An Initial Feasibility Study to Identify Loneliness Among Mental Health Patients from Clinical Notes

Abstract

There is increasing evidence that better health outcomes for patients can be achieved with improvement in mental health. Loneliness, one such condition, is itself toxic consequently driving increase in both chronic disease morbidity and mortality. Thus, the early identification of loneliness and interventions to address it are of urgent importance. One potential mechanism for identifying loneliness rests in analyzing care provider notes which include details regarding a provider's interaction with patients and often provide insights about both mental and physical health. To automatically determine which patients are suffering from loneliness, a data science analysis based on natural language processing techniques was performed on clinical notes from 12 care providers for 128 patients. The analysis techniques included co-occurrence of uni-gram, bi-gram, tri-gram and quad-gram words; sentiment analysis using AFINN sentiment lexicon scores; and word usage frequencies. The results surfaced key challenges associated with determining the presence of loneliness suggested the importance of including validated clinical questionnaires specifically designed for identifying loneliness.