Spine Publish Ahead of Print DOI: 10.1097/BRS.0000000000000892
Predicting In-Hospital Mortality in Elderly Patients with Cervical Spine Fractures: A Comparison of the Charlson and Elixhauser Comorbidity
EP TE D
Measures Authors’ full names, academic degrees and affiliations: Mariano E. Menendez, MD (1) David Ring, MD PhD (1)
Mitchel B. Harris, MD (2)
Thomas D. Cha, MD MBA (3)
Orthopaedic Hand and Upper Extremity Service, Yawkey Center, Suite 2100, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA
Department of Orthopaedic Surgery, Brigham and Women's Hospital, 75 Francis Street,
C C
Boston, MA, 02115, USA
Orthopaedic Spine Service, Yawkey Center, Suite 3A, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA
A
Corresponding author name and e-mail address: Thomas D. Cha, MD MBA, Orthopaedic Spine Service, Yawkey Center, Suite 3A, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA Tel: 617-724-8636 Fax: 617-726-7587 E-mail: [email protected]
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Ethical review committee statement: No IRB approval is mandatory for this study. The data is deidentified and free available online. The study has been performed in accordance with the ethical standards in the 1964 Declaration of Helsinki and has been carried out in accordance
Act (HIPAA). AWR: 1/19/15 ACK: 7/25/14
EP TE D
with relevant regulations of the US Health Insurance Portability and Accountability
1st Revise: 11/19/14 Accept: 2/9/15
The manuscript submitted does not contain information about medical
C C
device(s)/drug(s).
No funds were received in support of this work. Relevant financial activities outside the submitted work: Grants/grants pending, consultancy, expert testimony, royalties, stock, other (Deputy Editor Journal of Hand
A
Surgery, Deputy Editor Clinical Orthopedics and Related Research)
ABSTRACT Study design. Retrospective analysis of nationally representative data collected for the National Hospital Discharge Survey.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Objective. To compare the performance of the Charlson and Elixhauser comorbiditybased measures for predicting in-hospital mortality after cervical spine fractures. Summary of Background Data. Mortality occurring as a consequence of cervical spine fractures is very high in the elderly. The Charlson comorbidity measure has been
EP TE D
associated with an increased risk of mortality, but its predictive accuracy has yet to be compared to the more recent and increasingly used Elixhauser measure.
Methods. Using the National Hospital Discharge Survey for the years 1990 through
2007, we identified all patients aged 65 years or older hospitalized with a diagnosis of cervical spine fracture. The association of each Charlson and Elixhauser comorbidity with mortality was assessed in bivariate analysis using chi-square tests. Two main
multivariable logistic regression models were constructed, with in-hospital mortality as
the dependent variable, and one of the two comorbidity-based measures (as well as age, sex, and year of admission) as independents variables. A base model that included only age, sex, and year of admission was also evaluated. The discriminative ability of the
C C
models was quantified using the area under the ROC curve (AUC).
Results. Among an estimated 111,564 patients admitted for cervical spine fractures, 7.6% died in the hospital. Elixhauser comorbidity adjustment provided better prediction of in-hospital case-mortality (AUC=0.852, 95% CI 0.848-0.856) compared to the
A
Charlson model (AUC=0.823, 95% CI 0.819-0.828) and the base model with no comorbidities (AUC=0.785, 95% CI 0.781-0.790). In terms of relative improvement in predictive ability, the Elixhauser model performed 43% better than the Charlson model.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Conclusion. The Elixhauser comorbidity risk-adjustment method performed numerically better than the widely used Charlson measure in predicting in-hospital mortality after cervical spine fractures.
fractures; trauma.
Mini Abstract
EP TE D
Key words: Charlson; Elixhauser; comorbidity; risk adjustment; mortality; cervical spine
We assessed and compared the discriminative ability of the Charlson and Elixhauser
comorbidity measures for predicting in-hospital mortality in elderly patients admitted with cervical spine fractures. The Elixhauser comorbidity risk-adjustment method
performed numerically better than the widely used Charlson measure in predicting inpatient death after cervical spine fractures.
C C
Key Points
The overall rate of in-hospital mortality after cervical spine fracture was 7.6%.
Comorbidities that were found to be highly associated with inpatient mortality included congestive heart failure, prior myocardial infarction, peripheral vascular
A
disorders, alcohol abuse, and diabetes.
The Elixhauser comorbidity risk adjustment model outperformed the Charlson model in predicting inpatient mortality after cervical spine fractures.
In terms of relative improvement in predictive ability, the Elixhauser model performed 43% better than the Charlson model.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Introduction Inpatient mortality occurring as a consequence of cervical spine fractures is very high in the elderly; rates range from 5% to 15% 1-3. Pre-existing medical comorbidities have a significant impact on mortality after cervical spine fractures 4. Selecting
EP TE D
appropriate comorbidity risk adjustment methods is not only important for patient outcome prediction but also for provider performance assessment and reimbursement 5-7.
To statistically adjust these analyses for fair and meaningful calculations, relevant data on patients’ risk factors needs to be assessed.
Administrative data have been employed to assess the influence of baseline comorbidity status on inpatient outcomes 8-11. Originally developed in 1987 using
medical records 12, and subsequently adapted for use with administrative databases in
1992 13, the Charlson Comorbidity Index encompasses 19 medical conditions and is the
most commonly used comorbidity risk-adjustment method in patients with spine trauma 4,14,15
. The Elixhauser comorbidity measure 16 ––a newer approach containing 31
C C
conditions–– is currently the second most commonly employed risk-adjustment method in clinical research 17 and has been shown to perform better than the Charlson Index in predicting mortality in patients with cardiac, gastrointestinal, hepatobiliary, and oncologic conditions, and more recently among patients undergoing major orthopaedic 18-22
. Several prevalent comorbidities such as hypertension, obesity, malnutrition,
A
surgery
and depression are included in the Elixhauser but not the Charlson model 16. The Charlson comorbidity measure has been shown to predict mortality after
cervical spine trauma 4, but its predictive performance has yet to be compared to the more recent and increasingly used Elixhauser measure. Using nationally representative data,
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
we sought to assess and compare the discriminative ability of the Charlson and Elixhauser comorbidity measures for predicting in-hospital mortality in elderly patients admitted with cervical spine fractures.
EP TE D
Materials and Methods Data for this retrospective population-based study were extracted from the
National Hospital Discharge Survey (NHDS) for 1990 to 2007 23. Operated by the
National Center for Health Statistics, the NHDS is an annual probability sample survey of discharges from non-federal short-stay hospitals in all 50 U.S. states and the District of Columbia 24. By use of a stratified three-stage probability design that includes
population-weighting adjustments, the NHDS ensures an unbiased national sampling of
inpatient records 25. The three stages consist of (1) primary sampling units ––geographic areas such as counties or townships––, (2) hospitals within primary sampling units, and
(3) discharges within hospitals. In addition to patient demographic- and provider-related
C C
information, the NHDS collects up to 7 discharge diagnoses and 4 procedures with the
use of the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes 26,27. Recognizing its utility to address valuable clinical epidemiology questions, more than 500 studies have been published using NHDS data 5,28-30. Formal
A
approval by our Institutional Review Board was waived for the present study, as the data included no patient identifiers. All patients aged 65 years or older who had an ICD-9-CM primary diagnosis code
of cervical spine fracture, with (806.00-806.09) or without (805.00-805.08) spinal cord injury, were identified and included in the analysis. Consistent with a previous study by
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Harris et al 4 in patients with cervical spine fractures, we considered individuals aged 65 years or older because they generally present with multiple medical comorbidities, have a high risk of death after cervical spine fractures, and represent the fastest growing segment of the US population.
EP TE D
Comorbidity burden was quantified using validated Charlson (Deyo’s adaptation) and Elixhauser coding algorithms available for ICD-9-CM codes 13,16,31. We created dichotomous variables indicating the presence or absence of each Charlson and
Elixhauser comorbidity, and assessed their associations with inpatient mortality in
bivariate analysis using chi-square tests. Charlson and Elixhauser comorbidities with a p-
value of less than 0.10 in bivariate analysis and present in at least 2.0 % of the population 27,32
, were inserted into multivariable binary logistic regression analysis to evaluate the
individual contribution of each condition to inpatient mortality. Two main regression
models were constructed, each of them encompassing one of the two comorbidity-based measures, as well as age, sex, and year of admission, as independent variables. A base
C C
model that included only age, sex, and year of admission was also evaluated 7,20.
In order to determine which comorbidity measure best predicted inpatient
mortality, the regression models were compared on the basis of the area under the receiver-operating-characteristic (ROC) curve and its 95% confidence interval 33. The
A
area under the ROC curve (AUC) is a measure of discrimination that quantifies the ability of a model to accurately predict the value of an observation's response 34,35. In our study, the AUC quantified the ability of our regression models to assign a high probability of mortality to those patients who actually died. Values range from 0.50 to 1.0; the higher the AUC value, the better discrimination. In general, values less than 0.70 are considered
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
to show poor discrimination, between 0.70 and 0.80 acceptable discrimination, between 0.80 and 0.90 excellent discrimination, and above 0.90 outstanding discrimination (but they are extremely rare) 33. In addition to the absolute improvement in predictive performance, we calculated the difference between two AUCs in percent beyond the
EP TE D
predictive power of the base model including age, sex, and year of admission 7. For instance, a difference in AUC between the Charlson and Elixhauser comorbidity
measures of 0.85 and 0.90, when the baseline AUC is 0.80, corresponds to a 50% relative increase in AUC: [(0.90-0.80) – (0.85-0.80)]/(0.90-0.80) = 0.50. We calculated the
relative improvement in predictive performance of the Elixhauser method to the Charlson method. Global model performance was also compared using the Nagelkerke pseudo Rsquare, which represents a measure of the proportion of variability explained in the
response variable. To correct for multiple comparisons and the large weighted sample size, an alpha error of 0.001 was used to denote statistical significance. SPSS Version
C C
22.0 (SPSS, Chicago, IL) was employed for all statistical analyses and data modeling.
Results
An estimated number of 111,564 patients with a cervical spine fracture were
retrieved from the NHDS database between 1990 and 2007. The study population was
A
52% female, and predominantly white (94%). The majority of patients sustained an isolated cervical spine fracture (86%), had no neurological involvement (87%), and underwent non-operative treatment (86%). The overall rate of inpatient mortality was 7.6% (Table 1).
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Chronic lung disease (11%) and uncomplicated diabetes mellitus (11%) were the most prevalent comorbidities when employing the Charlson/Deyo coding algorithm (Table 2). Uncomplicated hypertension (31%) was the most frequently encountered condition among all 31 Elixhauser comorbidities, followed by cardiac arrhythmias (20%)
EP TE D
and congestive heart failure (12%) (Table 3). In multivariable logistic regression analysis, myocardial infarction (OR=8.9, 95% CI 8.0-9.9; p
Predicting In-Hospital Mortality in Elderly Patients with Cervical Spine Fractures: A Comparison of the Charlson and Elixhauser Comorbidity
EP TE D
Measures Authors’ full names, academic degrees and affiliations: Mariano E. Menendez, MD (1) David Ring, MD PhD (1)
Mitchel B. Harris, MD (2)
Thomas D. Cha, MD MBA (3)
Orthopaedic Hand and Upper Extremity Service, Yawkey Center, Suite 2100, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA
Department of Orthopaedic Surgery, Brigham and Women's Hospital, 75 Francis Street,
C C
Boston, MA, 02115, USA
Orthopaedic Spine Service, Yawkey Center, Suite 3A, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA
A
Corresponding author name and e-mail address: Thomas D. Cha, MD MBA, Orthopaedic Spine Service, Yawkey Center, Suite 3A, Massachusetts General Hospital, 55 Fruit Street, Boston, MA 02114, USA Tel: 617-724-8636 Fax: 617-726-7587 E-mail: [email protected]
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Ethical review committee statement: No IRB approval is mandatory for this study. The data is deidentified and free available online. The study has been performed in accordance with the ethical standards in the 1964 Declaration of Helsinki and has been carried out in accordance
Act (HIPAA). AWR: 1/19/15 ACK: 7/25/14
EP TE D
with relevant regulations of the US Health Insurance Portability and Accountability
1st Revise: 11/19/14 Accept: 2/9/15
The manuscript submitted does not contain information about medical
C C
device(s)/drug(s).
No funds were received in support of this work. Relevant financial activities outside the submitted work: Grants/grants pending, consultancy, expert testimony, royalties, stock, other (Deputy Editor Journal of Hand
A
Surgery, Deputy Editor Clinical Orthopedics and Related Research)
ABSTRACT Study design. Retrospective analysis of nationally representative data collected for the National Hospital Discharge Survey.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Objective. To compare the performance of the Charlson and Elixhauser comorbiditybased measures for predicting in-hospital mortality after cervical spine fractures. Summary of Background Data. Mortality occurring as a consequence of cervical spine fractures is very high in the elderly. The Charlson comorbidity measure has been
EP TE D
associated with an increased risk of mortality, but its predictive accuracy has yet to be compared to the more recent and increasingly used Elixhauser measure.
Methods. Using the National Hospital Discharge Survey for the years 1990 through
2007, we identified all patients aged 65 years or older hospitalized with a diagnosis of cervical spine fracture. The association of each Charlson and Elixhauser comorbidity with mortality was assessed in bivariate analysis using chi-square tests. Two main
multivariable logistic regression models were constructed, with in-hospital mortality as
the dependent variable, and one of the two comorbidity-based measures (as well as age, sex, and year of admission) as independents variables. A base model that included only age, sex, and year of admission was also evaluated. The discriminative ability of the
C C
models was quantified using the area under the ROC curve (AUC).
Results. Among an estimated 111,564 patients admitted for cervical spine fractures, 7.6% died in the hospital. Elixhauser comorbidity adjustment provided better prediction of in-hospital case-mortality (AUC=0.852, 95% CI 0.848-0.856) compared to the
A
Charlson model (AUC=0.823, 95% CI 0.819-0.828) and the base model with no comorbidities (AUC=0.785, 95% CI 0.781-0.790). In terms of relative improvement in predictive ability, the Elixhauser model performed 43% better than the Charlson model.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Conclusion. The Elixhauser comorbidity risk-adjustment method performed numerically better than the widely used Charlson measure in predicting in-hospital mortality after cervical spine fractures.
fractures; trauma.
Mini Abstract
EP TE D
Key words: Charlson; Elixhauser; comorbidity; risk adjustment; mortality; cervical spine
We assessed and compared the discriminative ability of the Charlson and Elixhauser
comorbidity measures for predicting in-hospital mortality in elderly patients admitted with cervical spine fractures. The Elixhauser comorbidity risk-adjustment method
performed numerically better than the widely used Charlson measure in predicting inpatient death after cervical spine fractures.
C C
Key Points
The overall rate of in-hospital mortality after cervical spine fracture was 7.6%.
Comorbidities that were found to be highly associated with inpatient mortality included congestive heart failure, prior myocardial infarction, peripheral vascular
A
disorders, alcohol abuse, and diabetes.
The Elixhauser comorbidity risk adjustment model outperformed the Charlson model in predicting inpatient mortality after cervical spine fractures.
In terms of relative improvement in predictive ability, the Elixhauser model performed 43% better than the Charlson model.
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Introduction Inpatient mortality occurring as a consequence of cervical spine fractures is very high in the elderly; rates range from 5% to 15% 1-3. Pre-existing medical comorbidities have a significant impact on mortality after cervical spine fractures 4. Selecting
EP TE D
appropriate comorbidity risk adjustment methods is not only important for patient outcome prediction but also for provider performance assessment and reimbursement 5-7.
To statistically adjust these analyses for fair and meaningful calculations, relevant data on patients’ risk factors needs to be assessed.
Administrative data have been employed to assess the influence of baseline comorbidity status on inpatient outcomes 8-11. Originally developed in 1987 using
medical records 12, and subsequently adapted for use with administrative databases in
1992 13, the Charlson Comorbidity Index encompasses 19 medical conditions and is the
most commonly used comorbidity risk-adjustment method in patients with spine trauma 4,14,15
. The Elixhauser comorbidity measure 16 ––a newer approach containing 31
C C
conditions–– is currently the second most commonly employed risk-adjustment method in clinical research 17 and has been shown to perform better than the Charlson Index in predicting mortality in patients with cardiac, gastrointestinal, hepatobiliary, and oncologic conditions, and more recently among patients undergoing major orthopaedic 18-22
. Several prevalent comorbidities such as hypertension, obesity, malnutrition,
A
surgery
and depression are included in the Elixhauser but not the Charlson model 16. The Charlson comorbidity measure has been shown to predict mortality after
cervical spine trauma 4, but its predictive performance has yet to be compared to the more recent and increasingly used Elixhauser measure. Using nationally representative data,
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
we sought to assess and compare the discriminative ability of the Charlson and Elixhauser comorbidity measures for predicting in-hospital mortality in elderly patients admitted with cervical spine fractures.
EP TE D
Materials and Methods Data for this retrospective population-based study were extracted from the
National Hospital Discharge Survey (NHDS) for 1990 to 2007 23. Operated by the
National Center for Health Statistics, the NHDS is an annual probability sample survey of discharges from non-federal short-stay hospitals in all 50 U.S. states and the District of Columbia 24. By use of a stratified three-stage probability design that includes
population-weighting adjustments, the NHDS ensures an unbiased national sampling of
inpatient records 25. The three stages consist of (1) primary sampling units ––geographic areas such as counties or townships––, (2) hospitals within primary sampling units, and
(3) discharges within hospitals. In addition to patient demographic- and provider-related
C C
information, the NHDS collects up to 7 discharge diagnoses and 4 procedures with the
use of the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes 26,27. Recognizing its utility to address valuable clinical epidemiology questions, more than 500 studies have been published using NHDS data 5,28-30. Formal
A
approval by our Institutional Review Board was waived for the present study, as the data included no patient identifiers. All patients aged 65 years or older who had an ICD-9-CM primary diagnosis code
of cervical spine fracture, with (806.00-806.09) or without (805.00-805.08) spinal cord injury, were identified and included in the analysis. Consistent with a previous study by
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Harris et al 4 in patients with cervical spine fractures, we considered individuals aged 65 years or older because they generally present with multiple medical comorbidities, have a high risk of death after cervical spine fractures, and represent the fastest growing segment of the US population.
EP TE D
Comorbidity burden was quantified using validated Charlson (Deyo’s adaptation) and Elixhauser coding algorithms available for ICD-9-CM codes 13,16,31. We created dichotomous variables indicating the presence or absence of each Charlson and
Elixhauser comorbidity, and assessed their associations with inpatient mortality in
bivariate analysis using chi-square tests. Charlson and Elixhauser comorbidities with a p-
value of less than 0.10 in bivariate analysis and present in at least 2.0 % of the population 27,32
, were inserted into multivariable binary logistic regression analysis to evaluate the
individual contribution of each condition to inpatient mortality. Two main regression
models were constructed, each of them encompassing one of the two comorbidity-based measures, as well as age, sex, and year of admission, as independent variables. A base
C C
model that included only age, sex, and year of admission was also evaluated 7,20.
In order to determine which comorbidity measure best predicted inpatient
mortality, the regression models were compared on the basis of the area under the receiver-operating-characteristic (ROC) curve and its 95% confidence interval 33. The
A
area under the ROC curve (AUC) is a measure of discrimination that quantifies the ability of a model to accurately predict the value of an observation's response 34,35. In our study, the AUC quantified the ability of our regression models to assign a high probability of mortality to those patients who actually died. Values range from 0.50 to 1.0; the higher the AUC value, the better discrimination. In general, values less than 0.70 are considered
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
to show poor discrimination, between 0.70 and 0.80 acceptable discrimination, between 0.80 and 0.90 excellent discrimination, and above 0.90 outstanding discrimination (but they are extremely rare) 33. In addition to the absolute improvement in predictive performance, we calculated the difference between two AUCs in percent beyond the
EP TE D
predictive power of the base model including age, sex, and year of admission 7. For instance, a difference in AUC between the Charlson and Elixhauser comorbidity
measures of 0.85 and 0.90, when the baseline AUC is 0.80, corresponds to a 50% relative increase in AUC: [(0.90-0.80) – (0.85-0.80)]/(0.90-0.80) = 0.50. We calculated the
relative improvement in predictive performance of the Elixhauser method to the Charlson method. Global model performance was also compared using the Nagelkerke pseudo Rsquare, which represents a measure of the proportion of variability explained in the
response variable. To correct for multiple comparisons and the large weighted sample size, an alpha error of 0.001 was used to denote statistical significance. SPSS Version
C C
22.0 (SPSS, Chicago, IL) was employed for all statistical analyses and data modeling.
Results
An estimated number of 111,564 patients with a cervical spine fracture were
retrieved from the NHDS database between 1990 and 2007. The study population was
A
52% female, and predominantly white (94%). The majority of patients sustained an isolated cervical spine fracture (86%), had no neurological involvement (87%), and underwent non-operative treatment (86%). The overall rate of inpatient mortality was 7.6% (Table 1).
Copyright © 2015 Wolters Kluwer Health, Inc. Unauthorized reproduction of this article is prohibited.
Chronic lung disease (11%) and uncomplicated diabetes mellitus (11%) were the most prevalent comorbidities when employing the Charlson/Deyo coding algorithm (Table 2). Uncomplicated hypertension (31%) was the most frequently encountered condition among all 31 Elixhauser comorbidities, followed by cardiac arrhythmias (20%)
EP TE D
and congestive heart failure (12%) (Table 3). In multivariable logistic regression analysis, myocardial infarction (OR=8.9, 95% CI 8.0-9.9; p
