Association of an Index of Healthy Aging With Incident Cardiovascular Disease and Mortality in a Community-Based Sample of Older Adults

Abstract

Background:

The healthy aging index (HAI) was developed as a marker of health in multiple systems that can identify individuals who age most successfully.

Methods:

We calculated an HAI in 934 Framingham Offspring Study participants aged 60 or older at baseline. Heart rate and C-reactive protein (CRP) were added in modified versions of the HAI. Cox proportional hazard models were used to quantify the association of the HAI with mortality, cardiovascular disease (CVD), and cancer. We used fully conditional specification to multiply impute missing values for HAI components, increasing the sample size by 44%.

Results:

Over 10 years of follow-up, there were 138 deaths, 103 incident cases of CVD, and 138 incident cases of cancer. In models adjusted for age, sex, and behavioral risk factors, the HAI was associated with mortality (hazard ratio [HR] per unit of HAI 1.24, 95% confidence interval [CI] 1.13–1.36) and with CVD (HR 1.27, 95% CI 1.13–1.42), but not with cancer (HR 1.01, 95% CI 0.91–1.11) in observed (non-missing) data. In multivariable models further adjusting for prevalent diseases, results were slightly attenuated. When including heart rate and CRP, a modified HAI gave stronger associations. Results with imputed data are similar to results from complete case analyses.

Conclusions:

In our large community-based sample, the HAI is a strong predictor of mortality and CVD. Other factors that are strongly associated with mortality, such as heart rate and CRP can improve the ability of the HAI to identify the healthiest older adults.

The healthy aging index (HAI), a modified version of the physiologic index of comorbidity (1), was developed in epidemiologic studies using noninvasive measures across multiple organ systems (2,3). The HAI score, which ranges from 0 (most healthy) to 10 (least healthy), is associated positively with mortality and disability in the Cardiovascular Health Study (CHS) and Health, Aging, and Body Composition Study (1–3) and negatively with longevity in the Long Life Family Study (3). All of these studies were conducted using participants with advanced old age and fairly symmetric HAI values (ie, few participants were characterized as very healthy or very unhealthy).

We examined the association of the HAI with all-cause mortality and incident cardiovascular disease (CVD) and cancer in a sample of participants in earlier old age from the Framingham Heart Study (FHS) Offspring Cohort. We hypothesized that participants with higher scores, indicating a greater burden of disease across multiple physiologic systems, would not only have increased mortality but have a higher incidence of CVD and cancer, the two leading causes of death in older adults. We assessed how including C-reactive protein (CRP), an established marker of systemic inflammation (4), and resting heart rate, a recognized measure of overall autonomic function and physical fitness (5), in the HAI may modify the observed associations between the HAI and mortality or CVD. CRP is particularly interesting because chronic inflammation is considered a marker of biological aging across multiple organ systems (6) and is a common pathway to all-cause mortality (7,8). Finally, we examined whether multiple imputation methods are appropriate to increase sample size for the HAI in this study of older participants under longitudinal follow-up. Missing data are a common problem, with up to 31% of the sample missing an individual component in prior reports (3).

Methods

Study Sample

The FHS Original Cohort participants were enrolled, beginning in 1948, to study risk factors for CVD. In 1971, the Offspring and Offspring spouses of the Original cohort were enrolled in the Framingham Offspring Study; they have been examined every 4 years–8 years (9). Of the 3,799 Framingham Offspring participants who attended the fifth examination cycle (1991–1995), 1,348 were at least 60 years of age and eligible for this study. This was the first exam where all the HAI components were measured, and we limited our study to participants aged 60 years and older because the HAI was developed in an older population and does not show much variability in younger adults. Complete HAI and covariate data were available for 934 participants. The Institutional Review Board of Boston University Medical Campus approved all study protocols, and all participants provided written informed consent.

Offspring Exam 5 Data

At each research examination, participants underwent a physician-administered medical history interview and physical examination including measurement of resting blood pressure, technician-administered questionnaires including cognitive function, pulmonary function testing, and laboratory measurements. The HAI comprised information from five physiologic systems: systolic blood pressure (cardiovascular), fasting glucose (metabolic), creatinine (kidney), Mini-Mental State Exam (brain), and forced vital capacity (pulmonary). Systolic blood pressure was computed by averaging two physician-obtained measurements, while fasting glucose and serum creatinine were collected from routine laboratory tests (10). The Mini-Mental State Exam was administered by trained interviewers following a standard protocol (11). Forced vital capacity was collected according to American Thoracic Society standards using a Collins Survey II spirometer (SandM Instruments, Doylestown, Pennsylvania) (12,13).

Healthy Aging Index

Data values for each component were arranged into three groups; the least healthy group was assigned a score of 2, the middle group a score of 1, and the healthiest group a score of 0. The component scores were summed to create the HAI, ranging from 0 (healthiest) to 10 (unhealthiest). Cutoffs for the three groups were replicated from the work of Sanders and coworkers (3), which used clinical cut points for glucose, tertiles from their samples for systolic blood pressure and Mini-Mental State Exam, and sex-specific tertiles for forced vital capacity and serum creatinine (see Supplemental Table 1). Participants using related medications or who had previously reported being diagnosed with a relevant disease were assigned a score of 2 for that component (see Supplemental Methods).

We investigated the relation of HAI with several other measures previously reported to be associated with mortality, including subjective health (14), CRP (4,15), and heart rate (5) (see Supplemental Methods).

In secondary analyses, three modified HAIs were constructed, adding CRP and heart rate, individually and then combined. We constructed tertiles based on all exam participants. Participants who had a pacemaker or reported use of cardiac medications were assigned a score of 2 for heart rate (see Supplemental Methods). We added these components to the original HAI, which ranged from 0 (healthiest) to 12 (unhealthiest) when adding in CRP or heart rate individually, or 0 to 14 when including both CRP and heart rate.

Covariates

Covariates include body mass index, smoking status, physical activity index, hypertension, diabetes, prevalent CVD (in all-cause mortality and cancer models), kidney disease, and pulmonary disease (see Supplemental Methods).

Outcomes: Mortality, CVD, and Cancer

The FHS follows all participants for CVD events and death. A panel of three physicians (or a panel of study neurologists for cerebrovascular outcomes) adjudicates all suspected CVD events and deaths using data collected from FHS examinations, hospitalization records, and physician office visit records (16). For this study, the three outcomes of interest are all-cause mortality, and incidence of cancer or CVD (defined as myocardial infarction, coronary insufficiency, stroke, cerebral embolism, and death due to CVD). The FHS validates cancer diagnoses with pathology reports; 5% of cancer cases in this study were validated based on death certificate or clinical diagnosis alone. Follow-up time was limited to the minimum of 10 years, date of event, date of death (in CVD and cancer analyses), or date of last contact.

Statistical Methods

We summarized continuous variables using means and standard deviations (SD), except for CRP, for which we used median and first and third quartiles due to the skewed distribution. We used frequency and percent to summarize categorical variables. Clinical characteristics were provided for the entire group of eligible participants (n = 1,348), and subsets with complete data (n = 934), or incomplete data (n = 414), with p values for comparison provided from t-tests (continuous), χ² tests (categorical), or Wilcoxon Rank Sum tests (CRP and self-reported health). We analyzed the HAI per unit of index and as a categorical variable (1–3).

To assess whether the HAI is predictive of any of the outcomes of interest (all-cause mortality or incident CVD or cancer), we used two different Cox proportional hazards regression models. The first model adjusts for age, sex, and behavioral risk factors: physical activity index, current smoking status, and body mass index. The second model also adjusts for comorbidities: prevalent cancer, hypertension, diabetes, CVD, kidney disease, and pulmonary disease. C-Statistics were used as a descriptive method (17) to assess model performance and the predictive power of HAI over traditional risk factors. We removed participants with prevalent CVD or cancer from analyses with respective outcome.

We examined the association of prevalent CVD, self-reported health, heart rate, and CRP with the HAI and its components. We used Spearman correlation to see how these measures tracked with HAI and to ensure that they were not highly correlated with the original components. Heart rate and CRP were also used as predictors in a series of three Cox regression models, including the two models described above, and a model adjusting for age, sex, behavioral risk factors, and HAI. We created modified indices of HAI by adding in heart rate and CRP as additional components. Analyses on the modified HAIs were the same as for the original HAI (described above). For the analyses per category of index, we kept the HAI groups the same when adding in only one additional component (extending the last group to include HAI values of 7–12). However, when we added both heart rate and CRP, we created an additional category for HAI values 11–14.

Due to missing data in some components (notably forced vital capacity and creatinine), we also used multiple imputation by fully conditional specification (18). Blom’s method of rank normalization was performed on all continuous variables prior to imputation (19). All HAI components were imputed using linear regression models (see Supplemental Methods and Supplemental Table 2). We looked at the association of the HAI with mortality and CVD using the imputed data in Cox regression models, and we compared the results to the analyses run on the observed data. All analyses were performed using SAS version 9.3 (SAS Institute, Cary, North Carolina).

Results

The mean age of the 934 participants with complete HAI and covariate data was 66 years, and 51% were female (Table 1). The distribution of HAI was right skewed with first and third quartiles of two and four, respectively; 39% of the participants fell into the healthiest HAI group with values less than 2 (Supplemental Figure 1). There were 414 participants with incomplete data: they were older and had a higher prevalence of CVD (12% vs 8%) and kidney disease (21% vs 14%) but lower prevalence of pulmonary disease (7% vs 13%) than those with complete HAI data.

Table 1.

Sample Characteristics

	Total Sample	Complete Data	Incomplete Data	p Value Complete vs Incomplete
	N = 1,348	n = 934	n = 414
Clinical characteristics
Age, y (min = 60)	65.9±4.5	65.6±4.4	66.6±4.8	0.0001
Female, N (%)	699 (52)	474 (51)	225 (54)	0.22
Body mass index, kg/m2	27.7±4.8	27.6±4.7	27.8±4.9	0.63
Physical activity index	33.8±5.7	33.8±5.6	33.7±6.1	0.80
Current smoker, N (%)	192 (14)	133 (14)	59 (14)	0.96
Cancer, N (%)	121 (9)	83 (9)	38 (9)	0.86
Hypertension, N (%)	721 (54)	489 (52)	232 (57)	0.14
Treatment for hypertension, N (%)	473 (35)	315 (34)	158 (39)	0.07
Cardiovascular disease, N (%)	128 (10)	77 (8)	51 (12)	0.02
Diabetes, N (%)	171 (13)	115 (12)	56 (14)	0.38
Treatment for diabetes, N (%)	92 (7)	59 (6)	33 (8)	0.26
Kidney disease, N (%)	177 (15)	131 (14)	46 (21)	0.02
Pulmonary disease, N (%)	150 (11)	123 (13)	27 (7)	0.0003
CRP, mg/dLa	2.6 (0.8, 6.8)	2.5 (0.7, 6.6)	2.8 (0.9, 7.4)	0.20
Heart rate, beats/min	66±12	66±12	67±12	0.07
Self-reported health				0.08
Excellent	509 (38)	367 (39)	142 (35)
Good	676 (50)	466 (50)	210 (52)
Fair	137 (10)	90 (10)	47 (12)
Poor	17 (1)	10 (1)	7 (2)
HAI components
Systolic blood pressure, mm Hg	134±20	134±19	134±20 (N = 413)	0.85
Fasting glucose, mg/dL	108±35	107±33	110±39 (N = 379)	0.09
Creatinine, mg/dL	0.92±0.32	0.92±0.31	0.95±0.36 (N = 225)	0.28
Forced vital capacity, L	3.47±0.85	3.48±0.86	3.43±0.82 (N = 180)	0.46
Mini mental status exam	28.4±1.8	28.5±1.7	28.2±2.0 (N = 405)	0.004
HAI	3.1±1.8	3.1±1.8	—	—

Notes. HAI = Healthy Aging Index, CRP=C-reactive protein. Values are shown as means ± standard deviation or percentages.

^aCRP values are shown as median (25th, 75th percentiles).

Association of HAI with mortality, incident CVD, and cancer

In the multivariable model adjusting for behavioral risk factors (Model 1), each unit of HAI was associated with a 24% higher hazard for death over an average follow-up time of 9.33 (SD = 1.95) years (Table 2). The HAI also improved prediction of mortality, increasing the C-statistic substantially from 0.703 to 0.730 in Model 1. The hazard ratios (HRs) for the categorized HAI showed a similar gradient: compared with the reference group (HAI 0–2), the group with HAI 7–10 had 3.4-fold greater risk of mortality in Model 1 (Table 2). In the multivariable model, further adjusting for comorbidities (Model 2) there was a 7% attenuation in the HAI effect (HR = 1.15, 95% confidence interval [CI]: 1.00–1.34) and including HAI incremented the C-statistic by 0.007. The rates of mortality increased across HAI category, as expected (Supplemental Table 4).

Table 2.

Associations of Modified HAI With All-Cause Mortality

Outcome	Events/N	HR (95% CI) Model 1	HR (95% CI) Model 2
Original HAI
HR per unit of index	138/934	1.24 (1.13,1.36)	1.15 (1.00,1.34)
HR per category of index
0–2	28/364	1.00	1.00
3–4	55/368	1.69 (1.06,2.68)	1.48 (0.89,2.48)
5–6	42/166	2.99 (1.81,4.92)	2.23 (1.13,4.40)
7–10	13/36	3.41 (1.70,6.83)	2.29 (0.89,5.88)
HAI including heart rate
HR per unit of index	132/914	1.24 (1.15,1.35)	1.22 (1.08,1.37)
HR per category of index
0–2	12/193	1.00	1.00
3–4	23/277	1.30 (0.64,2.62)	1.24 (0.60,2.53)
5–6	52/271	2.88 (1.53,5.46)	2.62 (1.30,5.29)
7–12	45/173	3.95 (2.05,7.62)	3.03 (1.30,7.05)
		—	—
HAI including CRP
HR per unit of index	132/914	1.25 (1.15,1.36)	1.21 (1.07,1.36)
HR per category of index
0–2	11/202	1.00	1.00
3–4	35/314	1.88 (0.96,3.72)	1.67 (0.83,3.37)
5–6	44/247	2.81 (1.43,5.51)	2.37 (1.13,4.95)
7–12	42/151	4.88 (2.45,9.74)	3.59 (1.53,8.42)
		—	—
HAI including heart rate and CRP
HR per unit of index	132/914	1.24 (1.16,1.34)	1.24 (1.12,1.37)
HR per category of index
0–2	7/115	1.00	1.00
3–4	9/203	0.72 (0.27,1.95)	0.64 (0.24,1.75)
5–6	39/261	2.48 (1.10,5.57)	2.36 (1.02,5.45)
7–10	67/310	3.51 (1.58,7.79)	2.95 (1.21,7.18)
11–14	10/25	6.57 (2.43,17.81)	4.88 (1.42,16.80)

Notes. HAI=Healthy Aging Index; HR = hazard ratio; CRP = C-reactive protein; CVD = cardiovascular disease; CI = confidence interval. Model 1 is adjusted for age, sex, physical activity index, smoking status, and body mass index. Model 2 is adjusted for the covariates in Model 1 and baseline cancer, hypertension, CVD, diabetes, kidney disease, and pulmonary disease.

In the subset of participants without prevalent CVD, association of HAI with incident CVD mirrored the results above with a mean (SD) follow-up time of 9.03 (2.28) years. In Model 1, each unit of HAI was associated with an HR of 1.27 (95% CI: 1.13–1.42) for CVD (Table 3) and showed a substantial increase in the C-statistic from 0.670 to 0.703. Associations were slightly attenuated in Model 2, with a 20% (95% CI: 1.01–1.42) increase per unit HAI in hazards for CVD and a 0.011 increase in the C-statistic. In the subset of participants without prevalent cancer, no association of the HAI with incident cancer in this sample of FHS participants was observed (Supplemental Table 5).

Table 3.

Associations of Modified HAI With CVD

Outcome	Events/N	HR (95% CI) Model 1	HR (95% CI) Model 2
Original HAI
HR per unit of index	103/857	1.27 (1.13, 1.42)	1.20 (1.01, 1.42)
HR per category of index
0–2	24/349	1.00	1.00
3–4	43/340	1.62 (0.97, 2.70)	1.41 (0.79, 2.53)
5–6	29/143	2.79 (1.58, 4.91)	1.98 (0.90, 4.36)
7–10	7/25	3.31 (1.38, 7.93)	2.47 (0.84, 7.25)
HAI including heart rate
HR per unit of index	103/839	1.25 (1.14, 1.38)	1.23 (1.07, 1.41)
HR per category of index
0–2	12/188	1.00	1.00
3–4	21/264	1.16 (0.57, 2.38)	1.10 (0.52, 2.32)
5–6	39/246	2.33 (1.20, 4.51)	1.98 (0.93, 4.24)
7–12	31/141	3.24 (1.62, 6.49)	2.31 (0.90, 5.93)
HAI including CRP
HR per unit of index	103/839	1.26 (1.14, 1.39)	1.23 (1.08, 1.41)
HR per category of index
0–2	6/197	1.00	1.00
3–4	36/294	3.72 (1.56, 8.85)	3.54 (1.46, 8.62)
5–6	33/227	4.03 (1.66, 9.76)	3.49 (1.35, 9.07)
7–12	28/121	7.78 (3.14, 19.26)	6.32 (2.20, 18.19)
HAI including heart rate and CRP
HR per unit of index	103/839	1.24 (1.14, 1.35)	1.23 (1.10, 1.39)
HR per category of index
0–2	4/112	1.00	1.00
3–4	13/196	1.85 (0.60, 5.69)	1.81 (0.58, 5.61)
5–6	33/244	3.80 (1.34, 10.79)	3.52 (1.20, 10.32)
7–10	48/272	4.94 (1.75, 13.94)	3.82 (1.22, 11.95)
11–14	5/15	9.42 (2.44, 36.38)	10.04 (2.18, 46.36)

Notes. HAI = Healthy Aging Index; CVD = cardiovascular disease; HR = hazard ratio; CI = confidence interval. Model 1 is adjusted for age, sex, physical activity index, smoking status, and body mass index. Model 2 is adjusted for the covariates in Model 1 and baseline cancer, hypertension, diabetes, kidney disease, and pulmonary disease.

Consideration of Additional Measures for HAI

The correlation between HAI and self-reported health was 0.28 (Supplemental Table 6a), and the two variables had similar distributions (Supplemental Figure 1). Participants who believed themselves to be in poorer health had higher HAI values (Supplemental Figure 2).

The Spearman correlations of heart rate, CRP, self-reported health, and prevalent CVD with the HAI (Supplemental Table 6a) indicated that the HAI tracked with other predictors of mortality. The HRs for heart rate and CRP across the series of Cox models remained consistent. After adjusting for HAI, the HR for heart rate (per 1 SD difference) was 1.39 (95% CI: 1.20–1.61), and the HR for CRP (per 1 SD of log difference) was 1.41 (95% CI: 1.19–1.67), indicating these markers were likely to increase the ability of HAI to predict mortality and CVD (Supplemental Table 7). Since these markers were not highly correlated with the original HAI components (Supplemental Table 6a), we created three modified HAI. The HRs per unit of index and per category of HAI were consistent across the models predicting mortality using the original and modified HAI (Table 2), though the C-statistics were 0.1–0.2 higher for the modified versus the original HAI. Including heart rate and CRP in the HAI maintains stable HR per unit of index when further adjusting for comorbidities in Model 2 (Table 2) and substantially increased the C-statistic from 0.728 to 0.753 (Supplemental Table 3). Compared with participants in the reference group (HAI 0–2), those with a modified HAI between 7 and 12 were at a 4-fold increased risk of mortality with heart rate included in the HAI and a 5-fold increased risk of mortality with CRP included. With both components included, participants with a modified HAI between 11 and 14 were at a 6.6-fold increased risk of mortality. Similar results were seen for CVD; however, the group with HAI 11–14 had a 10-fold increase in hazards for CVD as compared to the reference group (HAI 0–2) in Model 2 when using the HAI with both heart rate and CRP included (Table 3).

Imputation of Missing Values

In tertiary analyses, we used fully conditional specification to impute missing values for each component of the HAI. By imputing missing values, the sample size increased by 44% and the number of deaths by 69%. The results per unit of HAI were very similar to those in the complete case analysis, with a 26% increase in mortality per unit of HAI in Model 1 (Table 4). The results by HAI group were also similar to complete case analysis but with a steeper gradient in HR between the 5–6 and 7–10 groups (Figure 1). Participants with HAI values between 7 and 10 had 3.7-fold greater risk of mortality compared with the reference group (HAI 0–2). Results per unit of HAI were also similar to the complete case analysis for CVD with one anomaly: the HR for the 7–10 group was lower due to a small sample size, and of the 10 people imputed into this group, only 1 had CVD (Supplemental Figure 3).

Table 4.

Associations in Imputed Data Between HAI and All-Cause Mortality and CVD

Outcome	Events/N a	HR (95% CI) Model 1	HR (95% CI) Model 2
All-cause mortality
HR per unit of index	233/1,348	1.26 (1.17, 1.36)	1.20 (1.07, 1.35)
HR per category of index
0–2	51/523	1.00	1.00
3–4	86/526	1.50 (1.03, 2.19)	1.39 (0.91, 2.13)
5–6	72/246	2.70 (1.84, 3.96)	2.27 (1.35, 3.82)
7–10	24/53	3.66 (2.07, 6.45)	2.62 (1.18, 5.82)
CVDb
HR per unit of index	153/1,220	1.29 (1.17, 1.43)	1.14 (0.99, 1.33)
HR per category of index
0–2	33/498	1.00	1.00
3–4	64/480	1.85 (1.19, 2.87)	1.49 (0.91, 2.44)
5–6	48/207	3.52 (2.08, 5.95)	2.06 (0.98, 4.33)
7–10	8/35	3.02 (1.26, 7.23)	1.68 (0.63, 4.47)

HAI = Healthy Aging Index; CVD = cardiovascular disease; HR = hazard ratio; CI = confidence interval. Model 1 is adjusted for age, sex, physical activity index, smoking status, and body mass index. Model 2 is adjusted for the covariates in Model 1 and baseline cancer, hypertension, CVD (in all-cause mortality models), diabetes, kidney disease, and pulmonary disease.

^aMean # events/N per imputation.

^bCVD models exclude participants with prevalent CVD.

Figure 1.

Hazard ratio for all-cause mortality by HAI group in multivariable models comparing observed versus imputed.

Discussion

In our community-based sample of older adults, our main findings are threefold. First, the HAI was a strong predictor of mortality and incident CVD, but not cancer. Second, incorporation of heart rate and CRP into the HAI reduced the attenuation of the association of the HAI with mortality and CVD when adjusting for comorbidities, improved prediction, and better defined the risk in the unhealthiest HAI group. Third, multiple imputation methods worked well in this setting.

Previous studies have shown the HAI is associated with mortality, incident disability, mobility limitations, slow gait speed, and a decline in gait speed (1–3,20). The ability of HAI to predict mortality and multiple other age-related outcomes suggests that it can distinguish between individuals who age well and those who do not. Compared with previous studies done on HAI (in CHS (1) and Long Life Family Study (3)), our sample is 10 years younger, on average, and has a more balanced gender distribution. The FHS is also not as restrictive as the Health, Aging, and Body Composition study (70–79 years old and high functioning) (2). Our mortality results mirror those of Newman and coworkers in the CHS (1) but had a weaker association of HAI with mortality than Sanders and coworkers found in the Health, Aging, and Body Composition study (2) and CHS (3). We further extend the literature by showing that the HAI is associated with CVD.

Components of the original HAI represent signs of end-organ dysfunction rather than any specific process by which dysfunction develops. Biomarkers, such as CRP and heart rate, have been proposed as markers of the biological aging process that are not specific to a particular organ system and may be more representative of a global aging process. As our data suggest, biomarkers appear to add information about the biological aging process that is not necessarily represented by functional measures of any particular organ system(s). Capturing unhealthy aging processes that are present, even when measures of end-organ dysfunction are unchanged or only incrementally altered, could provide a more sensitive and potentially clinically useful metric of unhealthy aging.

We used multiple imputation by fully conditional specification to impute missing values, increasing the sample size. A comparison of those with and without HAI supports the assumption that the data are missing at random, with some relation between missingness and observed variables: there are statistically significant differences for age, prevalent CVD, kidney disease, pulmonary disease, and Mini-Mental State Exam between those with complete HAI and those missing at least one component. Therefore, we believe that the assumptions for multiple imputation hold.

Several limitations of our study merit consideration. FHS is a predominantly white population, limiting generalizability to other race/ethnic groups. We also did not have reliable data on all relevant comorbidities, and pulmonary disease is based mainly on self-report. Model 2 adjusts for several comorbidities that are associated with the HAI and may be overadjusted. An inadequate number of event subtypes precluded analyses of the HAI association with specific causes of death. Finally, in some individuals living in the community, very low heart rates may yet be associated with increased risk for adverse events, so analyses using heart rate as a biomarker of aging should be interpreted with caution.

Nevertheless, this study has several strengths. We replicated the association between HAI and mortality (1,2), extending prior results to a study of all comers who are in earlier older age and have a higher percent of females. To our knowledge, this is the first use of multiple imputation for the HAI, and it enabled us to increase the sample size and number of events. Since missing data is a common problem in studies of the aging population, it may benefit other studies.

In summary, we found an association between the HAI and mortality in a community-based study of folks in early older age as well as an association between the HAI and CVD, the leading cause of death among older people (21). Including heart rate and CRP as additional components created a modified HAI that has improved prediction of CVD per category of index, indicating that adding these to the HAI may be a valuable next step.

Supplementary Material

Please visit the article online at http://gerontologist.oxfordjournals.org/ to view supplementary material.

Funding

This work was supported by National Institutes of Health (contracts N01-HC-25195, HHSN268201500001I; grants R00-HL-107642 to S.C., R01-AG-29451 to J.M.M.); National Institute on Aging (grants 2U01AG023755-09 to K.L.L., A.B.N., J.M.M., R01-AG-023629 to A.B.N.); Ellison Foundation to S.C.; National Institute of General Medical Interdisciplinary Training Grant for Biostatisticians (T32 GM74905 to E.L.M.).

Conflicts of Interest

The authors have no conflict of interest to declare.