Correlations and Agreement Between Delta9tetrahydrocannabinol (THC

bs_bs_banner

RESEARCH REPORT doi:10.1111/j.1360-0443.2011.03757.x

Correlations and agreement between delta-9-tetrahydrocannabinol (THC) in blood plasma and timeline follow-back (TLFB)-assisted self-reported use of cannabis of patients with cannabis use disorder and psychotic illness attending the CapOpus

randomized clinical trialadd_3757 1123..1131

Carsten Rygaard Hjorthøj1, Allan Fohlmann1, Anne-Mette Larsen1, Mikkel Arendt2 & Merete Nordentoft1 Mental Health Centre Copenhagen and Faculty of Health Sciences, University of Copenhagen, Copenhagen NV, Denmark1 and Unit for Psychiatric Research, Aalborg Psychiatric Hospital, Aarhus University Hospital, Aalborg, Denmark2

ABSTRACT

Aims To assess correlations and agreement between timeline follow-back (TLFB)-assisted self-report and blood samples for cannabis use. Design Secondary analysis of a randomized trial. Setting Copenhagen, Denmark. Participants One hundred and three patients from the CapOpus trial with cannabis use disorder and psychosis, providing 239 self-reports of cannabis use and 88 valid blood samples. Measurements Delta-9-tetrahydrocannabinol (THC), 11-hydroxy-delta-9-tetrahydrocannabinol (11-OH-THC) and 11-nor-delta-9-tetrahydrocannabinol-9- carboxylic acid (THC-COOH) detected in plasma using high-performance liquid chromatography with tandem mass spectrometry detection. Self-report of cannabis-use last month by TLFB. Pearson’s r, sensitivity and specificity calculated as measures of correlation or agreement. Findings Correlations were strong; r = 0.75 for number of days and r = 0.83 for number of standard joints in the preceding month when excluding outliers. Including outliers, coefficients were moderate to strong (r = 0.49). There were differences in subgroups, mainly inconsistent, depending on inclusion or exclusion of outliers. Sensitivity and specificity for TLFB detecting the presence or absence of cannabis use were 95.7% [95% confidence interval (CI) 88.0–99.1%) and 72.2% (95% CI 46.5–90.3%), respectively. Using 19 days as cut-off on TLFB, they were 94.3% (95% CI 86.0–98.4%) and 94.4% (95% CI 72.2–99.9%), respectively.Area under the receiver operating characteristic (ROC) curve was 0.96. Conclusions Timeline follow-back (TLFB)-assisted self-report of cannabis use correlates highly with plasma-delta-9-tetrahydrocannabinol in patients with comorbid cannabis use disorder and psychosis. Sensitivity and specificity of timeline follow-back appear to be optimized with 19 days as the cut-off point. As such, timeline follow-back may be superior to analysis of blood when going beyond 19 days of recall.

Keywords Agreement, cannabis, correlation, dual diagnosis, psychosis, THC, timeline follow-back, validity.

Correspondence to: Carsten Rygaard Hjorthøj, Mental Health Centre Copenhagen and Faculty of Health Sciences, University of Copenhagen, Bispebjerg Bakke 23, Building 13A, 2400 Copenhagen NV, Denmark. E-mail: [email protected] Submitted 12 August 2011; initial review completed 21 September 2011; ﬁnal version accepted 6 December 2011

INTRODUCTION disorders [2]. Cannabis use in psychotic populations is associated with poorer prognosis and treatment outcome Cannabis is the most widely used illicit substance in the [3–7]. As such, both determination and quantiﬁcation of world [1], and its use and abuse are especially prevalent the substance is of high importance to practitioners and in psychiatric populations, e.g. patients with psychotic researchers. Two common ways to assess cannabis use

© 2011 The Authors, Addiction © 2011 Society for the Study of Addiction Addiction, 107, 1123–1131 1124 Carsten Rygaard Hjorthøj et al. are retrospective self-report and biological matrices such self-reported use of cannabis. The TLFB (and other as blood or urine. Instruments and techniques exist that assessment instruments) was administered by a trained aim to improve reliability of self-reported cannabis use. interviewer. One such instrument is the manual-based timeline Other information was collected at these three inter- follow-back (TLFB), which uses a visual calendar to views, and in the present investigation these are used to enhance recall [8]. In a systematic review and meta- conduct subgroup analyses. This included measures of analysis we showed that TLFB validly assesses illicit sub- psychopathology (Positive and Negative Syndrome Scale stance use compared to biological samples [9]. However, for Schizophrenia—PANSS) [18], Manchester Short particularly for cannabis, some studies reported sub- Assessment of Quality of Life (MANSA) [19], the World optimal agreement between TLFB and urine [10,11]. In Health Organization’s Disability Assessment Schedules addition, most of the validation in the review was based (WHODAS-II) [20] and cognitive tests including the on determination rather than quantification of cannabis Hopkins Verbal Learning Test (HVLT) [21] and Trail use, and no studies used blood for validation, even though Making Tests A and B [22]. blood holds advantages over both urine and hair for this purpose [12]. Finally, the meta-analysis showed slightly Self-reported cannabis use lower agreement rates in the presence of psychiatric By means of the TLFB instrument, self-reported cannabis comorbidity. use was assessed for the past 30 days. This length of time Training raters to use a manual-based instrument was appropriate in terms of the randomized trial, in such as TLFB is cheaper than using blood. Blood samples which the primary aim was to reduce cannabis use. For require the use of a laboratory, and often storage facilities that purpose, a longer retrospective period would not such as freezers. Also, cannabis is detectable even in have been of interest. Patients were asked to provide blood for a limited time, varying from hours to at least a information on events such as birthdays, parties, out- week post-abstinence in chronic users [13,14]. Similarly, patient treatment visits, visits to the dealer, visits to or 11-nor-delta-9-tetrahydrocannabinol-9-carboxylic acid from friends, etc. This information was entered into the (THC-COOH) detection times have been reported as TLFB calendar and used to enhance recall by use of ques- ranging in one study from 3.5 to 74.3 hours [15] to 25 tions along the lines of: ‘on the day you visited your days in a case study of a chronic cannabis user [12]. mother two weeks ago, did you use any cannabis that Conversely, self-report is limited only by respondents’ day? And how about the days before and after, do you memory,which may allow for longer retrospective assess- remember?’. Information on cannabis use was sought to ment. Finally, even though formulas have been developed be specific regarding amount (e.g. number of joints) and to establish the quantities and frequencies of cannabis type (e.g. cannabis resin, sinsemilla). Only as a last resort use from blood [12,16], such information would be were typical amounts, frequencies or guesses used. From readily available from TLFB. this information, it was possible to establish number of The aim of the present study was to assess the corre- days with cannabis use in the preceding month and total lations and agreement between TLFB and blood samples amount of cannabis used. For the latter, amounts were for cannabis use in patients with comorbid cannabis use converted into ‘standard joints’, defined as 0.5 g of can- disorders and psychosis [17]. nabis resin. For more potent types of cannabis, e.g. sinsemilla, a factor of 1.5 was applied—this was the case for METHODS 13 subjects at baseline, three post-treatment and four at follow up. Subjects comprised the 103 participants in the CapOpus trial, full details of which regarding design and overall Blood samples results are reported elsewhere [17]. These patients provided 239 self-reports of cannabis use and 88 valid Each blood sample was obtained with a Vacuette® blood blood samples. Briefly, CapOpus was a randomized trial collection set into a 9-ml Vacuette® tube containing of a psychosocial intervention to treat cannabis abuse in potassium (K3) ethylenediamine tetraacetic acid (EDTA). people with both cannabis use disorder and psychosis Each sample was put immediately into a centrifuge to (the F2 section of ICD-10). Patients were randomized to isolate plasma, which was stored subsequently in a -80°C either treatment as usual (TAU) or CapOpus plus TAU freezer. Samples were analysed at Odense University Hos- for 6 months. Patients were interviewed at baseline, pital for delta-9-tetrahydrocannabinol (THC) and the post-treatment and at 4-month follow-up. Assessments metabolites 11-hydroxy-delta-9-tetrahydrocannabinol included a detailed TLFB of self-reported cannabis use (11-OH-THC) and THC-COOH using high-performance in the past 30 days. Patients were also asked (but liquid chromatography with tandem mass spectrometry not required) to provide a blood sample to validate detection (LC/MS/MS) [23–25]. Values were rounded to

© 2011 The Authors, Addiction © 2011 Society for the Study of Addiction Addiction, 107, 1123–1131 TLFB versus THC in plasma 1125 the nearest integer, except for THC and 11-OH-THC false negatives (FN: negative TLFB, positive blood), true values below 0.5 ng/ml; as the detection limit reported by negatives (TN: both TLFB and blood negative), false posi- the laboratory was 0.2 ng/ml, values of these molecules tives (FP: positive TLFB, negative blood) and true positives below 0.2 ng/ml were rounded to 0 ng/ml, and between (TP: both TLFB and blood positive). From these propor- 0.2 ng/ml and 0.5 ng/ml were kept unchanged. We also TP tions we also calculated sensitivity (), specificity calculated the cannabis influence factor (CIF) [12], which TP+ FN is used in Germany to interpret acute effects in cases of TN TP driving while intoxicated, and with higher CIF indicating (), positive predictive value ()and nega- FP+ TN TP+ FP THC[] ng/ ml11−− OH THC[] ng/ ml TN + tive predictive value ()of TLFB. Exact binomial recent use: CIF = 314. 5 330. 5 . FN+ TN THC− COOH[] ng/ ml ×0.01 confidence intervals (CIs) were estimated for these four 344. 5 parameters. From sensitivity and 1-specificity, we drew Finally, we estimated time since last cannabis use accord- non-parametric receiver operating characteristics (ROC) ing to formulae published by Huestis et al. [16,26]. curves for various cut-off points of TLFB. Statistics

We correlated self-reported amounts of cannabis use RESULTS both as days and standard joints with amounts of THC detected in blood samples using the Pearson product– CapOpus enrolled 103 patients, all of whom participated moment correlation coefficient, r. Calculations were con- in the baseline assessment, 68 in the first follow-up and ducted on the entire data set, as well as by interview. In 68 in the second follow-up (57 of these patients also par- addition, we sought to identify subgroups with noticeable ticipated in the first follow-up), for a total of 239 inter- large or low values of r. Analyses were conducted both views. Blood samples were obtained at 90 (38%) of these including and excluding outliers detected in box-plots. interviews, but two were not analysed due to container Because number of days of cannabis use is restricted breakage, leaving 88 blood samples for LC/MS/MS analy- between 0 and 30 days, inclusive, we did not exclude sis. Those not giving blood samples reported a mean of outliers in these analyses. Number of joints and THC con- 2.4 (standard error 1.56) fewer days of cannabis use centrations are restricted downwards to 0, but theoreti- than those giving blood samples, but the difference was cally to positive infinity; hence, only large outliers in these not statistically significant, t(237) = 1.52, P = 0.13. The variables were identified formally as values above the same was true for number of monthly joints, where those third quartile + 1.5 times the interquartile range. Values not giving blood reported 8.4 (6.68) fewer monthly of r above 0.7 were determined a priori as indicating very joints, t(237) = 1.26, P = 0.21. Differences between these strong associations, values between 0.5 and 0.7 were two groups of patients were not significant within any considered strong, values between 0.30 and 0.50 were of the interviews. Furthermore, there was no difference considered moderate and anything below was considered by number of blood samples delivered throughout the weak or negligible. Analyses were carried out in STATA project by each patient on number of days of self-report version 11.2. Comparisons of correlation coefficients (F(3235) = 0.45, P = 0.71) or self-reported number of between subgroups were performed by first using the monthly joints (F(3235) = 0.38, P = 0.77). On average, Fisher Z-transform on both coefficients: patients smoked 40.4 (3.24) standard joints of cannabis on 13.8 (0.76) days last month. Average levels of THC 1+r ln 11-OH-THC and THC-COOH were 4.1 (0.76), 1.5 (0.29) ()− Zf = 1 r . and 29.5 (5.0), respectively. 2 Table 1 shows Pearson’s correlation coefficients for The level of significance was then computed from the each comparison. In the full data set, correlations were Z-value of the difference: on the boundary between moderate or strong, although removal of outliers increased the correlations to strong Zf12− Zf z = . or very strong. There were five outliers regarding plasma 1 1 + THC, in the range of 13–17 ng/ml, except one with nn−3 −3 12 58 ng/ml. Most of these patients self-reported daily use, Further analyses were conducted on classification of but one reported 8 days of use and one 22 days of use. qualitative cannabis use (absent or present) of TLFB using In addition, four outliers were identified regarding self- THC as the gold standard. For this purpose, plasma-THC reported monthly joints. These patients reported a range levels of 0.2 or above were considered indicative of can- of 163–196 joints in the last month. The weakest corre- nabis use. In this manner, we identified the proportions of lations were seen at baseline when not excluding outliers.

Table 1 Correlation coefﬁcients for delta-9-tetrahydrocannabinol (THC) in plasma versus timeline follow-back assisted self-report in 103 patients with psychotic disorder and cannabis use disorder.

THC versus number of days/month THC versus number of joints/month

Including outliers Excluding outliersa Including outliers Excluding outliersb

Full data set r = 0.49, n = 88 r = 0.75, n = 83 r = 0.49, n = 88 r = 0.83, n = 79 Baseline r = 0.43, n = 46 r = 0.73, n = 44 r = 0.38, n = 46 r = 0.72, n = 44 Post-treatment r = 0.65, n = 25 r = 0.80, n = 22 r = 0.78, n = 25 r = 0.82, n = 20 Follow-up r = 0.74, n = 17 r = 0.74, n = 17 r = 0.76, n = 17 r = 0.74, n = 13 Baselinec r = 0.43, n = 46 r = 0.73, n = 44 r = 0.38, n = 46 r = 0.77, n = 44 CapOpus r = 0.81, n = 18 r = 0.81, n = 18 r = 0.77, n = 18 r = 0.67, n = 16 Treatment as usual r = 0.71, n = 24 r = 0.81, n = 22 r = 0.80, n = 24 r = 0.84, n = 21 Men r = 0.49, n = 59 r = 0.66, n = 58 r = 0.46, n = 59 r = 0.74, n = 55 Women r = 0.71, n = 29 r = 0.71, n = 28 r = 0.83 n = 29 r = 0.63, n = 26 PANSS total Q1 r = 0.87, n = 25 r = 0.87, n = 25 r = 0.82, n = 25 r = 0.82, n = 25 PANSS total Q2-3 r = 0.57, n = 42 r = 0.79, n = 39 r = 0.61, n = 42 r = 0.82, n = 38 PANSS total Q4 r = 0.51, n = 21 r = 0.71, n = 20 r = 0.40, n = 21 r = 0.91, n = 17 PANSS N Q1 r = 0.70, n = 29 r = 0.77, n = 28 r = 0.82, n = 29 r = 0.75, n = 28 PANSS N Q2-3 r = 0.72, n = 38 r = 0.78, n = 37 r = 0.71, n = 38 r = 0.84, n = 35 PANSS N Q4 r = 0.51, n = 21 r = 0.61, n = 20 r = 0.39, n = 21 r = 0.68, n = 17 HVLT Q1 r = 0.68, n = 21 r = 0.68, n = 21 r = 0.72, n = 21 r = 0.59, n = 19 HVLT Q2-3 r = 0.62, n = 32 r = 0.65, n = 30 r = 0.70, n = 32 r = 0.66, n = 30 HVLT Q4 r = 0.83, n = 26 r = 0.83, n = 26 r = 0.87, n = 26 r = 0.85, n = 24

THC: delta-9-tetrahydrocannabinol; PANSS: the Positive and Negative Syndrome Scale for Schizophrenia; PANSS N: negative dimensions subscale; HVLT: Hopkins Verbal Learning Test; Q1: lowest quartile of scores. Q2–Q3: two middle quartiles of scores. Q4: highest quartile of scores. For PANSSand PANSSN,Q1 are the least symptomatic patients. For HVLT, Q1 are the patients with poorest cognitive skills. aExcluding outliers according to THC. bExcluding outliers both according to THC and monthly joints. cBaseline represents pre-randomization, and CapOpus and treatment as usual include the two post-randomization interviews. Bold type indicates statistically signiﬁcant differences in correlations between subgroups with P-values between 0.01 and 0.05.

Excluding two outliers from the baseline assessment to measure verbal learning and used in this context as a increased r from about 0.42 to about 0.75. The drastic proxy for memory. Excluding outliers and using standard increase in r was caused by a single person reporting joints for comparison, the correlation coefficients for the three standard joints everyday on average with a THC lowest (worst) and highest (best) quartile on the HVLT concentration of 58 ng/ml. Over time, all other correla- may be different at P = 0.08. Correlation coefficients tion coefficients were strong, sometimes bordering on were not extraordinarily low or high for any quartiles of very strong. MANSA, WHODAS or cognitive scores (data not shown), Correlation coefficients were not significantly different except for the highest (worst) scoring quartile on Trail- between the randomization groups. The correlation coef- Making Test B, where r = 0.38 (n = 14) for joints versus ficient between blood-THC and standard joints including THC excluding outliers. outliers was significantly higher for women than for Table 2 shows proportions of true and false positives men (P = 0.01), but the genders had similar correlation and negatives as well as sensitivity,specificity and positive coefficients in the other three comparisons (P > 0.15). and negative predictive values of TLFB. While the TLFB A tendency was seen of decreasing correlations with was administered to collect information for the past 30 increasing levels of symptoms, as measured both by days, Table 2 uses cut-off points from 1 to 30 days; for overall PANSS score and by the negative symptoms instance, the 7-day cut-off point row in Table 2 lists PANSS subscale. For overall PANSS, the differences values when only using the most recent 7 days on the between the highest (worst) and the lowest (best) quartile TLFB calendar for validation against plasma THC. of scores were significant only when including outliers Figure 1 shows information on sensitivity and specificity (P = 0.02 for number of days and P = 0.04 for number of plotted in a ROC curve. Area under the ROC curve was standard joints versus plasma-THC). On the negative 0.96 (95% CI 0.89–0.99). Ignoring the unsurprising dimensions subscale, only standard joints versus blood- 100% specificity for the shortest retrospective periods, THC and including outliers was significantly different sensitivity increased steadily to 94.3% (95% CI 86.0– between the two extreme quartiles (P = 0.02, all other 98.4%) without any drop in specificity [stable at 94.4% P-values above 0.32). (95% CI 72.2–99.9%) at the retrospective period of 19 There were no significant differences in correlations days]. Thus, 19 days of retrospective recall provided the based on quartiles of scores on the HVLT, a test designed best combination of sensitivity, specificity and positive

and negative predictive values. Using longer periods of TLFB retrospectively yielded only a slight increase in sensitivity,but a noticeable decrease in specificity.Validat- ctive value ing self-report against 11-OH-THC or THC-COOH gave slightly higher sensitivities but markedly lower specificity (data not shown). Median CIF value was 432, but vali- dating self-report against CIF gave both lower sensitivity and specificity (data not shown), indicating that the CIF is not applicable to quantification but rather an indicator of recent use. Sensitivity and negative predictive value of TLFB for short recall periods are low, reflecting the fact that in chronic cannabis users, THC is detectable in

’. plasma even after a lengthy period of abstinence [13,14]. Time since last cannabis use as estimated by Huestis et al.’s models gave a median (range) of 2.3 (0.3–26.5) hours and 1.0 (0.7–2.9) hours, respectively. Due to the nature of the models, this included only those patients with detectable levels of THC and THC-COOH in plasma.

sing 22, 23 or 24 days as cut-off point. False negative indicates THC in plasma For self-report, median time since last use in patients reporting at least 1 day of use last month was 1 day.

e between self-report and THC in plasma. However, 36% of these patients had a range of 2–25 days since last use, which is not reﬂected in the time ranges detected by Huestis et al.’s models.

DISCUSSION

Correlation analyses generally showed strong or very strong agreement between plasma-THC levels and TLFB- assisted self-reports of both days and standard joints used in the past month. For simple detection (presence or absence) of cannabis use, sensitivity and specificity of TLFB-assisted self-report were very high when using 19-day TLFB recall, with a drop in specificity for longer retrospective recall periods. This drop in specificity is caused by an increase in false positives, i.e. patients reporting cannabis use but without detectable THC in their blood. Deliberate over-reporting seems unlikely in view of the higher specificities obtained for shorter recall periods. Thus, for the longer recall periods, at least one of the following is probably true: patients do not remember abstinence and report cannabis use; or patients actually used cannabis on these days, but the LC/MS/MS analysis could no longer detect THC ingested so long ago. Our results indicate that TLFB estimates cannabis use validly in treatment-seeking patients with comorbid psychosis and cannabis use disorder, at least for a 19-day period. Beyond that, specificity drops dramatically, but

64 (72.7%)65 (73.9%) 6 (6.8%) 5 (5.7%)67 (76.1%) 17 (19.3%)67 (76.1%) 17 (19.3%) 3 (3.4%) 1 (1.1%) 3 (3.4%) 1 (1.1%) 14 (15.9%) 91.4%this (82.3%–96.8%) 13 (14.8%) 92.9% (84.1%–97.6%) 94.4% 4may (72.2%–99.9%) (4.5%) 94.4% 5 (72.2%–99.9%) (5.7%) 98.5% (91.7%–100%) be 98.5% (91.8%–100%) 95.7% (88.0%–99.1%) 73.9%an (51.6%–89.8%) 95.7% (88.0%–99.1%) 77.3% (54.6%–92.2%) 77.8% indicator (52.4%–93.6%) 72.2% (46.5%–90.3%) 94.4% (86.2%–98.4%) 93.1% (84.5%–97.7%) 82.4% (56.6%–96.2%) of 81.3% (54.4%–96.0%) TLFB being superior to LC/MS/MS for such non-recent use. Time since last cannabis use appeared to be underes- timated by the models by Huestis et al., as the maximum

Agreement rates for varying timeline follow-back (TLFB) cut-off points using plasma-based delta-9-tetrahydrocannabinol (THC) as ‘gold standard interval was 26 hours. These formulas have been vali- a a a a dated previously, but validation has generally been on Each cut-off point in range tested independently, yielding the same agreement for all days in this range; e.g. all values in table are the same whether u 1234567 488–16 (54.5%) 52 (59.1%) 54 22 (61.4%) (25.0%) 57 18 (64.8%) (20.5%) 59 16 (67.0%) (18.2%) 18 (20.5%) 60 13 (68.2%) (14.8%) 18 (20.5%) 62 11 (70.5%) (12.5%) 17 (19.3%) 0 (0%) 10 (11.4%) 17 (19.3%) 0 (0%) 8 17 (9.1%) (19.3%) 1 (1.1%) 17 (19.3%) 1 (1.1%) 68.6% 1 (56.4%–79.1%) (1.1%) 17 (19.3%) 74.3% 1 (62.4%–84.0%) (1.1%) 77.1% (65.6%–86.6%) 100% (81.5%–100%) 81.4% (70.3%–89.7%) 100% 94.4% 1 (81.5%–100%) (72.7%–99.9%) (1.1%) 84.3% (73.6%–91.9%) 94.4% 100% (72.7%–99.9%) (92.6%–100%) 85.7% (75.3%–92.9%) 98.2% (90.3%–100%) 94.4% 100% (72.7%–99.9%) (93.2%–100%) 98.3% (90.8%–100%) 45.0% 94.4% (29.3%–61.5%) (72.2%–99.9%) 88.6% (78.7%–94.9%) 51.5% 98.3% (33.5%–69.2%) (91.1%–100%) 50.0% (32.9%–67.1%) 56.7% 98.4% (37.4%–74.5%) (91.2%–100%) 94.4% (72.2%–99.9%) 60.7% (40.6%–78.5%) 63.0% 98.4% (42.4%–80.6%) (91.5%–100%) 68.0% (46.5%–85.1%) Table 2 TLFB cut-off: days True positives False negatives True negatives False positives Sensitivity Speciﬁcitya Positive predictive value Negative predi but negative self-report. False positive indicates positive self-report without THC in plasma. True positive or negative indicates correspondenc 17–18 19202122–24 25–30 66 (75.0%) 66 (75.0%) 66 (75.0%) 4individuals (4.5%) 4 (4.5%) 4 (4.5%) 17 (19.3%) 16 (18.2%) known 15 (17.0%) 1 (1.1%) 2 (2.2%)to 3 (3.4%) have 94.3% (86.0%–98.4%) 94.3% (86.0%–98.4%)smoked 94.4% (72.2%–99.9%) 94.3% (86.0%–98.4%) 88.9% (65.3%–98.6%) 98.5% (92.0%–100%) 83.3% (58.6%–96.4%) 97.1% (89.8%–99.6%) cannabis 81.0% 95.7% (58.1%–94.6%) (87.8%–99.1%) 80.0% (56.3%–94.3%) 78.9% (54.4%–93.9%) in the past 8

Figure 1 Receiver operating characteristic (ROC) curve

hours at highest. Possibly, our results indicate that the more patients who gave inaccurate self-reports (whether formulae developed by Huestis et al. do not estimate time intentionally or not) of cannabis use denied providing since last use accurately in individuals in whom this is a blood sample, overestimating correlations. Although several days ago. the difference in self-report between providers and non- To our knowledge, no other studies have compared providers of blood was not significant, non-providers TLFB-assisted self-report of cannabis use with blood reported 2.4 fewer days of cannabis use. However, in mul- samples. Against urine, our systematic review (currently tiple imputation analyses of the set of 239 self-reports, submitted for publication) identified studies estimating associations between self-report and THC, 11-OH-THC false-negative rates from 1 to 7% and false-positive rates and THC-COOH were not substantially different from the from 5 to 20% [11,27,28]. We found similar false- non-imputed associations (data not shown). negative rates but lower false-positive rates (1.1% at 19 Correlations increased when excluding outliers, but days retrospectively). In our study, sensitivity and nega- the reasons why these individuals have extreme values tive predictive value were somewhat low. At face value, are unknown. Possibly, the THC-based outliers were this may be an argument against the use of TLFB. caused by patients using cannabis very close to the time However, it is more likely to reflect the longevity of of blood collection, or with above-average THC contents. presence of THC in plasma of chronic cannabis-users, It is unlikely that these patients under-reported cannabis making it more likely that the problem is that plasma- use, as only one of them reported use on fewer than 22 THC does not distinguish between recent and past use in days in the past month. chronic cannabis users [13,14]. Previous studies have Patients were informed that they would be asked to reported sensitivity and specificity at 0.60 and 0.42 [10] give blood to validate self-reported consumption, which and Spearman’s correlation of approximately 0.4 [29], may have influenced the truthfulness of responses to well below what we found in the present study. Other the TLFB. However, there was no significant difference studies found kappas between 0.47 [11] and 0.9 [30], between self-reported consumption between those who intraclass correlation (ICC) of 0.62 [31] and Yule’s Y of provided a blood sample and those who did not. Also, 0.8 [32], being comparable or just below our correlation there were no consequences involved for being either coefficients. In the cases where our study shows better abstinent or using cannabis, and patients were not agreement than previous studies, it is difficult to state informed about the results of blood tests. whether this is due to the use of blood rather than urine, We converted self-reported amounts into a composite due to differences in sample populations or simply to measure of monthly ‘standard joints’. However, several statistical chance. studies have shown great variations in THC concentra- Our study has several limitations. Blood was not tions in seized cannabis products. We used a factor of 1.5 obtained at all interviews, so we cannot exclude that for sinsemilla, as this appeared to be in line with two

© 2011 The Authors, Addiction © 2011 Society for the Study of Addiction Addiction, 107, 1123–1131 TLFB versus THC in plasma 1129 publications using data from several countries [33,34]. If our findings are correct, they have implications that However, one study from the United Kingdom estimated benefit both researchers and practitioners who need to the median THC content in sinsemilla as almost four assess cannabis use. The use of self-reports is less costly times that of cannabis resin [35], and a study from the than the use of biological materials that require expen- United States found that THC concentrations in cannabis sive techniques for storage and analysis. Self-reports are resin were about twice as high as in sinsemilla [36]. more acceptable to many respondents as samples of, e.g. However, by far the prevailing type of cannabis used in blood and urine may be considered invasive. Finally, the our sample was cannabis resin, and repeating the analy- increase in false positives and decrease in specificity ses with, e.g. a factor of 4 for sinsemilla had little effect on for periods longer than 19 days may indicate that TLFB is correlation coefficients (data not shown). superior to blood samples for longer recall periods. A further limitation is the unclear extent to which our In conclusion, TLFB-assisted self-report was corre- findings can be generalized. While our findings should lated highly with THC in this study. Sensitivity and speci- hold for other trials involving patients with cannabis ficity of TLFB were very high when using 19 days as the use disorder and psychosis, validity may be lower in cut-off point on the TLFB. As such, TLFB may be superior other populations. Compared to non-using patients with to analysis of blood when going beyond 19 days of recall. schizophrenia, cannabis-using patients with schizophrenia may function better cognitively [37]. However, for all Clinical Trial Registration quartiles of scores on the HVLT, r was never below 0.59, The clinical trial, of which this is a secondary analysis, indicating strong agreement between self-report and was registered as follows: ClinicalTrials.gov registration biological measures even in those with poorest cognitive number NCT00484302. The regional ethics committee functioning. Moreover, median scores on cognitive tests for the greater Copenhagen area approved the protocol in CapOpus usually fell well below age- and sex-matched under the file number H-D-2007-0028, as did The normative median scores, indicating that CapOpus par- Danish Data Protection Agency under the file number ticipants were not cognitively well functioning. As such, 2007-41-0616. it seems likely that the recall-enhancing capabilities of TLFB in this population can also be generalized to other Acknowledgements populations. However, while never being decidedly low, and only when including outliers, there were significant The authors would like to thank Palle Fruekilde and drops in correlations as symptoms on the total PANSS his staff at Odense University Hospital for providing the scale or negative (but not positive) PANSS subscale laboratory work for this study. We also thank all our increased. If anything, this should also indicate that our colleagues at the Research Unit of Mental Health Centre results are also valid for individuals without psychiatric Copenhagen for constructive criticism of drafts of this comorbidity, as these would score low on these scales. paper. We thank the participants of the CapOpus trial, CapOpus was a randomized controlled trial (RCT) and finally we thank the Lundbeck Foundation, the aimed at reducing cannabis use, so participants were Municipality of Copenhagen, the Egmont Foundation, presumably motivated to alter their cannabis consump- the Health Insurance Foundation, the Ministry of Social tion, leading to a focus on amounts and frequencies Welfare, Aase and Ejnar Danielsen’s Foundation and the that enhanced recall beyond what might be expected, Wørzner Foundation for their financial support for the e.g. in cohort studies. This could have led to higher CapOpus trial. correlations in this study than in other types of studies. Conversely, RCTs also risk biasing participants towards Declarations of interest wanting to please the researchers, potentially leading The authors have no conflicting interests, and the foun- to under-reporting of actual cannabis use. This does dations, etc. providing funds and grants were not involved not appear to have been the case, as post-randomization in any manner in design and conduct of the study; correlation coefficients were at least as high as at base- collection, management, analysis and interpretation line, and did not differ significantly between treatment of the data; or preparation, review or approval of the groups. manuscript. Finally, patients knew they would be asked to deliver a blood sample when they were subjected to the TLFB. This References may have led to more truthful self-reports of cannabis use. Of note, however, is the result that those who did not 1. United Nations Office on Drugs and Crime. World Drug Report 2010. Vienna: United Nations Publications; 2010. give blood, for whatever reasons, did not self-report sig- 2. Green B., Young R., Kavanagh D. Cannabis use and misuse nificantly different amounts of cannabis use than those prevalence among people with psychosis. Br J Psychiatry giving blood. 2005; 187: 306–13.

3. Caspari D. Cannabis and schizophrenia: results of a 20. World Health Organization (WHO). World Health Organiza- follow-up study. Eur Arch Psychiatry Clin Neurosci 1999; tion Disability Assessment Schedule (WHODAS II). Geneva: 249: 45–9. WHO; 2000. 4. Grech A., van Os J., Jones P. B., Lewis S. W., Murray R. M. 21. Brandt J. The Hopkins Verbal Learning Test: development of Cannabis use and outcome of recent onset psychosis. Eur a new memory test with six equivalent forms. Clin Neurop- Psychiatry 2005; 20: 349–53. sychol 1991; 5: 125–42. 5. Hides L., Dawe S., Kavanagh D. J., Young R. M. Psychotic 22. Bowie C. R., Harvey P.D. Administration and interpretation symptom and cannabis relapse in recent-onset psychosis. of the Trail Making Test. Nat Protocols 2006; 1: 2277–81. Prospective study. Br J Psychiatry 2006; 189: 137–43. 23. Grauwiler S. B., Scholer A., Drewe J. Development of a 6. Kamali M., Kelly B. D., Clarke M., Browne S., Gervin M., LC/MS/MS method for the analysis of cannabinoids in Kinsella A. et al. A prospective evaluation of adherence to human EDTA–plasma and urine after small doses of Can- medication in first episode schizophrenia. Eur Psychiatry nabis sativa extracts. J Chromatogr B Analyt Technol Biomed 2006; 21: 29–33. Life Sci 2007; 850: 515–22. 7. Olfson M., Mechanic D., Hansell S., Boyer C. A., Walkup J., 24. Maralikova B., Weinmann W. Simultaneous Weiden P. J. Predicting medication noncompliance after determination of delta9-tetrahydrocannabinol, 11- hospital discharge among patients with schizophrenia. Psy- hydroxy-delta9-tetrahydrocannabinol and 11-nor-9- chiatr Serv 2000; 51: 216–22. carboxy-delta9-tetrahydrocannabinol in human plasma by 8. Sobell L. C., Sobell M. B. Timeline follow-back: a technique high-performance liquid chromatography/tandem mass for assessing self-reported alcohol consumption. In: Litten spectrometry. J Mass Spectrom 2004; 39: 526–31. R. Z., Allen J., editors. Measuring Alcohol Consumption: Psy- 25. Abbara C., Galy R., Benyamina A., Reynaud M., chosocial and Biological Methods. New Jersey: Humana Press; Bonhomme-Faivre L. Development and validation of a 1992, p. 41–72. method for the quantitation of delta9 tetrahydrocannabinol 9. Hjorthøj C. R., Hjorthøj A. R., Nordentoft M. Validity of in human plasma by high performance liquid chromato- Timeline Follow-Back for self-reported use of cannabis and graphy after solid-phase extraction. J Pharm Biomed Anal other illicit substances—systematic review and meta- 2006; 41: 1011–6. analysis. Addict Behav 2012; 37: 225–33. 26. Huestis M. A., Henningfield J. E., Cone E. J. Blood cannab- 10. Fals-Stewart W., O’Farrell T. J., Freitas T. T., McFarlin S. K., inoids. II. Models for the prediction of time of marijuana Rutigliano P. The timeline followback reports of psychoac- exposure from plasma concentrations of delta 9- tive substance use by drug-abusing patients: psychometric tetrahydrocannabinol (THC) and 11-nor-9-carboxy-delta properties. J Consult Clin Psychol 2000; 68: 134–44. 9-tetrahydrocannabinol (THCCOOH). J Anal Toxicol 1992; 11. Winstock A. R., Lea T., Copeland J. Lithium carbonate in 16: 283–90. the management of cannabis withdrawal in humans: an 27. Marijuana Treatment Project Research Group. Brief treat- open-label study. J Psychopharmacol 2009; 23: 84–93. ments for cannabis dependence: findings from a randomized 12. Musshoff F., Madea B. Review of biologic matrices (urine, multisite trial. J Consult Clin Psychol 2004; 72: 455–66. blood, hair) as indicators of recent or ongoing cannabis use. 28. Waldron H. B., Slesnick N., Brody J. L., Turner C. W., Peter- Ther Drug Monit 2006; 28: 155–63. son T. R. Treatment outcomes for adolescent substance 13. Huestis M. A., Henningfield J. E., Cone E. J. Blood cannab- abuse at 4- and 7-month assessments. J Consult Clin Psychol inoids. I. Absorption of THC and formation of 11-OH-THC 2001; 69: 802–13. and THCCOOH during and after smoking marijuana. J Anal 29. Bashford J., Flett R., Copeland J. The Cannabis Use Problems Toxicol 1992; 16: 276–82. Identification Test (CUPIT): development, reliability,concur- 14. Karschner E. L., Schwilke E. W., Lowe R. H., Darwin rent and predictive validity among adolescents and adults. W. D., Pope H. G., Herning R. et al. Do delta9- Addiction 2010; 105: 615–25. tetrahydrocannabinol concentrations indicate recent use in 30. Godley M. D., Godley S. H., Dennis M. L., Funk R., Passetti L. chronic cannabis users? Addiction 2009; 104: 2041–8. L. Preliminary outcomes from the assertive continuing care 15. Reiter A., Hake J., Meissner C., Rohwer J., Friedrich H. J., experiment for adolescents discharged from residential Oehmichen M. Time of drug elimination in chronic drug treatment. J Subst Abuse Treat 2002; 23: 21–32. abusers. Case study of 52 patients in a ‘low-step’ detoxifica- 31. Donohue B., Hill H. H., Azrin N. H., Cross C., Strada M. J. tion ward. Forensic Sci Int 2001; 119: 248–53. Psychometric support for contemporaneous and retros- 16. Huestis M. A., Barnes A., Smith M. L. Estimating the time of pective youth and parent reports of adolescent marijuana last cannabis use from plasma delta9-tetrahydrocannabinol use frequency in an adolescent outpatient treatment popu- and 11-nor-9-carboxy-delta9-tetrahydrocannabinol con- lation. Addict Behav 2007; 32: 1787–97. centrations. Clin Chem 2005; 51: 2289–95. 32. Stasiewicz P.R., Vincent P.C., Bradizza C. M., Connors G. J., 17. Hjorthøj C., Fohlmann A., Larsen A.-M., Madsen M. T., Maisto S. A., Mercer N. D. Factors affecting agreement Vesterager L., Gluud C. et al. Design paper: the CapOpus between severely mentally ill alcohol abusers’ and collater- trial: a randomized, parallel-group, observer-blinded clini- als’ reports of alcohol and other substance abuse. Psychol cal trial of specialized addiction treatment versus treatment Addict Behav 2008; 22: 78–87. as usual for young patients with cannabis abuse and psy- 33. McLaren J., Swift W., Dillon P., Allsop S. Cannabis potency chosis. Trials 2008; 9: 42. and contamination: a review of the literature. Addiction 18. Kay S. R., Fiszbein A., Opler L. A. The positive and negative 2008; 103: 1100–9. syndrome scale (PANSS) for schizophrenia. Schizophr Bull 34. King L., Carpentier C., Griffiths P., editors. Insights no. 6, An 1987; 13: 261–76. Overview of Cannabis Potency in Europe. European Monitor- 19. Priebe S., Huxley P., Knight S., Evans S. Application and ing Centre for Drugs and Drug Addiction. Luxembourg: results of the Manchester Short Assessment of quality of life Office for the Publications of the European Communities; (MANSA). Int J Soc Psychiatry 1999; 45: 7–12. 2004.

35. Potter D. J., Clark P., Brown M. B. Potency of delta-9-THC 37. Yücel M., Bora E., Lubman D. I., Solowij N., Brewer W. J., and other cannabinoids in cannabis in England in 2005: Cotton S. M. et al. The impact of cannabis use on cognitive implications for psychoactivity and pharmacology. J Foren- functioning in patients with schizophrenia: a meta-analysis sic Sci 2008; 53: 90–4. of existing findings and new data in a first-episode sample. 36. Mehmedic Z., Chandra S., Slade D., Denham H., Foster S., Schizophr Bull 2010; [Epub ahead of print]. Patel A. S. et al. Potency trends of delta-9-THC and other cannabinoids in confiscated cannabis preparations from 1993 to 2008. J Forensic Sci 2010; 55: 1209–17.

© 2011 The Authors, Addiction © 2011 Society for the Study of Addiction Addiction, 107, 1123–1131 This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material.