Validated frailty measures using electronic primary care records: a review of diagnostic test accuracy

Abstract

Introduction: Identification of people who have or are at risk of frailty enables targeted interventions, and the use of tools that screen for frailty using electronic records (which we term as validated electronic frailty measures (VEFMs)) within primary care is incentivised by NHS England. We carried out a systematic review to establish the sensitivity and specificity of available primary care VEFMs when compared to a reference standard in-person assessment.

Methods: Medline, Pubmed, CENTRAL, CINHAL and Embase searches identified studies comparing a primary care VEFM with in-person assessment. Studies were quality assessed using Quality Assessment of Diagnostic Accuracy Studies revised tool. Sensitivity and specificity values were extracted or were calculated and pooled using StatsDirect.

Results: There were 2,245 titles screened, with 10 studies included. These described three different index tests: electronic frailty index (eFI), claims-based frailty index (cFI) and polypharmacy. Frailty Phenotype was the reference standard in each study. One study of 60 patients examined the eFI, reporting a sensitivity of 0.84 (95% CI = 0.55, 0.98) and a specificity of 0.78 (0.64, 0.89). Two studies of 7,679 patients examined cFI, with a pooled sensitivity of 0.48 (95% CI = 0.23, 0.74) and a specificity of 0.80 (0.53, 0.98). Seven studies of 34,328 patients examined a polypharmacy as a screening tool (defined as more than or equal to five medications) with a pooled sensitivity of 0.61 (95% CI = 0.50, 0.72) and a specificity of 0.66 (0.58, 0.73).

Conclusions: eFI is the best-performing VEFM; however, based on our analysis of an average UK GP practice, it would return a high number of false-positive results. In conclusion, existing electronic frailty tools may not be appropriate for primary care-based population screening.

Keywords: frailty; older people; primary care; risk stratification; screening; systematic review.

Figure 1

PRISMA [6] flow diagram; 2,245 studies were identified from databases and 1,064 were identified from other methods after the removal of duplicates. In total, 195 studies (149 from databases and 46 from other sources) were full-text screened for eligibility with reasons for exclusion detailed. 10 studies met the inclusion criteria for this review.

Figure 2

Random effects proportion meta-analysis of the reported sensitivities and specificities of studies using the cFI or polypharmacy as the index test and FP as the reference standard from StatsDirect. (A) Shows the specificity values for each cFI study with the 95% confidence intervals as determined by Review Manager 5.4.1 alongside the weights given to each study as demonstrated by the grey boxes. The pooled cFI specificity is estimated at 0.8 (95% CI = 0.53, 0.98), shown by the diamond. (B) Shows the same information as described in (A) for the sensitivity values given for the cFI studies with the pooled figure given as 0.49 (95% CI = 0.23, 0.74). (C) Shows the specificity values for each polypharmacy study with the 95% confidence intervals as determined by Review Manager 5.4.1 alongside the weights given to each study as demonstrated by the grey boxes. The pooled specificity for polypharmacy is estimated at 0.66 (95% CI = 0.59,0.73), shown by the diamond. (D) Shows the same information as described in (C) for the sensitivity values given for the same polypharmacy studies with the pooled figure given as 0.61 (95% CI = 0.50, 0.72).

Figure 3

Correlation between the number of patients (aged ≥65) and the prevalence of frailty (%) on the performance of (A) eFI, (B) cFI and (C) polypharmacy.