Answers are grounded in this paper and cite their sources.
In Reply, We value the opportunity to respond to the letter by Dr Bruce Cohen et al1 in response to our recent meta-analysis of studies using the GeneSight Psychotropic Test entitled, “Meta-Analysis of Response and Remission Outcomes With a Weighted Multigene Pharmacogenomic Test for Adults With Depression.”2 The authors raise several issues relevant to the ongoing debate around the evidence base supporting the use of pharmacogenomic (PGx) testing in psychiatry, which we address below. INADEQUATE BLINDING The letter indicates that “none of the studies used in the meta-analysis had appropriate blinding,” and implies that the analysis is inappropriate and inflates the effect size. Complete blinding in clinical trials requires that the clinician, the patient, and the outcome assessor all remain uninformed about the treatment arm assignments. In pharmaceutical trials for major depressive disorder, complete blinding of all 3 components is theoretically possible, though the emergence of side effects frequently leads to functional unblinding of both clinician and patient,3 and is the major factor driving arguments for using “active placebos” in comparator arms.4 In psychotherapy trials, neither the clinician nor the patient can be blinded due to the nature of the intervention. Trials evaluating decision support tools (eg, PGx tests) to help guide treatment decisions of clinicians can fully blind patients and assessors, though blinding of clinicians is more challenging, as discussed in our meta-analysis.5 Hence, while complete blinding is the ideal, it is rarely achieved, even for psychiatry’s most strongly evidence-based treatments. Criticizing PGx trials for not achieving the blinding ideal thus holds PGx trials to an unreasonable standard and ignores aspects of trial designs that protected against unblinding bias. The letter also asserts that inclusion of studies with both blinded and unblinded participants is inappropriate and would bias the results towards a greater effect size. However, no heterogeneity was found for response (I2 = 0%, P = 0.58) nor remission (I2 = 0%, P = 0.94) when meta-analyzing the data combined across all 6 studies (2 open-label studies and 4 RCTs), indicating that the effect size across the trials was consistent. In addition, a recent meta-analysis of the 3 GeneSight trials6–8 that had both rater and patient blinding through their primary endpoint has been published. This analysis had similar findings for both response (double-blinded RCTs only analysis: OR = 1.44, P = 0.004; current analysis: RR = 1.30, P < 0.001) and remission (double-blinded RCTs only analysis: OR = 1.69, P = 0.001; current analysis: RR = 1.41, P < 0.001).8 We agree with the general statement that a meta-analysis does not correct for the limitations of the included studies, but in the case of our meta-analysis, there is no evidence that the results “falsely inflate statistical significance of effects attributed to PGx testing” or can be attributed purely to bias derived from unblinding. In addition, in the original publication, we noted that the PRESIDE trial,9 in which clinicians are blinded, is underway. As part of the extensive paragraph on blinding issues in the discussion section of our meta-analysis, we acknowledged this study as an interesting next step but also discussed that the study design has limited applicability to real-world clinical settings, as most clinicians interpret and apply PGx results themselves rather than having an intermediary interpret the data and make a suggestion. The letter’s discussion of this issue also assumes that the only component of PGx testing reports that clinicians use is the recommendation “bins” (eg, red, yellow, green) into which medications are assigned. In fact, PGx reports present much richer data on genotyping that clinicians can use when deciding whether to follow the binned recommendations.10 For example, the test report may bin a specific medication into a “caution” category. If the clinician cannot look up the genetic basis for that recommendation in the full PGx report, then their medication decision cannot adequately take into account other factors (eg, being on a CYP enzyme-inducing or enzyme-inhibiting comedication) that would impact their medication choice. Trial designs that only provide clinicians with binned recommendations instead of the full report do not adequately test the clinical utility of the test. COMPARING PHARMACOGENOMIC TESTING TO TREATMENT AS USUAL The letter claims that treatment-as-usual (TAU) is not an appropriate comparator arm to PGx-guided care because it allows for the use of medications that do not directly follow clinical guideline recommendations. The authors have speculated this previously by stating, “some medications used in the treatment-as-usual group and advised against in the PGx-guided group had known lesser efficacy or greater adverse effects and would not be used under available best-treatment clinical protocols.”11 However, no data were cited to substantiate that claim. Clinicians in both the TAU and PGx groups had equal access to clinical practice guidelines and, therefore, were able to implement them into treatment. Moreover, trial clinicians were not required to follow the guidance of the PGx test (and sometimes did not in the GUIDED trial), a design factor that would act to diminish effect sizes in the PGx arm.7 Thus, this critique appears conceptual rather than evidence‑based. In addition, clinical trials commonly compare new decision support tools to TAU, as TAU reflects treatments that patients actually receive. PGx testing clearly influences clinician treatment decisions,12 but PGx testing results are intended to be used alongside other information gathered during clinical evaluation. Clinicians who utilize PGx testing are encouraged to consider PGx as a decision support tool to be used in conjunction with numerous other nongenetic clinical factors, clinician experience, and patient preference.5,8 ISSUES REGARDING PHARMACOGENOMIC TESTING AND ACTIONABILITY The letter suggests our citations conflate the usefulness of PGx and cites a partial statement from the International Society of Psychiatric Genetics (ISPG) website13 to indicate that the Clinical Pharmacogenetics Implementation Consortium (CPIC) does not provide PGx recommendations for clinical care. However, CPIC itself refutes this assertion, stating: “One barrier to implementation of pharmacogenetic testing in the clinic is the difficulty in translating genetic laboratory test results into actionable prescribing decisions for affected drugs. CPIC’s goal is to address this barrier to clinical implementation of pharmacogenetic tests by creating, curating, and posting freely available, peer-reviewed, evidence-based, updatable, and detailed gene/drug clinical practice guidelines.”14 A more complete reading of the CPIC position reveals that CPIC has published guidelines providing actionable recommendations for SSRIs, SNRIs, and tricyclic antidepressants based on CYP2D6, CYP2C19, and CYP2B6.15,16 This is, by definition, providing guidance for use in clinical care, and we are puzzled at the letter authors’ suggestion otherwise. Further, the letter cites United Health Care’s (UHC) commercial policy for not covering multigene PGx panels as evidence against the clinical utility of PGx testing.17 However, payer policies, both favorable and unfavorable, often change over time. Before 2026, UHC’s policy covered multigene PGx testing panels, and its policy change cited no clinical trial evidence undermining previous demonstrations of clinical utility. As this meta-analysis was published in September of 2025, after the change in policy, we expect that it will help inform future policy updates. The warning letter issued by the FDA referenced by the authors' letter was published in 2019 and addressed unapproved claims made by specific testing companies (specifically naming Inova Genomics Laboratory) and cautioned against the use of tests with unsupported claims.18,19 This statement in no way equates to a category-wide statement that all psychiatric PGx tests lack evidence of benefit. To date, the test investigated in our meta-analysis, the GeneSight Psychotropic test, is the most studied psychiatric PGx test clinically available. In addition, in the years since 2019, the FDA has published on its website statements recognizing the clinical importance of gene-drug pairs through both its Table of Pharmacogenetic Associations20 and its Table of Pharmacogenetic Biomarkers in Drug Labeling.21 Regarding its Table of Pharmacogenetic Associations, the FDA states, “Pharmacogenetic tests, along with other information about patients and their disease or condition, can play an important role in drug therapy. When a health care provider is considering prescribing a drug, knowledge of a patient’s genotype may be used to aid in determining a therapeutic strategy, determining an appropriate dosage, or assessing the likelihood of benefit or toxicity.” On the Table of Pharmacogenetic Biomarkers and Labeling, the FDA states, “Pharmacogenomics can play an important role in identifying responders and nonresponders to medications, avoiding adverse events, and optimizing drug dose.” In addition, both sites recognize gene-drug associations with multiple antidepressants and genes such as CYP2D6 and CYP2C19. This highlights the nuance of the FDA’s position regarding PGx, that is, some claims have sufficient evidence and others require a more cautious approach, dependent upon the outcome of additional data. Our publication provides an important contribution to the evidence base supporting the use of PGx information in clinical care. Finally, the letter frames the 2024 American Journal of Psychiatry paper22 as a review from the American Psychiatric Association Council on Research (APA-COR) Workgroup on Biomarkers and Novel Treatments, seemingly supported by the APA. However, in response to learning that the review was not a statement by the APA, the journal published a correction, clarifying the article as a report rather than a position statement: “The findings, opinions, and conclusions of this report do not necessarily represent the views of the officers, trustees, or all members of the American Psychiatric Association. The views expressed are those of the authors.”23 REPORTING OF EFFECT SIZE The letter claims that relative risk (RR) below 2 is “considered a small effect.” We do not agree with this claim. For example, the current meta-analysis of the 4 RCTs produced a RR = 1.40 (P < 0.001) for remission. This means that, in relative terms, 40% more patients achieved remission when their treatment was guided by PGx compared with TAU. We strongly believe that this 40% relative increase is clinically significant and not “small.” The letter states that RR <2 is “difficult to interpret because of the possibility of confounding factors that have not been accounted for.” We do not believe that confounding factors materially affected the results of our meta-analysis, especially considering the results were consistent when the analysis was limited to only the 4 RCTs; the primary purpose of randomization is to balance any potential confounding factors between trial arms, such that the difference between the arms can be attributed to the effect of PGx and not to confounding factors. Indeed, as with the critique of imperfect blinding, this critique of confounding of the RCTs in the meta-analysis could be equally applied to essentially all prior pharmaceutical and psychotherapy trials for depression. The letter suggests that using RR can exaggerate the effect of PGx. We are not sure what this suggestion is based on. Using relative risk measures (such as RR and odds ratio) is considered more appropriate for conducting meta-analysis than using absolute risk differences because relative risks are more consistent between different studies within a meta-analysis.24,25 We agree that the number needed to treat (NNT) is a useful measure of clinical benefit. However, what the letter refers to as NNT may be more appropriately referred to as the number needed to genotype (NNG) when related to genetic testing. However, we disagree with the assessment that the estimated NNG = 21 for remission due to PGx is weak. Achieving remission in those with depression is difficult: according to the letter’s estimate, the remission rate due to medication treatment without PGx guidance is only 11%, which corresponds to NNT = 9. While an NNG of 21 associated with benefit due to PGx is higher than that of the NNT for a benefit from medication treatment, the associated harms should also be considered. Medications used to treat depression have a high rate of significant side effects. In contrast, using PGx is not associated with an increase in the rate of side effects.7 Thus, considering both efficacy and tolerability, an NNG of 21 with no associated increase in side effects represents a reasonable and meaningful therapeutic contribution relative to an NNT of 9 accompanied by an increased risk of significant adverse effects. Furthermore, the information provided by PGx testing may have additional implications for use outside of the initial first-time use of the information. Panel-based PGx testing, such as the GeneSight test, provides genotype information for multiple genes, some of which may be used for future treatments, including treatments outside of psychiatry. For example, CPIC has guidelines for a variety of medications, including but not limited to celecoxib, clopidogrel, hydrocodone, omeprazole, and ondansetron.26–30 REGARDING COMPARATIVE CLAIMS AND MAGNITUDE OF EFFECT We agree with the letter’s statement that only a portion of antidepressant response for major depression can be attributed to genetic factors, and that the exact contribution of pharmacogenetic variation is uncertain. However, the absence of a definitive estimate does not diminish the demonstrated value of PGx testing in guiding more personalized treatment decisions. In addition, we agree that PGx testing is not intended to be a “magic bullet.” Rather, it is a decision support tool used in conjunction with other clinical and nongenetic factors, as stated previously. Just as depression is not caused by a single factor alone, effective treatment rarely relies on a single factor. Incorporating PGx information into this broader clinical context does not negate its positive potential impact. And while the second largest RCT of PGx for major depression (the GUIDED trial) did not meet its primary symptom outcome endpoint at week 8, it nonetheless showed significantly greater response and remission rates in the PGx arm.7 In addition, the largest RCT of PGx for major depression (the PRIME Care trial) met both of its primary outcomes, finding that patients in the PGx arm had a greater likelihood of achieving remission over 24 weeks and had a greater likelihood of being prescribed an antidepressant with a lower gene-drug interaction potential compared with TAU.12 Thus, the 30% increased likelihood of response (P < 0.001) and 41% greater likelihood of remission observed in our meta-analysis represent meaningful clinical improvements, consistent with a valuable decision support tool, not a “magic bullet” for treatment selection and dosing in depression. REGARDING CONFLICTS OF INTEREST All funding and affiliations for our meta-analysis were disclosed, and similarly, funding sources and author affiliations of all studies included in the meta-analysis were disclosed in those studies. We, therefore, are not certain what the letter authors suggest should have been done differently. Concerning the PRIME Care study specifically, it was independently conducted and funded by the US Department of Veterans Affairs system, with standard conflict-of-interest disclosures for individual investigators.12 Further, the findings of the meta-analysis were not dependent upon the findings of any single trial. HOW TO EVALUATE THE RESULTS PRESENTED The letter authors note that concerns previously raised have been ignored; however, multiple of these concerns were addressed in the discussion of the meta-analysis, which explicitly outlines and discusses study limitations (including the majority of studies being industry-funded and differences in approaches to blinding). In addition, the meta-analysis addresses a specific concern noted in the 2024 American Journal of Psychiatry paper,22 which stated that “because most of the [previously conducted] meta-analyses [in the psychiatric PGx field] integrate effects across different PGx tools, there is concern that these tools actually differ substantially in their treatment recommendations and cannot properly be combined in an analysis.” Our meta-analysis addressed this concern by asking whether a single weighted multigene test (GeneSight Psychotropic) used to inform prescribing decisions in adult patients with MDD who have failed ≥1 treatment leads to improved response or remission compared with TAU. Across 6 prospective controlled studies (n = 3532), the answer was yes, with significant and consistent benefits and no detectable heterogeneity. CONCLUSION We agree with the letter’s sentiment that PGx tests should be supported by evidence. The test investigated in our meta-analysis, the GeneSight Psychotropic test, is the most studied psychiatric PGx test clinically available, and studies evaluating this test, including the current meta-analysis, contribute to the general PGx evidence base. While we welcome constructive discourse, we do not believe the letter identified any methodological inaccuracies or improprieties in our analysis; instead, it reiterates previously articulated broad concerns about PGx studies and that, in multiple cases, were explicitly addressed within our meta-analysis. We, therefore, reaffirm the robustness and validity of our findings and remain focused on producing high-quality evidence to move the field of personalized medicine forward in psychiatry.
No takes yet. Share an insight, caveat, or question.
Albers et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: