Identifying the need for infection-related consultations in intensive care patients using machine learning models

Abstract

Infection-related consultations on intensive care units (ICU) have a positive impact on quality of care and clinical outcome. However, timing of these consultations is essential and to date they are typically event-triggered and reactive. Here, we investigate a proactive approach to identify patients in need for infection-related consultations by machine learning models using routine electronic health records. Data was retrieved from a mixed ICU at a large academic tertiary care hospital including 9684 admissions. Infection-related consultations were predicted using logistic regression, random forest, gradient boosting machines, and long short-term memory neural networks (LSTM). Overall, 7.8% of admitted patients received an infection-related consultation. Time-sensitive modelling approaches performed better than static approaches. Using LSTM resulted in the prediction of infection-related consultations in the next clinical shift (up to eight hours in advance) with an area under the receiver operating curve (AUROC) of 0.921 and an area under the precision recall curve (AUPRC) of 0.541. The successful prediction of infection-related consultations for ICU patients was done without the use of classical triggers, such as (interim) microbiology reports. Predicting this key event can potentially streamline ICU and consultant workflows and improve care as well as outcome for critically ill patients with (suspected) infections.

Figure 1

Study design and data processing for three different data sources (hospital database, ICU database, and medical microbiology database). Note: (interim) microbiology reports were not included in the modelling process. Standard data cleaning processes are not shown (cf. main text in “Methods”).

Figure 2

Proportion of admissions with an infection-related consultation per quarter. No significant change in trend line (dashed) using a linear regression model.

Figure 3

Proportion of overall admissions and consultations stratified per weekday (weekdays duplicated to display two weeks for easier visual perception). The dashed line shows the proportion of patients receiving a consultation among all admissions, stratified per weekday.

Figure 4

Model performance on the held-out test set by area under the receiver operating curve (AUROC) for each model predicting a consultation at the ICU. LSTM with a time frame of 80 h showed the highest AUROC of 0.921.

Figure 5

Model performance on the held-out test set by area under the precision recall curve (AUPRC) for each model predicting a consultation at the ICU. The baseline represents the performance of a random classifier and reflects the occurrence of consultations in the study cohort (7.8% of all admissions). LSTM with a time frame of 80 h showed the highest AUPRC of 0.541.