By Julie E 
Photo Credit (Pixabay)
Policymakers must be informed on aging patterns among people over the age of 60 through global comparisons and large sample sizes. Using standardized data from the Gateway to Global Aging, we present a novel methodology for measuring healthy aging in 13 OECD nations.
Methods
First, we created an innovative measure of physiological age (PA), which is a measure of age weighted for the effects of frailty, daily living difficulties, and comorbidities. Second, we examined healthy aging indicators in 13 countries, ranking them according to the difference between estimated PA and chronological age (CA). Third, we investigated socioeconomic characteristics linked with healthy aging.
Results
We discovered a substantial association between our PA measurement and biological age. Italy, Israel, and the United States are the three countries with the greatest PA levels (independent of CA), indicating aging in bad health. In contrast, Switzerland, the Netherlands, Greece, Sweden, and Denmark have significantly lower PA than CA, indicating healthy aging. Finally, the PA-CA disparity is greater among the impoverished, less educated, and single older adults.
Conclusions
Countries with higher PA should establish or maintain healthy aging policies that target disadvantaged people.
The demographic transition affects all Organization for Economic Cooperation and Development (OECD) countries, but some appear to be better prepared than others.1 For example, Nordic countries (Norway, Sweden, and Denmark) are frequently used as benchmarks in global comparisons because their long-term care policies are more advanced than other countries’.2 However, little is known about healthy aging trends across countries. Recent findings have demonstrated improvements in the time spent in poor health conditions in several countries, although discrepancies (e.g., demographic characteristics, health measurements, sample sizes) in the surveys used for these analyses limit the comparability of these measures across nations.3
Documenting global healthy aging is challenging for two reasons: first, it is a relatively complex concept to describe, and second, it is extremely difficult to quantify. To address the need for global comparisons of healthy aging measures, a large body of literature has focused on the breakdown of cross-country differences in life expectancy versus quality-adjusted life expectancy, disability-adjusted life expectancy, or other combined measures of mortality and morbidity across countries. Several contributions highlight the relevance of using measures such as quality-adjusted life-years, disability-adjusted life-years, global burden of disease, years of life lost owing to mortality, and years lost due to disability to document differences in health statuses across different healthcare systems.4,5 These measures include both the prevalence of specific diseases (for example, strokes) and their relative impact on health, disabilities, and quality of life.
Although these scales are very useful for comparing health status across aging populations, they have two major limitations. First, they are not “true” measures of healthy aging, in the sense that they are not age-weighted for the impact of health and disability. Second, they are fundamentally demographic measures in the sense that they allow for the characterisation of population groups over a specific time period, but they do not reflect individual healthy aging differences over time. Currently, there are insufficient tools to enable global comparisons of healthy aging.
In this article, we present a unique measure of healthy aging based on harmonized data from the Gateway to Global Aging initiative. Our measure is based on Grossman’s conceptual framework 6, which states that aging is related to a decrease in people’s health capital. This methodology allows you to experimentally estimate the health capital degradation rate associated with aging and use that weight to determine the difference between people’s calendar age and their “true” physiological age (PA).
Although health disparities between countries and social determinants of health have been extensively researched, this paper makes three significant contributions to the literature. First, we create an original measure of physical activity at the person level, a measure of age weighted for the impact of frailty, activities of daily living (ADL) impairments, and comorbidities. Second, we use the difference between PA and chronological age (CA) to establish an individual-level measure of healthy aging and investigate the socioeconomic factors that influence it. Third, we rank nations in terms of healthy aging based on average (country-level) PA-CA discrepancies.
Our technique is unique in that we employ microeconometric models to compute individual-level indicators of healthy aging and then aggregate those data to estimate healthy aging at the national level. Our research helps to uncover markers of healthy aging and proposes a framework for aggregating these markers to measure PA. We demonstrate the need for identifying a holistic (vs. disease-specific) strategy for assessing healthy aging.
We used data from two surveys harmonized by the Gateway to Global Aging project: the Health and Retirement Survey (HRS)7 and the Survey of Health, Aging, and Retirement in Europe (SHARE)8 databases. We used six waves of SHARE data from 2004 to 2017, comprising the 12 nations observed over the time period: Austria, Belgium, Denmark, France, Germany, Greece, Israel, Italy, the Netherlands, Spain, Sweden, and Switzerland. The information for the United States was derived from HRS data spanning the years 2004 to 2017. We limited our study population to people aged 60 to 89, as well as those who were surveyed at least three times. After removing missing data, our sample included 39,164 persons and 121,705 person-wave observations, with an average of 3.11 observations per participant.
The Gateway to Global Aging statistics have been particularly standardized so that self-reported metrics can be compared. These data have been used in numerous articles to compare frailty and disability measurements around the globe. As previously stated, these harmonized data files can be easily used for cross-country and cross-temporal analysis because each country-level sample is representative of its 50+ population. Previous research has used these data to establish disparities in life expectancy between people with and without disabilities in numerous countries,10 so possible comparability and representativeness issues are minimal.
Conceptual Framework
Our PA measure is based on Grossman’s foundational model of health capital accumulation.6 In this paradigm, health is viewed as a durable capital stock that depreciates with age but may be increased through investment (for example, by purchasing preventive and curative medical care). This model has the benefit of being empirically testable, which allows it to quantify the contributions of numerous variables to the aging process. For example, previous research used this model to estimate the impact of difficult jobs on health and convert this impact into terms of aging.11,12 Similarly, our estimation strategy was to model the respective contributions of various health depreciation predictors on individuals’ self-reported health (SRH) and interpret these effects in terms of aging. This technique has several advantages: it is theoretically sound, has been utilized in prior studies, and is simple to implement, ensuring its replicability.
Our framework is based on four basic assumptions used to empirically implement Grossman’s model.13 First, health is a latent variable that may be approximated using an SRH measure (H1). Second, the age-related depreciation rate can be calculated by using age as a predictor of SRH in a regression model (H2). Third, health depreciation can be quantified using time variations in clinical characteristics that are established predictors of negative outcomes (hospitalization, disability, or death) (H3). Fourth, the effect of each health degradation variable is age-independent (H4). In accordance with the dynamic character of health evolution and the preceding empirical formulation of Grossman’s model (12, 13), our PA measure was based on the estimation of the general model described below:
is the structural age-related depreciation rate, evaluating the marginal (negative) effect of one more year on health (H2).
is the value of the kth indication of health degradation (for example, comorbidity index)
represents the worth of health in the previous year.
is a country-fixed impact.
is a time-invariant error term quantifying the effect of person-specific variables (e.g., biological predispositions) on health.
is an idiosyncratic (time-dependent) error term. The model has been updated for the effect of depression.
The linked coefficient is considered to have a considerable effect on SRH. SRH is a binary variable with values of 1 for “excellent,” “very good,” or “good” and 0 for “fair” or “poor health.”
We used the following health depreciation indicator dimensions:
Fried’s frailty index, 14 the number of limitations in ADL, the number of limitations in instrumental ADL (iADL), and a comorbidity index. Each dimension is detailed in Appendix Table A1 of Supplemental Materials, which may be obtained at https://doi.org/10.1016/j.jval.2022.05.007. Because our PA measure was intended for cross-country comparisons, we chose to include only objective comorbidity variables, including high blood pressure, diabetes, and cancer.
and
. Indeed, it has been suggested that SRH measures may suffer from cultural reporting biases.17, 18, 19 For instance, countries such as Spain or Italy tend to be more pessimistic in reporting their health than other countries.19,20 Conceptually, SRH can be modeled as a function of two components: the “true” latent health status
and a time-invariant reporting bias.
capturing pessimistic or optimistic attitudes.17,18 Following prior research, we assume a linear influence of both components on SRH; thus,
. The subscript c signifies the nation and indicates that the bias may be due to the cultural country effect. The inclusion of correlated random effects models enabled us to eliminate the cultural reporting bias caused by both between- and within-country disparities in health reporting. Furthermore, our dynamic model accounts for the fact that
Could change over time due to unobserved exogenous shocks (e.g., accidents). Indeed,
Such that Eq. (1) accounts for potential changes in reporting bias as represented by
The second stage involved converting the marginal impact of each health depreciation variable.
in terms of aging through calculation
can be understood as an “aging weight” (that is, a weight transformed into years of age). The weights for each dimension are presented in Appendix Tables A2 (weights for the comorbidity index) and A3 (weights for the health degradation indicators) in Supplemental Materials.
In the third stage, we calculated the PA by increasing the CA by each “aging weight” with the following formula:
It’s worth noting that this aggregation technique has three major limitations. First, it is based on assumptions that
is independent of population age structure (H4). Second, PA will undoubtedly be more than CA because the phrase
will always be optimistic. Finally, the model implicitly assumes that males and females have equal weights. These assumptions can be modified by accounting for the population’s age (and sex) structure using the formula below:
is the individual-specific value of indicator k (0 or 1) for sex g (male or female) at time t.
The related “aging weight” is estimated from Eq. (1) stratified by sex.
The scaling factor is determined by the relative values of each health degradation indicator in a population of the same gender and age group and is always between -1 and 1. Indeed, it enables for the “aging weight”
to be scaled while taking into account the age and gender distribution of the population. An example PA estimate for a 72-year-old male US citizen with frailty, diabetes, high blood pressure, and heart problems is provided in Appendix Table A4 in Appendix A in Supplemental Materials at https://doi.org/10.1016/j.jval.2022.05.007.
External Validation of the PA Measure.
We validated the PA measure using data from the Sarcopenia and Physical Frailty in Older People: Multicomponent Treatment Strategies database (ClinicalTrials.gov identifier: NCT02582138) comprising 1515 individuals from 11 European countries.20 We replicated our PA measure and created a measure of age weighted by the influence of biomarkers (“biomarker-based” measure of aging) using the multiple linear regression method described in the literature21 (see Appendix A for
We followed a two-step external validation technique. First, we verified that our PA measure was strongly linked with the “biomarker-based” metric. Second, we validated that both measures had the same explanatory power when predicting the value of a subject’s QOL by comparing the R2 between two models: one with only PA and the other with both PA and the “biomarker-based” measure. We discovered that the average anticipated survey-based and biomarker-based PA were similar (79.9 vs. 80.0). The correlation coefficient for the two measures was 0.4698. Both the PA and “biomarker-based” measures were significant predictors of QOL, with the R2 of the model including only PA being 0.1356 compared to 0.1354 for the models including both PA and the “biomarker-based” measure (see Tables B.2 and B.3, Appendix B in Supplemental Materials at https://doi.org/10.1016/j.jval.2022.05.007 for details). These findings demonstrate that, while not fully associated, the PA and “biomarker-based” assessments capture similar information about the subjects’ health.
The PA-CA Discrepancy: A Measure of Healthy Aging
To find variations in healthy aging, we first investigated the causes of the PA-CA discrepancy. In our approach, people are deemed to have healthy (or unhealthy) aging if their PA is less than (or greater than) their CA. We used various dynamic models (see Appendix A in Supplemental Materials at https://doi.org/10.1016/j.jval.2022.05.007 for details) to estimate (1) the marginal contribution of each health depreciation indicator to the PA-CA discrepancy and (2) the socioeconomic determinants of the PA-CA discrepancy. Our estimating technique decreases the possibility of lingering biases that could influence our measurements. Indeed, the inclusion of correlated random effects models enabled the elimination of systematic reporting disparities caused by time-invariant unobserved factors (between-subject variability). Furthermore, in all models, we adjusted for country-level and temporal fixed effects, as well as interactions between the two, allowing us to account for any changes (such as policy changes) that may have occurred over time. We utilized a heteroskedasticity-consistent standard error estimator in all models.
On average, the inhabitants of the United States, Israel, France, and Italy are in lower health than those of other countries (Figure 1). For example, in the United States, the proportion of older people with comorbidities is much higher (65.41% have high blood pressure, 19.46% have cancer, 29.97% have heart problems, 67.74% have arthritis, 52.23% have high cholesterol, and 33% have cataracts) than in Switzerland (41.43% have high blood pressure, 11.55% have cancer, 13.21% have heart problems, 33.22% have arthritis, 19.08% have high cholesterol, and 11.32% have cataracts).
ADL is for activity of daily living; HRS is the Health and Retirement Survey; IADL is instrumental activity of daily living; SHARE is the Survey of Health, Aging, and Retirement in Europe; and USA is the United States of America.
Table 1 analyzes the distribution of individual variables based on the PA-CA discrepancy. It specifically contrasts a group of persons whose PA is less than their CA (indicating “healthily” aging) to a group of people whose PA is bigger than their CA (indicating “unhealthily” aging). We can see that both groups have very different characteristics: people in the “healthy aging” group are more likely to report better self-perceived health (79.91% vs. 40.06%), are less likely to be frail (1.73% vs. 33.20%), face ADL/iADL limitations, and have fewer comorbidities (0.58 vs. 1.89). They also have reduced hospitalization rates (9.99% vs. 24.98%). Finally, they have higher levels of education and income, as well as a larger likelihood of living in partnerships (71.96% versus 66.98%).
Table 1. Descriptive statistics.
Note. The “healthy aging group” has a lower estimated PA than CA. In the “unhealthy aging group,” the estimated PA is greater than the CA. This table displays the column percentage. For example, within the “healthy aging group,” 79.91% report “good” (“excellent,” “very good,” “good”) self-perceived health, compared to 20.09% who report fair/poor self-perceived health.
ADL stands for Activities of Daily Living; iADL for Instrumental Activities of Daily Living; CA for Chronological Age; HRS for Health and Retirement Survey; PA for Physiological Age; and SHARE for Survey of Health, Aging, and Retirement in Europe.
Cross-country Ranking
Figure 2 compares countries based on their PA-CA disparity, as determined for adults aged 70 to 75 years. This age group is particularly important because handicap concerns are common around the age of 75.1 In the United States, Israel, and Italy, adults aged 70 to 75 are aging poorly; their PA is, on average, higher than their CA. The other countries indicate healthy aging, with average PA lower than CA. Switzerland, the Netherlands, Greece, and Sweden have the greatest rankings for healthy aging. For example, in Switzerland, the PA-CA disparity is minus 32 months, indicating that people in this nation mature healthily on average: their PA is nearly 3 years younger than their calendar age. Not unexpectedly, the PA disparities seen in Figure 2 are congruent with the health differences shown in Figure 1. This ranking is constant across age groups: in Switzerland, the Netherlands, Sweden, and Greece, the PA-CA discrepancy stays low in the oldest cohorts, indicating that healthy aging is maintained in old age (data not reported but available on request).
HRS stands for Health and Retirement Study, and SHARE stands for Survey of Health, Ageing, and Retirement in Europe.
The Socioeconomic Determinants of the PA-CA Discrepancy
Figure 3 depicts how individuals’ socioeconomic attributes affect the PA-CA disparity. Better education, being in a partnership, and having a better income are all protective factors against aging. Among these factors, educational level has the greatest impact: having completed tertiary education (as opposed to primary education) minimizes the PA-CA gap by 1.3 years, which is equivalent to being about 1 year and 3 months younger. Earning more than $50,000 per year has a four-times greater impact than earning less than $10,000. We did not detect significant interaction effects between income and education (see Appendix Fig. A1 in Appendix A in Supplemental Materials, available at https://doi.org/10.1016/j.jval.2022.05.007).
CA denotes chronological age; CI, confidence interval; HRS, Health and Retirement Study; PA, physiological age; ref, reference; SHARE, Survey of Health, Ageing, and Retirement in Europe; USA, United States of America.
Figure 4 depicts the occurrence of wealth differences in healthy aging across most countries. In the 70- to 75-year-old age group, the PA-CA disparity is 2.5 times lower among the top 25% than among the bottom 25%. Four countries have higher levels of inequality: the United States, France, Italy, and Greece. For example, in France, the 25% poorest persons aged 70 to 75 have an average PA that is 6 months bigger than their CA, whereas the 25% richest people have an average PA that is 12 months lower than their CA.
HRS stands for Health and Retirement Study, and SHARE stands for Survey of Health, Ageing, and Retirement in Europe.
Table 2 demonstrates that the weights used to construct our PA measures were resistant to different model settings. Indeed, all models produced fairly close estimates. In all models, the frailty and comorbidity measures were given higher weights. We also included separate variables for each health feature and achieved comparable results when ranking nations based on the PA-CA discrepancy (data available upon request).
Table 2. Weights assigned to each dimension in the PA computation in dynamic and nondynamic regressions.
Discussion
This article makes three major contributions: first, we create and validate a new measure of healthy aging that can be easily used in surveys and global comparisons; second, we provide a global comparison of healthy aging measures across 13 OECD countries; and third, we investigate the socioeconomic determinants of healthy aging and confirm that, across all countries, wealth and education are key factors in explaining aging inequalities. We provide evidence that factors such as human capital (as measured by individuals’ education), social capital (defined as the presence of siblings and friends who assist the older person when necessary), and income are strong predictors of PA, confirming the presence of socioeconomic inequalities in aging. Finally, we confirm earlier work demonstrating that an increase in disability in late age is not always found in all countries.10
Our strategy has various advantages. We compare the prevalence of health indicators across countries. This is accomplished by estimating the model on all countries and applying a scaling factor to account for country-specific differences in the distribution of each health depreciation indicator (when compared to all other countries). In other words, our strategy takes into consideration all disparities across countries. Furthermore, our method is more reliable than a country-specific technique. For example, the United States has the highest obesity rate of any country. Using a nation-specific model would have resulted in an underestimation of the weight associated with obesity in our PA calculation, whereas our scaling factor offers a “relative weight,” which is a more realistic approach to accounting for country variances.
Despite these strengths, our study has certain drawbacks. Although using harmonized data allows you to include more nations in your research, it has two drawbacks. The amount of information may vary slightly between SHARE and HRS data. We included the comorbidities found in both databases, which required us to omit Parkinson’s (missing from HRS data) and Alzheimer’s (misreported from SHARE data) disorders. Furthermore, our PA algorithm is based on the strong premise that the effects of each signal on self-rated health are additive. Future research could investigate if the influence of frailty, comorbidities, and ADL/iADL is multiplicative by introducing interaction effects among the health degradation indicators in the model. Finally, one could argue that the occurrence of biases in SRH measurements poses a possible issue in our research, which is wholly reliant on a cross-country comparison of outcomes. Previous research has revealed the importance of cultural differences in explaining SRH variations between nations.19 Jürges discovered that countries such as Denmark and Sweden overestimate their SRH, whereas Italy, France, Spain, and Germany underestimate their SRH. Nonetheless, we believe that this potential issue had no effect on our results for the reasons listed below. First, we calculated our PA measure by estimating the additive impact of health capital depreciation on SRH rather than using the SRH score’s value directly. As a result, the fact that the SRH score may be skewed due to cultural reporting habits does not necessarily mean that the marginal influence of age on health is biased. Second, our empirical technique allowed us to account for potential variability in reporting styles (for example, cultural bias), which could alter SRH assessment to the extent that reporting style and genuine latent health are additive. Our dynamic model also compensates for any time fluctuations in reporting patterns (by including the lag-dependent variable as a regressor). As a result, the coefficients computed and used to calculate our PA measure are unbiased as long as those cultural biases do not differ considerably between our cohort’s two waves. Third, our findings are robust to a variety of econometric specifications. When we compared the findings of two estimating procedures (ordinary least squares and random effect models), we discovered no significant changes in the effect of the health depreciation indicator on SRH, implying that cultural reporting styles had no influence on our estimations.
Our findings have several policy consequences. It is critical to better enlighten policymakers on healthy aging. Indeed, most social protection policies (for example, retirement age, immunization for the “most vulnerable”) are based on CA standards, despite the fact that PA measures would be more accurate. As a result, our health aging measure (CA-PA) would be more successful in prioritizing interventions for the elderly. For example, in France, the cutoff age for receiving the public allowance is 60 years; however, our study demonstrates that this law is fairly arbitrary. Furthermore, our metric could be utilized to improve the efficacy of initiatives aimed at older persons by calculating incremental cost-effectiveness ratios with gains in healthy aging years as the outcome. Our findings further suggest that healthcare policies should address age-related stigma issues, as CA is not a reliable indicator of people’s age and aging condition. The recent COVID-19 crisis has brought to light the risks of age-related discrimination in access to care, which have been documented over the last ten years among people with various diseases, such as colon cancer,22 and those undergoing surgeries, such as cardiac surgery.23 Healthcare interventions and preventive measures are now recommended based on age. Our findings indicate that these interventions and measures should be modified to precisely target patients with a higher need for secondary or tertiary prevention while accounting for their PA. Finally, our findings demonstrate how wealth and education disparities affect health24,25 and the aging process26, emphasizing the significance of giving assistance to the poorer and less educated populations.
In conclusion, this article provides a new tool to measure healthy aging across OECD countries, in line with previous work focusing on biological age, which reflects both physical and cognitive impairments that can be associated with the aging process.21,27,28 While measuring people’s biological aging requires the use of biomarkers, which are difficult to access and expensive to collect and are thus not available in global data, our PA measure can easily be implemented. Using our metric, future research might investigate the extent to which healthy aging influences healthcare utilization (particularly hospitalization) in older populations.
Article and Author Information.
Analysis and interpretation of data: Rapp, Ronchetti, Sicsic.
Drafting of the manuscript: Rapp, Ronchetti, Sicsic.
Rapp, Ronchetti, and Sicsic performed a critical revision of the work for essential intellectual content.
Conflict of Interest Disclosures: All authors disclosed that the Innovative Medicine Initiative and the Agence nationale de la recherche provided money to the Université Paris Cité. Dr. Rapp disclosed receiving an honorarium for consulting for Gerson Lehrman Group, Guidepoint, Ception, CreativCeutical, and Nextep. Dr. Rapp is an editor for Value in Health and did not participate in the article’s peer review. There were no additional disclosures noted. The publication of study results was not subject to the sponsor’s permission or censoring of the manuscript.
The funder/sponsor played no part in the study’s design and execution; data collection, management, analysis, and interpretation; manuscript preparation, review, or approval; or decision to submit the manuscript for publication.
Acknowledgment
This article draws on data from the Gateway to Global Aging, which is funded by the National Institute on Aging, National Institutes of Health (R01 AG030153, RC2 AG036619, R03 AG043052, R24 AG048024). Our findings were presented in June 2021 at an online research seminar co-hosted by researchers from King’s College London and Université Paris Cité, at the Caisse des Dépôts et Consignations’ Annual Aging workshop in Paris, France, and at the French Health Economics Association’s 2021 Annual Meeting. The authors thank everyone who attended these events for their useful feedback, especially Mauricio Avendano, Ludovico Carrino, and Alain Paraponaris.