Forecasting the consumptions of coagulation tests using a deep learning model

Abstract

Background: Laboratory professionals aim to provide a reliable laboratory service using public resources efficiently while planning a test's procurement. This intuitive approach is ineffective, as seen in the COVID-19 pandemic, where the dramatic changes in admissions (e.g. decreased patient admissions) and the purpose of testing (e.g. D-dimer) were experienced. A model based on objective data was developed that predicts the future test consumption of coagulation tests whose consumptions were highly variable during the pandemic. Methods: Between December 2018 and July 2021, monthly consumptions of coagulation tests (PTT, aPTT, D-dimer, fibrinogen), total-, inpatient-, outpatient-, emergency-, non-emergency -admission numbers were collected. The relationship between input and output is modeled with an external input nonlinear autoregressive artificial neural network (NARX) using the MATLAB program. Monthly test consumption between January and July 2021 was used to test the power of the forecasting model. Results: According to the co-integration analysis, the total number as well as the number of emergency and nonurgent examinations and the number of working days per month are included in the model. When the consumption of aPTT and fibrinogen was estimated, it was possible to predict the consumption of other tests. Fifty months of data were used to predict consumption over the next six months, and prediction based on NARX was the more robust approach for both tests. Conclusion: The deep learning model gives better results than the intuitive approach in forecasting, even in the pandemic era, and it shows that more effective and efficient planning will be possible if ANN-supported decision mechanisms are used in forecasting.

Keywords

• coagulation test
• test consumption
• test procurement
• deep learning
• artificial neural network
• narx (nonlinear autoregressive with external input) neural network