UCSF

Growing emphasis on the early diagnosis of cognitive impairment and Alzheimer’s disease and related dementias (ADRD) increases the demand for providers to make informed care decisions. Most of the care for people recently diagnosed with cognitive impairment or ADRD falls on primary care physicians (PCPs), as frontline providers for older adults. Many of the clinical decisions PCPs face when caring for patients with cognitive impairment are not currently supported by robust evidence. Given the heterogeneity of cognitive impairments and dementias and the expense of randomized controlled trials (RCTs), we cannot rely exclusively on RCTs to provide evidence for clinical decisions across the range of conditions and unique patient or family considerations. Unclear guidelines on treatment strategies lead to large inconsistencies in how patients are treated, potentially causing harm and exacerbating health inequities. There is a large and growing need for evidence on best practices in early dementia care using accessible, timely data, and rigorous methods.

Centering causal inference, this dissertation rigorously evaluates three important clinical decisions commonly encountered by PCPs caring for patients experiencing cognitive decline. Additionally, this dissertation leverages electronic health records from the University of California, San Francisco (UCSF) or health systems within the larger University of California (UC) system. Chapter 1 uses both traditional observational and econometric methods to explore whether referral to specialty memory care at first diagnosis of impairment improves patient outcomes. Chapter 2 applies an emulated target trial design to investigate whether prescriptions for antidementia medications (memantine or acetylcholinesterase inhibitors) are associated with patient outcomes. Finally, many conditions, including urinary tract infections (UTIs), though not specific to patients with cognitive impairment, are more common among this population. Among people with cognitive impairment, inappropriate UTI management, specifically discordant antibiotics, may be especially harmful. Chapter 3 evaluates the performance of LASSO models predicting discordant antibiotic treatment among older adults with suspected UTIs. This dissertation demonstrates how to leverage EHRs to evaluate the efficacy of current and future innovations in dementia clinical care across diverse patients. Additionally, this work includes in-depth discussion of the limitations of causal inference methods and EHRs in this applied setting.