Methods

Study population and measures

Data were from the Women’s Health and Aging Studies (WHAS) I and II, two prospective population-based cohort studies which recruited community-dwelling older women, complementary with respect to physical function status from the same Medicare eligibility sampling frame. Participants in WHAS I were eligible if they were aged ⩾65 years, had difficulty in two or more areas of physical function (thus representing the one-third most disabled older women in the community) and had a Mini-Mental State Examination score of at least 18. Participants in WHAS II were eligible if they were aged 70–79 years, had difficulty in at most one area of physical function (thus drawn from among the two-thirds of the least disabled women in the community) and had a Mini-Mental State Examination score of at least 24. WHAS participants were randomly selected from a Medicare sampling frame from 12 zip codes of Eastern Baltimore City and Baltimore County. Baseline assessments were performed in 1992–5 in WHAS I and 1994–6 in WHAS II. Seventy-one percent of the 1409 women who were screened and eligible for WHAS I and 49.5% of the 880 women who were screened and eligible for WHAS II agreed to participate. WHAS II used telephone recruitment rather than face-to-face recruitment which probably accounts for the lower participation rate. Detailed information on combining both samples and studies using the combined samples can be found elsewhere.7 16 17 18 19

Sampling weights

Data were weighted in all analyses to account for sampling design and to correct for non-participation. The weights were calculated for each participant based on the probability of selection into the study.16 These probabilities varied based on disability status, age and race. The sampling weights allow reference to the entire population of community-dwelling older women in and around Baltimore, Maryland.

Measures

Measurement of frailty

Frailty status, our outcome measure, was defined as frail, intermediately frail or robust using standardised criteria6 7 relating to weight loss/underweight, slow walking speed, weakness, self-reported exhaustion and low physical activity (see Fried et al6 for categorisation details). Weight loss/underweight was determined by ⩾10% weight loss since age 60 or body mass index <18.5 kg/m². The lowest quintile of customary pace walking speed was measured in m/s over 4 m. Grip strength was measured using a Jamar Dynamometer (model BK-7498; Fred Sammons Inc, Burr Ridge, Illinois, USA). Exhaustion was measured by whether a participant had a positive answer to any of the following three items: description of low energy level (defined as <3 on a 0–10 Likert scale with 0 being “no energy” and 10 being “the most energy you have ever had”), description of being unusually tired in the last month or unusually weak in the last month. Energy expenditure was measured with the Minnesota Leisure Time Activity Questionnaire; participants who reported ⩽90 kcal expended per week based on six activities such as walking and household chores were defined as exhibiting low energy expenditure.20 Participants were considered “frail” if they were positive for three of more of the above characteristics, “intermediately frail” if they had 1–2 of the same characteristics and “robust” if they had none. This categorisation has been validated in previous work.6 7

Socioeconomic status

SES was measured using education and household income. Education was grouped as <12, 12 or >12 years. Income was reported by participants; data for income were missing from 107 of the total 727 subjects. These missing data were imputed by a composite hot deck method using a regression with variables such as poverty status, receipt of Medicaid, age, race and number in household to predict income.21 We divided income into three categories of equal numbers of participants. These income cut-offs were <$10 000, ⩾$10 000–<22 250 and ⩾$22 250 per year.

Covariates

Covariates included a range of factors that have been associated with both SES and frailty: race, age, insurance status, smoking history and numbers of chronic diseases. Race was coded as black or white based on participant self-report. Age was measured in years and restricted to 70–79 (the age range of WHAS II). Smoking history was classified into the following three categories: never smoked, former smoker and current smoker. Insurance status was measured as Medicare Part A only (uninsured except for hospitalisation), Medicare/Medicaid or Medicare plus private health insurance. These three categories correspond to uninsured, insured for public providers and completely insured to see any healthcare provider.

Diseases and conditions were adjudicated by two physicians based on examination, medication list, radiographs, blood tests and medical records.16 The number of chronic diseases or conditions was measured as 0, 1, 2 or ⩾3 of the following: angina, myocardial infarction, congestive heart failure, degenerative disc disease, spinal stenosis, hip fracture and osteoporosis; osteoarthritis of the knee, hip and hand and rheumatoid arthritis; stroke, Parkinson’s disease, pulmonary disease, diabetes mellitus, peripheral arterial disease and cancer.

Analysis of data

We first calculated descriptive statistics for each of the key variables. To examine our first hypothesis, that SES is associated with frailty status, we used ordinal logistic regression with weights to correct for the sampling design. This approach models cumulative odds of frailty as a function of predictors. We tested each SES indicator (education and income) separately with frailty as the outcome. The unadjusted models tested the odds of being frail compared with intermediately frail and robust. The adjusted models tested the odds of being frail compared with intermediately frail and robust including age, race, insurance status (commercial plus Medicare vs Medicare/Medicaid vs Medicare only), smoking status and co-morbidity count. To test our secondary objective, that race is not related to frailty independent of SES, we used two separate ordinal logistic regression models with controls for income and for education, respectively. The proportional odds assumption was tested using a nested model approach comparing the ordinal logistic model with multinomial logistic regression. All analyses used Stata Version 9.0 (College Station, Texas, USA).

Results

The baseline characteristics of the participants are shown in table 1. Ten per cent of the women in the sample were frail, 46% were intermediately frail and 44% were robust. Forty-one percent of women had less than a high school education and 76% of the women reported their race as Caucasian. The mean income was $21 967, with 80% of participants having $5500–45 000. Sixty-nine percent of women had none or one chronic medical condition.

There were significant differences in the prevalence of frailty by race. The prevalence of frailty among African-American elderly women was higher than among Caucasians (13% vs 9%, p<0.05). Black women were disproportionately represented in the group of those who had not completed high school (composing 38% of the population not completing high school but only 24% of the study population).

The ordinal logistic regression analysis (table 2) shows that the measures of SES were significantly associated with frailty. In univariate analyses, those with <12 years of education had a relative odds for frailty of 3.51 (95% CI 1.99 to 1.54) compared to those with >12 years of education. Lower income was associated with greater odds of frailty (OR 2.69, 95% CI 1.84 to 1.93). Race was not a significant correlate of frailty when in the same model with any SES measure.

To further determine the relationship between SES and frailty we adjusted for potential variables that are associated with both SES and frailty: smoking status, insurance status and disease count. The association between SES indicators and frailty remained significant when adjusted for these potential confounders. Compared with those with more than a high school degree, the relative odds of being frail for those with less than a high school degree were 3.01 (95% CI 1.99 to 4.54). The relative odds of frailty for those with income <$10 000 was 2.01 (95% CI 1.28 to 3.16) compared with those with an income >$22 500. Because of the significant differences in education by race as well as the potential difference in quality and quantity of education received by women of different races when this cohort was young,22 we tested whether there was an interaction between race and years of school repeated; this interaction term was not significant.

We conducted a sensitivity analysis to determine whether our results would be altered by the categorisation of frailty. In these analyses we grouped participants who were robust and intermediately frail together to perform a binary logistic regression. This analysis provided similar results to those of the ordinal logistic regression models. Those with less education and lower income were more likely to be frail than their more advantaged counterparts. The relative odds of frailty was 3.01 for those with less than a high school degree compared with those with more than a high school degree (95% CI 1.99 to 4.54) in the fully adjusted model. The relative odds of frailty for those with <$10 000 income per year was 2.01 (95% CI 1.28 to 3.16) compared with those with a yearly income of >$22 500 in the full model.

Discussion

In this group of older community-dwelling women there was a significant association between SES and frailty. This association was present irrespective of the measure of SES and remained strong despite controlling for age, race, chronic disease, insurance status and smoking status. Based on the findings of Braveman et al,23 we analysed measures of SES separately. In our study, frailty was not related to race which is in contrast to a study by Hirsch et al that examined the independent effect of race on odds of frailty and found that race was a significant frailty predictor independent of SES.15 One possible reason for this different finding is that the Cardiovascular Health Study has both men and women participants while our study was restricted to women. The association between race and frailty may differ by sex or their finding could be affected by residual confounding.

In other respects this study extends the findings of others. Woods et al found that income and education were risk factors for frailty in the Women’s Health Initiative.5 The current study complements this finding using the objective measures in the Fried frailty definition which the Women’s Health Initiative was not able to use. Other large US cohorts have examined predictors of frailty. The descriptive tables contained in these studies show that participants with low education and income are disproportionately represented in the frail groups.6 24 25 26 However, these studies have not examined the contribution of SES factors to frailty independent of other covariates. In contrast to our findings, Hirsch et al found that neither education nor income was related to frailty in the Cardiovascular Health Study cohort.15

There are several biological mechanisms that could elicit the relationship between low SES and frailty. SES has been linked to inflammation,27 28 decreased physical tone,29 decreased serotonin30 and altered biological risk profiles.31 These same factors may be implicated in the origins of frailty as well. For example, researchers have posited that chronic inflammation may be a key factor in frailty,18 26 which has also been suggested to mediate the relationship between SES and morbidity due to chronically sustained psychosocial stressors.11 SES may also be linked to frailty status through decreased physical activity,32 which may lead to exhaustion and sarcopenia,33 which are key features of the frailty syndrome. SES may be linked to frailty through poor nutrition as those of low SES have decreased access to micronutrients34 and those with lower levels of micronutrients are more likely to become frail.35

As a second sensitivity analysis we examined whether neighbourhood SES (a composite of median income, wealth, education and proportion of residents with executive, managerial or professional specialty occupations)36 37 was associated with frailty status. We used generalised estimating equations to account for the fact that the SES of an individual is nested within the neighbourhood SES. Neighbourhood SES was a significant but weaker correlate of frailty status when adjusted for age and race (OR 1.26; 95% CI 1.03 to 1.54). Neighbourhood SES was no longer significantly associated with frailty (OR 1.18; 95% CI 0.97 to 1.45) once additional individual level covariates (smoking status, insurance status, disease burden) were added to the model.

Our study has the following limitations. The study is limited by its cross-sectional design. We cannot infer causality due to the cross-sectional design but reverse causation seems unlikely. Education is particularly resistant to reverse causation in older adults as it is usually attained in early life. Our study includes only African-Americans and white subjects. It is unclear whether these findings might apply to other races or ethnicities. Strengths of the current study include a population-based sample, objective and subjective measures of the frailty components and three different related measures of SES.

Summary

The current findings suggest that education and income are related to frailty. Whether the relationship is causal remains to be tested. We also found that the effect of race on frailty is confounded by socioeconomic position. The overall findings are important because the population of older adults with low education is increasing.