Severe hypoglycemia (SH) is one of the most prevalent diabetes complications and a critical public health concern; 25% of hospitalizations for adverse drug events is due to SH. Existing surveillance does not capture SH occurring in pre-emergency settings (e.g., treat and release by paramedics) because ambulance records are not included in the EMR. Thus EMR-based surveillance underestimates the incidence of SH. In this study, we will identify SH in records for ambulance calls linked to 235,000 type 2 diabetes patients from KPNC, and develop and validate a novel coding algorithm to identify SH events in ambulance records. This study will provide a new SH ascertainment tool to enable more comprehensive, valid and accurate surveillance efforts.