[Skip to Content]
[Skip to Content Landing]
Figure.
Association Between Hospital-Level ICU Utilization and Hospital Mortality for the 4 Study Conditions
Association Between Hospital-Level ICU Utilization and Hospital Mortality for the 4 Study Conditions

Increased utilization of the intensive care unit (ICU) was not associated with significant differences in hospital mortality for any of the 4 study conditions (diabetic ketoacidosis [DKA], pulmonary embolism [PE], upper gastrointestinal bleeding [UGIB], and congestive heart failure [CHF]). The slopes of the linear regression analyses between hospital mortality and ICU utilization were −0.0013% mortality per percent predicted ICU admission for DKA (P = .22); 0.0061% for PE (P = .59), −0.0061% for UGIB (P = .49), and 0.0083% for CHF (P = .37).

Table 1.  
Baseline Patient Characteristics by Clinical Condition and Hospital ICU Utilization Rate (Low vs High)a
Baseline Patient Characteristics by Clinical Condition and Hospital ICU Utilization Rate (Low vs High)a
Table 2.  
Baseline Hospital Characteristics by ICU Utilization Rate (Low vs High)a
Baseline Hospital Characteristics by ICU Utilization Rate (Low vs High)a
1.
Halpern  NA, Pastores  SM.  Critical care medicine in the United States 2000-2005: an analysis of bed numbers, occupancy rates, payer mix, and costs.  Crit Care Med. 2010;38(1):65-71.PubMedGoogle ScholarCrossref
2.
Wunsch  H, Angus  DC, Harrison  DA,  et al.  Variation in critical care services across North America and Western Europe.  Crit Care Med. 2008;36(10):2787-2793, e1-e9.PubMedGoogle ScholarCrossref
3.
Wunsch  H, Angus  DC, Harrison  DA, Linde-Zwirble  WT, Rowan  KM.  Comparison of medical admissions to intensive care units in the United States and United Kingdom.  Am J Respir Crit Care Med. 2011;183(12):1666-1673.PubMedGoogle ScholarCrossref
4.
Porter  ME.  What is value in health care?  N Engl J Med. 2010;363(26):2477-2481.PubMedGoogle ScholarCrossref
5.
Admon  AJ, Seymour  CW, Gershengorn  HB, Wunsch  H, Cooke  CR.  Hospital-level variation in ICU admission and critical care procedures for patients hospitalized for pulmonary embolism.  Chest. 2014;146(6):1452-1461.PubMedGoogle ScholarCrossref
6.
Gershengorn  HB, Iwashyna  TJ, Cooke  CR, Scales  DC, Kahn  JM, Wunsch  H.  Variation in use of intensive care for adults with diabetic ketoacidosis*.  Crit Care Med. 2012;40(7):2009-2015.PubMedGoogle ScholarCrossref
7.
Safavi  KC, Dharmarajan  K, Kim  N,  et al.  Variation exists in rates of admission to intensive care units for heart failure patients across hospitals in the United States.  Circulation. 2013;127(8):923-929.PubMedGoogle ScholarCrossref
8.
Healthcare Cost and Utilization Project (HCUP). HCUP Supplemental Variables for Revisit Analysis, June 2014. Agency for Healthcare Research and Quality, Rockville, MD. http://www.hcup-us.ahrq.gov/toolssoftware/revisit/revisit.jsp. Accessed June 29, 2016.
9.
United States Department of Health and Human Services; Centers for Medicare and Medicaid Services. Hospital Cost Report, 2011. https://www.cms.gov/Research-Statistics-Data-and-Systems/Downloadable-Public-Use-Files/Cost-Reports/Cost-Reports-by-Fiscal-Year.html. Accessed September 15, 2015.
10.
Cooper  GS, Chak  A, Lloyd  LE, Yurchick  PJ, Harper  DL, Rosenthal  GE.  The accuracy of diagnosis and procedural codes for patients with upper GI hemorrhage.  Gastrointest Endosc. 2000;51(4, pt 1):423-426.PubMedGoogle ScholarCrossref
11.
White  RH, Romano  PS, Zhou  H, Rodrigo  J, Bargar  W.  Incidence and time course of thromboembolic outcomes following total hip or knee arthroplasty.  Arch Intern Med. 1998;158(14):1525-1531.PubMedGoogle ScholarCrossref
12.
Wiener  RS, Schwartz  LM, Woloshin  S.  Time trends in pulmonary embolism in the United States: evidence of overdiagnosis.  Arch Intern Med. 2011;171(9):831-837.PubMedGoogle ScholarCrossref
13.
Chen  HF, Wang  CY, Lee  HY,  et al.  Short-term case fatality rate and associated factors among inpatients with diabetic ketoacidosis and hyperglycemic hyperosmolar state: a hospital-based analysis over a 15-year period.  Intern Med. 2010;49(8):729-737.PubMedGoogle ScholarCrossref
14.
Dharmarajan  K, Hsieh  AF, Lin  Z,  et al.  Diagnoses and timing of 30-day readmissions after hospitalization for heart failure, acute myocardial infarction, or pneumonia.  JAMA. 2013;309(4):355-363.PubMedGoogle ScholarCrossref
15.
Keenan  PS, Normand  SL, Lin  Z,  et al.  An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all-cause readmission rates among patients with heart failure.  Circ Cardiovasc Qual Outcomes. 2008;1(1):29-37.PubMedGoogle ScholarCrossref
16.
Quan  H, Parsons  GA, Ghali  WA.  Validity of procedure codes in International Classification of Diseases, 9th revision, clinical modification administrative data.  Med Care. 2004;42(8):801-809.PubMedGoogle ScholarCrossref
17.
Walkey  AJ, Wiener  RS, Lindenauer  PK.  Utilization patterns and outcomes associated with central venous catheter in septic shock: a population-based study.  Crit Care Med. 2013;41(6):1450-1457.PubMedGoogle ScholarCrossref
18.
Deyo  RA, Cherkin  DC, Ciol  MA.  Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases.  J Clin Epidemiol. 1992;45(6):613-619.PubMedGoogle ScholarCrossref
19.
Charlson  ME, Pompei  P, Ales  KL, MacKenzie  CR.  A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis. 1987;40(5):373-383.PubMedGoogle ScholarCrossref
20.
D’Hoore  W, Bouckaert  A, Tilquin  C.  Practical considerations on the use of the Charlson comorbidity index with administrative data bases.  J Clin Epidemiol. 1996;49(12):1429-1433.PubMedGoogle ScholarCrossref
21.
Angus  DC, Linde-Zwirble  WT, Lidicker  J, Clermont  G, Carcillo  J, Pinsky  MR.  Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care.  Crit Care Med. 2001;29(7):1303-1310.PubMedGoogle ScholarCrossref
22.
Healthcare Cost and Utilization Project (HCUP). HCUP Cost-to-Charge Ratio Files, July 2014. Agency for Healthcare Research and Quality, Rockville, MD. http://www.hcup-us.ahrq.gov/db/state/costtocharge.jsp. Accessed June 29, 2016.
23.
Talmor  D, Greenberg  D, Howell  MD, Lisbon  A, Novack  V, Shapiro  N.  The costs and cost-effectiveness of an integrated sepsis treatment protocol.  Crit Care Med. 2008;36(4):1168-1174.PubMedGoogle ScholarCrossref
24.
Morris  CN.  Parametric empirical Bayes inference: theory and applications.  J Am Stat Assoc. 1983;78(381):47-55.Google ScholarCrossref
25.
Normand  S-LT, Glickman  ME, Gatsonis  CA.  Statistical methods for profiling providers of medical care: issues and applications.  J Am Stat Assoc. 1997;92(439):803-814.Google ScholarCrossref
26.
Seymour  CW, Iwashyna  TJ, Ehlenbach  WJ, Wunsch  H, Cooke  CR.  Hospital-level variation in the use of intensive care.  Health Serv Res. 2012;47(5):2060-2080.PubMedGoogle ScholarCrossref
27.
Royston  P, Altman  DG, Sauerbrei  W.  Dichotomizing continuous predictors in multiple regression: a bad idea.  Stat Med. 2006;25(1):127-141.PubMedGoogle ScholarCrossref
28.
Streiner  DL.  Breaking up is hard to do: the heartbreak of dichotomizing continuous data.  Can J Psychiatry. 2002;47(3):262-266.PubMedGoogle Scholar
29.
Sjoding  MW, Prescott  HC, Wunsch  H, Iwashyna  TJ, Cooke  CR.  Hospitals with the highest intensive care utilization provide lower quality pneumonia care to the elderly.  Crit Care Med. 2015;43(6):1178-1186.PubMedGoogle ScholarCrossref
30.
Valley  TS, Sjoding  MW, Ryan  AM, Iwashyna  TJ, Cooke  CR.  Association of intensive care unit admission with mortality among older patients with pneumonia.  JAMA. 2015;314(12):1272-1279.PubMedGoogle ScholarCrossref
31.
Wunsch  H, Gershengorn  HB, Cooke  CR,  et al.  Use of intensive care services for Medicare beneficiaries undergoing major surgical procedures.  Anesthesiology. 2016;124(4):899-907.PubMedGoogle ScholarCrossref
Original Investigation
Less Is More
October 2016

Association Between Intensive Care Unit Utilization During Hospitalization and Costs, Use of Invasive Procedures, and Mortality

Author Affiliations
  • 1Division of Respiratory and Critical Care Physiology and Medicine, Los Angeles Biomed Research Institute, Harbor–University of California, Los Angeles Medical Center
  • 2Division of General Internal Medicine and Health Services Research, David Geffen School of Medicine, University of California, Los Angeles
JAMA Intern Med. 2016;176(10):1492-1499. doi:10.1001/jamainternmed.2016.4298
Abstract

Importance  Maximizing the value of critical care services requires understanding the relationship between intensive care unit (ICU) utilization, clinical outcomes, and costs.

Objective  To examine whether hospitals had consistent patterns of ICU utilization across 4 common medical conditions and the association between higher use of the ICU and hospital costs, use of invasive procedures, and mortality.

Design, Setting, and Participants  Retrospective cohort study of 156 842 hospitalizations in 94 acute-care nonfederal hospitals for diabetic ketoacidosis (DKA), pulmonary embolism (PE), upper gastrointestinal bleeding (UGIB), and congestive heart failure (CHF) in Washington state and Maryland from 2010 to 2012. Hospitalizations for DKA, PE, UGIB, and CHF were identified from the presence of compatible International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes. Multilevel logistic regression models were used to determine the predicted hospital-level ICU utilization during hospitalizations for the 4 study conditions. For each condition, hospitals were ranked based on the predicted ICU utilization rate to examine the variability in ICU utilization across institutions.

Main Outcomes and Measures  The primary outcomes were associations between hospital-level ICU utilization rates and risk-adjusted hospital mortality, use of invasive procedures, and hospital costs.

Results  The 94 hospitals and 156 842 hospitalizations included in the study represented 4.7% of total hospitalizations in this study. ICU admission rates ranged from 16.3% to 81.2% for DKA, 5.0% to 44.2% for PE, 11.5% to 51.2% for UGIB, and 3.9% to 48.8% for CHF. Spearman rank coefficients between DKA, PE, UGIB, and CHF showed significant correlations in ICU utilization for these 4 medical conditions among hospitals (ρ ≥ 0.90 for all comparisons; P < .01 for all). For each condition, hospital-level ICU utilization rate was not associated with hospital mortality. Use of invasive procedures and costs of hospitalization were greater in institutions with higher ICU utilization for all 4 conditions.

Conclusions and Relevance  For medical conditions where ICU care is frequently provided, but may not always be necessary, institutions that utilize ICUs more frequently are more likely to perform invasive procedures and have higher costs but have no improvement in hospital mortality. Hospitals had similar ICU utilization patterns across the 4 medical conditions, suggesting that systematic institutional factors may influence decisions to potentially overutilize ICU care. Interventions that seek to improve the value of critical care services will need to address these factors that lead clinicians to admit patients to higher levels of care when equivalent care can be delivered elsewhere in the hospital.

Introduction

Optimizing the utilization of intensive care units (ICUs) is an important health care priority in the United States. Overuse of ICUs for patients who can likely receive equivalent care in non-ICU settings may lead to invasive, potentially harmful, care and decrease access for patients for whom critical care services may be beneficial. In addition, US critical care services consume 13.4% of total hospital costs and 4.1% of national health expenditures; both these costs and the number of ICU beds are rising.1-3 The potential clinical consequences of overusing ICU care along with the rapid growth and high costs have made improving its value crucial for the viability of the US health care system.1,4

Central to improving value is identifying patients who derive the most benefit from critical care services and reducing variation in ICU admission practices that do not translate to improved clinical outcomes. However, lack of clear-cut guidelines for ICU admissions and differences in institutional resources, policies, and culture have resulted in significant variability in utilization among hospitals.2,5,6

Previous studies have shown that for diabetic ketoacidosis (DKA), pulmonary embolism (PE), and congestive heart failure (CHF), there is wide hospital-level variation in utilization of critical care services without a change in mortality.5-7 However, these studies were limited to single diseases, making it challenging to determine whether the variability is due to nuanced institutional differences in disease management or due to systematic factors related to overall ICU use. Determining whether variation in ICU utilization stems from disease-specific or institutional factors is an important step in developing interventions to reduce unwanted variation.

Some conditions, such as acute respiratory failure and septic shock, are nearly universally recognized as requiring ICU admission in the absence of an advanced directive to limit intensity of care. Other conditions, such as cellulitis or deep venous thrombosis, rarely require the ICU, and few patients are transferred there. For still other conditions, the potential need for and benefits from ICU care may be subject to variable opinions and practices. We studied 4 such medical conditions to examine (1) variability in ICU utilization across hospitals, (2) whether hospitals are more likely to use ICU services for all 4 conditions or whether there is significant variability within hospitals in level of ICU use across diseases, and (3) the association between hospitals that use ICU care more frequently and hospital-level mortality, use of invasive procedures, and hospital costs for each condition. We hypothesized that some hospitals utilize ICU care more frequently across multiple diseases and that this increased utilization is not associated with decreased mortality.

Box Section Ref ID

Key Points

  • Question Do hospitals have similar patterns of intensive care unit (ICU) utilization across 4 common medical conditions, and is increased ICU utilization associated with better clinical outcomes?

  • Findings In this retrospective cohort study, 94 hospitals were ranked by their predicted ICU utilization rates for diabetic ketoacidosis, pulmonary embolism, upper gastrointestinal bleeding, and congestive heart failure, and consistent patterns were found across all 4 conditions. Hospitals that used ICUs more frequently were more likely to perform invasive procedures and have higher costs but have no improvement in mortality.

  • Meaning Optimizing the value of ICU care will require assessments of systematic institutional factors that may lead clinicians to overutilize ICU care.

Methods

This study was approved by the institutional review board at the Los Angeles Biomedical Research Institute, waiving written informed consent.

Data Source and Study Population

Data were obtained from the Agency for Healthcare Research and Quality’s Healthcare Cost and Utilization Project State Inpatient Databases for Washington and Maryland.8 Hospital characteristics were obtained by linking the databases with the Healthcare Cost Report Information System.9 Additional details regarding these databases are provided in eMethods in the Supplement.

This was a retrospective cohort study of adult patients (age ≥18 years) hospitalized for DKA, PE, upper gastrointestinal bleeding (UGIB), and CHF from 2010 through 2012. Hospitalizations for each disease were identified based on compatible International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes as the principal diagnosis that was present at admission. The ICD-9-CM codes used to define the cohorts with DKA, PE, UGIB, and CHF were consistent with previous studies that have examined these diseases using administrative data and are provided in eMethods in the Supplement.5,6,10-15

There were 4 104 720 hospitalizations in 217 hospitals in the database during the study period. Hospitalizations were excluded from analysis if patients were younger than 18 years (n = 635 412), if they were hospitalized in facilities without ICU beds (n = 125 127; 113 hospitals), or information on hospital-level characteristics was missing (n = 4699; 10 hospitals), leaving 3 339 482 total hospitalizations in 94 hospitals available for analysis.

Study Variables
ICU Admission

The proportion of hospitalizations with ICU stays at each hospital was identified using the presence of ICU or coronary care unit room and board charges in the revenue codes (eMethods in the Supplement). Under this approach, the timing of ICU care (whether on admission or later during the hospitalization) was not available.

Outcomes and Adjustment Variables

The primary outcomes were risk-adjusted hospital-level mortality, use of invasive procedures, and hospital costs. We examined use of central venous catheters in all 4 medical conditions and mechanical ventilation in DKA, thrombolytics in PE, esophagogastroduodenoscopy in UGIB, and pulmonary artery catheterization in CHF. Utilization of these procedures and therapies was identified by the presence of corresponding ICD-9-CM codes (eMethods in the Supplement).5,10,16,17 Variables for risk adjustment included patient factors (age, race, sex, payer status, median income by zip code, comorbid conditions, and number of organ system failures) and hospital factors (number of beds, teaching status, percentage of hospital beds designated as ICU beds, and ICU occupancy). Comorbid conditions were identified using discharge data codes to calculate a Charlson Comorbidity Index.18-20 Organ system failures were identified by the presence of ICD-9-CM codes corresponding to respiratory, circulatory, renal, neurologic, hematologic, and hepatic failures as defined by Angus et al.21 Cost of hospitalization was calculated from total charges using hospital-specific cost-to-charge ratios and adjusted to 2012 dollar values using the medical component of the Consumer Price Index.22,23

Statistical Analysis

To assess variation in ICU admission rates across hospitals, we created hierarchical mixed-effects logistic regression models to estimate the odds of having an ICU admission in hospitalizations for DKA, PE, UGIB, and CHF in each hospital. We adjusted for patient- and hospital-level effects by adding adjustment variables as fixed effects and used individual hospitals as the random effect to account for clustering of hospitalizations within hospitals. Using empirical Bayesian posterior estimates from these logistic regression models, we determined the predicted ICU utilization rate (risk- and reliability-adjusted) and 95% CIs for each hospital during hospitalizations for DKA, PE, UGIB, and CHF.24-26 For each condition, hospitals were ranked based on the predicted ICU utilization rate. The ranges of these rates were used to examine the variability in ICU utilization across hospitals. We studied whether hospitals had consistent patterns of ICU utilization by examining the correlation between ICU utilization rates for DKA, PE, UGIB, and CHF among hospitals using the Spearman rank correlation.

We calculated risk-adjusted hospital-level mortality rates for each condition using empirical Bayesian posterior estimates from a separate set of multilevel logistic regression models using hospital mortality as the dependent variable. Patient-level adjustment variables were included as fixed effects, and individual hospitals as random effects to account for clustering of hospitalizations. We used a parallel approach to calculate risk-adjusted use of invasive procedures and therapies for each hospital in all 4 conditions. Similarly, we used multilevel linear regression models to calculate risk-adjusted hospital-level lengths of stay and costs for each condition. We log-transformed lengths of stay and costs prior to creating linear regression models owing to their nonnormal distributions.

The relationships between ICU admission rate (as a continuous variable) and hospital mortality, use of invasive procedures, and total costs were examined using linear regression analyses.27,28 In addition, the study outcomes were also compared after hospitals were dichotomized into higher (above 50th percentile for predicted ICU utilization rate) and lower (50th percentile and below) ICU utilization groups for each condition. These dichotomized analyses were performed because the results from these analyses have more intuitive clinical interpretations. All data analyses were performed using JMP software, version 11.0 (SAS Institute Inc), and SAS for Windows, version 9.4 (SAS Institute Inc).

Results
Baseline Characteristics

After exclusions, there were 94 acute-care hospitals with ICUs in Washington state and Maryland during the study period with 156 842 hospitalizations for the 4 study conditions, representing 4.7% of total hospitalizations in these institutions (Table 1). The median ICU admission rates for lower vs higher utilization hospitals were 43.6% and 67.2%, respectively, for DKA, 12.2% and 26.5% for PE, 23.4% and 34.2% for UGIB, and 9.6% and 28.7% for CHF (Table 2). Hospitalizations with ICU stays for these 4 conditions represented 7.8% of ICU hospitalizations in these institutions. For each condition, smaller hospitals (fewer hospital beds) were more frequently in the higher ICU utilization group (Table 2). Hospital and ICU occupancy were similar between the higher and lower ICU utilization groups for each condition (eTable 2 in the Supplement). Teaching hospitals were more frequently in the higher utilization groups for each condition (Table 2).

Patient characteristics for DKA, PE, UGIB, and CHF hospitalizations stratified by higher and lower ICU utilization groups and whether the hospitalization included an ICU stay are listed in Table 1. The distributions of age, sex, race, payer status, and median incomes were similar between the higher and lower ICU utilization groups for each condition (eTable 1 in the Supplement). With the exception of DKA, there were generally more comorbidities among ICU hospitalizations in lower utilization hospitals (Table 1). There were more organ system failures among ICU hospitalizations in lower utilization institutions for each condition (Table 1). The distributions of comorbidities and organ system failures among non-ICU hospitalizations between the lower and higher ICU utilization hospitals for DKA, PE, UGIB, and CHF were similar (Table 1).

Hospital-Level Variability in ICU Hospitalizations

Risk-adjusted ICU admission rates varied substantially for each condition (eFigure 1 in the Supplement). Adjusted mean ICU admission rates for hospitals ranged from 16.3% to 81.2% for DKA, 5.0% to 44.2% for PE, 11.5% to 51.2% for UGIB, and 3.9% to 48.8% for CHF (eFigure 1 in the Supplement). Spearman rank coefficients (ρ) showed extremely high correlations between ICU utilization rates for DKA, PE, UGIB, and CHF among hospitals (ρ ≥ 0.90 for all comparisons, with 1.0 representing perfect correlation of ranks; P < .01 for all comparisons). Risk-adjusted ICU admission rates for DKA, PE, UGIB, and CHF in each hospital are listed in eTable 4 in the Supplement.

Hospital Mortality and Use of Invasive ICU Interventions

Increased ICU utilization was not associated with significant differences in hospital mortality for any condition (Figure). The slopes of the linear regression analyses between hospital mortality and ICU utilization were −0.0013% mortality per percent predicted ICU admission for DKA (P = .22); 0.0061% for PE (P = .59), −0.0061% for UGIB (P = .49), and 0.0083% for CHF (P = .37). In dichotomized analyses, the respective risk-adjusted mortality rates for hospitals in lower and higher ICU utilization groups were 0.30% and 0.26% for DKA, 2.85% and 2.98% for PE, 2.08% and 2.03% for UGIB, and 2.99% and 3.02% for CHF (eTable 5 in the Supplement).

With respect to intensity of care delivery, the slopes of the linear regression analyses between ICU utilization and central venous catheter use showed that increased ICU utilization was associated with higher central venous catheter use for each condition (eFigure 2 in the Supplement). Similar relationships were seen between ICU utilization and use of all other invasive procedures (eFigure 3 in the Supplement). When hospitals were dichotomized to higher and lower utilization groups, rates of invasive procedures in all 4 conditions were greater in higher ICU utilization hospitals (eTable 5 in the Supplement). Hospital mortality rates and use of invasive interventions in higher and lower ICU utilization groups stratified by hospitalizations with and without ICU stays are listed in eTable 3 in the Supplement.

Hospital Costs and Length of Stay

Increased ICU use was associated with higher costs for DKA, PE, UGIB, and CHF (eFigure 4 in the Supplement). The slopes of the linear regression analyses between ICU utilization and costs were $33.85 per percent predicted ICU admission for DKA (P < .001), $30.18 for PE (P = .02), $29.89 for UGIB (P = .01), and $97.83 for CHF (P < .001) (eFigure 4 in the Supplement). The respective adjusted costs of hospitalization for institutions in lower and higher ICU utilization groups were $7141 and $8204 for DKA, $10 660 and $11 117 for PE, $10 164 and $10 851 for UGIB, and $10 175 and $13 587 for CHF (eTable 5 in the Supplement). Lengths of stay were similar between lower and higher ICU utilization hospitals for each condition (eTable 5 in the Supplement). Total costs and length of stay in the higher and lower ICU utilization groups stratified by hospitalizations with and without ICU stays are listed in eTable 3 in the Supplement.

Discussion

Overuse of ICUs among patients who can likely be treated in non-ICU settings may lead to inappropriately aggressive care and misallocation of resources away from patients who may truly need critical care services. These clinical concerns, as well as the high cost and resource utilization of critical care services, make it imperative to identify the clinical situations for which patients derive the most benefit when receiving ICU care. Our findings reflect the clinical challenge of identifying patients who benefit most from ICU admission and show significant variability in ICU utilization during hospitalizations for DKA, PE, UGIB, and CHF, despite adjusting for patient- and hospital-level risk factors. The medical conditions in our study frequently require physicians to make clinical judgments regarding whether ICU admission is necessary, and our findings provide insight into clinical practices for ICU utilization among patients who may or may not require ICU care. We found a consistent pattern of ICU utilization among institutions across these conditions, suggesting that systematic institutional factors strongly influence clinicians’ decisions to utilize the ICU. Our finding that in higher ICU utilization institutions, patients with ICU hospitalizations generally had fewer medical comorbidities and organ system failures suggests that their higher ICU utilization is not merely a reflection of caring for sicker patients. Instead institutional factors, such as numbers of ICU beds, nurse-to-patient ratios, protocols within non-ICU settings, and possibly even physician practice styles, likely contribute to greater propensity to utilize ICU care. The apparent variability in propensity for ICU utilization suggests that greater standardization of ICU admission practices might decrease costs, improve outcomes, and thus increase the value of critical care services. As such, a key question is what are the implications of greater ICU utilization for overall utilization of invasive and costly procedures, total hospital costs, and clinical outcomes?

We found that hospitals utilizing ICU care more often were more likely to perform invasive procedures overall on hospitalized patients for each condition studied. We also found that hospitalization costs were higher in institutions using ICU care more frequently, after controlling for patient factors. Yet, despite providing greater intensity of care at higher costs, higher utilization hospitals did not differ in hospital mortality rates for DKA, PE, UGIB, and CHF compared with lower ICU utilization institutions. If higher ICU utilization hospitals theoretically were to change their utilization patterns and lower their costs of hospitalization to match those of lower ICU utilization hospitals, the estimated cost savings for Washington and Maryland during the study period would be approximately $8 million for DKA, $3.5 million for PE, $6 million for UGIB, and $120 million for CHF. These estimates do not include certain cost factors, such as physician charges, which are also higher among ICU admissions. Furthermore, it is likely that there are other medical conditions for which overutilization of ICU services results in no significant difference in mortality and leads to more aggressive and costly health care delivery. Our findings are therefor likely conservative estimates of the potential to reduce costs and improve value through improvements in ICU utilization.

Our findings are consistent with studies by Gershengorn et al,6 Admon et al,5 and Safavi et al,7 who showed wide variations across hospitals in ICU use for DKA, PE, and CHF, respectively, without significant differences in mortality. Studies by these investigators and others have reported mixed results about the relationship between ICU utilization and costs.5,29-31 Differences between these studies and our findings may be due to differences in analytic methodologies and in populations studied. Our finding of consistent institutional patterns of ICU utilization across multiple conditions suggests that systematic institutional factors that lead clinicians to admit patients to the ICU more frequently should be explored as areas for interventions to improve the efficiency of ICU utilization, especially among higher-use institutions. However, our findings must also be interpreted in the context of recent findings by Valley and colleagues,30 who report that in elderly patients hospitalized with pneumonia and whose indications for critical care services appeared discretionary, 30-day mortality was lower when patients were admitted to ICUs vs hospital wards.30 Thus, for certain conditions, where the risks of clinical deterioration and mortality are higher, greater utilization of ICUs may be beneficial. For conditions like those examined in our study, where mortality rates are substantially lower and the need to escalate care is less frequent, it may be more appropriate to be selective with ICU utilization. Taken together, our findings and these previous studies suggest that prior to implementing institutional interventions to improve efficiency of ICU care, it is imperative that we identify the spectrum of clinical conditions for which higher ICU utilization improves clinical outcomes.

There are some limitations to our study related to the sources of data. First, administrative databases lack the granularity to fully adjust for medical complexity; residual confounding affecting the likelihood of ICU admission and mortality is likely to exist. Second, use of administrative codes may be relatively insensitive to identifying certain invasive procedures, such as central venous catheterizations, and could also potentially miss some hospitalizations and ICU admissions for the conditions studied.16,17 Third, the database does not allow examination of other outcomes besides hospital mortality because there is no information on posthospitalization outcomes or care. Examining these outcomes using data sources with more granular clinical information and approaches that can track patients after they leave the hospital will be important areas for further study. Finally, our findings do not explain the underlying mechanisms that drive hospitals to use ICUs more often. Studies that gather more detailed clinical information as well as information on policies and processes of care and professional attitudes to treatment will be needed to identify root causes and develop the most effective interventions to reduce unwarranted ICU use. In particular, hospital policies and institutional protocols in non-ICU settings that lead to overutilization of ICU care should be examined because they provide feasible opportunities for improvement.

Conclusions

In summary, hospitals that utilized ICU care more frequently for DKA, PE, UGIB, and CHF were more likely to perform invasive studies and have higher hospital costs with no improvement in mortality compared with lower ICU utilization institutions. These findings suggest that optimizing ICU utilization may improve quality and value of ICU care, but accomplishing that will require institutional assessments of factors that lead clinicians to admit patients to the ICU for cases in which that level of care may not be necessary.

Back to top
Article Information

Corresponding Author: Dong Chang, MD, MS, Department of Medicine, Harbor–UCLA Medical Center, 1000 W Carson St, PO Box 405, Torrance, CA 90509 (dchang@labiomed.org).

Accepted for Publication: May 23, 2016.

Published Online: August 8, 2016. doi:10.1001/jamainternmed.2016.4298

Author Contributions: Dr Chang had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Chang, Shapiro.

Acquisition, analysis, or interpretation of data: Chang, Shapiro.

Drafting of the manuscript: Chang.

Critical revision of the manuscript for important intellectual content: Chang, Shapiro.

Statistical analysis: Chang, Shapiro.

Obtained funding: Shapiro.

Administrative, technical, or material support: Chang.

Study supervision: Shapiro.

Conflict of Interest Disclosures: None reported.

Funding/Support: This study was supported by the National Institutes of Health/National Center for Advancing Translational Science UCLA Clinical and Translational Science Institute grant UL1TR000124.

Role of the Funder/Sponsor: The funding organizations had no role in the design and conduct of the study; the collection, management, analysis, and interpretation of the data; the preparation, review, or approval of the manuscript; or the decision to submit the manuscript for publication.

References
1.
Halpern  NA, Pastores  SM.  Critical care medicine in the United States 2000-2005: an analysis of bed numbers, occupancy rates, payer mix, and costs.  Crit Care Med. 2010;38(1):65-71.PubMedGoogle ScholarCrossref
2.
Wunsch  H, Angus  DC, Harrison  DA,  et al.  Variation in critical care services across North America and Western Europe.  Crit Care Med. 2008;36(10):2787-2793, e1-e9.PubMedGoogle ScholarCrossref
3.
Wunsch  H, Angus  DC, Harrison  DA, Linde-Zwirble  WT, Rowan  KM.  Comparison of medical admissions to intensive care units in the United States and United Kingdom.  Am J Respir Crit Care Med. 2011;183(12):1666-1673.PubMedGoogle ScholarCrossref
4.
Porter  ME.  What is value in health care?  N Engl J Med. 2010;363(26):2477-2481.PubMedGoogle ScholarCrossref
5.
Admon  AJ, Seymour  CW, Gershengorn  HB, Wunsch  H, Cooke  CR.  Hospital-level variation in ICU admission and critical care procedures for patients hospitalized for pulmonary embolism.  Chest. 2014;146(6):1452-1461.PubMedGoogle ScholarCrossref
6.
Gershengorn  HB, Iwashyna  TJ, Cooke  CR, Scales  DC, Kahn  JM, Wunsch  H.  Variation in use of intensive care for adults with diabetic ketoacidosis*.  Crit Care Med. 2012;40(7):2009-2015.PubMedGoogle ScholarCrossref
7.
Safavi  KC, Dharmarajan  K, Kim  N,  et al.  Variation exists in rates of admission to intensive care units for heart failure patients across hospitals in the United States.  Circulation. 2013;127(8):923-929.PubMedGoogle ScholarCrossref
8.
Healthcare Cost and Utilization Project (HCUP). HCUP Supplemental Variables for Revisit Analysis, June 2014. Agency for Healthcare Research and Quality, Rockville, MD. http://www.hcup-us.ahrq.gov/toolssoftware/revisit/revisit.jsp. Accessed June 29, 2016.
9.
United States Department of Health and Human Services; Centers for Medicare and Medicaid Services. Hospital Cost Report, 2011. https://www.cms.gov/Research-Statistics-Data-and-Systems/Downloadable-Public-Use-Files/Cost-Reports/Cost-Reports-by-Fiscal-Year.html. Accessed September 15, 2015.
10.
Cooper  GS, Chak  A, Lloyd  LE, Yurchick  PJ, Harper  DL, Rosenthal  GE.  The accuracy of diagnosis and procedural codes for patients with upper GI hemorrhage.  Gastrointest Endosc. 2000;51(4, pt 1):423-426.PubMedGoogle ScholarCrossref
11.
White  RH, Romano  PS, Zhou  H, Rodrigo  J, Bargar  W.  Incidence and time course of thromboembolic outcomes following total hip or knee arthroplasty.  Arch Intern Med. 1998;158(14):1525-1531.PubMedGoogle ScholarCrossref
12.
Wiener  RS, Schwartz  LM, Woloshin  S.  Time trends in pulmonary embolism in the United States: evidence of overdiagnosis.  Arch Intern Med. 2011;171(9):831-837.PubMedGoogle ScholarCrossref
13.
Chen  HF, Wang  CY, Lee  HY,  et al.  Short-term case fatality rate and associated factors among inpatients with diabetic ketoacidosis and hyperglycemic hyperosmolar state: a hospital-based analysis over a 15-year period.  Intern Med. 2010;49(8):729-737.PubMedGoogle ScholarCrossref
14.
Dharmarajan  K, Hsieh  AF, Lin  Z,  et al.  Diagnoses and timing of 30-day readmissions after hospitalization for heart failure, acute myocardial infarction, or pneumonia.  JAMA. 2013;309(4):355-363.PubMedGoogle ScholarCrossref
15.
Keenan  PS, Normand  SL, Lin  Z,  et al.  An administrative claims measure suitable for profiling hospital performance on the basis of 30-day all-cause readmission rates among patients with heart failure.  Circ Cardiovasc Qual Outcomes. 2008;1(1):29-37.PubMedGoogle ScholarCrossref
16.
Quan  H, Parsons  GA, Ghali  WA.  Validity of procedure codes in International Classification of Diseases, 9th revision, clinical modification administrative data.  Med Care. 2004;42(8):801-809.PubMedGoogle ScholarCrossref
17.
Walkey  AJ, Wiener  RS, Lindenauer  PK.  Utilization patterns and outcomes associated with central venous catheter in septic shock: a population-based study.  Crit Care Med. 2013;41(6):1450-1457.PubMedGoogle ScholarCrossref
18.
Deyo  RA, Cherkin  DC, Ciol  MA.  Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases.  J Clin Epidemiol. 1992;45(6):613-619.PubMedGoogle ScholarCrossref
19.
Charlson  ME, Pompei  P, Ales  KL, MacKenzie  CR.  A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.  J Chronic Dis. 1987;40(5):373-383.PubMedGoogle ScholarCrossref
20.
D’Hoore  W, Bouckaert  A, Tilquin  C.  Practical considerations on the use of the Charlson comorbidity index with administrative data bases.  J Clin Epidemiol. 1996;49(12):1429-1433.PubMedGoogle ScholarCrossref
21.
Angus  DC, Linde-Zwirble  WT, Lidicker  J, Clermont  G, Carcillo  J, Pinsky  MR.  Epidemiology of severe sepsis in the United States: analysis of incidence, outcome, and associated costs of care.  Crit Care Med. 2001;29(7):1303-1310.PubMedGoogle ScholarCrossref
22.
Healthcare Cost and Utilization Project (HCUP). HCUP Cost-to-Charge Ratio Files, July 2014. Agency for Healthcare Research and Quality, Rockville, MD. http://www.hcup-us.ahrq.gov/db/state/costtocharge.jsp. Accessed June 29, 2016.
23.
Talmor  D, Greenberg  D, Howell  MD, Lisbon  A, Novack  V, Shapiro  N.  The costs and cost-effectiveness of an integrated sepsis treatment protocol.  Crit Care Med. 2008;36(4):1168-1174.PubMedGoogle ScholarCrossref
24.
Morris  CN.  Parametric empirical Bayes inference: theory and applications.  J Am Stat Assoc. 1983;78(381):47-55.Google ScholarCrossref
25.
Normand  S-LT, Glickman  ME, Gatsonis  CA.  Statistical methods for profiling providers of medical care: issues and applications.  J Am Stat Assoc. 1997;92(439):803-814.Google ScholarCrossref
26.
Seymour  CW, Iwashyna  TJ, Ehlenbach  WJ, Wunsch  H, Cooke  CR.  Hospital-level variation in the use of intensive care.  Health Serv Res. 2012;47(5):2060-2080.PubMedGoogle ScholarCrossref
27.
Royston  P, Altman  DG, Sauerbrei  W.  Dichotomizing continuous predictors in multiple regression: a bad idea.  Stat Med. 2006;25(1):127-141.PubMedGoogle ScholarCrossref
28.
Streiner  DL.  Breaking up is hard to do: the heartbreak of dichotomizing continuous data.  Can J Psychiatry. 2002;47(3):262-266.PubMedGoogle Scholar
29.
Sjoding  MW, Prescott  HC, Wunsch  H, Iwashyna  TJ, Cooke  CR.  Hospitals with the highest intensive care utilization provide lower quality pneumonia care to the elderly.  Crit Care Med. 2015;43(6):1178-1186.PubMedGoogle ScholarCrossref
30.
Valley  TS, Sjoding  MW, Ryan  AM, Iwashyna  TJ, Cooke  CR.  Association of intensive care unit admission with mortality among older patients with pneumonia.  JAMA. 2015;314(12):1272-1279.PubMedGoogle ScholarCrossref
31.
Wunsch  H, Gershengorn  HB, Cooke  CR,  et al.  Use of intensive care services for Medicare beneficiaries undergoing major surgical procedures.  Anesthesiology. 2016;124(4):899-907.PubMedGoogle ScholarCrossref
×