Challenges in Implementing Clinical Decision Support Systems for the Management of Infectious Diseases

Background

Rapid and easy access to updated information is a sine qua non condition for physicians to make optimal clinical and therapeutic decision. Computer-based clinical decision support systems (CDSS) have been developed for their potential to answer this need (Garg et al., 2005). They are software which present individualized assessments or suggestions to prescribers towards clinical and/or therapeutic decisions, either providing unsolicited information (e.g., pop-up alerts for drug-drug interactions) or providing solicited information (e.g., diagnostic support systems). Early CDSS were based on expert systems, developed from recommendations elaborated by medical experts. For instance, electronical algorithms from patient care guidelines are among commonly used CDSS (Rawson et al., 2017; Durieux et al., 2000). The physician enters the patient’s clinical information and access the diagnostic or therapeutic guidelines adapted to the specific patient’s situation. Expert systems main limitation is their lack of flexibility in dealing with unexpected situations i.e. when experts did not explicitly code all possible situations beforehand. Machine learning (ML) was developed to meet the need for adaptability in decision making. Machine learning CDSS (ML-CDSS) find their own decision rules from massive volume of data. They are increasingly being developed and may replace knowledge-based CDSS (Peiffer-Smadja et al., 2020a).

Daily management of infectious diseases (ID) requires considering and adapting to constantly changing variables. The physician must take into account patient factors (such as demographics and medical history), the site of the infection, and the characteristics of the infecting organisms (e.g., pathogenicity and susceptibility to antimicrobial drugs). Thus, the potential success of a treatment relies on several changing factors such as the ability of an antimicrobial to penetrate the site of infection, the pharmacokinetics and pharmacodynamics (PK/PD) of the drug, and potential adverse events. Integration of large datasets in CDSS can thus be particularly useful in clinician decision-making by providing person-specific and population level data at the point decisions are made.

In ID, a number of CDSS have shown promising results to improve quality and safety of healthcare management (McGregor et al., 2006). However, most CDSS have not been evaluated in real-world clinical settings and are not integrated into clinical practice.