Website: https://www.haigekassa.ee/enBusiness domain: health servicesWorking language: Estonian/EnglishAvailable positions: 1Suitable curricula: Mathematics and Statistics, Actuarial and Financial EngineeringDescriptions of the topics:
Every time an insured person visits a medical specialist, the hospital invoices the Estonian Health Insurance Fund (EHIF) for the visit. The treatment invoice lists, among other data, all health services and diagnoses of the patient relevant to the visit. In the process, the EHIF has compiled data about diagnoses and accompanying health services through time. We could assume that patients with identical diagnoses might get similar health services, regardless of the particular hospital or medical specialist treating them. Significant deviations from the established standard practice might elicit closer inspection and analysis.
The project encompasses modeling the number of different health services provided to a person during a time period (e.g. a year) depending on the diagnoses of the person, and other health data available to EHIF. The aim is to detect “standard packages” of health services related to a particular history of diagnoses and highlight significant deviations from it.
Models for multiple response variables (incl. machine learning) can be the tools to be used, including neural networks and canonical correspondence analysis (or other methods used for analyzing ecological community data). Thus we presume some skill of R or Python.
As a simpler initial task, the total sum of all treatment invoices of a person during a time period can be modelled.
Certain chronic conditions (e.g. hypertension, diabetes) can be effectively managed by continuous monitoring and (appropriate updates of) treatment. However, if left untreated, the patient may experience a health deterioration and end up in hospital. This is not a desired outcome neither for EHIF nor for the patient. Therefore continuous monitoring of these risk-patients is necessary.
Patients with forementioned chronic conditions are already under increased attention of general practitioners. However, a pilot project has demonstrated that the target population can be somewhat more focused using machine learning. The aim of this project is to validate and build upon the pilot.
In essence, this is a binary classification task based on EHIF claims, prescriptions and insurance data. Therefore the choice of tools is wide, extending from logistic regression to the reach of imagination. A significant part of the day-to-day work is anticipated to be feature engineering. We presume some skill of R or Python.
The industrial master's programme is intended for first-year master students.
Submission deadline is October 12, 2020.
To apply, please send to meelis.kaarik [ät] ut.ee
Besides the CV and the motivational letter, the applicant's academic achievement is also taken into account.
After the submission deadline, interviews will be held at the Institute of Mathematics and Statistics (if needed) and also at the partner company. Exact dates will be announced on an ongoing basis.