Clinical decision support with automated text processing for cervical cancer screening

Area of Science:

• Medical Informatics
• Oncology
• Public Health

Background:

• Cervical cancer screening relies on accurate interpretation of Papanicolaou (Pap) test reports.
• Electronic Medical Records (EMR) contain valuable free-text data that is often underutilized.
• Developing automated systems can enhance the efficiency and accuracy of cervical cancer screening protocols.

Purpose of the Study:

• To develop and evaluate a computerized clinical decision support system (CDSS) capable of interpreting free-text Pap reports.
• To leverage natural language processing (NLP) for extracting actionable information from unstructured clinical data.
• To generate optimal patient-specific screening recommendations for cervical cancer.

Main Methods:

• A CDSS was created with two rulebases: one for interpreting free-text Pap reports and another for national screening guidelines.
• The free-text rulebase was developed using a large corpus of 49,293 Pap reports.
• The system was evaluated on 74 patients, with recommendations reviewed by a physician, and subsequently upgraded to include human papillomavirus (HPV) testing results.

Main Results:

• The CDSS achieved 98.6% accuracy in providing optimal screening recommendations (73 out of 74 patients).
• The system identified two potential gynecology referrals missed by the physician and assisted in amending six recommendations.
• An initial limitation regarding separate HPV test reporting was addressed, leading to 100% accuracy in the upgraded system.

Conclusions:

• An accurate CDSS for cervical cancer screening is feasible with standardized reporting and explicit guidelines.
• Natural language processing effectively utilizes free-text EMR data to create valuable clinical decision support tools.
• This technology has the potential to improve the consistency and effectiveness of cervical cancer screening programs.