5 things needed to transform oncology clinical trial prescreening with AI
Prescreening potential subjects in oncology clinical trials is one of the most significant limiting factors in identifying eligible participants which requires trained medical professionals to apply manual effort and resources to flag prospects, scan charts and confirm eligibility. The process of screening large number of patients and evaluating results is time consuming and expensive.
An effective pre-screening AI-enhanced tool can allow clinical trials of potentially lifesaving treatments tested by principal investigators to be more accessible to patients. There has been significant progress in pulling information out of large data sets by developing artificial intelligence (AI) tools. There is a need for automated AI method utilized to assist in clinical trial prescreening. The challenges faced include developing sophisticated algorithms to effectively deal with unstructured clinical oncology data; integrating clinical AI into practices, procedures, data flows and administrative clinical setting.
Patient Pre-screening Problems
80-90 percent of unstructured medical data consisting largely of scanned documents, pathology or lab reports and clinical images is legible only by humans but not machines making patient pre-screening a highly manual process. Prescreening is conducted by trained medical professional to scrutinize traditional chart reviews, medical records, extracted relevant matching attributes and physician notes of prospective subject’s for flagging potential match during the course of regular treatment. This costly and slow process imposes significant challenge by excluding possible eligible patients, burdening clinics and hospitals.
To meet increasingly complex and specific eligibility criteria of identifying appropriate patients for a specific oncology clinical trial requires understanding the disease state by screening large number of potential patients. Electronic medical records (EMRs) were originally designed to document provider assessments activities, facilitate billing and treatment plans. EMRs can be used to analyse and provide structured diagnostic data while AI and natural language programming can automate the pre-screening process. Most of the attention needs to be focused in sorting AI complexity of the clinical system.
Challenges with AI solution
The five specific factors essential for an effective AI-supported prescreening process are obviously neglected and surprisingly difficult to accomplish.
-
Data extraction
The fundamental need of AI is researched to extract information from unstructured mass of clinical data in medical records.
-
Natural Language Programming
The extracted and de-identified information needs to be fed into an oncology-specific engine for natural language processing (NLP) pertinent for trial matching. The clinical language is vague, ambiguous and implicit compared to ordinary language. AI must be specific to oncology therapeutic area to continuously train and improve these models. Domain-specific AIs can perform well with specialized data type and specific tasks of moderate complexity.
-
Guidance by clinical staff
The purpose of AI tool is to augment human abilities of clinicians experienced in oncology by refining and guiding AI recommendations.
-
Access to large clinical data sets
AI needs large amounts of healthcare and patient data from useful heterogeneous sources which is difficult to acquire. Healthcare data is stored in variety of formats in difficult-to-access locations guarded by a wide range of accessibility rules. Useful AI system for oncology clinical trial prescreening will need large number of hospitals and clinics to securely acquire, transfer, store and process reliable clinical data.
-
Process and contract integration
AI already has an intimate relationship with contracting, IT, finance and other stakeholders. AI must be an integral part of normal operations carried out in hospitals or clinics pertaining to security reviews, IP assessments, slotting budget cycle and entire IT on board process.