PROJECT SUMMARY
Alzheimer's disease and related dementias (ADRD) represent a looming public health crisis, affecting roughly
5 million people and 11% of older adults in the United States. Studies on patient transitions across healthcare
settings suggests that older adults with chronic conditions are vulnerable to inadequate transfer of information,
putting them at risk for diminished quality of care, medication errors and potentially preventable complications.
Patients at varied stages along the ADRD trajectory – especially those with multiple chronic conditions – may
experience unique risks due to the loss of information across care settings. In a recent study, we developed a
longitudinal dataset on a diverse cohort of 56,652 patients – mostly aged 65+ with multiple comorbidities –
receiving home health care (HHC) services from a large non-profit home care provider. For patients admitted
to HHC in 2010-2012, we identified subgroups of patients with ADRD diagnoses made prior to and after HHC
admission. Outside the scope of this prior study remains a vast and largely unexplored data source – nurses’
free-text clinical notes captured in the electronic health record. With roughly 1 million entries of nurses’ free-
text notes associated with the study population, there is a wealth of potential information from which to gain
new insights and a need for innovative methods to analyze this unstructured data source. In this study, we
propose to use natural language processing (NLP) techniques, a method for systematically analyzing free-text
content that draws upon machine learning. Using the dataset developed in the prior study, this study aims to:
1. Expand and improve an existing NLP system to automatically identify the following information in the
nursing free-text notes: (i) knowledge of the patient having a previously established ADRD diagnosis; (ii)
observations of cognitive symptoms and related patient/caregiver needs; and (iii) mentions of interventions
to address these needs.
2. For patients who do not have an ADRD diagnosis prior to HHC admission, determine whether nurses’ free-
text documentation of cognitive symptoms identified in the NLP predict subsequent ADRD diagnoses
during the 4-year follow-up period.
3. Among patients diagnosed with ADRD prior to HHC admission, determine whether nurses’ free-text
documentation patterns (e.g. knowledge of the patient’s ADRD diagnostic status, observations of cognitive
symptoms, and interventions) predict: (i) service use; and (ii) adverse health outcomes for which ADRD
patients are at heightened risk (e.g. hospitalizations due to urinary tract infection, dehydration, falls).
This study will allow us to examine HHC nurses’ practices, which are often difficult to observe systematically,
and identify strategies to address the complex needs of their patients with ADRD and un-diagnosed patients
who may be on a path toward ADRD diagnosis. The long-term goal of this research is to develop a home-
based intervention that aims to improve quality of life for patients and caregivers along the ADRD trajectory.
