Our natural language processing engine was built specifically for Brazilian medical Portuguese. It reads progress notes, nursing notes, prescriptions, and reports — extracting structured clinical variables from free text, at scale.
Over 100 million clinical texts processed in Brazilian medical Portuguese.
An NLP model trained on English — or even on formal Portuguese — cannot read a Brazilian medical record with clinical precision. The reason is simple: the language of Brazilian medical records is unique.
Physicians use abbreviations that vary by specialty, by region, and even by institution. 'ICC descompensada c/ edema MMII' is a phrase any Brazilian cardiologist understands instantly — but an NLP not trained on that vocabulary will misread or ignore it. The same goes for negation structures ('nega dispneia'), temporality ('há 3 semanas'), diagnostic uncertainty ('a esclarecer'), and complex clinical relations ('evolui com').
iHealth spent more than a decade collecting, annotating, and processing real clinical text to teach an NLP model to understand this language with precision. The result is an engine that doesn't just read — it interprets clinical text with the rigor that regulatory use and scientific research demand.
Automatically identifies and classifies clinical entities in the text: confirmed diagnoses and hypotheses, prescribed and administered medications, procedures performed, reported symptoms, exams and results, comorbidities and family history. Each entity is extracted with context — negation, temporality, and diagnostic certainty — because in clinical text, context completely changes meaning.
We maintain a living clinical dictionary with more than 50,000 abbreviations, acronyms, and regional variations of Brazilian medical Portuguese — continuously updated. Each abbreviation is disambiguated in context: 'DP' can mean peritoneal dialysis in nephrology, precordial pain in cardiology, or Parkinson's disease in neurology. Our NLP knows the difference.
The model identifies with high precision negation structures ('nega', 'sem', 'ausência de'), temporality ('há 3 meses', 'na admissão', 'ao longo da internação'), and diagnostic certainty ('confirmado', 'suspeito', 'a investigar'). These elements are essential for building accurate longitudinal clinical histories and for use in regulatory studies.
Beyond individual entities, the NLP extracts the relationships between them: which medication was prescribed for which diagnosis, which exam confirmed which suspicion, which symptom led to which procedure. This turns free text into a clinical knowledge graph that supports analyses of patient journey, treatment effectiveness, and drug safety.
The NLP output isn't just annotated text — it's structured, traceable, and auditable data. Every extracted clinical variable comes with a reference to the source text excerpt, the model that extracted it, the model version, and the processing date. Full traceability for submissions to CONITEC, ANVISA, FDA, and EMA.
Structures unstructured medical record data for use in comparative effectiveness studies, disease burden, patient journey, and external control arms — with regulatory-grade quality for CONITEC, ANVISA, and global submissions.
Cross-references a protocol's inclusion and exclusion criteria with free-text clinical records, identifying eligible candidates in minutes — not weeks of manual review.
Reads the full medical record — including free text — and identifies uncoded diagnoses and procedures that impact the DRG and, consequently, hospital billing.
Structures clinical record data into queryable variables, enabling care performance, patient safety, and strategic management dashboards to run on the institution's real data.
Talk to our team and understand how iHealth's technology pillars translate into real results — for clinical research, market access, hospital management, or data auditing.
