Expressive query language for arbitrary data ontologies (ER models) that can be easily perceptible by non-programmers (domain experts) is still an open problem. In this paper we introduce an important subset of data ontologies called semistar ontologies. Patient health records are typical examples of semistar ontologies. We propose a new natural language-based query language for semistar ontologies and provide an efficient implementation for the language. The expressiveness of the query language was demonstrated by turning it into a working language for Riga Children's Clinical University Hospital. Experiments with potential end-users have shown that the proposed query language can be taught to non-programmers in a couple of hours.