Abstract
BACKGROUND The explosive increase in medical literature has changed therapeutic strategies, but it is challenging for physicians to keep up-to-date on the medical literature. Scientific literature data mining on a large-scale of can be used to refresh physician knowledge and better improve the quality of disease treatment. MATERIAL AND METHODS This paper reports on a reformulated version of a data mining method called MedRank, which is a network-based algorithm that ranks therapy for a target disease based on the MEDLINE literature database. MedRank algorithm input for this study was a clear definition of the disease model; the algorithm output was the accurate recommendation of antihypertensive drugs. Hypertension with diabetes mellitus was chosen as the input disease model. The ranking output of antihypertensive drugs are based on the Joint National Committee (JNC) guidelines, one through eight, and the publication dates, ≤1977, ≤1980, ≤1984, ≤1988, ≤1993, ≤1997, ≤2003, and ≤2013. The McNemar's test was used to evaluate the efficacy of MedRank based on specific JNC guidelines. RESULTS The ranking order of antihypertensive drugs changed with the date of the published literature, and the MedRank algorithm drug recommendations had excellent consistency with the JNC guidelines in 2013 (P=1.00 from McNemar's test, Kappa=0.78, P=1.00). Moreover, the Kappa index increased over time. Sensitivity was better than specificity for MedRank; in addition, sensitivity was maintained at a high level, and specificity increased from 1997 to 2013. CONCLUSIONS The use of MedRank in ranking medical literature on hypertension with diabetes mellitus in our study suggests possible application in clinical practice; it is a potential method for supporting antihypertensive drug-prescription decisions.