Currently Viewing:
The American Journal of Managed Care January 2015
Disease-Modifying Therapy and Hospitalization Risk in Heart Failure Patients
Fadia T. Shaya, PhD, MPH; Ian M. Breunig, PhD; and Mandeep R. Mehra, MD, FACC, FACP, FRCP
Frequency and Costs of Hospital Transfers for Ambulatory Care-Sensitive Conditions
R. Neal Axon, MD, MSCR; Mulugeta Gebregziabher, PhD; Janet Craig, PhD, RN; Jingwen Zhang, MS; Patrick Mauldin, PhD; and William P. Moran, MD, MS
Celebrating Our 20th Anniversary
A. Mark Fendrick, MD, and Michael E. Chernew, PhD Co-Editors-in-Chief, The American Journal of Managed Care
Value-Based Insurance Design: Benefits Beyond Cost and Utilization
Teresa B. Gibson, PhD; J. Ross Maclean, MD; Michael E. Chernew, PhD; A. Mark Fendrick, MD; and Colin Baigel, MBChB
Changing Physician Behavior: What Works?
Fargol Mostofian, BHSc; Cynthiya Ruban, BSc; Nicole Simunovic, MSc; and Mohit Bhandari, MD, PhD, FRCSC
State of Emergency Preparedness for US Health Insurance Plans
Raina M. Merchant, MD, MSHP; Kristen Finne, BA; Barbara Lardy, MPH; German Veselovskiy, MPP; Casey Korba, MS; Gregg S. Margolis, NREMT-P, PhD; and Nicole Lurie, MD, MSPH
Relationship of Diabetes Complications Severity to Healthcare Utilization and Costs Among Medicare Advantage Beneficiaries
Leslie Hazel-Fernandez, PhD, MPH; Yong Li, PhD; Damion Nero, PhD; Chad Moretz, ScD; S. Lane Slabaugh, PharmD, MBA; Yunus Meah, PharmD; Jean Baltz, MMSc, MSW; Nick C. Patel, PharmD, PhD; and Jonathan R. Bouchard, MS, RPh
Currently Reading
Revisiting Hospital Length of Stay: What Matters?
Mollie Shulan, MD; and Kelly Gao
Value-Based Insurance Design and Medication Adherence: Opportunities and Challenges
Kevin A. Look, PharmD, PhD
New Start Versus Continuing Users on Aripiprazole: Implications for Policy
Rashid Kazerooni, PharmD, BCPS; Joseph B. Nguyen, PharmD, BCPS; Mark Bounthavong, PharmD, MPH; Michael H. Tran, PharmD, BCPS; and Nermeen Madkour, PharmD, CSP
Multiple Chronic Conditions in Type 2 Diabetes Mellitus: Prevalence and Consequences
Pei-Jung Lin, PhD; David M. Kent, MD, MSc; Aaron Winn, MPP; Joshua T. Cohen, PhD; and Peter J. Neumann, ScD
Prognostic Factors of Mortality Among Patients With Severe Hyperglycemia
Ya-Wun Guo, MD; Tzu-En Wu, MD, MS; and Harn-Shen Chen, MD, PhD
Survey Nonresponders Incurred Higher Medical Utilization and Lower Medication Adherence
Seppo T. Rinne, MD, PhD; Edwin S. Wong, PhD; Jaclyn M. Lemon, BS; Mark Perkins, PharmD; Christopher L. Bryson, MD; and Chuan-Fen Liu, PhD
Using Financial Incentives to Improve the Care of Tuberculosis Patients
Cheng-Yi Lee, MS; Mei-Ju Chi, PhD; Shiang-Lin Yang, MS; Hsiu-Yun Lo, PhD; and Shou-Hsia Cheng, PhD

Revisiting Hospital Length of Stay: What Matters?

Mollie Shulan, MD; and Kelly Gao
Over 50% of inpatients are unmarried and experience 22% longer LOS. Racial/income disparities are not unavoidable; how care is paid for and delivered may make a difference.
ABSTRACT
Objectives
To explore actionable information that can be used to reduce hospital acute care length of stay (LOS) and to assess racial and income disparities in LOS in an integrated healthcare network.

Study Design and Methods
Retrospective analysis of 8718 inpatients in an integrated healthcare network. The LOS was examined by using linear, log-linear, Poisson, generalized Poisson, and negative binomial (NB) models to control for confounding factors. The performances of the 5 models were compared, and the NB model was selected for the final analysis and report.

Results
Over 50% of the inpatients were not married. The LOS was 22% longer for the unmarried patients compared with their married counterparts after controlling for confounding factors. No income or racial disparities were found.

Conclusions
The prolonged LOS of the unmarried patients and the potential lack of post discharge care support warrant greater attention from discharge planners at hospital level and from policy makers at both the national and local levels. Racial and income disparities are not unavoidable; the way in which care is paid for and delivered may play a role.

Am J Manag Care. 2015;21(1):e71-e77
  • Over 50% of the inpatients were not married, and their length of stay (LOS) was 22% longer compared with their married counterparts after controlling for confounding factors.
  • The prolonged LOS of the unmarried patients, and the potential lack of post discharge care support, warrant greater attention from discharge planners at the hospital level and from policy makers at the national as well as local levels.
  • No racial and income disparities in LOS were found in an integrated healthcare network. How care is paid for and delivered may make a difference in reducing these disparities elsewhere.
Hospital length of stay (LOS) has long been a crucial barometer of hospital efficiency and quality of care. Longer stays result in higher costs and extra burdens on patients and their families. The clinical and financial ramifications have made LOS one of the most watched measures in all hospitals and healthcare systems.1-5

Substantial literature has been focused on how LOS is affected by hospital teaching mission, for-profit status, hospitalists, facility physician involvement, nurse staffing, hospital volume, and patient insurance status.6-16 However, analyses of LOS to identify the contributing factors that can be acted on have been relatively sparse compared with analyses of other measures, such as hospital readmissions; for example, racial and income disparities in readmissions have been extensively analyzed and reported, but little has been done to analyze their possible effects on LOS.17-19

The primary aim of this study is to explore actionable information that can be used to reduce LOS; to do so, we focus on demographic and socioeconomic factors such as marital status and income, while comprehensively controlling for confounding factors such as case mix. The secondary aim is to assess racial and income disparities in LOS in an integrated healthcare network.

METHODS

Study Population and Data Sources


This study analyzed all of the 8718 patients hospitalized for acute care in fiscal year (FY) 2011 in Veterans Healthcare Network Upstate New York (VISN 2)—1 of the 21 integrated service networks of the Veterans Administration (VA). VISN 2, with 5 medical centers (4 providing inpatient services) and 31 outpatient clinics across upstate New York, serves 140,000 patients with an annual budget of over $1 billion.

The VA’s centralized National Patient Care Database (NPCD), hosted at the Austin Information Technology Center, was the primary data source, containing detailed patient demographic, socioeconomic, and clinical information (eg, marital status, income, and diagnoses).20 For case mix or patient risk adjustment, we used the risk score produced by DxCG,21-25 which has been used by the VA to systematically measure all of its 5.7 million patients’ risks over the last decade.

Most of the data fields in the NPCD files, such as admission/discharge dates and International Classification of Diseases, Ninth Revision, Clinical Modification codes, are routinely and rigorously validated with strict business rules. Its income information is means tested. One exception is that race information is often incomplete because the VA does not mandate patients to report racial status. However, for the last several years, the VA has systematically gathered race information from other data sources (eg, Medicare and the Department of Defense), and as a result, the updated race status is deemed reliable.26,27

Dependent and Independent Variables

The dependent variable is the LOS of the first hospitalization in FY 2011, excluding hospitalizations of the patients transferred from other hospitals. LOS is defined as discharge date minus admission date plus 1 (sensitivity analysis was conducted without adding 1).1 Guided by the literature and based on data availability in the VA, the independent variables are extracted from 4 categories (as shown in Table 1): 1) demographic and socioeconomic variables, 2) prior year patient care cost (FY 2010), 3) hospital characteristics (captured by fixed effect), and 4) risk score and comorbidities.20

In analyzing the effects of independent variables on dependent variables, proper patient risk adjustment is imperative for reliable results. In the literature, the use of case mix has been rather ad hoc, ranging from a few co-existing conditions to comprehensive measures such as Charlson Comorbidity Index scores.19,28-31 In this study, we used DxCG risk score as the aggregated case-mix measure. DxCG is a well-validated algorithm that most studies have found superior to others in predicting resource use.22-25 In spite of its superiority, we supplemented DxCG with a set of the most prevalent and/or expensive chronic conditions: hypertension, diabetes, congestive heart failure, chronic obstructive pulmonary disease, cancer, and depression.

This study only used existing data that include no identifiable patient private information, and therefore was exempt from Institutional Review Board review under VA Title 38, Section 16.101(b)(4).

Analyses

In this study, we first conducted univariate analyses by using 1-way ANOVA, and then we fitted multivariate regressions to control for the confounders. In the literature, 5 regression models are often used to analyze LOS: ordinary least square (OLS),32-34 loglinear regression,10-16,35-37 Poisson,17 generalized Poisson, and Negative Binomial (NB) models.38-40 For reliable results, in this study we assessed the performances of the 5 models. We found OLS and log-linear regression were inappropriate because even the latter failed to produce an approximately normal distribution (Figure), and the normality of the log-linear regression residuals was also rejected by the Kolmogorov & Smirnov test (P = .01). We selected the NB model for the final analysis because it outperformed Poisson and generalized Poisson models with Pearson overdispersion scales of 1.79 (NB), 6.99 (Poisson), and 1.83 (generalized Poisson), respectively. 41,42 Further, to make sure the results were not affected by any potentially remaining overdispersion, we winsorized LOS at 3 weeks (21 days) and 4 weeks (28 days) as sensitivity analyses.43

Sound model selection is necessary for obtaining reliable results, but does not guarantee them. The accuracy of the estimates heavily relies on how the confounding factors are controlled for; to ensure reliable results, we configured 3 NB regression models to control for confounding factors at different levels. In model 1, we included demographic/socioeconomic variables, prior year patient cost, and medical center characteristics (fixed effect: 4 dummy variables and 1 was omitted as the reference); in model 2, we added DxCG risk score as the total measure of disease severity; and in model 3, we further augmented DxCG risk score with 6 chronic conditions to be sure the results were not skewed by imperfect casemix measurement.

We also conducted extensive sensitivity analyses, such as modeling age and income as categorical variables and reclassifying blacks and other minority groups to ensure the robustness of our results. All analyses were conducted by using Proc GLM and GLIMMIX of SAS 9.2 (Cary, North Carolina).

RESULTS

The results of the univariate and multivariate analyses are reported in Table 1 and Table 2. Among the variables of importance, black patients appeared to have longer LOS in the univariate analysis: in the first quartile 9.9% were black, while in the fourth quartile, 12.2% were black (P = .1253). However, this weak association faded in models 2 and 3 after controlling for patient risk and other confounders (P = .1543 and .2622). Equally important, there was no income disparity in LOS. The patient income changed little across the 4 quartiles (P = .6525) and there were hardly any associations between income and LOS in all 3 models (P = .8726, .9449, and .8303).

Of particular interest, unmarried patients were more likely to have longer LOS. In the univariate analysis, 56% of the patients were unmarried in the first quartile, while 65% were unmarried in the last quartile (P <.0001). After adjusting for different confounders, all 3 models confirmed that unmarried patients experienced longer LOS (all P values <.0001). The final model indicated that the LOS of the unmarried patients was 22% longer than that of their married counterparts.

The coefficient estimates of the confounders, such as age, comorbidities, and prior year cost, were consistent with prior findings.44 In particular, patients enrolled in Medicare were more likely to experience longer LOS (P = .2199, .0052, and .0058). As expected, DxCG risk score was positively associated with LOS (all P values <.0001) while only depression among the 6 chronic conditions appeared to be associated with longer LOS (P = .0254 and P = .0909). Finally, patients in all 3 medical centers (A, B, and C) on average had longer LOS compared with patients in medical center D (omitted in the regression as the reference; all P values <.0001).

DISCUSSION

In this study, we used the NB model to analyze the LOS of 8718 patients in a veterans’ integrated healthcare network. To ensure that the findings were reliable, we conducted extensive sensitivity analyses: 1) fitted 3 NB models with different levels of risk adjustments, and winsorized LOS at 21 and 28 days; 2) fitted log-linear, Poisson, and generalized Poisson regressions to confirm the findings; 3) regrouped racial status (ie, combined all minority groups together vs white); 4) reanalyzed the data by changing continuous variables such as age and income to categorical variables; 5) analyzed the data without adding 1 to LOS; and 6) examined the effect of readmissions. All of these analyses confirmed that the findings were consistent and robust.

Racial disparity has been found pervasive in other healthcare outcomes, such as readmissions18,19; however, racial disparity in LOS has rarely been studied. In fact, we identified only 1 study that examined racial disparities in LOS for asthma, which found LOS wasone-fourth to one-third shorter for nonwhites versus whites.17 In this study, we found no racial disparities in LOS in an integrated healthcare network, which suggests that racial disparities are not necessarily universal and may be dependent on how healthcare is paid for and delivered. Unlike most US healthcare systems, patients in the VA system have few financial barriers; the VA clinicians are salaried and have no financial incentive to keep patients in the hospital for shorter or longer stays. More importantly, the VA has reformed itself to focus on patient-centered care since the inception of integrated healthcare networks in 1995.45 These might be among the factors that could reduce disparities.

Income disparity in LOS was also not found in this study. Income disparity has not been well studied compared with racial disparity in the literature because, we speculate, income data are often unavailable. It also should be noted, without controlling for income, analyzing racial disparity could yield misleading results, since race and income are often correlated46; as a result, what appears to be a racial issue could actually be an income issue.

 
Copyright AJMC 2006-2018 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
x
Welcome the the new and improved AJMC.com, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up