NTNU Health

Computational sepsis mining and modelling

Capturing and preserving individual case histories using novel technology for integrating temporal interpretation of clinical text with structured data; Establishing knowledge and decision support for individualized, real-time, infection and sepsis monitoring; Demonstrate a clinical dashboard extending the computerized chart for process monitoring and patient risk mitigation. Indirectly, improve interventions and reduce drug resistance.

Objectives and potential for impact

  1. Using advanced semantic modelling, extract detailed knowledge about infectious disease prevalence and exposure from textual and other health records for Central Norway. This includes re-analysing already established large registers and sources of secondary health information for which the collaborating research groups/centres has permission and consent to use. Impacts: Enable omics-wide association studies based on existing health information. Develop empirically validated set of care/disease cases representative for care improvement and risk reduction, usable for decision support and baselining.
  2. Further research and develop verifiable and scalable technology for massive clinical temporal data analysis, building on theory and tools for process and event mining developed and validated locally and by partners. Impacts: Prepare the Platform for secondary use of health data. Establish operative Platform-agnostic, case-based decision support. Make Norway, NTNU and Central Norway Health Trust an important centre for computational data-driven health.
  3. Establish a gold standard, validated, annotated clinical corpus and corresponding knowledge models for sepsis. Impact: Help nailing precise Sepsis prevention and treatment. Becoming an attractive multi-centre partner, with novel technology and high quality data, for international big health data projects.


Research tasks

  1. Improve machine learning methods for recognizing and classifying clinical statements, modifiers, indicators and variables.
  2. Further develop ontologies of infection-related phenotypes, interventions, findings.
  3. Carefully annotate a sufficiently large training corpus of clinical records.
  4. Employ health care process mining methods to derive commonly occurring patterns and subprocesses so as to build knowledge about dependent and independent events and processes.
  5. Train statement-classifiers and data-fusion algorithms using process knowledge, annotated corpus and ontologies.
  6. Further develop theories of constraint-based computational clinical reasoning.
  7. Use research registers, prevalence surveys and other sources to evaluate, improve and validate quality of the computational patient model.