A common obstacle to laymen of medicine is that they usually do not know the associations of health problems. This study detected the associations by extending the meaning of health problem terms, and found the most practical method among the compared four types of approaches: the probability-based LDA topic model, the thesaurus-structure-based method, the concept-mapping-based method, and the document-similarity-based method. The detection is on consumer-level medical information instead of electronic health records or medical literature, with the consideration of laymen’s digestion. The findings can be used in consumer-oriented medical information applications. The tested methods were evaluated by: (1) correlation analysis with the experts’ rating to show the overall performance; and (2) P@N to reflect the ability of detecting the strong associations. The results show that the LDA topic model-based method outperforms when each health problem is represented by the averaged topic probability distributions of related documents that are gathered from quality consumer-level medical information source.