Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2019 May 1.
Published in final edited form as: Cancer. 2018 Feb 1;124(9):2018–2025. doi: 10.1002/cncr.31269

Adapting the Elixhauser Comorbidity Index for Cancer Patients

Hemalkumar B Mehta 1, Sneha D Sura 2, Deepak Adhikari 3, Clark R Andersen 4, Stephen Williams 1, Anthony J Senagore 1, Yong-Fang Kuo 4, James S Goodwin 5
PMCID: PMC5910176  NIHMSID: NIHMS935327  PMID: 29390174

Abstract

Backgrounds

We aimed to adapt the Elixhauser comorbidity index for four cancer-specific populations (breast, prostate, lung, and colorectal) and compare three versions of the Elixhauser comorbidity score (individual comorbidities, summary comorbidity score, and cancer-specific summary comorbidity score) with three versions of the Charlson comorbidity score in predicting 2-year survival in four cancers.

Methods

This cohort study used Texas Cancer Registry linked Medicare data from 2005–2011 of older patients diagnosed with breast (n=19,082), prostate (N=23,044), lung (n=26,047), or colorectal (n=16,693) cancer. For each cancer cohort, the data were split into training and validation cohorts. In the training cohort, competing risk regression was used to model the association of Elixhauser comorbidities with 2-year non-cancer mortality, and cancer-specific weights were derived for each comorbidity. In the validation cohort, competing risk regression was used to compare three versions of the Elixhauser comorbidity score with three versions of the Charlson comorbidity score. Model performance was evaluated using c-statistics.

Results

The 2-year non-cancer mortality rate was 14.5% (lung), 11.5% (colorectal), 5.7% (breast), and 4.1% (prostate cancer). Cancer-specific Elixhauser comorbidity scores (c-statistics: breast=0.773, prostate=0.772, lung=0.579, colorectal=0.680) performed slightly better than cancer-specific Charlson comorbidity scores, i.e., the National Cancer Institute combined index (breast=0.762, prostate=0.767, lung=0.578, colorectal=0.674). Individual Elixhauser comorbidities performed best (c-statistics: breast=0.779, prostate=0.783, lung=0.587, colorectal=0.687).

Conclusions

The cancer-specific Elixhauser comorbidity score performed as well as or slightly better than the cancer-specific Charlson comorbidity score in predicting 2-year survival. If sample size permits, using individual Elixhauser comorbidities may best control for confounding in cancer outcomes research.

Keywords: Charlson comorbidity score, Elixhauser comorbidity score, National Cancer Institute combined index, confounding control, comorbidity

INTRODUCTION

Comorbidity is common among cancer patients and affects cancer diagnosis, treatment, and outcomes.1 Given the increased risk of comorbidities in the aging cancer population, improved methods to assess comorbidities are needed. Accurate measurement and control of comorbidities in cancer outcomes research is crucial to reduce confounding. Different methods such as the Charlson comorbidity score, the National Cancer Institute (NCI) combined index, and the Elixhauser comorbidity score are available to control for comorbidities.24

A comorbidity measure developed or adapted for a specific disease population performs better than a generic score.5 The Charlson score includes 17 comorbidities and was originally developed to predict 1-year mortality using all inpatient data from one hospital and validated in breast cancer patients from another hospital.2 Klabunde et al. modified the Charlson comorbidity index for breast, prostate, colorectal, and lung cancer patients using the Surveillance, Epidemiology, and End Results (SEER)-Medicare data.3 This cancer-specific Charlson comorbidity score, called the NCI combined index, outperformed the original Charlson comorbidity score in predicting 2-year survival in cancer patients.

The Elixhauser comorbidity index includes 29 comorbidities.4,6 A systematic review found that the Elixhauser comorbidity score performs better than Charlson comorbidity score in predicting mortality beyond 30 days.7 In oral and colorectal cancer patients, two studies reported that the Elixhauser performed better than the Charlson comorbidity score in predicting three-year survival.8,9 In contrast, one study showed better performance of the Charlson comorbidity score in bladder cancer patients.10

No consensus exists on the optimal method to define comorbidities in cancer.1,11 While the Charlson comorbidity index has been adapted for cancer patients, the Elixhauser comorbidity index has not. Moreover, both Charlson and Elixhauser comorbidity measures can be used as: (i) individual comorbidities, (ii) a summary score using generic weights, and (iii) a cancer-specific summary score using cancer-specific weights.

Modifying the Elixhauser comorbidity index for cancer-specific populations may improve its performance in this population. Therefore, the study objectives were to (i) adapt the Elixhauser comorbidity index for four cancer-specific populations and (ii) compare the performance of three versions of the Elixhauser comorbidity score with three versions of the Charlson comorbidity score in predicting 2-year survival in the four most common cancers diagnosed in the United States (breast, prostate, lung, and colorectal).

METHODS

Data Source and Study Sample

The study used the Texas Cancer Registry (TCR) linked Medicare data from 2005 to 2011. The TCR is linked with Medicare data under the guidance of NCI, TCR, and the Centers for Medicare and Medicaid Services. This linked data set provides detailed information on elderly adults with cancer in Texas, and approximately 98% of all people aged 65 and older in TCR are matched with Medicare enrollment and claims files. Similar to SEER files, TCR provides detailed information on patient demographics and the clinical characteristics of cancer such as stage of disease, tumor size, the first course of therapy, and cause of death, i.e., cancer or non-cancer.12 Medicare data provide information on the patient’s health care utilization. Medicare files used for this study included the denominator file, the Medicare provider analysis and review file (MedPAR) for inpatient claims, the carrier claims file, and the outpatient Standard Analytical File.

Four cancer-specific cohorts were developed by including older adults (age >65 years) diagnosed with breast, prostate, lung, or colorectal cancer. Patients were included in the final cohorts if they had continuous enrollment in Medicare Parts A and B, with no health maintenance organization enrollment, in the year before cancer diagnosis.

Outcome

The study outcome was 2-year non-cancer mortality; cancer mortality was treated as a competing risk. The underlying cause of death was determined from death certificate data as recorded by the TCR. The TCR is equivalent to SEER in timeliness, completeness and data quality.21 The agreement between the cause of death reviewed from the medical charts compared to SEER reported cause of death is over 95%.13,14 Non-cancer mortality was selected as an outcome because of our focus on deaths attributable to comorbid conditions; this method was consistent with the Charlson, Klabunde and NCI methodologies.2,3,15

Covariates

Age, sex (for lung and colorectal cancer) and stage of cancer (local, regional, and distant) were included as covariates.

Comorbidity Scores

Medicare inpatient and outpatient claims data were used to identify comorbidities. We used a baseline one year to define comorbidity scores. Comorbidity was identified as when the patient had at least one diagnosis from the inpatient file or at least two distinct diagnoses recorded >30 days apart from the outpatient file in the 365 days before cancer diagnosis.3,16

Charlson comorbidity score

The Charlson comorbidity index includes 17 comorbidities. The original Charlson comorbidity score was derived using inpatient data from 607 patients from New York Hospital-Cornell Medical Center and validated in 685 women with breast cancer from Yale New Haven Hospital. Weights for each comorbidity was derived and summed to obtain a summary comorbidity score. Klabunde et al. adapted the Charlson comorbidity index for four cancers and derived cancer-specific weights.3 This cancer-specific Charlson comorbidity score was termed the NCI combined index. The documentation and SAS codes for the NCI combined index are available on the NCI website.17

In this study, we used three versions of the Charlson comorbidity score: (i) individual Charlson comorbidities, (ii) a summary Charlson comorbidity score derived using original weights,2 and (iii) a cancer-specific summary Charlson comorbidity score, i.e., NCI combined index, derived using cancer-specific weights.3

Adapting the Elixhauser Comorbidity Score for Cancer Patients

The Elixhauser comorbidity index includes 29 disease conditions.4 Walraven et al. developed weights for 29 Elixhauser comorbidities to operationalize the score as a summary score.6 However, the weights were derived using inpatient data from a hospital, and they were not cancer-specific. Therefore, first we constructed cancer-specific weights for the Elixhauser comorbidity score. To do so, we performed a 50:50 split to derive training and validation cohorts for each cancer. The derivation cohort was used to obtain cancer-specific weights for Elixhauser comorbidities. We derived weights for each cancer cohort separately.

We used three versions of the Elixhauser comorbidity index: (i) individual Elixhauser comorbidities, (ii) a summary Elixhauser comorbidity score derived using Walraven weights,6 and (iii) a cancer-specific Elixhauser comorbidity score derived using cancer-specific weights.

Statistical Analyses

Descriptive statistics were performed in the four cancer-specific cohorts to describe demographics, death rates, and comorbidity prevalence. A training sample was used to derive cancer-specific weights for the Elixhauser comorbidity score. Fine and Gray’s competing risk model was used to model the association of Elixhauser comorbidities with 2-year survival, while controlling for age, gender, and stage of cancer. The beta coefficients obtained from the competing risk model were multiplied by 10 and rounded to the nearest integer to obtain weights for each comorbidity.18 This was done separately for each cancer cohort.

We used the validation sample to compare the performance of three versions of the Elixhauser comorbidity score with three versions of the Charlson comorbidity score. Fine and Gray’s competing risk model was used to predict 2-year non-cancer mortality. First, we constructed a baseline model that included age, gender, and stage of cancer. Six models were constructed: three for the Elixhauser comorbidity score and three for the Charlson comorbidity score. All models were compared using c-statistics and Akaike information criteria (AIC). Statistical comparison between c-statistics was performed using bootstrapped estimates with 1,000 replicates.

All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC), STATA 13.2 (STATA Inc., College Station, TX) and R (R version 3.4.0). This study was considered non-human subjects research and was deemed exempt by the Institutional Review Board of the University of Texas Medical Branch.

RESULTS

Descriptive Statistics

The overall study sample included 19,082 breast, 23,044 prostate, 26,047 lung, and 16,963 colorectal cancer patients. Table 1 reports the descriptive statistics for the four cancers by stage of diagnosis. The mean age of patients ranged from 73.6 ± 5.8 years (prostate cancer) to 77.2 ± 7.3 years (colorectal cancer). Over 92% of breast and prostate cancer, 83% of colorectal cancer, and 50% of lung cancer patients were diagnosed with local or regional stage of cancer. The overall 2-year mortality rate was highest for lung (73.9%) followed by colorectal (33.8%), breast (12.1%), and prostate (8.1%) cancer. The 2-year non-cancer mortality rate followed a similar pattern: lung (14.5%), colorectal (11.5%), breast (5.7%), and prostate (4.1%) cancer. The cancer and non-cancer death rates were higher for patients with advanced stage of cancer.

Table 1.

Descriptive Statistics of Study Cohorts

Cancer site and stage Number (%) Age in years, mean ±
SD
Female Gender, % 2-year Death Rate
Cancer, Number (%) Non-cancer, Number (%)
Breast 19,082 (100) 75.8 ± 7.0 100 1,219 (6.4) 1,085 (5.7)
  Local 12,523 (65.6) 75.7 ± 6.9 100 286 (2.3) 566 (4.5)
  Regional 5,098 (26.7) 75.9 ± 7.1 100 376 (7.4) 344 (6.8)
  Distant 1461 (7.7) 76.5 ± 7.4 100 557 (38.1) 175 (12.0)

Prostate 23,044 (100) 73.6 ± 5.8 0 926 (4.0) 952 (4.1)
  Local/regional 21,642 (93.9) 73.3 ± 5.6 0 394 (1.8) 738 (3.4)
  Distant 1,402 (6.1) 77.6 ± 7.6 0 532 (38.0) 214 (15.3)

Lung 26,047 (100) 75.7 ± 6.5 48.4 15,474 (59.4) 3,775 (14.5)
  Local 6,084 (23.4) 76.0 ± 6.3 52.2 1,833 (30.1) 855 (14.1)
  Regional 6,705 (25.7) 75.4 ± 6.3 47.7 3,484 (52.0) 1,019 (15.2)
  Distant 13,258 (50.9) 75.8 ± 6.6 47.1 10,157 (76.6) 1,901 (14.3)

Colorectal 16,963 (100) 77.2 ± 7.3 53.3 3,778 (22.3) 1,955 (11.5)
  Local 7,730 (45.6) 77.2 ± 7.2 52.4 675 (8.7) 841 (10.9)
  Regional 5,900 (34.8) 77.4 ± 7.4 54.5 1,097 (18.6) 661 (11.2)
  Distant 3,333 (17.1) 76.9 ± 7.4 53.1 2,006 (60.2) 453 (13.6)

Abbreviation: SD, standard deviation.

Comorbidity Prevalence

Table 2 displays the prevalence of the Charlson and Elixhauser comorbidities among the four cancers. Among Charlson comorbidities, diabetes without chronic complications, chronic pulmonary disease, congestive heart failure, diabetes with complications, and moderate or severe renal disease were the most prevalent conditions. Among Elixhauser comorbidities, hypertension, uncomplicated diabetes, chronic pulmonary disease, hypothyroidism, and deficiency anemia were the most prevalent conditions.

Table 2.

Prevalence of Charlson/NCI combined index and Elixhauser Comorbidities in Four Cancer Cohortsa

Comorbidities, % Breast Prostate Lung Colorectal
N = 19,082 N = 23,044 N = 26,047 N = 16,963
Charlson Comorbiditiesb
  Myocardial infarction 1.3 1.6 3.3 2.6
  Congestive heart failure 6.4 4.7 11.7 10.8
  Peripheral vascular disease 3.1 3.3 8.8 5.3
  Cerebrovascular disease 4.4 3.9 7.1 6.0
  Dementia 1.3 0.5 1.4 1.7
  Chronic pulmonary disease 9.1 8.1 32.3 12.1
  Connective tissue disease/Rheumatologic disease 2.3 1.0 2.7 1.8
  Ulcer disease 0.6 0.5 1.0 1.3
  Mild liver disease 0.23 0.13 0.4 0.3
  Diabetes without chronic complications 16.4 15.4 17.0 18.8
  Hemiplegia or paraplegia 0.3 0.3 0.6 0.5
  Moderate or severe renal disease 3.8 3.7 5.4 5.5
  Diabetes with complications/ESRD 4.3 2.8 3.9 4.9
  Moderate or severe liver disease 0.09 0.02 0.13 0.16
  AIDS 0.01 0.05 0.07 0.04

Elixhauser Comorbidities
  Congestive heart failure 6.5 4.8 11.5 10.5
  Valvular disease 4.5 3.7 5.4 5.5
  Pulmonary circulation disorders 1.0 0.6 1.5 1.2
  Peripheral vascular disorders 6.0 5.3 13.1 8.8
  Hypertension 58.1 49.8 58.0 57.4
  Paralysis 0.7 0.6 1.0 1.1
  Neurodegenerative disorders 4.7 2.5 4.7 5.5
  Chronic pulmonary disease 10.4 9.1 34.0 13.0
  Diabetes, uncomplicated 20.5 18.1 20.7 23.5
  Diabetes, complicated 5.4 3.7 5.2 6.2
  Hypothyroidism 16.4 4.9 10.5 11.3
  Renal failure 3.7 3.7 5.5 5.3
  Liver disease 0.5 0.4 0.7 0.6
  Peptic ulcer disease, no bleeding 0.05 0.03 0.07 0.08
  AIDS/HIV 0.01 0.05 0.07 0.04
  Rheumatoid arthritis/collagen vascular diseases 3.1 1.3 3.3 2.3
  Coagulopathy 1.4 1.3 1.8 1.7
  Obesity 2.1 1.2 1.7 2.0
  Weight loss 1.0 0.9 3.4 2.9
  Fluid and electrolyte disorders 6.0 3.8 9.4 8.4
  Blood loss anemia 0.7 0.4 1.0 2.4
  Deficiency anemia 10.9 7.5 14.7 22.3
  Alcohol abuse 0.2 0.5 0.8 0.5
  Drug abuse 0.05 0.05 0.14 0.04
  Psychosis 2.6 1.2 2.4 2.3
  Depression 4.3 1.6 4.4 3.9
a

Charlson and Elixhauser used different algorithms to identify comorbidities. Therefore, the prevalnce of ceratin conditions may differ for both comorbidity indices.

b

Both Charlson comorbidity and National Cancer Institute (NCI) combined index have the same comorbidities. They differ only in the weights assigned to each.

Adapting the Elixhauser Comorbidity Score for Cancer Patients

The training dataset included 9,391 breast, 11,346 prostate, 12,804 lung, and 8,314 colorectal cancer patients. Using training datasets for the four cancers, we derived cancer-specific weights for Elixhauser comorbidities (Table 3). The weights for comorbidities differed slightly by cancer type. For instance, the weight for congestive heart failure were 7 for the breast cancer cohort, 9 for the prostate cancer cohort, 1 for the lung cancer cohort, and 3 for the colorectal cancer cohort. Four sets of cancer-specific weights were then used to calculate cancer-specific summary Elixhauser comorbidity scores.

Table 3.

Coefficients for Elixhauser Comorbidities for Four Cancers Derived from the Training Sample a,b

Elixhauser Comorbidities Breast Prostate Lung Colorectal
N = 9,391 N = 11,346 N =12,804 N = 8,314

β HR Weight β HR Weight β HR Weight β HR Weight
Congestive heart failure 0.71 2.03 7 0.86 2.36 9 0.11 1.11 1 0.34 1.41 3
Valvular disease 0.08 1.08 1 0.12 1.13 1 0.13 1.14 1 0.13 1.14 1
Pulmonary circulation disorders 0.46 1.58 5 −0.21 0.81 −2 0.31 1.36 3 0.09 1.10 1
Peripheral vascular disorders 0.39 1.48 4 0.27 1.31 3 0.21 1.23 2 0.21 1.23 2
Hypertension −0.11 0.90 −1 −0.08 0.93 −1 −0.04 0.96 0 −0.04 0.96 0
Paralysis 0.60 1.82 6 0.27 1.31 3 0.16 1.18 2 −0.25 0.78 −2
Neurodegenerative disorders 0.52 1.69 5 0.22 1.24 2 0.30 1.35 3 0.24 1.27 2
Chronic pulmonary disease 0.46 1.58 5 0.48 1.62 5 0.26 1.29 3 0.50 1.64 5
Diabetes, uncomplicated 0.33 1.40 3 0.22 1.25 2 −0.02 0.98 0 0.14 1.15 1
Diabetes, complicated −0.07 0.93 −1 0.23 1.26 2 0.15 1.16 1 0.21 1.23 2
Hypothyroidism 0.07 1.07 1 −0.38 0.69 −4 0.10 1.11 1 0.14 1.15 1
Renal failure 0.28 1.33 3 0.15 1.16 1 0.19 1.21 2 0.15 1.16 1
Liver disease 0.65 1.91 6 −0.42 0.66 −4 −0.02 0.98 0 0.65 1.91 6
Rheumatoid arthritis/collagen vascular diseases 0.00 1.00 0 0.01 1.01 0 0.30 1.35 3 0.11 1.11 1
Coagulopathy 0.20 1.22 2 0.03 1.03 0 −0.15 0.86 −1 0.18 1.20 2
Obesity −0.56 0.57 −6 −0.75 0.47 −7 −0.06 0.94 −1 −0.29 0.75 −3
Weight loss 0.28 1.32 3 0.09 1.09 1 0.26 1.30 3 0.28 1.32 3
Fluid and electrolyte disorders 0.17 1.18 2 0.40 1.49 4 0.03 1.03 0 0.26 1.30 3
Blood loss anemia 0.21 1.23 2 −0.63 0.53 −6 −0.03 0.97 0 0.13 1.14 1
Deficiency anemia 0.08 1.08 1 0.20 1.22 2 0.00 1.00 0 0.12 1.12 1
Alcohol abuse 0.19 1.21 2 0.94 2.56 9 −0.02 0.98 0 −0.09 0.91 −1
Psychosis 0.38 1.47 4 0.44 1.56 4 0.05 1.05 1 0.16 1.17 2
Depression 0.15 1.16 2 0.03 1.03 0 0.24 1.27 2 0.32 1.38 3

Abbreviations: β, regression coefficient; HR, hazards ratio.

a

Weight is calculated as β*10 and rounded to the nearest integer.

b

All models include age, gender, and stage of the cancer as covariates.

Comparative Performance of Elixhauser with Charlson Comorbidity Score: Six Models

The validation dataset included 9,691 breast, 11,698 prostate, 13,243 lung, and 8,649 colorectal cancer patients. Table 4 reports the c-statistics of the different models. The c-statistics of the baseline model that included age, gender, and cancer stage was 0.722 for the breast, 0.725 for the prostate, 0.548 for the lung, and 0.629 for the colorectal cancer cohort. The addition of any comorbidity score to the baseline model improved the c-statistics for all four cancers (Table 4).

Table 4.

Comparison of Comorbidity Scores in the Validation Sample, C-statisticsa

Breast Prostate Lung Colorectal
N = 9,691 N = 11,698 N = 13,243 N = 8,649
Baseline model 0.722 0.725 0.548 0.629

Elixhauser comorbidity score
  Individual Elixhauser comorbidities 0.779 0.783 0.587 0.687
  Summary Elixhauser comorbidity score 0.764 0.771 0.580 0.677
  Cancer-specific summary Elixhauser comorbidity score 0.773 0.772 0.579 0.680

Charlson comorbidity score / NCI combined index
  Individual Charlson comorbidities 0.770 0.772 0.579 0.679
  Summary Charlson comorbidity score 0.763 0.762 0.576 0.672
  Cancer-specific summary Charlson comorbidity score or NCI combined index 0.762 0.767 0.578 0.674
a
P-values for c-statistics comparison (significant p-values are bolded):
Model Comparison Breast Prostate Lung Colon
Elixhauser vs. Charlson – individual comorbidiites 0.0284 0.0114 0.0462 0.0424
Cancer-specific Elixhauser vs. generic Elixhauser comorbidity score 0.0027 0.8752 0.8878 0.5317
Cancer-specific Elixhauser vs. generic Charlson comorbidity score 0.0100 0.0323 0.4416 0.0341
Cancer-specific Elixhauser vs. NCI combined index 0.0037 0.2817 0.7018 0.1525
Generic Charlson comorbidity score vs. NCI combined index 0.7399 0.0349 0.5210 0.1606

For both Elixhauser and Charlson comorbidity indices, a model with individual comorbidities performed better than models with summary comorbidity scores (Table 4). Individual Elixhauser comorbidities included in the model performed better than Charlson comorbidities used as individual variables (breast: 0.779 vs. 770, p=0.028; prostate: 0.783 vs. 0.772, p=0.011; lung: 0.587 vs. 579, p=0.046; colorectal: 0.687 vs. 679, p=0.042). Among summary comorbidity scores, a cancer-specific summary Elixhauser comorbidity score (breast: 0.773, prostate: 0.772, lung: 0.579, colorectal: 0.680) performed as well as or slightly better than a summary Elixhauser comorbidity score (breast: 0.764, prostate: 0.771, lung: 0.580, colorectal: 0.677), a summary Charlson comorbidity score (breast: 0.763, prostate: 0.762, lung: 0.576, colorectal: 0.672) or a cancer-specific summary Charlson comorbidity score (breast: 0.762, prostate: 0.767, lung: 0.578, colorectal: 0.674) (Table 4). AIC results are presented in Table 5. AIC results were similar to those using c-statistics. A cancer-specific summary Elixhauser comorbidity score had slightly lower AIC compared to other summary comorbidity scores, indicating better performance.

Table 5.

Comparison of Comorbidity Scores in the Validation Sample, Akaike Information Criterion (AIC)a

Breast Prostate Lung Colorectal
N = 9,691 N = 11,698 N = 13,243 N = 8,649
Baseline model 9702 8400 35827 17441

Elixhauser comorbidity score
  Individual Elixhauser comorbidities 9539 8241 35726 17211
  Summary Elixhauser comorbidity score 9552 8266 35731 17237
  Cancer-specific summary Elixhauser comorbidity score 9522 8253 35735 17208

Charlson comorbidity score / NCI combined index
  Individual Charlson comorbidities 9551 8261 35754 17241
  Summary Charlson comorbidity score 9558 8283 35752 17253
  Cancer-specific summary Charlson comorbidity score or NCI combined index 9566 8259 35743 17240
a

The lower AIC indicates a better model fit.

DISCUSSION

We adapted the Elixhauser comorbidity index for cancer patients and compared the performance of three versions of the Elixhauser comorbidity index with three versions of the Charlson comorbidity index in their ability to predict 2-year survival. For both Charlson and Elixhauser comorbidity indices, the use of individual comorbidities performed better than summary comorbidity scores; Elixhauser comorbidities as individual variables performed the best. Among summary scores, we found that the cancer-specific summary Elixhauser comorbidity score performed as well as or slightly better than generic and cancer-specific summary Charlson comorbidity scores.

Klabunde et al. derived cancer-specific weights for the Charlson comorbidity score. However, to our knowledge, this is the first study to adapt the Elixhauser comorbidity score for cancer patients by deriving four cancer-specific weights, and comparing them with three commonly-used versions of the Charlson comorbidity score. Several studies have shown that comorbidity scores predict survival and health-related quality of life in colorectal cancer patients.19, 20, 21 Lieffers et al. reported that individual Elixhauser comorbidities (c=0.864) performed better than individual Charlson comorbidities (c=0.831) in predicting 2-year survival in colorectal cancer patients.8 Chang et al. showed that Elixhauser performed better than Charlson when using a summary comorbidity score (c=0.654 vs 0.646) or separate indicator variables for comorbidities (c=0.677 vs 0.651) in oral cancer patients.9 However, one study of bladder cancer patients showed that Charlson (c=0.798) performed better than Elixhauser (c=0.770) in predicting 5-year survival; the authors used the number of comorbidities (0, 1, 2, 3, 4, ≥5) rather than a summary score or individual comorbidities.10 Our findings extend the evidence on the performance of comorbidity indices in cancer outcomes research via our comprehensive comparison for four common types of cancer.

The use of Elixhauser comorbidities as individual variables had the best performance followed by the cancer-specific summary Elixhauser comorbidity score. This result is consistent with prior studies that showed that individual comorbidities have the best performance.18,22 Moreover, the better performance of the Elixhauser comorbidity index may be attributed to the fact that it has more comorbidities than the Charlson comorbidity index. If sample size is not an issue, the use of individual comorbidities should be the first choice. However, summary comorbidity scores can be useful to describe the overall burden of comorbidities in a single measure and they offer advantages when the sample size is low and higher statistical power is desired.

Comorbidities are common in older adults with cancer and affect the disease progression, stage at diagnosis, treatment, and outcomes. No gold standard method exists to control for comorbidities in cancer outcomes research.11 One study identified 21 separate approaches to measuring comorbidities in cancer patients. The authors concluded that, for administrative claims data, the Charlson comorbidity score, NCI combined index, and Elixhauser comorbidity scores were the best option, and the NCI combined index may be the best because of its cancer-specific weights.11 The cancer-specific weights we derived for Elixhauser showed the same or slightly better performance than a cancer-specific Charlson comorbidity score, i.e., the NCI combined index.

Some of the comorbidities (hypertension, obesity) were negatively associated with survival. This has been found in other studies.6 The explanation may vary with the comorbidity. For example, a hypertension diagnosis might be linked to access to medical care and to an increased likelihood that elevated blood pressure is recognized and treated. An obesity diagnosis indicates absence of cachexia, which is common in cancer patients and poor sign.

Our study had following strengths. We operationalized comorbidity indices in different ways and used Fine and Gray’s competing risk model to determine the asosciation of comorbidities with non-cancer mortality. However, our findings should be interpreted in the context of the study design. We used cancer registry data from Texas and our findings may not be generalizable to other populations. Moreover, we used TCR-Medicare linked data, and these findings may not be applicable to patients 65 years of age or less. Consistent with Charlson and Klabunde’s methods, we used non-cancer mortality as an outcome. The lower c-index among lung cancer patients suggests potential limitations in applying the ‘one size fits all’ approach using comorbidity scores even when considering cancer-specific scores. The inability to control for unknown confounders using large retrospective registry data is an inherent limitation. Due to high correlation between summary Charlson and Elixhauser comorbidity scores, we did not use them together in the same model. Future research can compare cancer-specific Elixhauser comorbidity scores with a pharmacy-based comorbidity index for cancer patients to determine if it improves the prediction of mortality.23 Use of the most appropriate comorbidity score will not be a panacea to control all confounding; several sources of cofounding such as disease severity, functional status, social support and health behavior need to be controlled to make unbiased treatment estimates in observational studies.

CONCLUSIONS

In conclusions, we adapted the summary Elixhauser comorbidity score for four cancer-specific populations. Elixhauser comorbidities used as individual variables in the model performed the best and, among summary scores, the cancer-specific summary Elixhauser comorbidity score performed equally to or slightly better than either the generic or cancer-specific Charlson comorbidity score. If sample size allows, one should use individual Elixhauser comorbidities to control for confounding due to comorbidities in cancer outcomes research. For small samples that require using a summary comorbidity score, cancer-specific summary Elixhauser comorbidity or Charlson comorbidity scores work equally well. The use of comorbidity scores is growing in cancer outcomes research. With changes in coding practices (ICD-9-CM to ICD-10-CM), future work should adapt algorithms to the new coding system and re-evaluate the applicability and comparative performance of comorbidity scores in cancer patients.

Acknowledgments

The authors thank Sarah Toombs Smith, PhD, Science Editor and Assistant Professor in the Sealy Center on Aging, University of Texas Medical Branch at Galveston, for her editorial assistance. Dr. Toombs Smith received no compensation for her assistance beyond her university salary.

Funding: Supported by the Cancer Prevention and Research Institute of Texas (RP1607674), National Cancer Institute (K05 CA134923), Agency for Healthcare Research and Quality (R24 HS022134), Clinical and Translational Science Award Mentored Career Development (KL2) Award (KL2TR001441) from the National Center for Advancing Translational Sciences, National Institutes of Health (NIH).

Footnotes

Conflicts of Interest: All authors report no conflicts of interest.

Author Disclosures: None

Author Contribution:

Hemalkumar B. Mehta: Conceptualization, methodology, formal analysis, writing - original draft, and writing - review and editing

Sneha D. Sura: Data curation, formal analysis, writing - review and editing

Deepak Adhikari: Data curation, formal analysis, writing - review and editing

Clark Andersen: Data curation, formal analysis, writing - review and editing

Stephen Williams: Methodology, writing - review and editing

Anthony J. Senagore: Methodology, writing - review and editing

Yong-Fang Kuo: Conceptualization, methodology, writing - review and editing

James S. Goodwin: Conceptualization, funding acquisition, methodology, writing - review and editing

References

  • 1.Sarfati D, Koczwara B, Jackson C. The impact of comorbidity on cancer and its treatment. CA Cancer J Clin. 2016;66(4):337–350. doi: 10.3322/caac.21342. [DOI] [PubMed] [Google Scholar]
  • 2.Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chron Dis. 1987;40(5):373–383. doi: 10.1016/0021-9681(87)90171-8. [DOI] [PubMed] [Google Scholar]
  • 3.Klabunde CN, Legler JM, Warren JL, Baldwin LM, Schrag D. A refined comorbidity measurement algorithm for claims-based studies of breast, prostate, colorectal, and lung cancer patients. Ann Epidemiol. 2007;17(8):584–590. doi: 10.1016/j.annepidem.2007.03.011. [DOI] [PubMed] [Google Scholar]
  • 4.Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8–27. doi: 10.1097/00005650-199801000-00004. [DOI] [PubMed] [Google Scholar]
  • 5.Iezzoni LI. Risk adjustment for measuring health care outcomes. 4. Chicago, Ill. Arlington, VA: Health Administration Press; 2013. [Google Scholar]
  • 6.van Walraven C, Austin PC, Jennings A, Quan H, Forster AJ. A modification of the Elixhauser comorbidity measures into a point system for hospital death using administrative data. Med Care. 2009;47(6):626–633. doi: 10.1097/MLR.0b013e31819432e5. [DOI] [PubMed] [Google Scholar]
  • 7.Sharabiani MT, Aylin P, Bottle A. Systematic review of comorbidity indices for administrative data. Med Care. 2012;50(12):1109–1118. doi: 10.1097/MLR.0b013e31825f64d0. [DOI] [PubMed] [Google Scholar]
  • 8.Lieffers JR, Baracos VE, Winget M, Fassbender K. A comparison of Charlson and Elixhauser comorbidity measures to predict colorectal cancer survival using administrative health data. Cancer. 2011;117(9):1957–1965. doi: 10.1002/cncr.25653. [DOI] [PubMed] [Google Scholar]
  • 9.Chang HJ, Chen PC, Yang CC, Su YC, Lee CC. Comparison of Elixhauser and Charlson methods for predicting oral cancer survival. Medicine. 2016;95(7):e2861. doi: 10.1097/MD.0000000000002861. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Boorjian SA, Kim SP, Tollefson MK, Carrasco A, Cheville JC, Thompson RH, Thapa P, Frank I. Comparative performance of comorbidity indices for estimating perioperative and 5-year all cause mortality following radical cystectomy for bladder cancer. J Urol. 2013;190(1):55–60. doi: 10.1016/j.juro.2013.01.010. [DOI] [PubMed] [Google Scholar]
  • 11.Sarfati D. Review of methods used to measure comorbidity in cancer populations: no gold standard exists. J Clin Epidemiol. 2012;65(9):924–933. doi: 10.1016/j.jclinepi.2012.02.017. [DOI] [PubMed] [Google Scholar]
  • 12. [Accessed November 9, 2017];About the Texas Cancer Registry. Available at: https://www.dshs.texas.gov/tcr/about.aspx.
  • 13.Penson DF, Albertsen PC, Nelson PS, Barry M, Stanford JL. Determining cause of death in prostate cancer: are death certificates valid? J Natl Cancer Inst. 2001;93(23):1822–1823. doi: 10.1093/jnci/93.23.1822. [DOI] [PubMed] [Google Scholar]
  • 14.Rink M, Fajkovic H, Cha EK, Gupta A, Karakiewicz PI, Chun FK, Lotan Y, Shariat SF. Death certificates are valid for the determination of cause of death in patients with upper and lower tract urothelial carcinoma. Eur Urol. 2012;61(4):854–855. doi: 10.1016/j.eururo.2011.12.055. [DOI] [PubMed] [Google Scholar]
  • 15.Stedman MR, Doria-Rose P, Warren JL, Klabunde CN, Mariotto A. The impact of different SEER-Medicare claims-based comorbidity indexes on predicting non-cancer mortality for cancer patients. [Accessed November 9, 2017];Comorbidity Technical Report. https://healthcaredelivery.cancer.gov/seermedicare/considerations/comorbidity-report.html.
  • 16.Klabunde CN, Potosky AL, Legler JM, Warren JL. Development of a comorbidity index using physician claims data. J Clin Epidemiol. 2000;53(12):1258–1267. doi: 10.1016/s0895-4356(00)00256-0. [DOI] [PubMed] [Google Scholar]
  • 17. [Accessed August 21, 2017];SEER-Medicare: Calculation of Comorbidity Weights. https://healthcaredelivery.cancer.gov/seermedicare/considerations/calculation.html.
  • 18.Mehta HB, Mehta V, Girman CJ, Adhikari D, Johnson ML. Regression coefficient-based scoring system should be used to assign weights to the risk index. J Clin Epidemiol. 2016;79:22–28. doi: 10.1016/j.jclinepi.2016.03.031. [DOI] [PubMed] [Google Scholar]
  • 19.Marventano S, Grosso G, Mistretta A, Bogusz-Czerniewicz M, Ferranti R, Nolfo F, Giorgianni G, Rametta S, Drago F, Basile F, Biondi A. Evaluation of four comorbidity indices and Charlson comorbidity index adjustment for colorectal cancer patients. Int J Colorectal Dis. 2014;29(9):1159–1169. doi: 10.1007/s00384-014-1972-1. [DOI] [PubMed] [Google Scholar]
  • 20.Marventano S, Forjaz M, Grosso G, Mistretta A, Giorgianni G, Platania A, Gangi S, Basile F, Biondi A. Health related quality of life in colorectal cancer patients: state of the art. BMC Surg. 2013;13(Suppl 2):S15. doi: 10.1186/1471-2482-13-S2-S15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Grosso G, Biondi A, Marventano S, Mistretta A, Calabrese G, Basile F. Major postoperative complications and survival for colon cancer elderly patients. BMC Surg. 2012;12(Suppl 1):S20. doi: 10.1186/1471-2482-12-S1-S20. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Austin SR, Wong YN, Uzzo RG, Beck JR, Egleston BL. Why summary comorbidity measures such as the Charlson Comorbidity Index and Elixhauser Score work. Med Care. 2015;53(9):e65–e72. doi: 10.1097/MLR.0b013e318297429c. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Sarfati D, Gurney J, Stanley J, Lim BT, McSherry C. Development of a pharmacy-based comorbidity index for patients with cancer. Med Care. 2014;52(7):586–593. doi: 10.1097/MLR.0000000000000149. [DOI] [PubMed] [Google Scholar]

RESOURCES