OBJECTIVES

Current data acquisition through digital technology (e.g. Artificial Intelligence [AI]) has allowed unprecedented characterization of large digital patient data infrastructures. Here we set out to investigate whether AI can facilitate collection and perform trend analysis of peer-reviewed evidence in survival outcomes in Multiple Myeloma.

METHODS

Using a previously reported AI algorithm, 13,504 PubMed records (published before Dec 30, 2020) were identified and screened for unique overall survival (OS) and progression free survival (PFS) outcomes in MM. Eligible abstract records were processed, publication and study characteristics with supporting survival outcomes were automatically extracted and subsequently downloaded into Excel for further synthesis. Descriptive statistics were used to summarize the data and analyze trends over time.

RESULTS

The AI algorithm identified 2,948 unique abstract records in MM that reported OS and/or PFS (751 randomized controlled trials (RCTs) and 2,197 observational studies). OS and PFS were reported 1,780 and 1,220 times, respectively. For drug intervention studies, OS and PFS duration reported by RCTs were stable across time (mean: 34.0 months [SD:21.2, n=288] and 18.7 months [SD:15.4, n=420], respectively); observational drug intervention studies showed similar stability with a slightly longer duration (mean: 37.3 months [SD:27.0, n=590] and 20.2 months [SD:15.4, n=425], respectively). For non-interventional observational studies (i.e. epidemiology studies) an upward trend in OS duration was observed, from 28.9 months (SD:23.8, n=44) before 2000 to 53.8 months (SD:31.6, n=757) in 2020 with PFS duration remaining stable over time (mean: 27.6 months [SD:17.6. n=267]).

CONCLUSIONS

Using AI enabled us to robustly analyze and synthesize outcome-related research using large volumes of published peer-reviewed literature in MM. Interestingly, although survival outcomes in MM drug-intervention studies remained stable over time, OS duration in non-interventional study-settings doubled to 4.5 years over the last two decades, likely due to study design, including longer follow-up periods, common to these types of studies.