University of Twente Student Theses

Login

Patterns in continuously recorded vital signs as predictor for deterioration in patients with suspected infection after admission to the emergency department

Wijk, R.J. van (2019) Patterns in continuously recorded vital signs as predictor for deterioration in patients with suspected infection after admission to the emergency department.

[img] PDF
370kB
Abstract:At the start of this century the incidence of sepsis was estimated to be 437 cases per 100,000 person-years and approximately 30{50% of the total amount of registered sepsis cases is admitted to the hospital via the emergency department (ED). Prediction of deterioration of patients admitted via the emergency department could help determining in an early stage which treatment is best for a patient. Using Heart Rate Variability (HRV) features derived from continuously recorded ECG signals possible predictors were explored and used to create a model for predicting deterioration. 171 patients that visited the ED of the University Medical Center Groningen (Groningen, the Netherlands) between January 2017 and December 2018 were included. ECG recordings were made from their arrival at the ED and continued after admission to the hospital, up to 48 hours. Patients were included when they were at least 18 years or older (1), were announced with suspected infection (2) and were presenting with at least 2 SIRS criteria (3). Patients were placed in one three groups: not deteriorated or deteriorated. The latter was split into increased disease or constant disease. HRV features were calculated over 5 minute windows, grouped into 3 hour timespans and the average and slope of each HRV feature was calculated. Differences between the three groups were tested for using a Mann-Whitney U test and results were used to select features for a predictive model. A logistic regression model was made using those features as well as vital parameters measured at baseline. The model was optimized using stepwise backwards elimnation and then validation using five-fold cross validation. The AVNN, VLF, LF, Total Power, normalized LF, normalized HF and LF/HF-ratio were significantly different between the disease progression group and the other two groups during the third to sixth hour after arrival at the ED. To prevent overfitting normalized LF, normalized HF and total power were removed before model fitting. Age, heart rate, mean arterial pressure en temperature at the ED were added as parameters for the model. After stepwise backwards elimination three parameters were left: AVNN, age and mean arterial pressure. Validation showed an accuracy of 0.90, sensitivity of 16% and a specificity of 95%. This study proved several HRV features measured early at the ED are associated with patient deterioration during their stay at the hospital. The optimal window and timespan was determined to find the best predictors. However, when using those features in a model to predict deterioration very low sensitivity was obtained. The current model is highly unlikable to contribute to clinical decision making at the ED. Improvement of the model using other vital parameters en more patients to train it is necessary.
Item Type:Essay (Master)
Faculty:TNW: Science and Technology
Subject:44 medicine
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/79932
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page