Pilot Findings of Pharmacogenomics in Perioperative Care: Initial Results from the First Phase of the ImPreSS Trial

Abstract

Background:

Pharmacogenomics, which offers a potential means by which to inform prescribing and avoid adverse drug reactions, has gained increasing consideration in other medical settings but has not been broadly evaluated during perioperative care.

Methods:

The ImPreSS Trial is a prospective, single-center study consisting of a pre-randomization pilot and a subsequent randomized phase. We describe findings from the pilot period. Patients planning elective surgeries were genotyped with pharmacogenomic results and decision-support made available to anesthesia providers in advance of surgery. Pharmacogenomic result access and prescribing records were analyzed. Surveys (Likert-scale) were administered to providers to understand utilization barriers.

Results:

Of eligible anesthesiology providers, 166/211 (79%) enrolled. 71 patients underwent genotyping and surgery (median 62y; 55% female; average ASA score 2.6; 58 inpatient and 13 ambulatory). No patients required post-operative intensive care or pain consultations. At least one provider accessed pharmacogenomic results before or during 41/71 surgeries (58%). Faculty were more likely to access results (78%) compared to house-staff (41%; p=0.003) and mid-level practitioners (15%)(p<0.0001). Notably, all administered intraoperative medications had favorable genomic results with the exception of succinylcholine administration to one patient with genomically increased risk for prolonged apnea (without adverse outcome). Considering composite prescribing in pre-operative, recovery, throughout hospitalization, and at discharge, each patient was prescribed a median of 35 (range 15–83) total medications, 7 (range 1–22) of which had annotated pharmacogenomic results. Of 2,371 prescribing events, 5 genomically high-risk medications were administered (all tramadol or omeprazole; with 2/5 pharmacogenomic results accessed) and 100 genomically cautionary mediations were administered (hydralazine, oxycodone, and pantoprazole; 61% rate of accessing results). Providers reported that although results were generally easy to access and understand, the most common reason for not considering results was because remembering to access pharmacogenomic information was not yet a part of their normal clinical workflow.

Conclusions:

Our pilot data for result access rates suggest interest in pharmacogenomics by anesthesia providers, even if opportunities to alter prescribing in response to high-risk genotypes were infrequent. This pilot phase has also uncovered unique considerations for implementing pharmacogenomic information in the perioperative care setting, and new strategies including adding the involvement of surgery teams, targeting patients likely to need intensive care and dedicated pain care, and embedding pharmacists within rounding models will be incorporated in the follow-on randomized phase to increase engagement and likelihood of affecting prescribing decisions and clinical outcomes.

INTRODUCTION

Adverse drug reactions can be unpredictable due to multiple causes, including germline genetic factors.¹ During anesthesia and postoperative care, particularly high-risk adverse drug outcomes include malignant hyperthermia, prolonged apnea, sedation, and inadequate pain management.^2–4 Pharmacogenomics is the study of how germline polymorphisms affect a person’s response to medications and is being increasingly recognized as a potentially important factor to guide medication prescribing and decrease adverse drug reactions.^5–10 However, there are barriers to pharmacogenomic application in clinical practice, such as availability of routine genetic tests, difficulty translating results into actionable prescribing information, lack of provider knowledge, and integration into the clinical workflow.^11–13

Our institution has deployed a broad clinical implementation infrastructure to overcome these barriers and enable pharmacogenomic testing and results delivery with decision-support at the point-of-care in various care settings. To our knowledge, there have been few studies to prospectively examine the use of pharmacogenomic information to guide prescribing in the perioperative setting.¹⁴ We recently initiated a prospective, randomized, controlled trial – the ImPreSS Trial: Implementation of Pharmacogenomic Decision Support in Surgery – to evaluate the feasibility and utility of preemptively-obtained pharmacogenomic results to guide perioperative medication decision making.¹⁵ Here, we describe the initial findings from the pre-randomization pilot phase of this study.

The hypothesis of this pilot was that preemptive pharmacogenomic testing and results delivery would be feasible, and that anesthesia providers would consider results a majority of the time to guide prescribing in the perioperative setting. The primary outcome was to assess initial adoption of pharmacogenomic information. The secondary outcomes were 1) to determine the most frequently administered and prescribed drugs with annotated pharmacogenomic information; and 2) to identify challenges and barriers to implementing pharmacogenomic information in the perioperative setting.

MATERIALS AND METHODS

Study Design

The ImPreSS Trial (clinicaltrials.gov NCT03729180, Principal Investigator: Peter H. O’Donnell, date of registration prior to patient enrollment: November 2, 2018) was approved by the University of Chicago Institutional Review Board and written informed consent was obtained from all subjects. The trial is a prospective, randomized, controlled, single-center study consisting of two phases: the pre-randomization pilot period and the subsequent randomized phase.¹⁵ The study design has been previously described; here, we focus on the findings from the initial pilot period. The purpose of this pilot period was to enroll a small cohort of patients to evaluate and refine the process of pharmacogenomic implementation and results delivery prior to enrolling a larger cohort of patients for the randomization phase. All consenting patients were genotyped and results were made available to participating providers via our institutional electronic medical record-embedded results delivery tool, the Genomic Prescribing System (GPS).¹⁶ Throughout the pilot period, the study team evaluated mediators and moderators of pharmacogenomics adoption by providers. The team regularly met to plan new strategies and make changes to the original process to improve adoption in the randomized phase. The primary outcome was to assess initial adoption of pharmacogenomic information.

Provider Enrollment

Department of Anesthesia and Critical Care providers [faculty physicians, house-staff physicians consisting of fellows and residents, mid-level practitioners consisting of certified registered nurse anesthetists and advanced practice nurses, and pharmacists] were approached for enrollment through a process of stakeholder engagement and informed consent. Stakeholder engagement was initiated through presentations at grand rounds, departmental meetings, and small groups to engage and consent providers. Those not recruited through these mechanisms were approached individually. All consenting providers allowed us to evaluate their prescribing behaviors (post-hoc) and assess their perceptions of pharmacogenomics.

Prescribing Evaluations

As part of the pre-planned analysis, all prescribing events and drug administrations were captured across the various stages of the perioperative care period: 1) prior to and during surgery in the operating room; 2) after surgery in the recovery room; 3) throughout inpatient stay (if the patient was admitted to the hospital) including therefore instances where pain consultations might occur; and 4) at the time of discharge. Participating providers were given access to GPS and pharmacogenomic results and gave permission for their prescribing behaviors to be analyzed; however, they were never instructed on how to administer or prescribe medications by the study team. Therefore, pharmacogenomic test results could be delivered to any participating providers on the care team during the course of the patients’ surgery encounter throughout any of these settings. Non-participating providers were not given access to GPS and pharmacogenomic results for their participating patients.

Patient Enrollment

Patients were screened for eligibility and approached for enrollment during their Anesthesia Perioperative Medicine Clinic pre-surgery appointment by a research coordinator between July 31 and December 13, 2019. The eligible population consisted of adults (>18 years) with an elective inpatient or ambulatory surgery scheduled at the University of Chicago at least two weeks from the date of enrollment. Eligibility criteria for patients did not require that they were assigned to an enrolled study provider at the time of consent and for their clinical care. Assignment of patients to their clinical provider was performed solely according to the department clinical rotation schedules, without respect to the study. The two-week limit was initially established to provide adequate turnaround time for delivery of results to participating providers prior to the patients’ surgery. Exclusion criteria can be found in the Supplementary Methods. All medications prescribed or administered (“total medications”) were recorded and analyzed during the patients’ surgery (in the operating room), during admission (from the time the patient is admitted into the hospital to the time they are discharged, excluding the operating room), and at discharge.

Preemptive Genotyping

Patients who consented to the study provided a sample of whole blood for genotyping in conjunction with other standard-of-care blood draws. Blood samples were sent to our Clinical Laboratory Improvement Amendments-certified and College of American Pathologists-accredited Clinical Pharmacogenomics Laboratory for genotyping across a panel of actionable germline variants known to affect drug disposition, response, or toxicity. The pre-determined variants of interest for genotyping were included based on currently available scientific evidence to support their clinical actionability and have been previously characterized^8,17 (see also Supplementary Digital Content Table 1 for a list of tested genes, variants, clinical implications, and literature references supporting the clinical recommendations). Additional details about genotyping can be found in the Supplementary Methods.

Genomic Prescribing System (GPS) and Best Practice Advisory

The GPS is a secured, protected-access, results portal, which has been integrated into our institutional electronic medical record to deliver pharmacogenomic results to participating study providers at the point-of-care. The development of the GPS has been previously described and employed in our other institutional pharmacogenomics implementation studies.^5,16 The GPS utilizes the traffic light iconography to designate a patient’s pharmacogenetic risk – a red ‘warning’ light indicates a high risk for toxicity or non-response to a medication; a yellow ‘caution’ light indicates an increased risk for toxicity or non-response to a medication; and a green ‘favorable’ light indicates the patient is expected to have good response based on their genetics (see Supplementary Digital Content Figure 1). GPS has the capability to deliver patient-specific results for medications the patient is currently taking as well as a comprehensive list of results for medications providers are considering for the patient (Supplementary Digital Content Figure 2). Providers have the option to click on each medication to view the extended clinical decision support summary for the specific patient. Providers’ GPS click-log records were recorded and analyzed in this study to evaluate provider rate of access to pharmacogenomic test results. In the context of this study, “pharmacogenomic medications” consisted of medications annotated with pharmacogenomic information in GPS.

Our institutional Best Practice Advisory was built and integrated into the electronic medical record to alert providers about patients enrolled in the trial (see Supplementary Digital Content Figure 3). The alert contains a hyperlink to the patient-specific pharmacogenomic test results. Participating providers received an alert for enrolled patients upon opening the patient’s chart for the first time, when prescribing a medication with pharmacogenomic information, or when verifying a pharmacogenomic medication (pharmacists). Although all participating providers received alerts for all enrolled patients, providers were not mandated to click on the hyperlink for GPS – providers could bypass the alert by acknowledging the alert and continuing to the patient’s electronic medical record without accessing GPS.

Provider Experience Surveys

Surveys consisting of Likert-scale questions were administered to providers after each surgery encounter. Each participating provider on the patient’s care team was invited by email to answer an encounter-specific on-line survey with a unique link. Providers were given separate ‘GPS Log-In’ and ‘No GPS Log-In’ surveys depending on if they accessed GPS for that surgery. Study data were managed using REDCap, a secure, web-based application.¹⁸

Statistical Analysis

Provider and patient enrollment are presented as enrollment rates (%). McNemar’s test was performed to examine the relationship between provider type and rate of accessing GPS. Each of the three provider types were independently compared with each of the others (e.g., faculty were compared to house-staff; faculty were compared to mid-level practitioners; and house-staff were compared to mid-level practitioners). A p value of < 0.05 was considered statistically significant. All other data are presented using descriptive statistics.

The pilot phase was designed to include patients enrolled for approximately the first 6 months, or when 100 patients were reached (whichever occurred first). The power calculations supporting the 1800-patient sample size for the follow-on randomized phase of the study are described in the Supplementary Methods.

RESULTS

Providers

Of 211 eligible Department of Anesthesia and Critical Care providers, 166 (79%) were enrolled (see Supplementary Digital Content Table 2). The median years in practice post-training was 8 (range 1–32; house-staff were excluded from this calculation). All 7 pharmacists were specialists in the intensive care unit and were not involved in the care of patients in the operating room.

Patients

A total of 473 patients were pre-screened, 18 of whom did not meet inclusion and exclusion criteria and were not considered for enrollment. Of 455 patients that passed the pre-screen, 303 (67%) were not approached for participation, the primary reason being their surgery date was <2 weeks away. A total of 80 patients were eligible and approached, and 75 of those (94%) consented to the study (see Supplementary Digital Content Figure 4). One enrolled patient was not genotyped because no blood sample was available and three others had their surgeries cancelled, leaving 71 evaluable surgeries (58 inpatient, 13 ambulatory). None of the 71 evaluable surgery encounters resulted in the patient being admitted to the intensive care unit or receiving a pain consult. Pharmacogenomic result access patterns for these 71 surgeries were utilized to evaluate and refine the processes of implementation before the randomized phase. Patient demographics are summarized in Table 1. In this pilot phase, there were no enrolled patients that were not cared for by at least one participating provider.

Table 1.

Patient and Surgery Demographics

Evaluable surgeries (N)	71
Age in years, [median (range)]	62 (23–93)
Female, [n (%)]	39 (55%)
Race, [n (%)]
White	47 (66%)
Black or African-American	21 (30%)
Asian	1 (1%)
More than one race/Other	2 (3%)
Education level, [n (%)]
Did not complete high school	1 (1%)
High school graduate/GED	14 (20%)
Attended some college or university	16 (23%)
College or university graduate	25 (35%)
Advanced degree	15 (21%)
Average number of medications at enrollment, [mean (range)]	8.6 (1–33)
Average number of comorbidities/medical problems, [mean (range)]	7.0 (2–23)
Average ASA physical status score	2.6
Average length of admission, [mean (range)]	2.9 (1–30)
Primary Diagnosis, [n (%)]
Diseases of the musculoskeletal system and connective tissue	38 (54%)
Neoplasms	13 (18%)
Diseases of the genitourinary system	5 (7%)
Endocrine, nutritional, and metabolic diseases	5 (7%)
Diseases of the digestive system	3 (4%)
Diseases of the eye and adnexa	2 (3%)
Factors influencing health status and contact with health services	2 (3%)
*Others (n=1)	3 (4%)
Surgery location, [n(%)]
Inpatient	58 (82%)
Ambulatory	13 (18%)
Surgery type, [n (%)]
Lower joints – replacement or fusion	32 (45%)
Female reproductive system – excision or resection	7 (10%)
Endocrine system – resection	5 (7%)
Upper joints – fusion	4 (6%)
Skin and breast – replacement or resection	4 (6%)
Male reproductive system – resection	3 (4%)
Peripheral nervous system – release	2 (3%)
*Others (n=1)	14 (20%)

Primary diagnoses and surgery types were characterized by mapping International Statistical Classification of Diseases and Related Health Problems (ICD) and Common Procedure Technology (CPT) codes, respectively, to Hospital Account Record (HAR) codes.

^*All other diagnoses and surgery types where n=1 were classified together as ‘Others’.

ASA: American Society of Anesthesiologists.

Pharmacogenomic Result Access

At least one participating provider accessed patient-specific pharmacogenomic results via GPS prior to or during 41/71 surgeries (58%). The rate and timing of pharmacogenomic results being accessed via GPS are summarized in Table 2. When results were viewed by a participating provider, two thirds of results (66%) were viewed prior to any medication being administered in the perioperative setting. Faculty were more likely to access GPS (n=32, 78%), compared to house-staff (41%, p=0.003) and mid-level practitioners (15%, p<0.0001). The rate of pharmacogenomic results access was generally constant over time during the pilot period (Supplementary Digital Content Table 3), after accounting for variance in the early estimates (accessed for 7 of the first 10 patients [70%], 12 of the first 20 patients [60%], 23 of the first 40 [58%], and 41/71 at the end of the pilot [58%]).

Table 2.

Pharmacogenomic Results Access via GPS

	N (%)
Evaluable Surgery Encounters	71
Unique Evaluable Surgery Encounters with GPS Access	41 (58%)†
Faculty Access Rate	32 (78%)*
House-staff Access Rate	16 (39%)**
Mid-level Practitioners Access Rate	5 (12%)
Timing of GPS Access
Day Before Surgery	27 (66%)
Day of Surgery, Pre-operatively	14 (34%)
During Surgery	11 (27%)
Day of Surgery, Post-operatively	3 (7%)
Post-op Day #1	1 (2%)

^†Note that more than one provider type could have accessed GPS at the same encounter, which is why the different provider type accessions do not sum to the overall total.

^*Faculty were more likely to access GPS, compared to house-staff (p<0.003) and mid-level practitioners (p<.0001).

^**House-staff were more likely to access GPS, compared to mid-level practitioners (p=0.02)

GPS: Genomic Prescribing System.

Top Pharmacogenomic Medications Administered or Prescribed

Patients were administered a median of 13 (range 4–20) total medications and 1 (range 0–3) pharmacogenomic medication during surgery (while in the operating room). The top three most frequently administered medications with pharmacogenomic results during surgery were sevoflurane, desflurane, and succinylcholine (60%, 27%, and 10% of surgeries, respectively). However, all administered surgery medications had favorable genomic (GPS green light) results with the exception of one patient who was administered a genomic yellow light medication, succinylcholine, with no significant adverse outcome.

We next examined the impact of pharmacogenomic information on medication prescribing and administration across the perioperative period (pre-operative, intra-operative, recovery area, inpatient stay, and at discharge). When considering this broader perspective, patients were prescribed or administered a median of 35 (range 15–83) total medications and 7 (range 1–22) pharmacogenomic medications. Of these, patients were administered a median of 13 (range 1–36) total medications and 3 (range 0–9) pharmacogenomic medications during the recovery period and post-surgery hospitalization. Patients were prescribed a median of 11 (range 3–32) total medications and 4 (range 0–11) pharmacogenomic medications at the time of discharge. The most commonly administered or prescribed medications with pharmacogenomic information are shown in Figure 1.

Figure 1. Most frequently administered and prescribed pharmacogenomic medications during surgery encounters.

The medications are ranked by total medications prescribed or administered (in the operating room), during post-surgery hospitalization (admission), and at discharge. Out of 2,371 prescribing events, there were 5 total high-risk medications administered (all tramadol or omeprazole; 2/5 rate of results access among providers) and 100 cautionary mediations administered (most commonly hydralazine, oxycodone, and pantoprazole; 61% rate of access among providers). The top pharmacogenomic medications administered in the operating room were sevoflurane, desflurane, and succinylcholine (60%, 27%, and 10% of surgeries, respectively). The top pharmacogenomic medications administered during admission were oxycodone, tramadol, and aspirin (49%, 42%, and 41% of all admissions, respectively). The top pharmacogenomic medications prescribed at discharge were aspirin, tramadol, and oxycodone (56%, 56%, and 51% of all discharges, respectively).

Top Medications with Potential Pharmacogenomic Results in GPS

All providers who access GPS land on the “All Drugs (compact)” page (see description in Supplementary Digital Content Figure 2). This page lists all the pharmacogenomic medications, organized by traffic light signals, with a hyperlink to the detailed clinical decision support for each. This section therefore summarizes the medications with potential pharmacogenomic results available.

The top medications with increased pharmacogenomic risk results (red and yellow lights) that were available in GPS across the patient encounters in this study are shown in Figure 2. Although there were 25 yellow light results for mivacurium for increased risk of prolonged apnea, only one provider viewed the detailed clinical decision support, likely because mivacurium is not routinely utilized at our institution. There were also 20 yellow light results for diazepam based on CYP2C19 association with prolonged emergence time from anesthesia, with 10 having a provider access the result in GPS, and one provider viewing the detailed clinical decision support (diazepam was not administered or prescribed to this patient). Of the remaining potential yellow light results for diazepam, the medication was administered or prescribed three times during the hospitalization or at discharge to two patients (but none administered in the operating room by an anesthesia provider).

Figure 2. Top medications with potential pharmacogenomic results available that could have been viewed in GPS.

At least one provider accessed GPS in 57.7% of all surgeries, compared to 42.3% of surgeries where no providers did. The number (and percentages) of red, yellow, and green light medications are shown for surgeries with and without access to GPS. The top medications with increased pharmacogenomic risk (red and yellow lights) with results available in GPS are listed (with the number of surgeries they were administered in, in parenthesis) for both groups; green light medications are not listed. For example, there were 25 yellow light results for atenolol in which at least one provider accessed logged into GPS, and there were 15 yellow light results for mivacurium where no providers accessed GPS.

There were 27 red light and 373 yellow light medications with potential results available in GPS where at least one provider accessed the patient’s composite results (Left panels). Of these, 5/27 (codeine and azathioprine being the highest viewed) and 21/373 (atenolol and aspirin being the highest viewed) had at least one provider view the detailed clinical decision support summary, respectively.

There were 36 red light and 317 yellow light medications with potential results available in GPS where there were no provider access (Right panels), which are considered potential “missed opportunities” to consider pharmacogenomics when prescribing.

*Medications most frequently clicked on and viewed the detailed clinical decision support summary by at least one provider. GPS = Genomic Prescribing System.

Provider experience surveys

For providers who accessed GPS, there were 17 completed surveys (35% response rate). For those who did not access GPS, there were 12 completed surveys (20% response rate). Responses for select questions are summarized in Table 3 and Table 4. There were no significant differences in responses regarding patients, medication safety and efficacy, and the medication selection process between providers who accessed GPS and those who did not.

Table 3.

Provider Experience Survey Responses from Encounters Where the Provider Accessed GPS Results

Survey Responses from Encounters where Providers Accessed GPS Results
	Agree Strongly N (%)	Agree Somewhat N (%)	Unsure N (%)	Disagree Somewhat N (%)	Disagree Strongly N (%)
The complicated nature of this patient limited my ability to personalize care.	0 (0%)	0 (0%)	0 (0%)	9 (53%)	8 (47%)
This patient is likely to experience side effects as a result of the medications I prescribed or administered.	0 (0%)	1 (6%)	1 (6%)	6 (35%)	9 (53%)
The medications I prescribed or administered carried a significant risk for harm to this patient.	1 (6%)	5 (29%)	0 (0%)	3 (18%)	8 (47%)
I used individualized information when choosing medications for this patient.	3 (18%)	6 (35%)	3 (18%)	4 (24%)	1 (6%)
I benefitted from clearly-defined guidelines or intuitional processes to guide my care for this patient.	1 (6%)	1 (6%)	7 (41%)	6 (35%)	2 (12%)
I consulted with other providers to arrive at my medication choices for this patient.	1 (6%)	0 (0%)	1 (6%)	8 (47%)	7 (41%)
The practice patterns of my peers influenced the prescribing/administering approaches I took for this patient.	1 (6%)	5 (29%)	1 (6%)	6 (35%)	4 (24%)
Survey Responses from Encounters where Providers Did Not Accessed GPS Results
	Agree Strongly N (%)	Agree Somewhat N (%)	Unsure N (%)	Disagree Somewhat N (%)	Disagree Strongly N (%)
The complicated nature of this patient limited my ability to personalize care.	1 (8%)	0 (0%)	1 (8%)	5 (42%)	5 (42%)
This patient is likely to experience side effects as a result of the medications I prescribed or administered.	1 (8%)	1 (8%)	1 (8%)	7 (58%)	2 (17%)
The medications I prescribed or administered carried a significant risk for harm to this patient.	2 (18%)	4 (36%)	1 (9%)	3 (27%)	1 (9%)
I used individualized information when choosing medications for this patient.	1 (8%)	7 (58%)	0 (0%)	4 (33%)	0 (0%)
I benefitted from clearly-defined guidelines or intuitional processes to guide my care for this patient.	0 (0%)	4 (33%)	3 (25%)	3 (25%)	2 (17%)
I consulted with other providers to arrive at my medication choices for this patient.	1 (8%)	3 (25%)	1 (8%)	4 (33%)	3 (25%)
The practice patterns of my peers influenced the prescribing/administering approaches I took for this patient.	2 (17%)	5 (42%)	2 (17%)	2 (17%)	1 (8%)

Table 4.

Provider survey responses to questions regarding their experiences with the Genomic Prescribing System (GPS).

Survey Responses for Providers Who Logged into GPS
	Agree Strongly N (%)	Agree Somewhat N (%)	Not Sure N (%)	Disagree Somewhat N (%)	Disagree Strongly N (%)
The GPS information was relevant to my patient’s situation.	1 (6%)	4 (24%)	4 (24%)	5 (29%)	3 (18%)
I decreased the risk of side effects for my patients by utilizing GPS information.	0 (0%)	4 (24%)	5 (29%)	2 (12%)	6 (35%)
Accessing the GPS was a natural part of my clinical workflow.	1 (6%)	4 (24%)	3 (18%)	6 (35%)	3 (18%)
I had enough time to evaluate the results presented in GPS.	5 (29%)	7 (41%)	3 (18%)	1 (6%)	1 (6%)
The GPS provided straightforward prescribing guidance.	4 (24%)	7 (41%)	5 (29%)	1 (6%)	0 (0%)
The GPS was simple to use.	5 (29%)	8 (47%)	3 (18%)	1 (6%)	0 (0%)
The clinical decision support summaries were too complicated or technical.	0 (0%)	0 (0%)	1 (6%)	7 (41%)	9 (53%)
Survey Responses for Providers Who Did Not Log into GPS
	Agree Strongly N (%)	Agree Somewhat N (%)	Not Sure N (%)	Disagree Somewhat N (%)	Disagree Strongly N (%)
I did not have time to log into the GPS.	4 (33%)	1 (8%)	2 (17%)	1 (8%)	4 (33%)
I did not log into the GPS due to the complexity of this particular patient.	1 (8%)	0 (0%)	0 (0%)	3 (25%)	8 (67%)
I did not remember this patient was in the study.	4 (33%)	3 (25%)	0 (0%)	3 (25%)	2 (17%)
I did not remember how to access the GPS.	2 (17%)	5 (42%)	1 (8%)	1 (8%)	3 (25%)
I have not found the GPS to be a useful tool in my practice because the information provided is typically not clinically relevant.	1 (8%)	1 (8%)	7 (58%)	2 (17%)	1 (8%)
I have not found the GPS to be a useful tool in my practice because the information provided is typically not supported by enough evidence.	1 (8%)	0 (0%)	8 (67%)	1 (8%)	2 (17%)
I have not found the GPS to be a useful tool in my practice because I have not been able to adapt the system to fit my needs.	1 (8%)	3 (25%)	5 (42%)	2 (17%)	1 (8%)
For this patient, I did not think there would be an obvious benefit from incorporating genetic information.	0 (0%)	2 (17%)	7 (58%)	2 (17%)	1 (8%)
Changing the drug or dose would have meant deviating from the accepted standard of care, so pharmacogenomic information was not relevant to my decision-making.	0 (0%)	1 (8%)	3 (25%)	5 (42%)	3 (25%)

About 70% of survey respondents who accessed GPS said they had enough time to evaluate the pharmacogenomic results. Approximately 64% of respondents said the GPS provided straightforward prescribing guidance, 76% said it was simple to use, and none felt like the clinical decision support summaries were too complicated or technical.

For providers who did not access GPS, the top reasons for not accessing were that they either did not remember the patient was in the study (58%) or they did not remember how to access GPS (58%). About 58% and 67% of respondents were not sure whether information found in the GPS was clinically relevant or supported by enough evidence, respectively. Furthermore, 58% of responders were not sure whether their patient would benefit from incorporating genetic information.

DISCUSSION

This pilot study demonstrated that while anesthesia providers showed interest in the potential to apply pharmacogenomic information in the perioperative setting, patient-specific results access was intermittent, and moreover, there were relatively few instances thus far of high-risk pharmacogenomic results to alter prescribing. Nevertheless, when considering the entire peri-surgical encounter including prescribing in the pre-operative setting, recovery area, during post-surgery hospitalization, and at discharge, patients were prescribed or administered a median of 7 unique medications with associated pharmacogenomic annotations, most of them genomically congruent, representing affirming relevance for clinical care. Relatedly, when results were viewed by a participating provider, two-thirds of results were viewed prior to any medication being administered in the perioperative setting, an important consideration for clinical workflows. We plan to formally explore the potential applicability and impact of preemptive pharmacogenomic genotyping in the follow-on randomized portion of this trial, which is now ongoing.

Important implementation lessons were among the key takeaways from this pilot. The most common barrier to considering pharmacogenomic results was that doing so was not yet second nature—that is, in the instances where results were not accessed, providers said they either failed to remember that pharmacogenomic results were available or were unsure how to access the results in GPS. This likely reflects the fact that accessing pharmacogenomic results outside the context of this study remains non-standard for providers, and also probably reflects the busy workflow of the operating room. Failing to recall that results were available or not knowing how to find those results in the medical record was reported despite integration of our results into the electronic medical record through Best Practice Advisory alerts which informed providers of available pharmacogenomic results in all patients. Studies have shown that although alerts can improve protocol compliance to lessen drug-related adverse events, the rate of alert overrides increases over time possibly due to alert fatigue.^19,20 Ancker et al. showed a marked decrease in a clinician’s likelihood of accepting alerts with repeated reminders for the same patient.²¹ To combat potential alert fatigue, we will consider limiting our alerts to only those with genomic high risk (GPS designated red and yellow light) results so as to also hopefully increase the consideration for more relevant pharmacogenomic results. Nevertheless, the overlay of human behavior is a major barrier to adoption of any new technology, and in addition to memory tools, other implementation solutions may need considered. Therefore, we are also implementing strategies to promote and sustain adoption of pharmacogenomics by embedding dedicated personnel (e.g., a pharmacist with pharmacogenomics expertise) within the clinical care settings when anesthesiologists are considering medication decisions, except within the operating room itself which is likely infeasible.

Other factors that may influence the perceived utility of pharmacogenomic information include the prevalence and the clinical impact of the most dangerous genetic-linked phenotypes, and the availability of effective and safe alternative medications. The relatively low observed rate of high risk (or genomically discordant) results in our pilot is not, however, interpreted as a barrier to larger genomic implementation in the perioperative setting. In fact, most previous pharmacogenomic implementations outside of anesthesia and critical care apply only to a small portion of the overall tested population. Abacavir, for instance, has been rapidly incorporated into routine medical practice among infectious disease practitioners despite a relatively low population prevalence of genomically at-risk patients.²² Similarly, screening strategies in oncology for patients at risk for potentially fatal toxicity from fluoropyrimidine anti-cancer drugs have shown that preemptive screening not only significantly alters rates of adverse events but also is cost-effective, despite the fact that only ~1–8% of patients carry at-risk genotypes.^23,24 The same idea may hold true for perioperative care. For example, the prevalence of malignant hyperthermia caused by polymorphisms in the ryanodine receptor 1 (RYR1) is <1%, but the potential clinical implication for administering volatile anesthetics or depolarizing muscle relaxants in these high-risk individuals may be fatal.² Furthermore, patients who are either heterozygous or homozygous for the high-risk variant in the butyrylcholinesterase (BCHE) gene have been shown to have an increased risk of prolonged apnea (from anywhere between minutes to several hours).²⁵ This information may be clinically meaningful especially to anesthesia providers in the ambulatory surgery setting, where there is a high number of surgeries performed daily and there is thus a high need for rapid patient turnaround time.

In this pilot, no patients required intensive care or postoperative pain consultations—two settings where the potential relevance of pharmacogenomics is likely to be considerably higher because of the types of medications being prescribed there. Several commonly-used opioids (oxycodone, tramadol, and codeine) are metabolized by CYP2D6, for example, and a significant body of accumulated evidence suggests that patients with high-risk variants in the CYP2D6 gene may benefit from consideration of pharmacogenomic information to guide prescribing of pain medications, especially at discharge²⁶. Separately, stress-induced gastric ulcers (relevant to the intensive care unit) may be impacted by CYP2C19 metabolism of proton pump inhibitors²⁷. And cardiovascular medications frequently used in the intensive care setting have intriguing pharmacogenomic relevance⁸. These, and other medications of particular relevance to intensive care, are highlighted in Supplementary Table 1. During the pilot phase we could not report changes-over-time in individual providers’ access of pharmacogenomic results as there were very few individual providers that had more than one opportunity to access results for their individual patients (there were 166 enrolled providers and only 71 evaluable surgeries in the pilot). This is exactly the type of data that we will track in the randomized phase of the study that will include 1800 patients, and we expect that provider engagement may actually increase with repeated opportunities for use.

There were several limitations to this study. First, we did not expect that no patients requiring intensive care would be enrolled, and our unplanned over-recruitment of patients undergoing relatively straightforward orthopedic surgeries likely led to this finding. To address this limitation, we have already begun to target recruitment in the randomized portion of our study of patients undergoing higher-risk surgeries (thus more likely needing intensive care and pain consultations), for whom pharmacogenomics may have an even greater benefit. This unintended enrollment invariance likely was due to our prior criterion requiring patients to have surgeries scheduled at least two weeks in advance, which resulted in the high number of orthopedic surgeries in this cohort and limited our ability to study potential differences in adoption across different procedures. To address this, our lab has decreased genotyping test turnaround time to one week, and we have adapted enrollment to target more diverse surgery types in the randomization phase. Second, we learned that at our institution like many institutions, surgery providers generally prescribe many of the post-operative medications and the discharge medications for surgery patients. In this pilot phase, we found that patients were prescribed a median of 11 total medications (with a median of 4 pharmacogenomic medications) at the time of discharge, suggesting a high potential applicability for delivering pharmacogenomic results to these providers at the time of discharge. Surgery providers were not initially enrolled to the study during the pilot phase and these were potential missed opportunities to implement pharmacogenomic information to guide prescribing. To address this, we are now targeting surgery teams for enrollment in the randomized phase. Separately, modest provider survey response rates (although on par with most physician survey-based research) were observed, and thus the survey results might not be representative of all providers. Relatedly, during the pilot phase we could not report changes-over-time in individual providers’ access of pharmacogenomic results as there were very few individual providers that had more than one opportunity to access results for their individual patients (there were 166 enrolled providers and only 71 evaluable surgeries in the pilot). This is exactly the type of data that we will track in the randomized phase of the study that will include 1800 patients, and we expect that provider engagement may actually increase with repeated opportunities for use, consistent with what has been observed with respect to pharmacogenomic likelihood of adoption over time in other clinical settings²⁸ and unpublished data from our inpatient study²⁹. Lastly, we acknowledge that there are limitations to using pharmacogenomics as a tool to provide medication prescribing in general, as there are many steps required in the expression of an adverse phenotype, including potentially regulation of protein synthesis from mRNA, histone regulation of gene expression, and metabolic derangements influencing activity of the protein besides the presence or absence of variants. Therefore, future studies may additionally include transcriptomics, metabolomics, and proteomics to more fully characterize potential medication inefficacy or adverse drug effects.

In conclusion, data from the pilot phase of the ImPreSS Trial suggested that while anesthesia providers were interested in the potential to apply pharmacogenomic information in the perioperative care setting, the rate of access to pharmacogenomic information was intermittent, and there were few occasions of high-risk genotypes to prompt changes in provider prescribing. Learnings from this pilot phase resulted in implementation of new strategies in the follow-on randomized phase to promote increased provider engagement and adoption of pharmacogenomic information, including involvement of surgery teams and identification of higher-risk patients more likely to receive pharmacogenomically-annotated medications such as patients needing post-operative intensive care or requiring pain consultations.

Supplementary Material

Supplementary Table 1

Supplementary Table 1 Genotyping Results with Potential Clinical Implications for the Pilot Population. Study subjects were genotyped across a panel which included the above listed genes and variants. For each corresponding medication, the results and clinical implications were made available to study providers in the Genomic Prescribing System (GPS), an electronic medical record-embedded results delivery tool. UM: ultrarapid metabolizers; NM: normal metabolizers; IM: intermediate metabolizers; PM: poor metabolizers; NSAID: non-steroidal anti-inflammatory drug; OR: odds ratio; min: minutes; hr: hour; H.Pylori: helicobacter pylori; GERD: gastroesophageal reflux disease; SBP: systolic blood pressure; HR: hazards ratio; LVEF: left ventricular ejection fraction; MAP: mean arterial pressure; MI: myocardial infarction

Supplementary Table 3

Supplementary Table 3 Interval and Cumulative GPS Results Access Rates. Interval access rates of GPS results for every 10 patients and the cumulative (total) access rates for the entire pilot phase. GPS: Genomic Prescribing System

Supplementary Figure 4

Supplementary Figure 4 Patient enrollment flow diagram. Patients were screened for eligibility and approached for enrollment during their Anesthesia Perioperative Medicine Clinic pre-surgery appointment by the clinical research coordinator. The eligible patient population consisted of adult (>18 years) patients with an elective inpatient or ambulatory surgery scheduled at the University of Chicago for at least two weeks from the date of enrollment. The two-week limit was initially established to provide adequate turnaround time for genotyping and delivery of results to participating providers prior to the patients’ surgery. Exclusion criteria included 1) patients who have undergone, or are being actively considered for, liver or kidney transplantation, 2) patients with known active or prior leukemia, 3) patients enrolled in another pharmacogenomics study, and 4) inability to understand and give informed consent to participate.

Supplementary Figure 1

Supplementary Figure 1 Sample detailed clinical decision support summary. This clinical decision support for tramadol in GPS consists of a traffic light signal, a summary of the patient’s genetic risk for toxicity or non-response to a medication, research studies to support the clinical decision support and their corresponding literature references, a list of pharmacogenomic alternatives, and a level of evidence for the clinical decision support based on published literature.

Supplementary Figure 2

Supplementary Figure 2 Sample view of all the potential patient-specific pharmacogenomic results in GPS. All providers who access GPS for a specific patient land on the “All Drugs (compact)” page, as shown here. This page summarizes all of the pharmacogenomic medications (annotated with pharmacogenomic information) with results for that specific patient, organized by traffic light signals. Each medication name has a hyperlink, which will lead to the detailed clinical decision support summary for that medication (Supplementary Figure 1). Therefore, while providers can view the summarized list of pharmacogenomic medications, clicking on each hyperlink to view the detailed clinical decision support summary is at the discretion of the provider. *Patient name, medical record number, and date of birth are fictionalized.

Supplementary Figure 3

Supplementary Figure 3 Best Practice Advisory. Best Practice Advisories are integrated into our institutional electronic medical record and notifies providers when a patient they are caring for is enrolled in the trial and contains a hyperlink to the patient-specific pharmacogenomic test results for that patient. Participating providers receive an alert upon first time opening the patient’s chart, when prescribing a medication with actionable pharmacogenomic information, or when verifying a medication with pharmacogenomic information (this is only relevant to participating pharmacists). Although all participating providers received alerts for all enrolled patients, providers were not mandated to click on the hyperlink for GPS – providers could bypass the alert by acknowledging the alert and continuing to the patient’s electronic medical record without accessing GPS.

Supplementary Table 2

Supplementary Table 2 Summary of Study Providers. Total number of eligible and enrolled providers in the Department of Anesthesia and Critical Care and their average years in practice. *Average years in practice post-training.

KEY POINTS.

• Question: What is the clinical feasibility of integrating pharmacogenomics into the perioperative setting?

• Findings: Pharmacogenomic information was preemptively obtained and made available in advance of surgery over a 6-month pilot period, and was considered by anesthesia providers during 58% of the 71 evaluable surgeries with a median of 7 pharmacogenomically informed medications prescribed per patient.

• Meaning: We identified unique considerations for implementing preemptive genotyping in the perioperative setting to deliver pharmacogenomics results to anesthesia providers to inform medication administration.

Glossary of Terms

• GPS

Genomic Prescribing System

• ImPreSS Trial

Implementation of Pharmacogenomic Decision Support in Surgery