Objectives. We sought to determine whether chronic conditions and functional limitations are equally predictive of mortality among older adults.

Methods. Participants in the 1998 wave of the Health and Retirement Study (N=19430) were divided into groups by decades of age, and their vital status in 2004 was determined. We used multivariate Cox regression to determine the ability of chronic conditions and functional limitations to predict mortality.

Results. As age increased, the ability of chronic conditions to predict mortality declined rapidly, whereas the ability of functional limitations to predict mortality declined more slowly. In younger participants (aged 50–59 years), chronic conditions were stronger predictors of death than were functional limitations (Harrell C statistic 0.78 vs. 0.73; P=.001). In older participants (aged 90–99 years), functional limitations were stronger predictors of death than were chronic conditions (Harrell C statistic 0.67 vs. 0.61; P=.004).

Conclusions. The importance of chronic conditions as a predictor of death declined rapidly with increasing age. Therefore, risk-adjustment models that only consider comorbidities when comparing mortality rates across providers may be inadequate for adults older than 80 years.

Numerous studies have shown that both chronic conditions and functional limitations are powerful independent predictors of mortality.14 However, a growing body of research suggests that some risk factors behave differently in people at different ages.510 Some researchers have found that well-established mortality risk factors among younger persons, such as hypertension,7,8,11 hypercholesterolemia,7,8,12 increased body mass index,7,9,13,14 heart disease,5,8,9 and cancer,5,9 may not continue to pose a risk to the oldest old, suggesting that the association between chronic conditions and mortality may be weaker in the elderly. Autopsy series have also supported this notion, showing that a definitive cause of death attributable to a single disease process is often not found among older people.15 These observations have spurred a growing recognition within the geriatrics community that our methods of measuring and accounting for the burden of disease may be inappropriate for our oldest patients.1618

Despite these concerns, chronic disease diagnoses remain at the center of clinical care and risk adjustment for older patients.17 However, if the association between chronic conditions and mortality is weaker in the elderly, risk adjustment tools that rely solely on chronic disease diagnoses (such as the Charlson Comorbidity Index19 and the Elixhauser method20) may be suboptimal for our oldest old. Therefore, the use of these methods to compare risk-adjusted outcomes as a proxy for the quality of care21,22 may lead to erroneous conclusions. Improved risk-adjustment methods may lead to improvements in targeting health care quality interventions, ultimately resulting in better population health outcomes.

To address these issues, we examined the ability of specific types of risk factors—chronic conditions, functional limitations, and demographic variables—to differentiate between people at high and low risk of death across a range of age groups. Based on previous research, we hypothesized that chronic conditions would be less predictive of death among older people. Because functional limitations often represent a final common pathway of decline regardless of underlying etiology,2325 we further hypothesized that functional limitations would be a stronger predictor of mortality than chronic conditions among our oldest participants.


We studied community-dwelling participants aged 50 to 99 years who were interviewed in 1998 as part of the Health and Retirement Study (HRS). HRS was expanded in 1998 to become a representative sample of all persons in the contiguous United States older than 50 years.26,27 Data were collected by field interviewers using computer-assisted personal- and telephone-interviewing techniques. The Survey Research Center at the University of Michigan in Ann Arbor led the 7-day interviewer training and data collection efforts.28 Data were collected primarily through telephone interviews, with an overall response rate of 81%.29

A total of 20005 community-dwelling participants aged 50 to 99 years were enrolled in HRS in 1998. We excluded 575 participants (2.9% of the sample) because critical data were missing, such as 2004 vital status (548), date of death (6), or baseline functional status (21), leading to our final analytic sample of 19430 participants. Of this final sample, 3457 individuals (18%) died by 2004.


The outcome of interest was death through 2004. We identified deaths using the HRS follow-up procedures, which entailed cross-referencing HRS information with the National Center for Health Statistics National Death Index to determine vital status and month and year of death.30

Respondents provided information about chronic conditions and functional status by self-report. For chronic conditions, participants were asked, “Have you ever had, or has a doctor told you, that you have/had X?” For many chronic conditions, follow-up questions provided information on the severity of disease. For example, participants who reported a history of cancer were asked if their cancer had spread. Thus, participants with a history of cancer were classified into 1 of 2 severity groups: those with a history of localized cancer and those with a history of cancer that had spread. We examined a total of 9 conditions with up to 3 levels of disease severity encompassing 6 of the top 7 causes of death in the United States31: hypertension (absent or present), diabetes mellitus (absent, or oral medications, insulin, or kidney disease), cancer (absent, or localized or spread), chronic lung disease (absent, or present but no medications, daily medications, or oxygen dependent), heart disease (absent, angina, heart attack in past 2 years, or heart failure), stroke (never, no current deficits, or current deficits), psychiatric disease (absent or present), arthritis (absent or present), and dementia (absent or present). If a participant qualified for more than 1 level of severity for a chronic condition (e.g., insulin and oral hypoglycemic for diabetes mellitus), the higher level of severity was assigned (e.g., insulin).

Participants were also asked whether they had specific functional limitations. These included 5 activities of daily living (ADL; i.e., bathing, dressing, eating, transferring in and out of bed, and toileting), 5 instrumental activities of daily living (IADL; i.e., shopping, preparing meals, using the telephone, managing medications, and managing finances), and walking. For each ADL and IADL task, participants were divided into those who had no difficulty completing the task, those who had difficulty with the task, and those who needed help with the task. For walking, participants were divided into 5 levels of severity: “can’t walk 1 block,” “difficulty walking 1 block,” “walks 1 to several blocks,” “walks several blocks but unable to jog 1 mile,” and “able to jog 1 mile.” Previous studies have shown these functional measures to be important predictors of institutional care and death.2,3,32

Statistical Analysis

Using maximum likelihood estimation multivariable Cox proportional hazards regression, we constructed models to predict 6-year mortality among different age groups. Our base model (A) consisted of age, gender, and a quadratic age term to account for the nonlinear relationship between age and mortality.2 Our functional limitations model (B) included our base model and 5 ADL tasks, 5 IADL tasks, and walking. Our chronic conditions model (C) included our base model and 9 chronic condition measures. Our full model (D) included our base model, functional limitations, and chronic condition measures. The level of disease severity and the level of functional limitation were entered into our model with indicator (dummy) variables.

To evaluate the ability of these risk domains to predict mortality across age groups, we stratified our sample into 5 decades of age (50–59, 60–69, 70–79, 80–89, and 90–99 years). We fit each of the 4 models outlined above using participants in a single age group, for a total of 20 separate regression models. We used the Harrell concordance (C ) statistic as our primary measure of the ability of risk factors to predict mortality in a given age group and obtained 95% confidence intervals through bootstrapping with replacement (1000 repetitions).33,34 The Harrell C statistic is the probability that for any 2 individuals, the model will correctly assign a higher risk of death to the individual who dies first. This statistic is analogous to the C statistic for logistic regression and varies from 1.0 (perfect prediction) to 0.5 (prediction no better than chance). To determine the statistical significance of any trends in Harrell C statistics across age groups, we used a nonparametric permutation test (100 000 repetitions) to determine whether the slope of a linear regression line between age groups and the Harrell C statistic equaled zero. For each age group, we tested the null hypothesis that the Harrell C statistic was equal for the functional limitations and chronic conditions models through bootstrapping with replacement (1000 repetitions).35

To test the robustness of our findings, we repeated our analysis with different measures and methodologies to determine whether our results would change. First, we reexamined our data with alternate measures of model fit besides the Harrell C statistic,3638 including the likelihood ratio χ2, McFadden pseudo8R2,39 and the adjusted Royston R2.40 Second, we performed analyses stratified by gender to determine whether our results would differ between men and women. Because the stratified analysis results were similar to our primary results, only the primary results are shown. Third, because of differences in sample sizes in the age groups, the Harrell C statistic was less precise and had larger standard errors in the models in our group of participants aged 90 to 99 years. To account for the variable precision of the Harrell C statistic across age groups in our test of trend, we performed weighted regression using the inverse of the standard error, allowing us to give more weight to more precise data. Results of this weighted regression trend analysis were similar to our primary analysis, so only the primary results are shown.

All statistical analyses were performed using Stata software version 9.0 (StataCorp LP, College Station, TX).

Characteristics of the Participants

Overall, 57% of participants were women, with the proportion of women generally increasing in older age groups. The proportion of participants who reported chronic conditions generally increased in older age groups, with 70% of participants aged 50 to 59 years reporting any chronic condition, compared with 90% of participants aged 90 to 99 years reporting any chronic condition. Similarly, the prevalence of functional limitations was also higher among older participants; for example, among participants aged 90 to 99 years, 68% reported difficulty with at least 1 ADL or IADL, compared with 15% of participants aged 50 to 59 years. Overall mortality also increased among older participants, reaching 76% in the group of participants aged 90 to 99 years (Table 1).

Predictive Performance of Chronic Conditions and Functional Limitations Across Age Groups

We found that among participants aged 50 to 59 years, chronic conditions were stronger predictors of all-cause mortality than functional limitations (Harrell C statistic 0.78 vs. 0.73, P=.001). Although the ability of both chronic conditions and functional limitations to predict all-cause mortality declined with older age, the decline was more pronounced with chronic conditions. As a result, in our oldest participants (aged 80–99 years), functional limitations were stronger predictors of mortality than chronic conditions (Table 2). For example, in our participants aged 90 to 99 years, the functional limitations model Harrell C statistic was 0.67, compared with the chronic conditions model Harrell C statistic of 0.61 (P=.004). In the intermediate age groups, we found a smooth transition between these extremes, so that in the group of participants aged 70 to 79 years, the functional limitations and chronic conditions models yielded a similar Harrell C statistic of 0.70 (P=.36).

We also found that in every age group, the weaker types of risk factors (i.e., either chronic conditions or functional limitations) only marginally improved the predictive power of the full model (model D). For example, in the group of participants aged 50 to 59 years, the addition of functional limitations improved the Harrell C statistic slightly, from 0.78 to 0.79. Similarly, among participants aged 90 to 99 years, the addition of chronic conditions to a model containing demographic characteristics and functional limitations marginally improved the model Harrell C statistic, from 0.67 to 0.69.

To determine the robustness of our findings, we reexamined our data using other measures of model fit and found similar results (Figure 1). For each measure of model fit (Harrell C statistic, likelihood ratio χ2, McFadden pseudo-R2, and the adjusted Royston R2), chronic conditions were stronger predictors of mortality among younger participants, whereas functional limitations were stronger predictors of mortality among older participants.

In a population-based sample of US adults aged 50 to 99 years, we found that chronic diseases were more predictive of all-cause mortality among younger participants (aged 50–59 years), and functional limitations were more predictive of mortality among older participants (aged 80–99 years). Our results suggested that for participants older than 80 years, functional status was a more important predictor of mortality than their disease diagnoses.

Implications and Possible Explanations

Our results have several implications. Because many of the current comparisons of risk-adjusted outcomes for health care quality measurement and performance incentives rely solely on diagnosis codes,21,22 our results raise concerns about the accuracy of these methods in participants older than 80 years. Our results, coupled with previous research suggesting that the use of functional data can significantly improve mortality prediction, suggest that systematic collection of functional status data for persons aged 80 years or older could lead to improved risk adjustment in this population.24 Our results also highlight the importance of accounting for function in observational studies of older persons.

Our results also validate geriatricians’ longstanding focus on function. Maybe the time has come to teach our students and ourselves to consider functional limitations to be essential clinical information. For example, instead of teaching medical students to start presentations with statements such as “an 87-year-old male with a history of coronary artery disease and hypertension,” we could teach them that the statement “an 87-year-old male with difficulty in toileting and showering independently” may be more informative and may present a clearer picture of the patient and his issues.

There are several possible explanations for our results. First, because of survivor effect, chronic diseases may, on average, behave differently among older participants. Second, the number of diseases associated with mortality may increase with increasing age, decreasing the impact of any single disease on mortality.41 In that case, our finite list of chronic conditions would have accounted for more of the observed mortality in our younger patients than in our older patients. Third, the difference between nondisease and disease may be less clear-cut in the elderly. For example, older patients may be more likely to have prehypertension (classified as normal) and mild hypertension (classified as hypertensive) than younger patients, who are more likely to be either normotensive or clearly hypertensive.16,18 This would cause hypertension to be a weaker predictor of mortality among older participants because of the relative similarity of the hypertensive and nonhypertensive groups. Regardless of the underlying explanation, many risk-adjustment methods also use a finite number of chronic conditions, which means our results still suggest that commonly used risk-adjustment methods may be less accurate in the oldest old.

Our results do not suggest that chronic conditions are unimportant in predicting death in the oldest old. Given the higher absolute rates of mortality among older participants, the lower relative hazards from chronic conditions in this age group still represented a substantial increase in absolute predicted risk because of chronic conditions. However, this distinction between relative and absolute predicted risk has an even greater effect on functional limitations, with higher relative hazard ratios translating into very large increases in absolute predicted risk because of functional limitations.

Findings From Other Studies

Previous research has shown that the importance of various risk factors changes with age.7 Epidemiological studies of the oldest old in Israel5 and Denmark9 suggested that well-established risk factors for mortality in younger populations, such as history of disease and disease count, were less important among the oldest old. Furthermore, observational studies among older participants have shown that well-established cardiovascular risk factors, such as hypertension and hypercholes-terolemia, are relatively less important in predicting mortality among the elderly.8,11,12 We have extended those findings by incorporating measures of disease severity along with disease diagnoses and by examining a cohort that includes younger and older participants, allowing for direct comparisons of the importance of various risk factors across the age spectrum. In addition, by examining function along with chronic conditions, we were able to show that the relative importance of chronic conditions and functional limitations differs markedly at different ages.

We found that the weaker risk factors added little to the predictive power of the model for every age group. Thus, for younger participants, chronic conditions were most important and functional limitations added relatively little to mortality prediction. Conversely, for older participants, functional limitations were most important and chronic conditions added little. These findings suggest that a primary benefit of including both chronic conditions and functional limitations in mortality prediction models is to enable the model to predict more accurately over a wider age range.2

We also found that regardless of the risk factors considered, mortality prediction was less accurate in our oldest participants. Previous researchers have suggested that this may be because of the rapidly changeable nature of health in the elderly, making any mortality prediction more difficult.42 Some have even suggested that survival in the oldest old is a random process, independent of individual characteristics and primarily a function of external chance events.9,43 Although our results suggest that survival in the oldest old is a predictable, nonrandom process, mortality prediction does appear to be more difficult for this population, and optimal prediction may require consideration of risk factors not considered in our study.

Strengths and Limitations

Our findings should be considered in the context of the limitations of this study. First, we relied on patient self-report to measure chronic conditions and function, leaving open the possibility that more objective measures of these concepts could lead to different results. However, previous research suggests that patient report of both function and chronic conditions are reliable and have predictive validity.44,45 Although some studies have suggested that the accuracy of self-reported chronic conditions decreases with age, others have concluded that self-reports are valid among both younger and older participants.46 Thus, our findings highlight the need to examine whether the decline in the prognostic value of chronic conditions with increasing age is replicated when chronic conditions are measured through administrative diagnosis codes and chart review.

Second, our measures for chronic conditions and function are not exhaustive, leaving open the possibility that more-comprehensive measures of these concepts could lead to different results. However, we believe our measures are robust, encompassing several types of functional limitations (mobility, ADLs, and IADLs) as well as the most common causes of death in the United States.31 Third, our observational cohort study design can provide evidence of predictive association but not of causation. Thus, although many of our risk factors have been shown in other studies to be etiologic causes of mortality, our goal in the present study was strictly prediction. Finally, because our study sample was a community-dwelling cohort, it is unclear whether our results can be extrapolated to other populations, such as nursing home residents or hospitalized patients.


Our results suggest that chronic conditions are a stronger predictor of mortality among younger participants, whereas functional limitations are a stronger predictor of mortality among older participants. Therefore, mortality indexes and risk-adjustment methods that only consider chronic conditions may be suboptimal for comparing outcomes among persons older than 80 years. Overall, our findings suggest that comorbidity, as it is usually measured, has different prognostic implications for oldest old and that further research is needed on the interactions between age and chronic conditions.

TABLE 1— Participant Characteristics, by Age Group: Health and Retirement Study, 1998
TABLE 1— Participant Characteristics, by Age Group: Health and Retirement Study, 1998
 Age 50– 59 Years (n = 5382), %Age 60– 69 Years (n = 6591), %Age 70– 79 Years (n = 4903), %Age 80– 89 Years (n = 2219), %Age 80– 89 Years (n = 335), %All Participants (n = 19 430), %
Chronic conditions
Diabetes mellitus
    Oral medications only7101110810
    Kidney disease222212
Heart disease
    Heart attack past 2 y123432
    Heart failure124593
Lung disease
    No medications455535
    No current deficits135763
    Current deficits235693
Any chronic conditions708388929082
Functional limitations
    No difficulty979694898195
    No difficulty989897928897
    No difficulty979592826893
Using the telephone
    No difficulty989794877895
Managing finances
    No difficulty979794857295
Any difficulty in an ADL or IADL151926466821
    Able to jog 1 mile241583114
    Able to walk several blocks576160453057
    Able to walk 1 block111417242115
    Difficulty walking 1 block7913212911
    Can’t walk 1 block0.8227192

Note. ADL = activities of daily living; IADL = instrumental activities of daily living. ADL include bathing, dressing, eating, transferring, and toileting. IADL include shopping, preparing meals, using the telephone, managing medications, and managing finances.

TABLE 2— Harrell C Statistic of Cox Mortality Models Including Different Risk Factors, by Age Group: Health and Retirement Study, 1998
TABLE 2— Harrell C Statistic of Cox Mortality Models Including Different Risk Factors, by Age Group: Health and Retirement Study, 1998
ModelAge 50–59 y, Harrell C (95% CI)Age 60–69 y, Harrell C (95% CI)Age 70–79 y, Harrell C (95% CI)Age 80–89 y, Harrell C (95% CI)Age 90–99 y, Harrell C (95% CI)P Trenda
Model A: Base (age and gender)0.62 (0.59–0.65)0.60 (0.58–0.62)0.61 (0.59–0.62)0.60 (0.58–0.62)0.58 (0.54–0.62).05
Model B: Base + functional limitations0.73 (0.71–0.76)0.72 (0.70–0.73)0.70 (0.69–0.72)0.67 (0.66–0.69)0.67 (0.63–0.70).017
Model C: Base + chronic conditions0.78 (0.75–0.81)0.72 (0.70–0.74)0.70 (0.68–0.71)0.65 (0.63–0.67)0.61 (0.58–0.65).016
Model D: Base + conditions + function0.79 (0.77–0.82)0.75 (0.73–0.77)0.73 (0.72–0.74)0.69 (0.68–0.71)0.69 (0.66–0.72).017
P value for model B–C comparison. 

aPermutation test of whether slope of regression between age group and Harrell C statistic = 0.

Sei J. Lee was supported by the National Institute on Aging (grant T32 AG000212-15). Alan S. Go was supported by the National Heart, Lung, and Blood Institute (grant HL091179) and the National Institute of Diabetes and Digestive and Kidney Diseases (grant DK060902). Kenneth E. Covinsky was supported by the National Institute on Aging (grants R01 AG023626 and K24 AG029812).


1. Walter LC, Brand RJ, Counsell SR, et al. Development and validation of a prognostic index for 1-year mortality in older adults after hospitalization. JAMA. 2001;285(23):2987–2994. Crossref, MedlineGoogle Scholar
2. Lee SJ, Lindquist K, Segal MR, Covinsky KE. Development and validation of a prognostic index for 4-year mortality in older adults. JAMA. 2006;295(7): 801–808. Crossref, MedlineGoogle Scholar
3. Inouye SK, Peduzzi PN, Robison JT, et al. Importance of functional measures in predicting mortality among older hospitalized patients. JAMA. 1998; 279(15):1187–1193. Crossref, MedlineGoogle Scholar
4. Inouye SK, Bogardus ST Jr, 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(1):70–83. Crossref, MedlineGoogle Scholar
5. Ben-Ezra M, Shmotkin D. Predictors of mortality in the old-old in Israel: the cross-sectional and longitudinal aging study. J Am Geriatr Soc. 2006;54(6):906–911. Crossref, MedlineGoogle Scholar
6. Gullion CM, Keith DS, Nichols GA, Smith DH. Impact of comorbidities on mortality in managed care patients with CKD. Am J Kidney Dis. 2006;48(2):212–220. Crossref, MedlineGoogle Scholar
7. Forette B. Are common risk factors relevant in the eldest old? In: Robine JM, Forette B, Franceschi C, Allard M, eds. Research and Perspectives in Longevity: The Paradoxes of Longevity. Berlin, Germany: Springer-Verlag; 1999:73–79. Google Scholar
8. Menotti A, Kromhout D, Nissinen A, et al. Short-term all-cause mortality and its determinants in elderly male populations in Finland, the Netherlands, and Italy: the FINE Study. Prev Med. 1996;25(3):319–326. Crossref, MedlineGoogle Scholar
9. Nybo H, Petersen HC, Gaist D, et al. Predictors of mortality in 2,249 nonagenarians—the Danish 1905-cohort survey. J Am Geriatr Soc. 2003;51(10):1365–1373. Crossref, MedlineGoogle Scholar
10. Lamarca R, Ferrer M, Andersen PK, et al. A changing relationship between disability and survival in the elderly population: differences by age. J Clin Epidemiol. 2003:56(12):1192–1201. Crossref, MedlineGoogle Scholar
11. Casiglia E, Mazza A, Tikhonoff V, et al. Weak effect of hypertension and other classic risk factors in the elderly who have already paid their toll. J Hum Hyper-tens. 2002;16(1):21–31. Crossref, MedlineGoogle Scholar
12. Lindqvist P, Bengtsson C, Lissner L, Bjorkelund C. Cholesterol and triglyceride concentration as risk factors for myocardial infarction and death in women, with special reference to influence of age. J Intern Med. 2002;251(6):484–489. Crossref, MedlineGoogle Scholar
13. Heiat A, Vaccarino V, Krumholz HM. An evidence-based assessment of federal guidelines for overweight and obesity as they apply to elderly persons. Arch Intern Med. 2001;161(9):1194–1203. Crossref, MedlineGoogle Scholar
14. Stevens J, Cai J, Pamuk ER, et al. The effect of age on the association between body-mass index and mortality. N Engl J Med. 1998;338(1):1–7. Crossref, MedlineGoogle Scholar
15. Kohn RR. Cause of death in very old people. JAMA. 1982;247(20):2793–2797. Crossref, MedlineGoogle Scholar
16. Fried LP, Ferrucci L, Darer J, Williamson JD, Anderson G. Untangling the concepts of disability, frailty, and comorbidity: implications for improved targeting and care. J Gerontol A Biol Sci Med Sci. 2004;59(3): 255–263. Crossref, MedlineGoogle Scholar
17. Tinetti ME, Fried T. The end of the disease era. Am J Med. 2004;116(3):179–185. Crossref, MedlineGoogle Scholar
18. Karlamangla A, Tinetti M, Guralnik J, et al. Comorbidity in older adults: nosology of impairment, diseases, and conditions. J Gerontol A Biol Sci Med Sci. 2007;62(3):296–300. Crossref, MedlineGoogle Scholar
19. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373–383. Crossref, MedlineGoogle Scholar
20. Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8–27. Crossref, MedlineGoogle Scholar
21. Werner RM, Bradlow ET. Relationship between Medicare’s hospital compare performance measures and mortality rates. JAMA. 2006;296(22):2694–2702. Crossref, MedlineGoogle Scholar
22. Bradley EH, Herrin J, Elbel B, et al. Hospital quality for acute myocardial infarction: correlation among process measures and relationship with short-term mortality. JAMA. 2006;296(1):72–78. Crossref, MedlineGoogle Scholar
23. Fried LP, Guralnik JM. Disability in older adults: evidence regarding significance, etiology, and risk. J Am Geriatr Soc. 1997;45(1):92–100. Crossref, MedlineGoogle Scholar
24. Covinsky KE, Justice AC, Rosenthal GE, Palmer RM, Landefeld CS. Measuring prognosis and case mix in hospitalized elders: the importance of functional status. J Gen Intern Med. 1997;12(4):203–208. MedlineGoogle Scholar
25. Stein RE, Gortmaker SL, Perrin EC, et al. Severity of illness: concepts and measurements. Lancet. 1987; 2(8574):1506–1509. Crossref, MedlineGoogle Scholar
26. Heerenga SCJ. Sample design overview. In: Health and Retirement Study. Ann Arbor, MI: Institute for Social Research; 1996. Available at: http://hrsonline.isr.umich.edu/docs/sample/sho_samp.php?hfyle=ref023b&xtyp=1. Accessed March 30, 2005. Google Scholar
27. Design history. In: Health and Retirement Study. Ann Arbor, MI: Institute for Social Research; 2003. Available at: http://hrsonline.isr.umich.edu/intro/sho_uinfo.php?hfyle=history&xtyp=2. Accessed March 11, 2008. Google Scholar
28. Codebook: assets and health dynamics among the oldest old (AHEAD). In: Health and Retirement Study. Ann Arbor, MI: Institute for Social Research; 1999. Available at: http://hrsonline.isr.umich.edu/meta/1993/core/codebook/ahintro.htm. Accessed December 27, 2007. Google Scholar
29. Willis R, Suzman R. Sample sizes and response rates. In: Health and Retirement Study. Ann Arbor, MI: Institute for Social Research; 2002. Available at: http://hrsonline.isr.umich.edu/intro/sho_uinfo.php?hfyle=sample&xtyp=2. Accessed March 30, 2005. Google Scholar
30. Willis R, Suzman R. Cross-year NDI cause of death file. In: Health and Retirement Study. Ann Arbor, MI: Institute for Social Research; 2006. Available at: http://hrsonline.isr.umich.edu/meta/years/iy6.php?iyear=1027. Accessed March 30, 2005. Google Scholar
31. Gorina Y, Hoyert D, Lentzner H, Goulding M. Trends in causes of death among older persons in the United States. Aging Trends. 2006;6:2–4. Google Scholar
32. Saliba D, Elliott M, Rubenstein LZ, et al. The vulnerable elders survey: a tool for identifying vulnerable older people in the community. J Am Geriatr Soc. 2001;49(12):1691–1699. Crossref, MedlineGoogle Scholar
33. Harrell FE Jr, Califf RM, Pryor DB, Lee KL, Rosati RA. Evaluating the yield of medical tests. JAMA. 1982; 247(18):2543–2546. Crossref, MedlineGoogle Scholar
34. Mooney C, Duval R. Bootstrapping: A Nonparametric Approach to Statistical Inference. Newbury Park, CA: Sage; 1993. Google Scholar
35. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics. 1988;44(3):837–845. Crossref, MedlineGoogle Scholar
36. Cook NR, Buring JE, Ridker PM. The effect of including c-reactive protein in cardiovascular risk prediction models for women. Ann Intern Med. 2006;145(1): 21–29. Crossref, MedlineGoogle Scholar
37. Harrell FE Jr. Regression Modeling Strategies. New York, NY: Springer-Verlag; 2001. Google Scholar
38. D’Agostino RB, Griffith JL, Schmid CH, Terrin N. Measures for evaluating model performance. In: Carriquiry AL, ed. American Statistical Association, Proceedings of the Biometrics Section. Alexandria, VA: American Statistical Association; 1997:253–258. Google Scholar
39. McFadden D. Conditional logit analysis of qualitative choice behavior. In: Zarembka P, ed. Frontiers in Econometrics. New York, NY: Academic Press; 1974: 105–141. Google Scholar
40. Royston P. Explained variation for survival models. Stata J. 2006;6(1):83–96. Google Scholar
41. Guralnik JM. Assessing the impact of comorbidity in the older population. Ann Epidemiol. 1996;6(5): 376–380. Crossref, MedlineGoogle Scholar
42. Grant MD, Piotrowski ZH, Chappell R. Self-reported health and survival in the Longitudinal Study of Aging, 1984–1986. J Clin Epidemiol. 1995;48(3):375–387. Crossref, MedlineGoogle Scholar
43. Poon L, Johnson M, Davey A, et al. Psychosocial predictors of survival among centenarians. In: Martin P, Rott C, Hagberg B, Morgan K, eds. Centenarians: Autonomy Versus Dependence in the Oldest Old. New York, NY: Springer; 2000:77–89. Google Scholar
44. Covinsky KE, Palmer RM, Counsell SR, et al. Functional status before hospitalization in acutely ill older adults: validity and clinical importance of retrospective reports. J Am Geriatr Soc. 2000;48(2):164–169. Crossref, MedlineGoogle Scholar
45. Katz JN, Chang LC, Sangha O, Fossel AH, Bates DW. Can comorbidity be measured by questionnaire rather than medical record review? Med Care. 1996;34(1): 73–84. Crossref, MedlineGoogle Scholar
46. Simpson CF, Boyd CM, Carlson MC, et al. Agreement between self-report of disease diagnoses and medical record validation in disabled older women: factors that modify agreement. J Am Geriatr Soc. 2004; 52(1):123–127. Crossref, MedlineGoogle Scholar


No related items




Sei J. Lee, MD, MAS, Alan S. Go, MD, Karla Lindquist, MS, Daniel Bertenthal, MPH, and Kenneth E. Covinsky, MD, MPHSei J. Lee, Daniel Bertenthal, and Kenneth E. Covinsky are with the San Francisco Veterans Affairs Medical Center, Health Services Research and Development Research Enhancement Award Program, San Francisco, CA. Sei J. Lee, Karla Lindquist, and Kenneth E. Covinsky are with the Division of Geriatrics, University of California, San Francisco. Alan S. Go is with the Division of Research, Northern California Kaiser Permanente, Oakland, CA. “Chronic Conditions and Mortality Among the Oldest Old”, American Journal of Public Health 98, no. 7 (July 1, 2008): pp. 1209-1214.


PMID: 18511714