Clustering and prediction of cancer symptom trajectories using longitudinal nursing documentation data

Currently, electronic health records (EHR) collect a growing amount of clinical data. This “big data” in healthcare, may help understand patients’ profiles not achievable through traditional research approaches. The purpose of this dissertation research was to identify groups of cancer patients with different symptom experiences and to predict future symptoms based on the time-series nursing assessment data, ultimately aimed to provide more individualized care. A retrospective and longitudinal analysis was performed using records of advanced cancer patients who received chemotherapy (n=208) hospitalized between 2008 and 2014. From the clustering analysis, healthcare professionals should be aware that cancer patients with stage 4 lung cancer might be at high risk of nausea, malnutrition, poor oral health, and abnormal mental health across chemotherapy treatments. Caregivers can provide early interventions and prompt follow-up with patients to address these symptoms and attempt to lessen symptom severity or progression. In addition, a recurrent neural network using long short-term memory units was trained based on previous symptoms experienced by patients to predict future symptom trajectories. The mean absolute errors (MAE) predicted values were lower than the MAE between the last observation and the next observation for six symptoms; pain, mobility, activity, nausea, nutrition, and appetite. We can predict patients’ symptom trajectories with a prediction model, using routinely collected nursing documentation, and this feature can be built into future EHR. The results of this project can be applied to better individualize symptom management to support cancer patients’ quality-of-life.