Predictive algorithm for citizens with COPD

Student thesis: Master thesis (including HD thesis)

  • Hannah Møldrup Derby
  • Mille Andersen
  • Rasmus Steengaard Holm
4. term, Clinical Science and Technology, Master (Master Programme)
Summary
Background
Chronic Obstructive Pulmonary Disease (COPD) is associated with increased socioeconomic costs, with hospitalizations being the largest. Therefore, telemedicine is offered to all citizens with COPD as a solution to prevent hospitalizations. However, studies show that this does not benefit the economy. Several studies suggest the use of prediction algorithms as an opportunity to identify which citizens can help reduce the socio-economic costs by using telemedicine.

Method
An overall group and 10 stratified groups are prepared, where prediction algorithms are developed based on both Multiple Linear Regression (MLR) and Random Forest Regression (RFR). The algorithms must predict which citizens are at risk for the most bed days, as the costs associated with a hospitalization depend on the amount of bed days. Each group's data set is preprocessed and divided into a training and test set. The prediction algorithms are validated by a correlation analysis, where a coefficient of determination is calculated. The predictive power of the best prediction algorithms is illustrated in scatterplots.

Results
The predictability of the prediction algorithms based on both MLR and RFR is not acceptable in relation to predicting which citizens are at risk for the most bed days. The highest coefficient of determination at MLR is 0.228, where at RFR it is 0.288.

Conclusion
The results in the thesis were not acceptable, neither in the overall group nor the stratified groups, so the developed prediction algorithms cannot be used to predict which citizens are at risk for the most hospitalization days.
LanguageDanish
Publication date31 May 2021
Number of pages42
ID: 413365016