Fænotypeinddeling og prædiktion af ældre borgeres longitudinale hjemmeplejebehov

Jensen, Steffen ; Gadensgaard, Mathilde Fischer

4. semester

Sundhedsteknologi (cand.polyt.), Kandidat

2020

2020-06-02

105 pages

Background The municipality-based eldercare is in the need of improving the efficiency on account of having to manage an increasing amount of tasks while limiting the expenses hereof. In this regard, the increasingly diverse group of elderly citizens poses a challenge. From this, it is investigated whether data from electronic home care records may be utilized in a data-driven application to extract and predict phenotypes of care requirements. Methods and Materials The present study developed and implemented a system consisting of two central parts. The first part is made up of a comparison of k-means clustering and hierarchical agglomerative clustering for the task of extracting care requirement phenotypes from the longitudinal trajectory of elderly citizens in the duration of a year. The second part constituted a classifier for prediction of the care requirement phenotype of a citizen, based on demographic data and information regarding the first visitation. In this regard, a Random Forest and a Neural Network were compared. Results The developed proof of concept-system, deduced six unique care requirement phenotypes using k-means clustering. In the subsequent classification of the care requirement phenotype, the system achieved a maximum ROC AUC of 0.77 and PR AP of 0.43 with the implemented Random Forest. Conclusion Data from electronic home care records may be utilized in a data-driven application for extracting phenotypes, using k-means clustering. Furthermore, the use of a Random Forest classifier applied on data from the first visitation of the citizen, constitute a potential method for estimation of the future care requirement

Clustering ; K means ; Neural network ; machine learning

Download
Vis denne rapport i AAU Studenterprojekter