Participant characteristics and the frailty index
Of the 178,874 participants that were considered for this study, the frailty index was missing for 17,725 individuals. The resulting 161,149 participants comprised our analytic sample (Table 1). Participants were predominantly women (60%), self-reported as white (82%), resided in urban communities (89%), and never smoked (55%). The largest age group 50–59 years (24%) and those participants aged 60 years or older comprised 22% of the overall sample. The median follow-up was 7.10 years for all participants with exception of the 80 years and older group, for which it was 7.05 years.
The mean frailty index for the entire cohort was 0.140 ± 0.085, with a 1st and 99th percentile of 0 and 0.397, respectively. As expected, frailty increased with age in a linear fashion and was significantly greater in women, regardless of age (Fig. 1A). When considered categorically, 38% were low (FI ≤ 0.1), 42% were mild (0.1 < FI ≤ 0.2), 15% were moderate (0.2 < FI ≤ 0.3) and 5% were high (FI > 0.3); both the high and moderate groups increased with age, while the low group decreased (Fig. 1B). Relative to sociodemographic- and lifestyle-related characteristics, trends for the frailty index were also as expected (Supplemental Table 2). In the fully adjusted model (adjusted r2 = 0.235 for 134,776 complete cases), frailty was significantly lower in males, individuals that are or have been married, those who have received a diploma or higher education, income greater than $50,000, rural residence, those who consume alcohol more than once per month and never smoked; dose dependant relationships were apparent where applicable. In contrast, frailty was significantly higher in older adults, residents in Northern Ontario and those self-identified as white.
A summary of the frailty index in the sample population, stratified by age and sex. A) The mean and standard error of frailty in women and men across age group, and B) the proportion of frailty categories across ages
Associations between frailty and all-cause mortality
Over the follow-up period, 6,951 deaths were observed. The cumulative incidence increased exponentially with frailty, where 5-year estimates for the low, mild, moderate and high groups were 0.5, 1.2, 3.5, and 7.9% (Fig. 2A). For the entire sample, a 0.1-unit increase in frailty was associated with a 1.87-fold increased hazard of death (95% CI = 1.83, 1.91) in univariable analysis, and a 1.47-fold increase (1.44, 1.51) when adjusted for age, sex, ethnicity, income, alcohol consumption and smoking status (5,934 events in 137,502 complete cases). However, hazard estimates were observed to differ quite substantially depending on the age group considered, and also for participant sex to a lesser extent (Fig. 3, upper left; Table 2; Supplemental Table 3). The hazard of death associated with frailty tended to decrease with age (supported by a significant age x frailty interaction (data not shown)) and was particularly high for women younger than 40; for example, the respective HR (95% CI) for 30–39 year old women and men was 2.34 (1.91, 2.86) and 1.76 (1.28, 2.41), as compared to 1.35 (1.20, 1.53) and 1.33 (1.25, 1.43) in 70–79 year old women and men. The hazard was dramatically higher in women aged 18–29 (2.86 [1.96, 4.18]) as compared to men (1.46 [0.95, 2.24]), although the error around these estimates was also higher, likely due to the relatively smaller observed number of deaths (50 vs. 42, respectively).
Cumulative incidence of health outcomes over the follow-up period, stratified by age group. A) All-cause mortality, B) outpatient admissions, and C) inpatient admissions
Forest plots depicting estimates from stratified, adjusted models for frailty in relation to all-cause mortality, inpatient and outpatient admissions and length of stay for inpatient admissions. The hazard ratio (HR) or incidence rate ratio (IRR) is relative to a 0.1-unit increase in frailty
Associations between frailty and healthcare utilization
A total of 270,005 hospital admissions were observed over the follow-up period, 177,186 of which were classified as outpatient (67%) and 92,819 as inpatient. For outpatient admissions, the mean cumulative incidence over 5 years in the low, mild, moderate and high frailty groups was 0.40, 0.67, 1.13, and 1.63 visits, respectively, whereas for inpatient admissions it was lower at 0.23, 0.31, 0.59, and 1.13 visits (Fig. 2B). In the entire sample, after adjusting for age, sex, ethnicity, marital status, education, income, rurality, region, alcohol consumption and smoking status (complete cases = 134,776, r2-outpatient = 0.377, r2-inpatient = 0.262), frailty was observed to be more strongly associated with the rate of inpatient admissions, where the incidence of inpatient admissions increased 1.60-times (95% CI = 1.59, 1.62) for every 0.1-unit increase in frailty, and outpatient visits only increased 1.35-times (1.34, 1.36). Although in both cases estimates generally decreased with age, as supported by a significant age x frailty interaction (data not shown), the difference in sex-specific patterns was notable (Fig. 3; Table 2; Supplemental Table 3). The rate of outpatient admissions associated with frailty decreased consistently with age, and generally, were slightly higher for men. The rate of inpatient admissions also decreased consistently with age after 40 years old, but were markedly different between sexes at younger ages (Fig. 2C). For men aged 18–39, the adjusted IRR was between 1.96 and 2.13, whereas for women it was only 1.17 and 1.29; this was actually lower than adults 80 years of age and older.
The mean of the number of days for a given inpatient admission across the entire sample was 3.4 ± 5.0. This increased with frailty, where the mean length of stay for the low, mild, moderate and high groups was 2.7, 3.3, 4.0, and 4.7 days, respectively. In a multivariable model, adjusting for the aforementioned covariates (complete cases = 78,020), the length of stay was observed to increase 1.12-times (95% CI = 1.10, 1.14) for every 0.1-unit increase in frailty. This association was mostly stable with age, and did not obviously differ between sexes (Fig. 3; Table 2). Notably larger estimates were observed for women aged 18–29 and even 30–39, although the error of these associations complicate their interpretation.
link