Background
With the recent uptake of novel therapeutic agents, such as immunotherapy (IO) to treat aLC, there is a need for real world data (RWD) to understand the shift in treatment patterns and inform strategies to optimize available therapies. While traditional approaches of manual chart review are labour intensive and error prone, innovative AI techniques have been shown to be a viable approach to RWD extraction. This study aims to explore the treatment patterns and outcomes of patients with aLC who had ever received IO, using data extracted from patient records using AI.
Methods
This retrospective chart review includes 2,435 patients (≥18 years) with aLC diagnosed at Alberta Health Services between January 1st 2010 and January 1st 2019. An AI engine, DARWEN™, was used to extract 21 clinical features including clinico-demographic, tumour, and treatment information. AI outputs were fully validated against a manually-curated dataset and exceeded the required data standards; all features had over 90% accuracy except for smoking status which had an overall accuracy of 82%. Traditional Cox regression models were used to assess the relationships between clinical covariates and treatment duration or overall survival.
Results
Of the total aLC cohort (n=2,435), 408 patients received IO, mostly as second line treatment (53%). Since 2017, aLC patients were increasingly receiving IO as first-line therapy, demonstrating longer treatment duration than those receiving IO at later lines (HR: -0.4, 95% CI: -0.6, -0.2; p<0.01). The use of steroids at any point during treatment was associated with shorter IO treatment duration (HR: -0.6, 95% CI: -1.0, -0.2; p<0.01) but not overall survival. Patients with a diagnosis of adenocarcinoma were twice as likely to survive than those with squamous cell carcinoma (HR: 2.1, 95% CI: 1.1, 4.1; p=0.03).
Conclusions
With the use of novel therapeutic agents in practice, real world patient experiences provide valuable insights into treatment outcomes and personalized care strategies. AI technology can be leveraged to extract accurate real-world patient-level data at scale demonstrating evolution of treatment patterns and clinical covariates which impact real-world patient outcomes.
