Predicting adverse health outcomes in older community-dwelling adults: a prospective cohort study

Background This thesis aimed to investigate if adverse health outcomes in older community-dwelling people can be predicted, through the application of measures of prescribing, multimorbidity and emergency admission risk models. There were five objectives: 1) to determine if there is a longitudinal association between potentially inappropriate prescribing (PIP) and future adverse drug events (ADEs), reduced health related quality of life (HRQOL) and increased use of Accident & Emergency (A&E) and emergency admissions; 2) to assess the performance of different measures of multimorbidity and vulnerability in predicting emergency hospital attendance and functional decline; 3) to conduct a systematic review of emergency admission risk prediction models developed for use in community-dwelling adults; 4) to systematically review and meta-analyse the validation studies of the Probability of repeated admissions (Pra) risk model; and, 5) to externally validate the Pra risk model in predicting emergency hospital admission over the following year. Methods A prospective cohort study with two year follow-up was conducted linked to the national Health Services Executive Primary Care Reimbursement Services (HSE-PCRS) pharmacy claims database (2010-2012). At baseline a total of 904 older (≥70 years) community-dwelling people were recruited from 15 general practices. The Screening Tool of Older Persons Prescriptions (STOPP) and Beers 2012 prescribing indicator sets were applied to the pharmacy data to elicit PIP. ADEs were recorded through patient interview with corresponding review of the GP medical record. HRQOL was determined through the Euro-Qul-5Dimensions (EQ-5D) administered through a patient questionnaire. Emergency attendance was ascertained through a detailed review of the GP medical record. Multilevel regression modelling was used to investigate if PIP was longitudinally associated with ADEs, HRQOL and emergency hospital attendance (Poisson (incidence rate ratio (IRR) (95% CI) and linear regression models (regression co-efficient (95% CI)). Different medication and diagnosis based measures of multimorbidity, the Vulnerable Elders Survey (VES-13) and the Pra model were investigated by examining their discrimination (the ability of the model to distinguish correctly the patients with different outcomes, c-statistic (95% CI)) and calibration (reflects how closely predicted outcomes agree with the actual outcomes, Hosmer-lemeshow statistic). Results Of 791 participants eligible for follow-up, 673 (85%) returned a questionnaire and 605 (77%) also completed an ADE interview. Baseline STOPP PIP prevalence was 42% and 445 (74%) patients reported ≥1 ADE at follow-up. In multivariable analysis, ≥2 STOPP PIP was associated with ADEs (adjusted IRR: 1.29 (95% CI 1.03, 1.85, p=0.03); poorer HRQoL (adjusted regression co-efficient: -0.11 (-0.16, -0.06; p Conclusions Older community-dwelling people, prescribed ≥2 PIP, as defined by the STOPP prescribing criteria, are more likely to report ADEs, poorer HRQOL and attend A&E over two year follow-up. Both medication and diagnosis-based measures of multimorbidity demonstrated similar performance in predicting emergency admission. The VES-13 may be useful in identifying older people at risk of functional decline in the community. In certain circumstances, while acknowledging the limitations of risk stratification, the Pra tool may have a role in targeting older people at higher risk of emergency admission.