(150) Identification of Frail Older Adults Using Routine Primary Care Data: Accuracy of Four Electronically Derived Algorithms

Publication Number: 824

Background: Several methods have been developed that use routinely collected clinical data based on electronic medical records (EMR) to identify frail patients in clinical practice. These instruments can be easily applied in research and care settings. However, hardly any study validated frailty instruments against a diagnostic reference standard such as clinical judgement.

Objectives: To estimate and compare the ability of four potential frailty instruments that make use of EMR data and are used in research and clinical practice, to identify patients who were considered frail according to their general practitioner (GP).

Methods: Data were obtained from 36 GP practices from the PHARMO Database Network that routinely coded frailty as part of elderly care programs. These practices served a total population of 31,511 patients aged 65 years or older.
The four instruments that were considered were the Dutch Polypharmacy Score (DPI), the Charlson Comorbidity Index (CCI), the Chronic Disease Score (CDS), and the Frailty Index (FI). GPs’ clinical judgement of patients' frailty status was considered the reference standard (ICPC code A05). The ability of the instruments to distinguish between frail and non-frail patients according to the GP was assessed by calculating the area under the receiver operating characteristic curve (AUC). The sensitivity and specificity were calculated for each instrument. The analyses were done in the total population as well as in subgroups of age and sex.

Results: In the total study population of 31,511 patients (mean age 75 years (SD 8), 46% men), 3,735 (12%) patients were classified as frail by their GP. The CCI showed the highest AUC (0.79, 95%CI: 0.78-0.79), followed by the CDS (0.69, 95%CI: 0.68-0.69). Moderate sensitivity to identify frailty was found when using the DPI or the CCI, while sensitivity was poor using the CDS and FI. Specificity was highest for the CCI (0.70, 95%CI: 0.69-0.70). Calibration was considered good for all instruments.
Different results were observed when applied in subgroups of age and sex. Overall, the four instruments showed poorer performance in men and women aged >85 years compared to younger age groups (AUC 0.55 to 0.58 in men and 0.58 to 0.61 in women). According to the GP, a large proportion of the patients aged >85 years classified as non-frail by the instrument were considered frail. In all age groups, the CCI showed the most favourable results.

Conclusions: This study showed that the four frailty instruments based on electronic health care registration in primary care practice compared well with the frailty assessment done by the professionals in these practices. However, the performance declined with higher age and is least accurate in the age group of most interest, limiting the use of these instruments in people over 85.