Next Article in Journal
Treatment Landscape of Nonmetastatic Castration-Resistant Prostate Cancer: A Window of Opportunity
Next Article in Special Issue
The Contribution of Pharmacogenetic Drug Interactions to 90-Day Hospital Readmissions: Preliminary Results from a Real-World Healthcare System
Previous Article in Journal
A Focus Group Study of Perceptions of Genetic Risk Disclosure in Members of the Public in Sweden: “I’ll Phone the Five Closest Ones, but What Happens to the Other Ten?”
Previous Article in Special Issue
A Cost–Consequence Analysis of Preemptive SLCO1B1 Testing for Statin Myopathy Risk Compared to Usual Care
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
JPM
Volume 11
Issue 11
10.3390/jpm11111192
jpm-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Unmanaged Pharmacogenomic and Drug Interaction Risk Associations with Hospital Length of Stay among Medicare Advantage Members with COVID-19: A Retrospective Cohort Study

by
Kristine Ashcraft
1,*,
Chad Moretz
1,
Chantelle Schenning
1,
Susan Rojahn
1,
Kae Vines Tanudtanud
2,
Gwyn Omar Magoncia
2,
Justine Reyes
2,
Bernardo Marquez
2,
Yinglong Guo
2,
Elif Tokar Erdemir
2 and
Taryn O. Hall
2
1
Invitae Corporation, San Francisco, CA 94103, USA
2
OptumLabs at UnitedHealth Group, Minnetonka, MN 55343, USA
*
Author to whom correspondence should be addressed.
J. Pers. Med. 2021, 11(11), 1192; https://doi.org/10.3390/jpm11111192
Submission received: 15 October 2021 / Revised: 2 November 2021 / Accepted: 9 November 2021 / Published: 12 November 2021
(This article belongs to the Special Issue Pharmacogenetic Testing in Primary Care and Prevention)
Browse Figures
Review Reports Versions Notes

Abstract

:
Unmanaged pharmacogenomic and drug interaction risk can lengthen hospitalization and may have influenced the severe health outcomes seen in some COVID-19 patients. To determine if unmanaged pharmacogenomic and drug interaction risks were associated with longer lengths of stay (LOS) among patients hospitalized with COVID-19, we retrospectively reviewed medical and pharmacy claims from 6025 Medicare Advantage members hospitalized with COVID-19. Patients with a moderate or high pharmacogenetic interaction probability (PIP), which indicates the likelihood that testing would identify one or more clinically actionable gene–drug or gene–drug–drug interactions, were hospitalized for 9% (CI: 4–15%; p < 0.001) and 16% longer (CI: 8–24%; p < 0.001), respectively, compared to those with low PIP. Risk adjustment factor (RAF) score, a commonly used measure of disease burden, was not associated with LOS. High PIP was significantly associated with 12–22% longer LOS compared to low PIP in patients with hypertension, hyperlipidemia, diabetes, or chronic obstructive pulmonary disease (COPD). A greater drug–drug interaction risk was associated with 10% longer LOS among patients with two or three chronic conditions. Thus, unmanaged pharmacogenomic risk was associated with longer LOS in these patients and managing this risk has the potential to reduce LOS in severely ill patients, especially those with chronic conditions.

1. Introduction

Hospitalizations represent nearly one-third of all U.S. healthcare expenditures [1]. Patient length of stay (LOS) is a longstanding measure of hospital efficiency as a longer LOS is associated with greater resource utilization and higher costs [2,3,4,5]. Among many factors associated with longer LOS are adverse drug reactions (ADRs), which are preventable with appropriate medication management [6,7,8]. Patient risk factors for ADR are inappropriate polypharmacy as well as the harboring of genetic variants that affect medication responses. Pharmacogenomics-guided medication management could therefore reduce both ADRs and LOS. Studies suggest that >99% of individuals harbor at least one variant associated with an atypical drug response and 24% of individuals may currently take a drug to which they may have an atypical response [9,10,11]. Given that a greater pharmacogenomic variant burden has been associated with longer LOS, [12,13] unmanaged pharmacogenomic risk may be an intervenable vulnerability for reducing LOS among high-risk patients.
Population ADR risk management could benefit hospital systems at any time and may benefit hospitals even more so when there is additional population-wide stress on the hospital system, such as the COVID-19 pandemic. The COVID-19 pandemic placed unprecedented demands for hospitals across the globe, and patient demand for hospitalization exceeded hospital resources in several regions [14,15,16,17,18]. In the US, roughly half of those hospitalized with COVID-19 have been Medicare and Medicare Advantage beneficiaries and the majority of these beneficiaries have been hospitalized for more than 5 days [19,20]. Medication management interventions may especially help the Medicare population as 54% of patients 65 and older report taking four or more prescription drugs daily [21]. This increases their risk for ADR-related hospitalization, which is reported to be four times greater among patients who are 65 years of age or older compared to younger patients [22]. Further, polypharmacy is a reported risk factor for developing COVID-19 [23]. Thus, hospitalized COVID-19 patients may be at heightened risk for ADR.
A recent study found that nearly 90% of patients hospitalized with COVID-19 had at least one medication order that could have been informed by pharmacogenomic testing and nearly one-quarter had orders for four or more actionable medications [24]. However, it is not clear whether pharmacogenomic risk or drug–drug interaction risk affects LOS among COVID-19 patients. To evaluate the impact of pharmacogenomic risk on LOS among high-risk patients, we conducted a retrospective analysis of administrative medical and pharmacy claims among Medicare Advantage members hospitalized with COVID-19. The main objective of this study was to determine if pharmacogenomic or drug–drug interaction (DDI) risk among Medicare beneficiaries hospitalized with COVID-19 was associated with LOS. We further examined this potential association by the number of diagnosed chronic conditions as well as four specific chronic conditions: chronic obstructive pulmonary disease (COPD), diabetes, hyperlipidemia, and hypertension.

2. Methods

2.1. Ethics Approval Statement

This study was granted a waiver of authorization by the institutional review board of the UnitedHealth Group Office of Human Research Affairs (FWA00028881, approved 23 February 2021) because the work is limited to a secondary analysis of de-identified claims data, and under the applicable exemption criteria, patient consent and HIPAA authorization are not required.

2.2. Data Source and Study Population

Data were obtained from an administrative database of medical and pharmacy claims from a large U.S. health insurance company. The claims include the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis codes, the National Drug Codes (NDC), patient demographic data, and other information.
Claims data for 10,206 Medicare Advantage members admitted to a hospital with COVID-19 between 1 January 2020 and 30 June 2020 were reviewed. COVID-19 diagnoses were identified through SARS-CoV-2 lab test records indicated by the Logical Observation Identifiers Names and Codes (LOINC) organization’s guidance for mapping SARS-CoV-2 and COVID-19-related LOINC terms [25]. Test information provided via LOINC terms indicated the test type (antibody, RT-PCR, etc.) as well as the test result (detected, not detected, not given/cancelled). Suspected COVID-19 inpatient cases were subjected to daily manual review by clinical staff via clinical notes to determine an individual’s COVID-19 status, flagging it as either negative, confirmed, presumed positive, or needing clinical review. All findings, except for confirmed status, were reviewed daily. Only patients’ first COVID-19 admission was included and evaluated, and patients who died during their hospital stay (n = 1604, or 16% of patients reviewed) were excluded from the study to prevent death from defining the outcome (LOS). Only patients continuously enrolled in Medicare Advantage Prescription Drug plans for at least 12 months prior to the index COVID-19 admission were eligible for inclusion. Pharmacogenomic testing was not an inclusion criterion for this study; instead, unmanaged pharmacogenetic risk was assessed as described below.

2.3. Independent and Outcome Variables

The primary independent variables were pharmacogenetic interaction probability (PIP) and DDI severity, as determined by Invitae’s YouScript clinical decision support tool. PIP is the probability that pharmacogenomic testing of 14 genes (CYP2C19, CYP2C9, CYP2D6, CYP3A4, CYP3A5, CYP2B6, CYP4F2, DPYD, HLA-B*57:01, IFNL3, SLCO1B1, TPMT, UGT1A1, and VKORC1) will result in the detection of one or more clinically significant pharmacogenomic interactions involving one or more medications on the patient’s current medication list. PIP was calculated based on publicly available phenotype prevalence data in the U.S. population and pharmacogenomic interaction data, which encompasses clinically actionable drug–drug or drug–drug–gene interactions in CPIC Guidelines and FDA labeling. More robust descriptions of YouScript are available in US patents [26,27]. PIP was categorized into the following: (1) low (PIP ≤ 25%), (2) moderate, (PIP 26–50%), and (3) high (PIP > 50%).
DDI indicates the potential for adverse interactions among 2 or more medications prescribed to a patient, irrespective of the patient’s pharmacogenomic profile. DDI severity was measured on an ordinal scale of increasing severity: minimal, minor, moderate, major, and contraindicated, with moderate or higher being tied to evidence indicating a need to consider changing a drug or dose. Among patients with multiple DDIs in their drug regimen, only the most severe DDI was used in this study.
All medications active or prescribed within 30 days prior to and during the first COVID-19 admission were screened for PIP and DDIs. Other independent variables were age, gender, race/ethnicity, residential location, median income, risk adjustment factor (RAF), hierarchical condition category (HCC), and Special Needs Plan (SNP) type. RAF scores and HCC, which are measures of disease burden used to determine the likelihood of a patient’s need for medical care, were based on the one-year window prior to the patient’s admission month. HCC are simplified diagnosis categories that are constructed using a standardized ICD-10 code mapping determined by the Centers for Medicare and Medicaid Services (CMS). HCC counts indicate the number of chronic conditions and were categorized as 0 or 1, 2 or 3, 4 or 5, or 6 or more, following previously reported CMS categorizations [28]. HCC, along with demographic data, were used in an algorithm to assign an RAF score, which is a measure of patient complexity [29]. SNPs are Medicare plan types restricted to individuals with specific diseases or characteristics. SNP types were: (1) chronic conditions SNP (C-SNP), for individuals with severe or disabling chronic conditions; (2) dual-eligible SNP (D-SNP), for individuals entitled to both Medicare and state Medicaid; and (3) institutional SNP (I-SNP), for individuals who need or are expected to need services in a long-term care skilled nursing facility or a similar high-level care facility for 90 days or longer [30]. The outcome variable of interest was LOS (in days).

2.4. Statistical Analysis

Zero-truncated negative binomial analyses (ZTNB) were used to assess the effects of independent variables on LOS. Two sets of ZTNB analyses were carried out. The first focused on the entire cohort and included the following models: (i) the baseline model that accounted for age, gender, PIP and DDI; and (ii) the regression model that adjusted for the variables identified by least absolute shrinkage and selection operator (LASSO) as covariates with a strong association with LOS [31]. The second set of analyses were subgroup analyses to limit the effects of underlying disease through model fitting and variable selection for each of the following: (1) patients stratified by HCC counts (i.e., number of chronic conditions); (2) patients stratified by each of the following chronic conditions regardless of other comorbidities: (i) COPD, (ii) diabetes, (iii) hyperlipidemia, and (iv) hypertension.
In both sets of ZTNB analyses, the point and confidence-interval estimate of the rate ratios and the corresponding p-values resulting from the fitted models were obtained. No medications were included as control variables in the regression models. Statistical significance was set at p < 0.05. Additionally, descriptive statistics were applied to profile the study population, including count, percentage, mean, and standard deviation (SD). Patients with any missing data, except for race/ethnicity, were excluded from the study cohort. Missing data on race/ethnicity are indicated as “Unknown” in Table 1. All analyses were conducted using the statistical computing software R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria) [32].

3. Results

A cohort of 6025 individuals met all inclusion criteria, including hospitalization with COVID-19 and 12-month continuous Medicare Advantage coverage (Figure 1). Most patients were female (61%), had a mean age of 77 years (SD, 11 years), were of white (non-Hispanic) ancestry (62%), lived in urban areas (47%), and had an average median household income of USD63,027 (SD, USD17,435) (Table 1). One-third of patients were enrolled in an I-SNP (34%), and 76% had two or more chronic conditions. More than half of the study population had hyperlipidemia (58%) or diabetes (52%).
Overall, 42% (2511 of 6025) of patients had a moderate or high PIP (Table 1), indicating that among these patients, there was a >25% chance that pharmacogenomic testing would reveal a clinically significant gene–drug or gene–drug–drug interaction. Further, 48% (2915 of 6025) had a moderate, major, or contraindicated DDI. Nearly one-quarter (24%) of patients (1422/6025) were prescribed metoprolol, a beta blocker associated with a minimum PIP of 46% (Table S1). Other frequently prescribed drugs associated with moderate or high PIP were proton-pump inhibitors pantoprazole (12%, or 730/6025 patients) and omeprazole (9%, or 571/6025 patients), and serotonin-reuptake inhibitors escitalopram (6%, or 365/6025 patients) and citalopram (3%, or 207/6025; Table S1).
Patients were hospitalized for a mean of 13 days (SD, 11 days; Table 1). Patients with moderate or high PIP were hospitalized for at least one day longer on average compared to those with low PIP, while LOS for patients with moderate, major, or contraindication DDI was similar to those with minimal or minor DDI (Figure 2 and Table S2). In unadjusted analyses, the LOS for patients with moderate and high PIP were 9% (CI: 4–15%; p < 0.001) and 16% longer (CI: 8–24%; p < 0.001), respectively, compared to those with low PIP (Table S3). Moreover, with other variables held constant, the LOS for male patients was 8% longer (CI: 3–13%; p = 0.001) compared to female patients, and for each one-year age increment, the expected LOS was 0.5% longer (CI: 0.3–0.7%; p < 0.001).
Among the entire cohort, adjusting for potential confounders did not attenuate the association between PIP and LOS (Table 2). When patients were stratified by HCC count (Table 2 and Table S4), PIP was significantly associated with LOS among all patient groups except those with two or three conditions (Table 2). In particular, among patients with zero or one, or four or five conditions, those with moderate PIP had 15% (CI: 4–28%; p = 0.007) and 13% (CI: 2–25%; p = 0.019) longer LOS, respectively, compared to those with low PIP. Moreover, among patients with zero or one, or six or more conditions, those with high PIP had 39% (CI: 15–67%; p < 0.001) and 16% (CI: 3–31%; p = 0.014) longer LOS, respectively, compared to those with low PIP. Notably, across all patients, moderate and high PIP scores were significant predictors of LOS but RAF scores were not (Table S5).
Varying associations were observed for LOS, HCC count, and DDI. Among patients with zero or one HCC, those with moderate or greater DDI had a 9% shorter LOS (CI: 0−18%; p = 0.045) compared to those with minimal or minor DDI (Table 2). Among patients with two or three HCC, those with moderate or more severe DDI had 10% longer LOS (CI: 2–18%; p = 0.010) compared to those with minimal or minor DDI. Lastly, across all HCC categories, enrollment in an I-SNP was associated with a 23–34% shorter LOS compared to those not enrolled in I-SNP (Tables S6–S10).
To explore the impact of individual comorbidities on LOS, patients were stratified by four pre-existing chronic conditions: COPD, diabetes, hyperlipidemia, and hypertension. Patients with high PIP most frequently had hyperlipidemia (Table S10). Within each chronic condition subpopulation (controlled for the other chronic conditions), patients with high PIP were hospitalized 12–22% longer compared to those with low PIP (Table 2). In particular, among patients with hypertension, those with high PIP were hospitalized 2.5 days longer than those with low PIP (Table 3). Moreover, patients diagnosed with hyperlipidemia or COPD who had moderate PIP had longer LOS by 8% (CI: 2–15%; p = 0.013) and 13% (CI: 3–24%; p = 0.009), respectively, compared to those with low PIP (Table 2).
When considering the potential impact of DDI on LOS, we found that patients diagnosed with diabetes and at least moderate or more severe DDI had 7% longer LOS (CI: 1–13%; p = 0.031) compared to those with minimal or minor DDI (Table 2). Lastly, patients in each chronic condition group who were enrolled in I-SNP had a 25–29% shorter LOS than those not enrolled (Tables S11–S14).

4. Discussion

In this study, we showed that greater pharmacogenomic risk was associated with longer LOS among Medicare Advantage members hospitalized with COVID-19. In particular, high PIP patients, who have a 50% or greater likelihood of harboring a genetics-driven atypical response to current medications, were hospitalized for almost two days longer than those with low PIP. Given the known associations between ADRs and LOS, [8] our findings suggest that incorporating pharmacogenomic testing into the clinical management of high-risk COVID-19 patients may shorten hospital LOS. Critically, to potentially reduce LOS, testing would have to occur prior to admission to allow for testing results to be incorporated into a patient’s medication management.
In a novel pandemic such as the COVID-19 crisis, the sudden and intense demands on hospitals present a clear need to reduce patient LOS. Should proactive pharmacogenomics-informed medication management for COVID-19 patients successfully reduce LOS, even non-COVID-19 patients could benefit from the reduced burden on overwhelmed healthcare systems that have delayed some patients’ care [17,33,34,35]. Moreover, proactive management of pharmacogenomic risk could benefit many different patient populations through reducing hospital LOS. A study of patients hospitalized for a variety of conditions (mostly cancer and cardiovascular and kidney conditions) found that a higher number of CPIC-defined actionable variants was associated with a longer LOS [13]. Another study of patients with major depressive disorder showed that patients who harbored genetic variants associated with the reduced metabolism of some psychotropic medications and other drugs had longer LOS in psychiatric hospitals than patients without those variants [12].
In addition to potential benefits to patient health, pharmacogenomic intervention could reduce healthcare costs. Estimating the potential cost savings from a reduction in LOS is challenging; however, recent data show that the average Medicare payment per COVID-19 hospitalization is USD23,587, [19] which is USD8688 higher than the average Medicare Advantage hospitalization (USD14,900) [36]. Considering the higher cost associated with complex COVID-19 patients, pharmacogenomic-informed medication management of Medicare beneficiaries could result in substantial savings for patients, hospitals, and the Medicare program.
Overall, we found that almost half of Medicare Advantage members hospitalized with COVID-19 could benefit from improved medication management, as 42% had a moderate or high PIP and 48% had a moderate, major, or contraindicated DDI for current medications [9,10,11]. Further, despite mandates to flag DDIs in electronic health records and pharmacy systems, we found that one-third of our cohort was prescribed medications for which the risks likely exceeded the benefits. Thus, comprehensive medication management is underutilized among Medicare members, leaving many at risk for ADRs. Incorporating pharmacy expertise into the clinical management of patients could reduce this risk, as pharmacist participation has been shown to reduce ADRs, LOS, and mortality [37].
Several findings in our subpopulation analyses warrant further study. When patients were stratified by number of chronic conditions, patients with zero or one HCC and high PIP had a greater LOS increase than those with six or more HCC (39% versus 16% increase), which may be due to uncontrolled confounders such as the severity of the COVID-19 infection. In addition, we did not observe a significant association between PIP and LOS among patients with two or three chronic conditions, despite finding a significant association for all other HCC counts. A larger sample size may reveal significant findings for those with two or three HCC and possibly identify additional confounders. In addition, the finding that patients with zero or one HCC and more severe DDI had shorter LOS compared to those with less severe DDI requires further study that examines the specific medications used by those patients. Separately, for each specific chronic condition (COPD, diabetes, hyperlipidemia, and hypertension), we found patients with high PIP had significantly longer LOS compared to those with low PIP. Further research may explore the impact of pharmacogenomic risk and ADRs among COVID-19 patients with these comorbidities. Finally, in both subpopulation analyses, I-SNP enrollment was associated with shorter LOS. This may be a result of potential ADRs receiving more attention in institutional settings, or conversely, an effect of more expedient discharge due to enhanced care coordination with institutional settings.
LOS is a long-standing concern for healthcare organization performance, patient outcomes, and expenditures [38,39,40]. Although this study focused on COVID-19, population-based pharmacogenomic testing that improves medication management could broadly reduce ADRs, which are reported to cost the USD528 billion per year, exceeding the costs associated with any major disease or the actual drugs prescribed [41]. A one-time test could aid medication optimization and potentially reduce healthcare costs throughout each patient’s life. In addition, future research should further explore the potential for PIP to predict LOS, as moderate and high PIP were both significant predictors of LOS, whereas RAF score, a metric used by CMS for benchmarking expenditures and allocating payments to healthcare providers, was not (Table S5).
Limitations to this study include that although this study showed an association between PIP and increased LOS, it was not a study demonstrating that increased LOS among COVID-19 patients was a direct result of adverse drug events that could have been prevented by proactive pharmacogenomic testing. Additional work, including prospective studies with large, controlled cohorts, is needed to demonstrate the impact of pharmacogenomics-guided medication management on hospital LOS. In addition, the analyses did not account for some potential confounders, such as the severity of COVID-19, the medication regimen complexity (including the types and counts of prescribed drugs, medication dosage or frequency), patient kidney or hepatic function, or patient response to active medications. Other potential confounders include the difference in racial and ethnic composition of the study cohort compared to the broad U.S. population frequencies used in the development of the PIP score. A general limitation of observational studies using claims data is that the presence of a code does not guarantee that a patient has been diagnosed or has received or taken a prescribed medication. Finally, our findings may have been affected by a survivor bias arising from the exclusion of patients who died during the study period.

5. Conclusions

Among otherwise risk-matched patients with COVID-19, a moderate or high pharmacogenomic risk was associated with longer LOS. Addressing pharmacogenomic risk prior to hospitalization for COVID-19 or other serious diseases may reduce LOS, decrease healthcare costs, and improve risk predictions in Medicare members, especially those with COPD, diabetes, hyperlipidemia, or hypertension.

Supplementary Materials

The following are available online at https://www.mdpi.com/article/10.3390/jpm11111192/s1: Table S1: Frequently prescribed drugs with moderate or high PIP among Medicare Advantage members hospitalized with COVID-19; Table S2: Distribution of length of stay among Medicare Advantage members hospitalized with COVID-19 by pharmacogenomic risk and drug–drug interaction; Table S3: Length-of-stay ratios of covariates in the baseline ZTNB regression model; Table S4: Distribution of patients by HCC count subpopulation, pharmacogenomic risk, and drug–drug interaction categories; Table S5: Length-of-stay ratios in the ZTNB regression model for all Medicare Advantage members hospitalized with COVID-19; Table S6: Expected length-of-stay ratios in the ZTNB regression model for patients with zero or one HCC; Table S7: Expected length-of-stay ratios in the ZTNB regression model for patients with two or three HCC; Table S8: Expected length-of-stay ratios in the ZTNB regression model for patients with four or five HCC; Table S9: Expected length-of-stay ratios in the ZTNB regression model for patients with six or more chronic conditions; Table S10: Distribution of patients by chronic condition subpopulation, pharmacogenomic risk, and drug–drug interaction risk; Table S11: Expected length-of-stay ratios in the ZTNB regression model for patients with chronic obstructive pulmonary disease; Table S12: Expected length-of-stay ratios in the ZTNB regression model for patients with diabetes; Table S13: Expected length-of-stay ratios in the ZTNB regression model for patients with hyperlipidemia; Table S14: Expected length-of-stay ratios in the ZTNB regression model for patients with hypertension.

Author Contributions

Conceptualization, K.A., K.V.T., Y.G., E.T.E. and T.O.H.; Data curation, K.V.T., G.O.M., J.R. and B.M.; Formal analysis, K.V.T., G.O.M., J.R. and T.O.H.; Methodology, K.A., C.M., C.S., K.V.T., G.O.M., J.R., B.M, Y.G., E.T.E. and T.O.H.; Writing—original draft, K.A., C.M., C.S., S.R. and T.O.H.; Writing—review and editing, K.A., C.M., C.S., S.R., K.V.T., G.O.M., J.R., B.M., Y.G., E.T.E. and T.O.H. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

The data analyzed in this study was obtained from UnitedHealth Group Clinical Discovery Portal. The data are proprietary and are not available for public use, but, under certain conditions, may be made available to editors and their approved auditors under a data-use agreement to confirm the findings of the current study. Further inquiries can be directed to Lauren Mihajlov of UnitedHealth Group.

Acknowledgments

The authors thank Kathy Tzeng of UnitedHealth Group (UHG) for the initial development of this project, Nicolas Moyer of Invitae for assisting with analyses, and Elaine Preimesberger of UHG and Melinda Tran of Invitae for their review and feedback on this draft.

Conflicts of Interest

Authors K.A., C.M., C.S. and S.R. are employees of the genetic testing company Invitae. Authors G.O.M., J.R., B.M., Y.G., E.T.E. and T.O.H. are employees of UHG, and authors K.V.T. and E.T.E. are former employees of UHG.

Abbreviations

COVID-19Coronavirus-induced disease 2019
CPICClinical Pharmacogenetics Implementation Consortium
DDIDrug–drug interaction
FDAU.S. Food and Drug Administration
HCCHierarchical conditions category
LOSLength of stay
PIPPharmacogenetic interaction probability
RAFRisk adjustment factor
SNPSpecial needs plan

References

  1. Mitchell, E.M. Concentration of Health Expenditures and Selected Characteristics of High Spenders, U.S. Civilian Noninstitutionalized Population, 2016. In Statistical Brief (Medical Expenditure Panel Survey (US)); Agency for Healthcare Research and Quality (US): Rockville, MD, USA, 2019. [Google Scholar]
  2. Bueno, H. Trends in Length of Stay and Short-term Outcomes Among Medicare Patients Hospitalized for Heart Failure, 1993–2006. JAMA 2010, 303, 2141–2147. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  3. Lu, M.; Sajobi, T.; Lucyk, K.; Lorenzetti, D.L.; Quan, H. Systematic Review of Risk Adjustment Models of Hospital Length of Stay (LOS). Med. Care 2015, 53, 355–365. [Google Scholar] [CrossRef] [PubMed]
  4. Salim, S.A.; Elmaraezy, A.; Pamarthy, A.; Thongprayoon, C.; Cheungpasitporn, W.; Palabindala, V. Impact of hospitalists on the efficiency of inpatient care and patient satisfaction: A systematic review and meta-analysis. J. Community Hosp. Intern. Med. Perspect. 2019, 9, 121–134. [Google Scholar] [CrossRef] [Green Version]
  5. Ward, C.; Patel, V.; Elsaid, M.I.; Jaisinghani, P.; Sharma, R. A case-control study of length of stay outliers. Am. J. Manag. Care 2021, 27, e66–e71. [Google Scholar] [CrossRef]
  6. Lazarou, J.; Pomeranz, B.H.; Corey, P.N. Incidence of adverse drug reactions in hospitalized patients: A meta-analysis of pro-spective studies. JAMA 1998, 279, 1200–1205. [Google Scholar] [CrossRef] [PubMed]
  7. Sultana, J.; Cutroneo, P.; Trifirò, G. Clinical and economic burden of adverse drug reactions. J. Pharmacol. Pharmacother. 2013, 4 (Suppl. S1), 73–77. [Google Scholar] [CrossRef] [Green Version]
  8. Formica, D.; Sultana, J.; Cutroneo, P.M.; Lucchesi, S.; Angelica, R.; Crisafulli, S.; Ingrasciotta, Y.; Salvo, F.; Spina, E.; Trifirò, G. The economic burden of preventable adverse drug reactions: A systematic review of observational studies. Expert Opin. Drug Saf. 2018, 17, 681–695. [Google Scholar] [CrossRef]
  9. Ji, Y.; Skierka, J.M.; Blommel, J.H.; Moore, B.E.; VanCuyk, D.L.; Bruflat, J.K.; Peterson, L.M.; Veldhuizen, T.L.; Fadra, N.; Peterson, S.E.; et al. Preemptive Pharmacogenomic Testing for Precision Medicine: A Comprehensive Anal-ysis of Five Actionable Pharmacogenomic Genes Using Next-Generation DNA Sequencing and a Customized CYP2D6 Genotyping Cascade. J. Mol. Diagn. 2016, 18, 438–445. [Google Scholar] [CrossRef] [Green Version]
  10. Chanfreau-Coffinier, C.; Hull, L.E.; Lynch, J.A.; Duvall, S.L.; Damrauer, S.M.; Cunningham, F.E.; Voight, B.F.; Matheny, M.; Oslin, D.W.; Icardi, M.S.; 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, e195345. [Google Scholar] [CrossRef]
  11. McInnes, G.; Lavertu, A.; Sangkuhl, K.; Klein, T.E.; Whirl-Carrillo, M.; Altman, R.B. Pharmacogenetics at Scale: An Analysis of the UK Biobank. Clin. Pharmacol. Ther. 2021, 109, 1528–1537. [Google Scholar] [CrossRef]
  12. Ruaño, G.; Szarek, B.L.; Villagra, D.; Gorowski, K.; Kocherla, M.; Seip, R.L.; Goethe, J.W.; Schwartz, H.I. Length of psychiatric hospitalization is correlated with CYP2D6 functional status in inpatients with major depressive disorder. Biomark. Med. 2013, 7, 429–439. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  13. Finkelstein, J.; Zhang, F.; Cabrera, M. Association Between Number of Actionable Pharmacogenetic Variants and Length of Hospital Stay. Stud. Health Technol. Inform. 2020, 272, 195–198. [Google Scholar]
  14. Shadmi, E.; Chen, Y.; Dourado, I.; Faran-Perach, I.; Furler, J.; Hangoma, P.; Hanvoravongchai, P.; Obando, C.; Petrosyan, V.; Rao, K.D.; et al. Health equity and COVID-19: Global perspectives. Int. J. Equity Health 2020, 19, 1–16. [Google Scholar] [CrossRef] [PubMed]
  15. Lal, A.; Erondu, N.A.; Heymann, D.L.; Gitahi, G.; Yates, R. Fragmented health systems in COVID-19: Rectifying the misalignment between global health security and universal health coverage. Lancet 2021, 397, 61–67. [Google Scholar] [CrossRef]
  16. Grasselli, G.; Pesenti, A.; Cecconi, M. Critical Care Utilization for the COVID-19 Outbreak in Lombardy, Italy: Early Experi-ence and Forecast During an Emergency Response. JAMA 2020, 323, 1545. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  17. Iacobucci, G. Covid-19: All non-urgent elective surgery is suspended for at least three months in England. BMJ 2020, 368, m1106. [Google Scholar] [CrossRef] [Green Version]
  18. Kerlin, M.P.; Costa, D.K.; Davis, B.S.; Admon, A.J.; Vranas, K.C.; Kahn, J.M. Actions Taken by US Hospitals to Prepare for Increased Demand for Intensive Care During the First Wave of COVID-19: A national survey. Chest 2021, 160, 519–528. [Google Scholar] [CrossRef]
  19. Centers for Medicare and Medicaid Services. Preliminary Medicare COVID-19 Data Snapshot. 19 March 2021. Available online: https://www.cms.gov/research-statistics-data-systems/preliminary-medicare-covid-19-data-snapshot (accessed on 27 April 2021).
  20. Di Fusco, M.; Shea, K.M.; Lin, J.; Nguyen, J.L.; Angulo, F.J.; Benigno, M.; Malhotra, D.; Emir, B.; Sung, A.H.; Hammond, J.L.; et al. Health outcomes and economic burden of hospitalized COVID-19 patients in the United States. J. Med. Econ. 2021, 24, 308–317. [Google Scholar] [CrossRef]
  21. Data Note: Prescription Drugs and Older Adults. 9 August 2019. Available online: https://www.kff.org/health-reform/issue-brief/data-note-prescription-drugs-and-older-adults/ (accessed on 27 January 2021).
  22. Beijer, H.; De Blaey, C. Hospitalisations caused by adverse drug reactions (ADR): A meta-analysis of observational studies. Pharm. World Sci. 2002, 24, 46–54. [Google Scholar] [CrossRef]
  23. McQueenie, R.; Foster, H.M.E.; Jani, B.D.; Katikireddi, S.V.; Sattar, N.; Pell, J.P.; Ho, F.K.; Niedzwiedz, C.L.; Hastie, C.E.; Anderson, J.; et al. Multimorbidity, polypharmacy, and COVID-19 infection within the UK Biobank cohort. PLoS ONE 2020, 15, e0238091. [Google Scholar] [CrossRef]
  24. Stevenson, J.M.; Alexander, G.C.; Palamuttam, N.; Mehta, H.B. Projected Utility of Pharmacogenomic Testing Among Individuals Hospitalized With COVID-19: A Retrospective Multicenter Study in the United States. Clin. Transl. Sci. 2021, 14, 153–162. [Google Scholar] [CrossRef]
  25. SARS-CoV-2 and COVID-19 Related LOINC Terms. Available online: https://loinc.org/sars-cov-2-and-covid-19/ (accessed on 26 February 2021).
  26. Coleman, H.C.; Oesterheld, J.; Patterson, R.D. Genetic Data Analysis and Database Tools. U.S. Patent 8,099,298, 17 January 2012. Available online: https://patentimages.storage.googleapis.com/25/02/74/0495cac7195d3a/US8099298.pdf (accessed on 11 January 2021).
  27. Coleman, H.C.; Patterson, R.D.; Oesterheld, J.; Pany, R.V.; Ashcraft, K. Systems and Methods for Quantification and Presentation of Medical Risk Arising from Unknown Factors. U.S. Patent 2019/0164637 A1, 30 May 2019. Available online: https://patentimages.storage.googleapis.com/2b/e7/da/75f4be1d8818f2/US20190164637A1.pdf (accessed on 11 January 2021).
  28. Office of Enterprise Data and Analytics, Centers for Medicare & Medicaid Services (CMS). Chronic Conditions among Medicare Beneficiaries. 2018. Available online: https://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Chronic-Conditions/Chartbook_Charts (accessed on 27 January 2021).
  29. Hierarchical Condition Category Coding. Available online: https://www.aafp.org/family-physician/practice-and-career/getting-paid/coding/hierarchical-condition-category.html (accessed on 29 January 2021).
  30. Special Needs Plans. Available online: https://www.cms.gov/Medicare/Health-Plans/SpecialNeedsPlans (accessed on 29 January 2021).
  31. Tibshirani, R. Regression Shrinkage and Selection via the Lasso. J. R. Stat. Soc. Ser. B (Methodol.) 1996, 58, 267–288. Available online: https://www.jstor.org/stable/2346178?seq=1 (accessed on 26 February 2021). [CrossRef]
  32. R Core Team. R: A Language and Environment for Statistical Computing; Version 3.6.1; R Foundation for Statistical Computing: Vienna, Austria, 2019. [Google Scholar]
  33. Moreno, R.; Díez, J.; Diarte, J.; Macaya, F.; Hernández, J.T.; Rodríguez-Leor, O.; Trillo, R.; Alonso-Briales, J.; Amat-Santos, I.; Romaguera, R.; et al. Consequences of canceling elective invasive cardiac procedures during Covid-19 outbreak. Catheter. Cardiovasc. Interv. 2021, 97, 927–937. [Google Scholar] [CrossRef]
  34. Ueda, M.; Martins, R.; Hendrie, P.C.; McDonnell, T.; Crews, J.R.; Wong, T.L.; McCreery, B.; Jagels, B.; Crane, A.; Byrd, D.R.; et al. Managing Cancer Care During the COVID-19 Pandemic: Agility and Collaboration Toward a Common Goal. J. Natl. Compr. Cancer Netw. 2020, 18, 366–369. [Google Scholar] [CrossRef] [Green Version]
  35. Bennett, S.; Søreide, K.; Gholami, S.; Pessaux, P.; Teh, C.; Segelov, E.; Kennecke, H.; Prenen, H.; Myrehaug, S.; Callegaro, D.; et al. Strategies for the Delay of Surgery in the Management of Resectable Hepatobiliary Malignancies during the COVID-19 Pandemic. Curr. Oncol. 2020, 27, 501–511. [Google Scholar] [CrossRef] [PubMed]
  36. Medicare Advantage versus the Traditional Medicare Program: Costs of Inpatient Stays, 2009–2017 #262. Available online: https://www.hcup-us.ahrq.gov/reports/statbriefs/sb262-Medicare-Advantage-Costs-2009-2017.jsp (accessed on 19 March 2021).
  37. Lee, H.; Ryu, K.; Sohn, Y.; Kim, J.; Suh, G.Y.; Kim, E. Impact on Patient Outcomes of Pharmacist Participation in Multidisciplinary Critical Care Teams: A Systematic Review and Meta-Analysis. Crit. Care Med. 2019, 47, 1243–1250. [Google Scholar] [CrossRef]
  38. Williams, T.; Ho, K.; Dobb, G.; Finn, J.; Knuiman, M.; Webb, S. Effect of length of stay in intensive care unit on hospital and long-term mortality of critically ill adult patients. Br. J. Anaesth. 2010, 104, 459–464. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  39. Marfil-Garza, B.A.; Belaunzarán-Zamudio, P.F.; Gulias-Herrero, A.; Zuñiga, A.C.; Caro-Vega, Y.; Kershenobich-Stalnikowitz, D.; Sifuentes-Osornio, J. Risk factors associated with prolonged hospital length-of-stay: 18-year retrospective study of hospitalizations in a tertiary healthcare center in Mexico. PLoS ONE 2018, 13, e0207203. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  40. Chang, A.M.; Lin, A.; Fu, R.; McConnell, K.J.; Sun, B. Associations of Emergency Department Length of Stay with Publicly Reported Quality-of-care Measures. Acad. Emerg. Med. 2017, 24, 246–250. [Google Scholar] [CrossRef] [Green Version]
  41. Watanabe, J.H.; McInnis, T.; Hirsch, J.D. Cost of Prescription Drug–Related Morbidity and Mortality. Ann. Pharmacother. 2018, 52, 829–837. [Google Scholar] [CrossRef]
Jpm 11 01192 g001 550
Figure 1. Selection of study patients.
Figure 1. Selection of study patients.
Jpm 11 01192 g001
Jpm 11 01192 g002 550
Figure 2. Length of stay by pharmacogenomic risk among Medicare Advantage members hospitalized with COVID-19. Average length of stay (LOS) in days among the entire cohort and subpopulations stratified by HCC count or chronic conditions was compared by (A) pharmacogenomic interaction probability (PIP) and (B) drug–drug interaction (DDI). Error bars show 95% confidence intervals. PIP: pharmacogenetic interaction probability; HCC: hierarchical condition categories; COPD: chronic obstructive pulmonary disease.
Figure 2. Length of stay by pharmacogenomic risk among Medicare Advantage members hospitalized with COVID-19. Average length of stay (LOS) in days among the entire cohort and subpopulations stratified by HCC count or chronic conditions was compared by (A) pharmacogenomic interaction probability (PIP) and (B) drug–drug interaction (DDI). Error bars show 95% confidence intervals. PIP: pharmacogenetic interaction probability; HCC: hierarchical condition categories; COPD: chronic obstructive pulmonary disease.
Jpm 11 01192 g002
Table
Table 1. Characteristics of hospitalized COVID-19 Medicare Advantage members.
Table 1. Characteristics of hospitalized COVID-19 Medicare Advantage members.
VariableCategoryValue
Demographic/socioeconomic factors
GenderFemale, No. (%)3653 (61)
AgeMean (SD)77 (11)
Race/ethnicityWhite (non-Hispanic), No. (%)3753 (62)
Black (non-Hispanic), No. (%)1791 (30)
Hispanic/Latino, No. (%)191 (3)
Other, No. (%)140 (2)
Asian/Pacific Islander, No. (%)99 (2)
Unknown, No. (%)51 (1)
Residential locationUrban, No. (%)2853 (47)
Suburban, No. (%)2200 (37)
Rural, No. (%)972 (16)
Median incomeMean (SD)USD63,027 (USD17,435)
Plan and clinical characteristics
C-SNPEnrolled, No. (%)268 (4)
D-SNPEnrolled, No. (%)332 (6)
I-SNPEnrolled, No. (%)2061 (34)
HCC count0 or 1, No. (%)1460 (24)
2 or 3, No. (%)1991 (33)
4 or 5, No. (%)1182 (20)
6 or more, No. (%)1392 (23)
Chronic conditionsCOPD, No. (%)1450 (24)
Diabetes, No. (%)3104 (52)
Hyperlipidemia, No. (%)3499 (58)
Hypertension, No. (%)1902 (32)
COVID-19 hospitalization (in days)Mean LOS (SD)12.6 (11)
PIPLow, ≤25%, No. (%)3514 (58)
Moderate, 26–50%, No. (%)1784 (30)
High, >50%, No. (%)727 (12)
DDI Minimal or minor, No. (%)3110 (52)
Moderate, No. (%)983 (16)
Major, No. (%)1542 (25)
Contraindicated, No. (%)390 (7)
C-SNP: chronic conditions special needs plan (SNP); D-SNP: dual-eligible SNP; I-SNP: institutional SNP; HCC: hierarchical conditions category; LOS: length of stay; PIP: pharmacogenetic interaction probability; DDI: drug–drug interaction.
Table
Table 2. Expected length-of-stay ratios by pharmacogenetic risk and drug–drug interactions among patients by HCC count and chronic conditions based on regression models.
Table 2. Expected length-of-stay ratios by pharmacogenetic risk and drug–drug interactions among patients by HCC count and chronic conditions based on regression models.
PopulationModerate PIP (26% to 50%) *High PIP (>50%) *Moderate, Major, or Contraindicated DDI **
Rate Ratio (95% CI)p-ValueRate Ratio (95% CI)p-ValueRate Ratio (95% CI)p-Value
Total cohort1.09 (1.04, 1.14)<0.0011.16 (1.09, 1.24)<0.0011.04 (1.00, 1.09)0.066
HCC count
0 or 11.15 (1.04, 1.28)0.0071.39 (1.15, 1.67)<0.0010.91 (0.82, 1.00)0.045
2 or 31.03 (0.95, 1.11)0.5221.08 (0.96, 1.21)0.2041.10 (1.02, 1.18)0.010
4 or 51.13 (1.02, 1.25)0.0191.13 (1.00, 1.29)0.0571.07 (0.97, 1.17)0.179
6 or more1.08 (0.98, 1.18)0.1121.16 (1.03, 1.31)0.0141.01 (0.93, 1.11)0.786
Chronic conditions
COPD1.13 (1.03, 1.24)0.0091.18 (1.05, 1.34)0.0061.01 (0.93, 1.10)0.796
Diabetes1.05 (0.99, 1.12)0.1191.15 (1.05, 1.25)0.0021.07 (1.01, 1.13)0.031
Hyperlipidemia1.08 (1.02, 1.15)0.0131.12 (1.02, 1.22)0.0141.05 (0.99, 1.11)0.096
Hypertension1.03 (0.95, 1.12)0.4781.22 (1.10, 1.35)<0.0011.02 (0.95, 1.1)0.522
* 0−25% as baseline; ** minimal or minor as baseline. HCCs: hierarchical conditions categories; PIP: pharmacogenetic interaction probability; DDI: drug–drug interaction; COPD: chronic obstructive pulmonary disease. Significance was set at p < 0.05.
Table
Table 3. Distribution of length of stay among patients hospitalized with COVID-19 by chronic condition and pharmacogenomic or drug–drug interaction risk.
Table 3. Distribution of length of stay among patients hospitalized with COVID-19 by chronic condition and pharmacogenomic or drug–drug interaction risk.
Mean LOS (95% CI) by PIPMean LOS (95% CI) by DDI
Chronic ConditionTotal Subpopulation,
Mean LOS (95% CI)		Low (≤25%) Moderate (26–50%)High (>50%)Minimal or Minor Moderate, Major, or Contraindicated
COPD13.0 (12.4, 13.6)12.1 (11.4, 12.8)13.9 (12.8, 15)14.4 (12.8, 16)13.1 (12.1, 14.1)12.9 (12.2, 13.6)
Diabetes12.5 (12.1, 12.9)12 (11.5, 12.5)12.8 (12.1, 13.5)13.6 (12.6, 14.6)12.3 (11.8, 12.8)12.7 (12.2, 13.2)
Hyperlipidemia12.7 (12.3, 13.1)12.2 (11.7, 12.7)13.3 (12.6, 14)13.6 (12.5, 14.7)12.5 (12, 13)12.9 (12.4, 13.4)
Hypertension13.0 (12.5, 13.5)12.5 (11.8, 13.2)12.9 (12.1, 13.7)15 (13.6, 16.4)12.9 (12.2, 13.6)13.1 (12.4, 13.8)
Data is presented in days. LOS: length of stay; PIP: pharmacogenomic interaction probability; DDI: drug–drug interaction.
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Ashcraft, K.; Moretz, C.; Schenning, C.; Rojahn, S.; Vines Tanudtanud, K.; Magoncia, G.O.; Reyes, J.; Marquez, B.; Guo, Y.; Erdemir, E.T.; et al. Unmanaged Pharmacogenomic and Drug Interaction Risk Associations with Hospital Length of Stay among Medicare Advantage Members with COVID-19: A Retrospective Cohort Study. J. Pers. Med. 2021, 11, 1192. https://doi.org/10.3390/jpm11111192

AMA Style

Ashcraft K, Moretz C, Schenning C, Rojahn S, Vines Tanudtanud K, Magoncia GO, Reyes J, Marquez B, Guo Y, Erdemir ET, et al. Unmanaged Pharmacogenomic and Drug Interaction Risk Associations with Hospital Length of Stay among Medicare Advantage Members with COVID-19: A Retrospective Cohort Study. Journal of Personalized Medicine. 2021; 11(11):1192. https://doi.org/10.3390/jpm11111192

Chicago/Turabian Style

Ashcraft, Kristine, Chad Moretz, Chantelle Schenning, Susan Rojahn, Kae Vines Tanudtanud, Gwyn Omar Magoncia, Justine Reyes, Bernardo Marquez, Yinglong Guo, Elif Tokar Erdemir, and et al. 2021. "Unmanaged Pharmacogenomic and Drug Interaction Risk Associations with Hospital Length of Stay among Medicare Advantage Members with COVID-19: A Retrospective Cohort Study" Journal of Personalized Medicine 11, no. 11: 1192. https://doi.org/10.3390/jpm11111192

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop