Exploring measures of multimorbidity in predicting health and social care outcomes using administrative and survey data

Abstract

Background: Multimorbidity is associated with adverse health and care outcomes, particularly in older populations. When quantifying multimorbidity, the appropriate measure varies by population, outcome under study and data available. Integrated health/social care, with a focus on the individual, improves patient satisfaction and health. In Scotland, clarity as to which measures/conditions are most strongly associated with health and care outcomes will help anticipate integrated care.

Aim: To identify which multimorbidity measures, conditions and comorbidities predict health and care outcomes in an older Scottish population.

Methods: Demographics, social care, admissions, and prescribing data for individuals 65+/resident in Scotland 2010-16 comprised three panel cohorts: for health (n=5,579,492), social (n=4,374,662) and informal care outcomes (n=2,449,229). Survey data linked to admissions were used for co-resident care (n=8,334). Panel logistic regression, using the receiver operating curve (ROC), identified the most predictive measures of multimorbidity for health/care. Further modelling was used to identify the strongest associated conditions/comorbidities, the impact of multimorbidity on social care by deprivation, and whether administrative outperforms survey data in predicting informal/co-resident care.

Results: The Charlson Comorbidity Index (CCI) performed best (ROC >0.8) in predicting mortality, proxy measures for other health outcomes (ROC >0.7 and <0.9), the Henery Chronic Disease Score 2 for social care (ROC >0.7 and <0.8) and informal care (ROC >0.8), and self-report measure (ROC >0.75) for co-resident care. Dementia is strongly associated with care, while comorbid interactions varied. An inverse effect between the relationship between multimorbidity and social care was found for local authority deprivation. Administrative data outperforms survey data at predicting informal care.

Conclusions: The varying performance of multimorbidity measures highlight the importance of a wide range of data when predicting use of health and care services. A national index tailored to a Scottish population derived from both diagnosis-based and medication-based data may have better precision. This, and findings regarding individual and comorbid conditions, such as dementia, as well as macro- and micro-level effects of deprivation on the relationship between multimorbidity and care, have the potential to improve existing risk predicting algorithms within Scotland.