A Randomized Hybrid‐Effectiveness Trial Comparing Pharmacogenomics (PGx) to Standard Care: The PGx Applied to Chronic Pain Treatment in Primary Care (PGx‐ACT) Trial

ABSTRACT

This trial aimed to identify the effects of providing pharmacogenomic (PGx) results and recommendations for patients with chronic pain treated in primary care practices compared to standard care. An open‐label, prospective, largely virtual, type‐2 hybrid effectiveness trial randomized participants to PGx or standard care arms. Adults with pain ≥ 3 months who were treated with tramadol, codeine, or hydrocodone enrolled. Alternative analgesics were recommended for CYP2D6 intermediate or poor metabolizers (IM/PMs). Prescribing decisions were at providers' discretion. The trial randomized 253 participants. A modified intent‐to‐treat primary analysis assessed change in pain intensity over 3 months among IM/PMs (PGx: 49; Standard care: 57). The PGx and standard care arms showed no difference in pain intensity change (−0.10 ± 0.63 vs. −0.21 ± 0.75 standard deviation; p = 0.74) or PGx‐aligned care (69% vs. 63%; standardized difference [SD] = 0.13). In IM/PMs, secondary analyses of pain intensity change suggested improvements with PGx‐aligned (n = 70; −0.21 ± 0.70) vs. unaligned care (n = 36; −0.06 ± 0.69) (SD = −0.22), with this difference increasing when examining IM/PMs with an analgesic change (aligned: n = 31, −0.28 ± 0.76; unaligned: n = 36, −0.06 ± 0.69; SD = −0.31). This approach to PGx implementation for chronic pain was not associated with different prescribing (i.e., similar proportions of PGx‐aligned care) or clinical outcomes. Secondary analyses suggest that prescribing aligned with PGx recommendations showed a small improvement in pain intensity. However, the proportion of patients with a clinically meaningful improvement (≥ 30%) in pain intensity was similar. Future efforts should identify effective implementation methods.

Summary.

• What is the current knowledge on the topic?

  • ○
    Prior evidence, including Clinical Pharmacogenetic Implementation Consortium (CPIC) guidelines and a nonrandomized prospective pragmatic trial, indicates that CYP2D6‐guided pain management may improve pain symptoms among those with chronic pain.
• What question did this study address?

  • ○
    What are the effects of providing pharmacogenomic (PGx) results and recommendations for patients with chronic pain treated with tramadol, hydrocodone, or codeine in primary care practices compared to standard care? This hybrid‐effectiveness trial assessed clinical (e.g., pain intensity, physical function) and implementation outcomes (e.g., prescribing alignment with PGx recommendations).
• What does this study add to our knowledge?

  • ○
    It found that participants hypothesized to benefit (CYP2D6 poor or intermediate metabolizers) experienced a similar change in pain intensity in the PGx and standard care arms. The two arms experienced similar care, as evidenced by the proportion of participants with PGx‐aligned care (PGx: 69% vs. Standard Care: 63%; SD = 0.13). Therefore, we conclude that the implementation methods used in this trial were ineffective in changing prescribing behavior. A secondary analysis found that when care was aligned with PGx results, it was associated with improved pain symptoms compared to when it was unaligned with PGx results.
• How might this change clinical pharmacology or translational science?

  • ○
    This study highlights the importance of a hybrid‐effectiveness trial design. Additional future efforts should identify effective implementation strategies for integrating PGx results into opioid prescribing.

1. Introduction

Twenty percent of US adults experienced chronic pain in 2021 [1]. Chronic pain has pervasive effects on quality of life, and treatment is notoriously difficult [2]. Opioids are effective for moderate‐to‐severe acute pain and their use in chronic pain has garnered enhanced scrutiny in light of the opioid epidemic [3]. Hydrocodone and tramadol are the most widely prescribed opioids in the United States, with 8.6 and 5 million patients prescribed them in 2021, respectively [4]. Prior studies indicate clinical responses to hydrocodone and tramadol are associated with genetic polymorphisms in cytochrome P450 2D6 (CYP2D6) [5].

The CYP2D6 enzyme bioactivates hydrocodone, tramadol, and codeine to more potent metabolites responsible for pain relief. These metabolites, hydromorphone, O‐desmethyl tramadol, and morphine, have a 100‐, 200‐, and 200‐fold higher affinity for the μ‐opioid receptor, respectively [5]. Patients with reduced CYP2D6 activity, termed CYP2D6 intermediate metabolizers (IMs) and poor metabolizers (PMs), may be at increased risk of therapeutic failure from these medications. Smith et al. conducted a non‐randomized cluster design pragmatic proof‐of‐concept trial that found CYP2D6‐guided analgesic prescribing was associated with improved pain symptoms in CYP2D6 IM/PMs prescribed tramadol or codeine compared to usual care [6]. A post hoc analysis identified that CYP2D6‐guided care may also improve pain symptoms among those prescribed hydrocodone. Although encouraging, a randomized trial could provide more conclusive evidence about the benefit of PGx testing.

Implementation methods also need to be tested. The utilization of pharmacogenomics (PGx; the use of genetics to guide medication therapy) has been impeded by a lack of proven strategies to integrate PGx into care [7]. We performed a randomized type 2 hybrid‐effectiveness trial to identify the effects of providing PGx results and recommendations for patients with CYP2D6 IM/PM and chronic pain treated in primary care practices compared to standard care.

2. Methods

2.1. Study Design

The Pharmacogenomics Applied to Chronic Pain Treatment in Primary Care (PGx‐ACT) trial was an open‐label randomized type 2 hybrid‐effectiveness trial (NCT04685304). A type 2 hybrid‐effectiveness trial examines both implementation outcomes (e.g., prescribing alignment with PGx recommendations) and clinical outcomes (e.g., pain intensity) [8, 9]. Participants were randomized to PGx vs. standard care (Figure 1) [8, 9]. The trial was designed to not require in‐person activities beyond usual care. A statistician not involved in the study randomized participants in mixed blocks of 6 and 4. Allocation was stratified by baseline opioid, with 1:1 to tramadol or codeine to hydrocodone, respectively.

FIGURE 1.

Study design.

2.2. Population

Eligibility included patients 18 years or older with chronic pain (i.e., pain lasting for 3 or more months) who were prescribed either tramadol, hydrocodone, or codeine. Patients had to be treated at a participating primary care practice site and prescribed one of the target opioids by a provider within the MedStar Health system (additional eligibility details in supplement). Providers could have participants randomized to either arm. The study was approved by the IRB at MedStar Health Research Institute, and all procedures were in accordance with the ethical standards of the Declarations of Helsinki. Electronic health records (EHR) were screened for patients at participating sites prescribed tramadol, codeine, or hydrocodone. Multiple recruitment modalities were used, including provider referrals, patient portal messaging, emails, text messages, and telephone calls. Participants provided physically or electronically signed informed consent before any study intervention (e.g., sample collection). Participants received $25 after trial completion.

2.3. Pharmacogenetic Testing and CYP2D6 Phenotype Assignment

PGx testing was performed in a Clinical Laboratory Improvement Amendments (CLIA) certified laboratory (Kailos Genetics Inc.) using targeted next‐generation sequencing on select genes (i.e., CYP2C19, CYP2C9, CYP2D6, CYP3A4, CYP3A5, SLCO1B1, TPMT, VKORC1). Allele coverage is available in Table S1, with all tier one CYP2D6 alleles covered [10] Allele function was assigned per the Clinical Pharmacogenetics Implementation Consortium (CPIC) [5]. CYP2D6 phenotype was assigned by CYP2D6 activity score thresholds per the most recent CPIC guidelines: 0, PM; 0.25–1, IM; 1.25–2.25, NM; above 2.25, UM [5]. In cases with gene multiplications, ranged phenotypes (e.g., IM to NM) communicated possible activity score ranges.

Drug interactions were incorporated into CYP2D6 phenotype assignment. CYP2D6 inhibitors can reduce enzyme function compared to genotype [5, 6, 11]. This process, called phenoconversion, was accounted for similar to prior work [6, 12]. In brief, CYP2D6 activity scores were converted to zero in the presence of FDA‐defined strong CYP2D6 inhibitors (i.e., bupropion, fluoxetine, quinidine, paroxetine, terbinafine) [13, 14]. For moderate CYP2D6 inhibitors (i.e., abiraterone, cinacalcet, duloxetine, fluvoxamine, lorcaserin, mirabegron), the CYP2D6 activity score was multiplied by 0.5, often leading to CYP2D6 NMs becoming IMs.

2.4. Intervention and Implementation Strategy

Prescribing decisions were at the treating provider's discretion. Participants provided a buccal sample via at‐home mailing kits or in‐office collection. Participants were randomized to PGx or standard care after the laboratory receipt of their sample. The PGx arm had samples processed and released as soon as they were available. Results were released for the standard care participants upon completion of active participation (i.e., 3 months after baseline).

A multimodal approach incorporated PGx results into clinical care. First, results (e.g., genotype, phenotype, activity score) were entered as structured data into the results review section of the EHR within a “Pharmacogenomics” tab. Although the trial focused on CYP2D6, structured data were also entered for CYP2C9, CYP2C19, CYP3A5, SLCO1B1, and TPMT. Second, targeted posttest interruptive electronic clinical decision support (CDS) alerts provided recommendations for alternative therapy upon order entry of tramadol or codeine for patients with CYP2D6 PM or UM phenotypes based on CYP2D6 genotype (Figure S1). The alerts triggered for any patient regardless of study enrollment. Therefore, therapeutic recommendations in alerts aligned with CPIC guidelines [5]. Alerts did not account for phenoconversion and did not trigger other medications related to the treatment of pain.

Third, PGx‐trained pharmacists used PGx results to create a consultation note (PharmD consult) to aid providers in interpreting and applying PGx results. PharmD consults were delivered as results were available and were asynchronous to patient visits. After discussion with providers before the trial, the PharmD consult was placed as a consultation note in the EHR and sent to the primary care provider (PCP), the provider who ordered the opioid (if different from the PCP), and any other relevant provider per the pharmacist's discretion. A PGx‐trained pharmacist was available upon request to discuss results with patients/providers directly, but this was rarely utilized. Pharmacists used CYP2D6 phenotypes informed by genotype and concomitant CYP2D6 inhibitor use. Recommendations are shown in Table 1 and were based on CPIC guidelines and the findings from the prior prospective trial by Smith et al. [5, 6]. Due to varying pain etiologies and treatment histories, when alternative therapy was recommended, it included broad options for providers to select from. For example, “select a non‐opioid analgesic (e.g., naproxen, gabapentin, acetaminophen) or a different opioid (e.g., morphine, hydromorphone, oxycodone).” We did not recommend oxycodone in UMs. Recommendations were provided for additional drug‐gene pairs per CPIC guidelines If the patient had a documented condition treated by a medication with CPIC guidelines. This was intended to provide the best patient‐centered care, and it is possible other drug–gene pairs could contribute to pain symptoms (e.g., CYP2C9‐NSAIDs, CYP2D6/CYP2C19‐antidepressants). Results for other drug–gene pairs are beyond the scope of this work.

TABLE 1.

Recommendations by CYP2D6 phenotype.

CYP2D6 phenotype a	Recommendations
Ultrarapid metabolizer	Select alternative therapy due to the risk of toxicity from tramadol, codeine, or hydrocodone
Normal metabolizer	No changes recommended
Intermediate metabolizer b	Recommendations for IMs were stratified based on the patient's pain relief at their next visit. ○ If the provider and patient considered analgesia adequate, providers could continue tramadol, hydrocodone, or codeine therapy or replace the opioid with a nonopioid analgesic and thereby use PGx as part of an opioid‐de‐escalation strategy. ○ If the provider and patient considered analgesia inadequate, then discontinue the enrollment opioid and prescribe a nonopioid analgesic or a different opioid (e.g., oxycodone)
Poor Metabolizer b	Select alternative therapy due to the risk of ineffectiveness from tramadol, codeine, or hydrocodone

^a Per CYP2D6 genotype and concomitant CYP2D6 inhibitor use.

^b If a patient was phenoconverted to IM/PM from NM, another option was to discontinue the CYP2D6 inhibitor to return to NM and continue the opioid.

2.5. Outcomes

The outcomes address multiple domains of the Reach, Effectiveness, Adoption, Implementation, and Maintenance (RE‐AIM) framework (Table S2), focusing on the effectiveness and implementation domains [15, 16]. The primary analysis was a modified intention‐to‐treat (mITT) analysis of those with a CYP2D6 IM/PM result who completed the trial (i.e., the follow‐up survey at 3 months). The primary endpoint was the change in pain intensity Patient Reported Outcome Measurement Information System (PROMIS) composite among CYP2D6 IM/PMs between baseline and 3 months [17, 18]. The patient‐reported outcome (PRO) instruments utilized at baseline and 3 months were PROMIS‐29 Profile v2.0 and PROMIS Scale v1.0 Pain Intensity 3a [17, 18]. The pain intensity composite was the average of the pain intensity current, worst, and average over the past 7 days [6]. The baseline visit was defined as the next scheduled appointment with their outpatient provider who prescribed the opioid of interest (i.e., visit after enrollment) if it was ≥ 10 days after the lab received the sample. The delay between enrollment and baseline visits allowed for time to process PGx results and provide a PharmD consult note, which ensured PGx result availability in the PGx arm. The outpatient provider and the participant determined the timing of baseline visits through usual care practices. The study coordinator distributed questionnaires electronically and, if necessary, via phone or in person.

Planned secondary endpoints included an assessment in CYP2D6 IM/PMs between baseline and 3 months: change in morphine milligram equivalents (MMEs) prescribed [19], pain interference (how much pain affects a patient's quality of life), physical function, proportion with a ≥ 30% improvement in pain intensity (clinically meaningful improvement), and proportion with PGx‐aligned care. PGx‐aligned care was defined as pharmacotherapy concordant with PGx recommendations provided in the PharmD consults. It is possible providers made pharmacotherapy decisions for participants in the standard care arm that happened to align with PGx results without knowledge of PGx results. An intention‐to‐treat (ITT) analysis was performed among all randomized participants with pain intensity composite data available at baseline and follow‐up. Implementation metrics included turnaround times between various steps within the workflow, recruitment success, clinician responses to CDS alerts, the proportion of PGx sample collection kits returned, and the number of pharmacists delivering PharmD consults. Data were securely stored in research electronic data capture (REDCap) [20].

2.6. Analysis

The primary analysis used penalized multiple linear regression with elastic‐net methods to assess the effect of other covariates on the primary endpoint, the change in pain intensity composite. This trial possessed 80% power to detect an effect size of 0.6 at an alpha of 0.05 when 90 IM/PMs completed the trial. This large effect size was selected based on a prior nonrandomized prospective study's results, further sample size calculation details are available in the supplement [6]. Elastic net methods shrink regression coefficients toward zero, thereby identifying variables with the most impact. Specifically, covariates such as age, sex, self‐reported race, opioid type, and baseline PROMIS measures were included in the initial model, along with the interactions of trial arm with age, sex, and race. In addition, site was included as a random effect to assess whether between‐site variability could be ignored. Two planned sensitivity analyses were performed. They repeated the primary analysis after (1) excluding participants who had unplanned surgical procedures that typically necessitate postoperative opioids that occurred during the study period (sensitivity analyses 1 and 2) excluding participants who had their baseline visit in‐between questionnaires but > 7 days from the baseline questionnaire (risk: symptoms may not be reflective of symptoms at time of baseline visit) or the baseline visit was with a non‐PCP or nonopioid‐ordering provider (risk: bias to the null as provider may have been less likely to modify pain management; sensitivity analysis 2).

Secondary analyses are considered preliminary and utilized standardized differences (SD) as SD is not dependent on sample size [21]. For proportions, SD is the difference in proportions divided by the standard deviation of the difference: SD_P = (P1—P2)/sqrt(variance1 + variance2)/2. Standard deviation and interquartile range (IQR) are reported as appropriate. Descriptive statistics were reported for implementation metrics.

3. Results

Twenty of 23 clinics approached to participate in the trial agreed to participate. Reasons clinics declined participation included COVID‐19 response activities (n = 1), providers not interested (n c= 1), and no response (n = 1). Training was delivered to 109 providers at these sites, and 86 providers provided care to at least one participant who enrolled in the trial. Of these 86 providers, the median number of enrolled participants was 3 [1, 5], with a maximum of 18 participants. Between January 2021 and December 2022, 4573 patients were evaluated, with 2731 ineligible (e.g., not prescribed a relevant opioid), 944 who did not respond to outreach, 428 who declined to participate, 154 who did not participate for an unknown reason, and 315 participants enrolled (Table S3). Consent was provided virtually (e.g., eConsent) in 267 of 315 (85%) enrolled participants.

Two hundred and fifty‐three participants were randomized to PGx (n = 128) and Standard Care arms (N = 125; Figure 2). Common reasons for attrition between enrollment and randomization were patients not returning the PGx collection kit (n = 42), screen failures (n = 13), and consent withdrawal (n = 7). Additional attrition occurred given that 223 completed the baseline visit out of 253 randomized participants. The most common reason was the patient not returning to the primary care practice for care (n = 20; Figure 2). Patients who completed the baseline visit were likely to complete the trial (97%; 217 of 223). Trial participation was often virtual; 175 (81%) of 217 participants completed all questionnaires electronically (e.g., link via text message or email). Follow‐up was completed in July 2023.

FIGURE 2.

Consort diagram.

Trial participants consisted mainly of older adults with pain for over 5 years (Table 2). Back pain was the most common pain management indication. The median pain intensity at baseline was 7 [5, 8], translating to moderate to severe pain. Tramadol was the most commonly prescribed opioid at baseline (176 [79%]). Some populations typically underrepresented in genomic research were well represented, as 85 (38%) participants self‐identified as Black or African American, and 164 (74%) were female. Baseline characteristics were similar between arms. CYP2D6 IM/PM phenotypes were present in 106 (49%) participants who completed the trial (Tables S4 and S5). Twenty‐two of 122 (18%) participants who completed the trial with NM per genotype were phenoconverted to IM or PM, with 11 experiencing it in each arm. Ten of 76 (13%) IMs per genotype were phenoconverted to PM (PGx: 4; Standard care: 6). The most commonly prescribed CYP2D6 inhibitors overall were duloxetine [18], bupropion [16], fluoxetine [8], mirabegron [5], and paroxetine [3].

TABLE 2.

Baseline characteristics of all participants who completed baseline visit.

Baseline characteristics a	PGx (n = 109)	Standard care (n = 114)	SD
Age, years	66.5 (59, 71)	64.1 (57, 72)	0.02
Sex, female b	82 (75)	82 (72)	0.07
Self‐reported race c
White	58 (53)	60 (53)	0.01
Black or African American	39 (36)	46 (40)	−0.09
Multiple races	6 (6)	2 (2)	0.20
Native American or Alaska Native	1 (1)	1 (1)	0.00
Native Hawaiian or other Pacific Islander	0 (0)	2 (2)	−0.19
Asian Indian	0 (0)	1 (1)	−0.12
Prefer not to say	5 (5)	2 (2)	0.16
Ethnicity
Hispanic or Latinx	2 (2)	1 (1)	0.08
Non‐Hispanic or Latinx	101 (93)	105 (92)	0.02
Prefer not to say	6 (6)	8 (7)	−0.06
Pain Management Indication d
Back pain	64 (59)	73 (64)	−0.11
Arthritis	58 (53)	49 (43)	−0.21
Musculoskeletal	17 (16)	14 (12)	0.10
Duration of Pain
< 1 year	5 (5)	4 (4)	0.05
1–5 years	26 (24)	31 (27)	−0.08
> 5 years	78 (72)	79 (69)	−0.05
Pain Intensity (11‐point scale)	7 (5, 8)	7 (5, 8)	−0.03
Pain Intensity Composite (5‐point scale)	3.0 (2, 4)	3.3 (2, 4)	−0.01
Baseline Opioid Use e
Tramadol	86 (79)	90 (79)	0.00
Hydrocodone/acetaminophen	11 (10)	14 (12)	−0.07
Codeine/acetaminophen	13 (12)	10 (9)	0.10
Two opioids	4 (4)	6 (5)	−0.08
MME prescribed	10 (10, 20)	15 (10, 20)	−0.02

Note: Reports data from the 223 participants who completed the baseline visit.

Abbreviations: IM/PM: CYP2D6 intermediate or poor metabolizer; MME: Morphine milliequivalents; SD: standardized difference—small effect: 0.2; moderate effect: 0.5; large effect: 0.8.

^a Continuous data shown as XX (XX, XX) are reported as median (25th percentile, 75th percentile), and continuous data shown as XX (XX) are reported as mean (standard deviation).

^b No patients identified as intersex or other.

^c No patients identified as Japanese, Korean, Vietnamese, or other Asian.

^d Top 3 most prevalent indications are shown. Patients may have more than one indication.

^e One patient in the PGx group was prescribed two enrollment opioids (codeine, tramadol).

The ITT analysis, among all participants who completed the trial (n = 217), found the treatment arm had little effect on the change in pain intensity composite: −0.21 ± 0.79 vs. −0.12 ± 0.83; SD = −0.12 in the PGx and standard care arms, respectively. The mITT primary analysis did not identify a difference in pain intensity between the 49 and 57 participants with CYP2D6 IM/PM assigned to PGx and Standard Care, respectively (−0.10 ± 0.63 vs. −0.21 ± 0.75; p = 0.74; additional data in Table S6). Sensitivity analyses (supplement) and other clinical endpoints (Table 3) were also similar. Notably, the proportion with PGx‐aligned care was similar between arms: 34 of 49 (69%) in PGx vs. 36 of 57 (63%) in the standard care arm; SD = 0.08. This suggests the clinical intervention was not effective in changing prescribing decisions.

TABLE 3.

Endpoints among CYP2D6 IM/PMs.

Primary endpoint	PGx n = 49	Standard care N = 57	p
Change in pain intensity composite	−0.10 ± 0.63	−0.21 ± 0.75	0.74 a

Secondary endpoints	PGx n = 49	Standard care N = 57	SD
Proportion with PGx‐aligned care	34 (69%)	36 (63%)	0.13
Proportion with a ≥ 30% improvement in pain intensity	4 (8%)	8 (14%)	−0.19
Physical function b	0.37 ± 4.1	0.89 ± 3.9	−0.13
Pain interference b	−0.01 ± 5.3	−1.9 ± 6.5	0.32
Change in MME Prescribed	−1.7 ± 13	−0.94 ± 13	−0.06

Abbreviations: IM/PM: CYP2D6 intermediate or poor metabolizer; MME: Morphine milligram equivalents; SD: standardized difference—small effect: 0.2; moderate effect: 0.5; large effect: 0.8.

^a Adjusted for baseline sleep t‐score, baseline social t‐score, baseline fatigue t‐score, baseline anxiety t‐score, Hx of hypertension, Hx of previous injury, and Hx of anxiety.

^b Per change in PROMIS T‐score between baseline and 3 months.

A planned secondary analysis examined the change in pain intensity among CYP2D6 IM/PMs by alignment with PGx recommendations (Table 4). This identified a minor reduction in pain intensity with PGx‐aligned care (n = 70; −0.21 ± 0.70) compared to unaligned care (n = 36; −0.06 ± 0.69; SD = −0.22). The magnitude of effect increased when limiting the analysis to those with at least one change in an analgesic medication, favoring PGx‐aligned care (n = 31) vs. unaligned care (n = 36; −0.28 ± 0.76 vs. −0.06 ± 0.69; SD = −0.31). However, the proportion of patients with a clinically meaningful improvement (≥ 30%) in pain intensity was similar (Table 4).

TABLE 4.

Endpoints by alignment with PGx recommendations in CYP2D6 IM/PMs.

Change in endpoint a	CYP2D6 IM/PMs
	PGx‐aligned care	Unaligned care	SD
	N = 70	N = 36
Pain intensity composite	−0.21 ± 0.70	−0.06 ± 0.69	−0.22
Proportion with a ≥ 30% improvement in pain intensity	8 (11%)	4 (11%)	−0.01
Physical function b , c	1.1 ± 3.7	−0.24 ± 4.4	0.33
Pain interference b , c	−1.3 ± 6.7	−0.61 ± 4.5	−0.11
MME prescribed b	−3.4 ± 13	2.8 ± 13	−0.49

Among those with ≥ 1 analgesic medication change	N = 31	N = 36
Pain intensity composite	−0.28 ± 0.76	−0.06 ± 0.69	−0.31
Proportion with a ≥ 30% improvement in pain intensity	5 (16%)	4 (11%)	−0.15
Physical function b , c	2.0 ± 4.0	−0.24 ± 4.4	0.54
Pain interference b , c	−3.2 ± 8.2	−0.61 ± 4.5	−0.41
MME prescribed b	−8.5 ± 16	2.8 ± 13	−0.80

Abbreviations: IM/PM: CYP2D6 intermediate or poor metabolizer; MME: Morphine milliequivalents; SD: standardized difference—small effect: 0.2; moderate effect: 0.5; large effect: 0.8.

^a Improvement is portrayed as a positive value for physical function and a negative value for pain intensity composite and pain interference.

^b Post hoc.

^c Per change in PROMIS T‐score between baseline and 3 months.

Of IM/PMs with a change in an analgesic medication, nonopioid analgesics were prescribed to 25 of (81%) of those with PGx‐aligned care vs. 32 (89%) with unaligned care (SD = −0.23). Opioid analgesics were prescribed to 9 (29%) of those with PGx‐aligned care vs. 36 (100%) with unaligned care (SD = −2.2). Post hoc analyses in this population also identified that PGx‐aligned care was associated with improved physical function (Table 4; SD = 0.54), pain interference (SD = −0.41), and a reduction in MMEs prescribed (SD = −0.79).

The improvement in pain intensity composite among those with PGx‐aligned care (n = 70) occurred regardless of treatment arm (−0.20 ± 0.65 vs. −0.22 ± 0.76; SD = 0.04 in PGx [n = 34] vs. standard care [n = 36]), which suggests the potential benefit of PGx‐aligned care was not the result of a placebo effect among those assigned to the PGx arm. This was also found in the subanalysis among those with an analgesic medication change (n = 31): −0.26 ± 0.80 vs. −0.30 ± 0.76; SD = 0.05. Changes in pain intensity by CYP2D6 activity score, opioid, and PGx alignment are shown in Tables S7–S13. Additional medication data and analyses by race are available in Tables S14–S16.

3.1. Implementation Metrics

Forty‐two (13%) enrolled participants did not return the PGx kit that was mailed to their home. The trial delivered PGx results for 249 of 253 (98%) randomized participants, albeit the standard care arm received them after trial participation completion. The four randomized participants who did not receive results were withdrawn (n = 3) or had a screen failure (n = 1) before the results were available. One buccal swab generated PGx results for 238 of 250 (95%) participants for whom sample processing was attempted. The other 12 participants received PGx results after a second sample collection.

Seventeen (16%) of 109 patients in the PGx arm had their PGx results mentioned in the provider's note at the baseline visit. The median (IQR) time between (1) enrollment and PGx result return: 38 (30, 49) days; (2) PGx result return and PGx consult note upload: 9 (5.5, 18) days; (3) PharmD consult note upload and baseline visit: 31 (8, 66) days. The goal was to deliver PharmD consults prior to the baseline visit, which meant the trial did not focus on rapid turnaround times. Sixteen pharmacists provided PharmD consults, with a median of 9 [5, 15] consults provided by each pharmacist. The targeted interruptive electronic CDS alerts were overridden due to “prior use with no reported problems” for all five participants with CYP2D6 PM who had an alert triggered for codeine or tramadol.

4. Discussion

This hybrid‐effectiveness trial found participants experienced a similar change in pain intensity in CYP2D6 IM/PMs in the PGx and standard care arms. The two arms experienced similar care, as evidenced by the proportion of participants with PGx‐aligned care (PGx: 69% vs. Standard Care: 63%; SD = 0.13). Therefore, we conclude that the implementation methods used in this trial were ineffective for this endeavor.

Since the two arms received similar care, similar pain symptoms are expected. A planned secondary analysis found a small improvement in pain intensity among CYP2D6 IM/PMs when care aligned with PGx recommendations compared to unaligned care (SD = −0.22). The magnitude of the effect increased when limiting the population to those with at least one change in an analgesic medication (SD = −0.31). Within this population, PGx‐aligned care was further associated with improved physical function and pain interference, which are recognized as core chronic pain outcomes [22, 23]. The effect sizes seen with a change in pain intensity among IM/PMs with PGx‐aligned vs. unaligned care can be considered a small effect, and there was minimal association with a ≥ 30% improvement in pain; however, they are consistent with interventions in other clinical trials studying chronic pain [24, 25, 26]. These improvements occurred in the setting of reduction in MMEs prescribed.

This trial builds upon the prior prospective trial by Smith et al., which found improved pain intensity among CYP2D6 IM/PMs prescribed tramadol, codeine, or hydrocodone [6]. Recent data also support the use of PGx‐guided opioid therapy in patients with acute pain and cancer pain [27, 28]. Additional data from the ongoing ADOPT‐PGx trial are forthcoming, which is a well‐powered trial with over 1000 participants [29]. One difference between the PGx‐ACT, ADOPT‐PGx, and the prior trial by Smith et al. is the cutoff between normal and intermediate metabolism. CPIC changed the translation of a CYP2D6 activity score of one from an NM to an IM [30]. All trials recommended alternative therapy for IMs. However, the original definition (activity score 1 = NM) was used in the Smith et al. trial and the ongoing ADOPT‐PGx trial. PGx‐ACT utilized the new phenotype translation and, therefore, had a broader definition of IM. Thus, alternative therapy was recommended for participants with a CYP2D6 activity score of 1, which was present in nearly a quarter of participants.

This trial had several limitations. First, the trial was not blinded; however, outcome data were collected by a coordinator who was not involved in the clinical care or the clinical intervention. Allocation was not actively communicated to participants, but participants could discover arm allocation. It is unlikely that a placebo effect confounded the analysis, as evidenced by the similar improvements in pain intensity among those with PGx‐aligned care regardless of arm assignment.

Second, despite the implementation approach applying with recommended practices, the implementation methods were unsuccessful in changing prescribing practices in IM/PMs between arms [31, 32, 33, 34]. The median number of enrolled participants for each enrolling provider was 3 [1, 5] over a 2‐year period, suggesting most providers had a limited opportunity to practice applying PGx. This infrequent opportunity may reduce the effectiveness of the intervention methods designed to enable PGx‐guided care. In addition, the trial occurred during the COVID‐19 pandemic, which may have impacted provider engagement. Providers later suggested the expansion of EHR‐based messaging (i.e., inbasket messaging) to notify them of PGx consult note availability for patients with upcoming visits. The CDS alerts were triggered rarely and were overridden in all five instances. We have since consulted with the MedStar National Center for Human Factors in Healthcare to redesign the alerts using human factors engineering principles [35]. Last, a clinical consideration in the recommendations for IMs may have biased PGx‐aligned care between arms to the null: if the provider and patient considered analgesia adequate, providers could continue tramadol, hydrocodone, or codeine therapy or replace the opioid with a nonopioid analgesic and thereby use PGx as part of an opioid‐de‐escalation strategy. While this is clinically appropriate based on the current evidence, it means a lack of any prescribing change in IMs is defined as PGx‐aligned care. However, this limitation is addressed by the analysis of those with at least one analgesic medication change. Although improvements in multiple outcomes were found with PGx‐aligned care, these analyses do not use randomized data and are considered exploratory.

The timing of the baseline visit, the next visit after randomization, was a pragmatic design element to ensure all participants had an opportunity to receive pharmacotherapy modifications. This came at the cost of increased attrition after randomization: 36 (14%) of 253 randomized participants. Thirty (83%) of the withdrawn participants did not have a baseline visit with an associated baseline questionnaire. Secondly, the timing of the baseline visit created a potential scenario where providers could have modified therapy between randomization and the baseline visit. To mitigate this risk: (1) providers were consulted during trial design and indicated they rarely modify analgesic therapy between visits, and (2) participants were withdrawn if their PGx results were available and they did not complete the baseline questionnaire prior to the visit (n = 9).

Last, the population studied may underestimate the value of PGx‐guided opioid prescribing. The trial required participants to be prescribed an opioid. This enabled a more efficient trial design, as a higher proportion of enrolled participants were eligible for the primary analysis than the trial that informed this design [6]. However, this may bias the sample to participants prescribed the enrollment opioid for an extended time, which may select participants satisfied with the enrollment opioid or less likely to change therapy, particularly with ~70% of patients reporting having pain for 5 or more years. These results are most reflective of patients already prescribed tramadol (Table 2); however, it is unknown how long participants were prescribed their enrollment opioid before enrollment or their adherence. The population hypothesized to receive the most benefit from PGx‐guided opioid prescribing is those naïve to tramadol, codeine, or hydrocodone.

This trial has several strengths, including that, to our knowledge, it is the largest randomized trial investigating the effectiveness of PGx testing for the management of chronic pain. Second, it included a diverse sample of participants, including 85 (38%) who self‐identified as Black or African American and 164 (74%) female participants. Third, this trial was operationally efficient and largely a virtual clinical trial: 85% provided consent virtually. Further, among participants who completed the trial, 81% completed all questionnaires electronically. Fourth, it utilized best practices in PGx implementation, including storing PGx results as structured data, creating CDS alerts, providing PharmD consults, delivering provider education, sufficient CYP2D6 allele coverage, incorporation of CYP2D6 inhibitors into phenotype assessments, and obtaining stakeholder input prior to trial initiation [31, 32, 33]. Finally, this pragmatic hybrid‐effectiveness design allowed for a real‐world examination of the effectiveness of PGx testing. Unrealistic trial conditions are a component behind the 17‐year lag between translating research to clinical practice [36]. A clinical trial solely focused on efficacy would lack the underlying implementation effectiveness data that mediate the role of CYP2D6 and clinical response to tramadol, hydrocodone, and codeine.

5. Conclusion

Providing PGx results and recommendations (an asynchronous PGx PharmD consult with supporting CDS alerts and provider education) was not an effective implementation strategy for patients with chronic pain treated with tramadol, hydrocodone, or codeine. However, a secondary analysis identified that prescribing aligned with PGx results was associated with improved pain symptoms and reduced MMEs prescribed. This study highlights the importance of hybrid‐effectiveness trial design. Additional future efforts should identify effective implementation strategies for integrating PGx results into opioid prescribing.

Author Contributions

D.M.S., R.B., and S.M.S., wrote the manuscript; D.M.S., A.D.W., M.B.J., B.N.P., and S.M.S., designed the research; D.M.S., T.A.Y., V.N., A.L., R.W., S.D., and S.Z. performed the research; D.M.S., P.K., and R.H.P., analyzed the data; T.M. contributed new reagents/analytical tools.

Ethics Statement

The study was approved by a single IRB, the IRB at MedStar Health Research Institute, and all procedures were in accordance with the ethical standards of the Declarations of Helsinki. Participants provided physically or electronically signed informed consent prior to any study intervention (e.g., sample collection for PGx testing). Identifiable information was collected, but data were deidentified for analysis and reporting.

Conflicts of Interest

DMS reports research funding to the institution from Kailos Genetics Inc. SMS reports personal Fees for consulting/advisory services/nonpromotional speaking: AstraZeneca, Daiichi‐Sankyo, Genentech/Roche, Sanofi, Merck, Lilly, Chugai Pharmaceutical Co.; Research support (to institution) Genentech/Roche, Kailos Genetics Inc.; Scientific Advisory Board: Napo Pharmaceuticals; Board of Directors: SEAGEN Stock and Stock options (end 12/14/2023), Immunome; Stipend and stock options: Immunome; Other support: Genentech/Roche and AstraZeneca (third‐party writing assistance); In Kind Travel: Seagen, Napo Pharmaceuticals, Sanofi, Daiichi Sankyo, Genentech/Roche, Chugai Pharmaceutical Co. All other authors report no conflicts of interest.

Supporting information

File S1.

Data S1.

File S2.

Data Availability Statement

The data sets generated and analyzed during PGx‐ACT are not publicly available, but summary data are available in the supplement. Regarding data sharing, the informed consent signed by all participants in this trial stated, “…will not be used or distributed for future research studies, even if all of your identifiers are removed.”