Aims: Evaluate the benefit of utilising artificial intelligence (AI) software to find and validate patients as trial eligible using a commercial product on retrospective cohorts.
Methods: The AI software was utilised to generate potential patient lists for 6 trials by entering relevant inclusion and exclusion criteria, from a limited electronic medical record extract from June 2017 to October 2018. The AI uses proprietary algorithms to identify medical concepts from unstructured data (e.g. physician’s notes) as well as structured data (e.g. ICD codes). These results were compared with trial screening logs. Each potential patient was manually classified by a physician as concordant (identified by both AI and standard recruitment); AI-missed (AI failed to identify a patient found through standard recruitment) and AI-augmented (AI detected and not in screening log). The time required for software usage was measured. Descriptive statistics were used.
Results: The data source included 6028 patients with clinical, pathology and diagnostic imaging documents totalling 8178197 data items. There were 2 trials each of phase 1, 2 and 3. The studies were open for an average of 526 days prior to the end of the data extract period. Three are still open. Out of 234 patients on AI pre-screening, 20 were trial eligible after manual review. Of those 20 patients, 14 were concordant, six patients were AI-augmented and 26 patients were AI-missed. AI-augmented patients constituted 15% additional potential patients above standard process. The time to design trial queries and virtual screen all patients was 513 minutes. Incomplete EMR data was a notable source of AI-missed cases.
Conclusions: Use of AI Software can augment conventional patient recruitment processes and maximise patient recruitment, in a time-efficient manner. The performance of the software is highly dependent on the physician’s language and presence of adequate EMR content.