Factors associated with frailty in primary care: a prospective cohort study

Frailty can be defined as a progressive loss of reserve and adaptive capacity associated with an overall deterioration in health that can result in disability, loss of independence, hospitalisation, extensive use of healthcare resources, admission to long-term care and death [1‐3]. Nevertheless, despite the widespread use of the term, there is no agreement on the definition of frailty [4, 5] or on an instrument to identify it [6] in a straightforward way, particularly in primary care.

Three approaches for the identification of frail individuals have been described in the literature [7]: the rules-based, the cumulative and the clinical judgment. Rules-based approaches are derived from multiple regression models and are based on the presence of a number of symptoms. The phenotypic approach would be included in this group [1]. Cumulative-based approaches are based on the consideration and addition of the number of impairments presented by a single person [8]. Finally, clinical judgment based approaches rely on the professional interpretation of the clinical record and physical examinations [7].

A frequently considered aspect in the aforementioned approaches is functional performance. Several instruments have been proposed to assess various aspects of functional performance such as motor strength, mobility and balance. Those most commonly used being the Timed Up and Go (TUG) [9, 10] and Gait Speed (GS) tests [11, 12]. Such tests have demonstrated their usefulness in terms of monitoring the health status, functional capacity and risk of falls of elderly people [11, 13‐15].

Early identification of frail individuals in primary care could have a double impact. On the one hand, it may help improve the health status of identified individuals, since it would be possible to address their needs in a comprehensive manner, delaying or preventing the natural progression towards dependence [16]. On the other hand, it may lead to reductions in resources use [2].

The objective of this study was to assess which factors are associated with frailty-related adverse outcomes in elderly individuals and on the basis of these results and from a rules-based perspective, to propose a suitable instrument for identifying these patients in primary care.

A total of n = 215 individuals were initially included in the study and of those n = 169 completed the proposed follow-up. No significant differences were found in terms of socio-demographic and clinical characteristics between lost to follow-up subjects and those who completed the study. At the end of the follow-up period, n = 119 subjects remained independent, while the other n = 50 participants (24 %) had a frailty-related adverse outcome: death (n = 8) or loss of independence (n = 42) (Fig. 1). The rates of independence loss in the first and second years were 8.3 and 2.6 %, respectively.

At baseline, participants had a mean age of 79.4 (SD: 4.1) years and 63 % were women. A high proportion of subjects presented comorbidity (72 %) and polipharmacy (61 %). Comparing the baseline status of those who eventually developed a frailty-related adverse outcome and those who did not, we observed the following. The adverse outcome group was on average 3 years older (p < 0.001), had a lower level of physical activity (p = 0.001) and was more likely to present polipharmacy (p = 0.001) or comorbidity (p = 0.032). In addition, it presented more hospital admissions in the previous year (p = 0.013) and declared a poorer self-perceived health status. No significant differences were found in the variables of sex, body mass index, visual or auditory deficits and accidental falls in the previous year (Table 1).

As far as functional status is concerned, participants who developed frailty-related adverse outcomes, despite being independent at baseline, had lower levels of functioning in terms of basic activities of daily living (Barthel’s index) and IADL (Lawton’s score). Significant differences were also observed in the functional performance tests of TUG and GS at baseline. Participants with frailty-related adverse outcomes performed worse in both (p < 0.001) (Table 2). No socio-demographic differences were observed between those finishing the study and drop-out subjects. The latter were half a year older than the rest (mean drop-out age 79.9 (SD: 4.0), p = 0.398); 39 % of them were men, compared to 36 % of those completing the study (p = 0.705). Both groups had similar baseline Barthel values (p = 0.901) and self-perceived health (p = 0.218).

Except for Barthel and Lawton scales, all other variables with p-values ≤0.10 presented in Tables 1 and 2 were implemented in the construction of the multivariate logistic regression model. It’s worth mentioning that due to their high correlation (r = −0.821), GS and TUG could not be both considered in the same model and the TUG was selected given its capacities [33]. Furthermore, the presence of comorbidity and having specific health problems (e.g., musculoskeletal disorders) were not significant in any of the models that included the polipharmacy variable, possibly due to the strong association between these factors. Both the forward and backward stepwise regressions resulted in the same multivariate model.

The multivariate logistic regression model that best fitted the data included age (OR: 1.14; 95 % CI: 1.03, 1.25), polipharmacy status (OR: 2.74; 95 % CI: 1.06, 7.06) and TUG time (in seconds) (OR: 1.28, 95 % CI: 1.14, 1.44) (Table 3). Our results suggest that these factors have good discriminant validity, with an area under the curve (AUC) of 0.822 (Fig. 2), adjusted R-squared of 0.384 and Hosmer Lemeshow p = 0.721 (Table 3).

The additional sensitivity analyses gave the following results. When all lost to follow-up (n = 46) were assumed to be finally dependent, the derived model included the variables of age, TUG and at least one hospital admission in the last year. When only those not located (n = 29) were considered dependent, the model included age and TUG. When all lost to follow-up (n = 46) or only those not located (n = 29) were assumed to be independent, the final models included the variables of age, TUG and polipharmacy. The AUCs of all four models oscillated between 0.733 and 0.795.

Proper identification of frail individuals in primary care represents an unresolved challenge. The current study, leaving aside discussions on the definition of frailty and its components, addresses this issue from a pragmatic point of view. Subjects able to perform basic daily living activities (BADL), so independent, are considered frail when they lose independence or die in a maximum period of two years. With this approach, the prevalence of frailty in our sample would be 24 %, higher than expected in our setting, but in agreement with other published studies [4, 34].

Our study shows that it is possible to identify clear differences between groups of community-dwelling elderly individuals that do and do not lose their independence within 2 years. The differences are significant in terms of age, level of physical activity, health status and functional capacity. The current results are in agreement with the literature on this topic, indicating association between frailty with age, comorbidity, self-perceived health and functional status [35]. No differences in terms of socioeconomic status, or family and social network were found in our sample.

Considering the prospective nature of the methodology used in this study and the rules-based approach proposed for the construction of operational definitions of frailty, we found that the factors associated with frailty-related adverse outcomes could be expressed in a single model including age, polipharmacy status and TUG time. We should underline that when constructing the model polipharmacy was selected over other comorbidity-related variables, as it is easier to assess than other comorbidity indices. As stated in the results section, none of these other variables was statistically significant when considered simultaneously with polipharmacy, most likely due to the existing relations between them. The sensitivity analyses performed seems to support the presented model. Even though hospital admission turned out to be significant, when all drop-outs were considered dependent, it is a fact that hospital admission and polipharmacy are not independent, rather highly related variables. Recent hospital admission may be indicative of a health problem which can end up in a negative outcome and has been often described as associated to frailty [36]. A high association between the two was also found in our in our sample (p = 0.002).

To date, a variety of approaches have been suggested for identifying frail individuals. One well-known approach uses the frailty phenotype proposed by Fried in 2001 [1], based on the presence of at least three of the following deficits: slow gait speed, weak grip strength, low level of physical activity, exhaustion and unintentional weight loss. Despite the widespread use of this phenotype in research, this tool has not been widely adopted in clinical practice, possibly due to the lack of population studies establishing diagnostic cut-offs for some of the criteria [3]. Other ways of identifying frail individuals involve frailty indices, based on cumulative approaches [37]. Indices published in recent years include the Survey of Health, Ageing and Retirement in Europe (SHARE) Frailty Instrument [38, 39], and the Program of Research to Integrate Services for the Maintenance of Autonomy (PRISMA) questionnaire [40]. These indices assess the existence of deficits in areas such as strength, balance, nutrition, resistance, mobility, physical activity and cognitive ability, but none of them have been taken up in primary care settings. Also, a third approach based on the clinical judgment it’s being explored. In this group, the Gérontopôle Frailty Screening Tool [41] and the EASY-Care TOS [42, 43] can be considered. The latter, with a specific focus on primary care settings, provides a robust and easy to perform two step tool for the identification of frail patients in primary care practices. It is interesting to note the parallelisms with this article especially regarding aim (primary care oriented) and methodology (longitudinal, adverse outcome based). Nevertheless, the main difference is the implemented frailty identification approach. Our tool is a rules-based one while EASY-Care TOs is based on clinical judgment. Additionally, our proposal overcomes one of the main limitations of the clinical judgment based tools, which is that they rely on a strong and long term patient –care professional relationship, whereas the proposed tool is based on three objective and easy to measure variables, not requiring previous knowledge of the patient’s clinical history. Hence, given the characteristics and simplicity of our model, we believe that once validated in a different and broader population, it could be considered as a screening test for identifying frail individuals in primary care.

The main limitation of our study is related to its cohort design and associated losses to follow-up. To minimise this bias, we thoroughly informed patients about the goals of the research, arranged informative meetings with them and carried out a campaign to raise awareness of the study in the participating health centres. There were more losses during the follow-up period than expected, but figures were similar to those of other studies and are considered acceptable [44]; further, we reached the number of frailty-related adverse outcomes expected, which has enabled us to achieve the desired robustness in the analysis. Another notable limitation is the fact that repeating performance tests may lead to a certain degree of training. In any case, this limitation would underestimate the onset of functional deterioration and hence would not affect the validity of the results [45]. To minimise bias from inter-observer differences, we standardised the criteria for the administration of the various different scales and tests and trained the nurses in charge of data collection.

The main strength of this study is the fact that it was conducted entirely in a primary care setting. Many studies in this field are conducted in geriatric settings, where population characteristics and patient needs are very different. A recently initiated research project (reference PI14/01905) will allow us to further validate the ability of the proposed model in identifying frail individuals in a larger sample of independent community dwelling elders aged 70 or more.

It is possible to identify frail individuals considering their age, polipharmacy status and functional capacity. These three factors can be assessed in a simple and quick way, fact which renders the proposed model suitable to use in primary care.