Multi‐Gene Pharmacogenomic Testing in a Community‐Based Setting Is Feasible and Reduces Total Healthcare Costs

ABSTRACT

Pharmacogenomic (PGx) testing using multi‐gene panels (mgPGx) is documented to improve clinical outcomes; however, real‐world data on its economic impact remain limited. This study aimed to evaluate the utility and economic value of mgPGx testing among Medicare patients within a community‐based health system. We identified Medicare Advantage patients within the primary care setting of a community‐based health system hospital who were taking ≥ 1 PGx‐guided medication using a stratification algorithm. In total, 1042 patients participated in mgPGx testing. We evaluated the prevalence of PGx medications, polypharmacy involving PGx medications, and actionable results (i.e., a phenotype with PGx guidance and a relevant PGx medication). A Total Cost of Care (TCOC) analysis was performed for a subset of patients (n = 548) who underwent PGx testing and were matched to a control group that did not undergo PGx testing using propensity score matching. Total medical expenses over 12 months, both before and after testing, were compared. Forty‐four percent (n = 454/1042) of patients were ≥ 3 PGx‐guided medications. Over one‐third of patients who were on ≥ 3 PGx medications had ≥ 2 actionable results (35.5%, n = 161/454). The TCOC analysis demonstrated a trend toward a net cost savings of $1827 per member per year (PMPY), with $1582 in medical savings and $245 in pharmacy savings. Polypharmacy with PGx medications is prevalent, and mgPGx led to cost savings. Further research with a larger sample size is needed to replicate the results and assess the long‐term impact on healthcare utilization and costs.

Study Highlights.

• What is the current knowledge on the topic?

  • ○
    Pharmacogenomic (PGx) testing using multigene panels has been shown to improve drug‐related outcomes. However, data on its impact on economic outcomes remain limited.
• What question does this study address?

  • ○
    This study evaluates the prevalence of polypharmacy involving medications with pharmacogenetic guidance in a Medicare population within a primary care setting. It also evaluates the total cost of care and healthcare utilization associated with multigene Pharmacogenomic testing.
• What does this study add to our knowledge?

  • ○
    The findings demonstrate that polypharmacy with PGx‐guided medications is common, highlighting the potential utility of multigene testing. Additionally, the study shows a reduction in the total cost of care among patients who underwent multi‐gene PGx testing compared to those who did not undergo testing.
• How might this change clinical pharmacology or translational science?

  • ○
    This study adds to the growing body of evidence supporting the clinical utility and potential cost savings of multigene PGx testing.

1. Introduction

Multi‐gene pharmacogenomic (mgPGx) panels provide insights into multiple pharmacogenes simultaneously, enabling preemptive, genotype‐guided prescribing [1, 2]. Many PGx‐guided medications are supported by the United States Food and Drug Administration (FDA), prescribing guidelines such as the Clinical Pharmacogenetics Implementation Consortium (CPIC), and the Dutch Pharmacogenetics Working Group (DPWG). Many of these PGx‐guided medications are frequently prescribed and rank among the top 200 prescribed drugs in the U.S. Specifically, the PGx‐guided medications within the top 200 prescribed drug list accounted for approximately 600 million prescriptions for over 170 million people in 2022 [3]. Additionally, the actionable phenotypes (i.e., phenotypes that prompt medication and/or dose adjustments, monitoring, or cautious titration) associated with these drugs are prevalent, with nearly all individuals carrying at least one actionable phenotype, highlighting the potential population impact the mgPGx can have [4, 5].

Emerging data show that mgPGx is linked to improved drug‐related outcomes. The PREPARE study, which included 7000 patients tested with a 12‐gene PGx panel, demonstrated a 30% reduction in drug‐related adverse events compared to patients who did not undergo testing [6]. Additionally, several studies have shown improvements in healthcare utilization, including reductions in hospitalizations and emergency room visits, when mgPGx was utilized [7, 8, 9].

The economic value of PGx testing has been primarily demonstrated through studies focused on single‐gene PGx testing [10]. While limited data exist on the economic value of multi‐gene panels, several studies have shown the benefits in specific indications like depression [11, 12, 13, 14]. Projections suggested the potential economic value of multi‐gene panels in targeted populations, such as those undergoing percutaneous coronary intervention [15, 16]. Given the low incremental cost of mgPGx compared to single‐gene testing and its potential value for informing the treatment of multiple indications, mgPGx testing has the potential to provide a significant preventive health tool with both clinical and economic value [6, 10].

Primary care settings, where patients frequently encounter polypharmacy PGx medications, present a unique opportunity to demonstrate the clinical and economic impact of mgPGx testing. Therefore, this paper describes a structured PGx testing program implemented in primary care settings within a community‐based healthcare system to demonstrate the economic value of mgPGx testing. The study aimed to (1) demonstrate the utility of mgPGx in a primary care setting by evaluating the prevalence of PGx medications, and actionable results (defined as a person with an actionable phenotype and had a relevant PGx medication) among a Medicare population treated within this health system, and (2) evaluate the economic value of mgPGx testing in this population compared to a matched group of patients who did not undergo testing.

2. Methods

2.1. Patient Identification and Testing

We included patients who were established with a primary care provider (PCP) within St. Elizabeth Healthcare and were enrolled in the Humana Medicare Advantage plan (Figure 1). Pharmacy insurance claims of these patients were used to identify patients for outreach efforts for free mgPGx testing. A proprietary statistical clustering algorithm was used to identify patients most likely to benefit from PGx testing. This algorithm was designed to identify high‐risk patients by evaluating factors that suggest current drug inefficacy or potential for adverse events. For example, the factors considered included: medication prescription history (drug name, drug class, and duration on therapy); demographics (age, site of care); diagnosis with known PGx‐related medications (e.g., mental health); and the availability of claims data. The algorithm assigned a higher risk score for the presence of claims for PGx‐evidence medications and for a pattern of “trial‐and‐error” prescribing (e.g., multiple different prescriptions within the same drug class). The score was lowered for medications that had been repeatedly filled, suggesting a stable regimen. An individual who started clopidogrel in the past 14 days following percutaneous coronary intervention would be identified as higher risk than a patient with a recent prescription for a proton pump inhibitor that was previously refilled. Initially, outreach was targeted at those with the highest risk, who were deemed to have a high potential benefit from PGx testing. Additional outreach was expanded to include lower‐risk members to explore the broader utility of pre‐emptive testing. Patients deemed eligible for PGx testing were contacted via multiple ways, including electronic health record (EHR) portal messages from hospital staff, physical mailing, and follow‐up telephone outreach conducted by OneOme, the laboratory that performed the PGx testing (Figure 1). Outreach efforts started in June 2021 and ended in December 2022. PCPs were informed about the ongoing PGx testing initiative during regular all‐provider meetings. Additionally, in‐basket messages were sent to each PCP by a St. Elizabeth staff member to notify them that their patient had undergone PGx testing.

FIGURE 1.

Overview of the study.

Patients interested in the PGx offering enrolled via the OneOme Laboratory portal and requested a buccal swab DNA collection kit to be sent to their homes. They self‐collected the DNA samples and shipped them to the laboratory. A 27‐gene panel (RightMed Comprehensive Pharmacogenomic Test, OneOme, Minneapolis, MN) was used for PGx testing. We focused on 17 genes with high pharmacogenetic evidence (Table S1) and associated clinical actionability (primarily CPIC Level A or B), including those encoding for the CYP450 (e.g., CYP2D6, CYP2C19, CYP2C9, CYP3A5), drug transporters (SLCO1B1), warfarin therapy (VKORC1, CYP4F2, CYP2C Cluster–rs12777823), toxicity/safety (DPYD, UGT1A1, TPMT, NUDT15, HLA‐A*31:01, HLA‐B*15:02, HLA‐B*57:01, HLA‐B*58:01) and efficacy (e.g., IFNL4). Results were returned and reviewed by PGx‐trained pharmacists. The PGx data were stored as discrete data elements in the EHR, allowing best practice advisory (BPA, Figure S1) alerts to trigger if providers prescribe a drug that interacts with a patient's phenotype. For example, prescribing 5‐fluorouracil for a patient with a DPYD intermediate metabolizer phenotype will trigger a BPA to recommend a 50% dose reduction (Figure S1). Additionally, pharmacists sent clinical notes to the PCPs addressing actionable results (Figure S2). Pharmacists commented on additional future medications that could interact with the patient's PGx results. Finally, patients received a summary of their PGx results through EHR portal messages. The study was approved by St Elizabeth Healthcare Institution Review Board.

2.2. Study Analytical Procedures

Medication data (Table S2) were collected from the EHR to capture active prescriptions at the time of PGx result return, reflecting current treatment and the utility of the PGx results. Specifically, only medications listed as “active” in the medication list at the time of results return were included. We also assessed PGx medications from insurance claims in the 12 months before the stratification date. Having two resources for medications ensured capturing PGx medications not reflected in claims data if they were filled through other programs or cash. Since polypharmacy could increase the percentage of actionable results, we created multiple groups of patients based on the number of PGx medications these patients were on (1, 2, 3, 4, etc.). Within each group, we calculated the proportion of patients with actionable results. For example, we calculated the number and percentage of patients with actionable results among those taking one PGx medication, two PGx medications, three PGx medications, etc. An actionable result was defined as a gene‐drug pair where the patient's phenotype prompts a change in medication, dosage, or a monitoring recommendation communicated to the provider via a clinical note or Best Practice Advisory (BPA), based on clinical PGx guidelines (CPIC or DPWG) or FDA guidance, and the patient is on the medication of interest. The institutional priority designated CPIC guidelines as the primary source, followed by DPWG when CPIC was unavailable or under review, and finally, FDA‐approved labeling recommendations. In certain cases, when CPIC guidance was classified as “insufficient evidence” or “optional,” certain phenotypes were nonetheless included as actionable for close monitoring or watchful titration based on an institutional risk‐mitigation strategy, to alert the provider to the patient's increased risk of adverse effects or suboptimal efficacy. For example, for oxycodone, where CPIC guidance is insufficient, CYP2D6 Poor Metabolizer (PM) and Ultra‐rapid Metabolizer (UM) phenotypes were included based on implementation studies or large real‐world data documenting the impact of reduced CYP2D6 activity or phenoconversion by inhibitors on the pain response [17, 18, 19, 20]. A further example relying on institutional discretion was atomoxetine, where our program defined CYP2D6 Ultra‐rapid Metabolizers as actionable due to potential suboptimal efficacy. To further demonstrate the utility of mgPGx, we calculated the percentage of actionable results under hypothetical scenarios if testing was limited to only one gene (CYP2D6), versus two (CYP2D6 and CYP2C19), three (CYP2D6, CYP2C19, and CYP2C9), or four genes (CYP2D6, CYP2C19, CYP2C9, and SLCO1B1). We focused on these genes for their relevance to the most prescribed drugs in this cohort.

For the total cost of care (TCOC) analysis, we prioritized patients in the highest‐risk stratification group (n = 548). A propensity score matching (PSM) technique was used to create a control group that did not receive PGx testing. The control group was matched based on age, gender, number of PGx medications, pre‐testing medical/pharmacy spending, and healthcare utilization (inpatient admissions, emergency room visits). To account for potential unmeasured high‐cost conditions, the matching process also incorporated variables reflective of the patients' Elixhauser Comorbidity Index scores, which account for the presence of multiple chronic, high‐cost conditions, including heart disease and oncology‐related diagnoses, enabling balanced profiles between the tested and control cohorts. For the PGx‐tested patients, 12 months of incurred medical and pharmacy claims, pre‐ and post‐each patient's index date, were analyzed. The index date was defined as the date PGx results were communicated to the PCP. A three‐month claims runout period was included after the post‐testing to account for any claims still in process or under appeal to ensure that all spending occurring up to the end of the 12 months was captured. Patients who did not have a full 12‐month runout period were excluded from the analysis (n = 196). For the control group, the index date used to define the 12‐month pre‐post period was the same as the one for the matched pair who underwent PGx testing. The total spending (medical and pharmacy) for the 12 months following the index date was compared to the total spending in the preceding 12 months for both the test (PGx) and control groups. The change in TCOC between the two matched groups was analyzed using a Difference‐in‐Differences approach with a paired t‐test. Of note, the TCOC savings results were derived from the sum of allowed amounts for every claim, including reversals and denied claims for each member, in the pre‐ and post‐period. The allowed amount represents the total contracted cost of care, benefiting both the payer and the patient (through reduced out‐of‐pocket expenses). Medication changes were monitored via pharmacy claims data, identifying changes from one PGx medication to another within the same therapy class within 3 months of testing among those with moderate or severe gene‐drug interactions for the original medication.

3. Results

Out of patients who established care with an affiliated PCP, 11,115 patients were identified as eligible for this program based on the stratification algorithm (Figure 1). Of these patients, 6820 had an active EHR portal account and received an invitation to undergo testing. These patients, as well as those without a portal account, received an invitation letter via mail. Additionally, phone call outreaches were made to 1522 patients. Of all patients outreached (N = 11,115), 1042 underwent PGx testing, reflecting a 9% engagement. Of these patients who received phone calls, 10.9% underwent testing. Most patients (98%) self‐identified as White, reflecting the demographics of the catchment area. Fifty‐nine percent (N = 611) were female, with an average age of 74 years (SD = 12).

At the time of test result return, 86% of patients (n = 899) had at least one active PGx medication recorded in the EHR. Statins were the most commonly prescribed PGx‐guided medication, prescribed to 58% of patients (n = 609/1042), followed by proton pump inhibitors (PPIs) at 41%, metoprolol at 22%, NSAIDs at 13%, and CYP2C19‐dependent selective serotonin reuptake inhibitors (SSRIs, e.g., escitalopram and citalopram) at 9% (Figure 2). The top ten most commonly prescribed PGx medications are displayed in Figure 2. Among patients prescribed these drugs or drug classes, the prevalence of the actionable phenotypes (Table S1) ranged from 6% (metoprolol) to 64% (for PPIs, Figure 2). Approximately one‐third of patients taking statins had an actionable phenotype related to SLCO1B1, and a similar proportion (32.6%) of patients on clopidogrel had an actionable phenotype in CYP2C19 (Figure 2). The prevalence of actionable phenotypes for antidepressant medications ranged between 40% for CYP2D6‐metabolized antidepressants (e.g., paroxetine, fluvoxamine, venlafaxine) and 59% for CYP2C19‐metabolized antidepressants (e.g., escitalopram, citalopram).

FIGURE 2.

Prevalence of actionable phenotypes for top 10 drug classes.

To assess the usability of mgPGx, we evaluated the percentage of actionable results among patients taking multiple PGx medications captured from claims data. Of the 1042 patients who completed PGx testing, 917 were confirmed by claims data to be prescribed at least one PGx‐guided medication. This subset, n = 917, served as the primary denominator for assessing actionable gene‐drug results. Forty‐four percent (n = 454/1042) of the total tested cohort were taking ≥ 3 PGx medications (Table 1). The proportion of actionable results increased with the number of PGx medications. For example, among patients taking two PGx medications, 47% (n = 124/263) had at least one actionable result. This proportion increased to 64% (n = 145/225) for those on three PGx medications and to 83% (n = 108/130) for those on four PGx medications (Table 1). Over one‐third of the patients (35.5%, n = 161/454) who were taking ≥ 3 PGx medications had at least two actionable results. When evaluating the actionable results by the number of genes tested, we found that 16% of patients would have at least one actionable gene‐drug interaction if only a single gene (e.g., CYP2D6) was tested. This percentage increased to 40%, 44%, and 54% when two, three, and four genes were included, respectively (Table S3).

TABLE 1.

Count and percentage of patients with actionable results among patients with ≥ 1 PGx medication (n = 917) assessed using claims data.

Number of PGx meds	0 actionable results (%)	≥ 1 actionable results (%)	≥ 2 actionable results (%)	≥ 3 actionable results (%)
One (n = 200, 19.2%)	148 (74%)	52 (26%)	0 (0%)	0 (0%)
Two (n = 263, 25.2%)	139 (53%)	124 (47%)	25 (10%)	0 (0%)
Three (n = 225, 21.6%)	80 (36%)	145 (64%)	58 (26%)	13 (6%)
Four (n = 130, 12.5%)	22 (17%)	108 (83%)	49 (38%)	12 (9%)
Five (n = 68, 6.5%)	13 (19%)	55 (81%)	34 (50%)	10 (15%)
Six (n = 23, 2.2%)	4 (17%)	19 (83%)	15 (65%)	10 (43%)
Seven (n = 8, 0.8%)	2 (25%)	6 (75%)	5 (63%)	2 (25%)

Note: n represents the count of patients on PGx medications. An actionable result is defined as having a phenotype with clinical guidance for adjusting the dose or switching medication or for using cautious titration or monitoring for efficacy and side effects. The total number of patients taking ≥ 3 PGx medications (n = 454) is the sum of patients in the 3–7 medication groups (225 + 130 + 68 + 23 + 8). The count of those with ≥ 2 actionable results (n = 161) is calculated by summing the number of patients with 2 or more actionable results (58 + 49 + 34 + 15 + 5).

The TCOC analysis included 548 patients identified as high‐risk with a high opportunity to benefit from PGx testing and 548 matched controls. There was no significant difference in baseline demographics or spending between the test and control groups during the 12‐month period prior to the index date (Table 2). In the 12 months following the index date, patients in the non‐tested control group had an increase in total healthcare spending of $3675, compared to an increase of $1848 in spending among the PGx‐tested group during the same period (Table 3). This resulted in a net difference of $1827 in total per member per year (PMPY) costs between the two groups, representing a 49% lower spending among the tested group. Of this $1827 difference, $1582 was attributed to savings in medical spending and $245 to pharmacy spending.

TABLE 2.

Comparison of demographic characteristics and healthcare utilization between tested patients and matched controls.

Baseline demographics	Test (N = 548)	Control (N = 548)	p
Average age (SD)	73.9 (10.8)	74.3 (9.6)	0.48
Gender, n (%)
Female	302 (55.1%)	303 (55%)	0.94
Male	208 (38.0%)	210 (38.3%)
Unknown	38 (6.9%)	35 (6.4%)
Average medical spending (SD)	$15,285 ($29,553)	$15,293 ($29,752)	0.93
Average pharmacy spending (SD)	$3621 ($9251)	$3077 ($9182)	0.28
Average inpatient admissions (SD)	0.23 (0.59)	0.22 (0.55)	0.71
Average Emergency Room visits (SD)	0.30 (0.69)	0.27 (0.61)	0.33
Unique PGx medications (SD)	4.19 (1.96)	4.00 (1.83)	0.09
Proportion of patients with Annual Wellness visits	0.86 (0.35)	0.85 (0.36)	0.86

Note: Spending is expressed in U.S. dollars.

TABLE 3.

Total healthcare spending in 12 months before and after index date plus 3 months of claims runout.

Group	Pre‐index PMPY	Post‐index PMPY	Annual PMPY difference per patient
Test (N = 548)	$18,906	$20,754	$1848
Control (N = 548)	$18,370	$22,045	$3675
Difference	−$1827

Note: Index date: The date when PGx results were available and sent to the PCP. PMPY: per member per year.

We observed that 64 patients of the 548 members with actionable results had medication changes based on claims data. When focusing on these 64 members, we observed higher cost savings compared to the larger cohort. These 64 members had an average spending of $21,735 in the 12 months before the medication change, which decreased to $15,149 in the 12 months following the change. This difference represented a $6586 reduction, with the largest decrease observed in inpatient spending ($2956). Further, a comparison of the change in utilization between the intervention and control groups showed a favorable trend, with a net reduction of 7 inpatient events and a net reduction of 14 emergency room (ER) visits in the 12‐month post‐index period.

4. Discussion

Our study included 1042 Medicare beneficiaries who underwent PGx testing using a multi‐gene panel within a community‐based hospital system's primary care setting. Testing was part of a broader PGx program with established infrastructure to integrate results into the EHR as discrete data elements. Pharmacists reviewed the results, integrated the PGx data, and provided clinical recommendations to the PCPs. Further, EHR‐integrated BPAs allowed ongoing clinical decision support, helping to prevent inappropriate prescribing based on gene‐drug interactions.

The patients included in the study were identified through outreach efforts targeting patients who were recently prescribed medications with pharmacogenetic implications, as determined by claims data. Among these patients, 86% were prescribed one or more medications for which pharmacogenetic guidance was available at the time of results return. Statins, proton pump inhibitors (PPIs), metoprolol, NSAIDs, and antidepressants were among the most commonly prescribed medications, affected by genetic variations in SLCO1B1, CYP2C19, CYP2D6, CYP2C9, and CYP2B6.

We also observed polypharmacy among this cohort, with 44% prescribed at least three PGx medications. The high prevalence of PGx polypharmacy aligns with previous research that evaluated the incidence of PGx medication use in a four‐year period among Medicare Supplemental (age ≥ 65) or Medicaid (age 40–64) [21]. In this study by Samwald et al., half of the patients had at least one PGx medication, and one‐quarter to one‐third had two or more PGx medications [21].

A high proportion of patients had actionable phenotypes and were prescribed the relevant medications. For example, among patients who were prescribed a cardiovascular medication such as statins or clopidogrel, almost one‐third had an actionable phenotype in SLCO1B1 or CYP2C19, respectively. For patients taking CYP2C19‐metabolized antidepressants, almost 60% had a relevant CYP2C19 phenotype that could influence prescribing decisions. Additionally, among patients with polypharmacy, actionable gene‐drug results were common, with over one‐third of the patients having at least two actionable results. The high prevalence of actionable results in our cohort is likely driven by the polypharmacy characteristic of Medicare patients in an outpatient setting and the high prevalence of pharmacogenetic variants. This is consistent with other studies that evaluated the prevalence of gene‐drug interactions. For example, Pasternak et al. reported that up to 60% of patients had at least one gene‐drug interaction that could trigger adjustments or monitoring of therapy [22]. Similarly, data from the “All of Us” initiative documented that 99% of participants had a PGx phenotype with corresponding prescribing recommendations, and over 20% of participants had an actionable phenotype and prior exposure to a relevant medication [4]. Krulikas et al., documented in a Veteran patient population that patients with multiple actionable PGx phenotypes were more likely to meet polypharmacy criteria when prescribed one or more PGx‐impacted medications [23]. Although our study did not test if multiple PGx phenotypes are associated with polypharmacy, it is plausible that the presence of multiple actionable PGx phenotypes may contribute to polypharmacy.

Our TCOC analysis of a subset of patients demonstrated an impact on healthcare costs. We observed a non‐significant reduction in both medical and pharmacy spending among these patients who received PGx testing compared to matched, non‐tested controls. The $1827 per member per year (PMPY) savings, primarily driven by a reduction in medical spending, suggest that PGx‐guided medication management may have helped prevent emergency room visits or hospitalizations. This reduction in the allowed amount can accrue financial benefit to the payer (by decreasing financial liability) and the patient (by reducing out‐of‐pocket costs associated with healthcare utilization). Our results of reduced healthcare costs in the PGx testing group are consistent with other studies [7, 8]. Brixner et al., compared elderly patients who underwent PGx testing with non‐tested, matched controls and found significant reductions in hospitalizations and emergency department visits, leading to cost savings of approximately $218 per patient [8]. Furthermore, a recent study evaluated the impact of a PGx, comprehensive medication management program on healthcare utilization and costs in a self‐insured U.S. employee population [24]. This study found over 26 months that the participants had a 39% reduction in inpatient and emergency department visits and a 21% increase in outpatient visits compared to a matched control group. Although pharmacy costs increased, medical expenses decreased, with a reduction in both inpatient and emergency room visit costs of $1726 and $33.36 per member per month, respectively, compared to non‐tested patients [24].

Our study had several strengths, including real‐world feasibility data from a structured PGx testing program integrated with EHR data. We analyzed 12 months of claims data, which provided insights into cost trends associated with PGx testing, suggesting that the benefit of testing may extend beyond the initial intervention driven by EHR integration and clinical decision support. Several anecdotal cases from our health system identified that PGx results were used in guiding new prescriptions after the index date of this study. For example, one patient who qualified for testing and inclusion in this study based on an initial statin prescription experienced a myocardial infarction 6 months after testing. The existing CYP2C19 result was used to guide antiplatelet selection following stent placement. In another case, a patient was diagnosed with colon cancer 9 months after PGx testing (also based on statin use), and the DPYD result was used to pre‐emptively guide the dosing of 5‐fluorouracil. In both cases, serious drug reactions—and the potential for hospitalization and associated costs—were likely prevented due to the availability of PGx results. We acknowledge the limitations of our study. Limited member engagement of 9% could be attributed to known barriers to PGx testing, such as lack of familiarity among patients and providers, the impact of the COVID‐19 pandemic, provider comfort with current prescribing practice, and the impersonal nature of the outreach. We observed a slightly higher participation of ~11% among patients who had telephonic outreach. While our results showed cost savings, the difference in total spending between the PGx‐tested and control groups was not statistically significant, potentially due to the small sample size. The uptake of provider recommendations was not formally assessed; however, a small number of patients had changes in medications based on claims data. The low number of observed changes may reflect patients being on stable treatment or suggest that some changes occurred before PGx testing. The fact that a trending cost savings was observed despite the low rate of changes may have been facilitated by the integration of PGx data into the EHR. This integration, through continuous clinical decision support, could have guided medication or dose selection beyond the initial post‐testing recommendation for a medication change. Adherence was not formally assessed in our studies. We observed ∼$544 lower baseline pharmacy spending in the control group, which may suggest a difference in adherence or prescribing patterns between the two groups. Nevertheless, our PSM process controlled for the number of PGx medications and pre‐testing pharmacy spending to minimize baseline utilization differences. Further studies focusing on adherence would be important to isolate the effect of mgPGx testing on adherence and TCOC. Finally, the TCOC analysis was limited by sample size and follow‐up duration, which could have led to high variability in claims data and resulted in a non‐significant trend toward cost savings. Additionally, our study focused the TCOC analysis on the highest‐risk patient group as defined by the algorithm. Thus, the observed cost savings may not be generalizable across all risk strata of the Medicare population. Therefore, future larger studies with increased statistical power are needed to test the economic impact of mgPGx testing in lower‐risk groups.

In summary, our study demonstrated the clinical utility and potential cost savings of PGx testing for patients with polypharmacy in primary care settings. Further research in a larger population is needed to confirm these cost‐saving trends, identify therapeutic areas and patient groups most likely to benefit from testing and assess the long‐term impact of PGx testing using multi‐gene panels.

Author Contributions

N.E.R., J.D.A., T.K., J.K., A.R., A.D.‐N., A.S., and J.W.: Wrote the manuscript. N.E.R., T.K., and A.D.‐N.: Designed the research. N.E.R., A.D.‐N., J.C., J.D.A., P.M., S.P., C.B., A.C., G.M., B.W., and J.G.: Performed the research. N.E.R., T.K., J.C., and S.P.: Analyzed the data.

Conflicts of Interest

N.E.R.: None. J.D.A.: None. T.K., J.K., P.M. and A.R. are employed by OneOme. A.D.‐N. and J.C. are employed by Humana.

Supporting information

Appendix S1: cts70418‐sup‐0001‐AppendixS1.docx.

El Rouby N., Allen J. D., Koep T., et al., “Multi‐Gene Pharmacogenomic Testing in a Community‐Based Setting Is Feasible and Reduces Total Healthcare Costs,” Clinical and Translational Science 18, no. 12 (2025): e70418, 10.1111/cts.70418.