NAT2 variants and toxicity related to anti-tuberculosis agents: a systematic review and meta-analysis

BACKGROUND: Tuberculosis (TB) patients receiving anti-tuberculosis treatment may experience serious adverse drug reactions (ADRs) such as hepatotoxicity. Variants of the N-acetyltransferase 2 (NAT2) gene may increase the risk of experiencing such toxicity events. OBJECTIVE: To provide a comprehensive evaluation of the evidence base for associations between NAT2 variants and anti-tuberculosis drug-related toxicity. METHOD: This was a systematic review and meta-analysis. We searched for studies in Medline, PubMed, EMBASE, BIOSIS and Web of Science. We included data from 41 articles (39 distinct cohorts of patients). We pooled effect estimates for each genotype on each outcome using meta-analyses stratified by country. RESULTS: We assessed the quality of the included studies, which was variable, with many areas of concern. Slow/intermediate NAT2 acetylators were statistically significantly more likely to experience hepatotoxicity than rapid acetylators (OR 1.59, 95%CI 1.26–2.01). Heterogeneity was not detected in the overall pooled analysis (I2 = 0%). NAT2 acetylator status was significantly associated with the likelihood of experiencing anti-tuberculosis drug-related hepatotoxicity. CONCLUSION: We encountered several challenges in performing robust syntheses of data from pharmacogenetic studies, and we outline recommendations for the future reporting of pharmacogenetic studies to enable high-quality systematic reviews and meta-analyses to be performed.

concern. Slow/intermediate NAT2 acetylators were statistically significantly more likely to experience hepatotoxicity than rapid acetylators (OR 1.59, 95%CI 1.26-2.01). Heterogeneity was not detected in the overall pooled analysis (I 2 ¼ 0%). NAT2 acetylator status was significantly associated with the likelihood of experiencing anti-tuberculosis drugrelated hepatotoxicity.
C O N C L U S I O N : We encountered several challenges in performing robust syntheses of data from pharmacogenetic studies, and we outline recommendations for the future reporting of pharmacogenetic studies to enable high-quality systematic reviews and meta-analyses to be performed. K E Y W O R D S : tuberculosis; pharmacogenetics; adverse events; evidence synthesis TUBERCULOSIS (TB) is one of the most important challenges in global health. There were an estimated 1.3 million TB deaths in 2016 among human immunodeficiency virus (HIV) negative people and 374 000 deaths among HIV-positive people. 1 The World Health Organization (WHO) recommends a combination of four first-line drugs for individuals with drug-susceptible TB: isoniazid (INH), rifampicin (RMP), ethambutol (EMB) and pyrazinamide (PZA). 1 TB patients receiving a combination of these drugs may experience adverse drug reactions (ADRs), the most serious of which is anti-tuberculosis druginduced hepatotoxicity (ATDH). Reported incidence rates of ATDH among patients treated with standard multidrug treatment vary from 2% to 28%, depending on the regimen given, definition of ATDH and patient characteristics such as age, race and sex. 2 ATDH can be fatal, with reported mortality rates of 6-12% if drugs are not promptly stopped. 3 ATDH and other anti-tuberculosis drug-related adverse effects also contribute to non-adherence, eventually leading to treatment failure, relapse and the emergence of drug resistance. 2 The proposed genetic risk factors for ATDH include polymorphisms of the N-acetyltransferase 2 (NAT2) gene, which codes for the drug-metabolising enzyme, NAT2. 4,5 NAT2 polymorphisms may affect the activity of the NAT2 enzyme, altering the chemical modification of anti-tuberculosis drugs and their metabolites in the liver, leading to hepatic adverse reactions. 6 Toxic metabolites may also cause other toxicity events, such as peripheral neuropathy and maculopapular eruption, although the majority of evidence on the pharmacogenetics of anti-tuberculosis drugs focuses on hepatotoxicity.
INH is the anti-tuberculosis drug for which the genetic contribution to ATDH has been most widely studied and is best understood. Specifically, it is thought that NAT2 acetylator status may be associated with INH-related hepatotoxicity because NAT2 is one of the main enzymes involved in INH metabolism in the liver. There are three phenotypes of acetylator status. Individuals who are slow NAT2 acetylators have higher plasma drug concentrations. This may be beneficial for treatment efficacy, but may also cause an accumulation of toxic metabolites as part of the metabolic activation of acetylhydrazine to harmless diacetylhydrazine. INH suppresses the acetylation of acetylhydrazine to produce more toxic metabolites, which contributes to the increased risk of hepatitis. 7 Fast acetylators have lower plasma drug concentrations, and so treatment may be less effective, but also less toxic. Intermediate acetylators fall between these two extremes.
RMP and PZA have also been reported to be hepatotoxic; 8 however, the mechanisms for RMPand PZA-induced hepatotoxicity are not known. 9 The OATP1B1*15 haplotype has been reported to be a predictor of RMP-induced liver injury; 10 no research into the genetic predictors of PZA-induced hepatotoxicity has been reported. 11 No hepatotoxicity has been described for EMB. 8 The objective of this systematic review and metaanalysis was to evaluate evidence on the effect of NAT2 on anti-tuberculosis drug-related toxicity in TB patients receiving anti-tuberculosis treatment. Meta-analyses investigating the effect of NAT2 on toxicity outcomes have been published, 6,12-15 but the conclusions from these have been conflicting. Our review and meta-analysis updates and adds to the evidence base on associations between NAT2 and anti-tuberculosis drug-related toxicity.

METHODS
This review was conducted in line with the methods outlined in our protocol (PROSPERO registration number: CRD42017068448). 16 A search strategy and study selection process enabled identification of studies that investigated the association between any genetic variant and anti-tuberculosis drug-related toxicity. However, in this article, we focus only on the subset of studies that considered NAT2 variants. Studies investigating associations between other genetic variants and anti-tuberculosis drug-related toxicity will be reported separately.

Selection criteria
Types of studies We included cohort studies, case-control studies and randomised controlled trials (RCTs). We did not include studies on case series because this type of study design would be inappropriate to investigate the effect of genetic variants on anti-tuberculosis drug-related toxicity. We did not require a minimum number of enrolled patients for a study to be included in our review.

Types of participants
We included studies that recruited TB patients who were either already established on anti-tuberculosis treatment or commencing treatment (at least one of INH, RMP, PZA or EMB), and who were genotyped to investigate the effect of genetic variants on antituberculosis drug-related toxicity. We only included studies where .50% of included patients were TB patients receiving anti-tuberculosis treatment.

Types of outcomes
We included studies that measured any drug-related toxicity outcomes.

Search strategy
An information specialist (EK) designed the search strategy (Appendix Tables A.1* and A.2), and searched for relevant studies in Medline, PubMed, EMBASE, BIOSIS and Web of Science (date of search: 3 March 2016). We searched reference lists from relevant studies manually, and contacted experts to identify eligible studies. We included studies published in English only. We did not restrict by year of publication or publication status.

Study selection
The search results were imported to Covidence. 17 We removed duplicates, and one author (MR) scanned the study abstracts to remove irrelevant studies. A second author (AJ, JK or KD) independently screened a sample of 10% of studies.
We obtained the full text for each potentially relevant study. One reviewer (MR) assessed eligibility based on the selection criteria. A second author (AJ, JK or KD) independently assessed a sample of 10% of studies for eligibility. Disagreements between the two reviewers at the abstract and full-text screening stages were resolved through discussion, and by consulting a third author if necessary.

Outcomes
The primary outcome of this review was hepatotoxicity by any definition used by the original investigators. The secondary outcomes were all other toxicity outcomes.

Data collection
We designed and piloted a data extraction form. We collected data on study design, participant characteristics, and treatment regimen and outcomes. One author (MR) extracted data in accordance with the methods outlined in the Cochrane Handbook 18 and The HuGENet HuGE Review Handbook. 19 A second author (AJ, JK or KD) independently extracted all outcome data. Disagreements between the two reviewers were resolved through discussion, and by consulting a third author if necessary. We contacted study authors if outcome data necessary for inclusion in a meta-analysis were not published in the paper.
We contacted individuals who were listed as authors of multiple included articles to enquire whether there was overlap between articles in terms of the patient cohorts. We examined locations, dates of recruitment and other study characteristics to identify articles that reported outcomes for the same patient cohort. If an author confirmed that multiple articles reported outcomes for the same patient cohort, or if we suspected this based on reported study characteristics, we assigned a group identifier (GI) to these articles, and ensured that no data for the same patient cohort were included more than once in any meta-analysis.

Quality assessment
One author applied criteria for the quality assessment of pharmacogenetic studies 20 to each study. A second author (AJ) independently assessed the quality of a sample of 10% of studies. Disagreements between the two reviewers were resolved through discussion. We obtained the number of studies meeting each criterion and summarised this information in the text.

Data synthesis
We performed meta-analyses for associations between NAT2 and any anti-tuberculosis drug-related toxicity outcome that were investigated by at least two studies. The effects of both NAT2 acetylator status (as predicted using genotyping methods) and individual NAT2 single-nucleotide polymorphisms (SNPs) were investigated.

Primary analysis
The primary analysis compared risk of hepatotoxicity for slow/intermediate acetylators in comparison with rapid acetylators. Data were pooled from studies that reported data for each acetylator group separately together with data from studies that combined slow and intermediate acetylator groups.
Two sensitivity analyses were conducted. The first was pairwise comparisons of slow vs. rapid acetylator status, and intermediate vs. rapid acetylator status. Here, it was only possible to include data from studies that reported on each acetylator group separately. The second was comparison of slow vs. rapid/ intermediate acetylator status. Here, data were pooled from studies that combined data for intermediate and rapid acetylator groups, and from studies that reported data for each acetylator group separately.

Secondary analysis
The secondary analysis compared the risk of hepatotoxicity between genotype groups for NAT2 SNPs. For each SNP, two pairwise comparisons were undertaken: heterozygous genotype vs. homozygous wild-type (wt), and homozygous mutant-type vs. homozygous wt. For SNPs investigated by one study only, odds ratios (ORs) comparing genotype groups were calculated and summarised in a table, together with the pooled estimates from the meta-analyses. There were insufficient data to perform metaanalyses for an association between NAT2 (acetylator status and individual SNPs) and other toxicity outcomes; ORs and 95% confidence intervals (CIs) for each pairwise comparison were calculated and reported in a table.
Meta-analyses were performed using Stata v 14 (metan package) (StataCorp, College Station, TX, USA); 21 ORs with 95%CIs were the chosen measure of effect. We used the random-effects model because we anticipated heterogeneity between studies due to differences in study design, methodological quality, ethnicity of participants and outcome definitions. The random-effects model used the method of DerSimonian and Laird, 22 with the estimate of heterogeneity being taken from the Mantel-Haenszel model. 23 If zero events were observed in one of the genotype groups, a continuity correction of 0.5 was used. Data were excluded from the analysis if there were no patients in one of the genotype groups in a comparison.
The HuGENet HuGE Review Handbook recommends that meta-analyses of genetic association studies be stratified by ethnicity, and that metaanalyses should only be performed if effect estimates for different ethnic groups appear sufficiently similar. 19 However, information on participants' ethnicity was sparsely reported in the studies included in our review. We therefore performed analyses stratified by the countries in which studies were conducted as a proxy for ethnicity.

Investigation of heterogeneity
We assessed heterogeneity by visually examining forest plots, and by referring to the I 2 statistic. If substantial heterogeneity had been observed (.50%), 18 we planned to undertake subgroup analyses according to study design, outcome definitions, treatment regimens and date of study publication.

Selective reporting
We assessed the possibility of selective reporting as part of the quality assessment. Potential sources of selective reporting considered were genetic variants, outcomes and modes of inheritance. 20

RESULTS
Included and excluded studies A Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow chart showing the selection of studies during the literature search is provided in Figure 1 (for more information, visit www.prisma-statement.org). 24 The initial search identified 77 articles investigating the association between any genetic variant and anti-tuberculosis drug-related toxicity, from which 52 distinct cohorts of patients were identified ( Figure 1).
Forty-six articles reported data for the association between NAT2 variants and anti-tuberculosis drugrelated toxicity; from these articles, 40 distinct patient cohorts were identified. In this review, we include data from 40 articles (39 distinct patient cohorts).  We did not include data from the remaining six articles. [65][66][67][68][69][70] Of those six articles, five reported data for patient cohorts for whom data were also reported in other articles (or we suspected that this was the case); for the sixth article, 70 the numbers of patients in each genotype group were not reported, and we were unable to obtain this information from the authors. The characteristics of studies included in this review are provided in Appendix Table A.3.

Quality assessment
Choosing which genes and SNPs to genotype Twenty-seven articles reported the reasons for choosing all genes and SNPs investigated. For the 13 articles 28, 31,33,37,48,51,53,57,59,60,62-64 that did not report this information, no articles limited their reporting to only statistically significant associations. Therefore, there was no evidence to suggest that selective reporting of genes and SNPs had occurred.

Sample size
The median sample size was 170 (interquartile range 108.5-285). Only two articles 26,63 provided details of the a priori power to detect pre-specified effect sizes.

Study design
Eleven articles described case-control studies, 27 articles described prospective cohorts, one article described a retrospective cohort and one article described an RCT. For one case-control study, 33 the case and control groups were not clearly defined. No articles describing case-control studies reported that the two groups were genotyped in mixed batches.

Reliability of genotypes
Only three articles 26,32,46 mentioned genotype quality control procedures, and only 12 articles 26,33,35,37,38,41,45,[49][50][51]53,55 compared the genotype frequencies of all investigated SNPs to those previously published for the same population. Of the articles describing case-control studies and retrospective cohorts, only two 45,46 mentioned that genotyping personnel were blinded to outcome status.
Missing genotype data For most articles (29/40), on comparison of the number of participants included in the analyses with the study sample size, it was apparent there were no missing genotype data. For the remaining 11 articles, 32,33,[42][43][44]53,56,58,60,63,64 only five articles 32,56,58,63,64 summarised the extent of missing data for all the genes and SNPs analysed. None of these articles described checking whether missing data were randomly distributed.

Population stratification
One article mentioned undertaking tests for population stratification; 53 no population stratification was identified. One article used a study design that ensured that the included patients were from a nondiverse ethnic group. 48 All other studies were at potential risk from confounding due to population stratification.

Mode of inheritance
Nineteen articles made a specific assumption reg a r d i n g t h e u n d e r l y i n g m o d e o f i n h e r itance. 25,29,31,34,35,40,43,44,48,50,53,[55][56][57][59][60][61]63,64 Of these, only two provided justification; 29,60 for the remaining 17 articles, there was a risk of selective reporting under different modes of inheritance. Two articles 42,58 applied models assuming different modes of inheritance to the genotype data, although only one of these articles 42 adjusted these analyses for multiplicity of testing.
Choice and definition of outcomes There was large variation in the definition of hepatotoxicity (Appendix Table A.4). Of the 37 articles reporting hepatotoxicity data, one did not provide a definition, 62 one provided a vague definition, 30 and the remaining 35 articles provided 31 different definitions. Definitions of other toxicity outcomes were generally not sufficiently detailed (Appendix Table A.5).
Nine articles did not provide justification for the choice of outcomes, but outcomes were in line with the main study aim as conveyed in the Introduction section of the article. 27,32,38,49,50,52,56,57,63 The remaining articles all provided justification for the choice of outcomes. There was therefore no evidence to suggest that selective reporting of outcomes had occurred.

Treatment adherence
Six articles 31,32,43,45,50,57 mentioned assessing treatment adherence. One article 48 reported that treatment was administered under DOTS; it was therefore not necessary to measure adherence. Of the six articles that reported assessing adherence, one did not report adjusting the analyses for adherence. 50 It was not necessary to adjust for adherence in the analyses of two articles because patients were reported to have good treatment adherence. 31,32 Association between NAT2 variants and antituberculosis drug-related toxicity NAT2 acetylator status and hepatotoxicity A forest plot displaying the results of the primary analysis is given in Figure 2. Slow/intermediate acetylators were significantly more likely to experience hepatotoxicity than rapid acetylators (OR 1.59, 95%CI 1.26-2.01). No heterogeneity was detected in this analysis (I 2 ¼ 0%).
Results of the sensitivity analyses are provided in Appendix Figures A.1-A.3. Results from the pairwise comparisons suggested that slow acetylators were significantly more likely to experience hepatotoxicity than rapid acetylators (OR 3.68, 95%CI 2.23-6.09, I 2 ¼ 60.0%), but there were no significant differences between intermediate and rapid acetylators (OR 1.12, 95%CI 0.87-1.45, I 2 ¼ 0%). The sensitivity analysis that compared slow acetylators with rapid/intermediate acetylators suggested that slow acetylator status significantly increased the risk of hepatotoxicity (OR 3.12, 95%CI 2.45-3.97, Moderate heterogeneity was observed in the sensitivity analyses of slow vs. rapid acetylator status, and slow vs. rapid/intermediate acetylator status. Such moderate heterogeneity may have been due to the variable distribution of genotypes in different geographic areas. The funnel plot for the primary analysis (Appendix Figure A.4) provided no evidence of publication bias.

NAT2 SNPs and hepatotoxicity
The included studies reported data for 12 NAT2 SNPs. A summary of all data for the association between NAT2 SNPs and hepatotoxicity is provided in Table 1. There were sufficient data to perform meta-analyses for six SNPs. Forest plots showing the results of these meta-analyses are provided in Figure  2. The four main findings from these meta-analyses are shown below.  NAT2 variants and other toxicity outcomes A summary of all data for the association between NAT2 variants and toxicity outcomes (other than hepatotoxicity) is provided in Table 2. Each reported result is based on data from a single study because there were no comparisons where more than one study provided data ( Table 2).
For peripheral neuropathy, no significant associations were reported for either of the pairwise comparisons conducted for acetylator status, 191G-A or 341T-C. Similarly, for skin rash and eosinophilia, the pairwise comparisons for acetylator status demonstrated no significant effects. None of the SNPs investigated by Kim et al. had a significant effect on anti-tuberculosis drug-induced maculopapular eruption. 43 Slow acetylators were significantly more likely to experience adverse drug-induced hepatotoxicity outcomes (definition unclear; OR 3.31, 95%CI 1.03-10.62), and ADRs (defined as at least one of the following: gastric, joint, neuromuscular or skin reactions, hepatotoxicity; OR 3.20, 95%CI 1.31-7.80) compared with rapid or intermediate acetylators. However, slow acetylator status was not found to increase the risk of gastrointestinal ADRs.

DISCUSSION
There is substantial evidence for the association between NAT2 variants and anti-tuberculosis drugrelated toxicity outcomes, as previously identified and as our systematic review confirmed. However, we established that performing robust synthesis of this evidence is challenging due to the variability between studies in terms of how participants are classified according to genotype; choice and definition of outcomes and variants to investigate; ethnicity of participants; and methodological quality. In conducting our review, we carefully considered these challenges, stratifying meta-analyses by genetic variants, genotype contrasts and outcomes. We also stratified further by the country where the study was conducted as a proxy for ethnicity, which has not been widely reported. We supplemented our data synthesis with a rigorous assessment of the methodological quality of included studies.

Meta-analyses
Where possible, meta-analyses were undertaken to improve the power to estimate genetic effects. We found that slow/intermediate acetylators were significantly more likely to experience hepatotoxicity than rapid acetylators. This result is consistent with the findings of several meta-analyses, 12-15 but not consistent with the meta-analysis reported by Sun et al., 6 who did not identify a significant association between slow acetylator status and hepatotoxicity. However, the search date for Sun et al. 6 (May 2007) is several years earlier than the search dates for the other metaanalyses, and many relevant studies have been published in recent years. As more studies are published, the power to detect a statistically significant association increases.
Meta-analyses on individual SNPs of the NAT2 gene have not been published, so our results add to the existing knowledge of the association between NAT2 variants and hepatotoxicity. INH remains an essential drug in the treatment of active TB and is the mainstay of chemoprophylaxis in latent tuberculous infection (LTBI), an intervention that is being rapidly expanded in recent strategies to eliminate TB as a public health problem. The global use of INH will therefore greatly increase worldwide in the coming decade. While transaminase testing is a readily available biomarker of possible ATDH, baseline values have modest predictive value and routine monitoring is not generally recommended. Where slow acetylator status is common, pharmacogenetic testing could make a clinically useful contribution to risk stratification for ATDH. However, the need for testing of a relatively large panel of SNPs and the current lack of a clear substitute to INH for LTBI chemoprophylaxis mean that such a strategy may not be cost-effective or feasible. Studies investigating the cost-effectiveness and/or feasibility of such a strategy would be beneficial. Nevertheless, based on the nearly three-fold increased risk of ATDH in slow acetylators observed in this review, pharmacogenetic epidemiology should certainly be a factor in national policymaking on the need for transaminase monitoring during treatment of active TB and LTBI locally.

Quality assessment
The quality of included studies varied, with some areas of concern. Most studies were significantly smaller than typically required to provide sufficient power, 20 and the reader was left unaware of the likelihood of false-negatives in all studies due to the lack of reported a priori power calculations. The fact that no studies described checking that missing data were missing at random is also a concern; missing genotype data are unlikely to be missing at random because heterozygotes are notoriously more difficult to call than homozygotes. 20 Few studies reported testing of HWE, which can highlight genotyping errors, population stratification and other problems. 20 Furthermore, in studies that did not adjust for treatment adherence, the proportion of variability explained by genetic variants may have been underestimated. 20 As the quality assessment was qualitative rather than quantitative, it was not possible to exclude studies from meta-analyses based on a single summary score. Although we identified issues of concern relating to some of the quality criteria, we did not identify any studies that were thought to be of particularly poor quality overall, so we did not deem it necessary to exclude any single study in sensitivity analyses.

Limitations
Most included studies did not report the ethnic background of participants. We therefore performed NAT2 and anti-tuberculosis drug-related toxicity analyses stratified by the country in which the study was conducted as a proxy variable for ethnicity. It is clear that this approach is not ideal as the population of any given country is often ethnically diverse. However, stratifying by country was deemed the most suitable approach in the absence of definitive information on ethnicity. An additional challenge was identifying distinct patient cohorts from the included articles. If multiple articles report data for the same patient cohort, data for this patient cohort must only be included in metaanalysis once, otherwise a unit-of-analysis error occurs. 18 We found that it was often not possible to determine from the articles alone whether the patient cohorts were identical. We contacted several study authors for clarification. For two articles, 51,68 we did not receive a response and, consequently, data from the older article 68 were excluded from a metaanalysis to which both articles contributed data. If the two articles reported data for two distinct cohorts, then information would have been lost by excluding one article. Furthermore, there may have been cases of multiple articles reporting outcomes for the same cohorts that we did not identify; if this was the case, some patients may have been doublecounted in the meta-analyses.
There was considerable variability in the definitions of hepatotoxicity in the included studies, which introduced heterogeneity into the meta-analyses. Jorgensen et al. 71 and Contopoulos-Ioannidis et al. 72 made similar observations about the variability of definitions of outcomes across pharmacogenetics studies. If outcome definitions were more consistent between pharmacogenetic studies, the amount of heterogeneity observed in meta-analyses would have been reduced.
Finally, an important limitation of the systematic review was a lack of evidence from studies conducted NAT2 and anti-tuberculosis drug-related toxicity in Africa. There is a great deal of NAT2 diversity across Africa, 73 where TB is endemic, but there has been little mapping of pharmacogenomic polymorphisms in African populations. Only four studies included in this review were conducted in Africa. The vast majority of evidence included in this review is therefore not representative of the global population most affected by TB.

Recommendations for authors of pharmacogenetic studies
We made several recommendations regarding the reporting of future pharmacogenetic studies to facilitate the conduct of high-quality systematic reviews and meta-analyses, and thus improve the power to detect genetic associations.
1 Report the number of patients in each genotype group; 2 Report outcomes for each genotype group separately (i.e., number of events for dichotomous outcomes, and mean and standard deviation values for continuous outcomes); 3 Report the rs number of each genotyped SNP; 4 Report the ethnicity of included patients; 5 If a study includes more than one ethnic group, provide the summary data specified in 1) and 2) per ethnic group; 6 Provide the reference to the published protocol; 7 Provide information on patient cohort overlap; 8 Report full details of all variants and outcomes investigated, and of all analyses undertaken; 9 Consensus should be reached between experts in specific areas of research on the definitions of outcomes that are commonly reported in pharmacogenetic studies of a particular treatment.
We also recommend that articles adhere to the criteria of the quality assessment tool 20 as improvement in the methodological quality of studies included in meta-analyses would in turn improve the strength of the evidence synthesised in metaanalyses. Furthermore, we recommend that STREGA reporting guidelines are referred to, which provide guidance on the reporting of genetic association studies in general. 74

CONCLUSION
This review showed that slow/intermediate acetylators were significantly more likely to experience hepatotoxicity than rapid acetylators. Therefore, pharmacogenetic testing may be useful in clinical practice in terms of risk stratification for ATDH during treatment of TB. However, more studies are needed to overcome the reported methodological limitations and to assess if this strategy might be feasible and cost-effective.   TITLE: (((aminosalicylic acid or diarylquinoline* or ethambutol* or ethionamide* or isoniazid* or prothionamide* or pyrazinamide* or thioacetazone* or capreomycin* or cycloserine* or enviomycin* or rifabutin* or rifampin* or viomycin*))) Approximately 49 386

#3
Search (#1 or #2) 151 329 #4 Search ((((Genetic or gene*) near/2 near/2 ))) AND associat*) AND ((studies or study or analys*)) 3 922 #5 Search (((genetic* or gene*))) AND ((suscept* or predisposit* or anticipat*)) 235 548 #6 Search ((single*) AND nucleotid*) AND polymorph* 98 071 #7 Search ((SNP or Genotyp* or Phenotyp* or Allele* or Pharmacogenet* or Pharmacogenom* or Polymorph*)) 997 538 #8 Search (((gene* or genetic*))) AND ((mutat* or variant*)) 743 819 #9 Search (#4 or #5 or #6 or #7 or #8) 1 553 428 #10 Search (((((TB or Tuberculosis* or Antitubercul*))))) 251  The causality of drug-induced hepatotoxicity was determined according to the report of an international consensus meeting. 1 These criteria include 1) an increase of liver transaminases levels of .2 times above the normal value (,40 IU/l) for AST and ALT, 2) an improvement of this pattern after the drug withdrawal, and 3) the absence of alternative causes of this disorder Gupta, 2013 (GI: GUPTA) Increase in ALT . 23 ULN or a combined increase in AST and bilirubin levels, provided one of them is .23 ULN, was defined as ATDH according to the international consensus meeting 1 Higuchi, 2007 DIH was defined according to the criteria of the international consensus meeting, 1 i.e., development of a 72-fold increase in serum ALT level above the ULN range: N (642 IU/l), or a combined increase of . 2 N in serum AST (N 6 33 IU/l) and total bilirubin (N 6 1.5 mg/dl) Ho, 2013 The criteria for the diagnosis of hepatotoxicity was an elevation in liver function tests, AST and/or ALT of . Anti-tuberculosis drug-induced hepatitis was diagnosed as 1) an increase in serum ALT level . 23 ULN during treatment, according to the criteria established by the international consensus meeting; 1 2) negative serum HBV surface antigen, IgM antibody to HAV, and antibody to HCV when ALT or AST is elevated; 3) without any other major hepatic or systemic diseases that may induce elevation of liver biochemical tests, such as alcoholic liver disease, autoimmune hepatitis, congestive heart failure, hypoxia, and bacteremia; and 4) a causality assessment score . 5 (when classified as 'probable' or 'highly probable' drug-induced hepatitis), as derived from the international consensus meeting 1 Khalili, 2011 Hepatotoxicity was defined as 1) increased levels of liver transaminases . 3 times above the normal value (,40 U/l for AST and ALT) with any other clinical signs and symptoms; or 2) elevation of transaminases . 53 ULN, if patients had no symptoms. For evaluation of causality, The Roussel Uclaf Causality Assessment Method scoring system was used 2 Kim, 2009 (GI: KIM) Anti-tuberculosis drug-induced hepatitis was defined as an elevation in the serum levels of ALT . 23 ULN (640 U/ml) during treatment and normalisation of these values after cessation of medication according to the criteria from the international consensus meeting 1 Leiro-Fernandez, 2011 ATDH was defined as an increase in serum transaminase (either AST or ALT) to values . 33 ULN (i.e., .120 IU/l) at any time during the treatment period Lv, 2012 ATDH was designated as an increase of .23 ULN value in ALT or a combined increase in AST and total bilirubin provided one of them is .23 ULN. In this study, the ULN of ALT, AST and total bilirubin were respectively 40 U/l, 40 U/l and 19 lmol/l Causality assessment result was highly probable, probable or possible based on the CIOMS scale 1 Mahmoud, 2012 DIH was diagnosed as 1) an increase in serum ALT level greater than twice the ULN during the treatment, according to the criteria established by the international consensus meeting; 1 2) negative serum HBV surface antigen, IgM antibody to HAV, and antibody to HCV when ALT or AST was elevated; 3) without any other major hepatic or systemic diseases that may induce elevation of liver biochemical tests, such as alcoholic liver disease, autoimmune hepatitis, congestive heart failure, hypoxia, and bacteremia; when the French imputability score 4 was classified as 'probable' or 'likely' or 'certain ' Ng, 2014 All cases of DILI met at least one of the following biochemical criteria for enrolment into this study: 1) ALT . 53 ULN, 2) ALP . 23 ULN, or 3) ALT . 33 ULN and bilirubin . 23 ULN Ohno, 2000 Hepatotoxicity was estimated as follows: AST and/or ALT . 1.53 ULN and 23 before administration Possuelo, 2008 (GI: POSSUELO) Criteria for the diagnosis of hepatotoxicity was an elevation in liver function tests, AST and/or ALT of .33 ULN (reference: respectively 40 and 65 U/l) and/or in total bilirubin up to .2.0 mg/dl in the presence of such gastrointestinal symptoms as anorexia, nausea, vomiting and/or jaundice, with a normalisation of serum ALT level after discontinuation of the anti-tuberculosis drugs Rana, 2014 (GI: RANA) ATDH was defined according to international consensus criteria. 1 Patients with a rise in serum AST or ALT levels 7 53 ULN, irrespective of symptoms and serum bilirubin levels, or patients with rise in serum AST or ALT levels 7 23 ULN with hyperbilirubinaemia and an absence of serological evidence of infection with hepatitis viruses (A, B, C and E) were considered as having ATDH NAT2 and anti-tuberculosis drug-related toxicity i    R E S U LTA D O S: La calidad de los estudios incluidos fue variable, con muchos aspectos que son fuente de preocupación. La probabilidad de hepatotoxicidad en los acetiladores lentos e intermedios de NAT2 fue significativamente mayor que en los acetiladores rá pidos (OR 1,59; IC95% 1,26-2,01). No se detectó heterogeneidad en el aná lisis global combinado (I 2 ¼ 0%). El fenotipo acetilador de NAT2 se asoció de manera significativa con la probabilidad de aparición de hepatotoxicidad debida a los medicamentos antituberculosos.
C O N C L U S I Ó N: Se encontraron dificultades importantes al tratar de realizar una síntesis sólida de los datos de estudios farmacogenéticos, por lo cual se proponen recomendaciones sobre la comunicación de los resultados de futuros estudios farmacogenéticos, que faciliten la realización de revisiones sistemá ticas y metaná lisis de gran calidad.