Abstract
Objectives
Investigate whether combinations of sociodemographic factors, chronic conditions, and other health indicators pose barriers for older adults to access Annual Wellness Visits (AWVs) and influenza vaccinations.
Methods
Data on 4999 individuals aged ≥65 years from the 2012 wave of the Health and Retirement Study linked with Medicare claims were analyzed. Conditional Inference Tree (CIT) and Random Forest (CIRF) analyses identified the most important predictors of AWVs and influenza vaccinations. Multivariable logistic regression (MLR) was used to quantify the associations.
Results
Two-year uptake was 22.8% for AWVs and 65.9% for influenza vaccinations. For AWVs, geographical region and wealth emerged as the most important predictors. For influenza vaccinations, number of somatic conditions, race/ethnicity, education, and wealth were the most important predictors.
Conclusions
The importance of geographic region for AWV utilization suggests that this service was unequally adopted. Non-Hispanic black participants and/or those with functional limitations were less likely to receive influenza vaccination.
Keywords
Introduction
Aging of populations worldwide has intensified the focus on appropriate receipt of preventive health services for the promotion of good health and wellbeing in old age.1-3 With the increasing prevalence of older adults living with multiple chronic conditions in the US,4,5 access to medical care and supportive environments is essential to ensure health and quality of life for these older adults. 3 It is, thus, worrisome from a public health perspective that recommended preventive healthcare services are substantially underutilized among older adults in the US. 6 To improve health and access to care for this segment of the population, Annual Wellness Visits (AWVs) offered at no cost for eligible Medicare beneficiaries were introduced in 2011. The primary aim of this initiative was to increase access to preventive healthcare services for older adults.7,8 Influenza vaccination is a critical part of these services as it is a high-impact and cost-effective strategy to maintain the health of older adults. 9 Following the introduction of AWVs, studies have elucidated favorable developments including increased referrals and use of preventive healthcare services.10-17 At the same time, primary care physician practices and geographical regions where AWVs are more frequently adopted do not deliver a higher amount of healthcare services overall. 18
Despite a gradual increase in utilization of AWVs after their introduction in 2011,17-21 less than a quarter of Medicare beneficiaries received an AWV in 2015. 19 Previous studies have identified underutilization of AWVs among older adults who are unmarried, members of a race/ethnic minority group, or living in rural or less affluent areas.11,18,20,21 For influenza vaccinations, while findings are inconsistent with regard to age, sex, and single chronic conditions,11,18,21 sociodemographic factors and number of chronic conditions have been consistently linked to influenza vaccination uptake among older adults in the US.15,22-27
None of the previous studies on AWVs and influenza vaccinations have focused on the interaction between individual characteristics, such as sociodemographic factors and health indicators. However, individual characteristics may converge to influence the receipt of AWVs and influenza vaccinations among Medicare beneficiaries. Better understanding of the relationship between these factors in the use of AWVs and influenza vaccinations will inform future prevention efforts to address potential inequities and barriers to accessing services among different demographic groups. This study was undertaken to explore the combinations of sociodemographic factors, chronic conditions, and other health indicators that characterize utilization of AWVs and influenza vaccinations among groups of older adults.
Methods
We used data from the 2012 wave of the Health and Retirement Study (HRS) linked through individual identification numbers with Centers for Medicare and Medicaid Services (CMS) claims. In brief, the HRS is a nationally representative survey of individuals aged ≥51 years. The survey has been conducted biennially since 1992 with refreshment samples added every 6 years. 28 The selection of participants (N = 4999) for the study was based on enrollment in Medicare Part A and B for 12 consecutive months during the study period or until death, age ≥65 years, and complete information for predictors (Figure A1). At least 3 years of enrollment in the Medicare fee-for-service program was required to identify the conditions based on the Center for Medicare and Medicaid Services’ Chronic Condition Data Warehouse (CCW) algorithms. The study protocol was approved by Oregon Health and Science University Institutional Review Board (STUDY00017034).
Conditional Inference Trees for (A) Annual Wellness Visits (AWV) and (B) Influenza vaccinations (Flu). Explanatory note: The circles (child nodes) show the characteristics (variables) for which the data is split into the final boxes (terminal nodes). The boxes show the proportion of individuals with these characteristics that receive (A) Annual Wellness Visits (AWV) and (B) Influenza vaccinations (Flu). The “n”s under the boxes provide the number of individuals in each terminal node.
The main exposure variables consist of sociodemographic factors, chronic conditions, and other health indicators. Sociodemographic factors were identified in HRS and include: age categorized as 65 to 69 years (reference) with 5-year increments to ≥90 years; sex (male as reference); race/ethnicity categorized as non-Hispanic White (reference), Hispanic, and non-Hispanic Black; geographical region based on Census Divisions: New England (reference), Mid-Atlantic, East-North Central, West-North Central, South Atlantic, East-South Central, West-South Central, Mountain Division, and Pacific Division (see Online Appendix Figure A2 for more details); HRS validated measure of self-reported wealth (See Hurd et al for further details 29 ) included as quintiles in US$ (1st: ≤4000 (reference), 2nd: >4000 to ≤150 000, 3rd: >150 000 to ≤327 500, 4th: >327 500 to ≤767 000, 5th:> 767 000); Medicare-Medicaid dual eligibility (no as reference); and education as years in school centered on 12 years (median).
Chronic conditions up to the time of HRS interview covered common chronic conditions in older adult populations as recommended by Goodman et al. 30 The chronic conditions were identified from linked Medicare beneficiary files and administrative claims using CMS Chronic Condition Data Warehouse (CCW) algorithms.31,32 A description of the methodology to ascertain each chronic condition can be found at the CCW website.31,32 Number of somatic conditions up until the time of the HRS interview was defined as a count of hypertension; congestive heart failure; coronary artery disease, coronary and ischemic heart disease; cardiac arrhythmias; hyperlipidemia; stroke; arthritis; asthma; cancer (breast, prostate, lung, colorectal, blood and endometrial); chronic kidney disease; chronic obstructive pulmonary disease; diabetes mellitus; and osteoporosis. Alzheimer’s disease and related disorders of senile dementia (ADRD) and depression were included separately to address the presence of mental or neurodegenerative conditions. Additional health indicators from HRS included: activities of daily living and instrumental activities of daily living (ADL/IADL) limitations by a continuous score (0-11) (ADL: dressing, walking across a room, bathing, eating, getting in and out of bed, toileting; IADL: preparing hot meals, grocery shopping, using telephone, taking medication, and managing money); body mass index (BMI) categorized as underweight/normal: ≤25 (reference), overweight: >25-30, obese grade I: >30-35, obese grade II: >35-40, obese grade III: ≥40; and proxy interview (no proxy/self-interview as reference).
The two outcomes were identified by Current Procedural Terminology (CPT) codes in CMS during 2-years follow-up from HRS interview date: (1) AWVs (CPT codes: G0438 and G0439) and (2) influenza vaccinations (CPT codes 90630, 90653–90657, 90661, 90662, 90672–90674, 90685–90688, Q2035–Q2039, and G0008). 17 AWVs and influenza vaccinations were formulated as dichotomous variables.
Descriptive statistics were conducted using means with standard deviations for continuous variables and frequencies with percentages for categorical variables for each of the outcomes. Statistical analyses were conducted in 3 steps. First, Conditional Inference Tree (CIT) analyses, implemented in the R package ‘partykit’, 33 were performed to identify combinations of sociodemographic factors, chronic conditions, and other health indicators predicting AWVs and influenza vaccinations. CIT is a non-parametric machine learning method that performs recursive binary partitioning to examine the relationship between multiple explanatory variables and a single outcome. In this process, a decision tree is constructed by testing the null hypothesis of independence between each variable and the outcome. If the hypothesis cannot be rejected or no division can be made without at least 60 individuals (1.2% of our sample) in each group, the algorithm was stopped. Otherwise, the variable with the greatest reduction of heterogeneity in the outcome is selected and a binary split of the variable is performed. The algorithm recursively repeats these steps until the stopping criteria are met. To build these models, we used 75% of the dataset for training and reserved 25% for testing. After constructing a CIT with the training data, we used the test dataset to identified the accuracy, sensitivity, and specificity of the CIT based on the cutoff point on the ROC curve closest to the optimal model based on Youden’s Index, the cutpoint on the ROC curve that optimizes both sensitivity and specificity. We repeated the analyses three times with different random seeds to confirm the robustness of the results. Second, a Conditional Inference Random Forest (CIRF) algorithm was implemented using the R package ‘party’ 34 on the full dataset to test whether the CIT identified the most important variables for predicting the outcomes. This method uses bootstrapping aggregation to create multiple decision trees, each using a random sample of variables as split candidates, and collects their results. Each forest was created using 1500 trees. We reported the ranking of the variable importance identified in the CIRF analyses. Third, multivariable logistic regression analyses were conducted using the full dataset to identify risk estimates (adjusted odds ratios [aORs]) and 95% confidence intervals (95% CI) and quantify the association between all exposure variables and each of the two outcomes.
In the machine learning models, all count and continuous variables (age, educational level, BMI, ADL/IADL, and wealth) were included as continuous variables. The previously described variable categorizations were used in the logistic regression analyses. R code to run CIT and CIRF are provided in the Online Appendix.
Results
Baseline characteristics for the overall population as well as stratifications by AWV and influenza vaccination status are presented in Table 1. In total, 1139 (22.8%) older adults received an AWV and 3292 (65.8%) an influenza vaccination.
Baseline Characteristics of Study Population, n (%).
Figure 1A shows results from the CIT analysis of AWVs during the 2-year period following participants’ HRS interview. It shows that combinations of geographical region, wealth, and education resulted in different proportions of AWV utilization. Individuals living in New England had the highest AWV utilization level of 40.4%. In the other geographical regions, individuals with ≤14 years of education and wealth ≤$159 000 had the lowest AVW utilization of 15.2%. The CIT for this analysis had an accuracy of 55.4% with a sensitivity of 52.8% and specificity of 56.2% in the testing data. Repeating the Conditional Inference Tree analyses with different seeds produced CIT that all selected geographical region, wealth and education, but with different splits resulting in slightly different CIT structures. One CIT additionally included age and another included ADL/IADL limitations.
The CIRF results for AWVs are shown in Figure 2A. They support findings from the CIT analyses, identifying geographical region and wealth as the two most important predictors of AWVs.
Conditional Inference Random Forest for (A) Annual Wellness Visits and (B) Influenza vaccinations. Explanatory note: The graphs illustrate the ranking of variable importance for predictors on the y-axis. The x-axis shows the mean decrease in accuracy if the specific variable is removed from a Conditional Inference Tree. The ranking of the variables are informative, whereas the variable importance value should not be interpreted. 35
Multivariable logistic regression results (Table 2) showed that relative to those living in New England, older adults living in all other regions had lower ORs of receiving AWVs. Residents in West-North Central (aOR: 0.32; 95% CI: 0.22, 0.45), West-South Central (aOR: 0.35; 95%CI: 0.24, 0.49), and Mid-Atlantic (aOR: 0.38; 0.26, 0.54) were associated with the lowest AWV utilization. A dose-response relationship was identified between quintiles of wealth and AWV utilization. Female sex, Hispanic ethnicity, and more years of education were also significantly associated with higher ORs of AWV utilization in the multivariable logistic regression analysis. The concordance statistic (C-statistic) for the multivariable logistic regression was 0.62.
Multivariable Logistic Regression Analyses of Annual Wellness Visits and Influenza Vaccinations.
Bold = statistical significant estimates.
Odds ratio and 95% confidence interval for non-Hispanic black vs. Hispanic: Annual Wellness Visits
Findings from the CIT analysis for receipt of influenza vaccinations during the 2-year period after HRS interview are presented in Figure 1B. These results show that various combinations of race/ethnicity, number of somatic conditions, educational level and partnership status differentially affected influenza vaccination uptake. The lowest levels of influenza vaccination uptake were among (1) Hispanic and non-Hispanic Black beneficiaries with ≤4 somatic conditions (42.1%) and (2) non-Hispanic White beneficiaries with ≤2 somatic conditions and no partner (43.8%). The highest level of influenza vaccination uptake was among non-Hispanic White beneficiaries with >9 somatic conditions (84.4%). The CIT had an accuracy of 64.4% with a sensitivity of 75.3% and a specificity of 43.2%. Repeating the CIT analysis with different seeds produced trees that selected the variables race/ethnicity, number of somatic conditions, and education yet with different splits. None of the other CIT included partnership status and one additionally incorporated ADL/IADL limitations.
The CIRF results for influenza vaccinations are shown in Figure 2B. As we found in CIT analyses, number of somatic conditions, race/ethnicity, educational level, and wealth were identified as the most important predictors of influenza vaccination uptake. In addition, the CIRF analysis identified partnership status as the 5th most important predictor.
Multivariable logistic regression results (Table 2) showed a dose-response relationship for somatic conditions (aOR: 1.19; 95%CI: 1.15, 1.22), years of education (aOR: 1.06; 95%CI: 1.03, 1.08), and wealth in quintiles. Non-Hispanic Black beneficiaries (aOR: 0.53; 95%CI: 0.44, 0.65 and aOR: 0.51; 95%CI: 0.37, 0.70) had lower adjusted odds of influenza vaccinations than non-Hispanic White and Hispanic beneficiaries, respectively. Influenza vaccination uptake did not vary between non-Hispanic White and Hispanic beneficiaries. Older adults with a partner had 19% (95% CI: 1.19, 1.37) higher odds of influenza vaccinations than those without. Female sex and ADL/IADL limitations were also significantly associated with higher adjusted odds of influenza vaccinations. The C-statistic for the multivariable logistic regression was 0.66.
Discussion
We conducted machine learning analyses to explore combinations of predictors for AWVs and influenza vaccinations among US Medicare beneficiaries. The most important predictors of receiving an AWV were geographical region and wealth, whereas number of somatic conditions, race/ethnicity, educational level, and wealth were the most important predictors of influenza vaccinations.
Utilization of AWVs has increased since its introduction in 2011.17-21 In this study, 22.8% received at least one AWV within a 2-year period following their 2012 HRS interview, which is comparable to previous findings in a 5% national sample of Medicare beneficiaries in 2011 to 2013. 14 However, a recent report has shown that AWV uptake for Medical Advantage patients increased from 2011 to 2015, leading to 40% greater uptake in these patients compared to fee-for-service patients. 36 Furthermore, another study found that healthcare practices with medically and socially complex patients provide less AWVs, whereas accountable Care Organizations and practices with higher rates of electronic health record incentive program participation provided more AWVs in 2015. 19 Yet, future studies are needed to determine whether the increase in AWVs since their introduction is explained by diffusion of recommendations from practice guidelines and policy to implementation in the healthcare practices and/or by incentives programs aimed at managed care beneficiaries.
We identified individuals living in New England to have the highest AWV utilization. Interestingly, we identified that no other characteristics were predictive of utilization within this region in the CIT analysis. For individuals living in the other geographical regions, AWV utilization was lowest among beneficiaries with fewer years of education (≤14) and lower levels of wealth (≤$159 000). This illustrates that individual socioeconomic factors, which are related to uptake of other preventive healthcare like influenza vaccination, 24 may be important in geographical regions with lower AWV utilization. To ensure adequate provision and use of preventive services, it is essential that AWVs are equally available and accessible across all geographical regions of the country. When this is met, our findings suggest that socioeconomic factors may be important for utilization.
Our finding that geographical region is the most important predictor of AWVs with highest utilization in New England is supported by a study of a 20% random sample of 2015 Medicare beneficiaries. Ganguli et al. (2018) showed that 51.2% of primary care practices did not provide AWVs; those that provided AWVs were clustered in the Northeast and urban areas, had a more stable patient assignment and a slightly healthier patient mix. Practices with lower AWV rates more often provided care for underserved populations, such as racial minorities and Medicare-Medicaid dual eligibility patients. 19 We also identified that socioeconomic factors including wealth and education were important predictors of AWV utilization. This is indirectly supported in a number of other studies that identified residence in more affluent areas and non-rural metropolitan areas to have higher AWV utilization.11,18,20,21
Hispanic ethnicity and female sex were significantly associated with greater AWV utilization in the multivariable logistic regression analysis. Previous studies of sex are inconsistent.11,21 Studies have shown lower AWV utilization among non-Hispanic Black and other race/ethnicity groups compared to non-Hispanic White Medicare beneficiaries.11,18,20,21 One study that investigated utilization among Hispanic beneficiaries found lower utilization of AWVs in comparison to non-Hispanic White beneficiaries in univariable analysis, but no association in multivariable analysis. 20 The surprising finding from our analysis that AWV utilization was higher in Hispanic beneficiaries compared to non-Hispanic White and Black beneficiaries may not be representative of the full Medicare population. Our study only had 6.3% Hispanics, whereas Hispanics represented 10.7% of the population in the previous study. 20 Hispanic particpants in the present study may be a more selected group and, thus, possibly healthier than the general Hispanic population in the US, because they represent a smaller proportion of the final study population compared to other studies. This may be explained by the strict selection criteria (Figure A1), which may have influenced the inclusion of this race/ethnic group more than non-Hispanic white and black participants.
This study elucidated that AWVs do not reach all Americans and utilization was especially low among beneficiaries with lower levels of wealth and education in geographical regions with overall low AWV utilization. This could lead to unforeseen negative consequences by increasing sociodemographic health disparities among older adults in the US. Due to the risk of increased health disparities, Tipirneni et al suggest implementation of a new version of AWVs that targets the root causes of poor health covering individual, social, and behavioral determinants of health in addition to addressing cognition, balance, and vision as predictors of poor health. 37 Our findings and those from previous studies suggest that a first step to increase access to and use of AWVs may be to increase incentives for healthcare providers throughout the US to administer AWVs.
Influenza vaccination coverage of 65.9% during the 2-year period in this study is in line with coverage (66.2% ± 0.8%) among adults aged ≥65 years in the US during the 2012-2013 season. 38
We identified number of somatic conditions, race/ethnicity, educational level, and wealth as important predictors of influenza vaccination uptake. Uptake was highest among non-Hispanic White beneficiaries with >9 somatic conditions. By contrast, influenza vaccination uptake was low among all race/ethnic groups with few somatic conditions – that is, non-Hispanic White beneficiaries with ≤2 somatic conditions and no partner and Hispanic and non-Hispanic Black beneficiaries with ≤3 somatic conditions. This may suggest an interplay between somatic conditions and race/ethnicity for influenza vaccination coverage.
Previous studies have also identified chronic conditions as positive predictors of influenza vaccination uptake in the US; 26 however, findings for single chronic conditions such as dementia, diabetes, and asthma are inconsistent.22,25,26 In this study, neither ADRD nor depression were identified as predictors of influenza vaccinations. Individuals with medical conditions are at the highest risk of complications including death from influenza infections; 39 hence, it is an important and positive public health finding that older adults with more somatic conditions have greater uptake.
In our multivariable logistic regression, we found that non-Hispanic Black beneficiaries had lower influenza vaccination uptake than both non-Hispanic White and Hispanic beneficiaries. Influenza vaccination, surprisingly, did not differ between non-Hispanic White and Hispanic beneficiaries, which may be explained by a healthy selection of Hispanics in our study population as previously discussed. Prior studies generally showed lower influenza vaccination uptake among race/ethnic minorities in the US. Two studies have shown that non-Hispanic Black and/or Hispanic older adults had lower uptake than non-Hispanic White older adults,15,23 and one study showed that non-Hispanic Black older adults more often refused influenza vaccination than non-Hispanic White older adults. 27 However, a fourth study showed that Hispanics had the lowest uptake of racial/ethnic groups examined and were less aware of recommendations, less informed about influenza virus and benefits of vaccination, and the least confident about the vaccine compared to both non-Hispanic Black and White older adults. 22 Lower influenza vaccination uptake in minority groups may be explained by differences in community-specific attitudes toward vaccination, inadequate access to care, marginalization, and lack of trust in the healthcare system. 24 The impact of sex on influenza vaccination coverage is mixed,15,25,27 whereas Medicare-Medicaid dual eligible beneficiaries have been found to have lower coverage. 15
A number of literature reviews have proposed mechanisms that may lead to underutilization of influenza vaccinations. In summary, a positive attitude toward influenza vaccination, high perceived utility and safety of vaccination, previous severe influenza experiences, cues to action including advice from health professionals and kin networks, habits (e.g., previous influenza vaccination uptake), and practical barriers including transport and access issues for the oldest older adults were identified as the major and most consistent factors influencing influenza vaccination uptake.24,40-42 The largest of the systematic reviews (N = 470 studies) argued that sociodemographic factors are only indirectly related to influenza vaccination utilization. 41
Our findings and those from previous studies show a continued need for the healthcare system and public health authorities to strive toward greater influenza vaccination coverage among older adults in the US. This is supported by a qualitative meta-analysis of 14 years of influenza-related communication research by U.S. Centers for Disease Control and Prevention (CDC). The qualitative meta-analysis showed that many people have an aversion toward influenza vaccinations and tend to overestimate the effect of other actions to decrease the risk of infection with influenza virus. 42 Further, the positive impact of AWVs on preventive healthcare utilization10-17 may suggest that increased implementation of AWVs throughout the US may help improve influenza vaccination uptake.
This study has several strengths. Through linkage of HRS and CMS, we were able to identify a comprehensive number of potential predictors of AWVs and influenza vaccinations including sociodemographic characteristics, chronic conditions, and other health indicators. We included information on chronic conditions and use of preventive healthcare services from CMS, which are not subject to recall bias. We were able to explore our research questions and the impact of multiple predictors by leveraging machine learning models instead of restricting our hypothesis to a priori knowledge. In CIT analyses, we obtained empirical information about the combinations of variables important for AWV utilization and influenza vaccination uptake, not restricted to a linear relationship. The differences in the two analytical approaches (logistic regression models versus machine learning models) resulted in slightly different findings, which is reasonable due to the different criteria for the two analytical approaches. The logistic regression model is a parametric model that builds on a specific form and estimates parameters based on the data. The logistic regression model is defined and the effects of the variables are quantified from the fitted data. On the other hand, the machine learning methods make predictions based on the data where there are no assumptions about specific functional forms or inferential models. Instead, an algorithm is presented with the training inputs and desired outputs and executes all steps to develop a rule to map inputs to outputs. These machine learning methods provide automated ways of assessing complex data for important patterns. In this study, the machine learning methods were applied to identify key predictors of the two outcomes, whereas the logistic regression model was used to quantify the relationships.
The generalizability of our results to all older Americans may be hampered by restricting the study population to Medicare FFS beneficiaries aged ≥65 years with ≥3 years of enrollment in the Medicare fee-for-service program at baseline. This was necessary to identify chronic conditions by the CCW algorithms31,32 and subsequent preventive healthcare use (Figure A1). There were also insufficient numbers of American Indian/Native Alaskan or Asians in HRS to include in the models, limiting inference for these groups. Although these approaches provide insights to factors associated with AWV and influenza vaccination, they do not infer a causal relationship. Finally, the study did not elucidate potential mediating factors explaining the sociodemographic factors and morbidities as predictors of AWVs and influenza vaccination. Especially, healthcare utilization may be an important factor explaining the findings, which should be investigated in future studies.
Conclusion
AWVs and influenza vaccinations are underutilized among the US Medicare population. Applying CIT and CIRF analyses, we identified geographical region and wealth as important factors for AWV utilization, whereas number of somatic conditions, race/ethnicity, wealth and educational level were predictive for influenza vaccination uptake. The importance of geographic region for AWV utilization suggests that this service was unequally adopted. Non-Hispanic black participants and/or those with functional limitations were less likely to receive influenza vaccination.
Supplemental Material
Appendix – Supplemental material for Annual Wellness Visits and Influenza Vaccinations among Older Adults in the US
Supplemental material, Appendix for Annual Wellness Visits and Influenza Vaccinations among Older Adults in the US by Terese Sara Høj Jørgensen, Heather Allore, Miriam R. Elman, Corey Nagel, Mengran Zhang, Sheila Markwardt and Ana R. Quiñones in Journal of Primary Care & Community Health
Supplemental Material
eAppendix_R_code – Supplemental material for Annual Wellness Visits and Influenza Vaccinations among Older Adults in the US
Supplemental material, eAppendix_R_code for Annual Wellness Visits and Influenza Vaccinations among Older Adults in the US by Terese Sara Høj Jørgensen, Heather Allore, Miriam R. Elman, Corey Nagel, Mengran Zhang, Sheila Markwardt and Ana R. Quiñones in Journal of Primary Care & Community Health
Footnotes
Authors’ Note
A previous version of the results was presented as a poster presentation on November 14th at the Gerontological Society of America 2019 Annual Scientific Meeting.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The work was supported by the National Institute of Aging [R01 AG055681-02S1, P30 AG021342, R01 AG047891]. Terese Sara Høj Jørgensen was also supported by the Social Inequality in Aging (SIA) project, funded by NordForsk [project number: 74637]. The sponsors had no role in the study design; the collection, analyses and interpretation of data; in the writing of the report; and in the decision to submit the article for publication.
Supplemental Material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
