The Impact of Genetic and Non-Genetic Factors on Warfarin Dose Prediction in MENA Region: A Systematic Review

Background Warfarin is the most commonly used oral anticoagulant for the treatment and prevention of thromboembolic disorders. Pharmacogenomics studies have shown that variants in CYP2C9 and VKORC1 genes are strongly and consistently associated with warfarin dose variability. Although different populations from the Middle East and North Africa (MENA) region may share the same ancestry, it is still unclear how they compare in the genetic and non-genetic factors affecting their warfarin dosing. Objective To explore the prevalence of CYP2C9 and VKORC1 variants in MENA, and the effect of these variants along with other non-genetic factors in predicting warfarin dose. Methods In this systematic review, we included observational cross sectional and cohort studies that enrolled patients on stable warfarin dose and had the genetics and non-genetics factors associated with mean warfarin dose as the primary outcome. We searched PubMed, Medline, Scopus, PharmGKB, PHGKB, Google scholar and reference lists of relevant reviews. Results We identified 17 studies in eight different populations: Iranian, Israeli, Egyptian, Lebanese, Omani, Kuwaiti, Sudanese and Turkish. Most common genetic variant in all populations was the VKORC1 (-1639G>A), with a minor allele frequency ranging from 30% in Egyptians and up to 52% and 56% in Lebanese and Iranian, respectively. Variants in the CYP2C9 were less common, with the highest MAF for CYP2C9*2 among Iranians (27%). Variants in the VKORC1 and CYP2C9 were the most significant predictors of warfarin dose in all populations. Along with other genetic and non-genetic factors, they explained up to 63% of the dose variability in Omani and Israeli patients. Conclusion Variants of VKORC1 and CYP2C9 are the strongest predictors of warfarin dose variability among the different populations from MENA. Although many of those populations share the same ancestry and are similar in their warfarin dose predictors, a population specific dosing algorithm is needed for the prospective estimation of warfarin dose.

summary of the evidence on warfarin pharmacogenetics in this region is needed in-order to identify the gaps and provide future directions. Our goal was to systematically review studies from the MENA region that estimate the impact of genetic and non-genetic factors on warfarin dose requirements.
Study objectives: to systematically review all observational studies that have explored the effect of both genetic and non-genetic factors on warfarin dose variability in the MENA region, estimate the prevalence of the studied genetic variations that are associated with warfarin dose variability, and compare the performance of different genetic-based algorithms in warfarin dose prediction.

Search Strategy
A systematic comprehensive search for observational studies was applied with no date or language restrictions. The search started on February 2016 and was completed by the end of March 2016.
Database search. PubMed, Scopus, MedLine, PharmGKB (Pharmacogenomics Knowledge Base), and PHGKB (Public Health Genomics Knowledge Base) were searched using different terms with the appropriate Booleans. Below is an example of the search terms and Boolean used to search PubMed. Similar terms were used to search other databases, for Scopus and MedLine results were refined by country.
Other resources. For grey literature, Google Scholar was searched without any language or date limits. Moreover, reference list of all kind of relevant review articles retrieved during our search, were hand searched to identify any potential study.

Study Types
Observational cross sectional or cohort studies were included. Randomized controlled studies, case control, case series, and qualitative studies were excluded from this review.

Participants
Adult patients from MENA region, who have been on warfarin for a sufficient amount of time to have a stable dose and their International Normalized Ratio (INR) within the therapeutic range.

Outcome Measure
Included studies were required to have mean warfarin dose as their primary outcome and to have estimated the variability in the dose explained by genetic and non-genetic factors.

Eligibility Criteria
Studies were considered eligible for this review if they have studied populations of the MENA region, have explored the effect of both genetic and non-genetic factors on warfarin dose, and have developed a dosing model.

Study Selection
Potentially relevant studies were screened first by title and abstract to exclude any irrelevant studies. Then studies were screened by text to exclude any study that does not apply the inclusion criteria. Both former steps were carried by two reviewers (HE and LB), independently. Whenever there was a disagreement, it was resolved by discussion.

Data Extraction and Management
Data were extracted with a pre-specified and piloted extraction tool that was prepared and reviewed by the authors. Abstracted data included: author and year of publication, population studied, sample size, genetic and non-genetic factors explored, mean warfarin dose, mean INR, and main findings.

Quality Assessment
Quality assessment tool for observational cohort and cross sectional studies, adopted from the NIH, National Heart, Lung, and Blood Institute [17] was used to assess the quality of all included studies.

Data Synthesis
All extracted data were summarized and presented descriptively.

Study Selection
A total of 17 studies were included in this review, including 8 different populations. Fig 1 shows a flow chart of the included studies. Out of 32 studies screened for eligibility, 12 studies were excluded for not targeting patients from the MENA region [7,9,10,[18][19][20][21][22][23][24][25][26]] and 3 studies were excluded [27-29] as they did not develop a dosing model. We identified studies from 8 different countries including: Egypt, Iran, Lebanon, Turkey, Sudan, Oman, Kuwait, and two studies in occupied Palestine on Israeli population. Characteristics of all studies are shown in Table 1.
After assessing for quality, most studies were found to be ranging from fair to good quality. However, the study by Alrashid et al [38] was considered of poor quality as there was no reporting of the r 2 value for the dosing model. The sample-size used in some of the studies was less than 100, which is relatively low and may have affected the power of their conclusions [31,32,34,39]. Testing for the Hardy-Weinberg equilibrium was reported in all studies except those by Yildirim et al [44] and Ghozlan et al [32].
Targeted populations were required to be on stable warfarin dose as part of the inclusion criteria. Due to the variability in the definition of stable warfarin dose among studies, we reported the different definitions used by authors of each included study (Table 2). Overall, having an INR within therapeutic range for three consecutive visits while on the same warfarin dose was considered "stable warfarin dose" description by almost all studies.

Prevalence of the studied genetic variants
In-order to estimate the prevalence of different genetic variants we reported the minor allele frequency (MAF) of these variants. For the most common genetic variants in VKORC1 and CYP2C9, MAF's are presented in Table 3. Variants in the VKORC1 gene had the highest frequencies among all populations followed by the Ã 2 variant of CYP2C9 gene and Ã 3, with Ã 2 being relatively higher in most of the populations.
Other less common genetic variants have been explored in Egyptians, Israeli, Omani, and Turkish populations. These include variants in the following genes: CYP4F2, EPHX1, GGCX1, MDR1, factor VII, APOE, CALU and PZ. The MAF for the CYP4F2 (rs2108622) was 40% among Turkish [42], 30% in Omani [40] and 42% in Egyptians [12]. The EPHX1 encodes for a putative subunit of the vitamin K epoxide reductase complex called: microsomal epoxide hydrolase 1 [10]. The MAF of EPHX1 (rs2292566) was 16% and 26.19% in Turkish and Egyptians, respectively [33,42], while another variant the (rs1051740) was Other gene variants that are associated with warfarin dose variability are the APOE and CALU [10]. The rs429358 and rs7412 variants of the APOE gene had a MAF of 6.7% and 7.4%, in Egyptians, respectively [12]. On the other hand, the rs339097 variant of the CALU gene was less common in Egyptians than the APOE variants with a MAF of 2.3% [12]. Other rare genetic variants that have been explored are the Factor VII (-401G>A) which had a MAF of 35% in Turkish [42], and the GGCX (rs699664) which had a MAF of 29.5% in Israeli [36].

Significant Predictors of Warfarin Dose
Using univariate and multivariate linear regression, authors of all studies were able to identify the most significant predictors of warfarin dose and the most significant dosing model which explained the highest percent of the dose variability. Table 4 shows the significant predictors in the different populations and to what extent were they able to predict warfarin dose.

Summary of Evidence
The term MENA refers to countries from Morocco to Iran and down to Sudan. Studies of maternally inherited mitochondrial DNA have suggested that all modern populations, including those of MENA [46][47][48], have originated through a single wave from Sub Saharan Africa and populated other parts of the world [49]. Later-on, several historical events took place and facilitated genetic migration like the early Islamic conquests, the Ottoman Empire expansion and the continuous migrations in and out of the Arabian Peninsula and Europe. Other examples include the silk and spice road, which used to connect China with Europe through the Middle East [50]. Countries like Egypt and Oman, their strategic crossroad position between Africa and Eurasia have left them with high intrapopulation diversity [46]. On the other hand, other countries like Qatar and Yemen have the highest rates of consanguinity, 44.7% and 54%, respectively, leaving them with less genetic variability [51]. All of the aforementioned reasons have made these countries of particular interest in genetic studies.
Pharmacogenetics studies have shown that variants of VKORC1 and CYP2C9 are strongly associated with warfarin dose. VKORC1(-1639G>A) alone can explain up to 40% of the dose variation in some populations [52]. Along with these genetic predictors, other genetic factors and non-genetic factors such as age, weight, gender, race, ethnicity, concurrent medication, and comorbidities can boost the preemptive estimation of warfarin dosing to more than 50% [10].
The fact that VKORC1 variants are the most significant predictors of warfarin dose followed by those of CYP2C9, was consistent across all studies included in this systematic review. However, the frequency of these genetic variations and the extent of their effect on dose variability varied from population to another. Similarities were seen between Sudanese and African Americans in that CYP2C9 Ã 5, Ã 6, and Ã 11 were better predictors of warfarin dose than the Ã 2 and Ã 3 variants [53]. Moreover, the variability explained by variants of VKORC1 (rs9934438, rs9923231, rs8050894) were comparable to those of African Americans [1,53]. CYP2C9 Ã 8 (rs9332094) is another rare variant that affects around 12% of African Americans and has been associated with reduced S-warfarin clearance [54]. This SNP was only investigated in two populations of MENA; Egyptians and Sudanese [12,41]. Despite some of the similarities found between these two populations and African Americans, the CYP2C9 Ã 8 did not shown any significant association with warfarin dose requirements in Egyptians and Sudanese as it did in African Americans. This might be in part due the fact that this variant was not very common in these populations, it showed a MAF of 0.008 and 0.01 in Egyptians and Sudanese, respectively [12,41]. On the other hand, the frequencies of CYP2C9 Ã 2 and Ã 3 and VKORC1 (3673G>A) reported in Turkish [43] were closer to Caucasians than to Asians and African Americans [1]. Variants of the VKORC1 gene were the most prevalent among all populations. Great discrepancy was seen in the frequency of the VKORC1 in Egyptians, it ranged from 30% in the study be Ghozlan et. al and up to 72% in the study by Ekladious et. al. [32]. Such discrepancy could be in part due to the small sample sizes used in the different studies concerning the Egyptian population, in particular the studies by Ghozlan, Ekladious, and Bazan and their colleagues [31, 32, 34]. Moreover, the Egyptian population is a heavily admixed population with different ethnic origins [12] which may have led to different frequencies depending on the ethnicity of the studied population. However, none of these studies have reported the different ethnic origins of their studied population and whether it was a contributing factor or not. Carriers of the -1639G>A variant required lower warfarin dose compared to the wild type in all studies. Shrif and colleagues were able to identify four different SNPs downstream from the VKORC1 gene (rs7199949, rs11865038, rs11864839, and rs750952). These SNPs were found to be located in newly identified gene which is only 2 kb downstream from the VKORC1 gene in the same reverse strand named: POL3S (polyserase 3S) [41]. Amino acid sequences of the POL3S are highly conserved among humans, chimpanzees, dogs, cows, and mice [55]. While Omani [40] • VKORC1 3673(GA & AA) • CYP2C9 (*2/*3, *3/*3)

• Atrial fibrillation 63%
Sudanese [41] • VKORC1(rs8050894,rs7199949, rs7294), • CYP2C9 (*2,*5,*6,*11) function of the encoded protein is still unknown in-vivo, it has shown to degrade the alpha chain of fibrinogen as well as pro-urokinase type plasminogen in vitro, indicating its potential role in the coagulation process [55]. Out of the identified SNPs in the POL3S, rs7199949 (p. Pro406Ala) proved to be a significant predictor of warfarin dose and its addition to the regression model explained 7% of the variability [41]. Variants in the CYP2C9 were less common than those of the VKORC1 in all populations with the Ã 2 variant being slightly more common than the Ã 3 variant in majority of cases. The Ã 2 variant was highest among Iranian (27%) [35], while the Ã 3 variant was highest among Turkish (27%) and it was not detected in Sudanese [41,44]. MAF of Ã 2 and Ã 3 were closely similar among Egyptians, Israeli, and Turkish [30,36,43]. Both Ã 2 and Ã 3 variants were associated with lower warfarin dose requirements in all populations.
The VKORC1 (-1639G>A) variant was the most significant predictor of warfarin dose in all population. This variant alone was able to explain 45% of the dose variability in Omani [40]. The VKORC1 Asp36Tyr was significantly associated with higher warfarin doses in Egyptians and when added to Shahin et al. dosing model it further explained the variability by only 2% [30].
CYP4F2 is the gene coding for the cytochrome P450 4F2 enzyme, which is a vitamin K oxidase. The V433M allele is associated with reduced metabolism of vitamin K1 and higher warfarin doses [56]. Caldwell et al. has shown that CYP4F2 V433M (rs2108622) variant has an impact on warfarin dose and that carriers of at least one minor allele required 4-12% higher doses than those with homozygous wild type genotype [57]. Only 3 studies in three different populations of MENA -Egyptian, Omani, and Turkish-have investigated the association between warfarin dose and that gene variant. These studies have shown no significant association between CYP4F2 gene variant and warfarin dose in Egyptians neither in Omani, despite its high frequency [12,40]. A significant association between CYP4F2 gene variant and warfarin dose was among Turkish by Ö zer et. al [42]. Furthermore, Turkish carriers of the variant allele required higher warfarin doses than those with the wildtype. The APOE gene encodes for an apolipoprotein that is involved in the hepatic clearance of vitamin K [10]. In the Egyptian cohort studied by Shahin. et. al., carriers of the APOE ε2 gene variant required lower warfarin dose compared to the wild type [12]. Moreover, that variant explained 2.53% of the dose variability. No association was found between MDR1 C3435T, EPHX1 H139R, GGCX and protein Z A-13G gene variations and warfarin dose in populations of MENA. However, when combined, these genetic variants were able to explain 3% of the dose variability in Egyptians [33]. Calumenin (CALU) is a regulator of vitamin K epoxide reductase (VKOR) as identified in studies of warfarin resistance rat and it have been linked to warfarin sensitivity [58]. Shahin et al. was able to confirm the association of CALU rs339097 gene polymorphism with higher doses of warfarin, however, when added to the regression model it failed to show any significance [12]. The authors reported that failing to show significance may have been due to lack of power, since only nine variant carriers were found in their cohort. In African Americans, it was shown, that for every CALU rs339097 variant allele warfarin dose increases by almost 11% [58]. Factor VII is a vitamin K dependent factor, factor VII -401G>T variant was able to predict only 2.2% of warfarin dose variability among Turkish, even though it had an MAF of 35% [44]. Carriers of the minor allele required lower warfarin dose as opposed to the carriers of the major allele [44].
After developing their dosing algorithm, Pathare et al. compared the performance of their algorithm with that of the International Warfarin Pharmacogenetics Consortium (IWPC) in predicting warfarin dose in their Omani cohort [40]. The IWPC algorithm was able to explain only 33.6% of the variability, while Pathare et al. algorithm explained 63% of the variability [40]. Moreover, warfarin dose predicted by the IWPC algorithm was 13% higher than the patients' actual dose [40]. Even though the IWPC algorithm has been developed and validated using data from different racial groups including Asians, Blacks, and Whites across four different countries, and even though the Omani population comes from Asian, African and Caucasian ancestries, the IWPC algorithm did not perform well in the Omani population [40]. Such variation in the algorithm performance could be owed to the fluctuation in racial definitions, clinical variations, social/lifestyle contexts, population migration and historical perspectives [59]. Such observation highlights the need for dosing algorithm that are more populationtailored.
Beside the genetic variants other clinical and demographic variables were shown to be significantly associated with warfarin dose. Age, weight, gender, body mass index and body surface area were the most significant predictors of warfarin dose and added up to the variability explained.

Limitations
Some of the limitations of this review were having only two reviewers carrying the whole process independently which may have led us to overlook some important findings or observations. Another limitation is that we only considered mean warfarin dose as our primary outcome, while other outcomes like Time to Therapeutic Range (TTR) and risk of bleeding might be predicted by genetic factors. Furthermore, SNPs like the GGCX, Factor VII, CALU, APOE, MDR1, EPHX1, and PZ were investigated in limited number of studies, which made it unfeasible for us to compare their effect on warfarin dose between different populations of MENA.