- Email: [email protected]
Prediction Model for Extended Hospital Stay Among Medicare Beneficiaries After Percutaneous Coronary Intervention
Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Contents lists available at ScienceDirect
Journal of Cardiothoracic and Vascular Anesthesia journal homepage: www.jcvaonline.com
Original Article
Prediction Model for Extended Hospital Stay Among Medicare Beneficiaries After Percutaneous Coronary Intervention Brittany N. Burton, MHS, MAS*, Boya Abudu, MPH*, Dennis J. Danforth, MDy, Saatchi Patell, BS*, Lizett Wilkins y Martinez, MD, MPHz, Byron Fergerson, MDy, ,1 Ahmad Elsharydah, MD, MBAx, Rodney A. Gabriel, MD, MASy,{ *
School of Medicine, University of California San Diego, San Diego, CA Department of Anesthesiology, University of California, San Diego, San Diego, CA z Department of Internal Medicine, Olive View-UCLA, Sylmar, CA x Department of Anesthesiology and Pain Management, University of Texas Southwestern Medical Center, Dallas, TX { Department of Medicine, Division of Biomedical Informatics, University of California, San Diego, San Diego, CA y
Objective: The authors conducted a retrospective analysis to develop a predictive model consisting of factors associated with extended hospital stay among Medicare beneficiaries undergoing percutaneous coronary intervention (PCI). Design: Retrospective cohort study. Setting: Multi-institutional. Participants: Data were obtained from the National (Nationwide) Inpatient Sample registry from 2013 to 2014 over a 2-year period. Interventions: None. Measurements and Main Results: The primary outcome was extended hospital stay, which was defined as an inpatient stay greater than 75th percentile for the cohort (5 d), among Medicare beneficiaries (fee-for-service and managed care) undergoing PCI. A multivariable logistic regression analysis was built on a training set to develop the predictive model. The authors evaluated model performance with area under the receiver operating characteristic curve (AUC) and performed k-folds cross-validation to calculate the average AUC. The final analysis included 91,880 patients. Inpatient hospital length of stay ranged from 0 to 247 days, with 3 and 5 days as the median and 3rd quartile hospital stay, respectively. The final multivariable analysis suggested that sociodemographic variables, hospital-related factors, and comorbidities were associated with a greater odds of extended hospital stay (all p < 0.05). The use of PCI with drug-eluting stent was associated with a 31% decrease in extended hospital stay (odds ratio 0.69, 95% confidence interval 0.66-0.72; p < 0.001). Model discrimination was deemed excellent with an AUC (95% confidence interval) of 0.814 (0.811-0.817) and 0.809 (0.799-0.819) for the training and testing sets, respectively. Conclusion: The authors’ predictive model identified risk factors that have a higher probability of extended hospital stay. This model can be used to improve periprocedural optimization and improved discharge planning, which may help to decrease costs associated with PCIs. Management of Medicare beneficiaries after PCI calls for a multidisciplinary approach among healthcare teams and hospital administrators. Ó 2019 Elsevier Inc. All rights reserved. Key Words: Medicare; percutaneous coronary intervention; hospital stay; prediction model
1
Address reprint requests to Rodney A. Gabriel, MD, MAS, Department of Anesthesiology, University of California, San Diego, 9500 Gilman Drive, MC 0881, La Jolla, CA 92093-0881. E-mail address: [email protected] (R.A. Gabriel). https://doi.org/10.1053/j.jvca.2019.04.022 1053-0770/Ó 2019 Elsevier Inc. All rights reserved.
ISCHEMIC HEART DISEASE (IHD) affects 16.5 million individuals in the United States. Recent data have shown a decline in the mortality rate; however, IHD continues to be a major cause of death, resulting in more than 33% of deaths in the United States among individuals older than 35.1,2 Due to
3036
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
the insidious nature of this disease, patients with advanced IHD often require interventions to reduce symptoms and prevent complications. Percutaneous coronary intervention (PCI) is a procedure that restores arterial circulation to myocardial tissues, and more than 600,000 procedures are performed annually in the United States. Important predictors of poor outcomes in PCI include advanced age, female sex, diabetes mellitus, poor systolic function, and multivessel disease.3-5 Hospital length of stay (LOS) is an important indicator of hospital efficiency and quality of care.6 Extended LOS places a financial burden on the patient and healthcare system and increases the risk for hospital complications and mortality.7 PCI has the highest aggregate cost of all cardiovascular procedures and the third highest aggregate cost of any surgical procedure, at $10 billion annually.8 Approximately 95% of patients may be discharged safely 30 hours post-PCI, with comorbidities increasing the likelihood of extended LOS.9 A more robust understanding of characteristics that predict extended LOS in PCI Medicare beneficiaries may provide insight into perioperative management to expedite recovery, reduce hospital costs, and improve the safety and quality of care. Methods and Materials Data Collection and Study Population Data were obtained from the National (Nationwide) Inpatient Sample (NIS) registry of the Healthcare Cost and Utilization Project.10 NIS is the largest national inpatient healthcare database in United States. All data collected were de-identified and were exempt from consent requirement by the institutional review board. The authors conducted a retrospective analysis to evaluate risk factors associated with extended hospital stay in Medicare beneficiaries who underwent PCI over a 2-year period, from 2013 to 2014. To define the patient population, International Classification of Disease, Ninth Revision, ICD-9-CM procedures codes were used to extract medical records of acutely hospitalized Medicare beneficiaries who underwent a PCI procedure.11 Medicare included fee-for-service and managed care beneficiaries. The primary outcome variable was extended hospital stay, defined as a hospital stay greater than 75th percentile for the cohort (5 d of inpatient stay). NIS contains a variable “LOS” (hospital length of stay), and this variable was dichotomized at the 75th percentile to create the outcome of interest. The ICD9-CM diagnosis codes were used to identify potential preoperative risk factors of extended hospital stay during a single inpatient hospital stay. Briefly, data of four sociodemographic variables (race, age, median household income in the patient’s hometown, and sex); four hospital variables (geographical division, teaching status, bed size, and ownership); 18 preoperative variables (body mass index [BMI], hypertension, diabetes mellitus [DM], chronic kidney disease [CKD], depression, pulmonary disease, pneumonia, obstructive sleep apnea [OSA], heart failure [HF], alcohol use, smoker, bleeding disorder, dyslipidemia, myocardial infarction [MI], atrial
fibrillation, PCI with drug-eluting stent, cardiogenic shock, and dyspnea); and inpatient mortality were collected. Per the Healthcare Cost and Utilization Project, the hospital’s ownership category was obtained from the American Hospital Association. These ownership categories were government nonfederal (public), private not-for-profit (voluntary), and private investor-owned (proprietary).12 Table 1 lists the ICD-9CM codes used to extract the medical history for each case. Sociodemographic and hospital data were provided by the registry. Additional information for median household income for the patient’s hometown and hospital bed size and geographical divisions can be found at https://www.hcup-us.ahrq.gov/db/ nation/nis/nisdde.jsp. Statistical Analysis R, a software environment for statistical computing (R version 3.3.2), was used to perform all statistical analyses. Pearson chi-square test was used for categorical variables, and independent samples t test was used for continuous data to compare mean differences between patients with hospital stay that was not extended and those with extended hospital stay. Mixed effect logistic regression was performed to identify predictors of extended hospital stay in this patient population. The random effect was “hospital identification number” (a unique value assigned in the NIS database for a specific institution). Using this data element allowed for the accounting of clustered observations within hospitals. The patient population was divided into 10 equal parts, and the mixed effect logistics regression analysis was built on a training set (9 parts). A univariable mixed effect logistic regression analysis was performed on the training set to evaluate the association of each potential risk factor with extended hospital Table 1 International Classification of Diseases, Ninth Revision, Codes Used to Select Medical History Characteristics
ICD-9-CM Codes
BMI 40 Hypertension Diabetes mellitus Chronic kidney disease Depression Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol dependence Tobacco use Bleeding disorder Dyslipidemia Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock Dyspnea
V854.x 401.x 250.x 585.x 311, 296.x 486, 484.x 490.x, 491.x, 494.x, 492.x, 493.x 327.23 428.x 303.x, 305.x 305.1 286.x 272.x 410.x 427.31 360.7 785.51 786.x
Abbreviation: BMI, body mass index; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; PCI, percutaneous coronary intervention.
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
stay. The initial mixed effect multivariable logistic regression model was built from the univariable analysis with p values <0.2. Backwards elimination was performed until all covariates in the model were p < 0.05.13 The odds ratio, 95% confidence interval (CI), and Wald test p value are reported for each independent variable. The two-tailed significance level was set at p < 0.05. Model performance was performed with area under the receiver operating characteristic curve (AUC). The AUC was calculated and reported for the training and testing sets. To further evaluate performance, k-folds cross-validation (k = 10) was performed to calculate the average AUC. An AUC of 1.0 indicates perfect discrimination, whereas an AUC of 0.5 represents no discrimination. Calibration was evaluated with the Hosmer-Lemeshow goodness-of-fit test using deciles of predicted risk.
3037
location, bed size, and ownership); and comorbidities (BMI, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock) were associated with a higher odds of extended hospital stay (all p < 0.05). However, the use of PCI with drug-eluting stent was associated with a 31% decrease in extended hospital stay (odds ratio 0.69, 95% CI 0.66-0.72; p < 0.001). Fig 1 shows the AUC for the training and testing sets. Model discrimination was deemed excellent with an AUC (95% CI) of 0.814 (0.811-0.817) and 0.809 (0.799-0.819) for the training and testing sets, respectively. The average AUC (95% CI) was 0.793. The Hosmer-Lemeshow goodness-of-fit test suggested poor calibration for both the training set (p < 0.05) and validation set (p < 0.05). Fig 2 demonstrates the prevalence of extended hospital stay for US geographical divisions.
Results Discussion For the present study, 91,880 Medicare beneficiaries who underwent PCI over a 2-year period from 2013 to 2014 were identified. Of the entire sample, one patient record was missing hospital LOS, and one patient record was missing age. After removing these patients, the final analysis included 91,878 study participants. Inpatient hospital LOS ranged from 0 to 247 days, with 3 days and 5 days as the median and 3rd quartile hospital stay, respectively. Table 2 reports the baseline overall study characteristics, the mean differences among the hospital stay cohorts, and the univariable logistic regression analysis of extended hospital stay. White patients comprised 75.1% (n = 68,990) of the sample, median household income in the patient’s hometown in the lowest quartile (ie, quartile 1) represented 29.9% (n = 26,885) of the study sample, and the mean (standard deviation) age was 72 (9.69) years old. Hospitals in the South Atlantic region accounted for 21.3% (n = 19,597) of study participants. Most patients were treated at urban teaching hospitals (60.9% [n = 55,988]). Dyslipidemia, MI, hypertension, and DM were the most prevalent comorbidities. Compared with Medicare beneficiaries who were discharged within 5 days of stay, patients with extended hospital stay were significantly more likely to be Black (7.1% [not extended] v 10.0% [extended]), Hispanic (5.9% [not extended) v 6.8% (extended]), or Asian or Pacific Islander (1.6% [not extended] v 2.0% [extended]); all p < 0.001). Patients with extended hospital stay were significantly older (71 y old [not extended] v 73 y old [extended]; p < 0.001) and female (36.8% [not extended] v 44.4% [extended]; p < 0.001). Extended hospital stay was associated with higher rates of BMI 40, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock (all p < 0.05). PCI with drug-eluting stent was associated with lower rates of extended hospital stay (p < 0.001). The rate of inpatient mortality was more than 2 times between Medicare beneficiaries with extended stay versus those without (2.1% v 4.5%; p < 0.001). Table 3 illustrates the results of the multivariable analysis of the training set. The final analysis suggested that sociodemographic variables (race [Black, Hispanic, and Asian or Pacific Islander]; age; sex; and median household income); hospitalrelated factors (geographical division, teaching status and
In this national registry retrospective analysis, the authors developed a predictive model of factors associated with extended hospital stay (5 d) among Medicare beneficiaries undergoing PCI. Several factors of extended hospital stay were identified, including 4 sociodemographic variables (race, age, sex, and median household income); 4 hospital-related factors (geographical division, teaching status and location, bed size, and ownership); and 12 inpatient comorbidities (BMI, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock). This prediction model may help to risk-stratify patients and reduce the need for extended hospital stay among Medicare beneficiaries. Periprocedural management and early discharge planning for high-risk patients and accelerated care pathways for low-risk cohorts may help to improve outcomes while offsetting costs. Per the Centers for Medicare and Medicaid Services, Medicare is a federal health insurance program for US citizens ages 65 years and older, younger patients with specific disabilities, and people with renal dialysis or transplantation.14 It has been shown across medical specialties that Medicare patients tend to have a higher comorbidity burden with worse health outcomes. In a retrospective analysis of HF patients, Kapoor et al.15 showed that Medicare patients had significantly higher rates of anemia, atrial fibrillation, atherosclerotic disease, and renal insufficiency. Compared with non-Medicare beneficiaries, patients who received Medicare had significantly longer hospital stays and were less likely to receive standard medical therapy for left ventricular systolic dysfunction. Similarly, in a national database study evaluating the effect of health insurance on short-term mortality after STsegment elevation MI, Pancholy et al. showed that compared with patients with private health insurance, Medicare beneficiaries were 63% more likely to experience inpatient mortality.16 A single-institution study using machine learning algorithms demonstrated that insurance status was associated with increased hospital stay among patients admitted to cardiology service.17 It is clear from these data that Medicare insurance status defines a vulnerable patient population. Among health plans, Medicare beneficiaries represent the largest proportion of patients undergoing PCI.18
3038
Table 2 Patient Study Characteristics Hospital Stay Not Extended 67,526
Hospital Stay Extended 24,352
68,990 (75.1) 7,235 (7.9) 5,680 (6.2) 1,579 (1.7) 389 (0.4) 2,492 (2.7) 5,513 (6.0) 72.28 (9.69)
51,226 (75.9) 4,811 (7.1) 4,013 (5.9) 1,094 (1.6) 273 (0.4) 1,837 (2.7) 4272 (6.3) 71.75 (9.60)
17,764 (72.9) 2,424 (10.0) 1,667 (6.8) 485 (2.0) 116 (0.5) 655 (2.7) 1,241 (5.1) 73.73 (9.79)
26,885 (29.3) 25,952 (28.2) 20,826 (22.7) 16,459 (17.9) 1,756 (1.9)
19,366 (28.7) 19,154 (28.4) 15,474 (22.9) 12,249 (18.1) 1,283 (1.9)
7,519 (30.9) 6,798 (27.9) 5,352 (22.0) 4,210 (17.3) 473 (1.9)
56,231 (61.2) 35,642 (38.8) 5 (0.0)
42,700 (63.2) 24,823 (36.8) 3 (0.0)
13,531 (55.6) 10,819 (44.4) 2 (0.0)
3,175 (3.5) 12,587 (13.7) 15,794 (17.2) 7,250 (7.9) 19,597 (21.3) 7,930 (8.6) 10,092 (11.0) 5,632 (6.1) 9,821 (10.7)
2,296 (3.4) 8,976 (13.3) 11,607 (17.2) 5,569 (8.2) 14,090 (20.9) 5,961 (8.8) 7,211 (10.7) 4,339 (6.4) 7,477 (11.1)
879 (3.6) 3,611 (14.8) 4,187 (17.2) 1,681 (6.9) 5,507 (22.6) 1,969 (8.1) 2,881 (11.8) 1,293 (5.3) 2,344 (9.6)
5,939 (6.5) 29,951 (32.6) 55,988 (60.9)
4,561 (6.8) 22,391 (33.2) 40,574 (60.1)
1,378 (5.7) 7,560 (31.0) 15,414 (63.3)
9,556 (10.4) 23,637 (25.7) 58,685 (63.9)
7,212 (10.7) 17,560 (26.0) 42,754 (63.3)
2,344 (9.6) 6,077 (25.0) 15,931 (65.4)
7,399 (8.1) 68,703 (74.8) 15,776 (17.2) 3,167 (3.4) 54,539 (59.4) 37,747 (41.1) 20,699 (22.5) 7,291 (7.9)
5,364 (7.9) 50,637 (75.0) 11,525 (17.1) 2,006 (3.0) 43,677 (64.7) 26,115 (38.7) 11,580 (17.1) 5,039 (7.5)
2,035 (8.4) 18,066 (74.2) 4,251 (17.5) 1,161 (4.8) 10,862 (44.6) 11,632 (47.8) 9,119 (37.4) 2,252 (9.2)
p Value
Univariable Analysis OR (95% CI)
p Value
<0.001
<0.001 <0.001
Reference 1.36 (1.29-1.44) 1.17 (1.10-1.26) 1.27 (1.13-1.43) 1.25 (0.98-1.58) 1.05 (0.95-1.17) 0.83 (0.76-0.90) 1.23 (1.21-1.25)
<0.001 <0.001 <0.001 0.069263 0.326515 <0.001 <0.001
Reference 0.91 (0.88-0.95) 0.88 (0.84-0.92) 0.86 (0.81-0.90) 0.93 (0.82-1.04)
<0.001 <0.001 <0.001 0.205
Reference 1.37 (1.33-1.42) 2.34 (0.37-14.97)
<0.001 0.37
<0.001
<0.001 Reference 1.11 (0.96-1.28) 0.96 (0.83-1.10) 0.78 (0.67-0.91) 1.02 (0.89-1.18) 0.85 (0.72-0.99) 1.08 (0.93-1.24) 0.77 (0.66-0.91) 0.86 (0.74-0.99)
0.1673 0.53899 0.00169 0.74347 0.04208 0.3109 0.0014 0.04159
<0.001 Reference 1.12 (1.02-1.22) 1.29 (1.18-1.41) <0.001 Reference 1.05 (0.98-1.13) 1.11 (1.04-1.19)
0.0139 <0.001
0.18303 0.00204
0.032
<0.001 <0.001 <0.001 <0.001 <0.001
Reference 0.91 (0.84-0.99) 0.93 (0.85-1.02) 1.63 (1.51-1.77) 0.44 (0.42-0.45) 1.43 (1.38-1.47) 2.90 (2.80-3.01) 1.26 (1.19-1.33)
0.0248 0.1238 <0.001 <0.001 <0.001 <0.001 <0.001 (continued on next page)
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Race White Black Hispanic Asian or Pacific Islander Native American Other Unknown Age (y), mean (SD) Median household income of hometown Quartile 1 Quartile 2 Quartile 3 Quartile 4 Unknown Sex Male Female Unknown Hospital division New England Middle Atlantic East North Central West North Central South Atlantic East South Central West South Central Mountain Pacific Hospital teaching status Rural Urban nonteaching Urban teaching Hospital bed size Small Medium Large Hospital ownership Government, nonfederal Private, not-profit Private, invest-own BMI 40 Hypertension Diabetes mellitus Chronic kidney disease Depression
Overall Study Population 91,878
23,240 (95.4) 1,107 (4.5) 5 (0.0) 66,122 (97.9) 1,398 (2.1) 6 (0.0) 89,362 (97.3) 2,505 (2.7) 11 (0.0)
NOTE. Extended hospital stay = greater than 75th percentile for cohort (5 d); p value = Pearson chi-square and independent samples t test for hospital stay cohorts; p value = Wald test. Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio; PCI, percutaneous coronary intervention; SD, standard deviation.
10.86 (9.96-11.83) 2.28 (2.17-2.40) 1.44 (1.36-1.53) 5.72 (5.52-5.92) 1.46 (1.31-1.63) 0.78 (0.75-0.82) 4.69 (3.77-5.84) 0.58 (0.56-0.60) 1.59 (1.54-1.65) 2.81 (2.71-2.92) 0.54 (0.52-0.56) 5.79 (5.40-6.20) 0.91 (0.73-1.15) <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.62 <0.001 2,738 (11.2) 3,735 (15.3) 2,132 (8.8) 13,397 (55.0) 560 (2.3) 3,420 (14.0) 255 (1.0) 15,502 (63.7) 16,606 (68.2) 7,350 (30.2) 17,829 (73.2) 2,769 (11.4) 117 (0.5) 813 (1.2) 4,956 (7.3) 4,276 (6.3) 12,007 (17.8) 1,066 (1.6) 11,728 (17.4) 149 (0.2) 50,793 (75.2) 39,079 (57.9) 9,035 (13.4) 56,119 (83.1) 1,493 (2.2) 344 (0.5) 3,551 (3.9) 8,691 (9.5) 6,408 (7.0) 25,404 (27.6) 1,626 (1.8) 15,148 (16.5) 404 (0.4) 66,295 (72.2) 55,685 (60.6) 16,385 (17.8) 73,948 (80.5) 4,262 (4.6) 461 (0.5)
Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol abuse Smoker Bleeding disorder Dyslipidemia Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock Dyspnea Inpatient mortality No Yes Missing
Hospital Stay Not Extended 67,526 Overall Study Population 91,878
Hospital Stay Extended 24,352
p Value
Univariable Analysis OR (95% CI)
Characteristics
Table 2 (continued )
3039
Table 3 Multivariable Logistic Regression Analysis for All Covariates for Predicting Extended Hospital Length of Stay Built on the Training Set
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.439
p Value
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Race White Black Hispanic Asian or Pacific Islander Unknown Age (10-y increase) Median household income of hometown Quartile 1 Quartile 2 Quartile 3 Quartile 4 Sex Male Female Hospital division New England Middle Atlantic West North Central Mountain Pacific Hospital teaching status Rural Urban nonteaching Urban teaching Hospital bed size Small Medium Large Hospital ownership Government, nonfederal Private, not-profit BMI 40 Diabetes mellitus Chronic kidney disease Depression Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol abuse Bleeding disorder Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock
OR (95% CI)
p Value
Reference 1.24 (1.16-1.32) 1.14 (1.06-1.23) 1.28 (1.12-1.47) 0.89 (0.81-0.98) 1.13 (1.11-1.15)
<0.001 <0.001 <0.001 0.021075 <0.001
Reference 0.95 (0.90-0.99) 0.92 (0.87-0.97) 0.89 (0.84 -0.94)
0.01925 0.001193 <0.001
Reference 1.31 (1.27-1.36)
<0.001
Reference 1.22 (1.05-1.42) 0.79 (0.67-0.94) 0.83 (0.70-0.98) 0.75 (0.64-0.87)
0.009627 0.005754 0.024349 0.000204
Reference 1.20 (1.09-1.32) 1.33 (1.20-1.46)
0.00022 <0.001
Reference 1.11 (1.03-1.20) 1.27 (1.18-1.37)
0.009549 <0.001
Reference 0.89 (0.82-0.97) 1.32 (1.21-1.44) 1.16 (1.12-1.20) 2.09 (2.01-2.17) 1.24 (1.17-1.32) 5.97 (5.46-6.52) 1.79 (1.69-1.89) 1.12 (1.05-1.20) 3.66 (3.53-3.80) 1.71 (1.51-1.92) 2.37 (1.86-3.02) 1.45 (1.40-1.51) 2.02 (1.94-2.11) 0.68 (0.65-0.71) 3.61 (3.35-3.89)
0.006332 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio.
Because Medicare does not reimburse hospitals for unnecessary services or for patients with extended hospital stay, it is prudent to identify and address risk factors of poor outcomes in this patient population. Between 2006 to 2016, the death rate owing to coronary heart disease declined by 31.8%; however, coronary heart disease continues to be a significant contributor to healthcare spending. The literature reporting factors associated with important health quality indices, such as extended hospital stay, remains sparse.19 To improve healthcare quality, the Institute of Medicine recommends performance indices—safety, effectiveness, patient centeredness,
3040
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Fig 1. Area under the receiver operating curve for predictive model fitted on (A) training set and (B) testing set. AUC, area under the curve; ROC, receiver operating curve.
Prevalence of Extended Hospital Stay by United States Georgraphical Division
Extended Hospital Stay (%)
5 10 15 20
Fig 2. The prevalence of extended hospital stay by US geographical region. Geographical regions with prevalence: New England (3.6%), Mid Atlantic (14.8%), East North Central (17.2%), and West North Central (6.9%), South Atlantic (22.6%), East South Central (8.1%), West South Central (11.8%), Mountain (5.3%), and Pacific (9.6%).
timeliness, efficiency, and equity—of which LOS helps to estimate healthcare efficiency.20 Given that PCI accounts for a significant portion of healthcare spending and that resource utilization and length of hospital stay are quality metrics that help to estimate cost, identifying and addressing modifiable risk factors associated with extended hospital stay should be a high priority. Not surprisingly, modifiable risk factors of coronary atherosclerosis, such as alcohol abuse and BMI, were associated with higher odds of extended hospital stay. Patients should continue to be educated on lifestyle modification and secondary prevention. Patient residence in a higher income
neighborhood was associated with lower odds of extended hospital stay. This finding is consistent with other reports. Compared with nonparticipants, participants in the Supplemental Nutrition Assistance Program, a federal program that provides food assistance to low-income families, had poorer health outcomes, which likely reflects differences in socioeconomic and behavioral characteristics.21,22 We identified inpatient comorbidities (DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, bleeding disorder, and MI) that may not be “modifiable” prior to PCI, and yet there is a clear mortality benefit to early PCI. Therefore, it is a challenge to
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
optimize these patients before PCI, as such, cost-effective alternatives to preprocedural optimization must be considered. Although the present study’s data lack information on the severity of comorbidities, the predictive model developed may help to guide periprocedural risk stratification, optimization, and rehabilitation. Early PCI has a proven mortality benefit in patients with acute MI, and therefore optimization before PCI may not be feasible. For these patients, postprocedural rehabilitation should be the focus, and involvement of multidisciplinary care teams with early discharge planning to acute rehabilitation may improve outcomes and reduce hospital LOS. It is likely that enhanced postprocedural pathways will be the most important mechanism of preventing extended hospital stay because PCI should be performed within 90 minutes of recognition of MI. However, in patients with angina the evidence suggesting early catheterization is less clear. In this patient population, preprocedural risk stratification and optimization should be considered. There are important limitations to the present analysis. NIS is an administrative inpatient database, and important clinical data that may affect the outcome are not available. NIS does not collect information regarding the cause of death; cardiovascular metrics (ejection fraction, degree of stenosis, and number of vessels involved); time to PCI; previous cardiac procedures; medications; lifestyle data; physical activity; laboratory values; intraoperative data; or events that transpire after discharge. The use of ICD-9CM diagnosis codes also pose limitations. Coding practices may vary among hospitals and providers; and therefore patients may be misclassified with respect to their exposure status. The administrative nature of the database with use of ICD-9-CM codes precludes the evaluation of comorbidity severity. Strengths of the present study include a large nationally representative sample size, allowing for the examination of hospital LOS. Rising healthcare costs and poor outcomes have prompted change in the healthcare system. In the last decade, a dramatic shift from volume- to quality-based healthcare has occurred in the United States. The model described here has an excellent AUC; however, it has poor calibration. Therefore, the panoply of risk factors identified in this article may predict patients at risk of extended hospital stay; however, additional prospective studies are needed to develop predictive models with excellent discrimination and calibration. This predictive model can be used to help with periprocedural risk-stratification, optimization, and rehabilitation to improve discharge planning. Conflicts of Interest The authors have no conflicts of interest to disclose. References 1 Benjamin EJ, Virani SS, Callaway CW, et al. Heart disease and stroke statistics-2018 update: A report from the American Heart Association. Circulation 2018;137:e67–e492.
3041
2 Rosamond W, Flegal K, Furie K, et al. Heart disease and stroke statistics— 2008 update: A report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 2008;117: e25–146. 3 Aggarwal B, Ellis SG, Lincoff AM, et al. Cause of death within 30 days of percutaneous coronary intervention in an era of mandatory outcome reporting. J Am Coll Cardiol 2013;62:409–15. 4 Rasoul S, Ottervanger JP, de Boer MJ, et al. Predictors of 30-day and 1-year mortality after primary percutaneous coronary intervention for STelevation myocardial infarction. Coron Artery Dis 2009;20:415–21. 5 Kaluski E, Alfano D, Randhawa P, et al. Length of hospital stay after percutaneous coronary interventions. J Cardiovasc Nurs 2008;23:345–8. 6 Thomas JW, Guire KE, Horvat GG. Is patient length of stay related to quality of care? Hosp Health Serv Adm 1997;42:489–507. 7 Chin CT, Weintraub WS, Dai D, et al. Trends and predictors of length of stay after primary percutaneous coronary intervention: A report from the Cath PCI registry. Am Heart J 2011;162:1052–61. 8 Amin AP, Patterson M, House JA, et al. Costs associated with access site and same-day discharge among Medicare beneficiaries undergoing percutaneous coronary intervention: An evaluation of the current percutaneous coronary intervention care pathways in the United States. JACC Cardiovasc Interv 2017;10:342–51. 9 Pedersen F, Butrymovich V, Kelbaek H, et al. Short- and long-term cause of death in patients treated with primary PCI for STEMI. J Am Coll Cardiol 2014;64:2101–8. 10 Healthcare Cost and Utilization Project. Overview of the National (Nationwide) Inpatient Sample (NIS). Available at: https://www.hcup-us.ahrq. gov/nisoverview.jsp. Accessed April 29, 2019. 11 Centers of Disease Control and Prevention. International classification of diseases, ninth revision, clinical modification (ICD-9-CM). Available at: https://www.cdc.gov/nchs/icd/icd9cm.htm. Accessed December 16, 2004. 12 American Hospital Association. Fast facts on United States hospitals. Available at: https://www.aha.org/statistics/fast-facts-us-hospitals. Accessed April 21, 2019. 13 Heinze G, Wallisch C, Dunkler D. Variable selection - A review and recommendations for the practicing statistician. Biom J 2018;60:431–49. 14 Centers for Medicare and Medicaid Services. Medicare program - general information. Available at: https://www.cms.gov/Medicare/Medicare-GeneralInformation/MedicareGenInfo/index.html. Accessed February 14, 2018. 15 Kapoor JR, Kapoor R, Hellkamp AS, et al. Payment source, quality of care, and outcomes in patients hospitalized with heart failure. J Am Coll Cardiol 2011;58:1465–71. 16 Pancholy S, Patel G, Pancholy M, et al. Association between health insurance status and in-hospital outcomes after ST-segment elevation myocardial infarction. Am J Cardiol 2017;120:1049–54. 17 Daghistani TA, Elshawi R, Sakr S, et al. Predictors of in-hospital length of stay among cardiac patients: A machine learning approach. Int J Cardiol 2019 Jan 19; [E-pub ahead of print]. 18 Curtis JP, Schreiner G, Wang Y, et al. All-cause readmission and repeat revascularization after percutaneous coronary intervention in a cohort of Medicare patients. J Am Coll Cardiol 2009;54:903–7. 19 Go AS, Mozaffarian D, Roger VL, et al. Heart disease and stroke statistics— 2014 update: A report from the American Heart Association. Circulation 2014;129:e28–292. 20 Agency for Healthcare Research and Quality. Examples of hospital quality measures for consumers. Available at: https://www.ahrq.gov/talkingquality/measures/setting/hospitals/examples.html. Accessed February 14, 2019. 21 Benjamin EJ, Muntner P, Alonso A, et al. Heart disease and stroke statistics-2019 update: A report from the American Heart Association. Circulation 2019;139:e56–528. 22 Murphy SL, Xu J, Kochanek KD, et al. Mortality in the United States, 2017. Available at: https://www.cdc.gov/nchs/data/databriefs/db328-h.pdf. Accessed February 14, 2019.
Contents lists available at ScienceDirect
Journal of Cardiothoracic and Vascular Anesthesia journal homepage: www.jcvaonline.com
Original Article
Prediction Model for Extended Hospital Stay Among Medicare Beneficiaries After Percutaneous Coronary Intervention Brittany N. Burton, MHS, MAS*, Boya Abudu, MPH*, Dennis J. Danforth, MDy, Saatchi Patell, BS*, Lizett Wilkins y Martinez, MD, MPHz, Byron Fergerson, MDy, ,1 Ahmad Elsharydah, MD, MBAx, Rodney A. Gabriel, MD, MASy,{ *
School of Medicine, University of California San Diego, San Diego, CA Department of Anesthesiology, University of California, San Diego, San Diego, CA z Department of Internal Medicine, Olive View-UCLA, Sylmar, CA x Department of Anesthesiology and Pain Management, University of Texas Southwestern Medical Center, Dallas, TX { Department of Medicine, Division of Biomedical Informatics, University of California, San Diego, San Diego, CA y
Objective: The authors conducted a retrospective analysis to develop a predictive model consisting of factors associated with extended hospital stay among Medicare beneficiaries undergoing percutaneous coronary intervention (PCI). Design: Retrospective cohort study. Setting: Multi-institutional. Participants: Data were obtained from the National (Nationwide) Inpatient Sample registry from 2013 to 2014 over a 2-year period. Interventions: None. Measurements and Main Results: The primary outcome was extended hospital stay, which was defined as an inpatient stay greater than 75th percentile for the cohort (5 d), among Medicare beneficiaries (fee-for-service and managed care) undergoing PCI. A multivariable logistic regression analysis was built on a training set to develop the predictive model. The authors evaluated model performance with area under the receiver operating characteristic curve (AUC) and performed k-folds cross-validation to calculate the average AUC. The final analysis included 91,880 patients. Inpatient hospital length of stay ranged from 0 to 247 days, with 3 and 5 days as the median and 3rd quartile hospital stay, respectively. The final multivariable analysis suggested that sociodemographic variables, hospital-related factors, and comorbidities were associated with a greater odds of extended hospital stay (all p < 0.05). The use of PCI with drug-eluting stent was associated with a 31% decrease in extended hospital stay (odds ratio 0.69, 95% confidence interval 0.66-0.72; p < 0.001). Model discrimination was deemed excellent with an AUC (95% confidence interval) of 0.814 (0.811-0.817) and 0.809 (0.799-0.819) for the training and testing sets, respectively. Conclusion: The authors’ predictive model identified risk factors that have a higher probability of extended hospital stay. This model can be used to improve periprocedural optimization and improved discharge planning, which may help to decrease costs associated with PCIs. Management of Medicare beneficiaries after PCI calls for a multidisciplinary approach among healthcare teams and hospital administrators. Ó 2019 Elsevier Inc. All rights reserved. Key Words: Medicare; percutaneous coronary intervention; hospital stay; prediction model
1
Address reprint requests to Rodney A. Gabriel, MD, MAS, Department of Anesthesiology, University of California, San Diego, 9500 Gilman Drive, MC 0881, La Jolla, CA 92093-0881. E-mail address: [email protected] (R.A. Gabriel). https://doi.org/10.1053/j.jvca.2019.04.022 1053-0770/Ó 2019 Elsevier Inc. All rights reserved.
ISCHEMIC HEART DISEASE (IHD) affects 16.5 million individuals in the United States. Recent data have shown a decline in the mortality rate; however, IHD continues to be a major cause of death, resulting in more than 33% of deaths in the United States among individuals older than 35.1,2 Due to
3036
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
the insidious nature of this disease, patients with advanced IHD often require interventions to reduce symptoms and prevent complications. Percutaneous coronary intervention (PCI) is a procedure that restores arterial circulation to myocardial tissues, and more than 600,000 procedures are performed annually in the United States. Important predictors of poor outcomes in PCI include advanced age, female sex, diabetes mellitus, poor systolic function, and multivessel disease.3-5 Hospital length of stay (LOS) is an important indicator of hospital efficiency and quality of care.6 Extended LOS places a financial burden on the patient and healthcare system and increases the risk for hospital complications and mortality.7 PCI has the highest aggregate cost of all cardiovascular procedures and the third highest aggregate cost of any surgical procedure, at $10 billion annually.8 Approximately 95% of patients may be discharged safely 30 hours post-PCI, with comorbidities increasing the likelihood of extended LOS.9 A more robust understanding of characteristics that predict extended LOS in PCI Medicare beneficiaries may provide insight into perioperative management to expedite recovery, reduce hospital costs, and improve the safety and quality of care. Methods and Materials Data Collection and Study Population Data were obtained from the National (Nationwide) Inpatient Sample (NIS) registry of the Healthcare Cost and Utilization Project.10 NIS is the largest national inpatient healthcare database in United States. All data collected were de-identified and were exempt from consent requirement by the institutional review board. The authors conducted a retrospective analysis to evaluate risk factors associated with extended hospital stay in Medicare beneficiaries who underwent PCI over a 2-year period, from 2013 to 2014. To define the patient population, International Classification of Disease, Ninth Revision, ICD-9-CM procedures codes were used to extract medical records of acutely hospitalized Medicare beneficiaries who underwent a PCI procedure.11 Medicare included fee-for-service and managed care beneficiaries. The primary outcome variable was extended hospital stay, defined as a hospital stay greater than 75th percentile for the cohort (5 d of inpatient stay). NIS contains a variable “LOS” (hospital length of stay), and this variable was dichotomized at the 75th percentile to create the outcome of interest. The ICD9-CM diagnosis codes were used to identify potential preoperative risk factors of extended hospital stay during a single inpatient hospital stay. Briefly, data of four sociodemographic variables (race, age, median household income in the patient’s hometown, and sex); four hospital variables (geographical division, teaching status, bed size, and ownership); 18 preoperative variables (body mass index [BMI], hypertension, diabetes mellitus [DM], chronic kidney disease [CKD], depression, pulmonary disease, pneumonia, obstructive sleep apnea [OSA], heart failure [HF], alcohol use, smoker, bleeding disorder, dyslipidemia, myocardial infarction [MI], atrial
fibrillation, PCI with drug-eluting stent, cardiogenic shock, and dyspnea); and inpatient mortality were collected. Per the Healthcare Cost and Utilization Project, the hospital’s ownership category was obtained from the American Hospital Association. These ownership categories were government nonfederal (public), private not-for-profit (voluntary), and private investor-owned (proprietary).12 Table 1 lists the ICD-9CM codes used to extract the medical history for each case. Sociodemographic and hospital data were provided by the registry. Additional information for median household income for the patient’s hometown and hospital bed size and geographical divisions can be found at https://www.hcup-us.ahrq.gov/db/ nation/nis/nisdde.jsp. Statistical Analysis R, a software environment for statistical computing (R version 3.3.2), was used to perform all statistical analyses. Pearson chi-square test was used for categorical variables, and independent samples t test was used for continuous data to compare mean differences between patients with hospital stay that was not extended and those with extended hospital stay. Mixed effect logistic regression was performed to identify predictors of extended hospital stay in this patient population. The random effect was “hospital identification number” (a unique value assigned in the NIS database for a specific institution). Using this data element allowed for the accounting of clustered observations within hospitals. The patient population was divided into 10 equal parts, and the mixed effect logistics regression analysis was built on a training set (9 parts). A univariable mixed effect logistic regression analysis was performed on the training set to evaluate the association of each potential risk factor with extended hospital Table 1 International Classification of Diseases, Ninth Revision, Codes Used to Select Medical History Characteristics
ICD-9-CM Codes
BMI 40 Hypertension Diabetes mellitus Chronic kidney disease Depression Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol dependence Tobacco use Bleeding disorder Dyslipidemia Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock Dyspnea
V854.x 401.x 250.x 585.x 311, 296.x 486, 484.x 490.x, 491.x, 494.x, 492.x, 493.x 327.23 428.x 303.x, 305.x 305.1 286.x 272.x 410.x 427.31 360.7 785.51 786.x
Abbreviation: BMI, body mass index; ICD-9-CM, International Classification of Diseases, Ninth Revision, Clinical Modification; PCI, percutaneous coronary intervention.
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
stay. The initial mixed effect multivariable logistic regression model was built from the univariable analysis with p values <0.2. Backwards elimination was performed until all covariates in the model were p < 0.05.13 The odds ratio, 95% confidence interval (CI), and Wald test p value are reported for each independent variable. The two-tailed significance level was set at p < 0.05. Model performance was performed with area under the receiver operating characteristic curve (AUC). The AUC was calculated and reported for the training and testing sets. To further evaluate performance, k-folds cross-validation (k = 10) was performed to calculate the average AUC. An AUC of 1.0 indicates perfect discrimination, whereas an AUC of 0.5 represents no discrimination. Calibration was evaluated with the Hosmer-Lemeshow goodness-of-fit test using deciles of predicted risk.
3037
location, bed size, and ownership); and comorbidities (BMI, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock) were associated with a higher odds of extended hospital stay (all p < 0.05). However, the use of PCI with drug-eluting stent was associated with a 31% decrease in extended hospital stay (odds ratio 0.69, 95% CI 0.66-0.72; p < 0.001). Fig 1 shows the AUC for the training and testing sets. Model discrimination was deemed excellent with an AUC (95% CI) of 0.814 (0.811-0.817) and 0.809 (0.799-0.819) for the training and testing sets, respectively. The average AUC (95% CI) was 0.793. The Hosmer-Lemeshow goodness-of-fit test suggested poor calibration for both the training set (p < 0.05) and validation set (p < 0.05). Fig 2 demonstrates the prevalence of extended hospital stay for US geographical divisions.
Results Discussion For the present study, 91,880 Medicare beneficiaries who underwent PCI over a 2-year period from 2013 to 2014 were identified. Of the entire sample, one patient record was missing hospital LOS, and one patient record was missing age. After removing these patients, the final analysis included 91,878 study participants. Inpatient hospital LOS ranged from 0 to 247 days, with 3 days and 5 days as the median and 3rd quartile hospital stay, respectively. Table 2 reports the baseline overall study characteristics, the mean differences among the hospital stay cohorts, and the univariable logistic regression analysis of extended hospital stay. White patients comprised 75.1% (n = 68,990) of the sample, median household income in the patient’s hometown in the lowest quartile (ie, quartile 1) represented 29.9% (n = 26,885) of the study sample, and the mean (standard deviation) age was 72 (9.69) years old. Hospitals in the South Atlantic region accounted for 21.3% (n = 19,597) of study participants. Most patients were treated at urban teaching hospitals (60.9% [n = 55,988]). Dyslipidemia, MI, hypertension, and DM were the most prevalent comorbidities. Compared with Medicare beneficiaries who were discharged within 5 days of stay, patients with extended hospital stay were significantly more likely to be Black (7.1% [not extended] v 10.0% [extended]), Hispanic (5.9% [not extended) v 6.8% (extended]), or Asian or Pacific Islander (1.6% [not extended] v 2.0% [extended]); all p < 0.001). Patients with extended hospital stay were significantly older (71 y old [not extended] v 73 y old [extended]; p < 0.001) and female (36.8% [not extended] v 44.4% [extended]; p < 0.001). Extended hospital stay was associated with higher rates of BMI 40, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock (all p < 0.05). PCI with drug-eluting stent was associated with lower rates of extended hospital stay (p < 0.001). The rate of inpatient mortality was more than 2 times between Medicare beneficiaries with extended stay versus those without (2.1% v 4.5%; p < 0.001). Table 3 illustrates the results of the multivariable analysis of the training set. The final analysis suggested that sociodemographic variables (race [Black, Hispanic, and Asian or Pacific Islander]; age; sex; and median household income); hospitalrelated factors (geographical division, teaching status and
In this national registry retrospective analysis, the authors developed a predictive model of factors associated with extended hospital stay (5 d) among Medicare beneficiaries undergoing PCI. Several factors of extended hospital stay were identified, including 4 sociodemographic variables (race, age, sex, and median household income); 4 hospital-related factors (geographical division, teaching status and location, bed size, and ownership); and 12 inpatient comorbidities (BMI, DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, alcohol abuse, bleeding disorder, MI, and cardiogenic shock). This prediction model may help to risk-stratify patients and reduce the need for extended hospital stay among Medicare beneficiaries. Periprocedural management and early discharge planning for high-risk patients and accelerated care pathways for low-risk cohorts may help to improve outcomes while offsetting costs. Per the Centers for Medicare and Medicaid Services, Medicare is a federal health insurance program for US citizens ages 65 years and older, younger patients with specific disabilities, and people with renal dialysis or transplantation.14 It has been shown across medical specialties that Medicare patients tend to have a higher comorbidity burden with worse health outcomes. In a retrospective analysis of HF patients, Kapoor et al.15 showed that Medicare patients had significantly higher rates of anemia, atrial fibrillation, atherosclerotic disease, and renal insufficiency. Compared with non-Medicare beneficiaries, patients who received Medicare had significantly longer hospital stays and were less likely to receive standard medical therapy for left ventricular systolic dysfunction. Similarly, in a national database study evaluating the effect of health insurance on short-term mortality after STsegment elevation MI, Pancholy et al. showed that compared with patients with private health insurance, Medicare beneficiaries were 63% more likely to experience inpatient mortality.16 A single-institution study using machine learning algorithms demonstrated that insurance status was associated with increased hospital stay among patients admitted to cardiology service.17 It is clear from these data that Medicare insurance status defines a vulnerable patient population. Among health plans, Medicare beneficiaries represent the largest proportion of patients undergoing PCI.18
3038
Table 2 Patient Study Characteristics Hospital Stay Not Extended 67,526
Hospital Stay Extended 24,352
68,990 (75.1) 7,235 (7.9) 5,680 (6.2) 1,579 (1.7) 389 (0.4) 2,492 (2.7) 5,513 (6.0) 72.28 (9.69)
51,226 (75.9) 4,811 (7.1) 4,013 (5.9) 1,094 (1.6) 273 (0.4) 1,837 (2.7) 4272 (6.3) 71.75 (9.60)
17,764 (72.9) 2,424 (10.0) 1,667 (6.8) 485 (2.0) 116 (0.5) 655 (2.7) 1,241 (5.1) 73.73 (9.79)
26,885 (29.3) 25,952 (28.2) 20,826 (22.7) 16,459 (17.9) 1,756 (1.9)
19,366 (28.7) 19,154 (28.4) 15,474 (22.9) 12,249 (18.1) 1,283 (1.9)
7,519 (30.9) 6,798 (27.9) 5,352 (22.0) 4,210 (17.3) 473 (1.9)
56,231 (61.2) 35,642 (38.8) 5 (0.0)
42,700 (63.2) 24,823 (36.8) 3 (0.0)
13,531 (55.6) 10,819 (44.4) 2 (0.0)
3,175 (3.5) 12,587 (13.7) 15,794 (17.2) 7,250 (7.9) 19,597 (21.3) 7,930 (8.6) 10,092 (11.0) 5,632 (6.1) 9,821 (10.7)
2,296 (3.4) 8,976 (13.3) 11,607 (17.2) 5,569 (8.2) 14,090 (20.9) 5,961 (8.8) 7,211 (10.7) 4,339 (6.4) 7,477 (11.1)
879 (3.6) 3,611 (14.8) 4,187 (17.2) 1,681 (6.9) 5,507 (22.6) 1,969 (8.1) 2,881 (11.8) 1,293 (5.3) 2,344 (9.6)
5,939 (6.5) 29,951 (32.6) 55,988 (60.9)
4,561 (6.8) 22,391 (33.2) 40,574 (60.1)
1,378 (5.7) 7,560 (31.0) 15,414 (63.3)
9,556 (10.4) 23,637 (25.7) 58,685 (63.9)
7,212 (10.7) 17,560 (26.0) 42,754 (63.3)
2,344 (9.6) 6,077 (25.0) 15,931 (65.4)
7,399 (8.1) 68,703 (74.8) 15,776 (17.2) 3,167 (3.4) 54,539 (59.4) 37,747 (41.1) 20,699 (22.5) 7,291 (7.9)
5,364 (7.9) 50,637 (75.0) 11,525 (17.1) 2,006 (3.0) 43,677 (64.7) 26,115 (38.7) 11,580 (17.1) 5,039 (7.5)
2,035 (8.4) 18,066 (74.2) 4,251 (17.5) 1,161 (4.8) 10,862 (44.6) 11,632 (47.8) 9,119 (37.4) 2,252 (9.2)
p Value
Univariable Analysis OR (95% CI)
p Value
<0.001
<0.001 <0.001
Reference 1.36 (1.29-1.44) 1.17 (1.10-1.26) 1.27 (1.13-1.43) 1.25 (0.98-1.58) 1.05 (0.95-1.17) 0.83 (0.76-0.90) 1.23 (1.21-1.25)
<0.001 <0.001 <0.001 0.069263 0.326515 <0.001 <0.001
Reference 0.91 (0.88-0.95) 0.88 (0.84-0.92) 0.86 (0.81-0.90) 0.93 (0.82-1.04)
<0.001 <0.001 <0.001 0.205
Reference 1.37 (1.33-1.42) 2.34 (0.37-14.97)
<0.001 0.37
<0.001
<0.001 Reference 1.11 (0.96-1.28) 0.96 (0.83-1.10) 0.78 (0.67-0.91) 1.02 (0.89-1.18) 0.85 (0.72-0.99) 1.08 (0.93-1.24) 0.77 (0.66-0.91) 0.86 (0.74-0.99)
0.1673 0.53899 0.00169 0.74347 0.04208 0.3109 0.0014 0.04159
<0.001 Reference 1.12 (1.02-1.22) 1.29 (1.18-1.41) <0.001 Reference 1.05 (0.98-1.13) 1.11 (1.04-1.19)
0.0139 <0.001
0.18303 0.00204
0.032
<0.001 <0.001 <0.001 <0.001 <0.001
Reference 0.91 (0.84-0.99) 0.93 (0.85-1.02) 1.63 (1.51-1.77) 0.44 (0.42-0.45) 1.43 (1.38-1.47) 2.90 (2.80-3.01) 1.26 (1.19-1.33)
0.0248 0.1238 <0.001 <0.001 <0.001 <0.001 <0.001 (continued on next page)
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Race White Black Hispanic Asian or Pacific Islander Native American Other Unknown Age (y), mean (SD) Median household income of hometown Quartile 1 Quartile 2 Quartile 3 Quartile 4 Unknown Sex Male Female Unknown Hospital division New England Middle Atlantic East North Central West North Central South Atlantic East South Central West South Central Mountain Pacific Hospital teaching status Rural Urban nonteaching Urban teaching Hospital bed size Small Medium Large Hospital ownership Government, nonfederal Private, not-profit Private, invest-own BMI 40 Hypertension Diabetes mellitus Chronic kidney disease Depression
Overall Study Population 91,878
23,240 (95.4) 1,107 (4.5) 5 (0.0) 66,122 (97.9) 1,398 (2.1) 6 (0.0) 89,362 (97.3) 2,505 (2.7) 11 (0.0)
NOTE. Extended hospital stay = greater than 75th percentile for cohort (5 d); p value = Pearson chi-square and independent samples t test for hospital stay cohorts; p value = Wald test. Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio; PCI, percutaneous coronary intervention; SD, standard deviation.
10.86 (9.96-11.83) 2.28 (2.17-2.40) 1.44 (1.36-1.53) 5.72 (5.52-5.92) 1.46 (1.31-1.63) 0.78 (0.75-0.82) 4.69 (3.77-5.84) 0.58 (0.56-0.60) 1.59 (1.54-1.65) 2.81 (2.71-2.92) 0.54 (0.52-0.56) 5.79 (5.40-6.20) 0.91 (0.73-1.15) <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.62 <0.001 2,738 (11.2) 3,735 (15.3) 2,132 (8.8) 13,397 (55.0) 560 (2.3) 3,420 (14.0) 255 (1.0) 15,502 (63.7) 16,606 (68.2) 7,350 (30.2) 17,829 (73.2) 2,769 (11.4) 117 (0.5) 813 (1.2) 4,956 (7.3) 4,276 (6.3) 12,007 (17.8) 1,066 (1.6) 11,728 (17.4) 149 (0.2) 50,793 (75.2) 39,079 (57.9) 9,035 (13.4) 56,119 (83.1) 1,493 (2.2) 344 (0.5) 3,551 (3.9) 8,691 (9.5) 6,408 (7.0) 25,404 (27.6) 1,626 (1.8) 15,148 (16.5) 404 (0.4) 66,295 (72.2) 55,685 (60.6) 16,385 (17.8) 73,948 (80.5) 4,262 (4.6) 461 (0.5)
Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol abuse Smoker Bleeding disorder Dyslipidemia Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock Dyspnea Inpatient mortality No Yes Missing
Hospital Stay Not Extended 67,526 Overall Study Population 91,878
Hospital Stay Extended 24,352
p Value
Univariable Analysis OR (95% CI)
Characteristics
Table 2 (continued )
3039
Table 3 Multivariable Logistic Regression Analysis for All Covariates for Predicting Extended Hospital Length of Stay Built on the Training Set
<0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.439
p Value
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Race White Black Hispanic Asian or Pacific Islander Unknown Age (10-y increase) Median household income of hometown Quartile 1 Quartile 2 Quartile 3 Quartile 4 Sex Male Female Hospital division New England Middle Atlantic West North Central Mountain Pacific Hospital teaching status Rural Urban nonteaching Urban teaching Hospital bed size Small Medium Large Hospital ownership Government, nonfederal Private, not-profit BMI 40 Diabetes mellitus Chronic kidney disease Depression Pneumonia Pulmonary disease Obstructive sleep apnea Heart failure Alcohol abuse Bleeding disorder Myocardial infarction Atrial fibrillation PCI with drug-eluting stent Cardiogenic shock
OR (95% CI)
p Value
Reference 1.24 (1.16-1.32) 1.14 (1.06-1.23) 1.28 (1.12-1.47) 0.89 (0.81-0.98) 1.13 (1.11-1.15)
<0.001 <0.001 <0.001 0.021075 <0.001
Reference 0.95 (0.90-0.99) 0.92 (0.87-0.97) 0.89 (0.84 -0.94)
0.01925 0.001193 <0.001
Reference 1.31 (1.27-1.36)
<0.001
Reference 1.22 (1.05-1.42) 0.79 (0.67-0.94) 0.83 (0.70-0.98) 0.75 (0.64-0.87)
0.009627 0.005754 0.024349 0.000204
Reference 1.20 (1.09-1.32) 1.33 (1.20-1.46)
0.00022 <0.001
Reference 1.11 (1.03-1.20) 1.27 (1.18-1.37)
0.009549 <0.001
Reference 0.89 (0.82-0.97) 1.32 (1.21-1.44) 1.16 (1.12-1.20) 2.09 (2.01-2.17) 1.24 (1.17-1.32) 5.97 (5.46-6.52) 1.79 (1.69-1.89) 1.12 (1.05-1.20) 3.66 (3.53-3.80) 1.71 (1.51-1.92) 2.37 (1.86-3.02) 1.45 (1.40-1.51) 2.02 (1.94-2.11) 0.68 (0.65-0.71) 3.61 (3.35-3.89)
0.006332 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio.
Because Medicare does not reimburse hospitals for unnecessary services or for patients with extended hospital stay, it is prudent to identify and address risk factors of poor outcomes in this patient population. Between 2006 to 2016, the death rate owing to coronary heart disease declined by 31.8%; however, coronary heart disease continues to be a significant contributor to healthcare spending. The literature reporting factors associated with important health quality indices, such as extended hospital stay, remains sparse.19 To improve healthcare quality, the Institute of Medicine recommends performance indices—safety, effectiveness, patient centeredness,
3040
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
Fig 1. Area under the receiver operating curve for predictive model fitted on (A) training set and (B) testing set. AUC, area under the curve; ROC, receiver operating curve.
Prevalence of Extended Hospital Stay by United States Georgraphical Division
Extended Hospital Stay (%)
5 10 15 20
Fig 2. The prevalence of extended hospital stay by US geographical region. Geographical regions with prevalence: New England (3.6%), Mid Atlantic (14.8%), East North Central (17.2%), and West North Central (6.9%), South Atlantic (22.6%), East South Central (8.1%), West South Central (11.8%), Mountain (5.3%), and Pacific (9.6%).
timeliness, efficiency, and equity—of which LOS helps to estimate healthcare efficiency.20 Given that PCI accounts for a significant portion of healthcare spending and that resource utilization and length of hospital stay are quality metrics that help to estimate cost, identifying and addressing modifiable risk factors associated with extended hospital stay should be a high priority. Not surprisingly, modifiable risk factors of coronary atherosclerosis, such as alcohol abuse and BMI, were associated with higher odds of extended hospital stay. Patients should continue to be educated on lifestyle modification and secondary prevention. Patient residence in a higher income
neighborhood was associated with lower odds of extended hospital stay. This finding is consistent with other reports. Compared with nonparticipants, participants in the Supplemental Nutrition Assistance Program, a federal program that provides food assistance to low-income families, had poorer health outcomes, which likely reflects differences in socioeconomic and behavioral characteristics.21,22 We identified inpatient comorbidities (DM, CKD, depression, pneumonia, pulmonary disease, OSA, HF, bleeding disorder, and MI) that may not be “modifiable” prior to PCI, and yet there is a clear mortality benefit to early PCI. Therefore, it is a challenge to
B.N. Burton et al. / Journal of Cardiothoracic and Vascular Anesthesia 33 (2019) 30353041
optimize these patients before PCI, as such, cost-effective alternatives to preprocedural optimization must be considered. Although the present study’s data lack information on the severity of comorbidities, the predictive model developed may help to guide periprocedural risk stratification, optimization, and rehabilitation. Early PCI has a proven mortality benefit in patients with acute MI, and therefore optimization before PCI may not be feasible. For these patients, postprocedural rehabilitation should be the focus, and involvement of multidisciplinary care teams with early discharge planning to acute rehabilitation may improve outcomes and reduce hospital LOS. It is likely that enhanced postprocedural pathways will be the most important mechanism of preventing extended hospital stay because PCI should be performed within 90 minutes of recognition of MI. However, in patients with angina the evidence suggesting early catheterization is less clear. In this patient population, preprocedural risk stratification and optimization should be considered. There are important limitations to the present analysis. NIS is an administrative inpatient database, and important clinical data that may affect the outcome are not available. NIS does not collect information regarding the cause of death; cardiovascular metrics (ejection fraction, degree of stenosis, and number of vessels involved); time to PCI; previous cardiac procedures; medications; lifestyle data; physical activity; laboratory values; intraoperative data; or events that transpire after discharge. The use of ICD-9CM diagnosis codes also pose limitations. Coding practices may vary among hospitals and providers; and therefore patients may be misclassified with respect to their exposure status. The administrative nature of the database with use of ICD-9-CM codes precludes the evaluation of comorbidity severity. Strengths of the present study include a large nationally representative sample size, allowing for the examination of hospital LOS. Rising healthcare costs and poor outcomes have prompted change in the healthcare system. In the last decade, a dramatic shift from volume- to quality-based healthcare has occurred in the United States. The model described here has an excellent AUC; however, it has poor calibration. Therefore, the panoply of risk factors identified in this article may predict patients at risk of extended hospital stay; however, additional prospective studies are needed to develop predictive models with excellent discrimination and calibration. This predictive model can be used to help with periprocedural risk-stratification, optimization, and rehabilitation to improve discharge planning. Conflicts of Interest The authors have no conflicts of interest to disclose. References 1 Benjamin EJ, Virani SS, Callaway CW, et al. Heart disease and stroke statistics-2018 update: A report from the American Heart Association. Circulation 2018;137:e67–e492.
3041
2 Rosamond W, Flegal K, Furie K, et al. Heart disease and stroke statistics— 2008 update: A report from the American Heart Association Statistics Committee and Stroke Statistics Subcommittee. Circulation 2008;117: e25–146. 3 Aggarwal B, Ellis SG, Lincoff AM, et al. Cause of death within 30 days of percutaneous coronary intervention in an era of mandatory outcome reporting. J Am Coll Cardiol 2013;62:409–15. 4 Rasoul S, Ottervanger JP, de Boer MJ, et al. Predictors of 30-day and 1-year mortality after primary percutaneous coronary intervention for STelevation myocardial infarction. Coron Artery Dis 2009;20:415–21. 5 Kaluski E, Alfano D, Randhawa P, et al. Length of hospital stay after percutaneous coronary interventions. J Cardiovasc Nurs 2008;23:345–8. 6 Thomas JW, Guire KE, Horvat GG. Is patient length of stay related to quality of care? Hosp Health Serv Adm 1997;42:489–507. 7 Chin CT, Weintraub WS, Dai D, et al. Trends and predictors of length of stay after primary percutaneous coronary intervention: A report from the Cath PCI registry. Am Heart J 2011;162:1052–61. 8 Amin AP, Patterson M, House JA, et al. Costs associated with access site and same-day discharge among Medicare beneficiaries undergoing percutaneous coronary intervention: An evaluation of the current percutaneous coronary intervention care pathways in the United States. JACC Cardiovasc Interv 2017;10:342–51. 9 Pedersen F, Butrymovich V, Kelbaek H, et al. Short- and long-term cause of death in patients treated with primary PCI for STEMI. J Am Coll Cardiol 2014;64:2101–8. 10 Healthcare Cost and Utilization Project. Overview of the National (Nationwide) Inpatient Sample (NIS). Available at: https://www.hcup-us.ahrq. gov/nisoverview.jsp. Accessed April 29, 2019. 11 Centers of Disease Control and Prevention. International classification of diseases, ninth revision, clinical modification (ICD-9-CM). Available at: https://www.cdc.gov/nchs/icd/icd9cm.htm. Accessed December 16, 2004. 12 American Hospital Association. Fast facts on United States hospitals. Available at: https://www.aha.org/statistics/fast-facts-us-hospitals. Accessed April 21, 2019. 13 Heinze G, Wallisch C, Dunkler D. Variable selection - A review and recommendations for the practicing statistician. Biom J 2018;60:431–49. 14 Centers for Medicare and Medicaid Services. Medicare program - general information. Available at: https://www.cms.gov/Medicare/Medicare-GeneralInformation/MedicareGenInfo/index.html. Accessed February 14, 2018. 15 Kapoor JR, Kapoor R, Hellkamp AS, et al. Payment source, quality of care, and outcomes in patients hospitalized with heart failure. J Am Coll Cardiol 2011;58:1465–71. 16 Pancholy S, Patel G, Pancholy M, et al. Association between health insurance status and in-hospital outcomes after ST-segment elevation myocardial infarction. Am J Cardiol 2017;120:1049–54. 17 Daghistani TA, Elshawi R, Sakr S, et al. Predictors of in-hospital length of stay among cardiac patients: A machine learning approach. Int J Cardiol 2019 Jan 19; [E-pub ahead of print]. 18 Curtis JP, Schreiner G, Wang Y, et al. All-cause readmission and repeat revascularization after percutaneous coronary intervention in a cohort of Medicare patients. J Am Coll Cardiol 2009;54:903–7. 19 Go AS, Mozaffarian D, Roger VL, et al. Heart disease and stroke statistics— 2014 update: A report from the American Heart Association. Circulation 2014;129:e28–292. 20 Agency for Healthcare Research and Quality. Examples of hospital quality measures for consumers. Available at: https://www.ahrq.gov/talkingquality/measures/setting/hospitals/examples.html. Accessed February 14, 2019. 21 Benjamin EJ, Muntner P, Alonso A, et al. Heart disease and stroke statistics-2019 update: A report from the American Heart Association. Circulation 2019;139:e56–528. 22 Murphy SL, Xu J, Kochanek KD, et al. Mortality in the United States, 2017. Available at: https://www.cdc.gov/nchs/data/databriefs/db328-h.pdf. Accessed February 14, 2019.