Methods
Population
Data were taken from Wave 1 of The Irish Longitudinal Study on Ageing (TILDA), a prospective study of a sample representative of the population of the Republic of Ireland aged 50 and older living in the community (not in long-term care facilities). Participants who reported a doctor's diagnosis of dementia or who were unable to consent personally to participation because of severe cognitive impairment (judged at an interviewer's discretion) were not included in Wave 1 of the study. The full cohort and assessment has been described in detail elsewhere, but in brief, 8,175 participants aged 50 and older were interviewed in their own homes and then underwent a nurse-administered health assessment at one of two dedicated health centers. Participants unable or unwilling to attend a health assessment center received a modified nurse-delivered health assessment at home. The response rate to the interview was 62.0%, and 61.6% of participants (n = 6,150) completed the health (n = 5,274) or home (n = 876) assessment.
Data from participants who had participated in a full health center assessment that included all of the physical and cognitive items needed to compose a measure of frailty together with the battery of neuropsychological measures were used.
Consistent with the definition of frailty by Fried and colleagues, participants who had Parkinson's disease (n = 16) or a history of stroke (n = 67); those taking antidepressants (n = 284), dementia medication (n = 6), or levodopa and carbidopa (n = 1); and those with severe cognitive impairment (MMSE score <18; n = 11) were excluded.
Assessment of Frailty
The Fried phenotypic definition of frailty was used because it incorporates a definition of prefrailty and frailty. Participants were classified as frail if they met three of the following five criteria: poor grip strength, slow gait speed, low levels of physical activity, unintentional weight loss, and exhaustion. Poor grip strength was assessed using a dynamometer. Two readings from the dominant hand were taken, and mean strength calculated. The 20th percentile was calculated, and population-specific cutoffs were derived after adjustment for age, sex, and body mass index. Gait speed was measured using the GAITRite portable electronic walkway system (CIR Systems, Inc., Havertown, PA). Participants walked at their usual pace along a 4.88-m (16-foot) walkway with an extra 2.5 m at each end to allow for acceleration and deceleration. Mean gait speed was calculated from two walks. Participants who fell below the sex- and height-adjusted 20th percentile were considered to have slow gait speed, as defined according to the Fried methodology. Physical activity was measured using the short form of the International Physical Activity Questionnaire, which converts levels of physical activity in various domains into predicted kilocalories expended per week. Unintentional weight loss was assessed using the question, "In the past year, have you lost 10 pounds (4.5 kg) or more in weight when you were not trying to." Exhaustion was assessed using two items from the 20-item Center for Epidemiologic Studies Depression Scale (CES-D): "I could not get going," and "I felt that everything I did was an effort," with four possible responses to each question: never, rarely, sometimes, often. A response of sometimes or often to either question was classified as exhaustion. Participants with three or more indicators were defined as frail, those with one or more as prefrail, and those with none as robust.
Assessment of Cognitive Function
The cognitive battery included tests of global cognition, attention, memory, executive function, and speed of processing. Composite scores were calculated for each of these domains. Although each of the tests measures more than one domain of cognitive function, they were classified based on the main cognitive component involved.
Global cognition was assessed using the MMSE and Montreal Cognitive Assessment (MoCA). Attention was assessed according to two tasks: part 1 of the Color Trails Test and the Sustained Attention to Response Task (SART). The Color Trails Test involves two parts; the first requires participants to draw a line connecting circles numbered 1 to 25 in consecutive order. The second part requires connecting numbered circles while alternating between pink and yellow circles (e.g., pink 1, yellow 2, pink 3). The first part assesses visual scanning and attention. The mental flexibility involved in the switching task of part 2 assesses executive function. The SART tests the ability to sustain attention for long periods of time during a repetitive task. Participants were shown a computer screen that repeated a sequence of numbers from 1 to 9 for approximately 4 minutes. Numbers appeared every 300 ms, and participants were asked to click in response to each number except the number 3.
Memory was assessed using three methods: the Picture Recognition and Recall subtests of the Cambridge Mental Disorders of the Elderly Examination (CAMDEX), a 10-word immediate and delayed recall test using a list of 10 words previously developed for the Health and Retirement Survey, and self-rated memory measured using the question: "How would you rate your day-to-day memory at the present time? Would you say it is: excellent, very good, good, fair, or poor?"
Executive function was assessed using three tasks: the Visual Reasoning subtest of the CAMDEX (participants are asked to identify which of six objects would complete a pattern of three similar objects), part 2 of the Color Trails Test, and the MoCA verbal fluency subtest (participants were given 1 minute to name as many words beginning with F as they could think of).
Processing speed was assessed using a computer-based program. Participants were instructed to press a central button on the keyboard and wait for a stimulus (yes/no) to appear on screen. When the stimulus appeared participants released the central button and pressed one of two other buttons labeled "yes" or "no" to correspond to the stimulus on screen. The full task involves approximately 100 repetitions. Cognitive processing speed is measured as the time taken to release the first button on presentation of the stimulus.
Composite scores for each cognitive domain were calculated by obtaining z-scores for each cognitive test and combining them to create total scores for global cognition (MoCA and MMSE), executive function (visual reasoning, Color Trail Test part 2, MoCA "F letter" subtest), memory (visual recall, visual recognition, immediate memory, delayed memory), attention (Color Trail Test part 1 and SART), and speed of processing (Choice Reaction Time). The standardized versions of these scores constituted the composite cognitive score for each domain. Using Z scores allows the tests to be combined, which leads to more-reliable estimates of the effects on each domain, reduces the number of independent tests being conducted, and allows the predictive value of frailty indicators for each domain of cognitive function to be compared in a standardized way. These lead to estimates of effects for each domain in the population, which can be approximately related back to differences in individual tests using the population standard deviations for each test.
Covariates and Exclusion Criteria. Education was categorized as primary and below, secondary, or tertiary and above. History of stroke and chronic conditions, including joint problems, cataracts, glaucoma, age-related macular degeneration, lung disease, asthma, arthritis, osteoporosis, any type of cancer, Parkinson's disease, peptic ulcer, liver disease, varicose ulcer, alcohol or substance abuse, chronic pain, and incontinence, were ascertained according to self-report of a doctor's diagnosis. Height and weight were measured and used to calculate sex-adjusted cutoffs for the frailty measures. Participants were also asked to record all of the medications taken on a regular basis, and when possible, interviewers asked to see the packages to confirm.
Statistical Analyses
Sociodemographic characteristics of the three groups (frail, prefrail, robust) were compared. The relationship between frailty and each of the individual cognitive tests was then assessed using linear regression models.
Owing to large differences in the mean age of each frailty group, it was not deemed useful to compare cognitive test scores between groups without taking demographic factors into account. Therefore, for descriptive purposes, the marginal mean score of each cognitive test was calculated for each frailty group using regression models adjusted for age (as a continuous variable, including an age-squared term to account for nonlinear relationship between cognitive function and age), sex, and educational attainment.
Regression models were then used to estimate the association between frailty and prefrailty and each cognitive domain as measured according to composite z-scores, and these were plotted to compare the relative effect of frailty on each cognitive domain. This analysis was adjusted for age, age squared, sex, education, all chronic conditions reported, and number of regular medications. Beta coefficients from these analyses indicate differences in cognitive scores between frailty groups in terms of that standard deviation of that cognitive domain in the population.
Finally, the effect of each frailty indicator on each domain of cognitive function was assessed using separate models, with the composite cognitive scores as an outcome variable and all frailty indicators as binary variables adjusted for age, age squared, sex, education, chronic conditions, and medications, as above. For models including individual indicators of frailty, an outcome was considered statistically significant at P < .008 (.05/6), to avoid potential false positives caused by the large number of hypothesis tests.
Analyses were performed using Stata version 12.1 (StataCorp LP, College Station, TX).