A Mixed Methods Approach Incorporating Text Analytics to Examine Persistence of Depression

BackgroundSome patients depression persists despite evidence-based interventions; understanding factors associated with depression persistence could inform screening and treatment. We used a novel mixed-methods approach to examine demographic, clinical, and social factors affecting depression persistence among older, low-income women; we also assessed the utility of this approach for evaluating intervention fidelity. MethodsData used for this study were generated from a comparative effectiveness study comparing the impact of prevention care management (PCM) versus a collaborative care intervention (CCI) on depression among women overdue for cancer screening: We reviewed 700 care manager logs to identify themes among patients experiences and analyzed language use using NVivo(R)s natural language processing (NLP) functionality. 757 women age 50-64 who screened positive for depression at baseline using the Patient Health Questionnaire (PHQ)-9 and were overdue for [≥]1 cancer screening test (breast, cervical, and/or colorectal) participated. All received primary care in XXX Federally Qualified Health Centers. We used NLP to quantify differences in language use across intervention groups and explored how often themes appeared in logs of participants whose depression did not meaningfully improve based on PHQ-9 scores. Differences in demographic, clinical, and social factors were examined. ResultsParticipants with persistent depression were more likely to discuss pain, fear, and transportation. Asthma and anxiety were associated with lower likelihood of depression remission, while no differences were observed in depression remission rates among those with diabetes or hypertension. Patient-centered words, including "needs" and "feelings", were more common in the CCI group, while procedure-related words, like "screening" and "mammography", appeared more frequently in the PCM group. ConclusionsPatient-related factors and social barriers contributed to depression persistence. NLP identified patterns of language use in case logs, suggesting unmet needs among depressed patients. NLP is an efficient, effective method for identifying themes in unstructured text and monitoring intervention fidelity.