Patient cohort is similar symptoms of group of patients over a time period. It is important to correctly identify patient cohort for observational study or interventional study. To identify patient cohort, we can easily retrieve information from large structured and unstructured data tables, but this information may not fulfill our interest. We need to obtain knowledge from the unstructured medical data for further detailed inspection. In this work, we have improved structured extraction by presenting context on a health record dataset for identifying cohort of diabetic patients. Results shows that traditional structured query-based data extraction methods accurately identified 97.14% positive patients to our question of interest where adding natural language processing supported technique have retrieved 98.37% precisely.