Development of an algorithm to find persons with intellectual disability in general practitioner data
Background
Within the academic collaborative center “Sterker op Eigen Benen”, we aim to improve healthcare for people with intellectual disabilities (ID). Research using routinely collected healthcare data plays an important role in this. General practitioner (GP) healthcare records are such a data source. Although GPs have the option to register if a person has ID by using the International Classification of Primary Care (ICPC) code for mental retardation, this code is often not used for a person with ID, partly because GPs are not familiar with the code or find it stigmatising, leading to under-registration.
CARMENDA is the data infrastructure program within the “Sterker op Eigen Benen” collaborative that currently applies various methods to find persons with ID in GP data, combining coded and textual searches. Although these methods are sufficiently efficient on a relatively small dataset containing approximately 500,000 GP records, there is much room for improvement now that Carmenda has gained access to four other GP datasets with approximately equal size.
Objective
The objective of this project is to improve the existing methods to find persons with ID in GP healthcare records and develop an algorithm that can predict the likelihood that a person has ID based on GP data.
Activities
- Describe
- What variations does one encounter when investigating the free text episode titles GPs can give to episodes of care for persons with ID?
- How are these registrations distributed among eight different categories that we have currently distinguished in GP records of persons with ID: mental retardation, Down syndrome, other syndromes, intellectual disability, mentally handicapped, low intelligence, psychomotor retardation, other?
- Improve case searching
- As existing search strings miss various combinations of possible misspellings of words, regex search will be applied to reduce the number of false negative cases.
- Outliers will be manually inspected and the search string will possibly be adjusted to reduce the number of false positives.
- Labeling & AI model development
- Label GP records into three possible classes: certain, possible, unlikely
- Split the dataset to create train, validation, and test sets
- Apply various AI algorithms and determine their performances after having selected the most relevant performance metrics
- Model evaluation
- Evaluate the performance of the different AI algorithms
- Motivate which model performs best and apply this model to the test set
- Reporting & Recommendation
- Write a summary report about your project
- Provide a recommendation for implementing the model in the Carmenda research infrastructure
Expected Results
- Improved (regex) search to find persons with ID in GP records
- Labelled dataset to obtain three classes of GP records
- Validated AI algorithm that predicts the likelihood that a person has ID
- Recommendation for implementing the model
Practical Information
- Level: BSc/MSc in Computer Science, AI, Data Science, Health Technology, or related field
- Location: Nijmegen, partly hybrid
- Start date: to be agreed upon
- Language: being fluent in Dutch is a prerequisite
Supervision & Contact Details
- Pim van Oirschot (Radboudumc / “Sterker op Eigen Benen”)
- Laurens Reumers (Radboudumc / “Sterker op Eigen Benen”)
