Development and validation of prediction models for predicting social care strengths and vulnerability in older people: Cohort study using routine data in Adult Social Care

In Adult Social Care, UK local authorities have statutory responsibilities for assessing needs and delivering services to ensure adults wellbeing. Administrative data collected during this process may help local authorities compliance with these duties. We developed and internally validated predictive models for older people (>- 60 years) receiving social care for whether they remained at home or were admitted to care homes, two years after index assessments, using administrative data from one English City authority. We enquired whether the right data, to predict older peoples vulnerability to adverse outcome (care home) or evidence their strengths (remaining at home), were present in local authority systems, and if accurate datasets for large numbers of older people could be constructed to allow robust modelling. Logistic regression models were created with binary outcome (remaining at home/entering care homes). Sample size calculations determined the maximum number of candidate predictors we could consider for model development. 20,128 older people in the data cohort indicated we could consider a maximum of 46 candidate predictor parameters. In our final analyses we considered 31 candidate predictor parameters, in the areas of: age, sex, ethnicity, deprivation, housing tenure, abilities in activities of daily living, access to carer, Primary Support Reason, diagnosis of dementia. We used all data for model development and internal validation using cross-validation. Models were robust as to assumptions, with no evidence of overfitting and good predictive accuracy. Circumstances predicting strongly that older people would not remain at home were that they were: from other (non-white) ethnic groups, privately rented tenants or other (unstable) tenancies, not able to eat or keep their home habitable, with their Primary Support Reason for personal care, mental health, or support with cognition. With public involvement partners, we created a prototype index for local authority use to inform professional decisions and/or local planning.