As people age, their health usually becomes more vulnerable,

BRIEF METHODOLOGICAL REPORTS Long-Term Risks of Death and Institutionalization of Elderly People in Relation to Deficit Accumulation at Age 70 Kenneth Rockwood, MD, Arnold Mitnitski, PhD, w Xiaowei Song, PhD, Bertil Steen, MD, z and Ingmar Skoog, PhD, MD OBJECTIVES: To measure relative fitness and frailty in older people without specific frailty instruments and to relate that measurement to long-term health outcomes. DESIGN: Retrospective cohort studies. SETTING: Two population-based studies of people aged approximately 70 at baseline and followed up to 0 years (in the Canadian Study of Health and Aging (CSHA)) or 26 years in the Gothenburg H-70 cohort study. PARTICIPANTS: Nine hundred sixty-two men and,78 women. MEASUREMENTS: Deficit accumulation (the exposure) was counted using self-reported (CSHA) or clinically designated (H-70) symptoms, signs, diseases, and disabilities. Relative fitness and frailty were measured in relation to the degree of deficit accumulation evaluated in four quartiles, representing those most fit to those most frail. The items that made up the frailty index were selected randomly without replacement in,000 iterations. The outcomes were risks of death or residential long-term care. RESULTS: Worse frailty, however measured, was associated with worse survival; the Kaplan-Meier curves of random iterations of the frailty definition showed virtually no interquartile overlap for mortality. For any given level of frailty, men died younger than women. Worse frailty was also associated with a higher risk of institutionalization. CONCLUSION: Frailty appears to be a robust concept that is readily operationalized, with the risk of adverse outcomes being largely established by age 70. J Am Geriatr Soc 2006. From the Geriatric Medicine Research Unit and w Department of Medicine, Dalhousie University, Halifax, Nova Scotia, Canada; and z Department of Geriatric Medicine and Institute of Clinical Neuroscience, Göteborg University, Göteborg, Sweden. Related paper presentations: Mitnitski A, Song X, Rockwood K. Random combinations of deficits in a frailty index to predict mortality. 57th GSA Annual Scientific Meeting, Washington, DC, November 2004. Gerontologist 2004;44 Special issue:4. Address correspondence to Dr. Kenneth Rockwood, 43 5955 Veterans Memorial Lane, Halifax NS, B3H 2E, Canada. E-mail: Kenneth.Rockwood@Dal.Ca DOI: 0./j.532-545.2006.00738.x Key words: aging; frailty; mortality; sex difference; resampling As people age, their health usually becomes more vulnerable, although not everyone is equally vulnerable; the state of marked vulnerability is often termed frailty. Although readily recognizable to clinicians, 2,3 how frailty should be defined is controversial.,4 The controversy arises chiefly from differing views about how to move from the many manifestations of frailty (presumed to arise from multiple mechanisms) to the many fewer items that pragmatically can be included in a given operational definition. Thus, for example, one widely held view is that frailty can be defined as a specific syndrome of wasting (manifest as weight loss, exhaustion, low physical activity, and slowness), 5 whereas others contend that frailty can be operationalized simply as slow gait speed. 6,7 Alternately, a less-specific approach, which assumes that the more things people have wrong with them, the frailer they will be, has been proposed. 8 By this account, summarizing what people have wrong with them is done simply by counting the deficits that individuals accumulate over time. This can be done robustly; for example, people from several countries accumulated deficits at a rate of about 3% per year. 9 Surprisingly, perhaps, this replicable rate of accumulation occurred even when exactly what was counted was not the same in each sample. This approach of counting deficits is also attractive, in that it makes the assessment of frailty widely available without special instrumentation, while adhering to the standard view that frailty is multiply determined. This apparently less-specific approach of simply counting the deficits that people have can allow for some hypotheses about frailty to be tested. Given that different samples with different measures showed comparable rates of deficit accumulation and that the number of deficits present in a given person is consistently related to their risk of adverse outcomes, 9 might it be that different measures from the same samples have similar relationships with JAGS 2006 r 2006, Copyright the Authors Journal compilation r 2006, The American Geriatrics Society 0002-864/06/$5.00

2 ROCKWOOD ET AL. 2006 JAGS adverse outcomes? If so, then in a given sample, grades of risk might vary little by the items that are included to define risk, so that the proportion of deficits that have accumulated, whatever they might be, would be more important than precisely which ones are included in a given deficit count from which the proportion present is estimated. One way to evaluate this is to construct the index variable by sampling deficits at random. In this way, it is possible to compare whether risk varies more by which variables are included or by the proportion of deficits that are present. In this study, people were evaluated at age 70 (an age that is sufficiently old for a large-enough level of deficits to be accumulated on average but at which variability is evident and the risk of adverse outcomes is important) and tested the hypothesis that, at a given age, frailty can be defined in relation to deficit accumulation, that it depends more on the number of deficits accumulated than on their nature, and that, although variable in populations, it is relatively fixed for individuals. METHODS Settings and Samples The longitudinal Gothenburg study of 70-year-olds (H- 70) 0 evaluates birth cohorts from their 70th birthday onward. The H-70 cohort born in 90/02 was first evaluated in 97/72 (N 5 966 (58 women) without missing data) and has been followed for 26 years, by which time, 98% of the cohort had died. It has had a particular focus on dementia in relation to cardiovascular risk factors.,2 The Canadian Study of Health and Aging (CSHA) is a national cohort study of elderly people that was assembled in 99/ 92 and followed for 0 years. 3,3,4 From the cohort of 0,267 respondents, the,208 community-dwelling people (660 women) aged 69 to 7 at baseline, of whom,74 had complete exposure and mortality data available, were selected. All community-dwelling people in CSHA had a screening questionnaire administered at home; those who were cognitively impaired and a comparison group were invited to a clinical examination. By the 0-year follow-up, 336 (29%) had died, and 9 (0%) had been institutionalized. Institutionalization status is not known for 207 people. (Those for whom institutionalization is not known showed no systematic differences in age, sex, or baseline frailty index values.) The Deficit Count As proposed elsewhere, 8 by counting deficits that people accumulated a frailty index can be operationalized as simply the proportion (from a given set) of deficits present in an individual, in relation to age. (For example, if 20 deficits are considered, and an individual has three present, then that person s index value is 3/20 5 0.5.) The frailty index is reproducible and highly correlated (correlation coefficient 0.95) with near-term (up to 5-year) mortality. 9,5 Although individual items can be weighted, 6 in most of its iterations, including those presented here, the frailty index treats all problems equally so that items (such as cancer) that carry a high risk of death have the same weight as others (such as skin disorders) that carry a lower risk. Although it is arguable whether the assumption of the equality of deficits should be maintained, the reproducibly high correlation between an equal-weights deficit count and mortality suggests that it should not be ignored. Thus, and in contrast to naming the specific parts of a frailty definition, an opposite approach was tested. Here whether random combinations of items can operationally define frailty is reported. The definition is tested by whether the resulting state predicts mortality. The rationale for predicting time to death is that, although not everyone who dies is frail, a higher risk of death is an indicator of vulnerability. Moreover, in general, to test a definition strictly, it must either be compared with a referentfa criterion standard, which in this case does not existfor its ability to predict a relevant and nonarbitrary outcome, such as death, must be tested. 7 Forty variables from CSHA and 5 variables from H-70 were used. 8 Whereas the H-70 variables come from many sources (self-report, examination, laboratory tests), all CSHA variables were self-reported. Although the data sets generally contain similar variables, they do not all contain the same variables. 9 Analysis The variables for each frailty index were recoded as binary, with value when the deficit was present and 0 when absent, so that higher scores indicate worse frailty.,8 Random sampling of 50% to 75% of the variables was used, repeated,000 times for each quartile of the index. The cutpoints varied slightly between men and women and between the H-70 and CSHA samples (e.g., in H-70, quartiles were cut at 0.08 for the fewest deficits (0.08 0.4, 0.5 0.20, and 40.20); in CSHA, the cutpoints were 0.06, 0.3, and 0.20). Kaplan-Meier survival curves were calculated for each quartile of the frailty index. The log-rank test was used to assess differences between quartiles, and Po.05 was considered to be statistically significant. Ethics The research reported in this paper complies with the ethical rules of the Declaration of Helsinki, including approval of each participating institutional review board for the collection of the original data. In addition, the Capital Health Research and Ethics Committee approved the proposal for secondary analysis. RESULTS In the H-70 cohort, long-term (26-year) mortality was related to the baseline count of deficits (Figure ). The Kaplan-Meier survival curves show that men (Figure A) and women (Figure B) who had more deficits at baseline (those who were frailer) had worse outcomes than those with fewer deficits (those who were fitter). Although the randomly composed indexes show within-quartile variability, there is little overlap across the quartiles in men or women in the various iterations (log-rank Po.05). The CSHA data show a similar pattern of higher mortality with more deficits over the 0 years of follow-up (Figure 2). Again, there is virtually no overlap between the survival curves in the four strata. In both the CSHA and H-70, differences in survival between men (Figures A, 2A) and women (Figures B, 2B) can be observed within each stratum (log-rank Po.00). At the same index value,

JAGS 2006 LONG-TERM RISK OF DEATH DEFINED BY AGE 70 3 A men B women Survival 0.4 0.2 0 0 5 0 5 20 25 30 0 5 0 5 20 25 30 Life span after 70 Figure. Survival in relation to the frailty index in the Gothenburg H-70 cohort, of people aged 70 at baseline, followed for 26 years: (A) men, n 5 448; (B) women, n 5 58. Survival depends on the number of deficits accumulated, here presented as quartiles of the frailty index designated by different shades. The quartile with the highest index value quartile is black, and the rest are in descending order of gray shades, interquartile cutpoints: 0.08, 0.5, 0.20 in men and 0.0, 0.5, 0.2 in women. Each line represents one Kaplan- Meier survival curve with randomly selected deficits. For illustration purposes, only 00 randomly selected lines per quartile are presented from the,000 resamplings. women, on average, showed higher survival than men (from 4 years for the lowest quartile to 6 years for the highest quartile). The relationship between institutionalization and the deficit count at baseline shows that the two quartiles with the most deficits had the highest rates of entering nursing homes (Figure 3) (log-rank Po.002). By contrast, the two least impaired had lower and overlapping rates (log-rank P 5.72). DISCUSSION These analyses demonstrate that random compositions of the deficits that make up a frailty index yield comparable estimates of the risk of adverse outcomes. Specifically, they show a dose response relationship (more deficits yielding higher risks) with respect to short-term and long-term mortality. Fewer data are available on institutionalization, but the general trend appears to hold there too. Men appear to be more vulnerable to deficit accumulation than women. The data must be interpreted with caution. Although there is little overlap in survival between quartiles, there is variability within quartiles. Any such variability would need to be considered when comparing between studies, or within studies over time, where it would be important to use the same items to try to make the most exact comparisons, recognizing that differences in sampling, operationalization, and loss to follow-up will remain. In addition, mortality prediction, although important, is not the only adverse outcome of interest. Although there were some data on institutionalization, it would be better if the frailty index could be tested in relation to other outcomes, such as hospitalization and use of other healthcare services, but such data are not available. The slopes of the mortality rates are steeper in the H-70 cohort than in the CSHA. This is an artifact of the differences in scale, reflecting the greater number of years of follow-up. Elsewhere, it has been shown that, on the same scale, mortality and the frailty index are closely related across studies. 9 A Survival 0.9 0.7 0.5 B 0.4 men women 0.3 0 2 4 6 8 9 0 0 2 4 6 8 9 0 Life span after 70 Figure 2. Survival in relation to the frailty index in the Canadian Study of Health and Aging cohort, selecting people aged 69 to 7 at baseline, followed for 0 years: (A) men, n 5 54; (B) women, n 5 660. Survival depends on the number of deficits accumulated, here presented as quartiles of the frailty index designated by different shades. The quartile with the highest index value quartile is black, and the rest are in descending order of gray shades, cutpoints: 0.06, 0.2, 0.20 in men and 0.06, 0.3, 0.2 in women. Each line represents one Kaplan-Meier survival curve with randomly selected deficits. For illustration purposes, only 00 randomly selected lines per quartile are presented from the,000 resamplings.

4 ROCKWOOD ET AL. 2006 JAGS Institutionalization free proportion 0.9 0.7 0.5 0 2 4 6 8 0 Time to institutionalization (years) Figure 3. Kaplan-Meier curves for time to institutionalization in the Canadian Study of Health and Aging cohort, selecting people aged 69 to 7 (n 5,74) at baseline. Institutionalization depends on the deficit count at baseline in randomly sampled deficits presented by quartiles designated by different shades. The quartile with the highest index value quartile is black, and the rest are in descending order of gray shades, cutpoints: 0.05, 0.3, 0.9. The two quartiles with the most deficits had the highest institutionalization rates, whereas the two least impaired had lower and overlapping rates. One caveat of secondary data analysisfthat not all relevant exposure data are availablefobtains less here. Indeed, the essential point is not just that more deficits indicate worse survival, but that the relationship holds regardless of which deficits are considered. In short, as people age, at a population level, it is not just the nature but also the number of problems that appears to be important. This has implications for how we understand deficits, both clinically and methodologically. From a methodological standpoint, multiplicity of measurements makes any individual item less important, and thus, overall, the approach is robust, without special instrumentation, although this is not to say that any variable will do, because they need to be age related and not to saturate too early (e.g., macular degeneration, cf. need for bifocal lens). In addition, given the variability within strata, it would be necessary, in following a given group of people over time, or in comparing two groups, to use deficit indexes that were strictly comparable, but the data suggest that there is considerable room for pragmatism and flexibility in deciding which data to collect. This is particularly important from the standpoint of secondary data analyses. From a clinical standpoint, the apparent imprecision in understanding deficits runs counter to the care that one pays to knowing, in a given patient, exactly what is wrong. It appears to be possible to couple clinical sensibility with a frailty index approach by counting the items that would be collected in the course of a routine comprehensive geriatric assessment. 3,9 In addition, many physicians will recognize the pattern of a patient who appears to be frail based on many small problems. It also appears that experienced physicians can stratify degrees of frailty with simple oneline descriptors in a way that yields results that are comparable to the index approach. 3 The clinical implications of this approach may extend beyond pragmatic flexibility in deciding how to measure relative fitness and frailty. By illustrating the complexity of the interactions between items, the data draw attention to other ways to think about the complex nature of elderly people who are frail. For example, if frail elderly people with high deficit index counts show increased vulnerability, it is easy to see why such vulnerability will take the form of failures such as delirium and falls. When complex systems fail, they fail with their highest-order functions first. Attention and concentration, maintenance of balance, and upright bipedal ambulation are such high-order functions, so that viewing frailty as a problem of complexity, with vulnerability arising from the inability to integrate responses in the face of a variety of insults, can be a powerful heuristic for clinicians. The data also support the validity of defining relative fitness and frailty in relation to deficit accumulation by showing a robust and replicable relationship between the counts and the adverse outcomes. Perhaps most intriguingly, it appears that, at least by the age of 70, vulnerability to adverse health outcomes can be predicted by considering a broad range of data, including self-reported information. Given that the baby boom cohort will be approaching this age within 0 years, this observation is of interest. Whether this vulnerability is mutable with specific interventions, whether individuals can move backward and forward between strata, and whether it is knowable at an earlier age are matters of some interest and are motivating further inquiries. ACKNOWLEDGMENTS Financial Disclosure: These analyses were supported by Canadian Institutes for Health Research (CIHR) operating Grants MOP62823 and MOP6469 and by the Dalhousie Medical Research Foundation (DMRF). KR is supported by the CIHR through an investigator award and by the DMRF as Kathryn Allen Weldon Professor of Alzheimer Research. BS and IS were supported by Swedish Research Council Grant 267 and Swedish Council for Working Life and Social Research Grant 2835. Each coauthor asserts no proprietary interest in the result and no financial conflict of interest. Author Contributions: Drs. Mitnitski and Rockwood had full access to the data and take responsibility for the integrity of the data and the accuracy of the data analysis, study concept, design, analysis and interpretation of data, drafting the manuscript, and obtaining funding. Drs. Skoog and Steen provided the H-70 data and contributed to and approved the final draft. Drs. Mitnitski and Song were responsible for statistical analyses. Sponsor s Role: Neither the CIHR nor the DMRF played a direct role in the study design, conduct, management, data analysis, review, or authorization for submission. REFERENCES. Hogan DB, MacKnight C, Bergman H et al. Models, definitions, and criteria of frailty. Aging Clin Exp Res 2003;5(Suppl 3): 29. 2. Studenski S, Hayes RP, Leibowitz RQ et al. Clinical Global Impression of Change in Physical Frailty: Development of a measure based on clinical judgment. J Am Geriatr Soc 2004;52:560 566. 3. Rockwood K, Song X, MacKnight C et al. A global clinical measure of fitness and frailty in elderly people. Can Med Assoc J 2005;73:489 495. 4. Rockwood K. Frailty and its definition: A worthy challenge. J Am Geriatr Soc 2005;53:069 070.

JAGS 2006 LONG-TERM RISK OF DEATH DEFINED BY AGE 70 5 5. Fried LP, Ferrucci L, Darer J et al. Untangling the concepts of disability, frailty, and comorbidity: Implications for improved targeting and care. J Gerontol A Biol Sci Med Sci 2004;59A:M255 M263. 6. Gill TS, Allore HG, Holford TR et al. The development of insidious disability in activities of daily living among community-living older persons. Am J Med 2004;7:484 49. 7. Gill TS, Allore HG, Holford TR et al. Hospitalization, restricted activity, and the development of disability among older persons. JAMA 2004;292:25 224. 8. Mitnitski AB, Graham JE, Mogilner AJ et al. Frailty and late-life mortality in relation to chronological and biological age. BMC Geriatr 2002;2:. 9. Mitnitski A, Song X, Skoog I et al. Relative fitness and frailty of elderly men and women in developed countries and their relationship with mortality. J Am Geriatr Soc 2005;53:284 289. 0. Steen BH, Djurfeldt H. The gerontological and geriatric population studies in Gothenburg, Sweden. Z Gerontol 993;26:63 69.. Skoog I, Nilsson L, Palmertz B et al. A population-based study of dementia in 85-year-olds. N Engl J Med 993;328:53 58. 2. Skoog I, Lernfelt B, Landahl S et al. 5-year longitudinal study of blood pressure and dementia. Lancet 996;347:4 45. 3. Canadian Study of Health and Aging. Study methods and prevalence of dementia. Can Med Assoc J 994;50:899 93. 4. The Canadian Study of Health and Aging Working Group. Disability and frailty among elderly Canadians: A comparison of six surveys. Int Psychogeriatr 200;3(Suppl ):59 67. 5. Mitnitski AB, Song X, Rockwood K. The estimation of relative fitness and frailty in community dwelling older adults using self-report data. J Gerontol A Biol Sci Med Sci 2004;59A:M627 M632. 6. Song X, Mitnitski AB, Rockwood K. Assessment of individual risk of death using self-report data: An artificial neural network compared to a frailty index. J Am Geriatr Soc 2004;52:80 84. 7. Streiner D, Norman G. Health Measurement Scales. A Guide to Their Development and Use, 3rd Ed. Oxford: Oxford University Press, 2003. 8. Supplementary Table. The List of Variables in H-70 and CSHA Used to Calculate the Frailty Index [on-line]. Available at http://myweb.dal.ca/amitnits/ H70-CSHA-variables.htm Accessed November, 2005. 9. Jones DM, Song X, Rockwood K. Operationalizing a frailty index from standardized comprehensive geriatric assessment. J Am Geriatr Soc 2004;52: 929 933.