Exploiting electronic health records to improve infection management
Gu Q.
The main goal of effective infection management is to prescribe antimicrobials with an appropriate spectrum to combat the infecting pathogen and with dosing regimens that are optimally adjusted to the patient’s characteristics to ensure efficacy. This first requires identifying the causative organism(s) and their antimicrobial susceptibilities to target therapy. However, this may be delayed due to the time taken to obtain culture results and then susceptibilities, or inconclusive if culture results are negative or only contaminating organisms are identified. Changing demographics, such as an ageing population and rising obesity rates, complicate dosing regimens, which are often developed from studies in healthy volunteers, leading to possible suboptimal outcomes. The widespread adoption of electronic health records (EHR) offers a major opportunity to refine antimicrobial practices. This thesis aims to exploit electronic health records for improving infection management, focusing particularly on how they can be used to improve antimicrobial prescribing and tracking response to infection. I first evaluated the effectiveness of vancomycin prescribing guidelines at Oxford University Hospitals. The results showed that despite good compliance with the new guidelines (70-80%), the proportion of drug levels within the target range remained suboptimal (~30%). Using the real-world pharmacokinetic data, I developed updated dose recommendations to optimise drug levels, taking into account patient age, weight, and renal function. I then explored how routinely collected clinical parameters could guide treatment decisions in patients presenting with presumed bloodstream infections (BSI; based on having a blood culture taken), particularly when blood culture results are pending or negative. I found that how C-reactive protein and vital signs measurements changed over time after blood was taken for culture (“response trajectories”) were associated with specific pathogen groups and infection sources in individuals with suspected BSI. Distinct patterns of clinical response trajectories were identified: early peaks (day 1 or 2) and typical recovery, slow recovery, a delayed peak (day 6), or persistently low levels. Centile reference charts were created based on the subgroups with “normal” responses to antibiotics as determined by latent class modelling to standardise the assessment of infection progression and treatment response in patients with suspected BSI; these could be used to guide management independently of microbiological test results. Finally, I examined the current clinical antibiotic prescribing patterns for suspected BSI and their association with the dynamics of these clinical parameters. Overall, the thesis demonstrates the potential of EHR as a pivotal tool in enhancing the quality of antimicrobial management in clinical settings. By integrating data from EHR with patient-specific characteristics and real-time clinical responses, more personalised treatment recommendations can be developed to improve outcomes in patients with varying demographics and health profiles.