Background
Approximately 1 in 4 older adults presenting to primary care are living with frailty and face the threat of declining health, poor quality of life, loss of independence, and greater reliance on higher levels of care [
1]. Frailty is a state of increased vulnerability to stressors involving loss of reserves in interrelated biological, psychological and social domains [
2‐
4]. Due to the detrimental impact of frailty and the potential to mitigate its adverse health outcomes with targeted interventions [
5], international consensus guidelines recommend case-finding of frailty in primary care as part of routine clinical practice [
6‐
8]. However, a key difficulty in widespread implementation of this recommendation has been the reliance on opportunistic case-finding in clinical practice using bedside instruments and questionnaires. Many of these tools require additional time, training, use of specialized equipment, or clinical resources, thus hindering efficiency and consistency in a busy primary care setting.
A recent breakthrough in the United Kingdom (UK) has been the development, validation and national implementation of an electronic frailty index (eFI) for frailty case-finding in primary care using data from electronic medical records [
9]. The eFI is based on the deficit accumulation approach to frailty. This approach, developed first by Rockwood, Mitnitski and colleagues [
10], identifies frailty based on a range of variables (e.g., signs, symptoms, diseases, disabilities, impairments, abnormal test values) collectively referred to as health deficits [
11]. According to this model, frailty can be measured by calculating a frailty index (FI) that can be generated from any appropriately populated healthcare database [
12‐
14] provided that there are a sufficient number of health deficits that satisfy certain criteria [
13,
15]. Primary care electronic medical records (EMRs) contain rich data on a patient’s health and psychosocial context that make it a promising dataset to generate a FI score. The eFI in UK has been developed and validated using routinely available primary care EMR data from around 900,000 patients [
9]. A pilot study in primary care in England demonstrated that the eFI was simple and quick to use (i.e. it is fully automated and scores are available at point of care), acceptable to practice staff, and was able to discriminate older patients referred for comprehensive geriatric assessment (CGA) from the total practice population [
16].
While there is a number of frailty indices developed so far, the eFI represents an interesting approach of using routinely available data in the primary care EMR for frailty identification. However, there is limited knowledge of the applicability and validity of such FI in the Canadian context. In this project, the convergent validity—a core component of construct validity of a test [
17]—of the eFI (calculated manually) from Canadian primary care EMR data with a validated frailty index tool, a frailty index based on a comprehensive geriatric assessment (FI-CGA) was examined.
Discussion
This is one of the first studies to investigate the relationship between the eFI and FI-CGA as different instruments designed to assess the construct of frailty as a state, and both applied to the Canadian primary care setting. The analysis demonstrated a linear relationship and strong correlation between the eFI and FI-CGA scores, with both lower and upper limits of the 95% confidence interval supporting this strong correlation. Thus, the study findings support the convergent validity of the eFI in relation to the FI-CGA, a core component of its construct validity.
The distribution of both scores approximated a normal distribution, which is expected in a population of oldest old (> 80 years old) who warranted CGA in a primary care setting. The distribution of the FI becomes less skewed as the mean age of the sample increases, and its relative heterogeneity diminishes [
27]. In our study both indices were free of ceiling and floor effects, which is consistent with the results reported in other studies [
28,
29] and might indicate that both indices were constructed with consideration of the selection criteria for health deficits outlined by Searle et al. [
13] Proper deficit selection is crucial in establishing the consistent ability of the FI to determine frailty levels [
28].
While strong correlation between two indices can be explained by the fact that they are based on the same theoretical framework of the cumulative deficit model of frailty, the results of the two indices were not in complete agreement. For example, the mean score for the eFI was significantly lower than that of the FI-CGA. This could be due to low prevalence or reporting of deficits derived from routine primary care data as reported elsewhere [
30]. Reasons for this may include suboptimal data entry, tendency of patients not to discuss all of their health and social concerns during a clinic visit, and the greater emphasis on comorbidities rather than function, mobility and health attitudes in this dataset. As such, if frailty-related data is systematically missing from the record, it may be assumed incorrectly to be absent [
30]. In our SCH cohort, the deficits were clearly defined and recorded as present if charted in the patient EMR, while eFI deficits not found in EMR were treated as absent. In contrast, the FI derived from the CGA intentionally explores and records challenging cognitive, psychosocial, and functional aspects of frailty and geriatric syndromes thus enriching the FI-CGA.
An important limitation was that the study sample was small; nevertheless, the 95% confidence interval was narrow, which indicates a statistically significant correlation even in such a small sample size. In addition, the sample consisted of community-dwelling older adults, not living in long-term care facilities, that were identified by family physicians as having ongoing concerns (in many cases, multiple concerns), and thus received an assessment by the SCH. This may limit the generalizability of the study findings to the very fit/robust or more functionally dependent or severely frail older adult populations. Other limitation is that the cross sectional design does not allow to compare the predictive performance of the eFI and FI-CGA. Future research may consider the predictive ability of the eFI generated from Canadian Primary Care EMR before being widely implemented. Another important limitation of this study was that the eFI calculation relied on labour-intensive review of medical records, which defeats the fundamental purpose of the eFI for rapid frailty case-finding in primary care. Since much of the data is in narrative/open text form in Canadian primary care EMRs, innovative technologies in computer science such as Natural Language Processing and Machine Learning could facilitate the future automation of the eFI in primary care.
Results indicated that measured by FI-CGA, women had higher levels of frailty than men. This finding is not new and confirmed by numerous studies [
1,
26,
31,
32]. Moreover, Herr et al. [
33], in their study of life expectancy in the state of frailty, also point to the fact that women live longer despite bearing a larger burden of health deficits than men. The authors explain this sex difference as an interplay of various social, behavioral, and biological factors. There was no difference in the eFI scores based on sex, which could be explained by the deficits included in the eFI; the comorbidities are more prevalent than functional measures.
Regardless of these differences, both indices correlated with age, number of chronic conditions and number of medications. In both the eFI and the FI-CGA, the strength of association was weak to moderate. However, the strength of these correlations was higher in the eFI, which may reflect its greater dependence on these deficits compared to the FI-CGA which includes additional assessment information, reflecting the complex and rich nature of frailty. Similar to other studies based on the FI [
30,
34], higher scores of both eFI and FI-CGA indices were observed in patients with 3 and more chronic conditions, polypharmacy, history of falls in the past 12 months and urinary incontinence. No significant difference in scores was found in those living alone. However, this finding should be interpreted with caution, as living alone does not mean having no social support, nor does it exclude the possibility of ‘assets’ that make such an individual more resilient.