((B-162)) (B-162) Artificial Intelligence-Based Estimation of the CGI-I to Address Gaps in Real-World Data and Increase Study Sample Size

Background: The Clinical Global Impression Scale - Improvement (CGI-I) is a widely-used, clinician-reported measure for global assessment of change in patients with major psychiatric disorders, such as major depressive disorder. The CGI-I is a useful tool for evaluating treatment response and patient outcomes over time, but documentation of the CGI-I is inconsistent in real-world data sources such as electronic medical records (EMRs). This limits the utility of these data for supporting large, heterogeneous real-world studies. A previous effort applied artificial intelligence (AI) methods to estimate CGI-I scores for patients using routinely-recorded clinical notes data with very good performance.

Objectives: This effort assessed the feasibility of using the estimation model to increase the sample size for RWD studies of treatment response in different drug classes.

Methods: The model was applied to the OM1 MDD PremiOM Dataset, a RWD source containing data on over 490,000 MDD patients with a diagnosis of depression and receiving treatment from a mental health professional. Patients met the following inclusion criteria: 1) a diagnosis of MDD; 2) new drug initiation (index date) for a drug indicated for depression treatment; 3) baseline observation within 90 days prior to 14 days after the index day; and 4) follow-up observation between 45 and 273 days post-index date.

Results: The cohort included 182,750 patients. Of these, 38,252 had at least one recorded CGI-I score in the study timeframe. The remaining 144,498 patients had estimated CGI-I (eCGI-I) scores generated by the AI model. Increases of 4.3x to 6.0x in available study sample size were observed across drug classes. Specifically, sample sizes increased for serotonin-specific modulators (6.0x), serotonin-norepinephrine reuptake inhibitors (5.1x), norepinephrine dopamine reuptake inhibitors (5.0x), serotonin antagonists and reuptake inhibitors (5.0x), atypical antipsychotics (4.9x), tetracyclic antidepressants (4.5x), tricyclic antidepressants (4.4x), and selective serotonin reuptake inhibitors (4.3x).

Conclusions: Use of an AI-based model to estimate CGI-I scores for patients with depression increased the number of patients available for RWD studies across drug classes.