Frailty assessment in older adults using upper-extremity function: index development

Community-dwelling and outpatient older adults (≥60 years) were recruited from Banner University Medical Center in Tucson and from the Banner Sun Health Research Institute in Sun City, Arizona. Exclusion criteria included: known disorders associated with severe motor deficits (including stroke, diagnosed Parkinson’s disease, and amputation); major mobility disorders (e.g., who were unable to walk a distance of ~10 m); and upper-extremity disorders (including severe bilateral shoulder or elbow rheumatoid or osteoarthritis). Assistive devices used routinely by participants in their daily activities, including canes and walkers, but excluding wheelchairs, were allowed. The study was approved by the University of Arizona and Banner Sun Health Research Institute’s Institutional Review Boards. Before participation, written informed consent according to the principles expressed in the Declaration of Helsinki [16] was obtained from all subjects (or an authorized person in case of lack of clinically assessed capacity for informed consent).

The Fried index was used as the gold standard, based on unintentional weight loss, self-reported exhaustion and low physical activity, weakness (grip strength), and slow walking speed [6]. Grip strength was assessed three times for right and left hands, and the average for both sides was reported. Walking speed was measured as the required time to walk 4.6 m (15 ft). Individuals were classified as frail with three or more positive Fried criteria, as pre-frail with one or two criteria, and as non-frail with none. In addition, mini-mental state examination (MMSE) [17], Montreal cognitive assessment (MoCA) [18], or both were collected to assess cognitive condition.

Per our previously validated UEF method [14], wearable motion sensors (tri-axial wearable gyroscope sensor, BioSensics LLC, Cambridge, MA) were used to measure forearm and upper-arm motion, and ultimately the elbow angular velocity. Each participant performed a 20-s trial of elbow flexion, within which they repetitively and fully flexed and extended their dominant elbow as quickly as possible in the seated position, while wearing the UEF system. Before the actual test, participants performed a short practice trial to become familiar with the protocol. The protocol was explained to participants and they were encouraged only once, before elbow flexion, to do the task as fast as possible (participants were not further encouraged during the task). All assessments were performed by trained researchers, and to assure consistency, they provided exactly the same verbal instruction before measurement: “As quickly as you can, bend and straighten your dominant arm over and over for 20 s. Make sure you straighten and bend your arm all the way”.

Several outcome measures representing kinematics and kinetics of elbow flexion were derived to quantify “slowness”, “weakness”, “exhaustion”, and “flexibility” as frailty markers [6, 19]. Outcome measures included: 1) speed; 2) flexibility; 3) power; 4) rise time; 5) moment; 6) speed variability; 7) speed reduction; and 8) flexion number (see Table 1 for definitions) [14]. Readers are referred to previous work [13, 14] for more details regarding validation of UEF and parameter descriptions.

Sociodemographic parameters were compared between three Fried frailty groups using separate analyses of variance (ANOVAs) or chi-square (χ²). UEF parameters were compared between males and females and correlations between UEF parameters and body mass index (BMI) were determined.

Two UEF indexes were developed. The first index (UEF categorical index) was developed to provide frailty status as non-frail, pre-frail, and frail, similar to the Fried index. Multivariable ordinal logistic models with the Fried frailty categories as the dependent variable, and UEF parameters plus demographic information (i.e., age, sex, and BMI) as independent variables were used to develop the UEF categorical index. To construct models, the following steps were followed: 1) descriptive analysis of UEF parameters: outliers detection within each frailty group using box plots and histograms and testing of distribution normality using Shapiro-Wilk W test; 2) univariate analysis for UEF parameters as independent variables: UEF parameters with significant association with the Fried frailty status were selected for subsequent steps; 3) testing of collinearity between UEF parameters: using variance inflation factor (VIF) values. A VIF cutoff value larger than 10 was considered an indication of presence of collinearity [20]; 4) stepwise parameter selection: UEF and demographic parameters were selected based on Akaike information criterion (AIC) values. Since BMI has been shown to have a U-shape association with frailty [21], both continuous and categorical variables (<20, 20–24.9, 25–29.9, 30–34.9, ≥ 35 kg/m²) were tested in separate analyses; and 5) model evaluation: using 10-fold cross-validation, in which the sample was randomly divided into k = 10 equal parts. At each k^th iteration, k − 1 partitions were used as the training dataset and the left out partition was used as the validation dataset. The average values of area under receiver operating characteristic (ROC) curves, AIC, and accuracy – (true positive + true negative) / sample size – were calculated.

Parameters selected from the categorical index were used to develop the UEF continuous score, using methods developed for the Framingham cardiovascular risk score [22]. For developing the UEF score, first, each continuous UEF parameter and BMI were split into three categories based on mean values for each frailty group. That is parameter mean values were considered as reference values (W) and the mid-points between reference values were considered as cut-offs. Next, the distance between each category and the lowest (reference) category in regression units was determined, by multiplying the β (parameter estimates from the categorical index model) by the difference between the category W and the reference category W_REF, or β (W-W_ref). Then, each categorical independent variable was assigned a point value by dividing the distance β (W-W_ref) by a base constant that represents a coefficient value corresponding to one point. Point values were rounded to the nearest integer. For simplicity, age (increased risk of frailty associated with a 4-year increase in age corresponding to age differences between frailty groups) was selected as the base constant, as it provided the smallest distance β(W-W_ref). Readers are referred to Additional file 2: Text S2 for details regarding UEF scorings. The UEF score (0 extreme resilience and 1 extreme frailty) for a given participant was defined as the sum of points corresponding to performance results and demographic information.

The UEF score distribution was plotted. Based on previous reports on frailty index/score distribution [23, 24], the UEF score was expected to show a skewed gamma distribution. Lastly, associations between the UEF score with age, the Fried index, and MMSE and MoCA scores were investigated using both Pearson and Spearman Rank correlations.

Three hundred fifty-two participants were recruited including 132 (37%) non-frail, 175 (50%) pre-frail, and 45 (13%) frail older adults, defined by the Fried index. Age, height, and MMSE score were significantly different among frailty groups (p < 0.01). On average, frail participants were ~3 years older and 1% shorter compared to pre-frail individuals; pre-frail participants were ~4 years older and 2% shorter compared to non-frails. Non-frail participants had 2% higher MMSE score compared to other groups. Sociodemographic parameters are presented in Table 2. Significant differences were observed in power, flexibility, and moment UEF parameters between male and female participants (p < 0.03). Male participants had 32% and 99% larger elbow power and moment, respectively, and were 3% less flexible in elbow motion compared to females. Also, participants with higher BMI showed less power (r = −0.13, p < 0.01) and higher elbow moment (r = 0.17, p < 0.01).

Using UEF parameters, (except power) in the forward stepwise approach, resulted in speed, flexibility, moment, speed variability, speed reduction, and flexion number as included parameters (Table 3). Of note, when age, sex, and BMI were individually entered into the logistic model in addition to UEF parameters only continuous BMI was significantly associated with the Fried index. Therefore, we removed age and sex as predictor parameters since frailty and pre-frailty predictions (ROC area under curve and accuracy) remained unchanged after removing these parameters. Using the ordinal logistic regression model, the final probability equation was derived from parameter estimates (Additional file 1: Text S1). Using the model with the whole sample, ROC area under curves were 0.86 and 0.91 for predicting pre-frail and frailty, respectively; respective values of AIC and accuracy were 461.72 and 0.70. Results from 10-fold cross-validation showed mean (standard deviation - SD) values of 0.77 (0.07), 0.80 (0.12), 56.73 (5.17), and 0.69 (0.08) for ROC area under curve for pre-frailty, ROC area under curve for frailty, AIC, and accuracy, respectively.

Highest UEF sub-scores were obtained for speed and flexibility UEF parameters (Additional file 2: Text S2). Similar to the UEF index, age and sex were removed as they demonstrated non-significant association with frailty. Scores with equivalent categorical cutoffs are presented in Additional file 2: Text S2. As expected, the UEF score histogram was described by a gamma distribution (p < 0.01). Results from Pearson tests showed significant positive correlations between the UEF score with MMSE (r = 0.22, p < 0.001) and MoCA (r = 0.37, p < 0.001), and significant negative correlation between the UEF score and age (r = −0.37, p < 0.001). Spearman Rank correlations between the same pairs of variables showed no appreciable difference from the Pearson correlations.

As hypothesized, UEF parameters were significantly associated with frailty and UEF accurately predicts Fried frailty categories. The theorized biological mechanism is that sarcopenia/dynapenia, as a manifestation of physical frailty, would influence the entire muscular structure of the human body, including both lower and upper-extremities [6, 25]. Therefore, speed of upper-extremity movement was measured as a surrogate of gait speed. Associations between frailty and weakness, exhaustion, and flexibility have also been investigated in several studies. Grip strength weakness measure and exhaustion doing physical activity demonstrated strongest association with adverse health outcomes [6, 26]. Of note, rather than exhaustion within daily physical activity, proposed UEF speed variation and reduction parameters during maximum pace elbow flexion assess muscle fatigue, which is defined as inability to maintain required/maximum force generating capacity of the muscle [27]. Assessing muscle fatigue for screening physical frailty has been recommended previously, because frailty and muscle fatigue share several contributory factors aside from aging, including inflammation, physical inactivity, malnutrition, hormonal deficiencies, as well as muscular factors such as muscle strength and size [28, 29]. Finally in regard to the UEF flexibility parameter, although limited research exists, Brown et al. reported significantly less range of motion in the shoulder joint among older adults with greater frailty [19].

As mentioned above, all UEF parameters were selected based on frailty features. Although direct association of a parameter with frailty is necessary, it is not sufficient for including it into a frailty index. There are five criteria that have been recently proposed by Searle et al., as requirements for frailty index parameters: 1) parameters must associate with health status; 2) parameters should increase in prevalence with age; 3) parameters should not saturate at early age; 4) selected parameters as a group must cover a wide range of frailty features; and 5) frailty index must contain the same parameters for different measurements on the same sample [24]. UEF parameters and index cover all of these five criteria. Each parameter is associated with age and health status, and would not saturate at early ages as it incorporate maximum pace muscle performance (criteria 1, 2 and 3). Further, the UEF index covers several features of frailty, and it incorporates a single algorithm based on sensor-based parameters (criteria 4 and 5).

Among demographic data, only BMI was a significant frailty predictor; age and sex were not independently associated with frailty (Table 3). Current findings suggest that older adults with higher BMI (> 27 kg/m²) were at higher risk of frailty. Reports from previous work revealed that there is a U-shape association between BMI and frailty; the level of frailty was higher at very low BMI (20 kg/m²) indicating wasting, and very high (30 kg/m²) BMI indicating obesogenic sarcopenia [21]. We did not observe an association between very low BMI and greater frailty, probably due to our small sample size and the fact that components of the implemented gold standard (i.e., the Fried index) has already been adjusted for BMI.

As presented above, UEF flexibility and moment parameters were different between sexes. It was expected that male participants would show higher strength in performing elbow flexion test as reflected within the moment UEF parameter. On the other hand, previous research suggested that older females have larger maximum voluntary range of motion of elbow and shoulder joints when compared to males [30]. We believe sex was not influential in the UEF model since sex-differences in UEF parameters cancel out the overall influence of sex in the model.

First, the current study lacked test-retest assessments. The assessment procedure is objective and requires minimum judgement from the test provider, and, therefore, we believe UEF would provide high test-retest reliability, as it was also evident from cross-validation results. Second, the UEF index and score application needs to be confirmed in prospective longitudinal studies for predicting health outcomes. Especially, the UEF index/score validity and reliability for predicting mortality, hospitalization, exacerbation, as well as in-hospital outcomes such as length of stay, discharge disposition, and readmission should be investigated in future research. Within the current study, UEF index and score were developed cross-sectionally; we validated the UEF accuracy for predicting frailty using the Fried index as the “gold standard”. However, the Fried index includes only physical measures and lacks other components of frailty such as cognitive impairment, depression, and comorbidity. Third, the focus of the UEF test was solely on physical frailty and lacks a measure of cognition. Although we showed associations between UEF score and cognitive tests (i.e., MMSE and MoCA), no direct conclusion can be made regarding cognitive frailty. As a future direction, UEF using dual-task performance may be a sensitive marker of cognitive status, based on simultaneous assessment of physical frailty and cognitive performance [31].