PillHarmonics: An Orchestrated Pharmacogenetics Medication Clinical Decision Support Service

Keywords

pharmacogenetics - decision support system - electronic health records - FHIR

Background and Significance

Pharmacogenetics (PGx) is increasingly important for achieving medication safety goals and efficacy outcomes. Studies show that over half of all primary care patients are exposed to drugs that have potential PGx therapeutic implications.[1] Over 500 drugs have Food and Drug Administration (FDA)-approved biomarker labeling, and 7% of FDA-approved medications, as well as 18% of the 4 billion prescriptions written in the United States per year, are affected by actionable PGx variants (i.e., variants that participate in a drug–gene interaction of potential clinical significance).[2] Nearly all individuals (91%) have at least one known, actionable variant by current Clinical Pharmacogenetics Implementation Consortium (CPIC) guidelines.[3] [4] [5] When 12 pharmacogenes with at least one known, actionable, inherited variant are considered, over 97% of the U.S. population have at least one potentially actionable finding (e.g., presence of a CYP2C19 diplotype containing a *2, *3 or *17 allele)[6] with an estimated impact on nearly 75 million prescriptions.[7] Chanfreau-Coffinier et al[8] have estimated that almost all veterans carry an actionable variant, and more than half had been exposed to a drug that is greatly affected by these variants.

PGx test result findings are most commonly not integrated into the electronic health record (EHR) or represented as other laboratory results, and are only available as nonactionable PDF reports.[9] Structured solutions are emerging,[10] [11] and several groups, including ourselves, are experimenting with the use of Health Level Seven (HL7) Fast Healthcare Interoperability Resources (FHIR)[12] and HL7 CDS Hooks[13] standards to support integration.[14] [15] (CDS Hooks is a new standard that specifies interactions between an EHR and a clinical decision support [CDS] server. Defined events in the EHR trigger a message to the CDS server, which can then gather additional data and execute rules before responding back to the EHR.) A common theme across these efforts is that PGx is implemented apart from other types of medication CDS, leading to overlapping or inconsistent medication alerts. This is problematic in that clinicians may not remember PGx results and their implications at the time of order entry, and may fail to recognize conflicting recommendations (e.g., PGx considerations suggest the need for a higher dose, whereas renal function suggests the need for a lower dose). Over 20 years ago, Hansten et al[16] suggested the need to integrate PGx with other types of medication knowledge. Evidence suggests that a holistic approach to medication CDS can address patient safety issues, such as by juxtaposing conflicting drug recommendations, and decrease alert fatigue through improving precision.[17]

Medication-related adverse events account for over 2 million hospital stays and 3.5 million physician office visits per year.[18] Medication CDS when implemented correctly can have a significant impact on patient safety and drug efficacy.[19] [20] [21] [22] [23] However, there are many challenges with medication CDS implementations.[24] [25] [26] For instance, irrelevant low-risk interactions being surfaced can lead to alert fatigue, conflicting recommendations can leave clinicians frustrated, and poor workflow integration can lead to failure to recognize clinically significant information. Considerable research continues to be devoted to human factors surrounding optimization of information delivery.[27] [28] While most health care systems today can filter medication alerts by severity, suppress various types of alerts, and leverage pieces of patient-specific clinical information (e.g., allergies, weight) to enhance specificity, clinicians are still “generally unsatisfied with the lack of patient specificity and inappropriate context” of such notifications.[27] Adding PGx CDS into an environment that already has many usability challenges risks obscuring the benefits of such alerts.[29] [30]

In response, we built a cloud-based orchestrated medication CDS service known as “PillHarmonics” under National Human Genome Research Institute (NHGRI) grant 1R43HG011832–01A1. As an orchestra blends sounds from multiple instruments to produce a pleasing, integrated experience, an “orchestrated” medication CDS service needs to blend a variety of medication-related knowledge to provide utility to the clinician. Patient safety can be enhanced through minimization of adverse drug events, and alert fatigue decreased via more precise surfacing of aggregated, relevant information. The PillHarmonics service logically blends PGx with other types of medication CDS (e.g., drug–drug, drug–allergy, drug–condition interactions), and presents the clinician with an overall concise summary of the individual pairwise interactions, along with actionable recommendations.

Objectives

In this study, we sought to integrate PGx drug–gene interaction reporting with other categories (e.g., drug–drug, drug–allergy, drug–condition) and deliver a holistic integrated alert to a clinician at the time of drug order entry.

Methods

Our approach was to (1) develop the PillHarmonics service as a functional prototype and (2) evaluate clinician-perceived usefulness of the service. The evaluation was deemed institutional review board (IRB) exempt, not requiring monitoring by an IRB.

Development of the PillHarmonics Service

We developed the PillHarmonics service according to specifications for the CDS Hooks protocol. The system was iteratively designed to respond to a triggering medication order in an EHR and compute an informative recommendation to return to a user. Logic was designed to incorporate other clinical data from the patient (conditions, allergies, current medications, demographics, and laboratory results), as well as genomic data housed in the Genomic Archiving and Communication System (GACS), accessed using FHIR Genomics Operations.[31] Knowledge sources leveraged include First Databank (FDB) cloud-screening services, PharmGKB,[32] and locally curated content. A high-level schematic of the system is shown in [Fig. 1] and further details on the system are in the Results section.

Fig. 1 PillHarmonics service. Components include: (1) EHR: Houses clinical data and serves as CDS Hooks client. The client triggers the service based on a medication order; (2) PillHarmonics service: Housed in a CDS Hooks server, the service computes and orchestrates all interactions, returning the results back to the EHR client; (3) Knowledge Sources: PillHarmonics draws knowledge from FDB, PharmGKB, and locally curated tables; (4) GACS: Genomic Archiving and Communication System that houses genomic data. GACS is accessed using FHIR Genomics Operations. CDS, clinical decision support; EHR, electronic health record; FDB, First Databank.

Evaluation of the PillHarmonics Service

We assessed the perceived usefulness of the service from currently active clinicians by performing a user study simulating order-entry-based invocation of PillHarmonics in synthetic patient scenarios. User feedback was collected via the Technology Acceptance Model (TAM) Questionnaire[33] coupled with recorded, structured interviews summarized using thematic analysis.[34] [35] See [Supplementary Material] (online only) for additional method details.

We recruited practicing clinicians and pharmacists, with experience prescribing clopidogrel and/or tacrolimus, via social media, targeted listserv mailings, and direct outreach. Pre-study sample size calculations suggested the need for at least four participants (see [Supplementary Material] [online only] for detailed sample size calculations). Thirteen respondents consented to participate. Each participant took part in a 1-hour virtual interview that included a review of 10 scenarios specific to the drug they had experience prescribing, representing point-of-care medication decision support. Scenario review was followed by completion of the TAM Questionnaire, and a structured discussion. All sessions were recorded in their entirety. Participants were compensated for their time.

Each of the 10 presented scenarios was a synthetic case (either based on clopidogrel or tacrolimus order entry). The case was presented, followed by a mockup of the resulting rendered CDS Hooks card from the PillHarmonics service. Participants were encouraged to think out loud about their impressions of the card, and ultimately describe what action they would plan to take upon viewing it.

The TAM Questionnaire measures perceived usefulness via a six-item scale, with scores ranging from 1 to 7 per item, with higher scores representing more agreement with the questionnaire statements. Items are designed to assess impressions along six major axes: perceived time savings, job performance, productivity, effectiveness, job ease, and overall usefulness. The specific wording for each of these themes is listed in [Table 1]. TAM's perceived usefulness is significantly correlated with both self-reported current usage (r = 0.63) and self-predicted future usage (r = 0.85).[36] We summarized TAM Questionnaire responses as means and standard deviations, medians and interquartile ranges, and the percent of responses greater or equal to 5 and then greater or equal to 6. Finally, we used a one-sided, one-sample t-test to determine if the mean responses were statistically higher than threshold cut-off values of 4.5 and 5 (where a score of 4 is neutral).

Table 1 Descriptive statistics of the responses to the TAM questionnaire

TAM question	Mean (SD)	Median [25th–75th percentile]	Responses ≥ 5, n (%)	Responses ≥ 6, n (%)
PillHarmonics will allow me to make a complex medication treatment decision more quickly [Work more quickly]	5.2 (0.8)	5 [5–6]	12 (92.3%)	4 (30.8%)
PillHarmonics will enhance patient safety [Job Performance]	5.9 (1.3)	6 [6–7]	12 (92.3%)	11 (84.6%)
PillHarmonics aggregation and presentation of relevant information will save me time [Increase Productivity]	5.2 (1.2)	6 [4–6]	9 (69.2%)	7 (53.8%)
PillHarmonics will make me more confident in my treatment decisions [Effectiveness]	5.3 (1.3)	5 [5–6]	11 (84.6%)	7 (53.8%)
PillHarmonics will make complex decision making easier [Makes Job Easier]	5.4 (1.4)	6 [5–6]	11 (84.6%)	6 (46.2%)
PillHarmonics will be useful in my job [Useful]	5.6 (1.1)	6 [5–6]	12 (92.3%)	7 (53.8%)

We further assessed PillHarmonics with a structured interview protocol. Open-ended questions were asked about what aspects of the service participants liked best and least, what kind of impact they would envision it having on their practice, and what other suggestions they would make for further development. The qualitative feedback was synthesized using a thematic analysis approach. We followed the methods of Braun and Clarke,[35] having three independent reviewers assess all interview content, codify findings, and jointly come to consensus on emergent themes.

Results

PillHarmonics Service

Under this project we successfully developed a working prototype of the PillHarmonics service. A video demonstration of a CDS Hooks client application that calls the PillHarmonics service can be viewed at https://vimeo.com/820674996. EHRs differ considerably in their CDS Hooks card rendering capabilities, and the HTML renderings shown in this video are examples of one possible rendering.

CDS Hooks is a client-server decision support model where the EHR serves as the client and triggers the service, and a cloud-based CDS engine is the server that receives the triggering order, processes the logic, and returns a response to the user. This response comes in the form of a “card,” which can be informational with display text only, or a suggestion card with actionable recommended orders. The specification describes attributes of each card, which include required “indicator” and “summary” fields with optional “detail” field. The “indicator” categorizes the severity or urgency of the content, and the “summary” is a brief notification message less than 140 characters. The “detail” field is additional text which can have additional formatting if desired.

The service is triggered by a drug order in the EHR. This prompts the EHR to send the ordered drug and prefetched FHIR-formatted clinical data (demographics, conditions, medications, allergies, and specific laboratory results) to the PillHarmonics service. The service then infers hepatic and renal function, using a combination of conditions and laboratories, before determining pairwise interactions. FDB CloudConnector API is used to determine drug–drug, drug–allergy, and drug–condition alerts. Drug–renal, drug–hepatic, and drug–race alerts are derived from locally curated content. Drug–gene interactions are based on genotype–phenotype correlations, derived from PharmGKB and housed locally in GACS. To determine PGx drug–gene interactions, the PillHarmonics service gathers a patient's genotype and drug metabolism phenotype from GACS using FHIR Genomics Operations[31] (FHIR Genomics Operations extend the capabilities of the base FHIR Genomics Implementation Guide[12] enabling advanced search scenarios). Having obtained all pairwise drug alerts, the PillHarmonics service then creates the CDS Hooks card that will be sent back to the EHR.

The card's “indicator” field is set to the highest severity seen in any of the pairwise interactions. The card's “detail” field contains all identified pairwise interactions. For each interaction, we provide a normalized criticality, and a normalized “effect” that indicates how the interaction alters the efficacy or potential toxicity of the newly ordered drug or a drug the patient is already taking. In some cases, a drug the patient is already taking also has a drug–gene interaction of its own, further potentiating or attenuating the drug–drug interaction with the ordered drug (sometimes referred to as a “drug–drug–gene” interaction[37]). For instance, the concurrent use of tacrolimus and phenytoin can result in decreased efficacy of tacrolimus through phenytoin's ability to induce CYP3A4. This effect can be potentiated when phenytoin's metabolism is reduced in patients who are CYP2C9 intermediate or poor metabolizers. These potential modifiers are also normalized and surfaced for each pairwise interaction.

The card's “summary” field is populated with an overall concise statement that considers the content and severity of all the individual pairwise interactions, along with an actionable recommendation. We refer to calculating the summary as “orchestration,” and consider multiple factors in the PillHarmonics proprietary orchestration logic. Criticality of the potential interaction (e.g., potentially life-threatening vs. moderate risk), drug interaction category (e.g., drug–drug, drug–allergy), effect on ordered and current drugs, potential toxicity, and concordance of pairwise recommendations are all incorporated. Additionally, the necessity of the specific ingredient ordered is considered (e.g., a recommendation to avoid tacrolimus is potentially more significant than a recommendation to avoid a specific cholesterol lowering medication, as the former has fewer alternative agents). For specific drugs, dose form is also a factor (e.g., an oral drug may interact differently than a similar drug given intravenously), as is the indication for the ordered drug.

The CDS Hooks card is returned back to the EHR (in JSON format), where it can be rendered and presented to the ordering clinician before order completion. In our prototype, card rendering was designed to provide interaction information aggregated by category and severity, an icon with explanation for what kind of effect the interaction produced (e.g., decreased efficacy, toxicity), and expandable text fields providing further detail. In addition, actionable buttons corresponding to “suggestions” in the CDS Hooks specification are presented to the user.

For example, Patient #8 is an 80-year-old male recently diagnosed with tuberculosis who presents with intermittent chest pain, and is diagnosed with non-ST elevation myocardial infarction. The patient has a history of hypertension, type 2 diabetes, stage 4 chronic kidney disease; is taking amlodipine, ramipril, insulin, acarbose, aspirin, INH/PZA/Rifampin; has a serum creatinine of 2.4 mg/dL; and PGx star alleles of CYP2C9 *1/*1, CYP2D6 *1/*27, CYP2C19 *1/*2, and CYP3A5 *1/*3. The clinical team orders clopidogrel, and the service calculates the card shown in [Fig. 2].

Fig. 2 PillHarmonics output for a (synthetic) 80-year-old male recently diagnosed with tuberculosis who comes in complaining of intermittent chest pain for 3 weeks, and is diagnosed with non-ST elevation myocardial infarction. The patient has a past medical history of hypertension, type 2 diabetes, stage 4 chronic kidney disease; is taking amlodipine, ramipril, insulin, acarbose, aspirin, INH/PZA/rifampin; has a serum creatinine of 2.4, CYP2C9 *1/*1, CYP2D6 *1/*27, CYP2C19 *1/*2 (intermediate metabolizer), and CYP3A5 *1/*3. The clinical team orders clopidogrel, and the service calculates the card shown.

Another example: Patient #9 is a 61-year-old male with end-stage renal disease due to type 2 diabetes, and hypertension. The patient is taking nicardipine, metoprolol, insulin, empagliflozin, carbamazepine, sevelamer, erythropoietin, vitamin D, and iron; and has CYP3A5 *1/*3, CYP2C9 *24/*52, CYP2D6 *1/*27, CYP2C19 *1/*2. Cadaveric kidney becomes available and transplant is planned. Tacrolimus is ordered, and the service calculates the card shown in [Fig. 3].

Fig. 3 PillHarmonics output for a (synthetic) 61-year-old male with end-stage renal disease due to type 2 diabetes and hypertension. The patient has additional history of peripheral neuropathy; is taking nicardipine, metoprolol, insulin, empagliflozin, carbamazepine, Renagel, erythropoietin, vitamin D, iron; has CYP3A5 *1/*3, CYP2C9 *24/*52, CYP2D6 *1/*27, CYP2C19 *1/*2. Cadaveric kidney becomes available and kidney transplant is planned. Tacrolimus is ordered, and the service calculates the card shown.

PillHarmonics Evaluation

[Table 1] summarizes responses to the TAM Questionnaire and presents descriptive statistics of the entire cohort (N = 13 subjects).

The majority of responses to each question were >5 and the mean and median for each response were all >5, indicating the perceived benefits of the service. To further differentiate the responses between questions, we examined how often scores were 6 or 7 for each question. While each question received a high mean score, the question regarding job performance consistently received the highest scores, indicating this is where the cohort felt that PillHarmonics would be most effective. The question related to “Work More Quickly” received the lowest scores (though again, the median score for this question was 5).

Next, we examined if the mean scores were statistically higher than various cut-offs. At a threshold level of 4.5, we found the p-values were <0.05 for every question. In addition, two questions were statistically greater than 5 (Question #2 on job performance and #6 on usefulness). Thus, on average, the subjects agreed that the PillHarmonics program was beneficial (i.e., better than “neutral”) for every question, and some questions elicited stronger positive responses.

Qualitative themes from structured interviews are summarized in [Table 2]. See [Supplementary Material] (online only) for additional details of thematic analyses. Users overwhelmingly found value with the PillHarmonics service and stated that they would use such a system if it were available in their practice. They remarked highly on aspects like decreasing a clinician's cognitive burden in a complex clinical situation and increasing decision confidence. Subjects were emphatically positive about having all categories of medication interaction effects aggregated on one screen. The ability of the service to synthesize concise and actionable recommendations, particularly for complex medications or for those patients on multiple medications, was seen as a novel and important advance in medication CDS.

Table 2 PillHarmonics qualitative themes from structured interviews

	Themes	No. of Responses (%)
1	Access to additional information (e.g., links to other sources) and transparency is highly valued, for all types of interactions.	12 (92.3%)
2	Aggregation of multiple types of knowledge in a single display is highly useful.	11 (84.6%)
3	The indication for use should factor into the recommendations and content (whether or not there are alternatives, what the risk/benefit ratio is).	11 (84.6%)
4	A system like this might have mixed effects on efficiency, but this generally would be acceptable due to making more informed decisions.	11 (84.6%)
5	A summary recommendation incorporating multiple types of interactions is very useful, but should be very specific.	10 (76.9%)
6	Additional actions like pharmacy or specialty consult are useful.	10 (76.9%)
7	Text should be concise and reflective of the content.	9 (69.2%)
8	The sequence of presented knowledge matters for understanding (example sequence: allergies, genomics, drug–drug interactions, other conditions).	8 (61.5%)
9	The varied severity of interactions is very important to distinguish critical from mild (e.g., color-coding severity, willingness to accept interruptive alerts for critical interactions, and desire to hide mild interactions).	7 (53.8%)
10	The “race” category of interactions is of questionable utility, and should either be renamed or removed.	6 (46.2%)
11	Graphics draw users attention and thus should be consistent and unambiguous.	5 (38.5%)

Users did not believe that the system would save them time during an encounter; however, they felt that the high quality of recommendations would likely outweigh this factor. This aligns with the findings of the TAM analysis, in showing that clinicians overall find PillHarmonics useful, but also recognize the time implications of better informed decision making. Of note, a recurrent principle throughout the development of the TAM was that usefulness is a more important factor in adoption of technology than is ease of use, in that “users are often willing to cope with some difficulty of use in a system that provides critically needed functionality.”[36] Subjects felt that the addition of a concise actionable summary statement enhanced the service.

Interviewees had several comments related to drug–race alerting. While the tacrolimus package insert[38] notes that “African-American patients required a higher dose to attain comparable trough concentrations compared with Caucasian patients,” several questioned the utility and content of this category of alerts, given that some drug–race interactions may be interdependent with drug–gene interactions, and given the variability in assigning a patient's race in the EHR.

Subjects suggested several potential enhancements, not only to the user interface and user experience, but also to the underlying PillHarmonics orchestration algorithm and creation of the concise summary statement. While some PGx resources provide specific dosage guidance, which we could leverage in some scenarios, more complex scenarios with multiple interactions of mixed effects do not lend themselves to precise dosage guidance. Clinicians voiced a desire for such specific dosage guidance, but recognize that this is highly complex and beyond the current state of pharmacologic and PGx knowledge. Several approaches for quantifying the overall impact of combined drug–drug and drug–gene interactions have been developed and will be considered in a future version. In addition, subjects suggested that using PillHarmonics in other clinical workflows (e.g., in a medication management application) would be more valuable in certain scenarios.

Discussion

Key findings from this study are that it is possible to logically merge and normalize a range of drug interaction knowledge into a singular display via the CDS Hooks protocol, and clinicians perceive great value in such an orchestrated system. This approach has potential for improving medication safety (e.g., by notifying clinicians of conflicting drug effects), for increasing drug efficacy (e.g., by helping select the right drug), and for decreasing alert fatigue through aggregating knowledge to generate more precise recommendations.

Our approach of using FHIR Genomics Operations with a backend genomic data server could enable access to a patient's entire genome, if available. While PGx-testing laboratories often limit reporting to specific star alleles, a growing number of studies are suggesting that variants from across a patient's genome can be relevant in medication optimization,[39] [40] and testing platforms such as the ThermoFisher PharmacoScan assay have expanded coverage to over 4,000 markers in over 1,000 genes, and provide genotype, star allele, and copy number states for key genes. Today's EHRs are generally not equipped to manage large volumes of complex genomic data[9] [41] [42] and instead are exploring the storage of genomic data outside or alongside the EHR, using a genomic data server.[43] [44] The use of FHIR Genomics Operations enables contextually relevant slices of a patient's entire genome to be surfaced for CDS.

While PillHarmonics is implemented here as a CDS Hooks application, future applications of PillHarmonics will include deployment via a SMART-on-FHIR medication management app, and may also be deployed using EHR-specific solutions. Enhancements will include tailoring alternate medication suggestions based on formulary, alternate medication interactions, and other factors. For instance, in [Fig. 2], while rifampin increases the efficacy of clopidogrel, it decreases ticagrelor levels,[45] and therefore it may not make sense to suggest ticagrelor as an alternative. Major enhancements will include increased content to identify a larger range of possible interactions, and enhanced orchestration logic.

A major caveat to orchestration and meeting clinician needs for concise actionable medication CDS is the need to balance the state of drug interaction science, particularly where there are multiple interactions, against patient safety concerns that can arise from overly specific recommendations. When there are several significant and variable effects from all relevant factors (e.g., as in [Fig. 3]), the best summary statement may simply be “Multiple significant and conflicting interactions. Manage/monitor based on clinical judgment.” But several strategies are being developed to better qualify, if not quantify, the effects of multiple interactions or alerts,[37] [46] [47] [48] some of which are based on evolving machine learning algorithms,[49] [50] [51] and many of which may warrant incorporation into orchestration algorithms.

Our study has limitations. While the developed service is functioning in a test environment with actual service calls and logic calculation in real-time, it has not yet been implemented in a clinical environment. Additionally, the user study was performed with static mockups. It was the intent to encourage the participants to focus on content and meaning apparent in the mockups, rather than risking any potential distraction from functional difficulty with a prototype. Additionally, as we could not leverage native EHR interfaces for our prototype, we wanted to present something that may have been more similar to what could be possible with modern EHR graphics. However, it is possible that not all user input that could have been elicited at this stage was obtained due to this approach.

Conclusion

Medication safety and optimizing efficacy of therapy regimens remains a significant issue. A comprehensive medication CDS system, which leverages patient clinical and genomic data, to perform a wide range of interaction checking and present a concise and holistic view of medication knowledge back to the clinician, is feasible and highly valuable for more informed decision-making. We further believe that, given the positive responses observed in this study, such an orchestrated solution will gradually become the new norm in medication decision support.

Clinical Relevance Statement

Medication safety and optimizing efficacy of therapy regimens remain significant issues. A comprehensive medication decision support system that leverages patient clinical and genomic data to perform a wide range of interaction checking (drug–drug, drug–allergy, drug–condition, drug–gene, drug–renal, drug–hepatic, drug–race) and presents a concise and holistic view of medication knowledge back to the clinician is feasible and perceived as highly valuable for more informed decision-making. Such a system can potentially increase drug safety and efficacy, as well as decrease alert fatigue through more consolidated, precise recommendations.

Multiple Choice Questions

1. When ordering clopidogrel in a patient with a clopidogrel intermediate metabolizer phenotype, which of the following actions is correct?

   • Decrease the clopidogrel dose to less than standard dose (75 mg/day).

   • Use clopidogrel at standard dose (75 mg/day).

   • Increase the clopidogrel dose to greater than standard dose or switch to an alternate agent such as prasugrel or ticagrelor.

   • Avoid all drugs that inhibit platelet reactivity.

   Correct Answer: The correct answer is option c. See [Fig. 2]. In patients with a clopidogrel intermediate metabolizer phenotype, PharmGKB and CPIC recommend an alternative antiplatelet therapy for CYP2C19 poor or intermediate metabolizers.

2. When ordering tacrolimus in a patient with a tacrolimus intermediate metabolizer phenotype, which of the following actions is correct?

   • Decrease the tacrolimus starting dose to less than the standard starting dose.

   • Use tacrolimus at the standard starting dose.

   • Increase the tacrolimus starting dose to greater than the standard starting dose.

   • Strictly avoid use of tacrolimus.

   Correct Answer: The correct answer is option c. See [Fig. 3]. In patients with a tacrolimus intermediate metabolizer phenotype, PharmGKB and CPIC recommend increasing the starting dose by 1.5 to 2 times the recommended starting dose.

3. When ordering a drug in the EHR, and drug-gene interactions are only available in a PDF report within the EHR, which of the following actions is correct?

   • The ordering provider must always review the PDF before ordering any drug.

   • The ordering provider may overlook important drug–gene interactions during the order entry process.

   • The ordering provider can ignore the PDF and safely assume that any important drug–gene interactions will be automatically surfaced during the order-entry process.

   • The ordering provider needs to defer all medication ordering to trained PGx pharmacists.

   Correct Answer: The correct answer is option b. Drug–gene interactions only available in PDF reports are problematic in that clinicians may not remember PGx results and their implications at the time of order entry, and may fail to recognize conflicting recommendations (e.g., PGx considerations suggest the need for a higher dose, whereas renal function suggests the need for a lower dose).

4. When ordering a drug in the EHR and presented with integrated drug-drug/allergy/condition/gene/renal/hepatic/race alerting prior to order confirmation, which of the following actions is correct?

   • The order-entry process may take longer but will be better informed.

   • The order-entry process will be quicker.

   • Order-entry decision support will provide precise dosing recommendations, even where there are multiple discordant drug interactions.

   • The ordering provider will need to consult with a PGx-trained pharmacist prior to ordering most drugs.

   Correct Answer: The correct answer is option a. Studies show that while clinicians are loathe to accept interventions that increase time per encounter, an important caveat is that time is not absolute, but rather, is relative, particularly where high-quality interventions instill greater clinician confidence in the safety and efficacy of their medication order.

• References

• 1 Bell GC, Crews KR, Wilkinson MR. et al. Development and use of active clinical decision support for preemptive pharmacogenomics. J Am Med Inform Assoc 2014; 21 (e1): e93-e99
• 2 Relling MV, Evans WE. Pharmacogenomics in the clinic. Nature 2015; 526 (7573): 343-350
• 3 Van Driest SL, Shi Y, Bowton EA. et al. Clinically actionable genotypes among 10,000 patients with preemptive pharmacogenomic testing. Clin Pharmacol Ther 2014; 95 (04) 423-431
• 4 McInnes G, Lavertu A, Sangkuhl K, Klein TE, Whirl-Carrillo M, Altman RB. Pharmacogenetics at scale: an analysis of the UK Biobank. Clin Pharmacol Ther 2021; 109 (06) 1528-1537
• 5 Clinical Pharmacogenetics Implementation Consortium (CPIC). Accessed July 15, 2020 at: https://cpicpgx.org/
• 6 Dunnenberger HM, Crews KR, Hoffman JM. et al. Preemptive clinical pharmacogenetics implementation: current programs in five US medical centers. Annu Rev Pharmacol Toxicol 2015; 55: 89-106
• 7 Bush WS, Crosslin DR, Owusu-Obeng A. et al. Genetic variation among 82 pharmacogenes: the PGRNseq data from the eMERGE network. Clin Pharmacol Ther 2016; 100 (02) 160-169
• 8 Chanfreau-Coffinier C, Hull LE, Lynch JA. et al. Projected prevalence of actionable pharmacogenetic variants and level A drugs prescribed among US Veterans Health Administration pharmacy users. JAMA Netw Open 2019; 2 (06) e195345-e195345
• 9 Williams MS, Taylor CO, Walton NA. et al. Genomic information for clinicians in the electronic health record: lessons learned from the clinical Genome Resource Project and the electronic medical records and genomics network. Front Genet 2019; 10: 1059
• 10 Caraballo PJ, Sutton JA, Giri J. et al. Integrating pharmacogenomics into the electronic health record by implementing genomic indicators. J Am Med Inform Assoc 2020; 27 (01) 154-158
• 11 Melton BL, Zillich AJ, Saleem J, Russ AL, Tisdale JE, Overholser BR. Iterative development and evaluation of a pharmacogenomic-guided clinical decision support system for warfarin dosing. Appl Clin Inform 2016; 7 (04) 1088-1106
• 12 HL7 FHIR Genomics Reporting Implementation Guide. Accessed May 1, 2020 at: http://hl7.org/fhir/uv/genomics-reporting/index.html
• 13 Hooks CDS. A “hook”-based pattern for invoking decision support from within a clinician's EHR workflow. Accessed July 15, 2020 at: https://cds-hooks.org/
• 14 Dolin RH, Boxwala A, Shalaby J. A Pharmacogenomics clinical decision support service based on FHIR and CDS hooks. Methods Inf Med 2018; 57 (S 02): e115-e123
• 15 Watkins M, Eilbeck K. FHIR lab reports: using SMART on FHIR and CDS Hooks to increase the clinical utility of pharmacogenomic laboratory test results. AMIA Jt Summits Transl Sci Proc 2020; 2020: 683-692
• 16 Hansten PD, Horn JR, Hazlet TK. ORCA: OpeRational ClassificAtion of drug interactions. J Am Pharm Assoc 2001; 41 (02) 161-165
• 17 Paterno MD, Maviglia SM, Gorman PN. et al. Tiering drug-drug interaction alerts by severity increases compliance rates. J Am Med Inform Assoc 2009; 16 (01) 40-46
• 18 Howard I, Howland I, Castle N. et al. Retrospective identification of medication related adverse events in the emergency medical services through the analysis of a patient safety register. Sci Rep 2022; 12 (01) 2622
• 19 Hicks LS, Sequist TD, Ayanian JZ. et al. Impact of computerized decision support on blood pressure management and control: a randomized controlled trial. J Gen Intern Med 2008; 23 (04) 429-441
• 20 Bates DW, Leape LL, Cullen DJ. et al. Effect of computerized physician order entry and a team intervention on prevention of serious medication errors. JAMA 1998; 280 (15) 1311-1316
• 21 Bates DW, Teich JM, Lee J. et al. The impact of computerized physician order entry on medication error prevention. J Am Med Inform Assoc 1999; 6 (04) 313-321
• 22 Kuperman GJ, Teich JM, Gandhi TK, Bates DW. Patient safety and computerized medication ordering at Brigham and Women's Hospital. Jt Comm J Qual Improv 2001; 27 (10) 509-521
• 23 Kawamoto K, Houlihan CA, Balas EA, Lobach DF. Improving clinical practice using clinical decision support systems: a systematic review of trials to identify features critical to success. BMJ 2005; 330 (7494): 765
• 24 Nanji KC, Seger DL, Slight SP. et al. Medication-related clinical decision support alert overrides in inpatients. J Am Med Inform Assoc 2018; 25 (05) 476-481
• 25 Strom BL, Schinnar R, Aberra F. et al. Unintended effects of a computerized physician order entry nearly hard-stop alert to prevent a drug interaction: a randomized controlled trial. Arch Intern Med 2010; 170 (17) 1578-1583
• 26 Tolley CL, Slight SP, Husband AK, Watson N, Bates DW. Improving medication-related clinical decision support. Am J Health Syst Pharm 2018; 75 (04) 239-246
• 27 Payne TH, Hines LE, Chan RC. et al. Recommendations to improve the usability of drug-drug interaction clinical decision support alerts. J Am Med Inform Assoc 2015; 22 (06) 1243-1250
• 28 Humphrey KE, Mirica M, Phansalkar S, Ozonoff A, Harper MB. Clinician perceptions of timing and presentation of drug-drug interaction alerts. Appl Clin Inform 2020; 11 (03) 487-496
• 29 Hicks JK, Dunnenberger HM, Gumpper KF, Haidar CE, Hoffman JM. Integrating pharmacogenomics into electronic health records with clinical decision support. Am J Health Syst Pharm 2016; 73 (23) 1967-1976
• 30 Khelifi M, Tarczy-Hornoch P, Devine EB, Pratt W. Design recommendations for pharmacogenomics clinical decision support systems. AMIA Jt Summits Transl Sci Proc 2017; 2017: 237-246
• 31 Dolin RH, Heale BSE, Alterovitz G. et al. Introducing HL7 FHIR genomics operations: a developer-friendly approach to genomics-EHR integration. J Am Med Inform Assoc 2023; 30 (03) 485-493
• 32 Gong L, Whirl-Carrillo M, Klein TE. PharmGKB, an integrated resource of pharmacogenomic knowledge. Curr Protoc 2021; 1 (08) e226
• 33 Davis FD. A Technology Acceptance Model for Empirically Testing New End-User Information Systems: Theory and Results. Thesis. Massachusetts Institute of Technology; 1985. Accessed July 15, 2020 at: https://dspace.mit.edu/handle/1721.1/15192
• 34 Castleberry A, Nolen A. Thematic analysis of qualitative research data: is it as easy as it sounds?. Curr Pharm Teach Learn 2018; 10 (06) 807-815
• 35 Braun V, Clarke V. Using thematic analysis in psychology. Qual Res Psychol 2006; 3 (02) 77-101
• 36 Davis FD. Perceived usefulness, perceived ease of use, and user acceptance of information technology. Manage Inf Syst Q 1989; 13 (03) 319-340
• 37 Malki MA, Pearson ER. Drug-drug-gene interactions and adverse drug reactions. Pharmacogenomics J 2020; 20 (03) 355-366
• 38 DailyMed - TACROLIMUS capsule. Accessed June 12, 2023 at: https://dailymed.nlm.nih.gov/dailymed/drugInfo.cfm?setid=bd447ffa-9196-4c3c-accf-5adf29b84665
• 39 Lin E, Kuo PH, Liu YL, Yu YWY, Yang AC, Tsai SJ. A deep learning approach for predicting antidepressant response in major depression using clinical and genetic biomarkers. Front Psychiatry 2018; 9: 290
• 40 Pharmacogenomics-Driven Prediction of Antidepressant Treatment Outcomes. A Machine-Learning Approach With Multi-trial Replication - Athreya - 2019 - Clinical Pharmacology &amp; Therapeutics - Wiley Online Library. Accessed March 10, 2023 at: https://ascpt.onlinelibrary.wiley.com/doi/10.1002/cpt.1482
• 41 Walton NA, Johnson DK, Person TN, Chamala S. Genomic data in the electronic health record. Adv Mol Pathol. 2019; 2 (01) 21-33
• 42 Ohno-Machado L, Kim J, Gabriel RA, Kuo GM, Hogarth MA. Genomics and electronic health record systems. Hum Mol Genet 2018; 27 (R1): R48-R55
• 43 Starren J, Williams MS, Bottinger EP. Crossing the omic chasm: a time for omic ancillary systems. JAMA 2013; 309 (12) 1237-1238
• 44 Masys DR, Jarvik GP, Abernethy NF. et al. Technical desiderata for the integration of genomic data into electronic health records. J Biomed Inform 2012; 45 (03) 419-422
• 45 Wang ZY, Chen M, Zhu LL. et al. Pharmacokinetic drug interactions with clopidogrel: updated review and risk management in combination therapy. Ther Clin Risk Manag 2015; 11: 449-467
• 46 Tod M, Rodier T, Auffret M. Quantitative prediction of adverse event probability due to pharmacokinetic interactions. Drug Saf 2022; 45 (07) 755-764
• 47 Le Corvaisier C, Capelle A, France M, Bourguignon L, Tod M, Goutelle S. Drug interactions between emergency contraceptive drugs and cytochrome inducers: literature review and quantitative prediction. Fundam Clin Pharmacol 2021; 35 (02) 208-216
• 48 Fermier N, Bourguignon L, Goutelle S, Bleyzac N, Tod M. Identification of cytochrome P450-mediated drug-drug interactions at risk in cases of gene polymorphisms by using a quantitative prediction model. Clin Pharmacokinet 2018; 57 (12) 1581-1591
• 49 Zhang Y, Deng Z, Xu X, Feng Y, Junliang S. Application of artificial intelligence in drug-drug interactions prediction: a review. J Chem Inf Model 2023 (e-pub ahead of print). Doi:10.1021/acs.jcim.3c00582
• 50 Jang HY, Song J, Kim JH. et al. Machine learning-based quantitative prediction of drug exposure in drug-drug interactions using drug label information. NPJ Digit Med 2022; 5 (01) 88
• 51 Mei S, Zhang K. A machine learning framework for predicting drug-drug interactions. Sci Rep 2021; 11 (01) 17619