes beyond simpler frequency counts of cannabis use alone.

• This method may be refined further, including estimates of cannabinoid strength.