![[ Visit Client Website ]](images/banner.gif)
Wednesday, September 2, 2009: 3:20 PM
Hanover E
The Iraq and Afghanistan wars, Katrina, and the World Trade Center all have contributed to a greater awareness of Post Traumatic Stress Disorder as a serious public health concern. Early detection of signs, symptoms, and manifestations of PTSD can assist with overcoming barriers and gaps in current treatment plans for the at-risk population of victims of trauma, leading to better health, quality of life and cost-effective care.
In our research, VA Compensation and Pension Examination (C&P) report documents were used as a platform for the development and testing of concept-based indexed natural language processing to detect terminology for signs, symptoms, and manifestations representing the case definitions and clinical sequelae of PTSD based on DSM-IV criteria. We utilized the Mt. Sinai Multithreaded Clinical Vocabulary Server (MCVS) indexing engine to process words and phrases from text reports into SNOMED CT encoded concepts. Rules for PTSD that model the DSM-IV criteria were applied to indexed records, and reviewers verified the true positives and negatives as well as the false positives and negatives.
Overall, positive predictive value was 78%, and sensitivity was 52%. These preliminary findings establish the lower limit for estimating the validity of using concept-based indexing for identification and classification of PTSD. The symptom clusters with greater rule extraction frequency in PTSD cases were Affective (37%), Physical (15%), and Cognitive (13%). Thus, affective symptoms predominated in PTSD cases.
Our findings demonstrate the ability to extract exposure, medical, and functional status information from the electronic record. We are refining this preliminary rule set by combining facts and qualifiers to form more sensitive use of the extracted information. This study suggests that NLP is a practical approach for the surveillance of medical records for PTSD signs and symptoms
In our research, VA Compensation and Pension Examination (C&P) report documents were used as a platform for the development and testing of concept-based indexed natural language processing to detect terminology for signs, symptoms, and manifestations representing the case definitions and clinical sequelae of PTSD based on DSM-IV criteria. We utilized the Mt. Sinai Multithreaded Clinical Vocabulary Server (MCVS) indexing engine to process words and phrases from text reports into SNOMED CT encoded concepts. Rules for PTSD that model the DSM-IV criteria were applied to indexed records, and reviewers verified the true positives and negatives as well as the false positives and negatives.
Overall, positive predictive value was 78%, and sensitivity was 52%. These preliminary findings establish the lower limit for estimating the validity of using concept-based indexing for identification and classification of PTSD. The symptom clusters with greater rule extraction frequency in PTSD cases were Affective (37%), Physical (15%), and Cognitive (13%). Thus, affective symptoms predominated in PTSD cases.
Our findings demonstrate the ability to extract exposure, medical, and functional status information from the electronic record. We are refining this preliminary rule set by combining facts and qualifiers to form more sensitive use of the extracted information. This study suggests that NLP is a practical approach for the surveillance of medical records for PTSD signs and symptoms
<< Previous Abstract
|
Next Abstract