Lessons learned from the development of the CancerLinQ prototype: Clinical decision support.
- 2013
Available online from MWHC library: 1999 - present, Available in print through MWHC library: 1999 - 2008
237 Background: CancerLinQ (CLQ) is a rapid learning system (RLS) for oncology in development by ASCO. CLQ is based on the transfer of electronic health records (EHR) from participating oncology practices to a data warehouse where data aggregation and de-identification occurs. A prototype was built using open source software and has collected de-identified data on 170,000+ pts with breast cancer (BC) from 31 community oncology practices using 4 different EHRs. The primary goals for the prototype were 1. Aggregate patient data from any EHR platform, process it and create a longitudinal record; 2. Develop quality reports from EHRs; 3. Point of care Clinical Decision Support (CDS) from ASCO guidelines; 4. Data visualization for hypothesis generation; 5. Demonstrate desire to share data for quality improvement; 6. Describe lessons learned (LL). This report focuses on LL about CDS. CONCLUSIONS: Conversion of ASCO's clinical guidelines into a MR format is possible. New and emerging methods such as GLIDES, BRIDGE-Wiz, and GEM-cutting provide excellent tools to migrate existing narrative recommendations into MR format that can populate CDS tools, such as those provided by CancerLinQ. METHODS: Physician experts identified specific elements from each ASCO BC guideline to make machine readable (MR). Abstractors then GEM-cut the elements using the GEM Abstraction Manual and Style Guide. The output reports were reviewed for comprehensiveness, accuracy, and style. Following verification of the GEM-cut content, reports were sent for meta-tagging, done by selecting widely used EHR vocabulary from the Unified Medical Language System (UMLS). The GEM-cut output and meta-tags were converted to DROOLS syntax and the resulting coded files were inserted into the DROOLS rules engine. When the rules engine encounters a combination of facts that match a rule, that rule is presented to the user. The enduring responses are collected using 'queries' and the CDS results are delivered to the EHR. RESULTS: Guidelines are often not written as "if"/"then" statements which is key for computer-based CDS. Any unintentional ambiguity must be removed for machine MR CDS. Using new methodologies, we have been able to convert narrative guidelines into MR CDS.