Summarization of clinical narratives is a long-standing research problem.\nHere, we introduce the task of hospital-course summarization. Given the\ndocumentation authored throughout a patient's hospitalization, generate a\nparagraph that tells the story of the patient admission. We construct an\nEnglish, text-to-text dataset of 109,000 hospitalizations (2M source notes) and\ntheir corresponding summary proxy: the clinician-authored "Brief Hospital\nCourse" paragraph written as part of a discharge note. Exploratory analyses\nreveal that the BHC paragraphs are highly abstractive with some long extracted\nfragments; are concise yet comprehensive; differ in style and content\norganization from the source notes; exhibit minimal lexical cohesion; and\nrepresent silver-standard references. Our analysis identifies multiple\nimplications for modeling this complex, multi-document summarization task.\n