A Pilot Study of the Application of Natural Language Interpretation to Identify Insights on Acute Lymphocytic Leukemia and Its Treatment from Social Media Data

Social media (SM) data is emerging as an important source of information on patient disease and treatment experiences. SM data provide researchers access to unsolicited information shared by patients/caregivers (CGs)/healthcare providers (HCPs), without burdens associated with traditional research methods or influence of interviewer bias. SM facilitates open communication on serious illness in an unstructured forum; offering snap shots of information that highlight important issues prioritized by patients/CGs/HCPs. SM data is growing exponentially, manual review of this noisy "Big Data" is time consuming and impractical. Natural language understanding (NLU) provides an optimal approach to extracting structured data elements from unstructured text. This pilot study aimed to use NLU to aggregate, analyze, and better understand various aspects of patients/CGs/HCPs experiences of Acute lymphocytic leukemia (ALL). ALL is a blood and bone marrow cancer with a significant patient symptom burden and detrimental impact on patient/CG health-related quality of life (HRQOL). Publicly available, English-language SM data were indexed from a 25-month period (Jul.2018-Aug.2020). Data were extracted using Sprinklr (Twitter), and Board Reader (boards/blogs/forums/reviews). NLU was applied to capture insights (e.g., emerging patterns/trends/relationships). Advanced analytics identified and analyzed relevant insights, and advanced logic determined user type and post contributor (e.g., patient/CGs/HCP). Posts underwent further manual evaluation using thematic content analysis to explore the experience of ALL shared by users and to support NLU-identified insights. 6,365 ALL-related posts were identified; 289 (4.5%) were contributed by patients, 657 (10.3%) by CGs, 1224 (19.2%) by HCPS and 4,195 (65.9%) by organizations. Topics included expressions of positive support/encouragement for patients/monetary donation requests/treatment type/impact of side effects. NLU identified notable discussion areas: patient/CG commentaries on treatment experiences, decisions, side effects and outcomes, and patient knowledge regarding ALL treatments. 189 patient/CG posts discussed ALL treatments in general, including in-hospital time: missing life events due to intensive treatment regimens and extended hospital stays. 195 patient/CG posts discussed specific ALL treatment experiences: chemotherapy, n=120; bone marrow transplant, n=22; stem cell transplant, n=16; immunotherapy, n=14; Kymriah, n=11; CAR-T, n=8; radiation, n=4. Chemotherapy was mentioned in relation to daily and/or durable short-and long-term effects (e.g., sickness, pain) which had major impacts for patients' and their family's HRQOL. Chemotherapy was associated with unforeseen impacts related to patients compromised immune systems (e.g., restricted social functioning due to increased risk of infection). Fear of infection and impact of treatment notably increased patients' needs to isolate which had a substantial impact on the broader family life. Bone marrow transplant posts described it as 'lifesaving'. Despite uncertainty of treatment success and negative treatment effects, patient/CG posts noted that side effects were an acceptable part of the journey to become cancer free. Posts also illustrated a general shift in patient/CG perception of ALL treatments, specifically that no one treatment works for everyone in the same way, and recent treatment developments mean that ALL is no longer perceived as a death sentence. 66 patient/CG posts commented on the cessation of treatment or lack of treatment due to remission, alternative treatments or end of life. Financial burden due to ALL treatment was an important issue for patients/CGs as it prevented the start of treatment or impaired patients'/CGs' lives. This pilot study shows how NLU can effectively extract SM data expressed by patients/CGs/HCPs relating to ALL. It is important to note that SM data are unregulated, not peer reviewed and inherently reliant on user self-identification. Thus, caution should be used interpreting SM data particularly as the information is not generalizable and/or reflective of the whole ALL patient population. Nevertheless, the pilot demonstrated how user-generated SM data can offer valuable insights on the experience and impact of ALL and its treatments that exist outside of the formal research context. Disclosures Crawford: Pfizer Inc: Consultancy. Welch: Pfizer Inc: Current Employment. Shah: Pfizer Inc: Current Employment. Doward: Pfizer Inc: Consultancy. Truchan: Pfizer Inc: Current Employment.