Prevalence And Implications Of Cytochrome P450 Substrates.pdf

The Pharmacogenomics Journal (2016), 1–4 © 2016 Macmillan Publishers Limited All rights reserved 1470-269X/16 www.nature.com/tpj

ORIGINAL ARTICLE

Prevalence and implications of cytochrome P450 substrates in Massachusetts hospital discharges TH McCoy1,2, VM Castro1,3,4, A Cagan1,3,4, AM Roberson1,2 and RH Perlis1,2 The cytochrome P450 (CYP450) system of drug-metabolizing enzymes may contribute to individual variation in drug response. We examined prevalence of CYP450 substrates at hospital discharge for patients in two cohorts: insurance claims of Massachusetts residents and the medical records of two academic medical centers. The claims cohort included 47 473 individuals (38.2%) treated with at least one CYP450 2D6, 2C19, 3A4 or 1A2 substrate. The electronic medical records cohort included 45 905 individuals (57.4%) treated with at least one substrate. In adjusted models, substrates of CYP450 2D6 and 2C19 were associated with greater risk for 90-day readmission in both cohorts (odds ratios of 1.104 and 1.128 (P o0.001), respectively). Presence of any CYP450 substrate was associated with increased monthly medical costs (+$397, P o0.003). These analyses of more than 300 000 admissions using two different cohorts and data types indicate that CYP450 substrates are associated with greater readmission rates and greater health-care cost. The Pharmacogenomics Journal advance online publication, 10 May 2016; doi:10.1038/tpj.2016.24

INTRODUCTION Despite recent enthusiasm about risk stratification and personalization of medication prescribing,1,2 the utility of assaying cytochrome P450 (CYP450) metabolism, a well-characterized source of interindividual variation in drug serum levels identified decades ago, remains unclear. The enzymes of the CYP450 system are responsible for phase I metabolism of a large subset of the pharmacopeia, up to 80% by one estimate,3 with the remainder either not metabolized or metabolized by other enzymes.4 The function of CYP450 enzymes is influenced by the presence of other medications that induce or inhibit enzymatic effects, by environmental factors such as diet and by well-described common genetic variation.5 By influencing serum drug levels for a given medication dosage, these factors may contribute to interindividual variation in adverse effects as well as efficacy.6 Interventions aimed at tailoring prescription and dosage based on other medications or genetic variation have been proposed as a means of reducing medication-related adverse outcomes. Among individuals undergoing genetic testing for CYP450 variation, the prevalence of gene–drug interactions has been reported to be 12%.7 However, the extent to which testing could improve outcomes in less-selected populations remains poorly characterized, particularly with regard to the P450 system itself: whereas prevalence of CYP450-mediated drug–drug interactions has been reported in specific clinical populations, the prevalence of substrate drugs alone is far less clear,8 with one estimate exceeding 58% of geriatric patients.9 To estimate the potential relevance and utility of such tests, we examined statewide Massachusetts claims data for hospital discharges, examining readmission and overall monthly medical cost following discharge. To further understand the robustness of association with outcomes using another data type, we applied 1

the same methodology to electronic medical record (EMR) data drawn from two large academic medical centers. Together, these parameters would allow estimation of an upper limit to benefit that could be anticipated by testing for CYP450 variation in the inpatient setting, a crucial value for health systems trying to pursue initiatives in precision medicine. MATERIALS AND METHODS Cohort derivation and overall design Claims cohort: The Massachusetts All-Payer Claims Database (APCD) was established as part of the Massachusetts health-care reform package, and includes all claims data paid for individuals residing in the state regardless of insurance type.10 For the present analysis, we identified all individuals aged ⩾ 40 years with at least one hospital admission in 2012; these include any hospitalization, whether at an academic medical center or a community hospital.11 Individuals with Medicare and no secondary insurance are not included in the APCD, but all claims for those with at least one secondary payer or supplemental plan are included. EMR cohort: We identified all individuals aged ⩾ 40 years admitted to either of the two large Massachusetts academic medical centers in 2007 or 2008. All available individuals in the two cohorts were included in analyses of outcome.

Derivation of clinical variables including outcomes The claims and EMR cohorts include limited sociodemographic data for all individuals, including age, sex and health insurance type. For the claims cohort, discharge medications were defined as any prescription fills within 15 days of hospital discharge, in order to distinguish them from medications used only during hospitalization. For the EMR cohort, medication reconciliation is done at discharge and hence discharge medications were extracted from the record and a datamart was

Center for Experimental Drugs and Diagnostics, Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA; 2Psychiatric and Neurodevelopmental Genetics Unit, Department of Psychiatry, Massachusetts General Hospital, Boston, MA, USA; 3Partners Research Computing, Partners HealthCare System, One Constitution Center, Boston, MA, USA and 4Laboratory of Computer Science and Department of Neurology, Massachusetts General Hospital, Boston, MA, USA. Correspondence: Dr RH Perlis, Center for Experimental Drugs and Diagnostics, Department of Psychiatry, Massachusetts General Hospital, Simches Research Building/MGH, 185 Cambridge Street, 6th Floor, Boston, MA 02114, USA. E-mail: [email protected] Received 25 February 2016; accepted 2 March 2016

Implications of cytochrome P450 substrates TH McCoy et al

2 created using Informatics for Integrating Biology and the Bedside server software (i2b2 v1.5, Boston, MA, USA).12 For both groups, those with at least one medication at discharge were evaluated. Determination of exposure to CYP450 substrates at discharge, according to individual enzyme, utilized algorithms previously reported13 incorporating annotated medication lists of substrates from the Indiana University Department of Medicine.14 (Inducers and inhibitors were not quantified here as they typically are not necessarily substrates themselves.) Analysis of adverse outcomes examined risk for all-cause hospital readmission within 90 days. The Partners institutional review board approved all aspects of this study; as only de-identified data were used, the institutional review board allowed a waiver of requirement for informed consent as detailed by 45 CFR 46.116. The APCD protocol was also approved by the institutional review board of the Massachusetts State Department of Health.

Analysis Logistic regression was used to examine association between presence or absence of individual substrates and hospital readmission within 90 days, a time period selected a priori for consistency with other readmission studies; odds ratios (ORs) were calculated with adjustment for potential confounding features including age, sex, presence of private insurance, total number of medications, any psychiatric ICD9 (International Classification of Diseases, Ninth Revision) code during admission and calculated Charlson comorbidity index. In the claims cohort, association with 30-day cost over up to 1 year of follow-up (excluding medications, and individuals with less than a 30-day follow-up) was modeled using linear regression to yield crude and adjusted coefficients. All regression models considered admissions to be clustered within patients to account for multiple hospitalizations, and hence robust estimates of standard errors were generated. (Limiting analysis to the first hospitalization per subject did not meaningfully change results; not shown.) We present only fully adjusted estimates of effect because exposed and unexposed patients are significantly different in all comparisons, and as such crude ORs would not be interpretable. For ease of interpretability, 30-day costs are presented (rather than per-day costs, for example) and only individuals with at least 30 days of follow-up post hospitalization are included. Analyses utilized Stata 13.1 (Statacorp, College Station, TX, USA).

RESULTS The available prescribing data from the claims cohort included a total of 173 632 hospital discharges involving 124 230 individuals aged ⩾ 40 years. The prescribing data from the EMR cohort included 128 092 index discharges involving 79 990 unique individuals aged ⩾ 40 years. The characteristics of these individuals are summarized in Table 1. For the two cohorts, the proportions of individuals with at least one CYP450 substrate at initial discharge are listed in Table 2. In the claims cohort, 47 473 individuals (38.2%) were treated with at least one substrate drug at discharge; in the EMR cohort, 45 905 (57.4%) were so treated at discharge. For each of 2C19, 2D6 and 3A4, 420% of discharged individuals in both cohorts had at least one relevant substrate. Proportion of total prescriptions for mostprescribed individual medications in the claims database are listed by enzyme system in Table 3. For each enzyme system, we next examined the implication of number of substrates on risk for all-cause hospital readmission as well as per-day health-care costs following discharge (Figure 1 and Table 4). In fully adjusted models, presence of at least one CYP450 substrate was associated with increased 90-day readmission risk in both cohorts (OR of 1.104 in claims and 1.128 in EMR; P o0.001 for both); as were CYP450 2D6 and 1A2 individually. (When total count of CYP450 substrates was included to examine linear association with risk, there was an OR of 1.026 (95% confidence interval 1.017–1.035) in claims and an OR of 1.028 (95% confidence interval 1.020–1.036) in EMR.) For individuals with at least one CYP450 substrate prescribed at discharge, adjusted 30-day health-care costs were significantly greater following discharge ($397.65, robust s.e. $130.60); with CYP450 substrate The Pharmacogenomics Journal (2016), 1 – 4

Table 1. Sociodemographic and clinical features of cohorts by the presence or absence of CYP450 substrate at discharge (a) Claims cohort (n = 124 230 individuals) No CYP substrates N

%

N

%

35 225 24 222 31 688

45.9 31.6 41.3

21 493 18 354 23 974

45.3 38.7 50.5

Mean

s.d.

Mean

s.d.

62.48 5.43 3.31

13.37 7.61 2.22

62 6.46 6.32

12.82 8.56 3.59

Feature Male sex Private insurance Psychiatric ICD9 code

1+ CYP substrates

Age (years) Charlson comorbidity Total drugs prescribed

(b) EMR cohort (n = 79 990 individuals) No CYP substrates N

%

N

%

15 757 16 502 8448

46.2 48.4 24.8

23 292 17 043 16 832

50.7 37.1 36.7

Mean

s.d.

Mean

s.d.

61.62 4.83 5.01

13.68 4.33 3.05

65.59 6.48 8.92

12.98 4.63 3.98

Feature Male sex Private insurance Psychiatric ICD9 code

1+ CYP substrates

Age (years) Charlson comorbidity Total drugs prescribed

Abbreviations: CYP450, cytochrome P450; EMR, electronic medical record; ICD9, International Classification of Diseases, Ninth Revision.

Table 2. Proportion of individuals in each cohort with at least one CYP450 substrate Claims cohort (n = 124 230 individuals)

1A2 2C19 2D6 3A4 Any of the above

EMR cohort (n = 79 990 individuals)

N

%

N

%

12 382 27 714 31 765 27 963 47 473

9.97 22.31 25.57 22.51 38.21

13 469 34 366 30 166 32 409 45 905

16.84 42.96 37.71 40.52 57.39

Abbreviations: CYP450, cytochrome P450; EMR, electronic medical record.

count included, incremental cost was $158.38 (robust s.e. $44.10) per additional substrate in fully adjusted models. DISCUSSION Among hospitalized Massachusetts residents, we found that 40– 60% of individuals were treated with at least one CYP450 substrate at discharge. These substrates were associated with greater subsequent health-care costs as well as hazard for 90-day readmission, after adjustment for multiple potentially confounding illness features including Charlson comorbidity, total medications prescribed and presence of a psychiatric diagnosis. Presence © 2016 Macmillan Publishers Limited

Implications of cytochrome P450 substrates TH McCoy et al

3 Table 3.

a

Most prevalent CYP450 substrates, by CYP450 enzyme

system Medication Acetaminophen Cyclobenzaprine Naproxen Warfarin Omeprazole Citalopram Warfarin Pantoprazole Diazepam Oxycodone Codeine Metoprolol Tramadol Fluoxetine Simvastatin Codeine Amlodipine Atorvastatin Zolpidem Hydrocortisone Trazodone Alprazolam Salmeterol Erythromycin Lidocaine Diazepam Dexamethasone

Enzyme

%

1A2 1A2 1A2 1A2 2C19 2C19 2C19 2C19 2C19 2D6 2D6 2D6 2D6 2D6 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4 3A4

3.5% 0.9% 0.7% 0.5% 1.6% 0.8% 0.5% 0.4% 0.4% 1.9% 1.1% 1.1% 0.7% 0.4% 2.1% 1.1% 1.0% 0.8% 0.6% 0.6% 0.6% 0.4% 0.4% 0.4% 0.4% 0.4% 0.4%

Abbreviation: CYP450, cytochrome P450. aAll medications with prevalence ⩾ 0.4%.

Odds of readmission within 90 days when CYP450 substrates present

Odds of readmission

1.15

1.10 Cohort Claims EMR 1.05

1.00

3A4

2D6

2C19

1A2

Any

Figure 1. Association between presence or absence of cytochrome P450 (CYP450) substrate and 90-day hospital readmission in 2 cohorts. APCD, All-Payer Claims Database; CI, confidence interval; EMR, electronic medical record.

of at least one CYP450 substrate was associated with 410% increase in odds of 90-day readmission—substantially greater than, for example, co-occurring psychiatric illness. Importantly, we observed this risk in two distinctly different data sets, one drawn © 2016 Macmillan Publishers Limited

from electronic medical records and the other from statewide claims data, suggesting the generalizability of the result and its applicability to different forms of pharmacy data. As both cohorts include Massachusetts state residents, some overlap in individuals (but not risk period) is possible, and hence the two cohorts should be considered as different perspectives on similar populations, not totally independent observations. Although estimates of drug–drug interactions in individual patient subsets have been reported,7 there are few estimates of CYP450 substrate medication use in large, unselected clinical populations. In one of the few such studies, among 396 individuals aged ⩾ 60 years (mean age, 72.1 years), 58% were treated with substrates.9 Here, in two large generalizable cohorts utilizing very different data sources, we derived estimates ranging from 57% (for academic medical centers) to 38% (for all state hospitals)— thus, in general our findings support and extend those of the two prior investigations.7,9 Variation in specific CYP450 substrate prevalence may be useful in planning the breadth of pharmacokinetic testing, particularly given their differential association with adverse outcomes and costs. Notably, each CYP450 enzyme individually was associated with increased readmission risk in at least one cohort after adjustment for multiple comparisons as well as for numerous potentially confounding sociodemographic and clinical features. CYP2D6 and 1A2 substrates were associated with greater risk in both cohorts. The consistency of these findings is particularly notable insofar as there are numerous differences between the cohorts: in claims data, prescriptions are recorded when filled, whereas in EMR data, medications are reconciled at discharge (that is, when prescribed). The EMR cohort includes two Boston-based academic medical centers that draw a national patient population but preferentially from Eastern Massachusetts, whereas the claims cohort includes every hospitalization for Massachusetts residents who met inclusion criteria. In particular, the claims cohort includes community hospitals rather than solely tertiary care facilities. As such, the general consistency of results across cohorts should greatly increase confidence in the robustness of the associations. Conversely, the discordance for 2C19 and 3A4 does not necessarily indicate a false positive association, but may reflect differences in the clinical population under study— for example, in terms of race and ethnicity, in terms of admitting diagnosis and in terms of prescribing patterns. They underscore the point that EMR and claims data represent two complementary, but distinct, means of characterizing treatment and outcomes. We note several limitations that bear on interpretation of our results. First, the available data do not allow us to distinguish the dosing of these substrates that might be adjusted to reflect known differences in metabolism or known interactions. Second, the claims cohort includes only a subset of the Medicare population (those individuals with at least one supplemental insurance) that likely contributes to underestimates of CYP450 substrate prevalence as we undersample older patients. Moreover, although the impact of CYP450 variation on blood levels is generally well established,3 and dosing guidelines have been created for some medications known to be substrates, for many putative CYP450 substrates the association between drug level, safety and efficacy is not clear. For some enzymes, including 3A4, the functional implications are unclear, although genetic variation is reflected in FDA (Food and Drug Administration) and other drug labels.15 Although our analysis relies on a curated list of common CYP450 substrates, it does not distinguish between degree of dependence on these enzymes that, in most cases, is not established. We also emphasize that association does not necessarily reflect causation, and that the associations we detect may still reflect residual confounding. Our results model the impact of general or specific comorbidity (for example, Charlson comorbidity index, psychiatric diagnoses or individual diagnostic categories), increasing confidence that the associations are unlikely to reflect The Pharmacogenomics Journal (2016), 1 – 4

Implications of cytochrome P450 substrates TH McCoy et al

4 Table 4.

Association between individual CYP450 substrates and odds of 90-day readmission, adjusted for sociodemographic and clinical featuresa Claims cohort OR (readmit)

2D6 2C19 3A4 1A2 Any

b

1.073 1.042 1.107b 1.116b 1.104b

EMR cohort 95% CI

1.038–1.110 1.007–1.078 1.068–1.146 1.067–1.166 1.071–1.139

OR (readmit) b

1.076 1.121b 1.023 1.098b 1.128b

Additional medical cost per 30 days (claims) 95% CI

Mean

s.e.

1.042–1.111 1.085–1.158 0.990–1.058 1.058–1.141 1.090–1.169

$175.45 $300.00 $569.17b $260.99 $397.65b

$157.48 $150.92 $153.22 $215.53 $130.60

Abbreviations: CI, confidence interval; CYP450, cytochrome P450; EMR, electronic medical record; OR, odds ratio. aAdjusted models include age, sex, insurance type, presence/absence of a psychiatric diagnosis by ICD9 (International Classification of Diseases, Ninth Revision) code, age-adjusted Charlson comorbidity index and total medications prescribed. bOdds ratios or regression coefficients significant at Po 0.003.

confounding by indication. Still, it remains possible that a particular diagnosis or combination of diagnoses may contribute to cost and readmission risk. Once again, the breadth and size of the study populations, in addition to inclusion of comorbidities as covariates, diminishes this possibility. Moreover, the most common substrates are not limited to a given disorder or set of disorders, underscoring the need for awareness of an entire medication list and the potential utility of strategies to efficiently summarize the implications of such lists.13,16–18 Taken together, our results suggest that the presence of particular CYP450 substrates among patients at hospital discharge is associated with greater risk for 90-day readmission as well as increased costs, even after accounting for overall burden of medical illness, across two types of large cohorts. As such, they suggest the potential usefulness of investigating the relevance of CYP450 characterization in high-risk patient populations. Beyond replication in other large populations, randomized controlled trials will be required to precisely establish the potential benefit, if any, of strategies to personalize dosing and/or minimize interactions based on such testing. CONFLICT OF INTEREST RHP has served on advisory boards or provided consulting to Genomind, Healthrageous, Perfect Health, Pfizer, Psybrain and RIDVentures. The other authors declare no conflict of interest.

ACKNOWLEDGMENTS Outside funding was not received for this study. We acknowledge the assistance of Massachusetts Department of Health staff in the generation of the All-Payer Claims Database.

REFERENCES 1 Office of the Press Secretary 2015. FACT SHEET: President Obama’s Precision Medicine Initiative. In: Statements and Releases. The White House. 2 IOM (Institute of Medicine). Policy issues in the development of personalized medicine in oncology. In: M Patlak, L Levit (eds). Workshop Summary. National Academies Press: Washington, DC, 2010.

The Pharmacogenomics Journal (2016), 1 – 4

3 Zhou SF, Liu JP, Chowbay B. Polymorphism of human cytochrome P450 enzymes and its clinical impact. Drug Metab Rev 2009; 41: 89–295. 4 Wienkers L, Heath T. Predicting in vivo drug interactions from in vitro drug discovery data. Nat Rev Drug Discov 2005; 4: 825–833. 5 Wang L, McLeod HL, Weinshilboum RM. Genomics and drug response. N Engl J Med 2011; 364: 1144–1153. 6 Tannenbaum C, Sheehan NL. Understanding and preventing drug-drug and druggene interactions. Expert Rev Clin Pharmacol 2014; 7: 533–544. 7 Verbeurgt P, Mamiya T, Oesterheld J. How common are drug and gene interactions? Prevalence in a sample of 1143 patients with CYP2C9, CYP2C19 and CYP2D6 genotyping. Pharmacogenomics 2014; 15: 655–665. 8 Kerr KP, Mate KE, Magin PJ, Marley J, Stocks NP, Disler P et al. The prevalence of co-prescription of clinically relevant CYP enzyme inhibitor and substrate drugs in community-dwelling elderly Australians. J Clin Pharm Ther 2014; 39: 383–389. 9 Cabrera MAS, Dip RM, Furlan MO, Rodrigues SL. Use of drugs that act on the cytochrome P450 system in the elderly. Clinics 2009; 64: 273–278. 10 Reis M, Nahmiash D. Validation of the indicators of abuse (IOA) screen. Gerontologist 1998; 38: 471–480. 11 Overview of the Massachusetts All-Payer Claims Database. Center for Health Information and Analysis. Available from http://www.chiamass.gov/assets/docs/p/ apcd/APCD-White-Paper-2015.pdf, 2015; last accessed on 27 November 2015. 12 Murphy SN, Mendis M, Hackett K, Kuttan R, Pan W, Phillips LC et al. Architecture of the open-source clinical research chart from Informatics for Integrating Biology and the Bedside. AMIA Annu Symp Proc 2007: 548–552. 13 Castro VM, McCoy TH, Cagan A, Rosenfield HR, Murphy SN, Churchill SE et al. Stratification of risk for hospital admissions for injury related to fall: cohort study. BMJ 2014; 349: g5863. 14 Indiana University Department of Medicine Clinical Pharmacology. P450 Drug Interaction Table. Available from http://medicine.iupui.edu/clinpharm/ddis/maintable/, 2015; last accessed on 27 November 2015. 15 PharmGKB. Gene CYP3A4 Drug Labels. Available from https://www.pharmgkb. org/gene/PA130, 2015; last accessed on 27 November 2015. 16 Hilmer S, Mager D, Simonsick E, Cao Y, Ling S, Windham B et al. A drug burden index to define the functional burden of medications in older people. Arch Intern Med 2007; 167: 781–787. 17 Cancelli I, Gigli GL, Plani A, Zanchettin B, Janes F, Rinaldi A et al. Drugs with anticholinergic properties as a risk factor for cognitive impairment in elderly people: a population-based study. J Clin Psychopharmacol 2008; 28: 654–659. 18 Hatah E, Braund R, Tordoff J, Duffull S. A systematic review and meta-analysis of pharmacist-led fee-for-services medication review. Br J Clin Pharmacol 2014; 77: 102–115.

© 2016 Macmillan Publishers Limited