Thiopurines’ Metabolites and Drug Toxicity: A Meta-Analysis

Many questions remain unanswered regarding therapeutic drug monitoring (TDM) utility with thiopurines. This study aims to establish a relationship between thiopurines’ metabolites and drug toxicity. We performed a systematic review with inclusion of studies evaluating the relationship between thiopurines’ metabolites and drug toxicity. Meta-analysis of mean difference (MD), correlations and odds ratio (OR) was performed. We identified 21,240 records, 72 of which were eligible for meta-analysis. Levels of 6-thioguanine nucleotides (6-TGN) were higher in patients with leukopenia (MD 127.06 pmol/8 × 108 RBC) and gastrointestinal intolerance (MD 201.46 pmol/8 × 108 RBC), and lower in patients with hepatotoxicity (MD −40.6 pmol × 108 RBC). We established a significant correlation between 6-TGN and leukocytes (r = −0.21), neutrophils (r = −0.24) and alanine aminotransferase levels (r = −0.24). OR for leukopenia in patients with elevated 6-TGN was 4.63 (95% CI 2.24; 9.57). An optimal cut-off of 135 pmol/8 × 108 RBC for leukopenia was calculated (sensitivity 75.4%; specificity 46.4%). 6-methylmercaptopurine ribonucleotides (6-MMPR) were significantly associated with hepatotoxicity (MD 3241.2 pmol/8 × 108 RBC; OR 4.28; 95% CI 3.20; 5.71). Levels of 6-MMPR measured in the first 8 weeks of treatment were associated with leukopenia. We conclude that TDM could be used to prevent thiopurines’ toxicity. As optimal metabolites level may vary according to indication, physicians may adapt posology to decrease toxicity without compromising efficacy.

As prodrugs, thiopurines have a complex metabolism which leads to the formation of 6-thioguanine nucleotides . Regarding conventional thiopurines, other pathways compete with the production of the active metabolite 6-TGN, leading to the formation of 6-methylmercaptopurine (6-MMP) and 6-MMP ribonucleotides . These metabolites can be determined by different methods, such as the Lennard [2] and Dervieux-Boulieu assays [3], that perform the measurement in red blood cells (RBC), with concentrations expressed as pmol/8 × 10 8 RBC.
Thiopurines present toxicity at distinct levels: myelosuppression, hepatotoxicity, pancreatitis and gastrointestinal intolerance, among others. Toxicity is an important cause of treatment cessation; in IBD, about 15% of patients discontinue thiopurines due to adverse events [4,5]. The toxicity of thiopurines can be divided into dose-dependent and idiosyncratic. Due to the distinct metabolisms, the safety profiles of thiopurines may differ. The most worrisome adverse event of 6-thioguanine is liver nodular regenerative hyperplasia (NRH), which still detracts some physicians from its use [6].
The balance between efficacy and toxicity can be achieved with tailored dosing and monitoring, using a weight-based regimen. However, the dose of thiopurines does not correlate with the levels of metabolites [7]. The level of metabolites, specifically 6-TGN, has been associated with improved clinical outcomes in ALL, renal transplantation, and IBD [8][9][10][11]. An optimal therapeutic range of~230 to 400 pmol/8 × 10 8 RBC is often cited for patients with IBD and other disorders [12,13]. Values of 6-TGN of 450 pmol/8 × 10 8 RBC and of 6-MMPR of 5700 pmol/8 × 10 8 RBC were reported as thresholds for myelotoxicity and hepatotoxicity, respectively [14,15]. However, the benefit of therapeutic drug monitoring (TDM) for thiopurines is still uncertain [16]. Based on the risk of myelosuppression, Food and Drug Administration (FDA) and the Clinical Pharmacogenetics Implementation Consortium recommend genotyping or phenotyping for thiopurine S-methyltransferase (TPMT) deficiency prior to starting thiopurines [17][18][19]. American Gastroenterology Association and proceedings of the first Thiopurine Task Force meeting [6,16] state that the benefit for routine TPMT testing is still uncertain for most patients, and some real-life studies support this statement [20].
Even though this topic is of great interest and can influence the clinical practice in several disciplines, as far as we know, scientific literature lacks a comprehensive study on the metabolites of thiopurines and their correlation with toxicity. The aim of this study was to conduct a systematic review and meta-analysis of the studies associating the levels of thiopurines' metabolites with the occurrence of toxicity.

Search Strategy
In this study, we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [21] and the Cochrane Collaboration [22] Guidelines. On November 2018, we performed a literature search on three electronic databases: Pubmed [23], Web of Science [24] and Scopus [25], without time restrictions. The following search words or Medical Subject Heading terms were used: (("Azathioprine" OR "Mercaptopurine" OR "6-thioguanine nucleotide" OR "6-methyl mercaptopurine") AND ("Drug-Related Side Effects and Adverse Reactions" OR "adverse effects" OR "leukopenia" OR "toxicity" OR "infection" OR "hepatotoxicity")). We performed a manual search of the list of references of all relevant studies to ensure that all pertinent articles were considered.

Eligibility and Inclusion/Exclusion Criteria
For our meta-analysis, we considered eligible for inclusion all the studies enrolling adult and/or pediatric patients in which the relationship between thiopurines' metabolites and toxicity was evaluated. The underlying disease was not a factor for eligibility. The following types of toxicity were considered: myelotoxicity (subdivided in leukopenia, lymphopenia, neutropenia, anemia, and thrombocytopenia), hepatotoxicity, infections, pancreatitis, and/or gastrointestinal intolerance. Oncological adverse events, effects in pregnancy or offspring outcomes, and postsurgical complications were not considered. Randomized controlled trials, cohort studies and case series with more than five patients were considered. No restriction in terms of publication dates was applied. Concomitant medications were allowed but listed.
The exclusion criteria were: (i) systematic reviews or guidelines; (ii) animal studies; (iii) individual case reports; and (iv) case series with up to five patients.

Study Selection and Data Collection
First, we screened the list of titles and the abstracts of the studies identified in the initial search. The list of references was screened by two reviewers and all studies not fulfilling the inclusion criteria were excluded from further analysis. In cases of disagreement, a third independent reviewer was consulted. For the remaining studies, we analyzed full text articles to determine eligibility. Rayyan application (Qatar Computing Research Institute, Doha, Qatar) was used during this process [26].
From the selected studies, we collected the following information: journal and authors' name, publication year, type of study, cohort's geographic origin, cohort's age group (pediatric vs. adult), number of enrolled patients, number of patients with measurement of metabolites, underlying disease for which the thiopurine was used, type of thiopurine and treatment duration, concomitant medication, 6-TGN and 6-MMPR serum levels and cut-offs, methodology used for metabolites quantification, time of metabolites quantification, type of toxicity, definition of toxicity, relationship between thiopurines' metabolites and drug toxicity.

Quality Assessment
We used funnel plot analysis to detect potential publication bias and/or systematic heterogeneity [22]. The quality of the studies was assessed following the quality assessment tool (QATSDD, Sirriyeh et al, Leeds, UK) [27]. For each study, the scores were added and divided by the maximum possible score (42) to obtain the overall quality score.

Statistical Analysis
In this meta-analysis, the main variable was the occurrence of toxicity. Three types of data were available:

1.
Mean values of metabolites concentration in patients with or without toxicity Since many studies provided medians, in order to avoid losing a significant amount of data by excluding them from the analysis, mean and standard deviation (SD) were calculated from the reported data, as described by Wan et al. (2014) [28]. The studies in which this calculation was performed are identified with an asterisk in the forest-plots. Studies providing full data-allowing the direct calculation of mean and SD-but with a skewed distribution of the variables are identified with two asterisks in the forest-plots. When data were only presented in the form of an image, we extracted the values with WebPlotDigitizer v4.2 (Ankit Rohatgi, Pacifica, CA, USA). Random-effects models were used to test whether mean 6-TGN, 6-MMP or ratio 6-MMPR/TGN values differed among patients with and without toxicity. Review manager v5.3 (Cochrane, London, UK) was used for the evaluation of mean metabolites/ratio differences.

2.
Odds Ratio (OR) When available, the proportion of patients with and without a specific toxicity with levels of 6-TGN/6-MMPR/ratio above and below the defined threshold values was extracted or calculated from each article. However, some studies only provided the final OR value. In these cases, when available, the OR from multivariate analysis was included. Stata 16 (StataCorp, Lakeway Drive, TX, USA) was used for estimating the pooled OR and its 95% confidence interval (95% CI), using a random-effects model. Since different studies used different methods to measure 6-TGN, a previously recommended strategy was used to compare data [29]. The Lennard assay was used as "standard" [2,30,31]; the values of 6-TGN in studies using the Dervieux-Boulieu assay [3] were divided by 2.6 [29], and those obtained with the Erdmann method [32] and with the commercial assay offered by Prometheus Laboratories Inc. (San Diego, CA, USA) were multiplied by 1.6 [33,34]. A high degree of correlation between methodologies has been demonstrated [31,33]. Concentrations of 6-MMPR were not converted, as it has been reported that these are similar in different assays [35][36][37].

Correlations
Since some toxicities can be presented as continuous variables, correlation analysis was also performed. The correlation coefficient was extracted or calculated from each article. The Schmidt-Hunter method was used to calculate the overall correlation coefficient (r) from a set of correlations [38]. Statsdirect v3.2.8 (StatsDirect Ltd., Birkenhead, Merseyside, UK) was used for this analysis.
All the presented p-values are two-sided and have a 5% significance level. Statistical heterogeneity was assessed using the I 2 statistic and by performing subgroup analyses on the following variables: (i) methodology used for the determination of metabolites; (ii) age group; (iii) underlying disease; (iv) geographical origin of the studies; (v) type of thiopurine; (vi) duration of treatment (< or ≥8 weeks); (vii) concomitant medication; and (viii) provided vs. calculated means.
An optimal cut-off for leukopenia was calculated according to the method described by Steinhauser [39].

Bibliographic Search and Study Selection
The adopted study selection strategy is detailed in Figure 1. From the initial 21,240 reports, after duplicates removal, initial screening and full-text retrieval and analysis, 127 articles met the criteria for qualitative synthesis. Of these, 72 had appropriate data to be included in the meta-analysis.

Description of the Studies
The details of the 72 studies included in the meta-analysis are presented in Table 1 and Table S1. The 72 included studies were published between 1983 and 2018. Most studies (n = 60, 83%) were from Western countries. The distribution between pediatric (n = 29, 40%) and adult cohorts (n = 30, 42%) was balanced. Most studies included patients with IBD only (n = 42, 58%), followed by ALL (n = 16, 22%). In the majority of reports, patients were treated with conventional thiopurines (n = 63, 89%) and were being treated for 8 weeks or longer at the time of the first metabolite measurement (n = 35, 49%). Most reports included patients concomitantly treated with other immunosuppressants or chemotherapy drugs. In 31 studies, data of patients without those concomitant drugs were provided and was used in our meta-analysis; still, steroids, mesalamine and/or antibiotics were allowed in most.
The measurement of metabolites followed the procedures described by Lennard [2,30] or were converted to approximate values in 33 studies, the methodology described by Dervieux-Boulieu [3] in 16 studies and the method described by Erdmann et al. [32] in 6 studies. In nine reports the measurements were performed in Prometheus Lab with a proprietary method. The study by Fangbin et al. (2016) [40] used the Dervieux-Boulieu methodology for measurement of 6-TGN and the Lennard method for 6-MMPR. Data for both metabolites (6-TGN and 6-MMPR) were available in 21 studies; 41 studies provided only usable data for 6-TGN and 10 studies for 6-MMPR. In six studies, it was also possible to evaluate the 6-MMPR/6-TGN ratio. We could identify some discrepancies concerning the nomenclature of methylated metabolites. The most common terms were 6-MMPR, 6-MMP, and methylated thioinosine monophosphate (meTIMP). In the methods commonly used for thiopurines' metabolites measurement, the methylated metabolites are hydrolyzed back to 4-amino-5-methylthiocarbonyl imidazole, a common derivative, rendering them indistinguishable. Therefore, both Dervieux and Lennard assays measure the sum of all methylated metabolites [13,41]. In this paper, the term used to describe these metabolites is 6-MMPR.

Description of the Studies
The details of the 72 studies included in the meta-analysis are presented in Table 1 and Table S1. The 72 included studies were published between 1983 and 2018. Most studies (n = 60, 83%) were from Western countries. The distribution between pediatric (n = 29, 40%) and adult cohorts (n = 30, 42%) was balanced. Most studies included patients with IBD only (n = 42, 58%), followed by ALL (n = 16, 22%). In the majority of reports, patients were treated with conventional thiopurines (n = 63, 89%) and were being treated for 8 weeks or longer at the time of the first metabolite measurement (n = 35, 49%). Most reports included patients concomitantly treated with other immunosuppressants or chemotherapy drugs. In 31 studies, data of patients without those concomitant drugs were provided and was used in our meta-analysis; still, steroids, mesalamine and/or antibiotics were allowed in most.
The measurement of metabolites followed the procedures described by Lennard [2,30] or were converted to approximate values in 33 studies, the methodology described by Dervieux-Boulieu [3] in 16 studies and the method described by Erdmann et al. [32] in 6 studies. In nine reports the measurements were performed in Prometheus Lab with a proprietary method. The study by Fangbin et al., (2016) [40] used the Dervieux-Boulieu methodology for measurement of 6-TGN and the Lennard method for 6-MMPR. Data for both metabolites (6-TGN and 6-MMPR) were available in 21 studies; 41 studies provided only usable data for 6-TGN and 10 studies for 6-MMPR. In six studies, it was also possible to evaluate the 6-MMPR/6-TGN ratio. We could identify some discrepancies concerning the nomenclature of methylated metabolites. The most common terms were 6-MMPR, 6-MMP, and methylated thioinosine monophosphate (meTIMP). In the methods commonly used for thiopurines' metabolites measurement, the methylated metabolites are hydrolyzed back to 4-amino-5-methylthiocarbonyl imidazole, a common derivative, rendering them indistinguishable. Therefore, both Dervieux and Lennard assays measure the sum of all methylated metabolites [13,41]. In this paper, the term used to describe these metabolites is 6-MMPR.  Multiple metabolites measurements per-patient were performed in 44 studies, whereas on 17 of them only one determination per patient was performed. When multiple determinations were performed, there was a wide variation concerning the value used in the analysis. When described, the used values are specified in Table 1. The type and definition of toxicities evaluated in each study are detailed in Table S1.
Following Cochrane recommendations, funnel plot asymmetry was only used when at least 10 studies were included [22]; even though in those cases forest plot analysis revealed a considerable amount of variability between studies, funnel plot did not suggest the existence of substantial publication bias.

Overall Adverse Events
In some reports, the evaluation of toxicity was performed in broader terms, with the inclusion of all toxicities related to thiopurines in the same group.

Myelotoxicity "General" Myelotoxicity
In some reports, toxicity combined different aspects of bone marrow suppression, here represented as "general" myelotoxicity.

6-TGN
In the pooled analysis of four studies, 6-TGN levels in patients with myelotoxicity showed no differences [46,48,58,80], with a mean difference of 169.14 pmol/8 × 10 8 RBC (95% CI −69.09; 407.37; p = 0.16). However, the heterogeneity was high (p < 0.01; I 2 = 93%). In the sub-analysis by duration of thiopurine treatment, only studies including patients with <8 weeks of therapy when metabolites were assessed showed numerically higher mean 6-TGN levels in patients with myelotoxicity ( Figure S3). A single study reported including patients using thiopurines in monotherapy [48]; in this study there were no differences in 6-TGN levels in patients with and without myelotoxicity, but only patients with more than 3 months of treatment were included.
Three studies provided data to calculate a pooled OR, all of them using a threshold of 450 pmol/8 × 10 8 RBC [36,46,58]. The risk of myelotoxicity was almost eight times higher in patients with elevated 6-TGN levels (OR = 7.78, 95% CI 1.67; 36.34; p < 0.01) (p = 0.24; I 2 = 31%). In one of these studies, thiopurines was used as part of a chemotherapy regimen, while in the other two concomitant medications were not specified.
Nine studies provided data to calculate a pooled OR [40,46,58,65,72,73,85,96,105], with thresholds ranging from 127 to 450 pmol/8 × 10 8 RBC. In these studies, patients with high 6-TGN levels were more than four times more likely to have leukopenia than those with lower levels (Figure 3c). In the pooled analysis of the three studies with thiopurines in monotherapy, patients with high 6-TGN levels were almost six times more likely to have leukopenia than those with lower levels (OR 5.87, 95% CI 3.27; 10.55; p < 0.01) (p = 0.58; I 2 = 0%).
Noticeably, in the sub-analysis by leukopenia definition, only studies defining leukopenia as a white blood count below 3 or 3.5 × 109/L had a significant association both in mean difference (MD) and OR analysis.
The data provided for different cut-offs enabled us to calculate an optimal cut-off of 135 pmol/8 × 10 8 RBC for leukopenia, with a sensitivity of 75.4% and specificity of 46.4% (area under the curve (AUC) = 0.629, 95% CI 0.432; 0.837) ( Nine studies provided data to calculate a pooled OR [40,46,58,65,72,73,85,96,105], with thresholds ranging from 127 to 450 pmol/8 × 10 8 RBC. In these studies, patients with high 6-TGN levels were more than four times more likely to have leukopenia than those with lower levels ( Figure  3c). In the pooled analysis of the three studies with thiopurines in monotherapy, patients with high 6-TGN levels were almost six times more likely to have leukopenia than those with lower levels (OR 5.87, 95% CI 3.27; 10.55; p < 0.01) (p = 0.58; I 2 = 0%).
Noticeably, in the sub-analysis by leukopenia definition, only studies defining leukopenia as a white blood count below 3 or 3.5 × 109/L had a significant association both in mean difference (MD) and OR analysis.

6-MMPR
The analysis of a pool of six studies demonstrated that the levels of 6-MMPR were not significantly higher in patients with leukopenia [46,58,72,85,92,105] (mean difference of 277.00 pmol/8 × 10 8 RBC (95% CI −237.06; 791.07; p = 0.29), with similar results when the analysis was limited to patients medicated with thiopurines in monotherapy (mean difference of 1161.78 pmol/8 × 10 8 RBC (95% CI −934.44; 3258.01; p = 0.28) The heterogeneity between studies was substantial (p < 0.01 for both; I 2 = 67% and 84%, respectively). In the subgroup analysis, method and time of treatment were significant variables. In fact, only studies using the Lennard method for metabolites assessment, and studies including patients with less than 8 weeks of treatment at the time of measurement reported higher levels of 6-MMPR with leukopenia ( Figure S5).
The pooled analysis of three studies that provided data for OR calculation revealed that there was no association between 6-MMPR levels and leukopenia (OR 2.02, 95% CI 0.37; 10.90; p = 0.42) (p = 0.09; I 2 = 59%). However, in the one study that only included patients with less than 8 weeks of treatment at the time of assessment [105], patients with 6-MMPR levels above 3525 pmol/8 × 10 8 RBC were almost six times more likely to develop leukopenia (OR 5.9, 95% CI 2.7-13.3) [46,58,105]. Additionally, of the three studies, this was the only study reporting the use of thiopurines in monotherapy.

Liver Toxicity
Altered Liver Enzymes
Nine studies provided data for the calculation of a pooled OR [14,36,44,50,57,75,82,102,104]. It was possible to conclude that patients with 6-MMPR levels above the defined thresholds, ranging from 3615 to 5700 pmol/8 × 10 8 RBC, were about four times more likely to develop hepatotoxicity (Figure 5c). When the analysis was restricted to studies reporting the use of thiopurines in monotherapy, the results were similar (OR 4.78; 95% CI 3.18; 7.19; p < 0.01) (p = 0.42; I 2 = 0%).

Veno-occlusive Disease
In the pooled analysis of the two studies that evaluated the relationship between 6-TGN levels and the occurrence of veno-occlusive disease (VOD) [42,89], no difference was found in the 6-TGN levels in patients with this type of toxicity (mean difference 7.95 pmol/8 ×

Pancreatitis, Infections and NRH
For each of these adverse events, only one study provided data on metabolites [66,80,109]. As such, it was not possible to perform meta-analysis, and these studies were not included.

Discussion
The wide use of thiopurines has been hindered by their inherent toxicity, which may also result in underdosing and lack of efficacy [5]. The measurement of thiopurines' metabolites can give physicians a safer context for prescription, if the levels are kept in the therapeutic range and below toxic thresholds. However, the therapeutic range may differ depending on the disorder and on the indication. In IBD, the most cited optimal range is of~230 to 400 pmol/8 × 10 8 RBC, when thiopurines are used in monotherapy [12,13]. Still, thiopurines are also important in combination with anti-TNF drugs, and may be the most important factor in reducing the immunogenicity of these drugs [110,111]. In combination therapy with biologics, lower levels (105 to 125 pmol/8 × 10 8 RBC) may be enough to improve anti-TNF pharmacokinetics [112,113].
In this study, we identified a relationship between thiopurines' metabolites and several adverse events: (i) 6-TGN were associated with leukopenia, neutropenia and gastrointestinal intolerance, and inversely associated with liver toxicity; and (ii) 6-MMPR were associated with liver toxicity and early leukopenia. As myelosuppression has long been linked to 6-TGN, we calculated an optimal 6-TGN threshold (135 pmol/8 × 10 8 ) for the occurrence of leukopenia. Although this cut-off is below the therapeutic levels for monotherapy with thiopurines in IBD, it is above the optimal cut-off for the levels demanded in combination therapy. This constitutes an additional argument for using lower doses of thiopurines when the drug is combined with infliximab. Regarding neutrophils and platelets, correlation with 6-TGN was only significant when the analysis was restricted to conventional thiopurines. Scientific evidence indicates that 6-TGN levels derived from 6-thioguanine have a different impact than those resulting from conventional thiopurines. Indeed, low doses of 6-thioguanine can lead to high 6-TGN levels without evidence of myelosuppression [114]. Some explanations can be pointed out for this fact. As most methods do not measure 6-TGN directly, but reduce it to thioguanine, the ingested 6-thioguanine is indistinguishable from 6-TGN, resulting in false high levels of 6-TGN if the drug is ingested close to the assay [115]. In addition, 6-MMPR are not produced with 6-thioguanine. Our results evidenced a possible association of early 6-MMPR assessment (i.e., in the first 8 weeks of treatment) with leukopenia. In previous studies, these metabolites were shown to be cytotoxic and to inhibit purine de novo synthesis, contributing to the antiproliferative properties of these drugs, responsible for both therapeutic and myelotoxic effects [116].
We have also confirmed a positive association of hepatoxicity with 6-MMPR, and a negative association with 6-TGN. In patients who metabolize thiopurines preferentially through the methylation pathway, generating high levels of 6-MMPR [13] (known as "shunters"), dose escalation will not always improve clinical outcomes. This explains thiopurines' inefficacy despite optimal weight-based dosage [68]. Some strategies can be used to improve the metabolite profile in these patients: (i) dose-splitting regimen [117]; (ii) addition of allopurinol [1,117]; or (iii) use of 6-thioguanine instead of a conventional thiopurine [114].
Gastrointestinal intolerance to thiopurines is one of the most frequent adverse events with thiopurines treatment, causing many patients to abandon treatment [118]. Some authors postulated that this adverse event could be related to the nitro-imidazole compound released in AZA metabolism to form 6-MP [119]. In this way, 6-MP could be an adequate alternative to AZA treatment in patients experiencing gastrointestinal intolerance, as was demonstrated in some studies [119,120]. However, we have found that 6-TGN levels were associated with the occurrence of gastrointestinal intolerance. Accordingly, in one of the studies included in the analysis, switch of AZA to 6-MP was only tolerated in a small proportion of patients [79].
The reported higher sensitivity of Asian populations to thiopurines, when compared to Western populations, deserves particular consideration [121]. In this context and whenever possible, we performed subgroup-analysis according to the geographical origin of the studies. Generally, we could not evidence a significant influence of ethnicity on toxicity. Differences in metabolism are probably responsible for these "different sensitivities". These results are strong arguments in favor of metabolites' monitoring rather than the traditional weight-based strategy. In fact, even though metabolism may vary according to ethnicity, the significance of concentrations of metabolites is likely similar across populations.
The results of this meta-analysis were impacted by several limitations. Most studies were retrospective, with small samples and, in many of them, the evaluation of the relationship between metabolites and toxicity was a secondary outcome. As such, toxicity events were possibly not always registered, and samples were often underpowered for conclusive results. The mean global quality score was only 58.9%, reflecting these factors. In most cases, studies displayed a high degree of heterogeneity and the definitions of toxic events were inconsistent, suggesting that the results should be interpreted with caution. This heterogeneity was, in part, caused by the different methods applied for the measurement of metabolites. Even though conversion factors for 6-TGN have been described to uniformize values, small variations in protocols can lead to significant differences in 6-TGN concentrations [1,31,35]. Levels of 6-MMPR are reported as being similar in different assays [35][36][37], but this causes a problem in the definition of optimal values for 6-MMPR/6-TGN ratios. Cut-offs obtained by a specific assay are not interchangeable [31] and thus the calculated optimal threshold for leukopenia should be used with caution in clinical practice. In the future, the standardization of procedures for the assessment of metabolites is of upmost importance [41,117]. Another limitation of this analysis is that most studies included patients with more than 8 weeks of treatment, but most adverse events with thiopurines occur in the first weeks/months [4,5,79]. Hence, a large proportion of patients had their doses of drug reduced (or even discontinued) at the time of the study. The strength of association between metabolites and adverse events could be stronger if more studies included patients in the beginning of thiopurine treatment. In fact, Wong et al. demonstrated that the assessment of metabolites at week 1 could predict the later occurrence of leukopenia and hepatotoxicity [104,105]. Some authors reduce the doses of thiopurines or change the treatment strategy in patients with high 6-MMPR based on similar experiences in their clinical practice, but this strategy is yet to be validated. [68] To perform this meta-analysis, we converted median values (reported in most studies) to mean values. This procedure could have introduced some error in the final analysis. However, whenever possible, we performed sub-analysis by "provided vs. calculated means", with no impact on the results. To finalize, many studies were excluded from the meta-analysis for not providing enough data for calculations. Most of these studies mentioned the absence of significant relationships between metabolites and adverse events, which might result in a positive results bias.
As many thiopurine-associated adverse events are related to the level of metabolites, physicians should take this information into account for dose selection, to achieve the best compromise between efficacy and toxicity. The importance of establishing a clear relationship between metabolite levels and toxicity may also be of value in patients receiving concomitant medications with similar toxicity profiles. In these cases, metabolites' measurement will help to determine the culprit. The same applies to disorders in which the clinical presentation resembles drug toxicity, as in the case of AIH flares. If the context enables TPMT and nudix hydrolase-15 (NUDT15) screening before starting the treatment to inform on eligibility or drug dosage, subsequent adjustments can be guided by the measurement of metabolites, in a tiered approach [117]. However, at this point, these strategies should be used as adjuncts in clinical practice and cannot yet replace blood and clinical monitoring for early detection of toxicity. A more personalized medicine should overcome the traditional weight-based dosing of thiopurines and rely more on TDM. Still, higher quality studies are needed to confirm this strategy.
Supplementary Materials: The following are available online at http://www.mdpi.com/2077-0383/9/7/2216/s1, Figure S1: Relationship between 6-TGN levels and overall toxicity: mean difference forest plot (AE-adverse events) with sub-analysis by age group; Figure S2: Relationship between 6-MMPR levels and overall toxicity: mean difference forest plot (AE-adverse events); Figure S3: Relationship between 6-TGN levels and overall myelotoxicity: mean difference forest plot with sub-analysis by duration of treatment; Figure S4: Relationship between 6-TGN levels and hemoglobin: correlation meta-analysis plot and sub-analysis of studies including patients with inflammatory bowel disease; Figure S5: Mean difference forest plot of 6-MMPR levels and leukopenia: (a) sub-analysis by method (b) sub-analysis by treatment duration; Figure S6: Relationship between 6 MMPR levels and leukopenia: correlation meta-analysis plot; Figure S7: Relationship between 6-MMPR/6-TGN ratio and leukopenia: mean difference forest plot; Figure S8: Relationship between neutrophils levels and 6-TGN levels: mean difference forest plot; Figure S9: Correlation meta-analysis plot of 6-TGN and platelets; Table S1: Type and definition of toxicities evaluated in each study included in the meta-analysis; Table S2: Measures of performance of different 6-TGN cut-offs for leukopenia occurrence (calculated with the method described in Steinhauser S,