Using molecular similarity to highlight the challenges of routine immunoassay-based drug of abuse/toxicology screening in emergency medicine

Similarity searching uses the 'find similar molecules by fingerprints' protocol in the library analysis module of Discovery Studio 2.0 (Accelrys, San Diego, CA). The MDL public keys are a fingerprint which uses a pre-defined set of definitions related to structural features [27]. A fingerprint is created based on pattern matching of the structure to this set of 166 keys. These MDL keys are used separately with the Tanimoto similarity coefficient and an input query molecule [21] and will be referred to as 'Tanimoto similarity'. It should be noted that this type of similarity algorithm does not recognize differences between stereoisomers (e.g., d- and l-amphetamine or their racemic mixture; citalopram and escitalopram). Sdf files of the structures of the database compounds are available on request from the authors.

Quetiapine fumarate was obtained from Sequoia Research Products (Pangbourne, United Kingdom). Quetiapine S-oxide, 7-hydroxyquetiapine, and 11-piperazin-1yl-dibenzo [b, f] [1, 4] thiazepine dihydrochloride (DBTP) were purchased from Molcan (Toronto, Ontario, Canada). These three quetiapine metabolites were tested for cross-reactivity with two different TCA screening immunoassays: (1) Emit^® tox™ serum (tricyclic antidepressants) run on Siemens (Dade-Behring) Viva-E analyzers and (2) Biosite Triage^® Tox screen. Both assays were performed following manufacturers' instructions on analyzers used for clinical testing. Urine samples were analyzed by GC/MS to identify a wide range of clinically important drugs and drug metabolites by methods previously described [28]. Patient samples from five University of Pittsburgh Medical Center hospitals (Children's, Montefiore, Presbyterian, Shadyside, and Western Psychiatric) showing positive immunoassay screens for PCP or TCAs were followed by GC/MS testing. The studies involving human samples in this report qualified as exempt, and the need for informed written consent was waived, as determined by the University of Pittsburgh Medical Center Institutional Review Board.

Cross-reactivity data for DOA/Tox immunoassays were retrieved and compiled from package inserts or the assay manufacturers' websites. Data for levofloxacin in Table 2 was obtained from a published article [13]. Complete data for each assay, with compounds sorted and color-coded by classification, is in Additional file 1 (tabs A-R, T). The most prescribed medications in the United States in 2007 [29] is also included in Additional file 1 (tab S). Information on the most prescribed medications in the United States from 1970–2006 is provided in Additional file 2.

We have used the MDL public keys, and a molecular similarity measure (Tanimoto coefficient), to compute the structural similarity between drugs and drug metabolites to the target compounds used in commercially marketed DOA/Tox immunoassay screens. We have illustrated this using PCP as an example. Figure 1 shows the similarity of PCP to 4-phenyl-4-piperidino-cyclohexanol (a PCP metabolite), dextromethorphan (cause of false positives on some PCP screening assays), meperidine (another potential cause of false positives), ketamine, and ibuprofen. PCP has the highest Tanimoto similarity (in descending order) to 4-phenyl-4-piperidino-cyclohexanol (0.784), dextromethorphan (0.565), and meperidine (0.538). All three of these compounds are known to cross-react with some marketed PCP assays, but with varying sensitivities as reported in the package inserts (Additional file 1, tab P).

For example, only 30 ng/mL of 4-phenyl-4-piperidino-cyclohexanol produces cross-reactivity equal to 25 ng/mL PCP in the Abbott Architect PCP assay. Dextromethorphan is reported to cross-react with 4 of 8 commonly marketed PCP assays, with 12,000 ng/mL providing cross-reactivity equal to 25 ng/mL PCP in the Syva EMIT assay but with 500,000 ng/mL needed to do the same for the Biosite Triage assay (Additional file 1, tab P). Meperidine only cross-reacts with 2 of 8 marketed assays (Abbott Architect and Syva EMIT). As shown in Figure 1, ketamine, despite similarity to PCP in terms of mechanism of action and clinical effects [30], is not that closely related to PCP structurally as measured by 2D similarity (0.333) and is also not known to cross-react with marketed PCP assays (Additional file 1, tab P). PCP has essentially no 2D similarity to ibuprofen (0.100), a widely used drug that does not cross-react at all with PCP assays (Figure 1).

We will now apply similarity calculation to our analysis and discussion of individual DOA/Tox screening assays.

Currently marketed amphetamine screening immunoassays in the United States use d-amphetamine, d-methamphetamine, or both drugs as the antigenic targets (Additional file 1, tab T). d-Amphetamine and d-methamphetamine have high similarity to one another (Tanimoto similarity = 0.765) and only 1 currently marketed amphetamine screening assay (Roche cobas c) has markedly different sensitivities for these two amphetamines (Figure 2A; Additional file 1, tab A). There is much more variability in detection by these assays for amphetamine derivatives such as MDMA/Ecstasy (Tanimoto similarity to amphetamine = 0.361) and 3,4-methylenedioxyamphetamine (MDA; Tanimoto similarity to amphetamine = 0.424). The low levels of 2D structural similarity of MDA and MDMA to amphetamine (or methamphetamine) are comparable or lower than those between amphetamine and bupropion (Tanimoto similarity = 0.321), ephedrine (Tanimoto similarity = 0.391), labetalol (Tanimoto similarity = 0.298), mexiletine (Tanimoto similarity = 0.500), phentermine (Tanimoto similarity = 0.778), and pseudoephedrine (Tanimoto similarity = 0.391).

This presents a difficult challenge in developing antibodies broad enough to detect a range of amphetamine derivatives but avoiding widely used drugs with potential for cross-reactivity such as bupropion, labetalol, or pseudoephedrine. Figure 2A shows the cross-reactivities of six marketed amphetamine assays for d-amphetamine, d-methamphetamine, MDA, MDMA, 3,4-methylenedioxyethylamphetamine (MDEA), and phentermine. As can be seen, there is wide variability in the ability of these assays to detect MDA, MDMA, and MDEA (note the ordinate in Figure 2A is on a logarithmic scale). One clinical consequence of this may be that a patient abusing MDMA can have opposing test results if evaluated by two different assay systems (e.g., because of transfer from one hospital to another). More recently, specific MDMA immunoassays that have good cross-reactivity with MDA and MDEA but essentially no cross-reactivity with d-amphetamine or d-methamphetamine have been developed and marketed (Additional file 1, tab T).

An additional challenge in interpreting amphetamine screening assay results is that prescriptions for amphetamine mixed salts (e.g., Adderall^®) are now common, ranking #66 in total volume of prescriptions in the United States in 2007 (Additional file 1, tab S; Table 3). A pharmacokinetic study of individuals taking Adderall^® for at least 5 consecutive days showed peak urine concentrations (5,739 to 19,172 ng/mL) that greatly exceed the 1,000 ng/mL cutoff often used in screening immunoassays, and in general urine amphetamine concentrations that were mostly above 1,000 ng/mL [31].

The plots in Figure 2A include dashed brackets to indicate clinically or toxicologically relevant urine concentrations from published studies. For amphetamine and methamphetamine, the urinary concentrations indicated by dashed brackets in Figure 2A are the ranges found in a pharmacokinetic study involving four consecutive daily doses of methamphetamine [32]. In this study, the urine concentrations achieved generally exceeded the 1,000 ng/mL positive cutoff for both amphetamine and methamphetamine. For MDMA and MDA, the range of urine concentrations indicated by dashed brackets in Figure 2A are from 25 antemortem urine concentrations in fatal cases associated with MDMA overdose [33]. Note that even these very high MDMA and MDA urine concentrations do not exceed the threshold for positivity on some amphetamine screening immunoassays. The dashed brackets in the phentermine plot in Figure 2A are the range of urine concentrations reported in forensic studies of phentermine overdose [34]. These very high phentermine urine concentrations would exceed the positive cutoff for only two marketed amphetamine screening immunoassays. Consequently, currently marketed amphetamine screening immunoassays generally do not cross-react with phentermine or do so only when this drug is taken in extreme overdose.

All currently marketed barbiturate immunoassays use secobarbital as a target compound, with some containing antibodies raised only against secobarbital, while others use antibodies raised against multiple barbiturates (Additional file 1, tab T). The choice of secobarbital as the antigenic target in first-generation barbiturate immunoassays followed from this intermediate-acting barbiturate being one of the most heavily prescribed and abused barbiturates of the 1960s and 1970s [35]. Based on similarity calculations, clinically used barbiturates do not possess as much 'within-class' structural variability as the amphetamines discussed above. Clinically important barbiturates have MDL similarities of 0.7 or greater to one another and low structural similarity to other classes of drugs, probably explaining why barbiturate assays have very few documented out-of-class cross-reactive compounds (Additional file 1, tab B). One known cross-reactive drug, aminoglutethimide (Tanimoto similarity = 0.567 relative to secobarbital), is not widely used in the United States and would be an uncommon cause of a barbiturate screening assay false positive. Prescriptions and abuse of barbiturates have been declining steadily in the United States for the past three decades [3]. For example, in the 1970s, six barbiturates were among the most highly prescribed medications in the United States (Additional file 2, figure S2-A). However, other medications such as benzodiazepines, eszopiclone, and zolpidem have steadily replaced barbiturates as safer hypnotics, anxiolytics, and sedatives (Additional file 2-figures S2-B,C). Currently, only two barbiturates (butalbital and phenobarbital) rank in the top most prescribed medications in the United States (Additional file 1, tab S and Additional file 2, figure S2-C) [29].

First-generation benzodiazepine screening assays of the 1970s used one of three antigenic targets [36, 37] – diazepam, nordiazepam, or oxazepam – with a recent shift towards using multiple benzodiazepines as antigenic targets (Additional file 1, tab T). The original choice of diazepam, nordiazepam, or oxazepam also followed from historical trends in usage of benzodiazepines. Diazepam was the most prescribed medication overall in the United States for over a decade (Additional file 2). Other commonly prescribed benzodiazepines of the 1970s, including chlordiazepoxide and clorazepate, are metabolized to nordiazepam and oxazepam (Additional file 2, figure S2-D). Using diazepam, nordiazepam, or oxazepam as target compounds thus fit the prescribing patterns of the 1970s well, either by targeting the most commonly prescribed benzodiazepine of that time (diazepam) or targeting metabolites common to multiple benzodiazepines.

However, three benzodiazepines (alprazolam, clonazepam, and lorazepam) are currently more commonly prescribed in the United States than diazepam (Additional file 2, figure S2-C; Table 3) [29]. None of these three 'newer' benzodiazepines are metabolized to nordiazepam or oxazepam; in addition, each has lower similarity to diazepam than does nordiazepam (Tanimoto similarities to diazepam: nordiazepam, 0.780; lorazepam, 0.673; clonazepam, 0.656; alprazolam, 0.610). The marketed benzodiazepine screening immunoassays therefore have difficulty in detection of clonazepam and lorazepam usage, as compared to the use of diazepam or other early generation benzodiazepines.

Figure 2B plots the cross-reactivities of marketed benzodiazepine assays towards diazepam, nordiazepam, oxazepam, 7-aminoclonazepam, and lorazepam glucuronide (note that cross-reactivity is not reported for all of these compounds for some of the assays). The upper brackets for diazepam, nordiazepam, and oxazepam in Figure 2B indicate the maximum urine concentrations detected in individual consuming a single diazepam dose of 10 mg or less [38]. As can be seen, even a single diazepam dose can result in urine concentration of diazepam and multiple metabolites that exceed the positive cutoff for benzodiazepine screening immunoassays (Figure 2B). Detection would be predicted to be even easier in patients on chronic therapy, where steady-state urine concentrations of diazepam and multiple metabolites would accumulate to even higher concentrations.

Currently marketed benzodiazepine screening assays have limited sensitivity to detecting use of clonazepam. Although marketed benzodiazepine assays have reasonably good sensitivity to clonazepam (parent drug), sensitivity is much lower to the major urinary metabolite 7-aminoclonazepam (Additional file 1, tab C). Following oral administration, the majority of clonazepam appears in the urine as the metabolite 7-aminoclonazepam [39, 40]. In a study of clonazepam pharmacokinetics following a single 3 mg dose [40], the highest peak urine concentration of 7-aminoclonazepam (~ 183 ng/mL) recorded in all study participants would still be below the reported concentrations of 7-aminoclonazepam necessary to produce a positive screening result in all currently marketed benzodiazepine screening immunoassays (dashed bracket in Figure 2B; Additional file 1, tab C). Even with chronic administration of clonazepam, urine concentrations of 7-aminoclonazepam may still be below the positive cutoff for most benzodiazepine screening immunoassays in clinical situations.

Currently marketed benzodiazepine screening assays also have difficulty in detecting the use of lorazepam. Studies of lorazepam pharmacokinetics following oral or parenteral administration show that very little unchanged drug is excreted in the urine, with the majority appearing as the glucuronide metabolite [41, 42]. Lorazepam glucuronide has low structural similarity to diazepam (Tanimoto similarity = 0.561) and is detected much more poorly by the marketed assays than unconjugated lorazepam (Additional file 1, tab C). Some marketed benzodiazepine immunoassays can include a separate step to cleave the glucuronide bonds (e.g., by enzymatic or chemical reaction), resulting in unconjugated drugs. For a drug such as lorazepam, where the glucuronide metabolite is the predominant form in the urine, cleaving the glucuronide bonds would be predicted to enhance the detection rate. Some marketed assays (e.g., Syva EMIT-H^® and Roche Online KIMS^®) incorporate a glucuronide cleavage step in the reaction, while still maintaining rapid analysis times [43, 44].

All cocaine screening immunoassays currently marketed in the United States use antibodies raised against benzoylecognine, one of the two major cocaine metabolites in humans [7], as the antigenic target (Additional file 1, tab T). Thus, the marketed assays can be termed more precisely 'cocaine metabolite screening assays' or 'benzoylecgonine screening assays'. Currently these marketed assays detect cocaine (parent drug) weakly, with cross-reactivities equal to 300 ng/mL benzoylecgonine only occurring at cocaine concentrations ranging from 10,000 ng/mL (Abbott AxSYM) to 80,000 ng/mL (Syva EMIT) (Additional file 1, tab F; Figure 3A). In clinical practice, this means that very recent use of cocaine, even in large amounts, may fail to trigger a positive screen if too little time has elapsed for the metabolism of the parent drug to benzoylecgonine to occur. The marketed assays also vary in detection of other cocaine metabolites such as ecgonine, ecgonine methyl ester (the second major cocaine metabolite in most individuals) [7], and benzylnorecgonine (Additional file 1, tab F; Figure 3A), potentially leading to different results if a patient sample is tested on more than one immunoassay system. The upper brackets for all compounds in Figure 3A except cocaethylene indicate peak urine concentrations of cocaine and metabolites following controlled administration of 40 mg cocaine by the inhalation route [45]. For cocaethylene (an adduct product of cocaine and ethanol), the upper bracket indicates the peak urine concentration in a controlled study of simultaneous cocaine and ethanol administration [46]. Figure 3A illustrates the low cross-reactivity of marketed cocaine metabolite immunoassays to all compounds except benzoylecgonine.

Molecular similarity also explains the generally low false positive rates for cocaine metabolite assays. In terms of similarity, benzoylecgonine has low structural similarity (Tanimoto similarity of 0.615 or lower) to common medications or illicit drugs. Despite being a local anesthetic (in addition to its other effects), cocaine has low structural similarity to other clinically important local anesthetics (e.g., bupivacaine, lidocaine, and procaine), which have Tanimoto similarities to benzoylecgonine of 0.377 or lower, effectively explaining why such compounds or their metabolites do not cause false positives on the currently available cocaine metabolite screening immunoassays (Additional file 1, tab F).

Unlike barbiturate and benzodiazepine screening immunoassays, where some manufacturers have used multiple drugs or drug metabolites as antigenic targets, all currently marketed opiate immunoassays use antibodies raised solely against morphine (Additional file 1, tab T). Based on our similarity calculations, this strategy would be predicted to be effective for the sensitive detection of opiates and metabolites that are structurally very close to morphine, including codeine (Tanimoto similarity to morphine = 0.943), heroin (Tanimoto similarity = 0.857), 6-AM (main metabolite of heroin; Tanimoto similarity = 0.891), and hydromorphone (Tanimoto similarity = 0.873). In fact, the package insert data reveals this generally to be true with few exceptions (e.g., a few assays are less sensitive to hydromorphone than to codeine; Additional file 1, tab N; Figure 3B). Currently marketed opiate immunoassays perform less well in their detection of oxycodone (Tanimoto similarity to morphine = 0.800) with 2 of 8 marketed assays being essentially insensitive to oxycodone and 3 additional assays only producing cross-reactivity equal to 300 ng/mL morphine at oxycodone concentrations of 16,000 ng/mL or greater (Additional file 1, tab N; Figure 3B; Table 1). These assays also respond weakly to oxymorphone (Figure 3B), itself a clinically used drug and also a main metabolite of oxycodone [47]. The lack of sensitivity of marketed opiate assays to oxycodone is a serious limitation as this drug is now prescribed more often in the United States than codeine, morphine, and propoxyphene (Additional file 2, figure S2-E; Table 3) and is also a frequently abused prescription medication [48].

The upper and lower brackets for morphine, codeine, hydromorphone, hydrocodone, and oxycodone in Figure 3B represent estimated steady-state urine concentrations in a 70 kg individual during chronic administration of daily 60 mg oral dose of codeine, 10 mg intramuscular dose of morphine, 5 mg oral dose of hydromorphone, 10 mg oral dose of hydrocodone, or 20 mg oral dose of oxycodone, respectively [49]. As can be seen in Figure 3B, the steady-state urine concentrations for all drugs except oxycodone will generally exceed the cutoff equivalent to 300 ng/mL morphine. For oxycodone, only 3 of 7 marketed assays have sensitivities to oxycodone sufficient to readily detect daily use of 20 mg oral oxycodone. Exact urine concentrations of oxymorphone following either oxymorphone or oxycodone administration have not been reported in the literature but are likely to be well below the assay sensitivities due to the extensive metabolism of oxymorphone prior to renal excretion [50, 51]. The upper bracket for 6-AM is the highest peak 6-AM urine concentration observed in a study of controlled heroin administration [52].

Marketed opiate assays do not cross-react with the mixed opiate agonist-antagonist buprenorphine (Tanimoto similarity to morphine = 0.783) (Additional file 1, tab N). Commonly used non-opiate opioid drugs (e.g., fentanyl, meperidine, methadone, propoxyphene) generally have low structural similarity to morphine (Tanimoto similarity range = 0.407 – 0.522) and either do not cross-react, or do so only at extremely high concentrations, with opiate screening immunoassays (Additional file 1, tab N).

As a drug of abuse in the United States, PCP has waxed and waned in popularity over time, with substantial regional differences in usage of this drug [53]. There are five well-documented cross-reactive compounds with PCP immunoassays: dextromethorphan, venlafaxine [14‐16], meperidine, thioridazine, and mesoridazine (Additional file 1, tab P), although high urine concentrations of these drugs are generally required to elicit a positive PCP screening result (Figure 4A; brackets indicate urine concentrations of PCP in patients abusing PCP [54]). While prescriptions for the latter three drugs have declined in the United States over the last decade, venlafaxine is widely prescribed in the United States (55^th most prescribed drug in 2008) [29], and dextromethorphan continues to be widely used as both a prescription and over-the-counter medication in anti-tussive remedies (Table 3) [29]. Supra-therapeutic doses of dextromethorphan-containing medications are sometimes abused for psychoactive effects, most frequently by adolescents and young adults [55].

In our own medical system, we had anecdotally observed that PCP abuse by patients presenting to our ED was rare, whereas intentional dextromethorphan and meperidine usage was more common, suggesting that there may be many false positive PCP screening tests. We therefore examined the causes of PCP positive screens in our medical system (Additional file 1, tab V). Over the course of 24 months (January 2007 through January 2009), we had 10 patient samples with positive PCP screens, nine on the Syva EMIT assay and only one on the Biosite Triage assay system (Additional file 1, tab V). Urine samples from these 10 patients were also analyzed by GC/MS, using a protocol to definitively identify a wide range of clinically important legal and illicit drugs [28]. Of these, only one patient showed the definitive presence of PCP by GC/MS (30 year old female); this positive screen occurred on the Syva EMIT system. The one positive PCP screen on the Biosite Triage assay in our sample occurred in a 48 year old female, with GC/MS analysis showing apparently very high urine concentrations of diphenhydramine, a medication reported to cause positives on the Biosite Triage PCP assay at high urine concentrations (Additional file 1, tab P). The eight patient samples (average age 23.6 years, range 2–44 years old; 5 males, 3 females) that had a positive PCP screen on the Syva EMIT assay not accounted for by the presence of PCP all showed apparently high urine concentrations of dextromethorphan by GC/MS that would be consistent with intentional or inadvertent overdose of dextromethorphan. One of these samples also showed the presence of apparently high urine concentrations of meperidine by GC/MS. Relative to PCP, dextromethorphan (Tanimoto similarity = 0.565) and meperidine (Tanimoto similarity = 0.538) have similarity coefficients higher than some of the PCP metabolites compared to their parent drug (Additional file 1, tab P).

The common misuse of dextromethorphan suggests that, in medical settings where PCP use is uncommon, false positives on some marketed PCP assays (e.g., Syva EMIT) due to dextromethorphan can occur more frequently than true positives. In these cases, PCP screening assays may become more effective as 'dextromethorphan overdose screens' than as PCP screens. However, it should be pointed out that at least five of the currently marketed PCP immunoassays are reported to be insensitive to dextromethorphan (Additional file 1, tab P; Figure 4A). The Biosite Triage system used in some hospitals in our medical system is an example of this. This may explain why all but one of the PCP screen positives we documented occurred on the Syva EMIT system, for which only 12,000 ng/mL of dextromethorphan, a concentration easily obtainable in patients taking dextromethorphan-containing medications in overdose [55, 56] will trigger a screen positive equal to 25 ng/mL PCP (Additional file 1, tab P). It should be noted that our laboratory has, over the last two years, analyzed many urine specimens where dextromethorphan is detectable by GC/MS but the PCP screen is negative, supporting that only high urine concentrations of dextromethorphan can cause a false positive PCP screen.

Currently marketed TCA screening immunoassays use either desipramine or imipramine, or multiple TCAs, as target compounds (Additional file 1, tab T). Our similarity calculations indicate that screening for TCAs is a difficult challenge for an immunoassay. In particular, several phenothiazines and other non-TCA drugs have a relatively high structural similarity to desipramine (or other TCAs), which may explain why some non-TCA compounds cross-react well with TCA screening assays (Additional file 1, tab R; Figure 4B) [57‐61]. Examples of the Tanimoto similarities of TCAs and other tricyclic compounds relative to desipramine are: amitriptyline (0.600), carbamazepine (0.460), chlorpromazine (0.630), cyclobenzaprine (0.565), doxepin (0.529), nortriptyline (0.628), prochlorperazine (0.630), and quetiapine (0.485) (Table 3).

An additional challenge for TCA screening assays is that prescriptions for TCAs have declined markedly in the United States in the last fifteen years as other medications such as the selective serotonin reuptake inhibitors (SSRIs) have assumed steadily increasing shares of the market for treatment of depression, obsessive-compulsive disorder, and other psychiatric conditions (Additional file 2, figures S2-G,H) [62]. This is illustrated in Figure 5A which shows the rank of TCAs, cyclobenzaprine, and quetiapine among the top prescribed medications in the United States in the time period from 1998 to 2007. In 2007 [29], only amitriptyline (#70) ranked in the top 100 most prescribed medications, likely due in part to the extensive use of amitriptyline for treating chronic pain [63], whereas nine non-TCA antidepressants rank in the top 100 most prescribed medications (sertraline, #23; escitalopram, #26; fluoxetine, #36; bupropion, #44; paroxetine, #49; venlafaxine, #55; citalopram, #56; trazodone, #59, and duloxetine, #79) (Additional file 1, tab S; Additional file 2, figures S2-G,H). As shown in Figure 5A, cyclobenzaprine was prescribed more often than amitriptyline in 2007, and quetiapine has also been approaching amitriptyline in total number of prescriptions. Meanwhile, prescriptions for the TCAs desipramine, doxepin, imipramine, and nortriptyline have steadily declined in the last decade with desipramine no longer ranking in the top most prescribed medications (Figure 5A, Additional file 1, tab S). Prescriptions for phenothiazines have also declined dramatically over the last decade in the United States, with 'atypical' antipsychotics increasing significantly in clinical use (Additional file 2, figures S2-H,I). This decline in use of phenothiazines means they will likely account for increasingly fewer false positives on TCA screening assays.

In our own medical system, we use two different assay methods for TCA screening (Biosite Triage^® and Syva Emit^® serum tox™). Over the course of 24 months (January 2007 through January 2009), we performed GC/MS analysis for the broad detection of drugs and drug metabolites on all samples that returned a positive screening result for TCA assays (109 on Triage and 15 on Syva) to determine the most likely cause of the positive result (Additional file 1, tab U). As shown in Figure 5B, 70% of the positive results were accounted for by the presence of amitriptyline (28.2%) or cyclobenzaprine (41.9%), a centrally acting muscle relaxant that differs from amitriptyline by the presence of one double bond [64]. Other TCAs accounted for 10.5% (1 imipramine and 10 doxepin cases) of the positive screens while two phenothiazines (chlorpromazine and prochlorperazine) accounted for only 3.2% of the positive screens. Thus, TCAs only accounted for 38.7% of the most likely causes for the positive screening results in our study. The remaining positive screens (16.0% of the total) included patients whose urine showed high concentrations of venlafaxine (n = 2), carbamazepine (n = 2), topiramate (n = 1), or quetiapine (n = 1). The frequency of drugs most likely causing positive TCA screens in our sample fits well with the overall prescription trends in the United States (Figure 5A; Additional file 1, tab S) and the known cross-reactivities of the Biosite Triage and Syva EMIT assay systems. For example, the package insert for the Triage assay states that a cyclobenzaprine urine concentration of only 2000 ng/mL will produce cross-reactivity equal to 1000 ng/mL TCAs (Additional file 1, tab R).

Given the current widespread use of cyclobenzaprine in the United States (Figure 5A) [29, 64], it is not surprising to encounter false positive TCA screening assay results due to the presence of this drug. The increasing use of cyclobenzaprine and quetiapine will likely result in more and more TCA positive screens resulting from these compounds [17‐19, 59, 65], especially combined with increasing reports of intentional misuse and overdoses with quetiapine [66‐68].

Previous studies have shown cross-reactivity of quetiapine (parent drug) with marketed TCA assays but did not look at quetiapine metabolites [17‐19, 59, 65]. We therefore tested the Syva EMIT and Biosite Triage TCA screening assays with drug-free serum (Syva) or urine (Biosite) spiked with pure reference standards of each of three quetiapine metabolites (DBTP, quetiapine S-oxide, 7-hydroxyquetiapine) or quetiapine itself. None of these four compounds, even at very high concentrations, cross-reacted with the Biosite assay. We did find, however, that cross-reactivity equal to 1000 ng/mL desipramine in the Syva assay was produced by 100,000 ng/mL quetiapine, 50,000 ng/mL DBTP, or 200,000 ng/mL quetiapine S-oxide. Although to our knowledge there is no published data on serum or urine concentrations of quetapine metabolites following quetiapine overdose, our data suggest that quetiapine metabolites may contribute to cross-reactivity with some TCA screening immunoassays.

As we have seen, several broad-specificity DOA/Tox immunoassays may fail to detect all clinically important members of a class of drugs (Table 1). To address this issue, manufacturers have developed and marketed assays for buprenorphine, heroin metabolite/6-AM, MDMA, and oxycodone (Additional file 1, tabs D, H, J, and O). The currently marketed assays for buprenorphine and oxycodone are reported to be highly specific for only these drugs and their main metabolites (i.e., buprenophine glucuronide and oxymorphone, respectively) [69]. One possible limitation of the oxycodone assay is that it will not distinguish between the use of either oxycodone and oxymorphone. The only currently marketed heroin metabolite immunoassay cross-reacts well with heroin and weakly with structurally related opiates (hydromorphine, morphine) (Additional file 1, tab H). The only currently marketed MDMA assay cross-reacts well with a number of designer amphetamines that are related structurally, e.g, MDA (Tanimoto similarity to MDMA = 0.889) and MDEA (0.850) but essentially does not cross-react with the less similar d-amphetamine (0.361) or d-methamphetamine (0.457) (Additional file 1, tab J).