This article proposes a coding scheme for identifying and assessing linguistic evidence of problematic understanding in health-care provider communication with patients affected by type 2 diabetes mellitus. Drawing on the existing literature in pragmatics and linguistics, the scheme is grounded on the distinctions between the different types of linguistic evidence of the occurrence of a misunderstanding or a problematic understanding, divided into three levels (stronger, acceptable and weak) based on their probative force. The application of the scheme is illustrated through a pilot study, conducted on an Italian corpus of 46 transcripts of videotaped consultations between six health-care providers and 13 patients affected by diabetes mellitus type 2. The most frequent types of linguistic evidence of problematic understanding were the categories of "acceptable" (amounting to 58% of the total) and the "strong" evidence (35%). Patients' problematic understanding was detected to occur significantly more frequently than health-care providers. Providers were also found to be significantly more aware of possible misunderstandings, tending to verify more frequently the correctness of their own interpretations. This pilot study represents a first step in the process of developing a productive evidence-based tool for detecting problematic understanding, which can be used for implementing linguistic strategies for helping prevent the risk of misunderstandings in health-care communication. Our findings show that misunderstandings are widespread between patients and that some linguistic strategies may be more effective than others in preventing the risk of misunderstandings, suggesting possible directions of research for improving health-care providers' communicative skills.