424

C OPYRIGHT Ó 2021 BY T HE J OURNAL OF B ONE AND J OINT S URGERY, I NCORPORATED

The CARDE-B Scoring System Predicts 30-Day

Mortality After Revision Total Joint Arthroplasty
Micheal Raad, MD, Raj Amin, MD, Varun Puvanesarajah, MD, Farah Musharbash, MD, Sandesh Rao, MD,
Matthew J. Best, MD, and Derek F. Amanatullah, MD, PhD

Investigation performed at The Johns Hopkins University School of Medicine, Baltimore, Maryland
Downloaded from http://journals.lww.com/jbjsjournal by BhDMf5ePHKav1zEoum1tQfN4a+kJLhEZgbsIHo4XMi0hCywCX1AWnYQp/IlQrHD3i3D0OdRyi7TvSFl4Cf3VC4/OAVpDDa8KKGKV0Ymy+78= on 03/22/2021

Background: There exists a substantial risk of having a perioperative complication after revision total joint arthroplasty
(TJA). The complex shared decision-making between surgeon and patient would benefit from a high-fidelity tool to identify
patients at risk for mortality after revision TJA. Therefore, we developed the CARDE-B score. CARDE-B is an acronym for
congestive heart failure, albumin or malnutrition (<3.5 mg/dL), renal failure on dialysis, dependence for daily living, elderly
(>65 years of age), and body mass index <25 kg/m2. We developed and validated the CARDE-B score to determine the risk
of death within 30 days of a revision TJA.
Methods: A total of 13,118 revision TJAs (40% hip and 60% knee) from the National Surgical Quality Improvement
Program (NSQIP) database were analyzed. A simple 1-point scoring system, CARDE-B, was created for predicting 30-day
mortality after revision TJA, based on a logistic regression model. The CARDE-B scoring system assigns 1 point to each
criterion in the acronym: congestive heart failure, albumin (<3.5 mg/dL), renal failure on dialysis, dependence for daily
living, elderly (>65 years of age), and body mass index of <25 kg/m2. The CARDE-B scoring system was compared with 2
commonly utilized scores: American Society of Anesthesiologists (ASA) physical status classification and the 5-factor
modified frailty index (mFI-5). The area under the curve (AUC) was used to assess the accuracy of each model. The Hosmer-
Lemeshow test was used to assess goodness of fit. Finally, the Nationwide Inpatient Sample (NIS) was used for external
validation of the CARDE-B score in 19,153 patients who underwent revision TJA in 2017.
Results: Eighty-eight patients (0.7%) did not survive 30 days after revision TJA. The AUC for the logistic regression model
was 0.88 in both the derivation and internal validation samples using NSQIP. The predicted probability of 30-day mortality
after revision TJA increased stepwise from <0.01% for a CARDE-B score of 0 points to 39% for a CARDE-B score of 5 points.
The AUC for the CARDE-B score predicting 30-day mortality after revision TJA was 0.85. This was more accurate (p <
0.001) than the ASA physical status classification (AUC, 0.77) and the mFI-5 (AUC, 0.67). The AUC for the CARDE-B score
in the NIS external validation set was 0.75. The Hosmer-Lemeshow p value for goodness of fit was 0.34, indicating
goodness of fit in the external validation sample.
Conclusions: The CARDE-B score is a simple system that predicts the risk of death within 30 days of a revision TJA,
offering surgeons and patients a valuable and validated risk-stratification tool.
Level of Evidence: Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.

R
evision total joint arthroplasty (TJA) is technically de-
manding and carries a higher risk of morbidity and
mortality when compared with primary TJA1. To this
end, a number of preoperative surgical risk scores have been
created to assess surgical risk, such as the 5-factor modified
frailty index (mFI-5)2 and the American Society of Anesthesi-
ologists (ASA) physical status classification3. Although the mFI-5
is associated with mortality after revision TJA4, its accuracy as a
risk-stratification tool has not been specifically investigated, to our
knowledge. The ASA classification was developed to assess anes-
thetic risk but has poor interrater reliability among different
specialists5. There are complex mortality risk calculators such as
the American College of Surgeons (ACS) National Surgical
Quality Improvement Program (NSQIP) Surgical Risk Calcu-
lator (SRC). The ACS-NSQIP-SRC involves >22 variables, and
some of the input variables are multilayered. An example of a

Disclosure: The authors indicated that no external funding was received for any aspect of this work. On the Disclosure of Potential Conflicts of Interest
forms, which are provided with the online version of the article, one or more of the authors checked “yes” to indicate that the author had a relevant financial
relationship in the biomedical arena outside the submitted work (http://links.lww.com/JBJS/G314).

J Bone Joint Surg Am. 2021;103:424-31 d http://dx.doi.org/10.2106/JBJS.20.00969

 425
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y

The remaining 13,118 revision TJAs (Table I) were analyzed

Fig. 1
Flowchart demonstrating the derivation of the final sample with com-
plete information for analysis after the listwise deletion of patients with
concomitant and concurrent procedures as well as incomplete
information.
multilayered input for the ACS-NSQIP-SRC is ASA physical
status classification. This subjects the ACS-NSQIP-SRC to the
same interrater reliability issues known with the ASA classifi-
cation, affecting its generalizability5,6.
Hence, we developed a simple, predictive, 1-point risk
assessment tool specific to revision TJA: the CARDE-B Scoring
System. CARDE-B is an acronym for congestive heart failure,
albumin (<3.5 mg/dL), renal failure on dialysis, dependence for
daily living, elderly (>65 years of age), and body mass index
(BMI) of <25 kg/m2. The CARDE-B scoring system was de-
rived after logistic regression using the ACS-NSQIP database to
predict 30-day mortality after revision TJA and validated in the
Nationwide Inpatient Sample (NIS). We hypothesized that
assigning 1 point to each diagnosis in the acronym of CARDE-B
would more accurately predict the 30-day mortality of a revision
TJA than the ASA physical status classification or mFI-5.
Materials and Methods
for several risk factors shown in previous studies to affect mortality
after arthroplasty as well as risk factors that may be considered
clinically relevant9,10. The variable “yrdeath” was used to identify
30-day mortality. We used the variable “fnstatu” to identify patients
who were dependent in performing their activities of daily living at
baseline. We defined pulmonary disease as either having an es-
tablished diagnosis of chronic obstructive pulmonary disease or
dyspnea on exertion. An albumin threshold of <3.5 mg/dL was
used to define hypoalbuminemia or malnutrition9. Anemia was
defined as a preoperative hematocrit of <40%. Infection was
identified using the International Classification of Diseases, Ninth
Revision (ICD-9), codes 996.66 to 996.67 and the ICD, Tenth
Revision (ICD-10), codes T84.5 to T84.7.
The analyzable revision TJAs were then randomly and
evenly divided using a computer program (Stata Statistical
Software, Release 15; StataCorp) into derivation and validation
cohorts (Fig. 1). Mortality rates for various risk factors were
then compared. The Fisher exact test was used for comparing
categorical variables. The Student t test was used to compare
continuous variables. Multicollinearity among risk factors was
estimated using the variance inflation factor (VIF) with a
threshold of <311. Logistic regression with robust estimates of
the variance was then utilized to derive a risk estimating equation
(REE). The REE was used to calculate the probability of death
using the entire cohort. We used receiver operating characteristic
curve analysis to calculate the area under the curve (AUC) and to
TABLE I Demographic Characteristics and Risk Profile from the
ACS-NSQIP of 13,118 Patients Who Underwent Revi-
sion Hip or Knee Arthroplasty
Characteristic No. of Patients*
Age >65 years 7,302 (55.7%)
Female sex 7,518 (57.3%)
BMI <25 kg/m2 2,080 (15.9%)

T he ACS-NSQIP database from 2015 to 2018 was utilized

for this study. The database contains prospectively col-
lected preoperative and 30-day postoperative outcomes for
BMI 25 to 34.9 kg/m2
BMI ‡35 kg/m2
6,971 (53.1%)
4,067 (31.0%)
Revision total hip arthroplasty 5,260 (40.1%)
randomly selected patients. Data are entered by trained surgical
Periprosthetic joint infection 1,668 (12.7%)
clinical reviewers and have an interrater reliability of >95%7.
The ACS-NSQIP database was queried for patients undergoing Hypoalbuminemia 2,078 (15.8%)
revision total hip arthroplasty using the Current Procedural Congestive heart failure 134 (1.0%)
Terminology (CPT) codes 27134, 27137, and 27138 and for Chronic obstructive pulmonary 1,410 (10.8%)
those undergoing revision total knee arthroplasty using CPT disease or dyspnea
codes 27486, 27487, and 27488. Patients who underwent addi- Renal failure on dialysis 97 (0.7%)
tional procedures or had incomplete data were excluded from our Anemia 6,940 (52.9%)
analysis using listwise deletion to create a complete-case Dependent for activities of daily living 676 (5.2%)
data set (Fig. 1). A comparison of multiple imputation and Hypertension 8,698 (66.3%)
listwise deletion (also known as complete-case analysis) to
Insulin-dependent diabetes mellitus 617 (4.7%)
deal with incomplete data within the ACS-NSQIP dem-
Tobacco use 1,704 (13.0%)
onstrated that listwise deletion was more appropriate 8 . A
comparison of characteristics between patients with com-
*The values are given as the number of patients, with the
plete and incomplete data may be seen in the Appendix, percentage in parentheses.
Supplemental Table 1.
 426
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y

illustrate the model’s accuracy in the derivation and validation Results

samples12. The Hosmer-Lemeshow test was used to assess good-
ness of fit13. Coefficients from the whole-sample REE were used to
assign scores for each of the identified variables and scoring sys-
tems were compared for accuracy and simplicity. The model-
predicted 30-day mortality as a function of the chosen score was
then displayed in the form of a box plot.
The CARDE-B score was derived as a practical risk
assessment tool specific to revision TJA after logistic regression
in the ACS-NSQIP database to predict 30-day mortality after
revision TJA. CARDE-B is an acronym of 6 risk factors where
each diagnosis is assigned 1 point within the scoring system:
congestive heart failure, albumin (<3.5 mg/dL), renal failure on
dialysis, dependence for daily living, elderly (>65 years of age),
and BMI of <25 kg/m2. The CARDE-B score was compared
with 2 commonly used risk-stratification scores, ASA physical
status classification3,14 and mFI-52, each frequently utilized in
the surgical setting and available for analysis in the ACS-NSQIP
database. The AUCs for various parameters were compared by
using a previously described theory on U-statistics to generate
an estimated covariance matrix15.
The NIS was used for external validation of the CARDE-
B score (Appendix, Supplemental Table 2). The NIS constitutes
hospital discharge-level data from states participating in the
Healthcare Cost and Utilization Project (HCUP) and is based on a
randomized stratified sampling designed to capture 20% of in-
patient discharges16. The NIS 2017 uses the ICD-10 Procedure
Coding System (ICD-10-PCS) and diagnosis codes (ICD-10
Clinical Modification [ICD-10-CM]). The NIS does not include
laboratory values, so the diagnosis code for malnutrition was used
to approximate hypoalbuminemia. Only patients with a hospital
length of stay of <30 days were included to approximate the
derivation and validation cohorts from the ACS-NSQIP database.
Categorical variables are presented as the number and
percentage, and continuous variables are presented as the mean
and the standard deviation. Significance was set as p < 0.05. All
statistical computing was performed using Stata Release 15.
Risk Factors That Contribute to an Early Death After
Revision TJA
E ighty-eight (0.7%) of 13,118 patients did not survive 30 days
after revision TJA. Using the derivation sample (Fig. 1), the
patients with and without each of the risk factors were compared
with respect to their 30-day mortality after revision TJA. Mortality
was significantly higher in patients who were >65 years of age
(1.1%) compared with those who were £65 years of age (0.1%) (p
< 0.001), had hypoalbuminemia (2.6%) compared with those
with a normal serum albumin (0.3%) (p < 0.001), had congestive
heart failure (7.6%) compared with those without heart failure
(0.6%) (p < 0.001), had renal failure on dialysis (4.1%) compared
with those who did not (0.7%) (p = 0.042), had anemia (1.1%)
compared with those with a normal serum hematocrit (0.2%)
(p < 0.001), underwent a revision total hip arthroplasty (1.0%)
compared with those who did not (0.5%) (p < 0.001), or were
dependent for activities of daily living (4.9%) compared with those
who were independent (0.5%) (p < 0.001) (Table II). The rate of
mortality within 30 days of a revision TJA was higher with a BMI of
<25 kg/m2 (1.8%) compared with a BMI from 25 to 34.9 kg/m2
(0.5%; p < 0.001) and a BMI of ‡35 kg/m2 (0.4%; p < 0.001). There
was no difference (p = 0.343) in the operative time of the revision
TJAs with early mortality (mean and standard deviation, 146 ± 49
minutes) and those without early mortality (137 ± 66 minutes).
Risk Model Development and Validation
All univariate risk factors with p values of <0.05 were included
in the logistic regression model, in addition to infection,
given its clinical importance in the setting of revision TJA17.
The mean VIF was 1.4, indicating low multicollinearity. The
AUC of the derivation sample was 0.88 (Hosmer-Lemeshow;
p = 0.52) (Fig. 2-A), and the AUC of the validation sample was
0.88 (Hosmer-Lemeshow; p = 0.09) (Fig. 2-B). Six independent
risk factors predicted death within 30 days of a revision TJA:
(1) congestive heart failure, (2) a serum albumin <3.5 mg/dL,
(3) renal failure on dialysis, (4) dependence for activities of daily

TABLE II Univariate Risk Factors for Death within 30 Days of a Revision TJA in the Derivation Sample

Risk Factor Deaths with Risk Present* Deaths with Risk Absent* P Value

Infection 10 (1.2%) of 865 34 (0.6%) of 5,694 0.072

Hypoalbuminemia 28 (2.6%) of 1,069 16 (0.3%) of 5,490 <0.001†
Congestive heart failure 5 (7.6%) of 66 39 (0.6%) of 6,493 <0.001†
Chronic obstructive pulmonary disease or dyspnea 5 (0.7%) of 690 39 (0.7%) of 5,869 0.805
Renal failure on dialysis 2 (4.1%) of 49 42 (0.7%) of 6,510 0.042†
Anemia 37 (1.1%) of 3,493 7 (0.2%) of 3,066 <0.001†
Dependent for activities of daily living 16 (4.9%) of 324 28 (0.5%) of 6,235 <0.001†
Hypertension 35 (0.8%) of 4,311 9 (0.4%) of 2,248 0.056
Insulin-dependent diabetes mellitus 4 (1.4%) of 292 40 (0.6%) of 6,267 0.131
Tobacco use 5 (0.6%) of 837 39 (0.7%) of 5,722 0.999

*The values are given as the number of deaths for each risk factor, with the percentage in parentheses. †Significant at p < 0.05.
 427
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y

Fig. 2-A

Fig. 2-B

Fig. 2 The receiver operating characteristic curve in the derivation sample (AUC, 0.88) (Fig. 2-A) and the validation sample (AUC, 0.88) (Fig. 2-B) for the risk
of death within 30 days of a revision TJA.

living, (5) elderly (>65 years of age), and (6) BMI <25 kg/m2
(Table III).
Creation of the CARDE-B Scoring System
CARDE-B is an acronym for the 6 independent risk factors that
predicted death within 30 days of a revision TJA. The CARDE-
B scoring system assigns 1 point for each independent risk
factor, with the lowest score being 0 points and highest score in
our sample being 5 points. Further analysis revealed that using
weights based on the model coefficients added a layer of
complexity to the scoring system with no significant change
in accuracy (p = 0.263) (Fig. 3). The rate of observed 30-day
mortality increased in a stepwise fashion per point added to
the CARDE-B score: 4,362 revision TJAs had a CARDE-B
score of 0 points and the 30-day mortality was <0.01%, 5,921
revision TJAs had a CARDE-B score of 1 point and the 30-
day mortality was 0.3%, 2,192 revision TJAs had a CARDE-B
score of 2 points and the 30-day mortality was 1.3%, 520
revision TJAs had a CARDE-B score of 3 points and the 30-
day mortality was 4.4%, 113 revision TJAs had a CARDE-B
 428
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y

TABLE III Multivariate Regression Identifying the Independent Risk Factors for Death within 30 Days of a Revision TJA in All Patients

Risk Factor OR* P Value

Age >65 years 6.83 (3.00 to 15.55) <0.001†

BMI <25 kg/m2 2.36 (1.46 to 3.81) <0.001†
BMI ‡35 kg/m2 0.99 (0.52 to 1.85) 0.964
Revision total hip arthroplasty 1.29 (0.81 to 2.05) 0.290
Infection 1.25 (0.72 to 2.19) 0.430
Hypoalbuminemia 3.99 (2.38 to 6.70) <0.001†
Congestive heart failure 4.88 (2.28 to 10.44) <0.001†
Chronic obstructive pulmonary disease or dyspnea 0.94 (0.52 to 1.72) 0.848
Renal failure on dialysis 4.97 (1.61 to 15.37) 0.005†
Anemia 1.78 (0.93 to 3.41) 0.081
Dependent for activities of daily living 2.67 (1.56 to 4.57) <0.001†

*The values are given as the OR, with the 95% CI in parentheses. †Significant at p < 0.05.

score of 4 points and the 30-day mortality was 16%, and 10
revision TJAs had a CARDE-B score of 5 points and the 30-day
mortality was 10%. The predicted risk of 30-day mortality after
revision TJA was plotted as a function of each unit of the CARDE-
B score (Fig. 4). The AUC using the CARDE-B scoring system for
30-day mortality after revision TJA was 0.85 (Fig. 5). A CARDE-B
score of ‡2 points has a sensitivity of 80%, a specificity of 79%, a
positive likelihood ratio of 3.7, and a negative likelihood ratio of
0.26. The ability of the CARDE-B scoring system to predict 30-day
mortality after revision TJA was statistically superior (p < 0.001) to
the ASA physical status classification (AUC, 0.77) and the mFI-5
(AUC, 0.67) (Fig. 5).
External Validation
There were 19,153 patients identified in the NIS database who
underwent a revision TJA in 2017 as defined by the ICD-10-
PCS corresponding to component removal as a part of the
surgical procedure. The AUC for the CARDE-B score in pre-
dicting mortality was 0.75. The Hosmer-Lemeshow p value for
goodness of fit was 0.34, indicating goodness of fit in the
external validation sample.
Discussion
T he CARDE-B scoring system accurately predicted the risk
of 30-day mortality after revision TJA. The ASA physical

Fig. 3
Receiver operating characteristic curves displaying the AUC of the simple CARDE-B scoring system, shown by the solid line (AUC, 0.85), compared with the
coefficient-weighted CARDE-B score, shown by the dashed line (AUC, 0.86).
 429
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y

Fig. 4
Box plot demonstrating the predicted probability of 30-day mortality after TJA as a function of the CARDE-B score. The box indicates the interquartile range,
the error bars indicate the minimum and maximum values, and the dots outside the error bars indicate the outliers.

status classification may be an accurate risk stratification tool
available for primary TJA10, but it is not routinely used or assessed
during the preoperative assessment by orthopaedic surgeons.
Agreement between physicians on ASA physical status classifica-
tion is poor, particularly between physicians of different special-
ties5. In contrast, we believe that the CARDE-B score can be easily
calculated using objective information available during a routine
clinical visit. This enables the use of the CARDE-B scoring system
in shared surgical decision-making prior to revision TJA.
The essential criteria for any new scoring system are to
outperform the existing systems and to offer easy implementation18.
We believe that the CARDE-B scoring system fulfills these criteria.
We intentionally chose to add a single point for each variable in the
CARDE-B score, compared with a complicated weighting system

Fig. 5
Receiver operating characteristic curves displaying the AUC of the CARDE-B scoring system, shown by the solid line (AUC, 0.85); the ASA classification,
shown by the dashed line (AUC, 0.77); and the mFI-5, shown by the dotted line (AUC, 0.67).
 430
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
that would impede practical implementation. Prior to evaluating the
CARDE-B system, a variable weighting system was evaluated and
offered only a small increase in accuracy (AUC, 0.86) with a sub-
stantial increase in the level of complexity required. With more than
one-third of orthopaedic surgeons in the United States practicing
in the private setting19 and the increasing burden of time con-
straints and documentation, a simple and easy-to-use clinical
scoring system is paramount. Furthermore, we were able to
demonstrate the accuracy of CARDE-B in an external, exclusive,
and nationally representative sample.
Revision TJA was conducted for a periprosthetic joint
infection 13% of the time in the ACS-NSQIP cohort. Published
rates of periprosthetic joint infection among revision TJAs have
ranged from 7% to 22%20,21. Periprosthetic joint infection is an
independent risk factor for postoperative mortality after revision
TJA17,22,23. We did not find a statistical association between peri-
prosthetic joint infection and mortality in our study. The potential
explanations for this discrepancy are that periprosthetic joint
infection at the time of revision TJA may be a proxy for poor host
status or that the 30-day follow-up time for this study was insuf-
ficient to demonstrate this association. In this study, we analyzed
hypoalbuminemia, which is a known indicator of general health
and frailty24. Hypoalbuminemia occurred more frequently when a
revision TJA was performed to treat a periprosthetic joint infection
(42%) than when a revision TJA was performed to treat a non-
septic etiology (12%, p < 0.001; data not shown). This highlights
the role of potential confounding, especially in the presence of
periprosthetic joint infection. However, when hypoalbuminemia
was removed from the regression model, periprosthetic joint
infection became a significant predictor of early mortality after
revision TJA (odds ratio [OR], 1.87 [95% confidence interval
(CI), 1.1 to 3.2]; p = 0.023; data not shown).
The CARDE-B scoring system assigns a single point to
both a BMI of <25 kg/m2 and an albumin level of <3.5 mg/dL.
Although hypoalbuminemia is a known manifestation of poor
nutrition24, it may also represent independent disease processes
such extravascular protein loss or tumor necrosis factor-alpha-
mediated critical illness25. Underweight patients are at an even
higher risk for death after revision TJA if they also have hypoalbu-
minemia, according to our findings. This highlights the importance
of preoperative patient optimization, particularly in the non-urgent
and non-emergency cases.
This study had several limitations. The use of CPT codes
for identifying procedures may be surgeon-dependent and may
not accurately reflect the procedure, especially when con-
current and concomitant procedure codes are entered. We
attempted to mitigate this by restricting our inclusion cri-
teria to patients undergoing a single principal procedure.
Although the CARDE-B score was internally and externally
validated, a factor that the CARDE-B score failed to assess, given
References
1. Weber M, Renkawitz T, Voellner F, Craiovan B, Greimel F, Worlicek M, Grifka J,
Benditz A. Revision surgery in total joint replacement is cost-intensive. Biomed Res
Int. 2018 Sep 25;2018:8987104.
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y
the limitations of large data sets, was the degree of preoperative
patient optimization. Given the complexity of sampling weights in
postestimation analysis, we chose not to include them in the
validation analysis. However, we believe that the demonstrable
accuracy of CARDE-B in the unweighted NIS data set provided
valuable information about its extrapolation to external samples.
Finally, we chose to exclude a substantial percentage of patients
with missing data to create a complete-case data set8. We attempted
to mitigate the potential bias that may arise from such a situation
by comparing the included and excluded patients (Appendix,
Supplemental Table 1).
To our knowledge, however, this is the first scoring sys-
tem designed to predict the risk of early mortality after revision
TJA with information easily obtainable during routine history-
taking. The accuracy and simplicity of the CARDE-B scoring
system make it readily available for further validation in expanded
data sets or subsets of other databases.
In conclusion, the CARDE-B scoring system provides
arthroplasty surgeons with a simple and valuable tool for
helping patients to decide between operative and nonoperative
management of complex arthroplasty-related complications.
Appendix
Supporting material provided by the authors is posted
with the online version of this article as a data supplement
at jbjs.org (http://links.lww.com/JBJS/G315). n
Micheal Raad, MD1
Raj Amin, MD1
Varun Puvanesarajah, MD1
Farah Musharbash, MD1
Sandesh Rao, MD1
Matthew J. Best, MD1
Derek F. Amanatullah, MD, PhD2
1Department of Orthopaedic Surgery, The Johns Hopkins University
School of Medicine, Baltimore, Maryland
2Department of Orthopaedic Surgery, Stanford Medicine, Redwood City,
California
Email address for D.F. Amanatullah: dfa@stanford.edu
ORCID iD for M. Raad: 0000-0001-8167-5808
ORCID iD for R. Amin: 0000-0001-7597-3542
ORCID iD for V. Puvanesarajah: 0000-0002-3296-6403
ORCID iD for F. Musharbash: 0000-0003-0739-2863
ORCID iD for S. Rao: 0000-0001-7346-3274
ORCID iD for M.J. Best: 0000-0002-4401-2834
ORCID iD for D.F. Amanatullah: 0000-0002-6203-5853
2. Subramaniam S, Aalberg JJ, Soriano RP, Divino CM. New 5-factor modified frailty
index using American College of Surgeons NSQIP data. J Am Coll Surg. 2018 Feb;
226(2):173-181.e8. Epub 2017 Nov 16.
 431
TH E JO U R NA L O F B O N E & JO I N T SU RG E RY J B J S . O RG
d
V O L U M E 103-A N U M B E R 5 M A R C H 3, 2 021
d d
3. Daabiss M. American Society of Anesthesiologists physical status classification.
Indian J Anaesth. 2011 Mar;55(2):111-5.
4. Traven SA, Reeves RA, Slone HS, Walton ZJ. Frailty predicts medical complica-
tions, length of stay, readmission, and mortality in revision hip and knee arthro-
plasty. J Arthroplasty. 2019 Jul;34(7):1412-6. Epub 2019 Mar 7.
5. Knuf KM, Maani CV, Cummings AK. Clinical agreement in the American Society of
Anesthesiologists physical status classification. Perioper Med (Lond). 2018 Jun 19;
7:14.
6. Bilimoria KY, Liu Y, Paruch JL, Zhou L, Kmiecik TE, Ko CY, Cohen ME. Develop-
ment and evaluation of the universal ACS NSQIP surgical risk calculator: a decision
aid and informed consent tool for patients and surgeons. J Am Coll Surg. 2013 Nov;
217(5):833-42.e1-3. Epub 2013 Sep 18.
7. Shiloach M, Frencher SKJ Jr, Steeger JE, Rowell KS, Bartzokis K, Tomeh MG,
Richards KE, Ko CY, Hall BL. Toward robust information: data quality and inter-
rater reliability in the American College of Surgeons National Surgical Quality
Improvement Program. J Am Coll Surg. 2010 Jan;210(1):6-16. Epub 2009
Nov 22.
8. Aziz KT, Best MJ, Shi BY, Srikumaran U. Missing data in the National Surgical
Quality Improvement Program database: how does it affect the identification of risk
factors for shoulder surgery complications? Arthroscopy. 2020 May;36(5):
1233-1239.e3. Epub 2020 Jan 16.
9. Kamath AF, Nelson CL, Elkassabany N, Guo Z, Liu J. Low albumin is a risk factor
for complications after revision total knee arthroplasty. J Knee Surg. 2017 Mar;
30(3):269-75. Epub 2016 Jun 30.
10. Gronbeck C, Cote MP, Lieberman JR, Halawi MJ. Risk stratification in primary
total joint arthroplasty: the current state of knowledge. Arthroplast Today. 2019 Feb
5;5(1):126-31.
11. Lewis-Beck M, Bryman A, Futing Liao T. Variance inflation factors. In: The SAGE
Encyclopedia of Social Science Research Methods. Sage; 2012.
12. Metz CE. Basic principles of ROC analysis. Semin Nucl Med. 1978 Oct;8(4):
283-98.
13. Hosmer DW, Lemesbow S. Goodness of fit tests for the multiple logistic
regression model. Commun Stat Theory Method. 1980;9(10):1043-69.
T H E CAR DE- B S C O R I N G S Y S T E M P R E D I C T S 3 0-D AY M O R TA L I T Y
A F T E R R E V I S I O N T O TA L J O I N T A R T H R O P L A S T Y
14. Mayhew D, Mendonca V, Murthy BVS. A review of ASA physical status - historical
perspectives and modern developments. Anaesthesia. 2019 Mar;74(3):373-9.
Epub 2019 Jan 15.
15. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or
more correlated receiver operating characteristic curves: a nonparametric approach.
Biometrics. 1988 Sep;44(3):837-45.
16. Agency for Healthcare Research and Quality. HCUP Fast Stats – most common
operations during inpatient stays. 2020 May. Accessed 2020 Dec 28. https://www.
hcup-us.ahrq.gov/faststats/NationalProceduresServlet
17. Zmistowski B, Karam JA, Durinka JB, Casper DS, Parvizi J. Periprosthetic joint
infection increases the risk of one-year mortality. J Bone Joint Surg Am. 2013 Dec
18;95(24):2177-84.
18. Ferguson MK. The rationale for developing scoring systems for clinical practice.
Thorac Surg Clin. 2007 Aug;17(3):343-51.
19. Lundy DW. A day at the office: is private practice orthopaedic surgery dead? Clin
Orthop Relat Res. 2016 Nov;474(11):2354-6. Epub 2016 Aug 16.
20. Boelch SP, Jakuscheit A, Doerries S, Fraissler L, Hoberg M, Arnholdt J, Rudert M.
Periprosthetic infection is the major indication for TKA revision - experiences from a uni-
versity referral arthroplasty center. BMC Musculoskelet Disord. 2018 Nov 10;19(1):395.
21. Lachiewicz PF, Soileau ES. Changing indications for revision total hip arthro-
plasty. J Surg Orthop Adv. 2005 Summer;14(2):82-4.
22. Boddapati V, Fu MC, Mayman DJ, Su EP, Sculco PK, McLawhorn AS. Revision
total knee arthroplasty for periprosthetic joint infection is associated with increased
postoperative morbidity and mortality relative to noninfectious revisions. J Arthro-
plasty. 2018 Feb;33(2):521-6. Epub 2017 Sep 23.
23. Jones MD, Parry M, Whitehouse MR, Blom AW. Early death following revision
total hip arthroplasty. Hip Int. 2018 Jul;28(4):400-6. Epub 2017 Apr 12.
24. Brock F, Bettinelli LA, Dobner T, Stobbe JC, Pomatti G, Telles CT. Prevalence of
hypoalbuminemia and nutritional issues in hospitalized elders. Rev Lat Am Enfer-
magem. 2016 Aug 8;24:e2736.
25. Levitt DG, Levitt MD. Human serum albumin homeostasis: a new look at the
roles of synthesis, catabolism, renal and gastrointestinal excretion, and the clinical
value of serum albumin measurements. Int J Gen Med. 2016 Jul 15;9:229-55.