atientMiner

Language and knowledge technology for interpreting medical data sources

Health Trajectories

Where we are


So far, the mapping of health trajectories has been mostly carried-out in a hypothesis-driven manner with focus on a few diseases or specific comorbidities to index diseases, constraining the description of trajectories to their closely limited complications. Recently, large volume data analyses using centralized electronic health registries have been successfully used to describe the temporal patterns of diseases in large population groups, uncovering hitherto neglected trajectories of health that link apparently benign conditions to long-term, life-threatening conditions [Jensen et al. 2014].

Mapping of health trajectories for patients with Chronic Obstructive pulmonary diseases using data on 6.2 M patients from Denmark from the centralized Danish NPR, which contains administrative information and primary and secondary diagnoses covering every hospital contact in Denmark. [Jensen et al., 2014]
Mapping of health trajectories for patients with Chronic Obstructive pulmonary diseases using data on 6.2 M patients from Denmark from the centralized Danish NPR, which contains administrative information and primary and secondary diagnoses covering every hospital contact in Denmark. [Jensen et al., 2014]

Challenges

The mapping and understanding of these trajectories and influencal factors is limited by several challenges:

  • lack of detailed information included in structured registries
  • consequently, limited capacity to predict individual pathways of progression towards severe diseases: difficulty to plan and develop prevention and early interventions
  • health registries only contain a fraction of the clinical information contained in individual medical records, and access to these data is limited by its volume, variability, velocity and heterogeneity

Therefore, it is essential to develop strategies for intelligent data mining of Electronic Health Records (EHR)  to map individual trajectories of health in order to support the development of prediction models and intervention strategies aimed at curbing these trajectories towards the least-unfavorable outcomes.

Solution

PatientMiner addresses these issues by developing a new strategy for linking unstructured data from EHR to map health trajectories based on an already existing running system developed in the frame of Synodos, a French collaborative project, dedicated to building technologies easy to use by medical staff for performing epidemiological studies.
With the development of EHRs, the linguistic analysis of textual data in the medical sector is receiving increasing interest. However, existing approaches vary in terms of the granularity and sophistication of linguistic processing carried out.