Sage Journals: Discover world-class research

Abstract

To compare the performance of the Chronic Conditions Warehouse (CCW) and the 38-condition Elixhauser Comorbidity Index in predicting all-cause mortality among Medicare beneficiaries with dementia, we used a national sample of 1,566,359 community-dwelling Medicare beneficiaries (age ≥65) with dementia, identified in 2018 claims data. Using elastic net logistic regression, we applied 30 CCW conditions and 38 Elixhauser comorbidities from 2018 to predict mortality at 30, 60, 180 days, and 1 year through December 31, 2019. Mortality rates were 2.42% (30 days), 4.27% (60 days), 10.77% (180 days), and 19.0% (1 year). All models demonstrated good discrimination (C-statistics: 0.696–0.731) and calibration, with no meaningful performance differences between the two measures. Elastic net models produced parsimonious predictors with performance comparable to traditional logistic regression. Both CCW and Elixhauser measures predicted all-cause mortality in dementia with similar accuracy. Elastic net offers a robust approach to claims-based mortality prediction.

Keywords

Elixhauser Comorbidity Index chronic condition data warehouse multimorbidity all-cause mortality elastic net model dementia

Introduction

Understanding multimorbidity—the presence of multiple chronic conditions—is essential for predicting patient outcomes, particularly all-cause mortality, in aging and high-risk populations (Barnett et al., 2012; Xu et al., 2017). While prediction of life expectancy at the individual level (ePrognosis, n.d.) is useful for patient care planning, systematic and standardized population-level mortality prediction is needed for policy development and program design (Suls et al., 2021). This is especially true for assessing outcomes in patients with dementia. Dementia is a debilitating neurological condition currently affecting 6.9 million Americans, a population that is expected to grow to 13 million by 2050 (Alzheimer’s Association, 2024). People with dementia have significantly higher mortality rates than older adults without this condition (Taudorf et al., 2021), related in large part to the high prevalence of multiple age-related chronic conditions (multimorbidity) (Stirland et al., 2025; Xin et al., 2025; Zhou et al., 2025). This high morbidity, coupled with the anticipated growth in the number of people with dementia, presents significant challenges for health care systems, particularly in predicting and managing mortality risk at a population level.

Robust measures of multimorbidity are important for evaluating mortality risk at the population level in dementia using national data sources such as Medicare claims. Two widely used claims-based measures for assessing ultimorbidity are the Chronic Conditions Data Warehouse (CCW), developed by the Centers for Medicare & Medicaid Services (CMS) with 30 conditions since 2017, and the Elixhauser Comorbidity Index, which includes 38 comorbidities according to the most recent guidelines from the Agency for Healthcare Research and Quality (AHRQ, 2022; CMS, n.d.-a, n.d.-b). Both measures offer objective, reproducible, and scalable methods for measuring multimorbidity (Maciejewski & Hammill, 2019; Quan et al., 2005).

Although both CCW and Elixhauser provide objective measures of multimorbidity, researchers face resource constraints when selecting a measure in their models, as these two comorbidity measures differ substantially in their data requirements and implementation (CMS, n.d.-a; Elixhauser et al., 1998; Quan et al., 2005; Research Data Assistance Center [ResDAC], n.d.-a). The 30 CCW conditions are predefined by CMS, facilitating consistent application across studies, whereas the Elixhauser comorbidities are typically coded in-house. While this flexibility allows researchers to tailor coding to specific study aims, it also introduces additional time and analytic effort. Furthermore, the two measures rely on different data sets, which may entail varying acquisition costs (ResDAC, n.d.-a). Thus, it is important to choose the right measure when estimating mortality risk.

To fully understand these considerations, it is useful to examine the features, similarities, and differences between the two measures. Although initially created for distinct purposes—the CCW to track chronic disease burden across populations and the Elixhauser to assess hospital resource use and predict inpatient mortality—the two indices share several similarities and some key differences (AHRQ, 2022; CMS, n.d.-a; Elixhauser et al., 1998). For example, both measures include chronic obstructive pulmonary disease (COPD). However, while some diseases appear in both, the level of detail can differ: the CCW includes a general diabetes category, whereas the Elixhauser distinguishes between diabetes with and without chronic complications. Similarly, the CCW tracks specific cancers such as breast or colorectal cancer, whereas the Elixhauser uses broader categories, such as metastatic cancer or solid tumors without metastasis, without specifying the exact cancer type. Certain conditions are unique to each measure—for instance, arthritis appears only in the CCW, while weight loss is included only in the Elixhauser.

Beyond differences in conditions, the measures were designed for different settings. The CCW classified individuals with chronic conditions and is widely used to evaluate disease burden, multimorbidity, and health care utilization across both inpatient and outpatient settings (Maciejewski & Hammill, 2019). In contrast, the Elixhauser was developed specifically for use with inpatient data (Elixhauser et al., 1998). Moreover, the CCW documents the first diagnosis of each condition, enabling the calculation of time since initial diagnosis—an important distinction from the Elixhauser. These differences underscore the importance of understanding each measure in health services research, as claims-based models inform population-level mortality estimation, risk stratification, resource allocation, policy implementation, and program design. A detailed overview of the two measures is provided below.

The CCW

Since 2017, the CCW has provided a comprehensive framework for identifying and tracking 30 chronic conditions in Medicare beneficiaries (CMS, n.d.-a), including diabetes, heart failure, chronic kidney disease, COPD, mood disorders, and Alzheimer’s disease, among others. Each condition is identified based on specific criteria, typically requiring at least one inpatient or two hospital outpatient (HOP) claims with relevant International Classification of Diseases, 10th Revision (ICD-10) diagnosis codes, or evidence of prescription drug use within a specified time frame (typically 1–2 years).

For data prior to 2017, the CCW provided files with algorithms for 27 chronic conditions spanning the years 1999 to 2021. These files covered the transition from the International Classification of Diseases, 9th Revision (ICD-9) to ICD-10 codes (CMS, n.d.-b). The update from 27 to 30 conditions involved the addition of three new conditions: urologic cancer, all-cause pneumonia, and Parkinson’s disease (including secondary Parkinsonism). These additions expanded the CCW’s capacity to capture a broader range of chronic conditions among Medicare beneficiaries. Importantly, the CCW only assigns condition flags to individuals with sufficient claims data to make a reliable diagnostic determination (e.g., 1–2 years of data).

The Elixhauser Comorbidity Index

The Elixhauser comorbidity index was developed to predict in-hospital mortality and hospital resource use with 30 binary diagnoses extracted from hospital discharge diagnoses and administrative data (Elixhauser et al., 1998) and expanded to 31 comorbidities (Quan et al., 2005). In 2022, the AHRQ updated the Elixhauser to include 38 diagnoses, incorporating an updated ICD-10 coding system (AHRQ, 2022) and input from clinical experts. This update introduced significant changes, including the modification of four measures to delineate disease severity (e.g., hypertension was split into hypertension with or without complication), the subdivision of the broad neurological disorders’ category into four specific groups, the addition of three new comorbidity measures (cerebrovascular disease, leukemia, and other thyroid), and the discontinuation of fluid and electrolyte disorders as a measure due to their classification as acute conditions.

Although the Elixhauser was designed for the inpatient setting, researchers have applied it in non-hospital settings. For instance, in a study predicting all-cause mortality in patients with rheumatoid arthritis and another study comparing four claims-based measures, including the Elixhauser, to predict non-cancer death in Medicare beneficiaries with colon cancer, the Elixhauser demonstrated good predictive performance (Baldwin et al., 2006; Y. J. Huang et al., 2021).

Machine Learning Models for Use With Claims Data—Elastic Net Models

Elastic net models are particularly useful for generalizing predictive models in large-scale claims data, especially for populations with high multimorbidity, as they address key limitations of traditional logistic regression. In such data sets, many beneficiaries have multiple chronic conditions, resulting in multicollinearity among the predictors. This makes it difficult for traditional logistic regression to distinguish the unique contribution of each variable. In addition, traditional logistic regression relies on p-values, which may detect statistically significant but practically insignificant differences in large samples. Models with many correlated or over-identified predictors may include an excessive number of covariates, increasing the risk of overfitting to a specific data set.

Elastic net combines L1 (lasso) and L2 (ridge) regularization to enable both variable selection and coefficient shrinkage, making it well-suited for high-dimensional, large-scale data with multicollinearity. Unlike logistic regression, which selects variables based on absolute statistical significance thresholds, lasso regularization excludes irrelevant variables by setting certain coefficients to zero through soft thresholding, while ridge regularization stabilizes estimates among correlated predictors (Mitchell, 1997; Zou & Hastie, 2005).

Effective parsimonious elastic net models offer at least five advantages: reduced data requirements, lower computational complexity, and improved system representation, transparency, and insightfulness (Daganzo et al., 2012). Elastic net models have demonstrated comparable predictive accuracy and power to traditional logistic regression models in previous studies while offering greater simplicity and robustness (Liu et al., 2023, 2025; Tibshirani, 1996; Zou & Hastie, 2005). These advantages are particularly important when modeling outcomes for individuals with chronic conditions such as dementia, a population characterized by complex multimorbidity patterns and high mortality risk.

Purpose of the Study

As the dementia population grows, there is an increasing need for accurate and scalable measures to predict mortality risk at the population level. Despite the similarities and differences between CCW and Elixhauser, no known studies to date have directly compared the CCW and the Elixhauser measures for predicting all-cause mortality in community-dwelling Medicare beneficiaries with chronic conditions, despite prior use of older versions of these measures in other settings (DuGoff et al., 2014; Thompson et al., 2015).

This study leverages elastic net models to evaluate the predictive performance of two widely used claims-based multimorbidity measures—the 30 predefined CCW chronic conditions and the 38 Elixhauser comorbidities developed by AHRQ—for predicting all-cause mortality among community-dwelling Medicare beneficiaries with dementia using comprehensive national claims. We also examine how the updated components of the CCW (e.g., all-cause pneumonia and Parkinson’s disease) and the Elixhauser (e.g., leukemia and refined comorbidity categories reflecting severity of illness) contribute to mortality prediction.

Method

Data Source and Variables

This study utilized a national sample of 2,836,382 Medicare beneficiaries aged 65 and older who had the end-of-year dementia flag in both the 27- and 30-condition versions of the CCW in 2018. Those who died in 2018 (19.56%) or had more than 100 days in a nursing home or hospital (19.57%) were excluded, as their mortality risk is heavily influenced by long-term institutionalization (Aneshensel et al., 2000; Mank et al., 2022), which differs from community-dwelling dementia beneficiaries and could bias the results. The final analytical sample comprised 1,566,359 community-dwelling beneficiaries with the chronic condition dementia.

Outcome Measures and Beneficiary Baseline Characteristics

The primary outcome variable was all-cause mortality, determined from beneficiaries’ death dates recorded in the 2019 CMS Master Beneficiary Summary File (MBSF). Four binary outcome measures were created to capture mortality at 30 days, 60 days, 180 days, and 1 year using a reference date of December 31, 2018. Thus, to create the mortality outcomes, we used mortality data from January 1 through December 31, 2019 (see Appendix Table A1 for detailed time elements of variables in the study). In addition, baseline information, including demographic information and dual eligibility for Medicare and Medicaid, was extracted from the 2018 MBSF (ResDAC, n.d.-b).

The 30 Predefined Chronic Conditions From the Chronic Condition Data Warehouse

We used a binary indicator for each of the 30 predefined CCW chronic conditions, which used the ResDAC-defined algorithm (CMS, n.d.-b) with predefined lookback windows (usually 1–2 years) ending December 31, 2018, to capture condition prevalence (Appendix Table A1). These algorithms draw from multiple sources, including carrier, inpatient, skilled nursing facilities (SNFs), HOP, and home health agency (HHA) claims. Each condition is defined using condition-specific criteria. For example, dementia is identified using at least one inpatient/SNF/HHA claim or two carrier/HOP claims with relevant diagnosis codes, while acute myocardial infarction requires only one inpatient claim.

We took advantage of the CCW’s ability to track the initial diagnosis of each condition, allowing the measurement of duration since diagnosis, which is essential for assessing clinical progression, understanding health care needs, and evaluating the effectiveness of chronic disease management programs (Maciejewski & Hammill, 2019). This information is particularly valuable in health services research for longitudinal analysis of health outcomes. In this study, the duration of each condition was calculated as the difference (in years) between January 1, 2019, and the first appearance of a diagnosis code for each condition, as captured in the 30 CCW chronic condition file of 2018.

The 38 Elixhauser Comorbidities Developed by the Agency for Healthcare Research and Quality

We used the most updated coding algorithm developed by AHRQ to code the 38 Elixhauser comorbidities (AHRQ, 2022). The Elixhauser, originally developed for inpatient settings, relies on hospital discharge diagnoses and administrative data. Although it has been used in non-hospital settings (Baldwin et al., 2006; Y. J. Huang et al., 2021; Song et al., 2021), no standardized algorithm has been specifically developed for coding the Elixhauser beyond the inpatient setting. Defining a reference period for coding in ambulatory care and community settings could enhance accuracy and relevance by capturing an individual’s longitudinal comorbidity history. In this study, we used all principal, primary, and secondary ICD-10 diagnosis codes in inpatient and carrier claims to determine the 1-year prevalence of comorbidities in 2018 when constructing the 38 AHRQ Elixhauser comorbidities. Each comorbidity was defined as having at least one inpatient or carrier claim for the condition. See Appendix Table A2 for algorithm differences between the predefined 30 CCW chronic conditions and the 38 AHRQ Elixhauser comorbidities.

Data Analyses

We first characterized the study sample and the four mortality rates using descriptive statistics. Next, we assessed the frequency of each individual Elixhauser comorbidity and chronic condition, as well as the descriptive statistics for the duration of each chronic condition.

We used elastic net models to assess the model performance of the 30 CCW chronic conditions and the 38 Elixhauser comorbidities on all-cause mortality. Before developing the model, we randomly divided the sample into two unequal subsamples. The model training sample comprised 70% (n = 1,096,451 beneficiaries) of the whole; the remaining 30% (n = 469,908 beneficiaries) served as the model validation sample. To minimize potential bias in selecting the validation data set, we used a 10-fold cross-validation split design by partitioning the training and validation data sets into 10 random, equal-sized subsamples, and the result is the average of all test results (Stone, 1974).

We assessed the predictive power of the models using the area under the receiver operating characteristic (ROC) curve, depicted as the C-statistic; C-statistic exceeding 0.7 is considered good and >0.8 strong (Hosmer & Lemeshow, 2000). In addition to discrimination, we assessed the calibration of the models using the Brier Score, E₉₀ (90th percentile of measurement error), and E_avg (average measurement error) (Y. Huang et al., 2020; Van Calster et al., 2019). The Brier score measures the accuracy of probabilistic predictions by calculating the mean squared difference between the predicted probability of a binary outcome and the observed outcome. A Brier score below 0.25 is considered acceptable, 0.05 to 0.10 is good, and less than 0.05 is excellent. E₉₀ and E_avg quantify miscalibration by summarizing the absolute differences between predicted and observed outcomes, using the 90th and average, respectively (Steyerberg, 2009). To complement these statistical evaluations, we visualized calibration using decile calibration plots, which provide a graphical representation of model calibration (Austin & Steyerberg, 2014).

For each outcome, we fit models using the 30 CCW chronic conditions, and then we refit the models by adding the duration of each CCW condition. We then fit the models using the 38 Elixhauser comorbidities. For all models, we included beneficiaries’ demographic information (age, race/ethnicity, and sex), and dual eligibility for Medicare and Medicaid. We reported model performance and calibration results for all four mortality outcomes, while we focused exclusively on the effects of the individual CCW chronic conditions and the Elixhauser comorbidities on two outcomes (30-day and 1-year mortality) to avoid redundancy. To evaluate model performance, we compared results from elastic net models with those from traditional logistic regression for these two outcomes. All data analyses were performed using R, using the glmnet package for the elastic net model analysis (Friedman et al., 2010) and the caret package for logistic regression (Kuhn, 2008).

Results

The sample included 62.6% women, with an average age of 82.3 years (standard deviation [SD]: 8.1; range: 65–113) (Appendix Table A3). The majority were White (78.2%) and female (62.6%). The mortality rates were notably high in this population, with 30-day, 60-day, 180-day, and 1-year mortality rates of 2.42%, 4.27%, 10.77%, and 19.0%, respectively.

Table 1 provides information on the prevalence and duration of each predefined CCW chronic condition, while Table 2 details the incidence of each Elixhauser comorbidity. Most of the study sample was diagnosed with non-Alzheimer’s dementia (95.3%), which includes unspecified, mixed, and other etiologies, while a third was diagnosed with Alzheimer’s dementia (36.4%). All beneficiaries had at least one type of dementia, and 31.7% had both types. Tables 1 and 2 confirm the high burden of multimorbidity in this population.

Table 1.

Descriptive Statistics and Odds Ratio of Mortality for 30 Predefined CCW Chronic Conditions for Medicare Beneficiaries With Dementia (N = 1,566,359).

Condition	N (%)	DurationM (SD)^a	Duration Mean (SD)^b	Odds ratio^c
Condition	N (%)	DurationM (SD)^a	Duration Mean (SD)^b	30-day Mortality	1-year Mortality
Alzheimer’s Disease	569,991 (36.39)	0.60 (0.94)	1.55 (0.89)	1.194	1.252
Non-Alzheimer’s Dementia	1,492,411 (95.28)	1.45 (0.92)	1.51 (0.89)	1.525	1.588
Acute Myocardial Infarction	35,754 (02.28)	0.06 (0.33)	1.29 (0.84)	1.572	1.320
Anemia	641,356 (40.95)	0.81 (1.05)	1.74 (0.88)	1.248	1.184
Asthma	132,974 (08.49)	0.19 (0.62)	1.85 (0.84)	0.861	0.864
Atrial Fibrillation	394,468 (25.18)	0.52 (0.99)	1.98 (0.91)	1.253	1.200
Benign Prostatic Hyperplasia	263,838 (16.84)	0.34 (0.79)	1.82 (0.83)	NA	0.957
Breast Cancer	79,375 (05.07)	0.11 (0.50)	1.99 (0.82)	NA	NA
Colorectal Cancer	38,228 (02.44)	0.05 (0.33)	1.80 (0.87)	1.086	1.082
Endometrial Cancer	8,818 (00.56)	0.01 (0.15)	1.67 (0.86)	NA	NA
Lung Cancer	21,812 (01.39)	0.03 (0.24)	1.75 (0.95)	1.884	1.928
Prostate Cancer	80,281 (05.13)	0.11 (0.51)	2.06 (0.81)	NA	NA
Urologic Cancer	14,206 (00.91)	0.02 (0.20)	1.80 (0.88)	1.037	1.085
Cataract	292,288 (18.66)	0.75 (1.09)	2.02 (0.81)	0.727	0.703
Chronic Kidney Disease	482,339 (30.79)	0.61 (1.01)	1.83 (0.90)	1.221	1.207
COPD	374,490 (23.91)	0.49 (0.94)	1.86 (0.88)	1.103	1.160
Mood Disorders	634,196 (40.49)	0.80 (1.07)	1.79 (0.88)	1.074	1.058
Diabetes	557,239 (35.58)	0.86 (1.21)	2.29 (0.79)	1.047	1.051
Glaucoma	257,548 (16.44)	0.42 (0.94)	2.24 (0.79)	0.902	0.872
Heart Failure	404,059 (25.80)	0.48 (0.91)	1.72 (0.91)	1.410	1.377
Hip/Pelvic Fracture	65,667 (04.19)	0.12 (0.47)	1.39 (0.87)	1.460	1.162
Hyperlipidemia	1,124,327 (71.78)	1.66 (1.12)	2.17 (0.73)	0.709	0.707
Hypertension	1,350,879 (86.24)	2.06 (1.00)	2.34 (0.69)	0.956	0.974
Hypothyroidism	471,544 (30.10)	0.66 (1.06)	2.07 (0.79)	NA	0.993
Ischemic Heart Disease	568,752 (36.31)	0.78 (1.10)	1.98 (0.85)	1.002	1.034
Osteoporosis	287,838 (18.38)	0.39 (0.84)	1.80 (0.85)	0.949	0.968
All-Cause Pneumonia	185,766 (11.86)	0.31 (0.71)	1.44 (0.86)	1.933	1.570
Parkinson’s Disease	130,991 (08.36)	0.16 (0.60)	1.87 (0.93)	1.305	1.458
Arthritis	778,355 (49.69)	1.08 (1.15)	1.92 (0.85)	0.902	0.891
Stroke	270,805 (17.29)	0.52 (0.93)	1.71 (0.89)	1.252	1.121

Note. Bolded conditions became available in the 30-condition version of the CCW, while not available in the 27-condition version. CCW = Chronic Condition Data Warehouse; COPD = Chronic Obstructive Pulmonary Disease; NA = not available when elastic net model did not select the comorbidity as a substantial factor for predicting mortality.

Duration is time since initial diagnosis and was coded as zero for those who did not have the chronic condition. ^b Only beneficiaries with the chronic condition were included in the analysis. ^c Results were from training data (n = 1,096,451); traditional p-values are not directly available in elastic net model.

Table 2.

Frequency Analysis and Odds Ratio of Mortality for 38 AHRQ Elixhauser Comorbidities for Medicare Beneficiaries With Dementia (N = 1,566,359).

Comorbidity	N (%)	Odds ratio^a
Comorbidity	N (%)	30-day Mortality	1-year Mortality
HIV/AIDS	1,523 (0.10)	NA	NA
Alcohol abuse	30,586 (1.95)	NA	NA
Anemia due to other nutritional deficiencies	316,215 (20.19)	1.179	1.122
Autoimmune conditions	49,913 (3.19)	NA	NA
Chronic blood loss anemia	23,125 (1.48)	NA	NA
Lymphoma	13,396 (0.86)	1.111	1.097
Leukemia	8,548 (0.55)	1.109	1.169
Metastatic cancer	17,382 (1.11)	2.939	3.026
Solid tumor without metastasis, in situ	5,515 (0.35)	NA	0.940
Solid tumor without metastasis, malignant	95,600 (6.10)	1.084	1.096
Cerebrovascular disease	183,019 (11.68)	NA	0.989
Heart failure	248,027 (15.83)	1.427	1.397
Coagulopathy	79,298 (5.06)	1.170	1.127
Dementia	760,350 (48.54)	1.446	1.343
Depression	250,932 (16.02)	NA	NA
Diabetes without chronic complications	279,746 (17.86)	NA	NA
Diabetes with chronic complications	222,574 (14.21)	1.095	1.097
Drug abuse	14,746 (0.94)	NA	NA
Hypertension, complicated	347,019 (22.15)	1.065	NA
Hypertension, uncomplicate	793,841 (50.68)	0.846	0.800
Liver disease, mild	45,115 (2.88)	NA	NA
Liver disease, moderate to severe	7,861 (0.50)	1.603	1.731
Chronic pulmonary disease	249,986 (15.96)	1.139	1.167
Neurological disorders affecting movement	99,719 (6.37)	1.127	1.245
Other neurological disorders	216,970 (13.85)	1.569	1.279
Seizures and epilepsy	70,412 (4.50)	1.103	1.127
Obesity	110,847 (7.08)	0.999	0.844
Paralysis	65,974 (4.21)	1.457	1.310
Peripheral vascular disease	158,217 (10.10)	1.025	NA
Psychoses	73,953 (4.72)	1.120	NA
Pulmonary circulation disease	56,044 (3.58)	1.234	1.216
Renal failure, moderate	232,867 (14.87)	1.046	1.041
Renal failure, severe	58,740 (3.75)	1.438	1.575
Hypothyroidism	286,245 (18.27)	0.995	0.969
Other thyroid disorders	36,790 (2.35)	0.968	0.923
Peptic ulcer disease x bleeding	22,867 (1.46)	NA	0.992
Valvular disease	156,881 (10.02)	NA	NA
Weight loss	122,360 (7.81)	2.057	1.575

Note. Bolded comorbidities or specific categories became available in the 38 AHRQ Elixhauser comorbidities, while not in traditional 31 or 30 Elixhauser comorbidities. NA = not available when elastic net model did not select the comorbidity as a substantial factor for predicting mortality; AHRQ = Agency for Healthcare Research and Quality.

Results were from training data (n = 1,096,451); traditional p-values are not directly available in elastic net model.

Effects of the 30 Chronic Conditions and 38 Elixhauser Comorbidities on Mortality

Table 1 highlights the effects of the 30 CCW chronic conditions on 30-day and 1-year mortality. Non-Alzheimer dementia had a stronger positive effect on mortality outcomes compared to Alzheimer’s disease. In addition to non-Alzheimer’s dementia, the conditions most strongly associated with increased 30-day mortality risk were all-cause pneumonia (odds ratio [OR] = 1.93), lung cancer (OR = 1.88), and hip/pelvic fracture (OR = 1.46). For 1-year mortality, the highest increased risks were observed for lung cancer (OR = 1.93), all-cause pneumonia (OR = 1.57), and Parkinson’s disease (OR = 1.46). All three new conditions in the 30 CCW conditions (all-cause pneumonia, Parkinson’s disease, and urologic cancer) were positively associated with 30-day and 1-year mortality. Some conditions were positively associated with mortality outcomes, while other conditions were negatively associated or not linked to the outcomes.

Based on Table 2, the Elixhauser comorbidities associated with the highest risk of 30-day mortality were metastatic cancer (OR = 2.939), weight loss (OR = 2.057), and moderate-to-severe liver disease (OR = 1.603). For 1-year mortality, the comorbidities associated with the highest risk were metastatic cancer (OR = 3.026), moderate-to-severe liver disease (OR = 1.731), severe renal failure (OR = 1.575), and weight loss (OR = 1.575). Some new comorbidities in the 38 AHRQ Elixhauser comorbidities (e.g., leukemia and autoimmune conditions not in the traditional 31/30 Elixhauser comorbidities) were associated with an increased risk of mortality (e.g., 30-day mortality: leukemia [OR = 1.109]). Some newly defined comorbidity categories reflecting the severity of illness had a differential impact on mortality outcomes (e.g., moderate to severe liver disease was positively associated with mortality risk, while mild liver disease showed no association). The elastic net models select variables that have substantial regression coefficients in relation to outcomes, rather than focusing on statistical significance.

Predictive Power and Calibration

Table 3 presents the predictive performance of each model. All models performed well, with all C-statistics ≥0.700, except for the 1-year mortality model using the 38 Elixhauser comorbidities (C-statistics = 0.696). Model performance, as measured by C-statistics, was similar between the CCW and Elixhauser models, regardless of whether condition duration was included; therefore, we focused on models without condition duration.

Table 3.

Model Predictive Performance and Calibration Metrics.^a

Metric	30-day Mortality	60-day Mortality	180-day Mortality	1-year Mortality
	30 CCW chronic conditions
C-statistics	0.722	0.713	0.701	0.700
Brier Score	0.023	0.040	0.091	0.142
E₉₀	0.004	0.006	0.010	0.012
E_avg	0.002	0.004	0.006	0.007
	30 CCW chronic conditions + condition duration
C-statistics	0.728	0.717	0.702	0.701
Brier Score	0.023	0.040	0.091	0.141
E₉₀	0.005	0.006	0.011	0.014
E_avg	0.003	0.004	0.007	0.008
	38 AHRQ Elixhauser comorbidities
C-statistics	0.731	0.720	0.701	0.696
Brier Score	0.023	0.039	0.09	0.14
E₉₀	0.004	0.007	0.007	0.009
E_avg	0.002	0.003	0.005	0.006

Note. E₉₀ = 90th of absolute prediction error; E_avg = average absolute prediction error; AHRQ = Agency for Healthcare Research and Quality.

Models were developed on training data (1,096,451 beneficiaries) and validated on validation data (n = 469,908).

Model predictive power decreased slightly for long-term mortality outcomes for both multimorbidity measures. For instance, C-statistics for the CCW models were 0.722, 0.713, 0.701, and 0.700 for 30-day, 60-day, 180-day, and 1-year mortality, respectively. Figure 1 visualizes the predictive power of different models.

Figure 1.

ROC Curves and Decile Calibration Plots of Using the CCW Versus the Elixhauser Methods to Predict All-Cause Mortality in Dementia.

Figure 1 also depicts decile calibration plots, illustrating the agreement between predicted and observed outcomes across different deciles of predicted probabilities. These plots demonstrate proper calibration for all models, with strong alignment between predictions and observations across the entire range of predicted probabilities for both multimorbidity measures.

Elastic Net Versus Logistic Regression: Differences in Results

Table 4 compares the performance of elastic net models with traditional logistic regression models. Elastic net models outperformed traditional logistic regression models by identifying fewer key factors (1 to 6 fewer than traditional logistic regression) while maintaining comparable predictive power.

Table 4.

Comparison of Number of Significant Predictors and C-Statistics Between Logistic Regression and Elastic Net Models.

	Logistic regression		Elastic net model
Model	Number of predictors	C-statistics	Number of predictors	C-statistics
CCW: 30-day mortality	30	0.701	25	0.700
CCW: 1-year mortality	28	0.725	27	0.725
Elixhauser: 30-day mortality	32	0.696	26	0.696
Elixhauser: 1-year mortality	29	0.731	26	0.731

Note. Results were from training data (n = 1,096,451).

Discussion

This study evaluated the predictive performance of the 30 predefined CCW chronic conditions and 38 Elixhauser comorbidities coded in-house for all-cause mortality in dementia, using large-scale national data on Medicare beneficiaries, with a focus on examining the impact of recent updates to both measures. Given the similar and reasonable model performance of the two measures, researchers can choose either one.

Our findings support the use of both claims-based multimorbidity measures (CCW and Elixhauser) for predicting all-cause mortality in individuals with dementia. These insights can support risk stratification and guide targeted interventions. As the dementia burden continues to grow, such models can also inform personalized care planning and resource allocation (Iyer et al., 2008; Lee et al., 2014).

This study found that all three updated conditions included in the 30-condition version of the CCW—all-cause pneumonia, Parkinson’s disease, and urologic cancer—were associated with increased short- and long-term mortality risk. These findings support the added-value of the 30-condition CCW measure over the earlier 27-condition version and suggest that conditions like pneumonia and Parkinson’s disease—although not included in the list of practically useful conditions in the previous study (Goodman et al., 2013)—may warrant consideration as meaningful additions.

The study also found that certain comorbidities newly included in the Elixhauser developed by AHRQ (e.g., leukemia) have a significant impact on mortality outcomes. In addition, refined comorbidity categories that capture varying levels of disease severity (e.g., mild, moderate, and severe liver disease) were associated with differential mortality risk. Similar to the updates made in the 30-condition CCW measure, these findings highlight the improved performance of the more granular Elixhauser comorbidity measure, which includes 38 comorbidities compared to earlier versions with 30 or 31.

The predefined CCW conditions leverage a broader range of data sources, while the Elixhauser offers the advantage of being implementable in-house using available diagnostic codes. The CCW uses 2 years of claims data from multiple sources (inpatient, carrier, SNF, HOP, and HHA claims) and applies a unique algorithm to define each chronic condition. In contrast, there is no universal standard for coding Elixhauser comorbidities in community settings. In this study, the 38 Elixhauser comorbidities were identified using the only available administrative data—1 year of inpatient and carrier claims. All comorbidities were identified using a uniform algorithm based on the presence of at least one inpatient or carrier claim. Despite these methodological differences, both measures demonstrated strong predictive performance for mortality, reinforcing their value in large-scale risk stratification.

This study also incorporates elastic net regression models, which are well-suited for balancing parsimony and predictive power. Although the overall predictive performance was similar to traditional logistic regression, elastic net models identified fewer key predictors (1–6 fewer in this study), yielding slightly more parsimonious models without a loss of accuracy. Parsimonious models are particularly valuable in making predictions, where overly complex models may be difficult to implement, interpret, and translate into actionable decision-making. This emphasis on balancing simplicity and fit reflects long-standing principles in developing predictive models, as highlighted by Vandekerckhove et al. (2015), who emphasized parsimony as a guiding principle for robust and generalizable models. The ability of the elastic net to identify parsimonious models is consistent with previous findings (e.g., Liu et al., 2023, 2025), reinforcing the robustness of this approach across different contexts. Moreover, with even larger data sets, traditional logistic regression models may flag a growing number of variables as statistically significant due to increased statistical power, which can obscure meaningful associations. In contrast, elastic net models mitigate this risk by shrinking or excluding less informative predictors using a soft threshold rather than an absolute cutoff of p-values. This perspective aligns with broader critiques within the research community, notably the commentary by Amrhein et al. in Nature (Amrhein et al., 2019), which cautions against over-reliance on p-values from traditional regression models and emphasizes the need for approaches that highlight substantive relevance. Thus, elastic net models provide an adaptable and pragmatic strategy by reducing reliance on statistical significance alone and focusing on informative predictors (Vandekerckhove et al., 2015).

Our study assesses the strength and direction of the association of comorbidities with mortality risk for beneficiaries with dementia. Co-occurring conditions like metastatic cancer, weight loss, and severe liver disease, not unexpectedly, were most strongly associated with both short- and long-term all-cause mortality (Taudorf et al., 2021). The results also suggest that dementia type is associated with differential mortality risk. Specifically, mortality in non-Alzheimer dementia was higher than in Alzheimer’s disease, consistent with findings from previous research (Alzheimer’s Society, n.d.).

Limitations and Future Research

This study has limitations. Comorbidities may be underreported in administrative data compared to medical charts (Chong et al., 2011). Negative associations of some diagnoses with mortality outcomes may reflect coding biases in administrative data, where healthier patients are more likely to have minor conditions documented (Elixhauser et al., 1998; Liu et al., 2023). In addition, the documentation for the onset of the 30 CCW chronic conditions is constrained by its reliance on ICD-10 codes, which were fully implemented in 2016. Therefore, the first full calendar year used to identify these conditions is 2016 (CMS, n.d.-b), which limits the accuracy of condition duration calculations. When coding the 38 Elixhauser comorbidities, we relied solely on available inpatient and carrier claims’ ICD-10 codes, which were available for about 81.4% of the study sample, and data from other sources, such as SNFs and HHA, were unavailable. Furthermore, Elixhauser comorbidities are coded using data from a single year, underestimating total comorbidity prevalence. Similarly, CCW may have unflagged conditions due to insufficient data. Incomplete ascertainment of dementia in administrative claims data is a well-recognized limitation (Alzheimer’s Association, 2024), common to all claims-based dementia studies, and it affects both the CCW and Elixhauser measures. In addition, the specificity of dementia subtypes in administrative data is limited due to the prevalent use of non-specific labels (e.g., “dementia NOS”) in medical coding. Neither the Elixhauser nor the CCW includes an indicator specifically for “dementia NOS.”

Efforts to create risk scores for predicting mortality in dementia, such as the nationwide cohort study in Taiwan (Cheng et al., 2019), have included additional clinical conditions alongside chronic comorbidities and achieved higher predictive power. Incorporating additional types of clinical data into future claims-based models could enhance predictive accuracy. Future research should also explore the predictive value of the models for other outcomes, such as hospitalizations or emergency room visits, which are critical indicators of health care utilization.

Another limitation is that we compared the two measures using data from Medicare beneficiaries with dementia. While our findings are important, future studies should test these two measures in a broader sample of Medicare beneficiaries. Using these models in the broader Medicare population will help researchers effectively evaluate their generalizability, identify condition-specific variations in predictive performance, and determine whether certain comorbidities carry different prognostic value across subgroups of beneficiaries.

Conclusion

Using large-scale claims data, this study evaluates the predictive performance of the 30 CCW chronic conditions and 38 Elixhauser comorbidities for all-cause mortality risk among community-dwelling Medicare beneficiaries with dementia, offering insights for future population-based risk stratification. Both measures effectively predict mortality, with the highest accuracy observed for 30-day mortality and the lowest for 1-year mortality. These findings underscore the continued utility of the 30 CCW conditions and 38 Elixhauser comorbidities—including updated and additional conditions and comorbidities—as a comprehensive resource for mortality risk prediction in large administrative data sets. In addition, our findings highlight the elastic net model’s potential as a robust method for all-cause mortality prediction in large-scale claim-based analyses.

Footnotes

Appendix

Table A3.

Characteristics and Outcomes of Medicare Beneficiaries With Dementia.

Variable	Total (N = 1,566,359)
Age
M (SD)	82.35 (8.05)
Gender n (%)
Female	979,779 (62.55)
Male	586,580 (37.45)
Race n (%)
American Indian/Alaska Native	6,975 (0.45)
Asian	52,563 (3.36)
Black	146,919 (9.38)
Hispanic	113,681 (7.26)
Non-Hispanic White	1,224,368 (78.17)
Other	12,038 (0.77)
Unknown	9,815 (0.63)
Dual eligibility status n (%)
No	1,220,602 (77.93)
Yes	345,757 (22.07)
All-cause mortality outcome n (%)
30-day	37,903 (2.42)
60-day	66,955 (4.27)
180-day	168,750 (10.77)
1-year	297,074 (18.97)

ORCID iDs

Jianfang Liu

Monica O’Reilly-Jacob

Jennifer Perloff

Lusine Poghosyan

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Institute on Aging (5 R01 AG069143-02).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

References

Agency for Healthcare Research and Quality. (2022). ICD-10-CM/PCS comorbidity software. https://www.hcup-us.ahrq.gov/toolssoftware/comorbidityicd10/comorbidity_icd10.jsp

Alzheimer’s Association. (2024). 2024 Alzheimer’s disease facts and figures. https://www.alz.org/getmedia/76e51bb6-c003-4d84-8019-e0779d8c4e8d/alzheimers-facts-and-figures.pdf

Alzheimer’s Society. (n.d.). What are the average life expectancy figures for the most common types of dementia? https://www.alzheimers.org.uk/about-dementia/symptoms-and-diagnosis/how-dementia-progresses/later-stages-dementia

Amrhein

Greenland

McShane

(2019). Scientists rise up against statistical significance. Nature, 567(7748), 305–307. https://doi.org/10.1038/d41586-019-00857-9

Aneshensel

C. S.

Pearlin

L. I.

Levy-Storms

Schuler

R. H.

(2000). The transition from home to nursing home mortality among people with dementia. The Journals of Gerontology: Series B, 55(3), S152–S162. https://doi.org/10.1093/geronb/55.3.s152

Austin

P. C.

Steyerberg

E. W.

(2014). Graphical assessment of internal and external calibration of logistic regression models by using loess smoothers. Statistics in Medicine, 33(3), 517–535. https://doi.org/10.1002/sim.5941

Baldwin

L. M.

Klabunde

C. N.

Green

Barlow

Wright

(2006). In search of the perfect comorbidity measure for use with administrative claims data: Does it exist? Medical Care, 44(8), 745–753. https://doi.org/10.1097/01.mlr.0000223475.70440.07

Barnett

Mercer

S. W.

Norbury

Watt

Wyke

Guthrie

(2012). Epidemiology of multimorbidity and implications for health care, research, and medical education: A cross-sectional study. The Lancet, 380(9836), 37–43. https://doi.org/10.1016/S0140-6736(12)60240-2

Centers for Medicare & Medicaid Services. (n.d.-a). 30 CCW chronic conditions algorithms: MBSF_CHRONIC_{YYYY} file. https://www2.ccwdata.org/documents/10280/19139421/chr-chronic-condition-algorithms.pdf

10.

Centers for Medicare & Medicaid Services. (n.d.-b). Chronic Conditions Data Warehouse: White papers. https://www2.ccwdata.org/web/guest/white-papers

11.

Cheng

C. M.

Chang

W. H.

Chiu

Y. C.

Yeh

H. W.

Lin

C. C.

Hsieh

H. M.

Chen

T. L.

(2019). Risk score for predicting mortality in people with dementia: A nationwide, population-based cohort study in Taiwan with 11 years of follow-up. The Journal of Clinical Psychiatry, 80(4), e1–e8. https://doi.org/10.4088/JCP.18m12629

12.

Chong

W. F.

Ding

Y. Y.

Heng

B. H.

(2011). A comparison of comorbidities obtained from hospital administrative data and medical charts in older patients with pneumonia. BMC Health Services Research, 11, 105. https://doi.org/10.1186/1472-6963-11-105

13.

Daganzo

C. F.

Gayah

V. V.

Gonzales

E. J.

(2012). The potential of parsimonious models for understanding large scale transportation systems and answering big picture questions. EURO Journal on Transportation and Logistics, 1(1), 47–65. https://doi.org/10.1007/s13676-012-0003-z

14.

DuGoff

E. H.

Canudas-Romo

Buttorff

Leff

Anderson

G. F.

(2014). Multiple chronic conditions and life expectancy: A life table analysis. Medical Care, 52(8), 688–694. https://doi.org/10.1097/MLR.0000000000000166

15.

Elixhauser

Steiner

Harris

D. R.

Coffey

R. M.

(1998). Comorbidity measures for use with administrative data. Medical Care, 36(1), 8–27. https://doi.org/10.1097/00005650-199801000-00004

16.

ePrognosis. (n.d.). Mortality risk calculator for community-dwelling older adults with dementia. University of California, San Francisco. https://eprognosis.ucsf.edu/dementia.php

17.

Friedman

J. H.

Hastie

Tibshirani

(2010). Regularization paths for generalized linear models via coordinate descent. Journal of Statistical Software, 33(1), 1–22. https://doi.org/10.18637/jss.v033.i01

18.

Goodman

R. A.

Posner

S. F.

Huang

E. S.

Parekh

A. K.

Koh

H. K.

(2013). Defining and measuring chronic conditions: Imperatives for research, policy, program, and practice. Preventing Chronic Disease, 10, E66. https://doi.org/10.5888/pcd10.120239

19.

Hosmer

D. W.

Lemeshow

(2000). Applied logistic regression (2nd ed.). Wiley.

20.

Huang

Macheret

Gabriel

R. A.

Ohno-Machado

(2020). A tutorial on calibration measurements and calibration models for clinical prediction models. Journal of the American Medical Informatics Association, 27(4), 621–633. https://doi.org/10.1093/jamia/ocz228

21.

Huang

Y. J.

Chen

J. S.

Luo

S. F.

Kuo

C. F.

(2021). Comparison of indexes to measure comorbidity burden and predict all-cause mortality in rheumatoid arthritis. Journal of Clinical Medicine, 10(22), 5460. https://doi.org/10.3390/jcm10225460

22.

Iyer

Naganathan

McLachlan

A. J.

Le Couteur

D. G.

(2008). Medication withdrawal trials in people aged 65 years and older: A systematic review. Drugs & Aging, 25(12), 1021–1031. https://doi.org/10.2165/0002512-200825120-00004

23.

Kuhn

(2008). Building predictive models in R using the caret package. Journal of Statistical Software, 28(5), 1–26. https://doi.org/10.18637/jss.v028.i05

24.

Lee

S. J.

Leipzig

R. M.

Walter

L. C.

(2014). Lag time to benefit for preventive therapies—Reply. JAMA, 311(15), 1567–1568. https://doi.org/10.1001/jama.2014.2325

25.

Liu

Bilazarian

Pollifrone

M. M.

Yoon

Siegel

Poghosyan

(2025). Incorporating the present-on-admission indicator to predict in-hospital mortality through Elixhauser measures: A Medicare data analysis. Medical Care, 63(12), 929–935. https://doi.org/10.1097/MLR.0000000000002192

26.

Liu

Glied

Yakusheva

Siegel

Bilazarian

Poghosyan

Yoon

(2023). Using machine-learning methods to predict in-hospital mortality through the Elixhauser index: A Medicare data analysis. Research in Nursing & Health, 46(4), 411–424. https://doi.org/10.1002/nur.22322

27.

Maciejewski

M. L.

Hammill

B. G.

(2019). Measuring the burden of multimorbidity among Medicare beneficiaries via condition counts and cumulative duration. Health Services Research, 54(2), 484–491. https://doi.org/10.1111/1475-6773.13124

28.

Mank

van Maurik

I. S.

Rijnhart

J. J. M.

Bakker

E. D.

Bouwman

F. H.

Lemstra

A. W.

van der Flier

W. M.

Scheltens

Sikkes

S. A. M.

(2022). Development of multivariable prediction models for institutionalization and mortality in the full spectrum of Alzheimer’s disease. Alzheimer’s Research & Therapy, 14, 110. https://doi.org/10.1186/s13195-022-01053-0

29.

Mitchell

T. M.

(1997). Machine learning. McGraw-Hill.

30.

Quan

Sundararajan

Halfon

Fong

Burnand

Luthi

J.-C.

Saunders

L. D.

Beck

C. A.

Feasby

T. E.

Ghali

W. A.

(2005). Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Medical Care, 43(11), 1130–1139. https://doi.org/10.1097/01.mlr.0000182534.19832.83

31.

Research Data Assistance Center. (n.d.-a). Fee list for RIFs: Physical research data request. https://resdac.org/sites/datadocumentation.resdac.org/files/2024-05/CMS%20Fee%20List%20for%20Physical%20Research%20Data%20Request_0.pdf

32.

Research Data Assistance Center. (n.d.-b). Master beneficiary summary file (MBSF) base A/B/C/D. https://resdac.org/cms-data/files/mbsf-base

33.

Song

Intrator

Twersky

Davagnino

Kinosian

Wieland

(2021). Utilization and cost effects of the VHA Caring for Older Adults and Caregivers at Home (COACH) Program. Medical Care Research and Review, 78(6), 736–746. https://doi.org/10.1177/1077558720929592

34.

Steyerberg

(2009). Clinical prediction models: A practical approach to development, validation, and updating. Springer.

35.

Stirland

L. E.

Choate

Zanwar

P. P.

Zhang

Watermeyer

T. J.

Valletta

Torso

Tamburin

Saeed

Ridgway

G. R.

Moukaled

Lusk

J. B.

Loi

S. M.

Littlejohns

T. J.

Kuźma

James

S. N.

Grande

Foote

I. F.

Cousins

K. A. Q.

. . . Suryadevara

(2025). Multimorbidity in dementia: Current perspectives and future challenges. Alzheimer’s & Dementia, 21(8), e70546. https://doi.org/10.1002/alz.70546

36.

Stone

(1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B (Methodological), 36(2), 111–133.

37.

Suls

Bayliss

E. A.

Berry

Holtrop

J. S.

Lonergan

McQuillan

Shetterly

West

Glasgow

R. E.

(2021). Measuring multimorbidity: Selecting the right instrument for the purpose and the data source. Medical Care, 59(8), 743–756. https://doi.org/10.1097/MLR.0000000000001566

38.

Taudorf

Nørgaard

Brodaty

Laursen

T. M.

Waldemar

(2021). Dementia increases mortality beyond effects of comorbid conditions: A national registry-based cohort study. European Journal of Neurology, 28(7), 2174–2184. https://doi.org/10.1111/ene.14875

39.

Thompson

N. R.

Fan

Dalton

J. E.

Jehi

Rosenbaum

B. P.

Vadera

Griffith

S. D.

(2015). A new Elixhauser-based comorbidity summary measure to predict in-hospital mortality. Medical Care, 53(4), 374–379. https://doi.org/10.1097/MLR.0000000000000326

40.

Tibshirani

(1996). Regression shrinkage and selection via the lasso. Journal of the Royal Statistical Society: Series B (Methodological), 58(1), 267–288.

41.

Van Calster

McLernon

D. J.

van Smeden

Wynants

Steyerberg

E. W

. (2019). Calibration: The Achilles heel of predictive analytics. BMC Medicine, 17, 230. https://doi.org/10.1186/s12916-019-1466-7

42.

Vandekerckhove

Matzke

Wagenmakers

E. J.

(2015). Model comparison and the principle of parsimony. In Busemeyer

J. R.

Wang

Townsend

J. T.

Eidels

(Eds.), The Oxford handbook of computational and mathematical psychology (pp. 300–319). Oxford University Press. https://doi.org/10.1093/oxfordhb/9780199957996.013.15

43.

Xin

Zhang

Jiang

(2025). Association between multimorbidity and the risk of dementia: A systematic review and meta-analysis. Archives of Gerontology and Geriatrics, 131, 105760. https://doi.org/10.1016/j.archger.2025.105760

44.

Mishra

G. D.

Jones

(2017). Evidence on multimorbidity from definition to intervention: An overview of systematic reviews. Ageing Research Reviews, 37, 53–68. https://doi.org/10.1016/j.arr.2017.05.003

45.

Zhou

You

Zhang

Yuan

(2025). Multimorbidity and risk of dementia: A systematic review and meta-analysis of longitudinal cohort studies. The Journal of Prevention of Alzheimer’s Disease, 12(7), 100164. https://doi.org/10.1016/j.tjpad.2025.100164

46.

Zou

Hastie

(2005). Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 67(2), 301–320. https://doi.org/10.1111/j.1467-9868.2005.00503.x

Predicting All-Cause Mortality Using Two Claims-Based Measures in Medicare Beneficiaries With Dementia

Abstract

Keywords

Introduction

The CCW

The Elixhauser Comorbidity Index

Machine Learning Models for Use With Claims Data—Elastic Net Models

Purpose of the Study

Method

Data Source and Variables

Outcome Measures and Beneficiary Baseline Characteristics

The 30 Predefined Chronic Conditions From the Chronic Condition Data Warehouse

The 38 Elixhauser Comorbidities Developed by the Agency for Healthcare Research and Quality

Data Analyses

Results

Effects of the 30 Chronic Conditions and 38 Elixhauser Comorbidities on Mortality

Predictive Power and Calibration

Elastic Net Versus Logistic Regression: Differences in Results

Discussion

Limitations and Future Research

Conclusion

Footnotes

Appendix

ORCID iDs

Funding

Declaration of Conflicting Interests

References