Supplemental Digital Content:
Validation of an administrative definition of intensive care unit admission using revenue center codes
Gary E. Weissman, MD, Rebecca A. Hubbard, PhD, Rachel Kohn, MD, George L. Anesi, MD, MBE, Scott Manaker, MD, PhD, Meeta Prasad Kerlin, MD, MSCE, Scott D. Halpern, MD, PhD
September 4, 2016
Classification error analysis
In order to understand patterns of misclassification using revenue center codes (RCCs), we built a logistic regression model to make inferences about clinical, demographic, and
administrative factors associated with erroneous identification. This approach is not intended to improve the classification of the RCC approach, but to provide insight into the patterns of its errors. To accomplish this, we built two separate models, one each to study false positive and false negative errors.
Methods
We used multivariable logistic regression to examine factors associated with
misclassification of ICU admission through use of RCCs. We built two models to separately examine the errors among those admissions with and without ICU-related RCCs. The outcome for each model was a binary indicator of a misclassification error (false positive or false
negative). We chose covariates that were likely to be available to researchers in an administrative
dataset. We included age, comorbidities, and hospital length of stay as linear, continuous
covariates, and the following as categorical variables: gender, race, discharge status (alive/dead), admission type, severe sepsis, mechanical ventilation, vasopressor use, and diagnosis-related group (DRG) codes that are commonly associated with ICU admission. We report multivariable adjusted odds ratios (ORs) and their associated 95% CIs.
Definitions
The comorbidity burden was derived from all International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM) reported present-on-admission diagnoses and calculated using the van Walraven modification (1) to the Elixhauser comorbidity measure (2).
The calculation of the latter was implemented with the medicalrisk software package (3) using the Quan approach (4) to adjust for severity of related diagnoses. We chose this measure because it has reasonable discrimination for predicting in-hospital mortality, and has superior
performance to the Charlson index (1). Admission type was categorized as either medical or surgical based on the MS-DRG code for that admission (5). Presence of severe sepsis was identified with the "Angus" definition (6) using a translation into R of a previously published classification algorithm in SAS (7). Mechanical ventilation was identified using ICD-9-CM procedure codes for mechanical ventilation (96.7x) or endotracheal intubation (96.0x) (8).
Vasopressor use was identified using an ICD-9-CM procedure code (00.17) (9). "High-risk" MS- DRG codes (64, 65, 189, 193, 208, 247, 280, 287, 291, 292, 309, 310, 313, 378, 638, 682, 871, 918) for ICU admission have been previously identified using RCCs in a large, nationally representative dataset (10). In that dataset, the MS-DRG codes associated with the highest rates of ICU-related RCCs were "208 Respiratory system diagnosis with ventilator support less than 96 hours" and "280 Acute myocardial infarction, discharged alive with MCC" (10).
Results
Among the 27,169 hospital admissions with an ICU-related RCC, we identified 3,978 (14.6%) false positive classifications. The odds of a false positive error were lower among patients with billing codes for severe sepsis (OR 0.50, 95% CI 0.42 - 0.60), mechanical ventilation (OR 0.23, 95% CI 0.18 - 0.28), high-risk diagnosis-related group codes (OR 0.09, 95% CI 0.07 - 0.11, and among male (OR 0.53, 95% CI 0.49 - 0.58) and surgical (OR 0.58 95%
CI 0.53 - 0.63) patients.
Among the 100,511 hospital admissions without an ICU-related RCC, we identified 1,321 (1.3%) false negative classifications. Odds of a false negative error were higher among male patients (OR 1.42, 95% CI 1.27 - 1.60), and those with severe sepsis (OR 1.16, 95% CI 1.16 - 1.65) or mechanical ventilation (OR 9.51, 95% CI 7.18 - 12.47).
Discussion
The odds of a true positive result were higher in patients admitted with a primary surgical diagnosis, or had billing codes suggesting increased severity of illness as defined by severe sepsis, mechanical ventilation, or a high-risk DRG. Most of these measures of severity of illness also increased the odds of a false negative result among patients without ICU-related RCCs.
Black patients were more likely to have erroneous RCCs whether or not they were actually in an ICU. We cannot determine whether race was independently associated with classification errors, or was associated with other unobserved variables such as payer status or treatment intensity that were associated with differential utilization and billing patterns.
References
1. van Walraven C, Austin PC, Jennings A, et al: A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care 2009;626–
633
2. Elixhauser A, Steiner C, Harris DR, et al: Comorbidity measures for use with administrative data. Med Care 1998;36:8–27
3. McCormick P, Joseph T: Medicalrisk: Medical risk and comorbidity tools for ICD-9-CM data.
Available at: https://CRAN.R-project.org/package=medicalrisk. Accessed July 1, 2016 4. Quan H, Sundararajan V, Halfon P, et al: Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005;1130–1139
5. Medicare & Medicaid Services C: Appendix A: list of MS-DRGs version 28.0. In: Draft ICD- 10-CM/PCS MS-DRGv28 definitions manual. U.S. Department of Health & Human Services 2016. Available at: https://www.cms.gov/icd10manual/fullcode_cms/P0029.html. Accessed July 1, 2016
6. Angus DC, Linde-Zwirble WT, Lidicker J, et al: Epidemiology of severe sepsis in the United States: Analysis of incidence, outcome, and associated costs of care. Crit Care Med
2001;29:1303–1310
7. Iwashyna TJ, Odden A, Rohde J, et al: Identifying patients with severe sepsis using
administrative claims: Patient-level validation of the Angus implementation of the International Consensus Conference definition of severe sepsis. Med Care 2014;52:e39–43
8. Kerlin MP, Weissman GE, Wonneberger KA, et al: Validation of administrative definitions of invasive mechanical ventilation across 30 ICUs. Am J Resp Crit Care Med 2016; In press
9. Fawzy A, Bradford M, Lindenauer PK, et al: Identifying vasopressor and inotrope use for health services research. Ann Am Thorac Soc 2016;13:414–418
10. Barrett M, Smith M, Elixhauser A, et al: Utilization of intensive care services, 2011. Agency for Healthcare Research and Quality Statistical Brief 2014;185:1-14
Supplemental Table 1: UB-92 revenue center codes relevant to critical illness
Revenue Center Code Description Included in proposed definition
173 Nursery - intermediate care
174 Nursery - intensive care
200 Intensive care - general classification X
201 Intensive care - surgical X
202 Intensive care - medical X
203 Intensive care - pediatric X
204 Intensive care - psychiatric X
206 Intensive care - intermediate care
207 Intensive care - burn care X
208 Intensive care - trauma X
209 Intensive care - other intensive care X
210 Coronary care - general classification X
211 Coronary care - myocardial infraction X
212 Coronary care - pulmonary care X
213 Coronary care - heart transplant X
214 Coronary care - intermediate care
219 Coronary care - other coronary care X
233 Incremental nursing charge rate - ICU 234 Incremental nursing charge rate - CCU Abbreviations: ICU = intensive care unit, CCU = coronary care unit
Note: Sometimes revenue center codes are displayed with a leading zero, for example “0200”, which is omitted here for clarity.
Supplemental Table 2: Multivariable logistic regression model of factors associated with a false positive error among those hospital admissions with intensive care unit-related revenue center codes, and with a false negative error among those hospital admissions without intensive care unit-related revenue center codes
False positive error False negative error
Variable OR 95 % CI p-value OR 95% CI p-value
Service
Medical 1 1
Surgical 0.58 0.53 - 0.63 <0.001 0.97 0.84 - 1.11 0.612
Unknown 0.69 0.52 - 0.90 0.007 1.23 0.89 - 1.66 0.186
Modified Elixhauser comorbidity score
0.96 0.96 - 0.97 <0.001 1.01 1.00 - 1.02 0.004 Hospital length of stay (days) 0.92 0.91 - 0.93 <0.001 1.02 1.02 - 1.03 <0.001
Age (years) 0.97 0.97 - 0.97 <0.001 1.01 1.01 - 1.01 <0.001
Gender
Female 1 1
Male 0.53 0.49 - 0.58 <0.001 1.42 1.27 - 1.60 <0.001
Race
White 1 1
Black 1.12 1.03 - 1.22 0.009 1.25 1.11 - 1.41 <0.001
Other 1.13 0.95 - 1.35 0.162 1.10 0.83 - 1.42 0.492
Unknown 0.83 0.68 - 1.00 0.060 1.74 1.30 - 2.28 <0.001
Discharge status
Alive 1 1
Dead 1.59 1.20 - 2.09 0.001 9.47 7.09 -
12.55
<0.001
Unknown 14.55 4.76 - 44.57 <0.001 1.32 0.21 - 4.50 0.706
Severe sepsis 0.50 0.42 - 0.60 <0.001 1.38 1.16 - 1.65 <0.001
Intubation or mechanical ventilation 0.23 0.18 - 0.28 <0.001 9.51 7.18 - 12.47
<0.001
Vasopressors 0.75 0.42 - 1.24 0.295 5.51 1.90 -
16.59
0.002
"High-risk" DRG 0.09 0.07 - 0.11 <0.001 2.40 2.07 - 2.78 <0.001 Abbreviations: OR = odds ratio, CI = confidence interval, DRG = diagnosis related group code
Supplemental Table 3: 2x2 table comparing classification by the CART algorithm to the gold standard electronic patient location data
True ICU admission
+ -
Predicted ICU Admission
+ 4,526 539
- 376 20,094
The table describes the performance of the CART algorithm for n=25,535 hospital admissions in the test set. CART
= classification and regression tree, ICU = intensive care unit.
Supplemental Figure Legend 1: Patterns of clinical and demographic associations with true admission to the intensive care unit (ICU)
A) Density plot of modified Elixhauser comorbidity score by ICU admission status, B) Density plot of patient age in years at admission by ICU admission status, C) Density plot of hospital length of stay in days by ICU admission status, D) Stacked bar plot of hospital admissions by gender indicating ICU admission status, E) Stacked bar plot of hospital admissions by type, defined by diagnosis related group, indicating ICU admission status, F) Stacked bar plot of hospital admissions by discharge status indicating ICU admission status, G) Stacked bar plot of presence of four severity measures and breakdown of ICU admission status for each. Note that the cumulative densities within each panel sum to 1. The regions of the stacked bar plots bounded with solid red or dashed blue lines represent those patients with and without an ICU stay, respectively.