OBJECTIVES:  To determine the sensitivity and specificity of software based on natural language processing to classify PubMed abstracts relevant to the humanistic or economic burden of disease. METHODS:  We developed an online database of abstracts of over 100,000 studies identified by a systematic search of PubMed on the humanistic and economic burden of disease (www.heoro.com). We manually indexed 10,000 abstracts to one or more study types: PRO study (with subtypes PRO validation study and Utility study), Costs and Resource use studies (Direct and indirect costs, resource use, treatment patterns and adherence), Economic models (cost-effectiveness, cost-utility, cost-benefit and other models) and mortality, as well as geographical location. We used this training set and expert assessment to iteratively develop classifiers from text, MeSH headings and metadata in the abstracts. We then assessed the accuracy of the classifiers on samples of the remaining 90,000 abstracts by expert checking of 200 abstracts scoring positive and 400 negative for each study subtype. RESULTS:  The classifiers had a sensitivity and specificity of 96% for PRO study identification, sensitivity of 87 to 99% and specificity of 93 to 100% for economic model types, sensitivity of 79 to 99% and specificity of 92 to 98% for costs and resource use study subtypes, and sensitivity of 82% and specificity of 97% for mortality studies. They also identified randomised controlled trials and systematic reviews with 93-95% sensitivity and 99% specificity. Indexing to geographical location was 97% accurate in an analysis of 200 abstracts. CONCLUSIONS: With overall accuracy of around 95%, the classifiers compare well with human indexing of study types. As 90,000 abstracts could be indexed accurately within hours, this method facilitates a highly streamlined approach to identifying relevant data for health economics and outcomes research.