Proposing a framework to quantify the potential impact of pharmacokinetic drug–drug interactions caused by a new drug candidate by using real world data about the target patient population

Abstract

Drug development teams must evaluate the risk/benefit profile of new drug candidates that perpetrate drug–drug interactions (DDIs). Real‐world data (RWD) can inform this decision. The purpose of this study was to develop a predicted impact score for DDIs perpetrated by three hypothetical drug candidates via CYP3A, CYP2D6, or CYP2C9 in type 2 diabetes mellitus (T2DM), obesity, or migraine. Optum Market Clarity was analyzed to estimate use of CYP3A, CYP2D6, or CYP2C9 substrates classified in the University of Washington Drug Interaction Database as moderate sensitive, sensitive, narrow therapeutic index, or QT prolongation. Scoring was based on prevalence of exposure to victim substrates and characteristics (age, polypharmacy, duration of exposure, and number of prescribers) of those exposed. The study population of 14,163,271 adults included 1,579,054 with T2DM, 3,117,753 with obesity, and 410,436 with migraine. For T2DM, 71.3% used CYP3A substrates, 44.3% used CYP2D6 substrates, and 44.3% used CYP2C9 substrates. For obesity, 57.1% used CYP3A substrates, 34.6% used CYP2D6 substrates, and 31.0% used CYP2C9 substrates. For migraine, 64.1% used CYP3A substrates, 44.0% used CYP2D6 substrates, and 28.9% used CYP2C9 substrates. In our analyses, the predicted DDI impact scores were highest for DDIs involving CYP3A, followed by CYP2D6, and CYP2C9 substrates, and highest for T2DM, followed by migraine, and obesity. Insights from RWD can be used to estimate a predicted DDI impact score for pharmacokinetic DDIs perpetrated by new drug candidates currently in development. This score can inform the risk/benefit profile of new drug candidates in a target patient population.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

During development, new drug candidates that may be identified as potential perpetrators of pharmacokinetic (PK)‐based drug–drug interactions (DDIs) are usually further evaluated for risk based on static mechanistic models, physiologically‐based PK modeling, and/or clinical DDI studies with specific probe substrates, selective inhibitors or inducers, or drugs likely to be co‐administered with the drug candidate. Real‐world data (RWD) are not routinely used in this assessment.

• WHAT QUESTION DID THIS STUDY ADDRESS?

This study sought to answer the question “How can we evaluate the potential impact of PK‐based DDIs perpetrated by a new drug candidate at a population level using RWD about the target patient population?”

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study developed a novel scoring system to evaluate the potential impact of PK‐based DDIs perpetrated by new drug candidates currently in development using RWD about the target patient population.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Incorporating RWD into the assessment of new drug candidates that perpetrate PK‐based DDIs could provide insights about the potential impact of DDIs if the new drug was eventually approved and used in a target patient population.

INTRODUCTION

A pharmacokinetic (PK) drug–drug interaction (DDI) occurs when the co‐administration of two or more drugs alters the absorption, distribution, metabolism, or excretion of a drug. Many PK DDIs occur when one drug (the perpetrator) affects the elimination or uptake of another drug (the victim) by inhibiting or inducing the metabolic enzymes or transporters that are responsible for the disposition of that drug. Such changes can increase or decrease the plasma concentration of the victim drug, which may result in adverse events (AEs) or adverse drug reactions (ADRs).

The potential for a new drug candidate to act as a perpetrator or victim of common metabolic enzyme or transporter pathways is usually identified early in drug discovery with in vitro studies. In recent years, regulatory agencies and the biopharmaceutical industry have developed systematic, risk‐based methods for evaluating PK‐based DDIs in new drug candidates and communicating these risks to the scientific and medical communities. Further evaluation of this risk can then be conducted with mechanistic static models, physiologically‐based pharmacokinetic (PBPK) modeling, and/or clinical DDI studies with specific probe substrates, selective inhibitors or inducers, or drugs likely to be co‐administered with the drug candidate. ¹ , ²

Cross‐functional drug development teams must weigh a variety of factors when evaluating the risk/benefit profile of new drug candidates, including the target product profile, treatment options, expected efficacy against standard of care, and potential to perpetrate DDIs. The information available about DDIs for new drug candidates (e.g., in vitro studies, PBPK modeling, and clinical DDI studies) provides limited insight into the potential impact that such DDIs would have in the target patient population (e.g., if DDIs impact only a small portion of patients and can easily be managed, development of a new drug candidate can proceed).

A better understanding of the potential impact of DDIs identified during drug development could help mitigate the health burden presented by DDIs, AEs, and ADRs. In the United States, ADRs are estimated to occur in 4.7%–6.7% of hospitalized patients and are responsible for nearly 3 million visits to emergency departments (EDs) each year, or 0.8%–2.9% of all ED visits. ³ , ⁴ , ⁵ Although the severity of ADRs can vary widely, an estimated 106,000 deaths per year are attributed to ADRs in the United States. ³ In 1995, an economic analysis estimated the cost of ADRs and other drug‐related problems in the United States at $76.6 billion/year; this estimate was revised to $177.4 billion in 2001. ⁶ , ⁷

Various patient characteristics have been identified as risk factors for DDIs, including older age, higher number of concurrently prescribed medications (i.e., polypharmacy), and a higher number of prescribers. ⁸ , ⁹ , ¹⁰ , ¹¹ Insights about the prevalence of these characteristics among patients in the target population could help evaluate the potential impact of DDIs related to a new drug candidate. In populo studies using real‐world data (RWD) have been proposed to complement data from in vitro and clinical DDI studies and evaluate this impact. ¹⁰ , ¹¹

RWD refers to patient‐level health information collected in electronic health records (EHRs), insurance claims, and other sources during routine health care encounters. ¹² RWD captures patient characteristics (e.g., age, gender, race, ethnicity, and location), comorbidities (e.g., renal disease and hepatic disease) and medications dispensed (e.g., name, type, formulation, quantity, and duration). Although previous studies have analyzed RWD to estimate the prevalence and sequelae of DDIs for marketed drugs, RWD is not commonly used to evaluate the potential impact of DDIs associated with new drug candidates prior to launch. ¹³ , ¹⁴

The primary objective of this study was to develop a framework to evaluate the potential impact of PK‐based DDIs perpetrated by new drug candidates using RWD on concomitant medications and selected risk factors for DDIs in the target patient population.

METHODS

Study design

This was an observational, retrospective analysis of existing and de‐identified RWD to evaluate the potential PK‐based DDI risk of hypothetical drug candidates in three target populations; authors deemed that review by an ethics review board was not required.

Data sources

We used the Market Clarity (MC) database from Optum Inc., which combines medical claims, pharmacy claims, and EHRs for individuals with commercial or Medicare Advantage health plans managed by UnitedHealth Group. Medications in pharmacy claims were cross‐referenced with the April 2023 edition of the University of Washington (UW) Drug Interaction Database (DIDB) to determine if they were CYP3A, CYP2D6, or CYP2C9 moderate sensitive substrates, sensitive substrates, substrates with narrow therapeutic index (NTI), or substrates with QT prolongation; the UW DIDB was acquired by Certara in June 2023.

The DIDB classifies drugs according to definitions in the 2020 US Food and Drug Administration (FDA) guidance on DDIs. ¹⁵ Moderate sensitive substrates are defined as drugs whose area under the curve of substrate drugs (AUCR) increases two to five times when administered with a known strong index inhibitor or in individuals with genetic polymorphisms related to a specific enzyme. Sensitive substrates are defined as drugs whose AUCR increases at least five times when administered with a known index inhibitor or in poor metabolizers. When multiple studies are available to estimate changes in AUCR for a given drug, DIDB classifies that drug according to the largest reported increase in AUCR (i.e., worst case scenario). The DIDB classifies substrates with NTI based on information from DrugBank.

Analysis period

We examined data from October 1, 2019, to September 30, 2021, subdivided into two periods. The baseline period (October 1, 2019, to September 30, 2020) was used to determine patient eligibility and identify the target populations of interest. The follow‐up period (October 1, 2020, to September 30, 2021) was used to examine use of prescription medications that could be victim substrates for DDIs perpetrated by the hypothetical drug candidates and to identify selected risk factors for DDIs in those patients.

Study population

Individuals who met the following criteria were eligible for this study:

1. Age 18+ years at start of baseline period,

2. Continuous medical and pharmacy benefits throughout the entire study period,

3. Known gender.

The entire study population of eligible individuals was used as a control group for providing the background prevalence for the use of victim substrates and selected risk factors for DDIs. The three target patient populations were type 2 diabetes mellitus (T2DM), obesity, and migraine, as described in the case definitions in Table S1 and below.

Type 2 diabetes mellitus

Individuals were identified as having T2DM if they met any of the following criteria:

1. One or more medical claim in a hospital with a relevant International Classification of Disease‐10th revision (ICD‐10) diagnosis code,

2. Two or more medical claims in a physician office with a relevant ICD‐10 diagnosis code,

3. Two or more pharmacy claims for a relevant medication,

4. Sixty or more days' supply in pharmacy claims for a relevant antidiabetic medication.

This definition was based on a study that compared ICD‐10 diagnosis codes for T2DM in medical claims with EHR data, HbA1c, blood glucose tests, or use of antidiabetic medications. ⁹ A case definition of T2DM similar to that used in our study had a sensitivity of 90.0%, specificity of 97.7%, positive predictive value of 82.6%, and negative predictive value of 98.8%. ⁹

Obesity

Individuals were identified as having obesity if they met any of the following criteria:

1. One or more medical claim in any setting with a relevant ICD‐10 diagnosis code,

2. One or more encounters in the EHR with a body mass index (BMI) of greater than or equal to 30 kg/m².

This definition was based on two studies that reported ICD‐10 diagnosis codes alone had a low sensitivity for obesity. ⁸ , ¹⁶ A study using the MC database compared ICD‐10 diagnosis codes with BMI in EHR data and reported that a case definition based on 1+ medical claim over 12 months resulted in a specificity of 98.7% but a sensitivity of only 25.2%. ⁸ A similar study compared BMI in EHRs to ICD diagnosis codes and reported a specificity of 97.0% but a sensitivity of only 40.4%. ¹⁶

Migraine

Individuals were identified as having migraine if they met any of the following criteria:

1. One or more medical claim in a hospital with a relevant ICD‐10 diagnosis code,

2. Two or more medical claims in a physician office with a relevant ICD‐10 diagnosis code,

3. Two or more pharmacy claims for a relevant medication,

4. Sixty or more days' supply in pharmacy claims for a relevant migraine medication.

This definition was based on two studies that identified migraine using RWD. ¹⁷ , ¹⁸ One study defined migraine based on one inpatient claim with relevant ICD‐10 diagnosis codes, two outpatient claims, one outpatient claim, and one pharmacy claim for triptans or ergotamines, or two pharmacy claims for triptans or ergotamines. ¹⁷ Another study compared migraine in claims with structured patient interviews based on International Headache Society (IHS) criteria for migraine. ¹⁸ Among the 1265 patients who met IHS criteria for migraine, only 449 (35.5%) had a relevant diagnosis code or migraine medication, resulting in a specificity using RWD of 91.6% with a sensitivity of only 35.5%.

Descriptive characteristics

Patient characteristics, including demographics (age, gender, race, ethnicity, region, and payer type), prescription medications (based on pharmacy claims), and selected risk factors for DDIs (see below) were summarized in each target population overall and within subgroups taking prescription medications classified as victim substrates of the CYP3A, CYP2D6, or CYP2C9 pathways.

Prescription medications

Our analyses focused on prescription medications likely to result in systemic exposure (rather than local exposure) and therefore excluded medications where the route of administration was unlikely to result in substantive systemic exposure (e.g., intraocular, ophthalmic, otic, and topical). Medical supplies and nutritional supplements in pharmacy claims were also excluded. Pharmacy claims where days’ supply was zero were also excluded.

General risk factors for DDIs

Age

Patient age was based on age at the start of the study period, with subgroups aged 65+ and aged 80+ years based on common definitions of older and elderly adults.

Duration of exposure

This was based on total duration of exposure to any medication of that type during the study follow‐up period, with subgroups for 90+ and 180+ days. Figure 1 presents a hypothetical patient with pharmacy claims for five different medications classified as CYP3A sensitive substrates.

FIGURE 1.

Illustration of method used to determine total duration of exposure for each type of victim substrate in each pathway of interest.

Polypharmacy

Number of concomitant medications was determined by identifying pharmacy claims with at least 1 day of overlapping supply during the study follow‐up period, with subgroups for five plus and 10+ medications. Figure 2 presents a hypothetical patient with polypharmacy for five plus medications.

FIGURE 2.

Illustration of method used to determine number of medications involved in polypharmacy for each type of victim substrate in each pathway of interest.

Number of prescribers

Number of prescribers was determined by the number of unique prescribing provider identification numbers in pharmacy claims, with subgroups for five plus and 10+ prescribers.

Predicted DDI impact score

A scoring system was devised to quantify and summarize the predicted impact of PK‐based DDIs perpetrated by each of the hypothetical drug candidates on the three pathways of interest (CYP3A, CYP2D6, and CYP2C9) in each of the three target patient populations, as explained below.

Prevalence of exposure to victim substrates

Points were first awarded for the proportion of individuals in each target patient population exposed to victim substrates. Different scoring thresholds were used for different types of substrates to reflect the perceived severity of potential DDIs involving those substrates. For moderate sensitive substrates, two points were awarded for every 10% increment in the prevalence of use, with a maximum of 10 points. For sensitive substrates, two points were awarded for every 5% increment in prevalence of use, with a maximum of 10 points. For NTI and QT, two points were awarded for every 1% increment in prevalence in use, with a maximum of 10 points (see Table 1).

TABLE 1.

Predicted DDI impact scoring system based on prevalence of victim substrate use within each target patient population.

Score	Prevalence of moderate sensitive substrate use	Prevalence of sensitive substrate use	Prevalence of substrate with NTI use	Prevalence of substrate with QT prolongation use
0 points	0%–10%	0%–5%	0%–1%	0%–1%
2 points	10%–20%	5%–10%	1%–2%	1%–2%
4 points	20%–30%	10%–15%	2%–3%	2%–3%
6 points	30%–40%	15%–20%	3%–4%	3%–4%
8 points	40%–50%	20%–25%	4%–5%	4%–5%
10 points	>50%	>25%	>5%	>5%

Abbreviations: DDI, drug–drug interaction; NTI, narrow therapeutic index.

Prevalence of selected risk factors for DDIs

Points were then awarded based on the characteristics of individuals exposed to victim substrates. Characteristics of interest were general risk factors for DDIs identified in scientific literature, such as patient age, duration of exposure (to victim substrates), polypharmacy (multiple concurrent medications), and polyprescribing (multiple prescribers). Two tiers were considered for each risk factor, including age 65+ and 80+ years, duration of exposure for 90+ days and 180+ days, polypharmacy with five plus and 10+ concurrent medications, and polyprescribing with five plus and 10+ prescribers.

One point was awarded for every 10% difference in the prevalence of a risk factor in a target patient population relative to its prevalence in the control group, with a maximum of five points. For example, if the proportion of patients age 65+ years among those taking CYP3A sensitive substrates was 37.6% in T2DM versus 23.6% in the control group, the difference (37.6%–23.6% = 14.0%) was awarded one point. This score for selected risk factors for DDIs was only calculated for victim substrates that scored two plus points on the use score described above (see Table 2).

TABLE 2.

Predicted DDI impact scoring system based on prevalence of selected risk factors for DDIs within each target population relative to control group. ^a

Difference in prevalence of selected risk factors for DDIs in target patient population vs. control group a	Points
<10%	0
10%–20%	1
20%–30%	2
30%–40%	3
40%–50%	4
>50%	5

Abbreviations: DDI, drug–drug interaction; MC, Market Clarity.

^a Control group consisted of all eligible patients in the MC database during the study period (n = 14,163,271).

RESULTS

Study populations

See Figure 3 for a patient flow diagram outlining the four populations in this study (i.e., 3 target patient populations and control group). Baseline characteristics are summarized in Table 3.

FIGURE 3.

Patient flow with eligibility criteria and target populations of interest. EHR, electronic health record.

TABLE 3.

Baseline characteristics of study populations.

Variable	T2DM		Obesity		Migraine		Control group a
	Number	%	Number	%	Number	%	Number	%
Total individuals	1,579,054	100.0%	3,117,753	100.0%	410,436	100.0%	14,163,271	100.0%
Crude 12‐month prevalence	–	11.1%	–	22.0%	–	2.9%	–	–
Sex
Female	828,298	52.5%	1,860,323	59.7%	344,011	83.8%	7,927,888	56.0%
Male	750,756	47.5%	1,257,430	40.3%	66,425	16.2%	6,235,383	44.0%
Age, years
Age 18–64	991,145	62.8%	2,500,855	80.2%	378,148	92.1%	11,616,060	82.0%
Age 65–79	467,346	29.6%	523,637	16.8%	29,478	7.2%	1,985,119	14.0%
Age 80+	120,563	7.6%	93,261	3.0%	2810	0.7%	562,092	4.0%
Mean (standard deviation)	59.85	13.84	50.90	15.68	45.32	13.83	47.89	17.39
Race
African American	216,283	13.7%	430,489	13.8%	30,900	7.5%	1,349,969	9.5%
Asian	43,717	2.8%	31,138	1.0%	4403	1.1%	356,174	2.5%
White	955,786	60.5%	2,111,547	67.7%	296,156	72.2%	9,220,492	65.1%
Other/unknown	363,268	23.0%	544,579	17.5%	78,977	19.2%	3,236,636	22.9%
Ethnicity
Hispanic	101,463	6.4%	170,303	5.5%	17,837	4.4%	670,131	4.7%
Not Hispanic	1,038,867	65.8%	2,232,489	71.6%	291,814	71.1%	9,327,086	65.9%
Unknown	438,724	27.8%	714,961	22.9%	100,785	24.6%	4,166,054	29.4%
Primary payer type
Commercial	783,517	49.6%	1,905,745	61.1%	280,609	68.4%	9,576,997	67.6%
Commercial + Medicaid	27,323	1.7%	65,922	2.1%	10,020	2.4%	227,630	1.6%
Commercial + Medicare	53,745	3.4%	68,167	2.2%	4491	1.1%	247,634	1.8%
Medicaid	179,119	11.3%	433,057	13.9%	61,533	15.0%	1,575,869	11.1%
Medicare + Medicaid	22,479	1.4%	31,126	1.0%	3806	0.9%	105,928	0.8%
Medicare	456,417	28.9%	486,747	15.6%	34,013	8.3%	1,835,291	13.0%
Other	56,454	3.6%	126,989	4.1%	15,964	3.9%	593,922	4.2%

Abbreviations: T2DM, type 2 diabetes mellitus; MC, Market Clarity.

^a Control group consisted of all eligible patients in the MC database during the study period (n = 14,163,271).

Medication use

Some differences were noted in the 10 most prescribed medications in each target patient population, with considerable overlap between patients with T2DM and obesity (Table 4). Omeprazole, a CYP3A moderate sensitive substrate, was among the 10 most prescribed medications for all target patient populations and the control group. Atorvastatin, a CYP3A moderate sensitive substrate, was the most common medication for obesity (18.3%) and the control group (11.2%) and the second most commonly prescribed medication for T2DM (35.2%).

TABLE 4.

Top 10 prescription medications in pharmacy claims for different study populations.

Medication	T2DM		Obesity		Migraine		Control a		Victim substrate type b
	%	Rank	%	Rank	%	Rank	%	Rank
Acetaminophen/hydrocodone	12.2%	11	11.5%	8	16.1%	4	7.7%	7
Albuterol	14.5%	7	14.1%	3	18.2%	2	8.8%	2
Amlodipine	18.7%	4	12.2%	6	6.5%	42	7.4%	9	CYP3A moderate sensitive substrate
Atorvastatin	35.2%	2	18.3%	1	11.3%	19	11.2%	1	CYP3A sensitive substrate
Cyclobenzaprine	7.7%	25	8.1%	18	14.4%	7	5.2%	18
Gabapentin	15.8%	5	10.5%	11	16.0%	5	6.3%	13
Ibuprofen	9.1%	21	10.9%	10	13.6%	10	7.7%	6	CYP2C9 sensitive substrate
Insulin glargine	12.5%	10	3.7%	52	1.8%	110	1.6%	73
Levothyroxine	14.3%	8	11.3%	9	13.0%	12	7.8%	5
Lisinopril	23.3%	3	12.9%	4	7.2%	40	7.6%	8
Losartan	14.9%	6	9.0%	15	5.0%	62	5.1%	19	CYP2C9 moderate sensitive substrate
Metformin	50.9%	1	14.6%	2	7.3%	39	6.9%	12
Methylprednisolone	6.7%	35	8.2%	16	13.8%	9	5.9%	15	CYP3A moderate sensitive substrate
Omeprazole	13.9%	9	11.6%	7	14.1%	8	7.0%	10	CYP3A moderate sensitive substrate
Prednisone	11.1%	13	12.3%	5	17.9%	3	8.5%	3
Sumatriptan	1.1%	137	1.8%	106	25.6%	1	1.3%	85
Topiramate	2.1%	107	2.7%	65	15.2%	6	1.4%	80

Abbreviations: DIDB, Drug Interaction Database; T2DM, type 2 diabetes mellitus; MC, Market Clarity.

^a Control group consisted of all eligible patients in the MC database during the study period (n = 14,163,271).

^b Only victim substrate types related to CYP3A, CYP2D6, and CYP2C9 are shown here; data are from April 2023 version of DIDB.

Overall, use by patients in the target populations was highest for CYP3A substrates, followed by CYP2D6 substrates, and CYP2C9 substrates (Table 5). The predicted DDI impact subscores based on prevalence of use of victim substrates in the target patient populations were highest for T2DM (CYP3A score = 34, CYP2D6 score = 24, and CYP2C9 score = 20) and lowest for the control group (CYP3A score = 22, CYP2D6 score = 16, and CYP2C9 score = 4).

TABLE 5.

Prevalence and predicted DDI impact scores for use of different types of victim substrates in study populations.

Type of victim substrate	Type 2 diabetes (%)	Points	Obesity (%)	Points	Migraine (%)	Points	Control group a (%)	Points
CYP3A moderate sensitive substrates	54.4%	10	47.2%	8	56.1%	10	33.8%	6
CYP3A sensitive substrates	49.9%	10	30.0%	10	30.3%	10	19.6%	6
CYP3A substrates with NTI	2.0%	4	1.4%	2	1.2%	2	0.9%	0
CYP3A substrates with QT prolongation	10.5%	10	9.8%	10	14.9%	10	7.1%	10
Subtotal score b	71.3%	34	57.1%	30	64.1%	32	41.4%	22
CYP2D6 moderate sensitive substrates	26.9%	4	21.0%	4	31.1%	6	13.9%	2
CYP2D6 sensitive substrates	26.3%	10	19.9%	6	22.6%	8	12.7%	4
CYP2D6 Substrates with NTI	0.1%	0	0.0%	0	0.1%	0	0.0%	0
CYP2D6 Substrates with QT prolongation	8.0%	10	8.6%	10	15.5%	10	5.9%	10
Subtotal score b	44.3%	24	34.6%	20	44.0%	24	23.1%	16
CYP2C9 moderate sensitive substrates	26.1%	4	13.9%	2	7.5%	0	7.7%	0
CYP2C9 sensitive substrates	27.1%	10	21.7%	8	24.3%	8	14.5%	4
CYP2C9 substrates with NTI	3.3%	6	1.7%	2	0.9%	0	1.0%	0
CYP2C9 substrates with QT prolongation	0.0%	0	0.0%	0	0.0%	0	0.0%	0
Subtotal score b	44.3%	20	31.0%	12	28.9%	8	19.9%	4

Abbreviations: DDI, drug–drug interaction; T2DM, type 2 diabetes mellitus; MC, Market Clarity; NTI, narrow therapeutic index.

^a Control group consisted of all eligible patients in the MC database during the study period (n = 14,163,271).

^b Subtotals represent the proportion of patients within a study population taking any of the four types of victim substrate in a pathway (e.g. 71.3% of patients with T2DM take a CYP3A moderate sensitive substrate, CYP3A sensitive substrate, CYP3A substrate with NTI, or CYP3A substrate with QT prolongation).

Selected risk factors for DDIs

The prevalence of selected risk factors for DDIs among individuals in the target populations exposed to different types of victim substrates is summarized in Table S2.

Total predicted DDI impact scores

Total predicted DDI impact scores based on prevalence of use for different types of victim substrates and selected risk factors for DDIs among individuals exposed to victim substrates in each of the three target patient populations are presented in Table 6.

TABLE 6.

Predicted DDI impact total scores for use of victim substrates and selected risk factors for DDIs among those using such medications.

Victim substrate type	Type 2 diabetes (%)			Obesity (%)			Migraine (%)
	Points for prevalence of use	Points for selected risk factors	Total score	Points for prevalence of use	Points for selected risk factors	Total score	Points for prevalence of use	Points for selected risk factors	Total score
CYP3A moderate sensitive substrates	10	9	19	8	1	9	10	6	16
CYP3A sensitive substrates	10	4	14	10	1	11	10	6	16
CYP3A substrates with NTI	4	5	9	2	0	2	2	5	7
CYP3A substrates with QT prolongation	10	8	18	10	2	12	10	7	17
Subtotal	34	26	60	30	4	34	32	24	56
CYP2D6 moderate sensitive substrates	4	7	11	4	2	6	6	4	10
CYP2D6 sensitive substrates	10	5	15	6	2	8	8	5	13
CYP2D6 substrates with NTI	0	0	0	0	0	0	0	0	0
CYP2D6 substrates with QT prolongation	10	8	18	10	2	12	10	6	16
Subtotal	24	20	44	20	6	26	24	15	39
CYP2C9 moderate sensitive substrates	4	2	6	2	0	2	0	0	0
CYP2C9 sensitive substrates	10	10	20	8	1	9	8	7	15
CYP2C9 substrates with NTI	6	2	8	2	0	2	0	0	0
CYP2C9 Substrates with QT prolongation	0	0	0	0	0	0	0	0	0
Subtotal	20	14	34	12	1	13	8	7	15

Abbreviations: DDI, drug–drug interaction; NTI, narrow therapeutic index.

Type 2 diabetes mellitus

For patients with T2DM, the total predicted DDI impact score was highest for CYP3A substrates (60), followed by CYP2D6 (44) and CYP2C9 substrates (34). No score was assigned for selected risk factors for patients with T2DM taking CYP2D6 substrates with NTI or CYP2C9 substrates with QT because use of those victim substrates was below the minimum threshold (<1%) for scoring.

Obesity

For patients with obesity, the total predicted DDI impact score was highest for CYP3A substrates (34), followed by CYP2D6 (26) and CYP2C9 substrates (13). No score was assigned for selected risk factors for patients with obesity taking CYP2D6 substrates with NTI or CYP2C9 substrates with QT because use of those victim substrates was below the minimum threshold (<1%) for scoring.

Migraine

For patients with migraine, the total predicted DDI impact score was highest for CYP3A substrates (56), followed by CYP2D6 (39) and CYP2C9 (15) substrates. No score was assigned for selected risk factors for patients with migraine taking CYP2D6 substrates with NTI, CYP2C9 moderate sensitive substrates, CYP2C9 substrates with NTI, or CYP2C9 substrates with QT prolongation because use of those victim substrates was below the minimum thresholds (<1% or < 10%, respectively) for scoring.

DISCUSSION

Our study found notable differences in exposure to victim substrates and prevalence of risk factors for DDIs across three target patient populations. The predicted DDI impact scores were highest for DDIs involving CYP3A, followed by CYP2D6 and CYP2C9 substrates and highest for patients with T2DM, followed by migraine and obesity. Many of the individuals using victim substrates were aged 65+ or 80+ years, used victim substrates for 90+ days or 180+ days, were exposed to polypharmacy with five plus or 10+ concurrent medications, and were prescribed medications by five plus or 10+ prescribers, suggesting they may be at increased risk of DDIs if co‐exposed to the new drug candidates.

Our estimates of the prevalence of polypharmacy are much higher than those reported by the National Center for Health Statistics, which estimated that 4.2% of those aged 18–44, 18.0% of those aged 45–64, and 41.9% of those aged 65+ years used five plus prescription medications in the past 30 days. ¹⁹ It is possible that RWD overestimates polypharmacy by assuming that patients take the full quantity of every medication dispensed. It is also possible that RWD overestimates the number of prescribers involved in an individual patient's care because it relies on unique identifiers that ignore how these healthcare providers (HCPs) might be related (e.g., a family practice with 1 physician, 1 physician's assistant, and 1 nurse practitioner would appear in pharmacy claims as 3 different prescribers).

Based on our framework to analyze RWD for the target patient populations, the predicted DDI impact scores were highest for DDIs involving CYP3A, followed by CYP2D6 and CYP2C9 substrates. These scores suggest, for example, that a drug candidate perpetrating PK‐based DDIs related to CYP3A could have a greater impact at a population level than one perpetrating DDIs related to CYP2D6 or CYP2C9. Although these findings may seem intuitive because more drugs are metabolized by CYP3A than by other pathways, our framework provides a method for quantifying this exposure to victim substrates and verifying this assumption. Without analyzing RWD, it is unclear how to precisely estimate potential exposure to victim substrates in target patient population.

With all else being equal, a new drug candidate with a lower predicted DDI impact score would be preferred over another drug candidate with similar characteristics as a perpetrator of DDIs but a higher predicted DDI impact score. The predicted DDI impact scores were highest for T2DM, followed by migraine, and obesity, which suggests that the role of a drug candidate in perpetrating DDIs could be viewed differently based on the target patient population (e.g., DDIs in patients with T2DM could be more problematic than DDIs in patients with obesity). These predicted DDI impact scores could also be used to inform on appropriate risk‐mitigation strategies (e.g., could recommend dose adjustment for patients at lower risk of predicted DDIs vs. label warning to avoid common victim substrates for patients at higher risk of predicted DDIs).

Previous studies on related topics

To our knowledge, few DDI risk tools have been developed to predict the risk of DDIs for drug candidates in development. The Drug–Drug Interaction Risk Calculator (DDIRC) from PharmaPendium (Elsevier, Inc; The Netherlands) uses different sources of preclinical data to predict the risk of DDIs for drug candidates based on mechanistic static models. ²⁰ The DDIRC is described as compliant with 2020 FDA guidance on in vitro interaction studies, consistent with scientific literature, and can be updated with new information throughout drug discovery and development. Although the proprietary methodology of the DDIRC cannot readily be assessed, its reliance on preclinical data does not appear to overlap with our use of RWD for the target patient population.

Several recent studies have used RWD to examine DDIs for approved products with known DDI and safety profiles. For example, a study by Bykov et al. ²¹ in 2021 analyzed the IBM/Truven Health MarketScan database with medical and pharmacy claims for ~100 million individuals in the United States. When comparing different methods to identify DDIs related to muscle toxicity or bleeding in RWD, it reported that the case‐crossover design and self‐controlled case series yielded comparable results. Similarly, a study by Chiang et al. ²² in 2018 compared DDIs related to myopathy reported in the FDA Adverse Event Reporting System (FAERS) and in an EHR database from the Indiana Network of Patient Care. The odds ratio for incident myopathy with a combination of omeprazole, fluconazole, and clonidine was 6.41 in FAERS versus 18.46 in the EHR database, suggesting consistent findings with different types of RWD.

A study by Smith et al. ²³ in 2021, also evaluated medications that could potentially be repurposed to treat coronavirus disease 2019 (COVID‐19). A medication risk score (MRS) was calculated using RWD for 527,471 members covered by commercial or managed Medicare plans before and after adding each medication evaluated. At baseline, the MRS predicted that 94.8% of members had a low risk of ADRs, 3.2% a moderate risk, and 2.0% a high risk. After adding lopinavir with ritonavir for patients with COVID‐19, the MRS estimated that 75.5% now had a low risk of ADRs, 13.5% a moderate risk, and 10.9% a high risk. Conceptually, the MRS appears similar to our predicted DDI impact score but uses a proprietary methodology that cannot readily be compared. In addition, it is unclear that the MRS could be generated for new drug candidates that are in development.

A study by Yee et al. ²⁴ in 2021, evaluated DDIs for both approved products with known safety profiles and compounds currently in development to evaluate 25 medications that could potentially be repurposed to treat COVID‐19. Their analyses identified transporter‐mediated DDIs using cell lines, evaluated each DDI using scientific literature and product labeling to estimate maximum plasma concentration, and predicted DDIs using EHR data for 2.9 million patients. It reported that 20 of 25 (80%) medications evaluated would likely result in clinically relevant DDIs if used in the target population.

Another study by Patrick et al. ²⁵ in 2021, combined RWD with data on gene expression and molecular structure to predict DDIs for 54 drugs used to treat psoriasis and 132 drugs used to treat comorbidities based on in silico analyses. Drugs were cross‐referenced against the DrugBank and MergedPDDI databases of DDIs and machine learning was used to predict 41 categories of DDIs (e.g., bleeding, drowsiness, and nephrotoxicity). The prevalence of predicted DDIs was validated using EHR data from 4 million patients. ADRs associated with new potential DDIs included ototoxicity from cyclosporine and furosemide as well as dyspepsia from calcitriol and trazodone.

In sum, most of the literature on RWD and DDIs to date has focused on predicting or identifying DDIs rather than quantifying the proportion of individuals in a target patient population who are susceptible to DDIs based on use of victim substrates and prevalence of risk factors for DDIs. Our manuscript could be one of the first to propose a framework to combine traditional data sources that identify the potential for a new drug to perpetrate DDIs (e.g., static models, PBPK models, and clinical DDI studies) with RWD to understand how such DDIs could impact the target population based on the prevalence of risk factors for DDIs (e.g., age, concomitant medications, and duration of exposure).

Limitations

The indications evaluated (T2DM, obesity, and migraine) are common drug targets for conditions seen in primary care settings. The risk/benefit profile of new drug candidates developed for life‐threatening indications – or conditions for which no therapies exist – will likely differ. The case definitions used in this analysis (e.g., 1+ hospital claim or 2+ non‐hospital claims) have not been validated and are based on ICD‐10 diagnosis codes that may not be accurate due to coding errors, use of specific codes to facilitate reimbursement, working diagnoses that are later ruled out, and failure to include all relevant codes on each claim. Although it is unlikely that individuals without the target diseases were identified using our case definitions, the lower prevalence for two of the three indications suggest that RWD may overlook individuals with the diseases of interest.

The 1‐year prevalence of obesity (21.5%) in our study was much lower than the 42.4% reported in the 2017–2018 National Health and Nutrition Examination Survey (NHANES), whereas the 1‐year prevalence of migraine (2.5%) was much lower than the 7.6% reported in the American Migraine Prevalence and Prevention study. ²⁶ , ²⁷ This may suggest that our case definitions did not identify all individuals with obesity or migraine. However, the 1‐year prevalence of T2DM (9.9%) in our study was consistent with the 11.0% prevalence of diabetes (type 1 or type 2) in adults aged 20+ years in the 2017–2018 NHANES, suggesting that RWD may accurately capture some target patient populations. ²⁸

Pharmacy claims reflect only prescription medications reimbursed through pharmacy benefits and therefore do not consider over‐the‐counter medications, nutritional or herbal supplements, prescription medications paid for out of pocket, or medications administered directly by HCPs; this approach could underestimate true exposure to victim substrates. Pharmacy claims also do not provide any information about how the medications were consumed. Days' supply are based on prescribed dosage (e.g., 20 tablets b.i.d. = 10 days' supply) rather than actual use. Nevertheless, the large sample sizes in our target populations should still provide a robust estimate of overall prescription medication use.

Because our predicted DDI impact scores have not been validated by comparing them to actual DDIs observed after a new drug is on the market, we hesitate to be more prescriptive in our approach (e.g., drug candidate scoring above X should not be developed any further due to DDI liabilities). The proposed conceptual framework is offered as a starting point to build a quantitative approach based on RWD about the target patient populations to inform decisions related to DDIs that must be made before more information is available. We anticipate that this scoring system could evolve as more drug candidates are evaluated, new risk factors for DDIs are uncovered, and eventually by comparing the predicted DDI impact scores with the prevalence and sequelae of actual DDIs that occur after a drug candidate is approved. Future iterations of this scoring system could also use different approaches to identifying an appropriate control group for risk factors (e.g., controls could be matched on age or other confounders rather than all adults ages 18 or older).

Another limitation to our approach is that we focused only on the role of new drugs in development as perpetrators of DDIs and did not also consider their role as substrates that could be themselves victimized by concomitant medications. We acknowledge this limitation but feel it is appropriate to primarily focus on the role of new drugs as perpetrators rather than victims of DDIs, although both are important. Furthermore, we did not attempt to differentiate our framework based on the estimated strength of the new drug under development as a perpetrator of DDIs. It is possible that a different scoring rubric could be warranted when evaluating stronger or weaker perpetrators (e.g., stronger perpetrators could be awarded more points for various risk factors), although it may be premature to do so without collecting information about predicted DDIs versus actual DDIs.

Our study is also limited by its reliance on the DIDB as a key source of information about DDIs, which is subject to change over time. For example, the April 2023 version of DIDB used in our analyses categorized ibuprofen as a CYP2C9 “sensitive” substrate based on the AUCR of 8.71 reported in a study by Garcia‐Martin et al. ²⁹ The July 2023 version of DIDB categorized ibuprofen as a CYP2C9 “moderate sensitive” substrate based on an AUCR of 2.96, also citing the same study by Garcia‐Martin et al. as the source. It appears that the initial AUCR of 8.71 was based on only a few participants whereas the revised AUCR of 2.96 was more reflective of what was observed in most participants, but no explanation was provided by DIDB for this reclassification.

Similarly, atorvastatin was categorized as a CYP3A “sensitive” substrate in the April 2023 DIDB used in our analyses based on an AUCR of 5.58 reported in a study by Prueksaritanont et al. ³⁰ However, the October 2023 version of DIDB categorized atorvastatin as a CYP3A “moderate sensitive” substrate based on an AUCR of 3.32 reported in a study by Kantola et al. ³¹ Other studies, including those by Mazzu et al. and Jacobson, also reported an AUCR with atorvastatin of less than 5.0, supporting its status as a “moderate sensitive” CYP3A substrate in DIDB. ³² , ³³ Readers are reminded that our understanding of drugs as perpetrators of DDIs can evolve over time with new scientific evidence.

CONCLUSION

Evaluating the potential impact of CYP‐mediated PK‐based DDIs identified early in the development of drug candidates could be informed by analyzing RWD to estimate exposure to victim substrates and prevalence of selected risk factors for DDIs in a large population of individuals with the target indication. This information could be summarized using a predicted DDI impact score to compare drug candidates in development and favor those with lower scores; it could also help inform the development of clinical DDI studies based on the most common co‐exposures. Although this manuscript focuses on DDIs mediated through CYP3A, CYP2D6, or CYP2C9 in patients with T2DM, obesity, or migraine, the proposed methodology could be easily applied to other CYP‐mediated or transporter‐mediated DDIs and other patient populations. We anticipate that this approach would be useful for many stakeholders involved in drug development who must decide whether a new drug candidate remains viable despite evidence that it perpetrates PK‐based DDIs. External stakeholders including regulators, healthcare systems, health plans, HCPs, and researchers may also find this information relevant when evaluating newly approved drugs.

AUTHOR CONTRIBUTIONS

S.D., C.L., C.C., E.W., and V.S. wrote the manuscript. S.D., C.L., C.C., E.W., and V.S. designed the research. S.D. performed the analysis. S.D., C.L., C.C., E.W., and V.S. interpreted the data.

FUNDING INFORMATION

No funding was received for this work.

CONFLICT OF INTEREST STATEMENT

All authors are full‐time employees and equity owners of Pfizer, Inc.

Supporting information

Table S1

Dagenais S, Lee C, Cronenberger C, Wang E, Sahasrabudhe V. Proposing a framework to quantify the potential impact of pharmacokinetic drug–drug interactions caused by a new drug candidate by using real world data about the target patient population. Clin Transl Sci. 2024;17:e13741. doi: 10.1111/cts.13741