[MA 2022 15] Machine learning–based prediction tools to optimise treatment of bloodstream infections
Amsterdam UMC, Dpt. of Medical Microbiology, Dpt. of Infectious Diseases
Proposed by: Dr. R.P. Schade [r.schade@amsterdamumc.nl]
Title
Machine learning–based prediction tools to optimise treatment of bloodstream infections
Place of the SRP Project
Amsterdam UMC, Dpt. of Medical Microbiology, Dpt. of Infectious Diseases
Background
Bloodstream infection (BSI) is a serious, life-threatening condition, with significant morbidity and mortality. In patients with sepsis, rapid installment of adequate antibiotic therapy is associated with better outcome. The gold standard to diagnose BSI is a positive bloodculture, but time-to-results can take up to 24-48 hours. For this reason empirical therapy is usually started directly after sampling, and therapy is adjusted when results of bloodcultures become available.
In patients with suspected BSI, antibiotic therapy is usually started with broad-spectrum antibiotics, covering both gram-positive and gram-negative bacteria. While broad-spectrum antibiotics have the benefit of providing covering of all possible causative pathogens, the downside is that patients are empirically treated with unnecessary antibiotics.
Previous research in the emergency department has shown that machine learning models can accurately predict (at a given timepoint) whether a blood culture will become positive. Studies conducted in the Amsterdam UMC have shown, both retrospectively and prospectively that these models can be used to reduce the number of unnecessary blood samples.
In the current project, we will extend this knowledge. The projects aims at designing models that can predict the positivity of a bloodculture sample but also the pathogen. In addition, we will investigate whether the model can predict the optimal empirical therapy to use in the patient (i.e. broad-spectrum vs small-spectrum antibiotics).
Research questions
- Can machine learning models be used to identify patients at high risk for bloodstream infections
- Which model has the best performance in this scenario
- Can machine learning models differentiate between gram-positive and gram-negative BSI
- Can we use these models to predict resistance in gram-negative bloodstream infections
Methods
The aim of the SRP is to design and validate a machine learning model that can be used in clinical practice, to identify patients at risk for bacteremia (positive blood cultures).
The model will be targeted at the patients with hospital acquired infections, ie infections that occur in patients that have already been admitted to the hospital for other reasons.
Dataset
For this SRP, data is used that has been stored in the data-warehouse of the department of clinical microbiology and infection control. The following parameters are available:
- Diagnoses (infectious syndromes)
- Clinical data (e.g.: fever, blood pressure)
- Laboratory data (e.g.: leucocyte count, c-reactive protein)
- Microbiological data (e.g.: bloodcultures, surveillance cultures)
- additional data from the hospital electronic health records (EPIC) if necessary
Activities
Design, analyze and validate a machine learning model to optimize detection and treatment of bloodstream infections. This will be done by investigating and analyzing the raw data, and experiment with different models and various parameter settings.
During this project, you will work in a multi-disciplinary team and have the opportunity to make connections at various clinical and supportive departments at both locations of the Amsterdam UMC (ie. Medical Microbiology, Infectious Diseases, Internal Medicine, ICU, Emergency Medicine, A-team, Pharmacy, Business Intelligence, EPIC, etc).
The intended result of the SRP is a scientific article and a presentation for the A-team.
Last but not least, you will get acquainted with the facets of implementing a digital tool in practice!
Time period
7 months
Contact
Interested? Feel free to contact me for questions!
Dr. R.P. Schade, Clinical Microbiologist, Dept Medical Microbiology and Infection Prevention
r.schade@amsterdamumc.nl
References
Goto M et al. (2013) Overall burden of bloodstream infection and nosocomial bloodstream infection in North America and Europe. Clin Microbiol Infect
Curren EJ et al. (2021) Advancing Diagnostic Stewardship for Healthcare-Associated Infections, Antibiotic Resistance, and Sepsis. Clin Infect Dis
Eliakim-Raz N et al.(2015) Predicting bacteraemia in validated models-a systematic review. Clin Microbiol Infect
Boerman AW et al. (2022) Using machine learning to predict blood culture outcomes in the emergency department: a single-centre, retrospective, observational study. BMJ Open
Schinkel M et al. (2022) Diagnostic stewardship for blood cultures in the emergency department: A multicenter validation and prospective evaluation of a machine learning prediction tool. EBioMedicine
van den Broek AK et al. (2021) A mandatory indication-registration tool in hospital electronic medical records enabling systematic evaluation and benchmarking of the quality of antimicrobial use: a feasibility study. Antimicrob Resist Infect Control.