Development and validation of a model to identify digitally excluded older people using electronic health records
Abstract
Introduction
Digital exclusion affects many aspects of health and wellbeing, and older people are especially vulnerable. Digital support and training are available but are not offered systematically and do not always reach those most in need. The aim of this study was to develop and validate a model that predicts digital exclusion in older people, based on information recorded in primary care.
Methods
A survey on digital use was mailed to patients aged 65+ registered at three GP practices in Yorkshire. The survey collected demographic information and internet use (including 22 online activities categorised into four groups: Entertainment, Keeping in touch, Healthcare, and Buying things/ managing money). Practices linked survey responses to frailty (eFI) scores, and whether the patient had an email address recorded in primary care. Data from two practices were used for model development and one for external validation. Digital exclusion was classed as completing <1 online task in any of the 4 groups of online activities. Logistic regression was performed with digital exclusion as the outcome and the following predictors: age, sex, frailty, English as first language, and recording of email address. Model discrimination was assessed using the C-statistic and calibration using the calibration slope and Calibration-in-the-Large (CiTL).
Results
There were 2971 participants in the development cohort and 417 in the external validation (recruitment rates=49.7% and 34.8%). Digital exclusion was similar in the development and external validation cohort (41.2% and 39.3%). In the development cohort, discrimination was very good (C-statistic=0.803), with excellent calibration (C-slope=0.996, Calibration-in-the-Large=0.0002). In the external validation, discrimination remained very good (C-statistic=0.836), although there was some miscalibration (C-slope=1.253, CiTL=-0.255).
Conclusions
The model could be used in primary care to identify patients unlikely to use digital systems and communication methods. This could support tailored patient engagement, including offline access routes, and targeted digital training.