OBJECTIVES: We explored how Artificial Intelligence (AI) can assist researchers in answering questions in oncology care. Specifically, we leveraged AI to assess whether published clinical outcomes have improved over time, and how that answer varied across different cancer types - Multiple Myeloma (MM), Acute Myeloid Leukemia (AML) and Glioblastoma (GB). The goal was to use AI to generate trends in overall survival (OS), over time, for analysis.

METHODS: We used a previously constructed AI to surface the relevant observational studies and extract the survival results for analysis. The AI automatically parsed OS results from 1,392 MM abstracts, 1,067 AML abstracts and 1,245 GB abstracts (after trimming extreme values), starting from 11,497, 16,603, and 8,378 search results, respectively.

RESULTS: For the trendline, we plotted the 5-year moving average of the mean overall survival values, per year, and fit it using linear regression (the moving average smoothed the data for fitting). For MM, there was a clear, positive trend in OS, as the trendline had a positive slope of 0.73 (r² = 0.80 indicating a good fit to the data). Therefore, OS for MM appears to improve over time. This was in contrast to AML, where the improvement still exists (the slope was 0.2 and positive), but it was weaker (the r² of the fit was 0.26). Finally, within GB we saw almost no relationship between OS and time (the slope was only 0.03, and the relationship was weak with an r² of 0.025).

CONCLUSIONS: Using AI to extract OS from thousands of abstracts, we identified trends in care with potential clinical implications. The correlations of OS over time were consistent with clinical progress in treating these conditions, giving confidence in the clinical validity of this approach. Further, we could examine many data points, across time, comparing care in different cancer types.