Terrence E Murphy1, Gail J McAvay2, George O Agogo2, Heather G Allore3. 1. Department of Internal Medicine, Section of Geriatrics, Yale School of Medicine, New Haven, CT; Department of Biostatistics, Yale School of Public Health, New Haven, CT. 2. Department of Internal Medicine, Section of Geriatrics, Yale School of Medicine, New Haven, CT. 3. Department of Internal Medicine, Section of Geriatrics, Yale School of Medicine, New Haven, CT; Department of Biostatistics, Yale School of Public Health, New Haven, CT. Electronic address: heather.allore@yale.edu.
Abstract
PURPOSE: We define personalized concurrent risk (PCR) as the subject-specific probability of an index outcome within a defined interval of time, while currently at risk for a separate outcome, where the outcomes are not mutually exclusive and can be jointly modeled with a shared random intercept. We further define typical concurrent risk as the risk obtained by setting the random intercept to null. METHODS: Drawing data from the Medical Expenditure Panel Survey (cohorts 2008-2013), we jointly model limitations in social activity and mobility over two years among older community-dwelling persons with both hypertension and chronic obstructive pulmonary disease. The joint model uses inverse probability of treatment weighting based on each participant's baseline propensity of polypharmacy (≥5 classes of medication). RESULTS: Even among participants with the same covariates, older persons with multiple chronic conditions exhibit wide-ranging heterogeneity of the treatment effect from polypharmacy, a risk factor for negative health outcomes among older persons. The magnitude of the PCRs is dominated by the value of the subject-specific random effect. CONCLUSIONS: Estimates of PCR and typical concurrent risk can be calculated from national or institutional data sets and may facilitate the practice of personalized care for older patients with multiple chronic conditions.
PURPOSE: We define personalized concurrent risk (PCR) as the subject-specific probability of an index outcome within a defined interval of time, while currently at risk for a separate outcome, where the outcomes are not mutually exclusive and can be jointly modeled with a shared random intercept. We further define typical concurrent risk as the risk obtained by setting the random intercept to null. METHODS: Drawing data from the Medical Expenditure Panel Survey (cohorts 2008-2013), we jointly model limitations in social activity and mobility over two years among older community-dwelling persons with both hypertension and chronic obstructive pulmonary disease. The joint model uses inverse probability of treatment weighting based on each participant's baseline propensity of polypharmacy (≥5 classes of medication). RESULTS: Even among participants with the same covariates, older persons with multiple chronic conditions exhibit wide-ranging heterogeneity of the treatment effect from polypharmacy, a risk factor for negative health outcomes among older persons. The magnitude of the PCRs is dominated by the value of the subject-specific random effect. CONCLUSIONS: Estimates of PCR and typical concurrent risk can be calculated from national or institutional data sets and may facilitate the practice of personalized care for older patients with multiple chronic conditions.
Authors: M H Gail; J P Costantino; J Bryant; R Croyle; L Freedman; K Helzlsouer; V Vogel Journal: J Natl Cancer Inst Date: 1999-11-03 Impact factor: 13.506
Authors: Martin Fortin; Lise Lapointe; Catherine Hudon; Alain Vanasse; Antoine L Ntetu; Danielle Maltais Journal: Health Qual Life Outcomes Date: 2004-09-20 Impact factor: 3.186