Beyond Comorbidity: Expanding the Definition and Measurement of Complexity Among Older Adults Using Administrative Claims Data

Chrischilles, Elizabeth MS, PhD*; Schneider, Kathleen PhD†,‡; Wilwert, June BA, RN; Lessman, Gregory MSc; O’Donnell, Brian PhD; Gryzlak, Brian MS*; Wright, Kara MS*; Wallace, Robert MD, MS*

Medical Care:
doi: 10.1097/MLR.0000000000000026
Original Articles

Background: Studies of patients with multiple chronic conditions using claims data are often missing important determinants of treatments and outcomes, such as function status and disease severity. We sought to identify and evaluate a class of function-related indicators (FRIs) from administrative claims data.

Population: The study cohort comprised US Medicare beneficiaries aged 65 years or older with Parts A and B fee-for-service and Part D coverage, with a hospitalization for acute myocardial infarction during 2007.

Methods: Measures during the year before admission included the FRIs, demographics, conventional comorbidity measures, and prior hospitalization. Outcomes were receipt of cardiac catheterization during the index hospitalization and 12-month mortality. Model development used a random sample (n=72,056) with an equal sample for validation.

Results: In addition to prior cardiovascular conditions (85%), 40% had ≥1 comorbid condition, 30% were hospitalized in the prior 6 months, and 65% had ≥1 FRI [eg, delirium/dementia (22.7%), depression (16.7%), mobility limitation (16.1%), and chronic skin ulcers (12.6%)]. Including the FRIs improved mortality and cardiac catheterization prediction models (C-statistics 0.71 and 0.77, respectively). Patients with more cardiovascular conditions received less cardiac catheterization [minimally adjusted odds ratio (OR) 0.83; 95% confidence interval (CI), 0.82–0.83], as did patients with more comorbidities (minimally adjusted OR 0.70; 95% CI, 0.69–0.71), but this was attenuated by adjusting for functional status (fully adjusted OR for cardiovascular conditions 0.95; 95% CI, 0.94–0.96 and for comorbid conditions 0.94; 95% CI, 0.92–0.95).

Conclusions: Claims data studies that include indicators of potentially diminished patient functional status better capture heterogeneity of patients with multiple chronic conditions.

Author Information

*Department of Epidemiology, The University of Iowa College of Public Health, Iowa City

Schneider Research Associates LLC

Buccaneer, a General Dynamics Company, Des Moines, IA

Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's Website,

Supported by grant number R24HS019440 from the Agency for Healthcare Research and Quality. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality.

Reprints: Elizabeth A. Chrischilles, MS, PhD, College of Public Health, The University of Iowa, 105 River Street, Iowa City, IA 52242. E-mail:

Article Outline

Patient health complexity is often defined on the basis of having multiple chronic condition (MCC) diagnoses. Established comorbidity indexes such as the Charlson Comorbidity Index1 and the Elixhauser conditions2 may be repurposed for complex patient research. When used with Medicare administrative claims data, comorbidity indexes identify codes for specific diagnoses (eg, diabetes, chronic obstructive pulmonary disease, heart failure, and peripheral vascular disease) from healthcare provider reimbursement claims. However, claims data studies have important weaknesses because they often do not include important determinants and confounders of health outcomes such as mortality. Perhaps the most important omissions are measures of functional status, an important component of illness severity.3

A broad conceptualization of illness severity includes the underlying diagnosis, comorbidities, and physical, cognitive, and emotional functional status of a patient.4 Related concepts include frailty, geriatric syndromes, and serious illness, which are associated with adverse outcomes in older adults above and beyond the traditional comorbidity indices.5–11 Yourman et al12 recently reviewed 16 prognostic indices for older adults. Across all of the indices, the most common final predictors of mortality included functional status and comorbidities (both present in 12 or more of the indices). The data sources for these indices varied, with many using physical assessment or patient-reported characteristics as predictors. One index incorporated function-related indicators (FRIs) recorded in medical records for 1246 hospitalized elders in 27 Connecticut hospitals: walking impairment, dementia diagnoses, urinary incontinence, and depression.13 Three indices used data from computerized administrative data only. As an example, Gagne et al14 combined medical diagnosis codes from 2 well-known comorbidity indices in a claims database of Pennsylvania Medicare enrollees, and this resulted in higher C-statistics from models predicting mortality than did either comorbidity index alone. None of the 3 indices that evaluated computerized administrative data alone included FRIs.

The objective of this study was to explore the use of a new class of patient FRIs, derived from claims records, to better control for the confounding effects of dysfunction and disability in predicting patient outcomes. If the FRI measure can be demonstrated to improve the assessment of illness severity in claims data beyond using traditional demographic and comorbidity measures, it might serve as a useful stratification factor when assessing the effectiveness of treatments or performance of healthcare facilities and systems. To test the value of the FRIs in controlling for confounders, we used data from hospitalized acute myocardial infarction (AMI) patients and 2 cardiovascular disease (CVD)-related outcomes: receipt of cardiac catheterization during the index hospitalization (an important measure of healthcare utilization), and 1-year all-cause mortality. Hospitalization for AMI is a useful setting for this evaluation because unmeasured functional capacity is thought to potentially explain the observation that AMI patients with MCC are less likely to be prescribed evidence-based therapies for their CVD (also known as the “treatment-risk paradox”).3 If the FRI measure is valid, patients with more FRIs should have higher mortality and have lower rates of catheterization (catheterization is less likely to be attempted if a patient’s function status is judged to be poor). Assuming that part of the explanation for treatment-risk paradox among patients with MCC is because some individuals with MCC have poor function status, adding the FRI measure to prediction models should also attenuate the relationships of MCC with mortality and catheterization.

Back to Top | Article Outline


Study Population

We used nationwide Medicare administrative claims data from the Chronic Condition Data Warehouse (CCW)15 The CCW is populated with CMS Medicare administrative enrollment and claims data for all Medicare beneficiaries. Enrollment data are available for 100% of Medicare beneficiaries, and fee-for-service Medicare Part A and B claims, and all Part D event data are included. Our cohort consisted of beneficiaries hospitalized with AMI in 2007; they were identified by claims with an International Classification of Diseases, Ninth Revision (ICD-9)16 code in the first or second position on an inpatient claim for the subset of codes that included 410.x1. The start date for the inpatient claim had to have been in 2007 (3304 beneficiaries whose start date was in 2006 and discharge date was in 2007 were excluded), and the diagnosis code must have appeared on a claim for a critical access hospital or acute care hospital, rather than some other type of inpatient facility (367 beneficiaries who only had a claim in an additional inpatient facility type, such as an inpatient rehabilitation hospital or long-term care hospital, were excluded). When the index inpatient claim was followed without a gap by another inpatient claim, these were considered to represent 1 index hospitalization. The admission date for this index hospitalization was the cohort start date used to demarcate the predictor variables measured during the year before admission and the outcome variables measured either during the index hospitalization (receipt of catheterization) or during 12 months after admission (all-cause mortality). Patients were required to have Part A and B fee-for-service coverage for all 12 months before the index hospitalization. There was no posthospitalization enrollment requirement, and data for the mortality outcome were obtained from Medicare enrollment data. One person with an invalid date of death (ie, died before the index AMI admission and had no record in the 2008 Beneficiary Summary File) was excluded. Finally, 23,308 patients with AMI were excluded as they were less than 65 years of age at the time of the index. In addition, because this was part of a larger study to evaluate the comparative effectiveness of secondary prevention medications, all subjects had 1 or more months of Part D coverage. The final AMI cohort size was 144,112.

Back to Top | Article Outline
Outcome Variables

Outcome variables were receipt of cardiac catheterization during the index hospitalization (yes/no) and 12-month all-cause mortality. Attempted cardiac catheterization was defined by determining whether the claims for the index hospitalization indicated that the patient used the cardiac catheterization laboratory. A single revenue center code on the claim(s) for the index hospitalization was used (0481). All-cause mortality was defined from dates of death in Medicare enrollment data.

Back to Top | Article Outline
Predictor Variables

Predictor variables included demographic measures, cardiovascular conditions, comorbidities, previous hospitalization, and the FRIs (defined below). Medicare claims for the 1-year period before the index hospitalization date were used for the medical conditions, and previous acute or critical access inpatient hospitalization for any reason was recorded from the 6 months before the index hospitalization.

Demographic variables included age, sex, and race. We used Medicare enrollment data from the CCW15 Beneficiary Summary File to calculate beneficiary age at the time of the index hospitalization, and created a dichotomous indicator of very old age (85+ or not). We also included beneficiary sex and race code from the Social Security records, which we recoded to 3 categories: White, African American, or Other/unknown.

Cardiovascular conditions, including coronary heart disease risk equivalents, were previous AMI, stroke, transient ischemic attack, heart failure, arrhythmia, peripheral vascular disease, chronic kidney disease, diabetes, hypertension, hyperlipidemia, ischemic heart disease, and a history of revascularization. Algorithms used for these conditions are indicated in Table 1.

Comorbidity measures included all conditions from the Charlson and Elixhauser comorbidity measures2 (excluding cardiovascular conditions); when a condition was included in both lists we chose whichever was more broadly defined. For each condition, the algorithm that was used is indicated in Table 1.

The FRIs, as noted in the introduction, were developed out of a need to identify a set of variables within claims databases that could be applied to older people for the purpose of controlling for the confounding effects of dysfunction and disability in outcomes, comparative effectiveness, and other types of health services research. Such variables are normally not represented in studies using administrative data. We selected relatively common chronic, disabling conditions of older people that were not present in the commonly used comorbidity indices. These were identified using interviews with geriatricians who manage older people with MCC and/or disability. We also examined the literature for each of the candidate conditions to determine whether the condition could be identified using claims data. In addition to identifying conditions using patient diagnoses, we further selected indicators of dysfunction and disability that were represented in therapeutic codes (eg, chronic oxygen use) or in purchases of durable equipment (eg, wheelchairs). These have face validity as FRIs, recognizing that not all persons with dysfunctions use them. These FRIs and their respective ICD-9, HCPCS codes, and revenue center codes are listed in Table 2.

Patient FRIs were derived using Medicare claims for the 1-year period before the index hospitalization. We examined frequency distributions for each of the following conditions: mobility limitations (defined by claims for cane, walker, wheel chair, hospital bed, etc.), blood transfusion, use of oxygen, supplemental nutrition, hip or pelvic fracture, chronic skin ulcer, pneumonia, delirium/dementia/Alzheimer disease, bone marrow failure/agranulocytosis, depression, use of urinary catheter, respiratory failure/insufficiency/arrest, sepsis, and malnutrition/unintentional weight loss, fall-related injury, and syncope (Table 2).8,10,11,17 All conditions were coded as binary variables.

Back to Top | Article Outline
Statistical Analyses

For the development phase of this study we used a split sample methodology with a random 50% sample from the AMI cohort (n=72,056). Models were estimated using logistic regression in SAS v9.3 software, for the following outcomes: receipt of cardiac catheterization during the index hospitalization (yes/no), and 12-month mortality. For the 2 outcome variables, the model included age, sex, race, prior cardiovascular conditions, comorbidities, prior hospitalization, and the FRIs. Predictors were specified a priori. Beginning with all predictors in the model, demographic and cardiovascular conditions were not allowed to leave the models, whereas comorbidities, FRIs, and the indicator of prior hospitalization were eliminated if their P-value in both outcome models was ≥0.20.18 This threshold was set at a fairly high level for removal as the RFIs were selected because they had face validity.

Once the reduced models were in place, we also created summary variables that counted the number of conditions, to help evaluate the cumulative effect of multiple conditions: (1) number of cardiovascular conditions (0–11); (2) number of comorbid conditions (0–10); and (3) number of FRIs (0–13). These 3 variables were continuous. For the 2 outcome variables, we constructed separate minimally adjusted (ie, for age, sex, race) logistic regression models to examine the explanatory power of each of the summary measures and prior hospitalization. Next, we constructed fully adjusted models, including all summary measures. Adjusted odds ratios (ORs), 95% confidence intervals (CIs), and C-statistics were calculated for all models.

For validation, the coefficients from the developmental model were used to score the validation sample, which consisted of the remainder of the population (n=72,056). Receiver operating characteristic (ROC) curves and C-statistics were examined for the validation models. In addition, the scoring generated predicted probabilities of each outcome for the beneficiaries in the validation sample,19 and the accuracy of the models was evaluated using methods similar to those used by Levine et al.20 First, risk of death for each subject was estimated on the basis of the final logistic regression model in the development cohort. Similarly, probability of cardiac catheterization was estimated. The subjects were next divided into quartiles of these probabilities. The predictive accuracy of the logistic regression model was then determined by comparing predicted and observed mortality and cardiac catheterization percentages in the validation cohort.

Back to Top | Article Outline


Developmental Study

Table 1 displays the characteristics of the development and validation samples. Comorbid conditions were common; 85.4% of the population had a prior cardiovascular condition and 40% had 1 or more noncardiovascular condition. Nearly 65% had 1 or more indicator of potentially diminished functional status, most commonly: delirium or dementia (22.7%), fall-related injury (21.9%), pneumonia (17.9%), depression (16.7%), mobility limitation (16.1%), chronic skin ulcers (12.5%), malnutrition or unintentional weight loss (11.5%), oxygen use (11%), and respiratory failure (10.6%). Thirty percent had a hospitalization in the 6 months before the index hospitalization.

The outcomes we examined were also common. Slightly fewer than half of the beneficiaries received cardiac catheterization during the index AMI hospitalization [33.898 (47.0%)] and 38% died within 12 months of the index hospitalization [27,352 (37.96%)]. Table 3 displays the multivariate outcome model results. All but 7 of the comorbid conditions and FRIs were retained at an elimination P-value of P≥0.20. The strongest associations were observed for old age, history of heart failure, chronic kidney disease, metastatic solid tumor, delirium/dementia, oxygen use, and bone marrow failure (Table 3).

When the summary measures were evaluated (Table 4), any of the summary measures alone (models 1–4) resulted in a reasonable model as indicated by C-statistics in the 0.66–0.77 range. People with a greater number of cardiovascular covariates (OR=0.83; 95% CI, 0.82–0.83), other comorbid conditions (OR=0.70; 95% CI 0.69–0.71), FRIs (OR=0.69; 95% CI, 0.69–0.70), or a prior hospitalization (OR=0.43; 95% CI, 0.41–0.45) were significantly less likely to receive cardiac catheterization during the index AMI. When all of the summary measures were placed together in the model to fully adjust for all the factors (model 5), the count of FRIs was most strongly associated with catheterization, and ORs for the remaining measures were substantially attenuated. Results were similar, but ORs were in the opposite direction, for 12-month mortality (Table 4). After adjustment for demographics and cardiovascular and comorbid conditions in the fully adjusted regression (model 5), for each additional FRI, the odds of receiving a cardiac catheterization decreased by 26% and the odds of dying within 1 year increased by 28%.

Back to Top | Article Outline
Validation Study

The validation sample consisted of the remaining AMI cohort (n=72,056) after removal of the 72,056 beneficiaries used for model development. The ROC curves and corresponding C-statistics for the models from Tables 3 and 4 were identical when applied in the validation sample. ROC curves for predicting cardiac catheterization in an index hospitalization for an acute myocardial infarction, and for 12-month mortality after the admission, are displayed for the developmental and validation samples in the (Supplemental Digital Content 1,; Supplemental Digital Content 2,; Supplemental Digital Content 3,; Supplemental Digital Content 4,

To further examine the fit of the fully specified models in Table 4, quartiles of predicted probabilities were estimated and observed proportions were tabulated for each quartile (Table 5). In the developmental cohort, observed cardiac catheterization percentages ranged from 16% in the lowest probability quartile to 77% in the highest quartile. Similar results were found in the validation cohort, with the lowest and highest probability groups having 16% and 76% cardiac catheterization, respectively. Twelve-month mortality ranged from 17% to 62% in the developmental cohort and 17% to 61% in the validation cohort.

Back to Top | Article Outline


In this study, we sought to determine whether an expanded definition of illness severity with functional status implications would better capture patient heterogeneity using claims data than relying on common comorbidity measures alone. Including indicators of potentially diminished functional capacity, present in two thirds of patients before an index hospitalization for an AMI, resulted in improved discrimination (better C-statistics) in predictive models for receipt of cardiac catheterization during hospitalization and 12-month mortality (Table 4; model 5 vs. the minimally adjusted models 1–4). There was good calibration in the models, with close agreement between observed and predicted probabilities across risk quartiles. Strengths of the study included its large sample size, that the model performed well for >1 outcome, and that the administrative data required are readily available in most healthcare systems and insurer databases. The FRIs are not disease specific and future work may establish their utility for noncardiovascular conditions.

Model fit was in the range reported for other prognostic indices. In a recent systematic review of prognostic indices for mortality, 10 indices had C-statistics between 0.70 and 0.79 and none of the C-statistics was >0.9.14 In addition, this level of discriminatory ability is comparable to that of other widely used indices for decision making in cardiovascular care, for example, the Framingham9 coronary heart disease prediction score (C-statistic=0.63–0.83)21 and the Thrombolysis in Myocardial Infarction risk score for unstable angina/non-ST elevation myocardial infarction (C-statistic=0.65).22

The multivariate models increased our understanding of individual predictors. In addition to large influences of metastatic cancer, advanced age, and cardiovascular risk factors, 11 of 16 FRIs and 6 of 22 conventional comorbidities were significantly associated with mortality. Eleven FRIs and 5 comorbidities were associated with catheterization rates.

Diminished functional status is often a result of other underlying diseases and comorbidities. Inclusion of the FRIs resulted in substantially attenuated ORs for cardiovascular and other comorbidity measures in both mortality and cardiac catheterization models. Confounding by severity of illness23 is a recognized limitation in treatment effectiveness studies using healthcare claims data. The FRI measure may help control for this bias and result in more accurate estimates of treatment impact.

Patients 85 years of age or older (29.4% of patients in this study) were much less likely to have a cardiac catheterization and much more likely to die in the 12 months after the AMI. ORs for old age were not appreciably attenuated by adjustment for any of the measures. It is not surprising that age is strongly associated with mortality, and prognostic models routinely include age adjustment.12,23 For instance, the age 85 years and older places an individual in the highest-risk category for mortality or functional decline on the Vulnerable Elders Survey.24 Although prognostic indices offer a theoretical means of moving beyond decisions based on age alone, the large independent contribution of age itself to these models underscores that there remains considerable uncertainty about what factors determine treatment decisions and prognosis among the oldest patients.

Similar to previously published prognostic indices, the C-statistics for the 12-month mortality models were not sufficient to use as the basis for clinical decisions for individual patients. The 12-month mortality model may be most useful in sorting patients into groups of risk. The hospital admission is an important opportunity for recognizing patients who are at risk for further decline or death. This need not imply a choice between palliation and cure. Although it was not an objective of this study, another possible use of the FRIs would be to prospectively identify patients who have a reasonable chance of dying. This is compatible with a broad view of palliative care in which the purpose is to provide aggressive symptomatic and supportive care irrespective of whether an individual also pursues active or curative treatment for a disease.25

The number of FRIs ranged from 0 to 13, and the change in risk associated with each additional FRI was substantial. Risk stratification has gained considerable importance in CVD with a goal of individualizing disease prevention.26 Model development has focused on finding new biomarkers with which to improve the ability to predict cardiovascular events. Most guidelines categorize all patients with existing coronary heart disease or with coronary heart disease equivalents as high risk according to these models. However, many clinical decisions in cardiovascular care are also influenced by life expectancy.12 Although healthcare claims data cannot measure key components of individuals’ CVD risk (eg, smoking status, serum lipids), risk stratification for the purpose of examining whether treatment effectiveness varies with life expectancy may be useful. The FRI measure provides additional information with which to create such strata.

Inclusion of FRIs in prognostic models may also help understand why patients who seem to be most in need of a treatment are less likely to receive it. For instance, patients with more cardiovascular risk factors and those with more comorbidities were less likely to receive a cardiac catheterization, but this treatment-risk paradox was substantially attenuated by adjusting for indicators of potentially diminished functional status. These results are consistent with findings from the APPROACH study3 suggesting that treatment-risk paradoxes in CVDs are attributable to baseline imbalances in functional status. Interventions to address treatment-risk paradox should recognize that patients with more advanced illness are less likely to receive cardiovascular treatments. Although many of the patients in our study already had CVD and thus had previously been evaluated, having a cardiac catheterization was a useful example of in-hospital utilization that also has bearing on whether patients will receive invasive cardiovascular treatments. Future research should be directed toward determining whether the benefits of noninvasive life-extending cardiology interventions such as statin therapy extend to these patients.

Back to Top | Article Outline


An expanded concept of general illness severity that incorporates indicators of potentially diminished functional status can identify older patients who are unlikely to receive a cardiac catheterization and who are at high risk of mortality in the year after hospitalization for an AMI. The advantage of the FRIs measure is that it uses claims data and can be readily implemented in the same manner as conventional comorbidity indices.

We conclude that investigators using administrative claims data for studying healthcare utilization and health outcomes among older adults should include an expanded set of conditions and characteristics to better capture patient complexity. Indicators of potentially diminished functional status should be included in risk-adjustment models when studying performance of healthcare systems providing care for older adults. Future work is needed to validate the utility of this approach with other patient populations.

Back to Top | Article Outline


1. Klabunde CN, Potosky AL, Legler JM, et al..Development of a comorbidity index using physician claims data.J Clin Epidemiol.2000;53:1258–1267.
2. Quan H, Sundararajan V, Halfon P, et al..Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.Med Care.2005;43:1073–1077.
3. McAlister FA, Oreopoulos A, Norris CM, et al..Exploring the treatment-risk paradox in coronary disease.Arch Intern Med.2007;167:1019–1025.
4. Smith MA, Nitz NM, Stuart SKKane RL.Severity and comorbidity.Understanding Health Care Outcomes Research.2006.Sudbury:Jones and Bartlett.
5. Rockwood K, Mitnitski A.Frailty defined by deficit accumulation and geriatric medicine defined by frailty.Clin Geriatr Med.2011;27:17–26.
6. Fried LP, Tangen CM, Wlaston J, et al..Frailty in older adults: evidence for a phenotype.J Gerontol.2001;56A:M146–M156.
7. Afilalo J, Karunananthan S, Eisenberg MJ, et al..Role of frailty in patients with cardiovascular disease.Am J Cardiol.2009;103:1616–1621.
8. Hastings SN, Purser JL, Johnson KS, et al..A frailty index predicts some but not all adverse outcomes in older adults discharged from the emergency department.J Am Geriatr Soc.2008;56:1651–1657.
9. Cigolle CT, Ofstedal MB, Tian Z, et al..Comparing models of frailty: The Health and Retirement Study.J Am Geriatr Soc.2009;57:830–839.
10. Steinman MA, Lund BC, Miao Y, et al..Geriatric conditions, medication use, and risk of adverse drug events in a predominantly male, older veteran population.J Am Geriatr Soc.2011;59:615–621.
11. Desai MM, Bogardus ST, Williams CA, et al..Development and validation of a risk-adjustment index for older patients: The High-Risk Diagnoses for the Elderly Scale.J Am Geriatr Soc.2002;50:474–481.
12. Yourman LC, Lee SJ, Schonberg MA, et al..Prognostic indices for older adults. A systematic review.JAMA.2012;307:182–192.
13. Inouye SK, Bogardus ST, Vitagliano G, et al..Burden of illness score for elderly persons. Risk adjustment incorporating the cumulative impact of diseases, physiologic abnormalities, and functional impairments.Med Care.2003;41:70–83.
14. Gagne JJ, Glynn RJ, Avorn J, et al..A combined comorbidity score predicted mortality in elderly patients better than existing scores.J Clin Epidemiol.2011;64:749–759.
15. .Available at: Accessed March 27, 2013.
16. .Available at: Accessed March 27, 2013.
17. Ensrud KE, Ewing SK, Taylor BC, et al..Comparison of 2 frailty indexes for prediction of falls, disability, fractures, and death in older women.Arch Intern Med.2008;168:382–389.
18. Agresti A.Categorical Data Analysis.2002:2nd ed.Hoboken, NJ:John Wiley & Sons Inc..
19. Hand DJ.Construction and Assessment of Classification Rules.1997.New York:John Wiley & Sons.
20. Levine SK, Sachs GA, Jin L, et al..A prognostic model for 1-year mortality in older adults after hospital discharge.Am J Med.2007;120:455–460.
21. D’Agostino RB Sr, Grundy S, Sullivan LM, et al..Validation of the Framingham coronary heart disease prediction scores: results of a multiple ethnic groups investigation.JAMA.2001;286:180–187.
22. Antman EM, Cohen M, Bernink PJ, et al..The TIMI risk score for unstable angina/non-ST elevation MI: a method for prognostication and therapeutic decision-making.JAMA.2000;284:835–842.
23. Brookhart MA, Sturmer T, Glynn RJ, et al..Confounding control in healthcare database research. Challenges and potential approaches.Med Care.2010;48:S114–S120.
24. Saliba D, Elliott M, Rubenstein LZ, et al..The Vulnerable Elders Survey: a tool for identifying vulnerable elders in the community.J Am Geriatr Soc.2001;49:1691–1699.
25. Lynn J.Serving patients who may die soon and their families: the role of hospice and other services.JAMA.2001;285:925–932.
26. Lloyd-Jones DM.Cardiovascular risk prediction. Basic concepts, current status, and future directions.Circulation.2010;121:1768–1777.

multiple chronic conditions; Medicare; functional status

Supplemental Digital Content

Back to Top | Article Outline
© 2014 by Lippincott Williams & Wilkins.