OBJECTIVES: With increasing demand for including NLP of text data for use in health economics and outcomes research (HEOR), there is a need to document and identify best practices.  Qualitative data provides greater detail beyond the existing fields captured in EMR/EHRs, and have helped bolster patient-related information to be used by the practice and in research.  However, the use of NLP is wrought with challenges stemming from incomplete, missing, or inaccurately entered data. We present one case study with the process, issues, and potential recommendations for using text information in HEOR.   METHODS: Researchers identified key terms available from unstructured fields and established search criteria in a dataset of CML patients.  Key words through NLP were established to identify information from free text, indicating patient’s disease state: chronic phase, accelerated, blast crisis.  Barriers to abstracting qualitative data through NLP were documented, and recommendations were made.    RESULTS: Of 152 patients, 85 (56%) had available text data on disease state.  Each patient could have more than one record, yielding 112 records overall. There were seven instances where multiple conflicting terms were recorded for the same patient on the same date.   Our research team applied clinical assumptions by using treatment information and took the most advanced disease state, i.e., if a patient had “accelerated” and “blast crisis” on the same date, then “blast crisis” was assumed and verified using treatment orders.  If multiple dates existed, we took the most recent date to the index date.  We were able to eliminate 24% of the multiple records and supplement the findings for these 85 patients. CONCLUSIONS: There is growing potential for the use of qualitative data in HEOR.  Research is needed to address the potential problems, degree of completeness and limitations of NLP that will help future mining techniques and the quality of the data uncovered.