Advertisement

Discrimination ability of comorbidity, frailty, and subjective health to predict mortality in community-dwelling older people: Population based prospective cohort study

  • Sasmita Kusumastuti
    Correspondence
    Corresponding author at: Section of Social Medicine, Department of Public Health, University of Copenhagen, Oster Farimagsgade 5, PO Box 2099, DK-1014 Copenhagen K, Denmark.
    Affiliations
    Section of Social Medicine, Department of Public Health, University of Copenhagen, Copenhagen, Denmark

    Center for Healthy Aging, University of Copenhagen, Copenhagen, Denmark
    Search for articles by this author
  • Thomas Alexander Gerds
    Affiliations
    Section of Biostatistics, Department of Public Health, University of Copenhagen, Copenhagen, Denmark
    Search for articles by this author
  • Rikke Lund
    Affiliations
    Section of Social Medicine, Department of Public Health, University of Copenhagen, Copenhagen, Denmark

    Center for Healthy Aging, University of Copenhagen, Copenhagen, Denmark
    Search for articles by this author
  • Erik Lykke Mortensen
    Affiliations
    Center for Healthy Aging, University of Copenhagen, Copenhagen, Denmark

    Section of Environmental Health, Department of Public Health, University of Copenhagen, Copenhagen, Denmark
    Search for articles by this author
  • Rudi G.J. Westendorp
    Affiliations
    Section of Social Medicine, Department of Public Health, University of Copenhagen, Copenhagen, Denmark

    Center for Healthy Aging, University of Copenhagen, Copenhagen, Denmark
    Search for articles by this author

      Highlights

      • Calendar age holds most of the ability to accurately predict individual mortality.
      • Discrimination ability of current health indicators decreases with increasing age.
      • Stochastic variation due to ageing causes infinite heterogeneity between old people.
      • Heterogeneity in old age cannot be sufficiently captured by health indicators.
      • Clinicians must be vigilant in comprehending the limitations of health indicators.

      Abstract

      Objective

      To investigate the added value of comorbidity, frailty, and subjective health to mortality predictions in community-dwelling older people and whether it changes with increasing age.

      Participants

      36,751 community-dwelling subjects aged 50–100 from the longitudinal Survey of Health, Ageing, and Retirement in Europe.

      Methods

      Mortality risk associated with Comorbidity Index, Frailty Index, Frailty Phenotype, and subjective health was analysed using Cox regression. The extent to which health indicators modified individual mortality risk predictions was examined and the added ability to discriminate mortality risks was assessed.

      Main outcome measures

      Three-year mortality risks, hazard ratios, change in individual mortality risks, three-year area under the receiver operating characteristic curve (AUC).

      Results

      Three-year mortality risks increased 41-folds within an age span of 50 years. Hazard ratios per change in health indicator became less significant with increasing age (p-value < 0·001). AUC for three-year mortality prediction based on age and sex was 76·9% (95% CI 75·5% to 78·3%). Information on health indicators modified individual three-year mortality risk predictions up to 30%, both upwards and downwards, each adding <2% discriminative power. The added discrimination ability of all health indicators gradually declined from an extra 4% at age 50–59 to <1% in the oldest old. Trends were similar for one-year mortality and not different between sexes, levels of education, and household income.

      Conclusion

      Calendar age encompasses most of the discrimination ability to predict mortality. The added value of comorbidity, frailty, and subjective health to mortality predictions decreases with increasing age.

      Keywords

      1. Introduction

      Prognosis is the cornerstone of medicine, yet predicting outcome is one of the biggest challenges professionals face in day-to-day clinical practice [
      • Moons K.G.M.
      • Royston P.
      • Vergouwe Y.
      • Grobbee D.E.
      • Altman D.G.
      Prognosis and prognostic research: what, why, and how?.
      ]. Not only does it reflect the trajectory of how a disease will develop along with its associated outcomes, it also guides decision making on the character and the timing of interventions. Without an accurate prediction of what is going to happen in future, it is sheer impossible to weigh up benefits and risks of implementing a potentially harmful strategy over watchful waiting, and subsequently for patients to make a properly informed decision. Estimating prognosis in old age is an even more strenuous task as many patients have atypical presentations of several diseases at the same time and use manifold pharmaceutical treatments [
      • Fabbri E.
      • Zoli M.
      • Gonzalez-Freire M.
      • Salive M.E.
      • Studenski S.A.
      • Ferrucci L.
      Aging and multimorbidity: new tasks, priorities, and frontiers for integrated gerontological and clinical research.
      ]. The underlying pathogenesis of these comorbidities is a random accumulation of permanent damage [
      • Kirkwood T.B.L.
      Understanding the odd science of aging.
      ], a notion that sparked an intensive search for a ‘biomarker of ageing’ that accurately reflects the functional status of our body. Currently we lack markers that reflect biological age better than calendar age [
      • Engelfriet P.
      • Jansen E.
      • Picavet H.
      Biochemical markers of aging for longitudinal studies in humans.
      ], and hence we rely on clinical disease markers for decision making [
      • Lee S.
      • Lindquist K.
      • Segal M.
      • Covinsky K.
      Development and validation of a prognostic index for 4-year mortality in older adults.
      ,
      • Gagne J.
      • Glynn R.
      • Avorn J.
      • Levin R.
      A combined comorbidity score predicted mortality in elderly patients better than existing scores.
      ].
      Even in the absence of a clinical diagnosis of disease, the ageing process results in a poor resolution towards homeostasis after a stressful event, which has been coined as ‘frailty’ [
      • Clegg A.
      • Young J.
      • Iliffe S.
      • Rikkert M.O.
      • Rockwood K.
      Frailty in elderly people.
      ]. In order to prevent the risk of adverse outcomes of (pharmaceutical) interventions, numerous studies have developed and validated indicators of frailty [
      • Clegg A.
      • Bates C.
      • Young J.
      • Ryan R.
      • Nichols L.
      • Ann Teale E.
      • et al.
      Development and validation of an electronic frailty index using routine primary care electronic health record data.
      ,
      • Song X.
      • Mitnitski A.
      • Rockwood K.
      Prevalence and 10-year outcomes of frailty in older adults in relation to deficit accumulation.
      ,
      • Theou O.
      • Brothers T.D.
      • Peña F.G.
      • Mitnitski A.
      • Rockwood K.
      Identifying common characteristics of frailty across seven scales.
      ], as well as subjective health [
      • Lee S.
      • Moody-Ayers S.
      • Landefeld C.
      The relationship between self-rated health and mortality in older black and white Americans.
      ,
      • Ishizaki T.
      • Kai I.
      • Imanaka Y.
      Self-rated health and social role as predictors for 6-year total mortality among a non-disabled older Japanese population.
      ] to help identify older people at risk. The generalized message from these studies is that these frailty indicators serve as warning signs in identifying older people at risk and provide a decision point to start or withhold specific interventions in order to improve outcomes of older people. However, there is only limited evidence as yet that the routine use of these instruments for decision making improves the outcomes of our interventions [
      • Moons K.G.M.
      • Altman D.G.
      • Vergouwe Y.
      • Royston P.
      Prognosis and prognostic research: application and impact of prognostic models in clinical practice.
      ].
      To have a better understanding on to what extent current health indicators capture information for decision making, we set out to determine how much mortality prediction based on age and sex can be improved by information on comorbidity, frailty, and subjective health using data from SHARE; a European effort of 27 European countries and Israel on determinants of health, ageing, and retirement. First, we assessed how health indicators and age interact in their ability to discern mortality risks in subjects from the general population. Second, we explored to what extent the health indicators modify individual mortality risk predictions based on age and sex. Third, we prospectively assessed the added value of health indicators to the discrimination ability of mortality predictions and analysed whether the added value changes with increasing age.

      2. Material and methods

      2.1 Study population

      This study was based on a secondary data analysis of the Survey of Health, Ageing, and Retirement in Europe (SHARE); a multinational database including information from sequential surveys – ‘Waves’ – on health, well-being, working conditions, retirement, socio-economic status and social networks of approximately 120,000 individuals from 27 European countries and Israel [
      • SHARE. Survey of Health, Ageing and Retirement in Europe
      SHARE Proj.
      ]. The SHARE target population consisted of community-dwelling persons aged 50 years and older who have regular domicile in their respective SHARE country, as well as their spouses/partners living in the same household independent of age. SHARE data used in this study were from Wave 1 until 4 (Release 5.0.0).
      We constructed a study sample consisting of participants who started at either SHARE Wave 1 or 2 and were censored for vital status in Wave 2 through Wave 4 (see Fig. 1). Out of all countries, Ireland was excluded because Ireland only started in Wave 2 and did not perform any censoring for vital status afterwards. Baseline data for demographics and health indicators were obtained from Wave 1 (2004/2005) or Wave 2 (2006/2007). New entries from the third wave were not included as that particular wave focused on retrospective interviews and did not provide data on health measurements. Dates of deaths and dates when participants were last surveyed alive were obtained from the second (2006/2007), third (2008/2009), and fourth (2011) waves of SHARE. Participants in all countries but Israel and Greece were followed up to Wave 4 for censoring of vital status. For Greece, we used Wave 3 for censoring as Wave 4 did not provide any information on vital status. For Israel, we included participants from Wave 1 in 2005 and used information from Wave 2 in 2010 for censoring vital status as Wave 2 in Israel was exceptionally late when compared to other countries. For the purpose of this study, participants with unknown birth date or aged younger than 50 were excluded. At baseline, there were 41,750 participants included in our study. We then excluded participants with any missing information on demographics and health indicators at baseline, resulting to the final study sample of 36,751 participants.
      Fig. 1
      Fig. 1Flowchart generating study sample from SHARE Wave 1 and 2.
      The selected participants started from either SHARE Wave 1 or 2 and then followed-up until Wave 4.
      aConsists of participants who were excluded due to no follow-up: all participants from Ireland (N = 1,007) and participants from Israel who started at Wave 2 (N = 407).
      Median duration of follow up after the initial health survey was 48·0 ± 33·6 months with an IQR of 75 months; it varied widely between countries due to differences in sampling regiments. As a result, censoring for vital status was available for 23,339 of the participants of whom the demographic variables and health indicators at baseline were not different from those who had no censoring information. We compared one-year Kaplan-Meier estimates with the expected mortality risk based on the mortality registers of the participating countries (see Supplementary Fig. 1) [

      University of California, Berkeley (USA) and MPI for DR (Germany). Human mortality database n.d. http://www.mortality.org.

      ]. Observed mortality in women aged over 85 years was lower than expected based on estimates from the population at large.

      2.2 Demographics and health indicators

      Demographic variables at baseline included continuous age, sex, education, and household income. Education was categorized according to 1997 International Standard Classification of Education (ISCED-97) [
      • United Nations Educational S, and Cultural Organization
      1997 International Standard Classification of Education.
      ]. Household income was based on the question “Is the household able to make ends meet?”. The Comorbidity Index was measured by making use of 7 available items from SHARE out of the 17 items from the Charlson Comorbidity Index ranging from 0 to 8 points following the original scoring system (see Supplementary Table 1) [
      • Charlson M.
      • Pompei P.
      • Ales K.
      A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.
      ]. The Frailty Index was established by Rockwood et al. [
      • Rockwood K.
      • Song X.
      • MacKnight C.
      • Bergman H.
      • Hogan D.B.
      • McDowell I.
      • et al.
      A global clinical measure of fitness and frailty in elderly people.
      ] and measured by the accumulation of health deficits; including symptoms, diseases, and disabilities. Following previous scholars [
      • Brothers T.D.
      • Theou O.
      • Rockwood K.
      Frailty and migration in middle-aged and older Europeans.
      ,
      • Peña F.G.
      • Theou O.
      • Wallace L.
      • Brothers T.D.
      • Gill T.M.
      • Gahbauer E.A.
      • et al.
      Comparison of alternate scoring of variables on the performance of the frailty index.
      ], we selected 70 items from the physical, behavioural, cognitive, mental health, and use of health care domains of the SHARE survey (see Supplementary Table 2). The items were coded into a dichotomous interval and calculated for each participant separately by dividing the sum of all deficits out of the total number of non-missing items. Participants missing information for >20% of the items were excluded from the analysis. The resulting range for this Frailty Index was 0 to 1. The Frailty Phenotype was originally developed by Fried et al. and characterized by shrinking/weight loss, weakness, exhaustion/poor endurance, slowness, and low activity [
      • Fried L.P.
      • Tangen C.M.
      • Walston J.
      • Newman A.B.
      • Hirsch C.
      • Gottdiener J.
      • et al.
      Frailty in older adults: evidence for a phenotype.
      ] and applied as described before (see Supplementary Table 3) [
      • Santos-Eggimann B.
      • Cuénoud P.
      • Spagnoli J.
      • Junod J.
      Prevalence of frailty in middle-aged and older community-dwelling Europeans living in 10 countries.
      ,
      • Romero-Ortuno R.
      • Walsh C.D.
      • Lawlor B.A.
      • Kenny R.A.
      • Ahmed N.
      • Mandel R.
      • et al.
      A frailty instrument for primary care: findings from the survey of health, ageing and retirement in Europe (SHARE).
      ]. A score of 1 was allocated for each of five criteria fulfilled, resulting in the full range of the Frailty Phenotype from 0 to 5. Subjective or self-perceived health was obtained from response to the question “Would you say your health is…” with the answer recoded to range from 1 to 5 with the levels being poor, fair, good, very good, and excellent.

      2.3 Statistical methods

      For our main analysis, we included participants with complete information on demographics and health indicators at baseline (see Fig. 1, N = 36,751 participants). Subjects were categorized into four groups of 50–59, 60–69, 70–79, and 80–100 year olds. Education was dichotomized into none/basic and higher education, household income was dichotomized into with and without difficulty making ends meet, and health indicators were dichotomized into Comorbidity Index score of 0 and ≥1, Frailty Index score of ≤0·2 and ≥0·3, Frailty Phenotype score of 0 and ≥1, and subjective health ratings of excellent/very good/good and fair/poor. The Kaplan-Meier method was used to describe mortality risks in the time period following baseline. Country specific Cox regression analyses were performed to predict one and three-year risks of mortality based on health indicators. Each Cox model included only one health indicator, all models were adjusted for age (residual age within age group if applicable) and sex. Further models were also adjusted for education and household income. The effects of the health indicators were allowed to change across age categories. We reported average hazard ratios across the countries (fixed effect meta-analysis) with corresponding 95% confidence limits. We examined how much the health indicators changed the subject specific mortality risk predictions compared to predictions based on age and sex alone, conditional on the status at the respective prediction horizon. The time-dependent area under the one and three-year receiver-operating-characteristic curves (AUC) was used to assess the added value of the health indicators to the discrimination ability based on age and sex alone. At the t-year horizon, AUC can be interpreted as the probability that a subject who died within t years had received a higher mortality risk compared to a subject who survived t years. All statistical analyses were performed using R version 3.3.1 [
      • R Development Core Team
      R: a language and environment for statistical computing.
      ].

      3. Results

      3.1 Descriptions of the study sample

      The study sample consisted of participants from northern (Denmark, Sweden), southern (Spain, Italy, Greece), western (Austria, Germany, Netherlands, France, Switzerland, Belgium), eastern (Czech Republic, Poland) Europe, and Israel. In total, we included the 36,751 community-dwelling older people with complete survey data ranging from 50 through 100 years old with an average age of 63·7 ± 9·7 (SD) years and the 25th, 50th, and 75th percentile being 56, 62, and 71 years. Some 19,682 (53·6%) were female. The proportion of participants across different age and gender groups (Table 1) conformed to the population distribution in Europe around the period when the initial surveys were performed (2004–2007) [

      Population Pyramid of EUROPE in 2016 n.d. https://populationpyramid.net/europe/2004/ (accessed May 10, 2016).

      ]. Supplementary Fig. 2 shows the heterogeneity of educational and economic levels over age in the study sample. In the younger age categories, the majority of participants have had secondary education while in the oldest old; most have had only primary education. In terms of economic levels, the proportions were similar over age categories. Overall, around 10% participants had great difficulties meeting their household needs while the rest were relatively equally divided between participants who had some difficulty, managed fairly easily, and easily making ends meet.
      Table 1Characteristics of the study sample at baseline.
      Age categories in yearsParticipants

      N [%]
      Females

      N [%]
      All36,751 [100·0]19,682 [53·6]
      50–547,503 [20·4]4,193 [21·3]
      55–597,236 [19·7]3,917 [19·9]
      60–646,314 [17·2]3,306 [16·8]
      65–695,433 [14·8]2,802 [14·2]
      70–744,380 [11·9]2,235 [11·4]
      75–793,178 [8·6]1,691 [8·6]
      80–841,865 [5·1]1,033 [5·2]
      85–89639 [1·7]379 [1·9]
      90–100203 [0·6]126 [0·6]

      3.2 Effect modification of age on the associations between health indicators and mortality risks

      Fig. 2 shows the distribution of health indicators across age categories showing the gradual accumulation of diseases, frailty, and worsening subjective health over age. Above age 65 years, >30% of the participants suffered from one major disease. Above age 70, >50% of the participants showed signs of frailty. The proportion of participants who considered themselves in good health gradually decreased in ascending age categories, especially in those aged 75 and older.
      Fig. 2
      Fig. 2Distribution of health indicators across age categories.
      Fig. 3 shows the three-year mortality risks that increased up to 41 folds within an age span of 50 years. Within the youngest age categories, comorbidity, frailty indices, and subjective health discriminated three-year mortality risks, i.e. the mortality trajectories varied depending on the scores of various health indicators, but three-year mortality risks became more similar with increasing age. Above age 80, the various health indicators had little discriminative power; especially the Comorbidity Index was less predictive compared to the frailty indices and subjective health, which appeared to discriminate three-year mortality risks better. As expected, three-year mortality risks in males were up to two-fold higher when compared to females. All health indicators appeared to be more discriminative in males when compared to females but with similar trajectories over age in the two sexes (data not shown).
      Fig. 3
      Fig. 3Three-year mortality risks stratified by health indicators across age categories (Kaplan-Meier method).
      Table 2 shows that for all health indicators except the Frailty Phenotype, the observed hazard ratios decreased significantly over ascending age categories, indicating weakening of risk estimation (all p for effect modification < 0·001). As for Frailty Phenotype, the hazard ratios increased from 1·7 to 2·4 in age 50s to 60s and then decreased again in the older age categories (Table 2, p for effect modification = 0·196).
      Table 2Hazard ratios associated with changes in health indicators stratified by age categories.
      Age categories in yearsNumber of people at riskNumber of deathsComorbidity IndexFrailty IndexFrailty PhenotypeSubjective Health
      Hazards ratio (95% CI)p-ValueHazards ratio (95% CI)p-ValueHazards ratio (95% CI)p-ValueHazards ratio (95% CI)p-Value
      50–5914,7392612·4 (2·3–2·5)<0·0012·9 (2·8–3·0)<0·0011·7 (1·6–1·8)<0·0013·4 (3·3–3·6)<0·001
      60–6911,7475312·0 (2·0–2·1)<0·0012·8 (2·7–2·9)<0·0012·4 (2·3–2·5)<0·0013·0 (2·9–3·0)<0·001
      70–797,5588181·6 (1·6–1·6)<0·0012·3 (2·3–2·4)<0·0012·1 (2·0–2·1)<0·0012·2 (2·2–2·3)<0·001
      80–1002,7076931·3 (1·3–1·4)<0·0011·9 (1·9–2·0)<0·0011·7 (1·6–1·7)<0·0011·9 (1·9–1·9)<0·001
      Trend36,7512,3030·8 (0·8–0·8)<0·0010·9 (0·9–0·9)<0·0011·0 (1·0–1·0)0·1960·8 (0·8–0·8)<0·001
      Analyses were performed separately for each country, then merged and reported as fixed effect meta-analysis averages.
      Hazard ratios presented are per change in health indicators and adjusted for residual age, sex, education, and household income.

      3.3 The added value of health indicators to individual mortality risk predictions above age and sex

      Fig. 4 present the changes in per individual predicted mortality risks when including information on health indicators when compared to corresponding predictions based on age and sex only (zero indicating no change). These individual predictions are grouped dependent on the participants' vital status three years later. Generally, we expect the health indicators to increase the mortality risks in those who died within three years and to decrease in those who survived within three years. Before age 60, among the subjects who died, adding the Frailty Index wrongly decreased the median predicted three-year mortality risk in by −0·1% with range from −1·1% to 5·4%. In contrast, adding the Comorbidity Index, the Frailty Phenotype, or subjective health correctly increased the median predicted mortality risk up to 0·5% with range from −1·4% to 3·0%. Among the subjects who survived, all health indicators correctly decreased the median predicted three-year mortality risk up to −0·2% with range from −1·6% to 3·5%. At age 60–69, among subjects who ultimately died, the Frailty Index wrongly decreased the median predicted mortality risk while all other health indicators correctly increased the median predicted mortality risk. Among the subjects who survived, all health indicators correctly decreased the median predicted mortality risk. At age 70–79, among those who died all health indicators correctly increased the median predicted mortality risk. Among those who survived, Comorbidity Index and Frailty Phenotype wrongly increased the median predicted risk while the Frailty Index and subjective health correctly decreased the median predicted risk. In the oldest old, among the subjects who died, all health indicators correctly increased the median predicted mortality risks up to 4·4% with range from −25·2% to 22·2%. Among the oldest old who survived, adding subjective health correctly decreased the median predicted three-year mortality risk by −3·0% (range from −20·4% to 25·7%) but adding Comorbidity, the Frailty Index, or the Frailty Phenotype wrongly increased the median mortality risks up to 1·5% (range from −19·2% to 7·0%).
      Fig. 4
      Fig. 4Absolute change in three-year individual mortality risk predictions using different health indicators across age categories.
      Absolute change in per-individual predicted three-year mortality risk using health indicators when compared to using information on age and sex only.
      Box plots represent medians, 25–75% range, and total range.
      Analyses were performed separately for each country, then merged and reported as fixed effect meta-analysis averages.

      3.4 The added value of health indicators to discrimination ability of age and sex

      Table 3 summarizes the discrimination ability of demographics and health indicators to predict one-year and three-year mortality among all participants aged 50 years and older. Age is responsible for the bulk of the discriminative power with an AUC of 79·8% (95% CI 77·5% to 82·0%) and 75·6% (74·2% to 77·1%) for one and three-year mortality respectively. Sex improved discrimination ability to predict vital status, and amounted to an extra 1·9% (1·3% to 2·5%) and 1·0% (0·7% to 1·3%) at one and three-year follow-up, while education and household income barely added any predictive value. Table 3 also shows the additional predictive value of the various health indicators to be little, i.e. each separate indicator added <2% points in AUC. Among the health indicators, subjective health appeared to improve AUC the most, followed by frailty indices, whereas comorbidity had the least added value. When using all available health information on top of age and sex, AUC increased from 84·6% (82·8% to 86·5%) to 84·7% (82·8% to 86·6%) for outcomes at one-year and from 76·9% (75·5% to 78·3%) to 81·7% (80·5% to 82·9%) for outcomes at three-year follow-up.
      Table 3Discrimination ability to predict one-year and three-year mortality based on demographics and health indicators.
      AUCOne-year mortalityThree-year mortality
      N deaths = 259N deaths = 1,087
       Added effect of AUCAUC (95% CI)p-ValueAUC (95% CI)p-Value
      Age79·8 (77·5–82·0)<0·00175·6 (74·2–77·1)<0·001
       Plus sex 1·9 (1·3–2·5)<0·001 1·0 (0·7–1·3)<0·001
      Age and sex84·6 (82·8–86·5)<0·00176·9 (75·5–78·3)<0·001
       Plus education 0·1 (−0·1–0·2)0·331 0·1 (−0·0–0·1)0·112
       Plus household income −0·0 (−0·1–0·1)0·950 −0·0 (−0·1–0·0)0·736
       Plus Comorbidity Index 0·5 (0·2–0·8)<0·001 0·8 (0·6–0·9)<0·001
       Plus Frailty Index 1·3 (1·0–1·7)<0·001 1·5 (1·3–1·7)<0·001
       Plus Frailty Phenotype 0·8 (0·4–1·1)<0·001 1·0 (0·8–1·2)<0·001
       Plus Subjective Health 1·3 (1·0–1·5)<0·001 1·7 (1·5–1·8)<0·001
      Analyses were performed separately for each country, then merged and reported as fixed effect meta-analysis averages.
      As the presence of diseases, frailty, and mortality is markedly different between the sexes, Table 4 summarizes the discrimination ability of age and health indicators to predict three-year mortality when participants were grouped according to gender. Overall, AUC in females was higher compared to males but the added value of health indicators to predict mortality risks is higher among males. Similar to Table 3, information on health indicators contributed very little to the discrimination ability of age, with subjective health contributing the most among all health indicators.
      Table 4Discrimination ability to predict three-year mortality based on demographics and health indicators stratified by gender.
      AUCFemalesMales
       Added effect of AUCAUC (95% CI)p-ValueAUC (95% CI)p-Value
      Age78·7 (76·5–80·8)<0·00173·7 (71·7–75·6)<0·001
       Plus education 0·0 (−0·0–0·1)0·393 0·0 (−0·1–0·1)0·771
       Plus household income 0·0 (−0·0–0·1)0·399 0·1 (0·0–0·2)0·032
       Plus Comorbidity Index 0·4 (0·2–0·6)<0·001 0·3 (0·2–0·5)<0·001
       Plus Frailty Index 0·1 (−0·1–0·2)0·361 2·1 (1·7–2·4)<0·001
       Plus Frailty Phenotype 0·2 (0·1–0·3)0·003 1·5 (1·2–1·9)<0·001
       Plus Subjective Health 0·6 (0·4–0·8)<0·001 2·4 (2·1–2·7)<0·001
      Analyses were performed separately for each country, then merged and reported as fixed effect meta-analysis averages.
      Additional analyses were also performed by stratifying participants based on educational levels and household income (data not shown). Discrimination ability of prediction based on age and sex in higher educational levels and socioeconomic groups was 2% to 3% higher compared to the lower levels, but stratified analyses shows similar patterns in which health indicators added very little to the ability of age (and sex) to discriminate mortality risks.
      Fig. 5 demonstrates the extent to which information on age (and sex) and health indicators accurately predicted three-year mortality in separate age categories. Overall, the ability of age (and sex) to correctly predict vital status three years after the initial health survey was highest in the youngest age categories and lowest in the oldest age categories. The added value of health indicators was highest in the youngest age categories and gradually decreased with increasing age. At age 50–59, among all health indicators, the added value of Comorbidity Index was 4·2% (1·8% to 6·7%, p < 0.001) on three-year AUC while others added <1% to three-year AUC. At age 60–69, Frailty Index added the most with 3·4% (1·8% to 4·9%, p < 0·001) while Comorbidity Index added the smallest with 0·5% (0·1% to 0·9%, p = 0·009). At age 70–79, all health indicators added <1·5% AUC, and in the oldest old it was <1%. The latter was not different when we analysed the participants aged 80 to 84 only.
      Fig. 5
      Fig. 5Discrimination ability of age and sex and health indicators to predict three-year mortality across age categories.
      Analyses were performed separately for each country, then merged and reported as fixed effect meta-analysis averages.
      Data are presented as point estimates of Area Under the receiver-operating-characteristic Curve (AUC).
      Although there was heterogeneity in the AUC trajectories over age between the different countries, we did not observe significant trends. We performed a sensitivity analysis excluding four countries that appeared to be most at odds (Israel, Spain, the Netherlands, Switzerland) but this did not alter our findings.

      4. Discussion

      This study aimed to investigate the discrimination ability of comorbidity, frailty, and subjective health to predict mortality risk among community-dwelling older persons. These widely used health indicators discriminated mortality risks of groups of people well, especially in the younger age categories but hazard ratios became significantly smaller with increasing age. When estimating prognosis of individual persons, the gradual occurrence of diseases, frailty, and worsening subjective health in the older age categories incrementally changed predictions of mortality risk based on age (and sex), both upwards and downwards. However, calendar age comprised most to the ability to accurately predict mortality of individual persons and extra information on health added only little to the discrimination ability, whereas the additional value of the various health indicators was smallest in the older age categories. The discrimination ability of health indicators was not different between sexes, countries, levels of education, and strata of household income.

      4.1 Limitations of health indicators to capture heterogeneity in old age

      It is paradoxical that widely used health and frailty indicators distinguished mortality risks of groups in the population at large but added only little to the discrimination ability predicting mortality of older individuals. This is not a new finding; both clinicians [
      • Ierodiakonou K.
      • Vandenbroucke J.
      Medicine as a stochastic art.
      ] and statisticians [
      • Henderson R.
      • Keiding N.
      Individual survival time prediction using statistical models.
      ] have emphasized the little predictive value of clinical signs and symptoms, as well as statistical models and indices to accurately estimate prognosis of individual persons, and both pointed to the fact that the distinction between what is achievable at the group and at an individual level is not well understood. This poor predictive accuracy of realistically estimating survival time of individuals reflects the challenge professionals face in every day clinical practice and marks the heterogeneity as it is perceived by older people themselves. A possible explanation for the phenomenon is the inherent complexity of the ageing process. The contributing factors that explain why and how individuals age can be distinguished into three principle categories, i.e. genetic variation, environmental variation, and stochastic variation, the latter being intrinsic to all biological processes. While research has revealed manifold gene-environmental interactions that explain ageing trajectories in experimental models and humans, capturing these pathophysiologic mechanisms in generalized archetypes is insufficient to accurately predict disease outcome and death of particular individuals as the intrinsic stochastic variation is not taken into account [
      • Finch C.E.
      • Kirkwood T.B.
      Chance, development, and aging.
      ]. Stochastic variation gives rise to unique etiological trajectories of ageing and an infinite heterogeneity between older people. Current mathematical models are not sufficiently able to capture this inter-individual variation that is often referred to as ‘frailty’. It should not come as a surprise that using parameter estimation of these imperfect models on groups of people for predicting outcomes of individual persons, that the large unexplained variation converts itself into a low discriminatory accuracy.
      Our findings also show that the incremental damage evoked by the ageing process to cells, tissues, and organs make older people intrinsically frail, a fact that cannot be sufficiently captured by neither clinical diagnoses nor by the scoring systems used in clinical practice. Nevertheless the modest association of these health indicators with mortality in groups of people have led others to recommend the use of these health indices for estimating outcome of individual persons [
      • Rockwood K.
      • Song X.
      • MacKnight C.
      • Bergman H.
      • Hogan D.B.
      • McDowell I.
      • et al.
      A global clinical measure of fitness and frailty in elderly people.
      ,
      • Fried L.P.
      • Tangen C.M.
      • Walston J.
      • Newman A.B.
      • Hirsch C.
      • Gottdiener J.
      • et al.
      Frailty in older adults: evidence for a phenotype.
      ,
      • Sundararajan V.
      • Henderson T.
      • Perry C.
      • Muggivan A.
      • Quan H.
      • Ghali W.A.
      New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality.
      ,
      • Song X.
      • Mitnitski A.
      • Rockwood K.
      Prevalence and 10-year outcomes of frailty in older adults in relation to deficit accumulation.
      ]. Moreover, estimates from the models that have been applied in these studies most often included age (and sex) either as a variable for stratification, as a covariate to adjust for confounding, or as an underlying time variable with the consequence that the predictive power of the various health indicators is overestimated. As a typical example, here we present an 80-year old male without comorbidity and a low frailty index that tends to be treated routinely because he is considered healthy, while a 50-year old female with comorbidities and a high frailty index tends to be treated with great caution. The data however, clearly indicate that these two seemingly very different persons have the same mortality risk.

      4.2 Calendar age as the driving force of mortality risk

      Above all it should be emphasized that the hazard ratios per change in the various health indicators were modest when compared to the hazard that comes with increasing age (and between the sexes) reiterating the importance of the ageing process as the driving force of disease and mortality [
      • Rae M.J.
      • Butler R.N.
      • Campisi J.
      • de Grey A.D.N.J.
      • Finch C.E.
      • Gough M.
      • et al.
      The demographic and biomedical case for late-life interventions in aging.
      ]. Furthermore, the discrimination ability of the various health indicators to predict individual outcome above the information on age (and sex) was lowest in the oldest age categories, and there were no particular groups in which they were more predictive. Other studies investigating the predictive power of genetics, physiological, lifestyle predictors [
      • Walter S.
      • Mackenbach J.
      • Vokó Z.
      • Lhachimi S.
      • Ikram M.A.
      • Uitterlinden A.G.
      • et al.
      Genetic, physiological, and lifestyle predictors of mortality in the general population.
      ], physical functioning [
      • Carey E.C.
      • Walter L.C.
      • Lindquist K.
      • Covinsky K.E.
      Development and validation of a functional morbidity index to predict mortality in community-dwelling elders.
      ,
      • Panas L.J.
      • Siordia C.
      • Angel R.J.
      • Eschbach K.
      • Markides K.S.
      Physical performance and short-term mortality in very old Mexican Americans.
      ,
      • Studenski S.
      • Perera S.
      • Patel K.
      • Rosano C.
      • Faulkner K.
      • Inzitari M.
      • et al.
      Gait speed and survival in older adults.
      ], comorbidity [
      • Newman A.B.
      • Boudreau R.M.
      • Naydeck B.L.
      • Fried L.F.
      • Harris T.B.
      A physiologic index of comorbidity: relationship to mortality and disability.
      ,
      • Quail J.M.
      • Lix L.M.
      • Osman B.A.
      • Teare G.F.
      Comparing comorbidity measures for predicting mortality and hospitalization in three population-based cohorts.
      ], and other biomarkers [
      • Wang T.J.
      • Gona P.
      • Larson M.G.
      • Tofler G.H.
      • Levy D.
      • Newton-Cheh C.
      • et al.
      Multiple biomarkers for the prediction of first major cardiovascular events and death.
      ,
      • Ganna A.
      • Ingelsson E.
      • Oortwijn W.
      • Nelissen E.
      • Adamini S.
      • van den Heuvel S.
      • et al.
      5 year mortality predictors in 498,103 UK Biobank participants: a prospective population-based study.
      ] came to similar conclusion as we presented here, with discrimination ability of age and sex alone ranging from 65% to 76%, whereas the health indicators these scholars investigated added only little predictive power.
      Our findings show that comorbidity and frailty indices perform similarly, which strengthens the idea that although frailty and comorbidity are defined differently, the two concepts are closely related [
      • Theou O.
      • Rockwood M.R.H.
      • Mitnitski A.
      • Rockwood K.
      Disability and co-morbidity in relation to frailty: how much do they overlap?.
      ] because both reflect the outcomes of the lifelong accumulation of bio-molecular damage [
      • Kirkwood T.B.L.
      Understanding the odd science of aging.
      ]. Also, it is interesting to note how well subjective health performs in comparison to comorbidity and frailty indicators. Single item self-reported scales such as subjective health tend to be overlooked as they are perceived as less reliable and more sensitive to contextual effects [
      • Sternhagen Nielsen A.B.
      • Siersma V.
      • Conradsen Hiort L.
      • Drivsholm T.
      • Kreiner S.
      • Hollnagel H.
      Self-rated general health among 40-year-old Danes and its association with all-cause mortality at 10-, 20-, and 29 years' follow-up.
      ,
      • Puvill T.
      • Lindenberg J.
      • Slaets J.P.J.
      • de Craen A.J.M.
      • Westendorp R.G.J.
      How is change in physical health status reflected by reports of nurses and older people themselves?.
      ]. Our findings show that the subjective feeling of one's health may be able to capture the same if not more than objective physical assessments such as having been diagnosed with diseases and or symptoms of frailty. This finding substantiates the notion to also take into account how older persons feel about their health and refrain from focusing solely on clinical impairments.

      4.3 Strengths and limitations

      This thorough exploration of the SHARE study has several strengths. It involves a large number of community-dwelling older participants from middle age up to the oldest old and of varying socioeconomic background, mimicking the heterogeneity of patients that present themselves for medical services in every day practice. Kaplan Meier estimates were used to examine mortality risks and a comparison with population mortality data shows that the lack of censoring information for a part of the sample has not introduced a major bias. Receiver-operating-characteristics and change in individual risk predictions were used to examine the added value of health indicators and essentially weigh the ratio of the true- over the false-positive predictions making the outcomes of the analyses intuitive on both population and individual level. Beyond age and sex, the added values of established health indicators were compared together with the subjective feeling of health as appreciated by older persons themselves as a measure to gauge one's health. It is a limitation that the health indicators used in this study were operationalized based on variables available to SHARE, which introduced restrictions to calculate the (rather elaborate) comorbidity index. Arguably, our shortened comorbidity index included major debilitating diseases such as heart attack, stroke, lung disease, and cancer. Diagnosis of dementia was not consistently recorded and included between the two waves of SHARE, therefore not included in the Comorbidity Index. There were no such restrictions to calculate frailty and the Frailty Index captures the functional consequences of cognitive deficits.

      4.4 Implications of the study

      The implication of this study is that in (very) old people, clinicians must be extra vigilant in comprehending the effect that the ageing process has incurred and the limitations of current health indicators to avoid potentially harmful stressors and interventions, regardless of how apparently healthy older people may appear. Decision making is further complicated by on-going innovations of severe (medical) interventions which have maximized outcomes at minimal adverse costs and some of these interventions have therefore become applicable to (very) old people. Due to heterogeneity in ageing, some patients with multi-morbidity are remarkably resilient to further decline and successfully undergo significant interventions whereas other apparently healthy elders suffer a fatal outcome when confronted with a minor event - leaving the question whether older patients are being under- or over-treated unanswered [
      • Moynihan R.
      • Doust J.
      • Henry D.
      Preventing overdiagnosis: how to stop harming the healthy.
      ]. Finally we acknowledge that as we increasingly live longer, in part because later birth cohorts have accumulated less damage over the life course, a nowadays 80-year old is generally in a far better shape than those who were 80 in the 1950s [
      • Christensen K.
      • Thinggaard M.
      • Oksuzyan A.
      Physical and cognitive functioning of people older than 90 years: a comparison of two Danish cohorts born 10 years apart.
      ].
      What kind of advice then should be given by professionals helping patients to make informed decisions? It is a moral imperative to adequately inform patients about our detailed knowledge on prognosis of groups of people but at the same time convey the message that there are uncertainties when making individual predictions [
      • Hollnagel H.
      Explaining risk factors to patients during a general practice consultation. Conveying group-based epidemiological knowledge to individual patients.
      ]. It is a daunting task to arrive at a good compromise when communicating this information effectively and at the same time avoiding spurious impressions of precision of individual forecasts.

      Acknowledgements

      SK, TAG, RL, ELM, and RGJW contributed to the conception of the study, the analysis, and interpretation of data. SK was in charge of drafting the work and together with TAG, RL, ELM, RGJW revised it critically for important intellectual content. SK, TAG, RL, ELM, RGJW gave final approval of the version to be published and agreed to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. RGJW is the guarantor of this paper.
      SK and RGJW had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
      All authors have completed the ICMJE uniform disclosure form at www.icmje.org/coi_disclosure.pdf and declare: no support from any organisation for the submitted work; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work.
      This work is supported by Nordea Fonden.
      There is no involvement of the funding source in study design; in the collection, analysis, and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.
      This paper uses data from SHARE Waves 1, 2, 3 (SHARELIFE), and 4 (DOIs: 10.6103/SHARE.w1.500, 10.6103/SHARE.w2.500, 10.6103/SHARE.w3.500, 10.6103/SHARE.w4.500), see Börsch-Supan et al. (2013) for methodological details.*
      The SHARE data collection has been primarily funded by the European Commission through FP5 (QLK6-CT-2001-00360), FP6 (SHARE-I3: RII-CT-2006-062193, COMPARE: CIT5-CT-2005-028857, SHARELIFE: CIT4-CT-2006-028812) and FP7 (SHARE-PREP: No. 211909, SHARE-LEAP: No. 227822, SHARE M4: No. 261982). Additional funding from the German Ministry of Education and Research, the U.S. National Institute on Aging (U01_AG09740-13S2, P01_AG005842, P01_AG08291, P30_AG12815, R21_AG025169, Y1-AG-4553-01, IAG_BSR06-11, OGHA_04-064) and from various national funding sources is gratefully acknowledged (see www.share-project.org).
      Relevant SHARE publications:
      • -
        Börsch-Supan A. Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 1 [Internet]. Data set; 2016. Available from: http://www.share-project.org/home0/wave-1.html
      • -
        Borsch-Supan A, Brandt M, Hunkler C, Kneip T, Korbmacher J, Malter F, et al. Data Resource Profile: The Survey of Health, Ageing and Retirement in Europe (SHARE). Int J Epidemiol [Internet]. Oxford University Press; 2013 Aug 1 [cited 2016 Aug 29];42(4):992–1001. Available from: http://www.ije.oxfordjournals.org/cgi/doi/10.1093/ije/dyt088
      • -
        Börsch-Supan A and HJ, editor. The Survey of Health, Ageing and Retirement in Europe – Methodology. [Internet]. Mannheim: Mannheim Research Institute for the Economics of Aging (MEA).; 2005. Available from: http://www.share-project.org/uploads/tx_sharepublications/SHARE_BOOK_METHODOLOGY_Wave1.pdf
      • -
        Börsch-Supan A, Brugiavini A, Jürges H, Mackenbach J, Siegrist J, Weber G, et al. Health, Ageing and Retirement in Europe First Results from the Survey of Health, Ageing and Retirement in Europe. 2005 [cited 2016 Aug 29]; Available from: http://www.mea.uni-mannheim.de
      • -
        Börsch-Supan A. Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 2 [Internet]. Data set; 2016. Available from: http://www.share-project.org/home0/wave-2.html
      • -
        Alcser KH, Andersen-Ranberg K, Angelini V, Attias-Donfut C, Avendano M, Benson GD, et al. First Results from the Survey of Health, Ageing and Retirement in Europe. 2004;
      • -
        Börsch-Supan A. Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 3 – SHARELIFE [Internet]. Data set; 2016. Available from: http://www.share-project.org/home0/wave-3-sharelife.html
      • -
        Börsch-Supan A, Brandt M, Hank K, Schröder M, editors. The Individual and the Welfare State [Internet]. Berlin, Heidelberg: Springer Berlin Heidelberg; 2011 [cited 2016 Aug 29]. Available from: http://link.springer.com/10.1007/978-3-642-17472-8
      • -
        Schröder M, Alcer K, Benson G, Blom AG, Börsch-Supan A, Das M, et al. Retrospective data collection in the Survey of Health, Ageing and Retirement in Europe. SHARELIFE methodology. Mannheim: Mannheim Research Institute for the Economics of Aging (MEA).; 2011.
      • -
        Börsch-Supan A. Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 4 [Internet]. Data set; 2016. Available from: http://www.share-project.org/home0/wave-4.html
      • -
        Börsch-Supan A, Brandt M, Litwin H, Weber G, editors. Active ageing and solidarity between generations in Europe [Internet]. Berlin, Boston: DE GRUYTER; 2013 [cited 2016 Aug 29]. Available from: http://www.degruyter.com/view/books/9783110295467/9783110295467/9783110295467.xml
      • -
        Malter F, Börsch-Supan A, Abduladze L, Balster E, Czaplicki C, Das M, et al. SHARE Wave 4 Innovations &amp; Methodology. 2013;

      Appendix A. Supplementary data

      References

        • Moons K.G.M.
        • Royston P.
        • Vergouwe Y.
        • Grobbee D.E.
        • Altman D.G.
        Prognosis and prognostic research: what, why, and how?.
        BMJ. 2009; 338: b375https://doi.org/10.1136/bmj.b375
        • Fabbri E.
        • Zoli M.
        • Gonzalez-Freire M.
        • Salive M.E.
        • Studenski S.A.
        • Ferrucci L.
        Aging and multimorbidity: new tasks, priorities, and frontiers for integrated gerontological and clinical research.
        J Am Med Dir Assoc. 2015; 16: 640-647https://doi.org/10.1016/j.jamda.2015.03.013
        • Kirkwood T.B.L.
        Understanding the odd science of aging.
        Cell. 2005; 120: 437-447https://doi.org/10.1016/j.cell.2005.01.027
        • Engelfriet P.
        • Jansen E.
        • Picavet H.
        Biochemical markers of aging for longitudinal studies in humans.
        Epidemiologic. 2013;
        • Lee S.
        • Lindquist K.
        • Segal M.
        • Covinsky K.
        Development and validation of a prognostic index for 4-year mortality in older adults.
        JAMA. 2006;
        • Gagne J.
        • Glynn R.
        • Avorn J.
        • Levin R.
        A combined comorbidity score predicted mortality in elderly patients better than existing scores.
        J Clin. 2011;
        • Clegg A.
        • Young J.
        • Iliffe S.
        • Rikkert M.O.
        • Rockwood K.
        Frailty in elderly people.
        Lancet. 2013; 381: 752-762https://doi.org/10.1016/S0140-6736(12)62167-9
        • Clegg A.
        • Bates C.
        • Young J.
        • Ryan R.
        • Nichols L.
        • Ann Teale E.
        • et al.
        Development and validation of an electronic frailty index using routine primary care electronic health record data.
        Age Ageing. 2016; 45: 353-360https://doi.org/10.1093/ageing/afw039
        • Song X.
        • Mitnitski A.
        • Rockwood K.
        Prevalence and 10-year outcomes of frailty in older adults in relation to deficit accumulation.
        J Am. 2010;
        • Theou O.
        • Brothers T.D.
        • Peña F.G.
        • Mitnitski A.
        • Rockwood K.
        Identifying common characteristics of frailty across seven scales.
        J Am Geriatr Soc. 2014; 62: 901-906https://doi.org/10.1111/jgs.12773
        • Lee S.
        • Moody-Ayers S.
        • Landefeld C.
        The relationship between self-rated health and mortality in older black and white Americans.
        J Am Geriatr. 2007;
        • Ishizaki T.
        • Kai I.
        • Imanaka Y.
        Self-rated health and social role as predictors for 6-year total mortality among a non-disabled older Japanese population.
        Arch Gerontol Geriatr. 2006;
        • Moons K.G.M.
        • Altman D.G.
        • Vergouwe Y.
        • Royston P.
        Prognosis and prognostic research: application and impact of prognostic models in clinical practice.
        BMJ. 2009; 338: b606https://doi.org/10.1136/bmj.b606
        • SHARE. Survey of Health, Ageing and Retirement in Europe
        SHARE Proj.
        (accessed June 28, 2016)
      1. University of California, Berkeley (USA) and MPI for DR (Germany). Human mortality database n.d. http://www.mortality.org.

        • United Nations Educational S, and Cultural Organization
        1997 International Standard Classification of Education.
        2006
        • Charlson M.
        • Pompei P.
        • Ales K.
        A new method of classifying prognostic comorbidity in longitudinal studies: development and validation.
        J Chronic. 1987;
        • Rockwood K.
        • Song X.
        • MacKnight C.
        • Bergman H.
        • Hogan D.B.
        • McDowell I.
        • et al.
        A global clinical measure of fitness and frailty in elderly people.
        CMAJ. 2005; 173: 489-495https://doi.org/10.1503/cmaj.050051
        • Brothers T.D.
        • Theou O.
        • Rockwood K.
        Frailty and migration in middle-aged and older Europeans.
        Arch Gerontol Geriatr. 2014; 58: 63-68https://doi.org/10.1016/j.archger.2013.07.008
        • Peña F.G.
        • Theou O.
        • Wallace L.
        • Brothers T.D.
        • Gill T.M.
        • Gahbauer E.A.
        • et al.
        Comparison of alternate scoring of variables on the performance of the frailty index.
        BMC Geriatr. 2014; 14: 25https://doi.org/10.1186/1471-2318-14-25
        • Fried L.P.
        • Tangen C.M.
        • Walston J.
        • Newman A.B.
        • Hirsch C.
        • Gottdiener J.
        • et al.
        Frailty in older adults: evidence for a phenotype.
        J Gerontol A Biol Sci Med Sci. 2001; 56: M146-M157https://doi.org/10.1093/gerona/56.3.M146
        • Santos-Eggimann B.
        • Cuénoud P.
        • Spagnoli J.
        • Junod J.
        Prevalence of frailty in middle-aged and older community-dwelling Europeans living in 10 countries.
        J Gerontol A Biol Sci Med Sci. 2009; 64: 675-681https://doi.org/10.1093/gerona/glp012
        • Romero-Ortuno R.
        • Walsh C.D.
        • Lawlor B.A.
        • Kenny R.A.
        • Ahmed N.
        • Mandel R.
        • et al.
        A frailty instrument for primary care: findings from the survey of health, ageing and retirement in Europe (SHARE).
        BMC Geriatr. 2010; 10: 57https://doi.org/10.1186/1471-2318-10-57
        • R Development Core Team
        R: a language and environment for statistical computing.
        2008
      2. Population Pyramid of EUROPE in 2016 n.d. https://populationpyramid.net/europe/2004/ (accessed May 10, 2016).

        • Ierodiakonou K.
        • Vandenbroucke J.
        Medicine as a stochastic art.
        Lancet. 1993; 341: 542-544https://doi.org/10.1016/0140-6736(93)90294-Q
        • Henderson R.
        • Keiding N.
        Individual survival time prediction using statistical models.
        J Med Ethics. 2005;
        • Finch C.E.
        • Kirkwood T.B.
        Chance, development, and aging.
        Oxford University Press, 2000
        • Sundararajan V.
        • Henderson T.
        • Perry C.
        • Muggivan A.
        • Quan H.
        • Ghali W.A.
        New ICD-10 version of the Charlson comorbidity index predicted in-hospital mortality.
        J Clin Epidemiol. 2004; 57: 1288-1294https://doi.org/10.1016/j.jclinepi.2004.03.012
        • Song X.
        • Mitnitski A.
        • Rockwood K.
        Prevalence and 10-year outcomes of frailty in older adults in relation to deficit accumulation.
        J Am Geriatr Soc. 2010; 58: 681-687https://doi.org/10.1111/j.1532-5415.2010.02764.x
        • Rae M.J.
        • Butler R.N.
        • Campisi J.
        • de Grey A.D.N.J.
        • Finch C.E.
        • Gough M.
        • et al.
        The demographic and biomedical case for late-life interventions in aging.
        Sci Transl Med. 2010; (2:40cm21)https://doi.org/10.1126/scitranslmed.3000822
        • Walter S.
        • Mackenbach J.
        • Vokó Z.
        • Lhachimi S.
        • Ikram M.A.
        • Uitterlinden A.G.
        • et al.
        Genetic, physiological, and lifestyle predictors of mortality in the general population.
        Am J Public Health. 2012; 102: e3-10https://doi.org/10.2105/AJPH.2011.300596
        • Carey E.C.
        • Walter L.C.
        • Lindquist K.
        • Covinsky K.E.
        Development and validation of a functional morbidity index to predict mortality in community-dwelling elders.
        J Gen Intern Med. 2004; 19: 1027-1033https://doi.org/10.1111/j.1525-1497.2004.40016.x
        • Panas L.J.
        • Siordia C.
        • Angel R.J.
        • Eschbach K.
        • Markides K.S.
        Physical performance and short-term mortality in very old Mexican Americans.
        Exp Aging Res. 2013; 39: 481-492https://doi.org/10.1080/0361073X.2013.839021
        • Studenski S.
        • Perera S.
        • Patel K.
        • Rosano C.
        • Faulkner K.
        • Inzitari M.
        • et al.
        Gait speed and survival in older adults.
        JAMA. 2011; 305: 50https://doi.org/10.1001/jama.2010.1923
        • Newman A.B.
        • Boudreau R.M.
        • Naydeck B.L.
        • Fried L.F.
        • Harris T.B.
        A physiologic index of comorbidity: relationship to mortality and disability.
        J Gerontol A Biol Sci Med Sci. 2008; 63: 603-609
        • Quail J.M.
        • Lix L.M.
        • Osman B.A.
        • Teare G.F.
        Comparing comorbidity measures for predicting mortality and hospitalization in three population-based cohorts.
        BMC Health Serv Res. 2011; 11: 146https://doi.org/10.1186/1472-6963-11-146
        • Wang T.J.
        • Gona P.
        • Larson M.G.
        • Tofler G.H.
        • Levy D.
        • Newton-Cheh C.
        • et al.
        Multiple biomarkers for the prediction of first major cardiovascular events and death.
        N Engl J Med. 2006; 355: 2631-2639https://doi.org/10.1056/NEJMoa055373
        • Ganna A.
        • Ingelsson E.
        • Oortwijn W.
        • Nelissen E.
        • Adamini S.
        • van den Heuvel S.
        • et al.
        5 year mortality predictors in 498,103 UK Biobank participants: a prospective population-based study.
        Lancet (London, England). 2015; 386: 533-540https://doi.org/10.1016/S0140-6736(15)60175-1
        • Theou O.
        • Rockwood M.R.H.
        • Mitnitski A.
        • Rockwood K.
        Disability and co-morbidity in relation to frailty: how much do they overlap?.
        Arch Gerontol Geriatr. 2012; 55: e1-e8https://doi.org/10.1016/j.archger.2012.03.001
        • Sternhagen Nielsen A.B.
        • Siersma V.
        • Conradsen Hiort L.
        • Drivsholm T.
        • Kreiner S.
        • Hollnagel H.
        Self-rated general health among 40-year-old Danes and its association with all-cause mortality at 10-, 20-, and 29 years' follow-up.
        Scand J Public Health. 2008; 36: 3-11https://doi.org/10.1177/1403494807085242
        • Puvill T.
        • Lindenberg J.
        • Slaets J.P.J.
        • de Craen A.J.M.
        • Westendorp R.G.J.
        How is change in physical health status reflected by reports of nurses and older people themselves?.
        J Gerontol Ser A Biol Sci Med Sci. 2016; (glw103)https://doi.org/10.1093/gerona/glw103
        • Moynihan R.
        • Doust J.
        • Henry D.
        Preventing overdiagnosis: how to stop harming the healthy.
        BMJ. 2012; 344e3502https://doi.org/10.1136/bmj.e3502
        • Christensen K.
        • Thinggaard M.
        • Oksuzyan A.
        Physical and cognitive functioning of people older than 90 years: a comparison of two Danish cohorts born 10 years apart.
        Lancet. 2013;
        • Hollnagel H.
        Explaining risk factors to patients during a general practice consultation. Conveying group-based epidemiological knowledge to individual patients.
        Scand J Prim Health Care. 1999; 17: 3-5