Chapter  146:  Testing for Cytochrome P450 Polymorphisms in Adults With Non-Psychotic Depression Treated With Selective Serotonin Reuptake Inhibitors (SSRIs)

Question 1 (Overarching Question)

No studies were identified that directly addressed any aspect of Question 1.

Question 2 (CYP450 Test Analytic Validity)

We identified 12 published articles and two documents from the FDA website (on performance of the Roche AmpliChip^®) that described methods for genotyping various CYP450 enzymes (nine pertaining to CYP2D6, three to CYP2C19, two to CYP2C8, and one each to CYP2C9 and CYP1A1). Of the studies of CYP450 enzymes most relevant to SSRI metabolism (CYP 2D6, 2C19, and 2C9), only four used the gold standard comparison (DNA sequencing), while others were methods comparisons. Notably, very few of the known polymorphisms of the CYP enzymes were tested. Sensitivity and specificity were high (in the range of 94 to 100 percent) for these studies, but confidence intervals for analytic sensitivity for most genotypes were very wide because of the relatively few samples tested. Gene deletion and duplication studies had lower sensitivity and specificity, further compounded by the limitation that there is no accepted gold standard for such tests.

Question 3a (CYP450 Genotypes and Metabolism of SSRIs)

Sixteen studies met our inclusion criteria, of which five were conducted in healthy adults after a single dose of an SSRI. Of these, three showed that CYP2C19 PMs have significantly higher area under the curve (AUC), longer half-life, and reduced oral clearance of the parent drug, and significantly lower AUC, and lower maximum plasma concentration (C_max) of the metabolite of each drug than EMs (drugs studied were sertraline, fluoxetine, and citalopram). Similar results were found in a study of CYP2D6*10 (associated with PM status) in healthy volunteers after a single dose of paroxetine, while another study of CYP2D6 using multiple doses of paroxetine found no significant difference between PMs and EMs. The remaining 11 studies were in clinical patients in treatment with SSRIs, were heterogeneous, had small sample sizes, and showed mixed results with respect to the association between CYP2D6/CYP2C9/CYP2C19 polymorphisms and SSRI blood levels.

Question 3b (CYP450 Testing and Efficacy of SSRIs)

We identified only five studies, three of which involved cohorts of depressed patients in antidepressant treatment. Of these, one found no differences in the proportion of responders among CYP2D6 EMs, IMs, and PMs treated with fluvoxamine. The second found that although plasma concentrations varied significantly between groups (with respect to 2D6 and 2C9 metabolizer status), levels above or below the lower limit of presumed therapeutic levels did not predict response. The third found no differences in depression scores between two groups, CYP2D6 UMs + EMs versus PMs + IMs, treated with paroxetine. The other two studies found significantly higher proportions of CYP2D6 PMs in non-responders to CYP2D6 metabolized SSRIs compared to the general population. The studies had several limitations including non-randomized designs, inadequate power, studying several SSRIs together as a group, and not accounting for other genetic factors that may influence SSRI efficacy (e.g., genetic variations in serotonin transporter proteins or serotonin receptor proteins).

Question 3c (CYP450 Testing and Adverse Drug Reactions)

We identified nine studies, three of which reported adverse effects in CYP PMs only as a secondary finding. Of the other six, three reported no differences in rates of adverse effects between CYP2D6 PMs and EMs, while a fourth reported no differences in adverse effects between the combined PM + IM and EM + UM groups. One study found a greater prevalence of gastrointestinal adverse effects in PMs compared to EMs. This study also found that the combination of CYP2D6 polymorphism and serotonin receptor 5HT2A polymorphism predicted gastrointestinal adverse effects. Two studies found a significantly higher prevalence of PMs in depressed patients with adverse effects than in the general population. The studies had several limitations including non-randomized design, inadequate power, and not accounting for other genetic factors that may influence SSRI tolerability (e.g., genetic variations in serotonin receptor proteins).

Questions 4 and 5 (Management Decisions, Clinical Outcomes, Decisionmaking, and Harms)

No studies were identified that directly addressed any aspect of these questions.

Model of Treatment for Major Depression

As a complement to the evidence review, we constructed a basic decision model to consider the circumstances under which testing for CYP polymorphisms could improve clinical outcomes, or favorably impact costs. We examined four strategies: (1) use a non-CYP metabolized SSRI without testing; (2) test and choose a non-CYP or CYP metabolized SSRI based on the result; (3) test and choose the dose of a CYP metabolized SSRI based on the result; and (4) use a CYP metabolized SSRI without testing. In no plausible scenario was a testing strategy predicted to improve expected outcomes of treatment at 6 weeks. The efficacy of a test strategy could approach the efficacy of use of a non-CYP metabolized drug, although this required the condition that a high correlation exist between genotype and phenotype (metabolizer status), as well as between phenotype and clinical outcomes. Current evidence does not support the conclusion that such high correlations apply. Moreover, the cost of testing is not offset by treatment savings if treatment duration is less than approximately 9 months.

Therapeutic Drug Monitoring in SSRI Treatment

In general, no clear relationship has been found between blood concentration and clinical response with SSRIs at usual doses, nor has any threshold been identified that defines toxic concentrations. Citalopram showed no significant correlation between steady-state plasma concentration and final Montgomery-Åsberg Depression Rating Scale (MADRS) scores (measure of response) in two studies, with numbers of patients ranging from 13 to 16, and doses ranging from 5 to 60 mg/d.^{14, 15} Paroxetine studies have found no statistically significant differences in plasma levels of paroxetine between responders and non-responders.¹⁶ No correlation has been found between Hamilton Rating Scale for Depression (HAM-D) scores (measure of response) and plasma levels of paroxetine. Studies had numbers of subjects ranging from 16 to 44, and doses from 20 to 60 mg/day.¹⁷–¹⁹ Similarly, studies of fluoxetine with small numbers of patients have suggested either no relationship between plasma concentration of the drug and clinical response,^{20, 21} or have suggested a curvilinear relationship between clinical response and plasma concentrations.²²–²⁴ The limitation of most of these studies is that they may not have been adequately powered. Perhaps the biggest study of plasma concentration and response has been of fluoxetine,²⁵ a multicenter study in which plasma concentrations were available for 615 patients receiving 20 mg/day of fluoxetine. No apparent relationship was observed between plasma concentration and drug response, and plasma concentrations of fluoxetine, norfluoxetine, active moiety, or fluoxetine/norfluoxetine ratio did not differ between responders and non-responders. This is probably the only study with adequate power to be meaningful. However, one limitation of this study was the fact that it was a fixed dose study of fluoxetine at 20 mg/day, raising the possibility that a dose-response relationship could exist at higher doses, or a threshold effect may be possible at lower doses.

Adverse effects of SSRIs, although not generally life-threatening, are typically dose-related. Therapeutic drug monitoring is not routinely recommended for SSRI treatment, but is thought to be of value for ascertaining compliance, for patients who do not respond to multiple SSRIs, or have poor tolerability.²⁶

Sources

The primary source of literature was MEDLINE^® (1966-May 2006). Additional databases searched included the Cochrane Database of Abstracts of Reviews of Effects (DARE), PsychInfo, HealthSTAR, and CINAHL. Searches of these databases were supplemented by reviews of the reference lists contained in all included articles and in relevant review articles. We also included data from the U.S. Food and Drug Administration (FDA) website describing the operating characteristics of the Roche AmpliChip^® CYP450 Test.^{40, 41} On the advice of our technical expert panel, we did not undertake a comprehensive search of the grey literature.

Search Strategies

The basic search strategy used the National Library of Medicine's Medical Subject Headings (MeSH) key word nomenclature developed for MEDLINE.^® Searches were limited to articles published in English. The exact search string used is given in Appendix A.^* The searches yielded a total of 1,200 citations, whose records are maintained in a ProCite (Thompson ISI ResearchSoft, Berkeley, CA) database.

Abstract and Full-Text Screening

Paired researchers from the Duke research team independently reviewed all abstracts and classified each as “included” or “excluded” according to project-specific criteria, which they developed. The exclusion criteria were:

• Single case.

• SSRI inhibition of CYP enzymes (unless the study examines how this is related to genotype).

• Outside the scope of the report.

An abstract was included for further review if at least one of the paired reviewers recommended that it be included. A total of 140 abstracts were included for review at the full-text stage. Inter-rater reliability for include/exclude decisions at the abstract stage was tested by having five pairs of readers review 862 abstracts. Agreement (kappa statistic) ranged from -0.037 to 0.613.⁴⁹

At the full-text review stage, paired researchers independently reviewed the articles and indicated a decision to “include” or “exclude” the article for data abstraction. When two reviewers returned different decisions about whether to include or exclude an article, they were asked to reconcile the difference. Detailed full-text exclusion criteria are listed immediately below.

Full-Text Screening Criteria

Studies were excluded at the full-text screening stage if any of the following applied:

• Single case.

• Patient age < 18 years.

• No gold standard comparison or methods comparison (for articles on analytic validity).

• Study falls outside study scope (e.g., there were several good reviews, including one that made pharmacogenetics-based therapeutic recommendations,⁵⁰ that did not answer any of the key questions directly).

At the full-text stage, studies were further identified as addressing one or more of the following criteria:

• A

  Clinical tests for polymorphisms. These include studies of commercial (e.g., AmpliChip^®) and other tests that may be used for determining genetic polymorphisms in a clinical setting.

• B

  Gold standard. DNA sequencing is the accepted gold standard for genotyping. Because very few studies used a gold standard comparison, a decision was made also to include studies that used methods comparisons (e.g., polymerase chain reaction and restriction fragment length polymorphism [PCR-RFLP]). In keeping with the clinical diagnostic test literature, these methods are referred to here as a reference standard, acknowledging that they provide a lower level of evidence than gold standard comparisons.

• C

  Predicted metabolism of SSRIs. This includes metabolizer status of an individual with respect to a particular SSRI, e.g., “poor metabolizer” (PM) or “ultra-rapid metabolizer” (UM), and is distinct from PM or UM of a probe drug for a given CYP enzyme. Because an SSRI may not be exclusively metabolized by a certain CYP enzyme, its metabolism may vary from that of the probe drug for that enzyme in a person carrying a function-altering mutation of that CYP enzyme.

• D

  Decisionmaking. This includes decisionmaking by patients and providers; medical, personal, and public health decisionmaking.

• E

  Health outcomes of interest. Heath outcomes included: drug efficacy, adverse drug reactions, and other outcomes such as improved prognosis and quality of life.

• F

  Harms. Harms associated with testing or with subsequent management decisions.

Studies were then classified as addressing one or more of the key questions. For example:

Question 2 (analytic validity): A + B

Question 3a (metabolism of SSRIs): (A or B) + C

Question 4b (improved outcomes versus not testing): (A or B) + E

Please note that although (A or B) + E would apply to all health outcomes questions, we did not expect to find many studies addressing these, and therefore we did not break down E further.

Summaries of the results of the abstract screening and full-text review are provided in Tables 4 and 5. A list of excluded articles, with reasons for exclusion, is provided in Appendix B ^*

Approach

For purposes of this report, we adopted the definition of analytic validity and its components from the Analytic validity, Clinical validity, Clinical utility and associated Ethical, legal and social implications (ACCE) model (Appendix E ^*), which reads:

It is notable that the definitions of sensitivity and specificity above are most directly applicable to tests with dichotomous results (mutation present or absent). Because there are multiple CYP450 polymorphisms that can be assessed, and each study may provide information on only a subset of polymorphisms, we defined analytic sensitivity operationally as the proportion of known genotype challenge samples that are correctly identified by the test under evaluation. Similarly, analytic specificity was defined operationally as the proportion of known wild-type challenge samples that are correctly identified by the test under evaluation.

Our assessment of analytic validity focuses on tests that are actually used, or are likely to be used, in clinical settings. The gold standard method for CYP450 genotyping is unequivocally the bidirectional sequencing of the specific genetic region of the gene of interest. However, many reference methods exist due to the complexity and high costs involved with sequencing of large populations. To date, there is only one technology approved by the U.S. Food and Drug Administration (FDA) specifically for CYP450 genotype testing (the Roche AmpliChip^®), and one technology approved for genetic testing of a different gene target (Invader Assay for UGT1A1 genotyping) which has been employed in one of the studies for CYP2D6 genotyping.⁵³ Other laboratories currently performing CYP450 tests in clinical settings generally employ traditional methods, including polymerase chain reaction and restriction fragment length polymorphism (PCR-RFLP) or allele-specific polymerase chain reaction (AS-PCR, also referred to as allele-specific amplification, or ASA).

In the absence of a substantial number of studies comparing the test under evaluation to the gold standard (bidirectional DNA sequencing), we decided to include studies that used a traditionally accepted methods comparison, typically PCR-RFLP or AS-PCR, acknowledging that a methods comparison would be a lower level of evidence regarding analytic sensitivity and specificity than a gold standard comparison. Consequently, we refer to the comparator tests as a “reference standard.” It should be noted that in most cases even DNA sequencing for the purpose of assay validation may not have been done bidirectionally (not reported), but is referred to as a gold standard nonetheless.

Few studies reported the ethnic makeup of the tested sample populations, and even when details were provided there was no standard format followed, or description provided of the source of ethnicity data (e.g., based on self-reported or medical or other documentation). We therefore summarize all studies by the common denominator of general ethnic group (e.g., Caucasian).

Some studies provided information about test performance in assessing individual alleles rather than genotypes. Although these are less clinically relevant, they are included to complement the information about genotypes.

Several studies addressed the issue of gene duplication and deletion. While these are clinically important, there is a lack of an accepted gold standard; in this case, comparisons were based on any reported comparator. Current methods commonly used for assessment of CYP2D6 gene copy number are based on two different approaches, both of which are sensitive to the location of primers used in assay design and are prone to produce erroneous results in rare cases of rearrangement variants or occurrence of mutations in the positions targeted by these primers. The first approach compares CYP2D6 copy number to a gene that is known to have no variation in gene copy number, computing the ratio between them. Most of the analytic validity studies employing these methods do not discriminate further to see which allele is duplicated (hence these are referred to in Table 7, below, as “duplication” and “deletion”). Alternatively, others amplify a duplication-specific fragment, but may miss duplications, depending on the particular primers utilized. These traditionally used fragments also allow limited or no genotyping, since they carry little or no coding regions (intergenic region amplified).

On the advice of the technical expert panel, we reviewed studies for polymorphisms in all CYP450 enzymes. Results for those most relevant to SSRI metabolism (CYP2D6, 2C19, and 2C9) are presented in this section; results for the remaining enzymes are provided in Evidence Table 1 (Appendix D ^*).

Results

Results for analytic sensitivity and specificity for genotype polymorphisms are presented below by CYP450 enzyme. All calculations were performed using FastPro version 1.8 (Academic Press, 1992). Summaries of evidence regarding sensitivity and specificity for individual alleles, gene deletion and duplication, laboratory quality control, and robustness issues are also provided. Detailed results relating to these latter issues are included in Evidence Table 1 (Appendix D ^*).

CYP2D6. We identified nine reports that compared clinical methods for genotyping CYP2D6 enzyme polymorphisms to a reference standard (Tables 6 and 7). Of these, eight were published studies⁵³–⁶⁰ and one was reported on the FDA website.⁴⁰ Only two studies^{40, 55} provided a comparison to the gold standard, DNA sequencing. One study⁶⁰ provided results in allele counts only and thus is not considered further here (for details see Evidence Table 1, Appendix D ^*). Five of the studies tested different 2D6 variants (Table 6), while five reported results on gene copy number (Table 7); two studies^{40, 59} reported both.

In all studies analytic sensitivity and specificity for each tested genotype ranged from 94.12 to 100 percent, with the exception of Schaeffeler et al.,⁵⁷ which reported sensitivity of 91.67 percent to detect the duplication/(single copy) genotype and specificity of 99.79 percent. However, only 26 of approximately 100 known CYP2D6 polymorphisms (www.cypalleles.ki.se) were evaluated in the included studies, with most studies focusing on only a handful of these variants.

Analytic sensitivity and specificity based on combined allele counts (as opposed to genotype calls presented in Table 6) ranged from 98 to 100 percent in all studies, again with the exception of Schaeffeler et al.,⁵⁷ which reported sensitivity of 73.1 percent and specificity of 98 percent. However, correct allele counts do not necessarily reflect correct genotype calls (which are assumed to predict treatment outcomes) and are therefore less relevant in the clinical context (details on allele counts, sensitivity, and specificity are presented in Evidence Table 1, Appendix D ^*).

CYP2D6 gene copy number methods exhibit relatively high sensitivity and specificity, although two of the four studies reporting results on duplication variants reported failures, resulting in sensitivity of 91.67 percent to identify duplication/(single copy) genotypes, and a homogeneity p-value of 0.06. It should be noted that traditional assays designed to identify the deletion variant *5 fail to depict some rearrangement-deletion alleles (the most common of which in Caucasians are *13 and *16 [0.5 to 1 percent], not tested in any of the studies above). These are non-functional alleles and result in the same metabolic phenotype as *5 (i.e., poor metabolizer [PM]).

Quality control procedures include the integration of negative and positive controls into the genotyping process in most studies^{40, 53, 57}–⁵⁹ to ensure that results fall within the specified assay limits. Robustness was not measured and reported in all studies. The effect of pre-analytic variables (such as blood sample collection and DNA extraction method) was reported only by Chou et al.,⁶⁰ but they did not perform a comprehensive comparison between the two DNA extraction methods employed. The effect of analytic variables was measured in three of the studies, each reporting a different measurement: Schaeffeler et al.⁵⁷ reported inter- and intra-assay variability in a subset of the tested samples separately; Soderback et al.⁵⁸ reported the reproducibility of linear regression coefficients; and Roche investigators⁴⁰ reported reproducibility scores by combining the effects of sites, replicates, dates, and reagent lots. All three studies reported high rates of robustness, but these were not comparable and could not be summarized together (details are presented in Evidence Table 1, Appendix D ^*).

CYP2C19. We identified three reports that compared clinical methods for genotyping CYP2C19 enzyme polymorphisms to a reference standard (Table 8). Of these, two were published studies^{54, 61} and one was reported on the FDA website.⁴¹ Only one study⁴¹ provided a comparison to the gold standard, DNA sequencing.

All three studies reported a high sensitivity and specificity (96.43 to 100 percent). However, each study focused on detection of two out of the three common CYP2C19 alleles (*2, *3, and *4). Similar results were obtained when performing calculations based on allele calls (Evidence Table 1, Appendix D ^*). Quality control procedures employed by the above studies included the incorporation of polymorphism surrounding sequence in assay design as an internal control,⁵⁴ and testing positive and negative control samples routinely.⁴¹ Only Roche Molecular Systems, Inc.,⁴¹ reported the robustness of their technology for CYP2C19 allele detection by means of reproducibility of correct genotype calls (99.6 percent; for details see Evidence Table 1, Appendix D ^*).

CYP2C9. We identified one report that compared clinical methods for genotyping CYP2C9 enzyme polymorphisms to a reference standard.⁵⁴ This study did not use the gold standard, DNA sequencing. Investigators reported lack of detection of homozygotes for the *2 and/or the *3 alleles. They also stated that compound heterozygotes (*2/*3 genotype) were identified, but they provided no genotype counts, preventing calculation of genotype analytic sensitivity and specificity.

Calculation of analytic sensitivity based on allele counts was 100 percent. Due to an unreported number of compound heterozygotes, it is impossible to calculate confidence intervals of assay specificity, but it is implied that mean specificity is 100 percent.

No measures of robustness were reported. Quality control featured interrogation of the surrounding sequence, along with the variable positions tested and providing internal controls.

Other CYPs. We identified two studies that compared clinical methods for genotyping CYP2C8 enzyme polymorphisms to a reference standard (Table 9), and one that did the same for CYP1A1 polymorphisms (Table 10). Only one study (of CYP2C8) provided a comparison to the gold standard, DNA sequencing.⁶²

All studies reporting assay performance for the detection of CYP2C8 and CYP1A1 exhibit 100 percent analytic sensitivity and specificity. Calculations based on allele calls reflect the same findings. Quality control procedures employed include the incorporation of positive and negative controls into the genotyping process.⁶² Robustness was assessed only by Wu et al.,⁶⁴ by means of inter- and intra-assay variability. The intra-assay coefficients of variance were reported to be lower than 11.2 percent for both CYP1A1 assays, and the inter-assay coefficients of variance were lower than 14.3 percent. Weise et al.⁶³ implied 100 percent inter-assay reproducibility of results obtained by four different investigators.

Discussion

Based on emerging standards, analytic validity of genetic tests includes not only the ability of the test to accurately identify challenge genotypes (as assessed by a gold standard test), but also quality control and robustness. We identified only a few studies of test performance relative to the gold standard of DNA sequencing (bidirectionally or unidirectionally), applied to a limited number of samples (as reflected by the wide confidence intervals calculated for analytic sensitivity and specificity), and covering but a small set of possible genetic variants. Many studies appear to be in the realm of preclinical evaluations and are not clearly relevant to the domain of clinical practice.

These data do suggest that the analytic sensitivity and specificity of available tests are generally high. One concern may be that in the evaluation of gene deletions and duplications, assessing the magnitude of the potential problem is limited by the lack of an established gold standard for gene copy number. Another concern is that few CYP450 variants are included in the studies we identified, which focused particularly on the more common variants in Caucasians and African-Americans. However, variants that are rare in these populations may be more frequent, and thus more clinically relevant, in other populations. In the same context, it should be noted that most studies focus on developing reliable methods for the genotyping of CYP2D6 variants known to be non-functional (PM). Of these, the most common in Caucasians and African-Americans are *3, *4, *5, and *6, and the majority of studies target their assays at capturing these variants. Even the AmpliChip^®, which targets the largest set of CYP2D6 variants (n = 26), fails to capture a large set of rare variants leading to deficient enzyme activity.

Many analytic validity studies, particularly those published less recently, tend to report results by stating allele counts and frequencies, without direct comparison to a reference standard. While this format allows comparison to published allele frequencies in populations similar to the ones employed in their studies, it is not helpful in estimating genotype-specific performance. For completeness, however, data summarizing allele frequencies are provided in Evidence Table 1 (Appendix D ^*).

Conclusions

Although these results suggest that analytic validity for detecting some of the CYP450 genotypes more frequently encountered in the Caucasian population is good, overall the data are limited, with relatively small numbers of samples and a relatively narrow range of polymorphisms tested. In addition to studies addressing these limitations, research should include closer examination of the issue of deletions and duplications. Furthermore, practical concerns of quality control and robustness deserve greater investigation based on emerging standards for such studies.

Approach

There is definitive literature supporting the association between certain CYP450 genotypes and their predicted phenotypes (i.e., how they would metabolize probe drugs specific for that CYP enzyme). Our question sought to address how well a certain genotype (or its corresponding predicted phenotype) predict metabolism of particular SSRIs. For example, does a CYP2D6 *5*5 or predicted poor metabolizer (PM) of the probe drug dextromethorphan also metabolize fluoxetine poorly? To address this question, we sought to identify all studies in which patients on SSRIs were tested for CYP450 genetic polymorphisms. Studies were included irrespective of the method used for genotyping. Because of the overall paucity of data, we included studies that had diagnoses other than non-psychotic depression as an indication for SSRI treatment, as clinical outcomes in such scenarios may be indicative of genotype effects. We also included studies in which only a subgroup of patients was treated with SSRIs, while others were treated with other antidepressants, including tricyclics.

Results

Note: Here, as throughout Chapter 3, the terms “poor metabolizer (PM),”“extensive metabolizer (EM),” etc., refer to general phenotypes (for a probe drug) as predicted by genotyping.

We identified 16 studies that met the inclusion criteria for this question (see Evidence Table 2, Appendix D ^*). Five studies examined SSRI metabolism in healthy adults,^{43, 65}–⁶⁸ while the other 11 looked at SSRI metabolism in patients who had achieved a steady state after multiple doses.⁶⁹–⁷⁹

The first group of five studies examined the effect of different genotypes of CYP2C19 and CYP2D6 on SSRI metabolism in healthy volunteers. Results are summarized in Table 11.

All these studies used standard measures such as area under the curve (AUC, which is an assessment of bioavailability of the drug), half-life (time taken to eliminate half the total ingested quantity of the drug from the body), and oral clearance (pertains to distribution and elimination of drug) as measures of rate of metabolism. Three of the five studies^{43, 65, 67} included young, healthy, male, non-smoking, Chinese subjects who were free of medications and alcohol for at least 2 weeks prior to the study. These studies looked at the effect of CYP2C19 genotypes and predicted phenotypes (EM vs. PM) on the metabolism of three different SSRIs, namely, fluoxetine, sertraline, and citalopram. All three studies found significantly higher AUC, longer half-life, and reduced oral clearance of the parent drug, and significantly lower AUC and lower maximum plasma concentration (C_max) of the metabolite of each drug, in PMs as compared to EMs. The fluoxetine and citalopram studies also found a gene dose effect such that heterozygous EMs showed values between homozygous EMs and PMs.

Of the remaining two studies, the first⁶⁸ was carried out in 16 healthy, young, Korean subjects and examined the effect of the CYP2D6*10 allele (predictive of poor metabolism) on paroxetine metabolism. Investigators found that homozygotes and heterozygotes for *10 alleles showed significantly greater volume of distribution, greater AUC, and lower oral clearance of paroxetine than wild type homozygotes.

Thus, all studies in healthy adults using a single dose of an SSRI found that PMs, as predicted by genotyping, metabolized the SSRI more slowly than EMs, irrespective of particular SSRI.

The other study⁶⁶ was a multiple-dose study that looked at paroxetine pharmacokinetics in 17 healthy, young, non-smoking Caucasian subjects who received paroxetine 20 mg/day for at least 5 days (range, 5 to 15 days). It found that heterozygous EMs had twofold higher median paroxetine steady-state concentrations than homozygous wild type EMs (n = 10); the difference was not statistically significant (p = 0.2).

The 11 studies conducted in clinical populations (with sample sizes ranging from 11 to 146) examined the effects of genotypes of 2D6, 2C9, and 2C19 on the pharmacokinetics of various SSRIs. Results are summarized by SSRI in Table 12 for all studies of individual SSRIs.

Table 13 reports confidence intervals for differences in mean SSRI levels between extensive metabolizers and comparison groups for those studies reporting the necessary data.

All 11 studies examined some or all genotypes of CYP2D6, while CYP2C9 and 2C19 were considered in three studies each. The studies were heterogeneous with respect to patient population, some with psychiatric outpatients, some with major depression defined either by Diagnostic and Statistical Manual for Mental Disorders, 4^th edition (DSM-IV) criteria or using other definitions; one with patients in treatment with an antipsychotic (risperidone) to which fluoxetine was added;⁷¹ one with breast-feeding mothers on antidepressants and their infants;⁶⁹ and one with patients who developed hyponatremia while on antidepressants.⁷⁷ All studies were cross-sectional in design. The studies were carried out in different ethnic groups. One study⁷² not included in Table 12 or 13 studied several antidepressants, including SSRIs and tricyclics, and the results could not be broken down by particular antidepressant type for different CYP polymorphisms.

The studies for individual SSRIs are summarized in Table 12. Paroxetine was the most studied SSRI, with six studies examining the effect of CYP2D6 polymorphism on paroxetine steady state plasma concentration. Fluoxetine was studied in four studies with respect to CYP2D6, 2C9, and 2C19 polymorphisms. Fluvoxamine and CYP2D6 polymorphism were studied in one study, and citalopram was studied with respect to CYP2C19 polymorphism in one study.

Data for calculating confidence intervals for differences in mean SSRI levels between EM and comparator groups (PM, heterozygous EM, etc.) were available for four studies of paroxetine, four of fluoxetine, and one of fluvoxamine (Table 13). The results for both paroxetine and fluoxetine were mixed, with some studies showing significant differences, and others not showing differences. The studies typically had small numbers of subjects, and hence the confidence intervals for differences in means were very wide, as shown in Table 13. For fluoxetine, two studies^{73, 78} showed a significant difference in active moiety between CYP2C9 EMs and PMs, but failed to show differences in active moiety levels between EMs and PMs of CYP2D6 (data for Scordo et al., 2005⁷⁸ are not shown in Table 13, as results were reported as median and range rather than means ± standard deviation).

Discussion

The 16 included studies provide mixed evidence regarding the first part of Question 3a (on possible correspondences between CYP450 genotypes and metabolism of particular SSRIs). In multiple dose studies of SSRIs, inconsistent results were obtained for individual SSRIs, and also for individual CYP enzymes.

Methodological issues in studies addressing this question include:

• (1)

  Single-dose studies in healthy volunteers:^{43, 65, 67, 68} Clinical situations may be very different from single-dose studies, because of the possible effects of the medication on CYP enzymes over time. Data from single-dose experiments cannot be extrapolated to long-term drug therapy, as saturation pharmacokinetics, irreversible enzyme blockade, or enzyme up- or down-regulation might change the outcome with multiple dosing.^{3, 81, 82}

• (2)

  Small sample sizes: All the studies had very small samples of the PM or ultra-rapid metabolizer (UM) groups, and thus may not have been powered adequately to detect significant differences, as shown by wide confidence intervals.

• (3)

  Heterogeneity: The studies were quite variable in terms of the population of interest, specific SSRIs considered, and specific CYP450 polymorphisms.

• (4)

  Not accounting for multiple CYP enzymes that may be involved in metabolism of a certain SSRI: Only one study⁷³ took into account the possibility that more than one CYP enzyme might be involved in the metabolism of a certain SSRI and therefore controlled for polymorphisms in another enzyme.

• (5)

  Not accounting for active metabolites of certain SSRIs like fluoxetine: Two studies measured active moiety rather than parent drug alone,^{73, 78} whereas two others did not.^{70, 71}

• (6)

  Most studies for accounted for co-medications that may be inhibitors or substrates for the enzyme being studied; one did not.⁶⁹ Benzodiazepines were typically allowed in these psychiatric cohorts, as these drugs are metabolized mainly by CYP3A4 and have no influence on the enzymes studied.

• (7)

  Diet was not taken into account in any study.

• (8)

  One study⁷² combined SSRIs and other antidepressants and examined effects of polymorphisms of various CYP enzymes. Combining various SSRIs, and moreover SSRIs with other antidepressant medications, may have confounded results because of variability in the contribution of different CYP enzymes to metabolism of different SSRIs and other antidepressants, and variability in CYP inhibition by different SSRIs.

The quality assessment criteria we applied to individual studies in this report⁵² (Appendix E ^*) yielded a range of scores between “3b” and “4.” For the suggestion that the genotypes affect metabolism of SSRIs, the grade of recommendation based on available data would be “C.”

Conclusions

In depressed patients treated with SSRIs, the existing data (a series of heterogeneous studies in small samples) do not support a clear correlation between CYP metabolizer status as predicted by genotyping and SSRI concentrations.

Approach

To address this question, we sought to identify all studies in which patients treated with SSRIs were tested for CYP450 genetic polymorphisms. Studies were included irrespective of the method used for genotyping. Because of the overall paucity of data, we included studies that had diagnoses other than non-psychotic depression as an indication for SSRI treatment, as clinical outcomes in such scenarios may be indicative of genotype effects. We also included studies in which only a subgroup of patients was treated with SSRIs, while others were treated with other antidepressants, including tricyclics.

Results

Note: Here, as throughout Chapter 3, the terms “poor metabolizer (PM),”“extensive metabolizer (EM),” etc., refer to general phenotypes (for a probe drug) as predicted by genotyping.

We identified only five studies that examined the association between CYP450 genotypes and SSRI efficacy (see Evidence Table 3, Appendix D ^*). All studies were cross-sectional in design. All five studied CYP2D6 polymorphisms. One additionally studied 2C9 and 2C19 polymorphisms. Three studies examined depressed patients on antidepressant treatment (Table 14). Gerstenberg et al.⁸³ found no differences in the proportion of responders among CYP2D6 EMs, intermediate metabolizers (IMs), and PMs treated with fluvoxamine. There were also no differences between the three groups on final Montgomery-Åsberg Depression Rating Scale (MADRS) scores, percent improvement, or amelioration score. Fluvoxamine steady-state concentrations were found to be significantly higher in responders than in non-responders, but were not reported by genotype. Grasmader et al.⁷² found that although plasma concentrations varied significantly between groups (with respect to 2D6 and 2C9 metabolizer status), levels above or below the lower limit of presumed therapeutic levels did not predict response. Murphy et al.⁷⁴ found no differences in depression scores between two groups, CYP2D6 UMs + EMs versus PMs + IMs, treated with paroxetine. The groups were combined because of low numbers of UM and PM phenotypes.

The remaining two studies examined the prevalence of different CYP2D6 genotypes in non-responders to adequate antidepressant treatment, including SSRIs (Table 15). Rau et al.⁸⁴ found a three-fold increase in the frequency of CYP2D6 UMs in a group of 16 German depressed patients non-responsive to antidepressants (only five treated with SSRIs) in comparison to the general population. Kawanishi et al.⁸⁵ found a significantly greater prevalence of 2D6 UM phenotypes in non-responders (subgroup of 81 Nordic Caucasian patients treated with 2D6 metabolized drugs) compared to the general population. “Worst week” Hamilton Rating Scale for Depression (HAM-D) scores were found to be higher in UMs than in EMs.

Discussion

Based on the available evidence, a definitive association of CYP450 (2D6, 2C9, 2C19) genotypes and efficacy of SSRIs cannot be inferred.

Methodological issues in studies addressing this question include:

• (1)

  Study-design and power: None of the studies was a prospective randomized trial. Three^{72, 74, 83} were observational or correlational studies, and two^{84, 85} were pilot studies of prevalence of CYP polymorphisms in non-responders to antidepressant treatment. All the studies had very small numbers of patients in the UM groups.

• (2)

  Only two studies^{74, 83} studied individual SSRIs (fluvoxamine and paroxetine respectively), while the others grouped the SSRIs together or with groups of other antidepressants. Combining various SSRIs, and moreover SSRIs with other antidepressant medications, may have confounded results because of variability in contribution of different CYP enzymes to metabolism of different SSRIs and other antidepressants, and variability in CYP inhibition by different SSRIs.

• (3)

  The two prevalence studies considered^{84, 85} have the obvious shortcoming of comparing CYP2D6 UM prevalence in depressed non-responder patients to the UM prevalence in the general population. It is possible that CYP2D6 UM phenotype itself is associated with presence of severe depression that is treatment-resistant, which may have accounted for high prevalence of this phenotype in non-responders to antidepressant treatment. It would be more meaningful to compare prevalence rates between responders and non-responders to a given SSRI, which would require a very large sample. In addition, neither of these studies specified exclusion criteria.

• (4)

  The data considered do not lead to any conclusions about the possible impact of race/ethnicity, diet, or other medications on the association between CYP450 genotypes and SSRI efficacy.

• (5)

  Genetic factors affecting serotonin receptor proteins, membrane transporters, and signal transduction molecules could also have important pharmacodynamic effects that could affect SSRI efficacy.⁸⁶ Thus, examining the impact of pharmacokinetic variability resulting from CYP enzyme polymorphisms on SSRI efficacy in isolation may not be optimal.

The quality assessment criteria we applied to individual studies in this report⁵² (Appendix E ^*) yielded a range of scores between “3b” and “4.” For the suggestion that CYP450 genotypes do not affect SSRI efficacy, the grade of recommendation based on available data would be “C.”

Conclusions

Because of the poor quality of relevant data that could be identified to address the question, no firm conclusions can be drawn about the relationship between CYP450 genotypes and efficacy of SSRI treatment in patients with non-psychotic depression.

Approach

To address this question, we sought to identify all studies in which patients treated with SSRIs were tested for CYP450 genetic polymorphisms. Studies were included irrespective of the method used for genotyping. Because of the overall paucity of data, we included studies that had diagnoses other than non-psychotic depression as an indication for SSRI treatment, as clinical outcomes in such scenarios may be indicative of genotype effects. We also included studies in which only a subgroup of patients was treated with SSRIs, while others were treated with other antidepressants, including tricyclics. Studies that specifically examined adverse effects were particularly sought.

Results

Note: Here, as throughout Chapter 3, the terms “poor metabolizer (PM),”“extensive metabolizer (EM),” etc., refer to general phenotypes (for a probe drug) as predicted by genotyping.

We identified nine studies that met our inclusion criteria (see Evidence Table 4, Appendix D ^*). Of these, three reported the incidence of adverse effects in PMs, but this outcome was not central to the aims of the study.^{43, 72, 87} These studies do not add any information of value to the discussion and are not considered further here. The remaining six studies are summarized in Table 16.

All six studies examined CYP2D6 polymorphisms only. Three of the six studies reported no differences in rates of adverse effects between PMs and EMs,^{83, 84, 89} while a fourth⁷⁴ reported no differences in adverse effects between the combined PM + IM and EM + UM groups. One study found a greater prevalence of gastrointestinal (GI) adverse effects in PMs compared to EMs.⁹⁰ This study also found that the combination of CYP2D6 polymorphism and serotonin receptor 5HT2A polymorphism predicted GI adverse effects, such that PM + GG and PM +AG had a significantly greater risk of developing GI side effects compared to EM + AA.

Two studies^{84, 88} found a significantly higher prevalence of PMs in depressed patients with adverse effects than in the general population. One of these⁸⁸ also found the PM phenotype to be more frequent in depressed patients with adverse effects than in a random group of depressed patients. Studies that reported types of adverse effects reported a range of typical SSRI adverse effects including but not limited to anxiety, agitation, restlessness, nausea, GI upset, headache, sleep disturbance, and sexual dysfunction.^{83, 84, 89} The most common adverse effect reported in studies was nausea.

Discussion

Although four studies did not find any differences in adverse effects in PMs versus EMs, these studies are heterogeneous, with major methodological problems, including:

• (1)

  Study design and power: None of the studies was a prospective randomized trial. Four^{74, 83, 89, 90} were observational or correlational studies, and two^{84, 88} were pilot studies of the prevalence of CYP polymorphisms in patients who had adverse effects with antidepressant treatment. All the studies had very small numbers of patients in the PM groups.

• (2)

  Three studies examined individual SSRIs^{74, 83, 89} (paroxetine, fluvoxamine, and fluoxetine, respectively), whereas the other two grouped the SSRIs together or with groups of other antidepressants. Combining various SSRIs, and moreover SSRIs with other antidepressant medications, may have confounded results because of variability in contribution of different CYP enzymes to metabolism of different SSRIs and other antidepressants, and variability in CYP inhibition by different SSRIs.

• (3)

  The two prevalence studies considered^{84, 88} did not specify exclusion criteria. Moreover, comparing a group of patients with adverse effects to a particular SSRI to a group of patients with no adverse effects to that SSRI may have been more meaningful, but will require a large number of patients.

• (4)

  The data considered do not lead to any conclusions about the possible impact of race/ethnicity, diet, or other medications, on the association between CYP450 genotypes and adverse effects to SSRIs.

• (5)

  Genetic factors affecting serotonin receptor proteins, membrane transporters, and signal transduction molecules could also have important pharmacodynamic effects that could affect SSRI tolerability.⁸⁶ Thus, examining impact of pharmacokinetic variability resulting from CYP enzyme polymorphisms on SSRI tolerability in isolation may not be optimal. Only one study⁹⁰ addressed this issue and did in fact show combined effects of CYP2D6 and 5HT2A polymorphisms on GI adverse effects, further supporting this point.

The quality assessment criteria we applied to individual studies in this report⁵² (Appendix E ^*) yielded a range of scores between “2b” and “4.” For the suggestion that CYP450 genotypes do not affect SSRI tolerability, the grade of recommendation based on available data would be “C.”

Conclusions

Because of the poor quality of relevant data that could be identified to address the question, no firm conclusions can be drawn about the relationship between CYP450 genotypes and tolerability of SSRI treatment in patients with non-psychotic depression.

Background

In deciding whether to use CYP450 genotype testing to guide depression therapy, it would be ideal to have direct scientific studies demonstrating that use of genotype testing leads to improved clinical outcomes. In the absence of such direct evidence, decision modeling can be used to provide indirect evidence based, for example, on the relationship between genotype and specific serotonin selective reuptake inhibitor (SSRI) metabolism (phenotype), and the relationship between phenotype and responsiveness to therapy. Examining these clinical relationships is of paramount importance. Genetic testing, like all forms of diagnostic tests, should only be promoted if the potential benefits (such as improved response to treatment for depression) outweigh the potential harms (such as increased adverse effects).

Decision analysis is a tool that provides a mechanism for inferring the likely outcomes of competing options by modeling the relationship between each option and the outcome of interest. Such decision models provide a framework for linking information from multiple sources (e.g., epidemiological studies, test performance studies, treatment efficacy studies, and surveys of patient preferences and quality of life). In addition to providing a “best guess” about the impact of a particular decision, decision models can offer insight into the dynamic relationship between various clinical inputs and decision relevant outcomes - under what circumstances is one decision preferred over others? This use of a decision model is especially valuable when the input data are not particularly strong, as is the case here.

We constructed and evaluated a decision model to address the question: Under what circumstances would genetic testing for CYP isoenzymes during the initial evaluation of an individual with non-psychotic major depression lead to a better clinical outcome, when compared to empiric SSRI therapy?

Methods

Population. The population of interest for the model was treatment-naïve adults who met the DSM-IV criteria for major depression. They were otherwise generally healthy and not taking medications that could interact with SSRIs.

Model structure. The model is a simple tree structure (Figure 2) with four options: (1) do not test and treat empirically with an SSRI not affected by genotype; (2) test and use the results to select an SSRI that is or is not affected by genotype; (3) test and use the results to select the dose of an SSRI affected by genotype; or (3) do not test and treat empirically with an SSRI affected by genotype. In this analysis, we were not focusing on any particular SSRI; however, the issue is illustrated by the examples of fluoxetine and sertraline. Fluoxetine is a generic and inexpensive medication that is primarily metabolized by the 2D6 enzyme, and sertraline is a more expensive medication that is not.⁸⁷ Although the overall response rate for standard dosing of fluoxetine (40 mg) is similar to that for sertraline, the adverse event rate is likely to be higher in those with poor metabolism (due to higher than normal serum concentrations), and the response rate lower in those with ultra-rapid metabolism (due to lower than normal serum concentrations). In this setting, empiric therapy with the non-CYP metabolized medication will always be the most effective strategy since it eliminates the potential for complications related to enzyme metabolism; however, the higher cost limits its widespread availability.

For each strategy, an individual could have one of three phenotypes: ultra-rapid metabolizer, extensive metabolizer/intermediate metabolizer, or poor metabolizer, with a probability based on the distribution of phenotypes in the population. We combined the extensive and intermediate metabolizers into a single phenotype to simplify the model since there was little data to support a difference in response to therapy for these two groups. For the first model strategy (use of a non-CYP metabolized SSRI without testing), the likelihood of treatment success is assumed to be the same for all phenotypes. For the second option (use of genetic testing to select SSRI), patients with genotypes that correspond to phenotypes with a high probability of treatment failure (ultra-rapid and poor metabolizers) would receive the more expensive non-CYP metabolized medication, while those not at high risk (extensive and intermediate metabolizers) would receive the less expensive CYP metabolized one. For the third option (use of genetic testing to select dose of SSRI), results of the genetic test are used to adjust the dose of the CYP metabolized medication. A low dose would be used for poor metabolizers, a standard dose for extensive and intermediate metabolizers, and a high dose for ultra-rapid metabolizers. For the fourth option (use of a CYP metabolized SSRI without testing), the likelihood of treatment success depends upon phenotype.

The model was created as a decision tree using TreeAge ProSuite 2006 (TreeAge Software Inc, Williamstown, MA)

Model parameters. The estimate and source for each model parameter are shown in Tables 15 and 16. For the purposes of this model our estimates for efficacy of therapy were based upon fluoxetine for the CYP metabolized SSRI and sertraline for the non-CYP metabolized SSRI.

In clinical decisionmaking, a key question is the probability that any particular genotype will correspond to a particular level of drug metabolism (phenotype). This question is paramount since the phenotype is purported to effect the likelihood of treatment success, both effectiveness and adverse effects. However, the available literature presents limited data on these essential probability estimates. In the absence of data, we used the technique of bootstrapping to backwards calculate probabilities which were consistent with the two correlation coefficients (0.2 and 0.8). Specifically, we created a series of tables (genotype x phenotype) in which synthetic patient samples were assigned to cells with the target correlation coefficient; the cells were divided by the row totals, and the resulting elements were the estimated probabilities that a specific genotype would be associated with a specific phenotype. We repeated this exercise for both levels of correlation on each genotype to phenotype pair. For example, when the correlation between genotype and phenotype is 0.8 (high), the estimated probability that an ultra-rapid phenotype will have an ultra-rapid genotype is 0.5; if the correlation is 0.2 (low), the estimated probability is only 0.38.

The clinical predictive value of phenotype is reflected in the model as the probability that an individual with a specific phenotype will respond to a specific SSRI. These estimates were based upon expert opinion; however, their clinical plausibility was verified by comparing calculated overall population response rates (using the estimates and known prevalence rates) to published response rates.

As a practical strategy for examining the impact of variations in the relationship between genotype and phenotype, and between phenotype and clinical response, we created four scenarios for levels of linkage between genotype and clinical outcome. These four scenarios corresponded to the four possible combinations of level of correlation between genotype and phenotype (high or low), and correlation between phenotype and clinical response (high or low).

Response rates for the non-CYP metabolized medication were assumed to be the same for all three genotypes since metabolism is not affected significantly by any one of the polymorphisms. For the purposes of this model we assumed that the analytic sensitivity and specificity of the genetic testing used in the field compared to a gold standard genetic testing was 100 percent.

In order to understand the impact of each strategy on quality of life, patient outcomes were adjusted by a quality of life multiplier. This multiplier intended to represent patient preferences for a given health state as a utility. We used the utility of moderate depression to represent those individuals who did not respond to medication by 6 weeks, and a utility very close to that of non-depressed healthy individuals for those who did.⁸⁸

Outcomes. We estimated two different clinical outcomes at 6 weeks: percent response to medical therapy and cumulative quality-adjusted survival at 6 weeks (in years). Response to medical therapy was defined as a 50 percent or greater improvement as measured by the HAM-D scale. We chose to measure these outcomes at 6 weeks, since response to an initial 6-week trial predicts both ultimate success with a medication and adherence to it. Longer time frames do not improve the response to initial therapy, and since adverse effects are rarely serious, the greatest potential benefit of genetic testing will be to improve initial response rates. In addition, we calculated the average cost for each strategy over a single trial of therapy (6 weeks).

Analyses. In decision modeling it is typical to create a best-guess or “base case” estimate of outcomes. Given the lack of high quality data permitting a credible point estimate for model inputs, we chose to provide results for each of the four levels of linkage, described above. For each of these levels, we also performed one-way sensitivity analysis on all other model inputs (that is, other than probabilities related to the levels of linkage.)

Results

The outcomes for each of the four analyses are presented in Figures 3 and 4.

For each of the four scenarios, treating with a non-CYP metabolized SSRI without testing was the most effective strategy, while treating with a CYP metabolized SSRI was the least effective. Of the two testing strategies examined, using testing to guide use of a CYP- versus a non-CYP metabolized SSRI was superior to using testing to guide the dose of a non-CYP SSRI, both in terms of response rates and quality-adjusted life. However, as the level of linkage between genotype and phenotype increased, the difference in efficacy between the two testing strategies and between the testing strategies and the dominant strategy narrowed, such that at the high linkage level both testing strategies approached the efficacy of the optimal strategy of using a non-CYP metabolized SSRI. For example, in the low linkage scenario, the difference between the two testing strategies was 7.92 % in response rate and 0.04 years for cumulative quality-adjusted survival, while in the high linkage scenario the difference was only 0.78% in response rate and 0.005 years for cumulative quality-adjusted survival at 6 weeks.

One-way sensitivity analyses were performed for the following variables: prevalence of each phenotype, utility of depression, probability of responding to sertraline, cost of fluoxetine, cost of sertraline, and cost of genetic testing. The results of these analyses (not shown) were robust, with the relationship between the various options remaining similar at all levels of linkage between genotype and clinical response.

Because of the non-trivial cost of testing, 6-week costs are always greater for the testing strategy (results not shown), even when compared to the strategy of using a non-CYP metabolized drug without testing. For example, using genetic testing to guide medication choice cost $909 more than empiric therapy with a non-CYP medication, while using genetic testing to guide CYP dosing cost $882 more. The least effective strategy was also the least expensive- empiric treatment with a CYP metabolized medication cost $118 less than the empiric treatment with a non-CYP medication. However, if the length of treatment is expected to exceed approximately 9 months, the cost of the test strategies break even.

Discussion

In this analysis of the potential impact of CYP450 genotype testing on treatment outcomes in a trial of SSRI therapy, use of a non-CYP metabolized SSRI without testing was always the most effective strategy, and use of a CYP metabolized SSRI was always least effective. The two genetic testing strategies considered (testing as a guide to use of a CYP or non-CYP metabolized SSRI) had intermediate efficacy. The degree of efficacy depended primarily on the linkage between genotype and clinical outcome. At relatively low levels of linkage testing provides little benefit over use of a CYP metabolized SSRI without testing. Testing approached the optimal efficacy only at the highest levels of linkage between genotype and clinical outcome. Further, the modeling exercise suggests that the most important element of the link is the ability of genotype to predict phenotype. It is notable that these results apply even though it was assumed that the analytic validity of the test used (ability of the test to discern true genotype) was perfect.

Given the lack of evidence regarding many of the model inputs, it is important not to overstate the specific numerical results. However, the analysis does provide insight into the reasons why various strategies may or may not be clinically desirable. What is easiest to explain is the superiority of the strategy using a non-CYP drug without testing. The reason is that we assume that non-CYP medications do not have increased adverse event rates or reduced response rates in the poor and ultra-rapid metabolizers, respectively, and the CYP drug was assumed to never be superior for any phenotype. What may be less evident is why neither testing strategy was optimal for any combination of plausible model inputs. The explanation is that an imperfect genetic test (i.e., one that provides less than perfect guidance to metabolism, efficacy, or adverse effects) can lead to worse outcomes for misclassified individuals. When ultra-rapid or poor metabolizers are misclassified as extensive metabolizers, they are mistakenly managed with higher risk treatments. In the strategy in which testing is used to guide use of a CYP or non-CYP metabolized SSRI, misclassified individuals are given a CYP metabolized SSRI at standard doses, increasing their risk for adverse effects or lowering the probability of responding. In a strategy in which genetic testing is used to adjust the dose of a CYP metabolized SSRI, misclassified individuals are offered either very high or very low doses of the CYP metabolized SSRI, effectively doubling their risk of a poor outcome.

This basic analysis suggests that when non-CYP metabolized SSRIs are available, they should be used. When this approach is not feasible, CYP genotyping may provide similar patient outcomes if the test results can be shown to be highly predictive of clinical response. A difficulty in supporting the use of CYP450 genotype testing is the lack of evidence regarding the ability of CYP genotyping to guide treatment; if the correlation between genotype and outcomes is only modest, testing strategies are unlikely to be much more effective than treating with a CYP metabolized SSRI without testing. Also, since testing has its own cost, testing strategies do not save costs, even for the optimistic “high correlation” scenario, unless expected treatment duration exceeds approximately 9 months.

Clearly, studies of the relationship between genotype and clinical outcomes present a high value target for future research. Additional modeling which includes variable lengths of treatment, the possibility of treatment changes would help clarify the likely impact of CYP 450 genotype testing on long-term benefits, risks, and costs.

Question 1

We did not find any data to address directly the overarching question of whether testing for CYP450 polymorphisms in adults entering SSRI treatment for non-psychotic depression leads to improvement in outcomes, or whether testing results are useful in medical, personal, or public health decisionmaking.

Question 2

We identified only a few studies of test performance relative to the gold standard of DNA sequencing, applied to a limited number of genetic variants. Many studies appear to be in the realm of preclinical evaluations and are not clearly relevant to the domain of clinical practice.

These data do suggest that the analytic sensitivity and specificity of available tests are generally high. One concern may be that in the evaluation of gene deletions and duplications, assessing the magnitude of the potential problem is limited by the lack of an established gold standard for gene copy number. Another concern is that few CYP450 variants are included in the studies we identified, particularly less common variants.

Question 3a

In healthy CYP2C19 PMs, there is evidence of slower metabolism of SSRIs after a single dose, whereas in CYP2D6 PMs, the evidence is weaker. In depressed patients who have reached a steady-state concentration of an SSRI, the existing data (a series of heterogeneous studies in small samples) do not support a clear correlation between CYP metabolizer status and SSRI concentrations.

Question 3b

In depressed patients, the existing data (a series of heterogeneous studies in small samples) do not support a clear correlation between CYP metabolizer status and the efficacy of SSRIs.

Question 3c

In depressed patients, the existing data (a series of heterogeneous studies in small samples) do not support a clear correlation between CYP metabolizer status and the tolerability of SSRIs.

Questions 4a, 4b, 4c

We did not identify any studies that addressed whether CYP450 testing influences depression management decisions by patients and providers in ways that could improve or worsen outcomes, or whether testing for CYP450 polymorphisms in adults entering SSRI treatment for non-psychotic depression leads to improved clinical outcomes compared to not testing. Also, there were no data examining whether testing results are useful in medical, personal, or public health decisionmaking.

Question 5

There were no data on possible direct or indirect harms associated with testing for CYP450 polymorphisms and subsequent management options.

Model of Treatment for Major Depression

As a complement to the evidence review, we constructed a basic decision model to consider the circumstances under which testing for CYP polymorphisms could improve clinical outcomes, or favorably impact costs. We examined four strategies: (1) use a non-CYP metabolized SSRI without testing; (2) test and choose a non-CYP or CYP metabolized SSRI based on the result; (3) test and choose the dose of a CYP metabolized SSRI based on the result; and (4) use a CYP metabolized SSRI without testing. In no plausible scenario was a testing strategy predicted to improve expected outcomes of treatment at 6 weeks. The efficacy of a test strategy could approach the efficacy of use of a non-CYP metabolized drug, although this required the condition that a high correlation exist between genotype and phenotype (metabolizer status), as well as between phenotype and clinical outcomes. Current evidence does not support the conclusion that such high correlations apply. Moreover, the cost of testing is not offset by treatment savings if treatment duration is less than approximately 9 months.