The decision on which antimicrobial to prescribe is often difficult for clinicians, whether empirically or using laboratory data. This project proposed the development of an algorithm-based app to assist the physician's decision in the face of a bacterial infection, whether moderate or severe, in-hospital or community-based, thus mitigating the selective pressure resulting from incorrect antimicrobial prescription. SMART-EP is a smart system that uses an algorithm based on artificial intelligence to provide a multiparametric analysis and integration of patient data, microorganisms, and drugs. According to the clinical case, the system indicates the various first-line antibiotics, thus supporting the clinical decision for the best antimicrobial treatment option. The system will allow adult physicians, pediatricians, and infectious disease experts to prescribe the best antibiotics while improving antimicrobial stewardship by making antibiotic prescription safer and more accurate, reducing antimicrobial resistance and treatment cost in these infections. The algorithm uses a system of weights and parameters such as patient´s age and sex, place of residence, and type of infection (community versus in-hospital). The system calculates renal function, risk of sepsis, and history of drug allergies, but mainly compiles the results of antibiotic sensitivity in patients similar to the index case, based on a national database.
How was the experiment
The first step involved building an artificial intelligence system using decision flow, developed by IT specialists with the support of a multidisciplinary team of infectious disease specialists (physicians, pharmacists, and microbiologists). This step assessed variables potentially impacting antimicrobial treatment outcomes such as age, weight, renal function, immune system status, infection site, the drug´s pharmacokinetic and pharmacodynamic properties, pharmacology, minimum inhibitory concentration, microorganism involved, drug association, drug cost, local epidemiology, and antimicrobial resistance in recent isolates. Then, the SMART-CDSS treatment recommendation feature was validated. In a double-blind study, the suggested treatments were tested and validated using hypothetical clinical cases of community-dwelling and hospitalized patients, where the team of experts defined the first three antimicrobial prescription choices. After making the adjustments and receiving feedback from the specialists, better performance was ensured. A smartphone app was developed to automatically rank the antimicrobials for each clinical case by informing the patient's clinical data and the infectious agent.
We performed a survey of the drugs used in Brazilian hospitals, including the necessary adjustments in cases of impaired renal function or risk of sepsis, besides weight, age, cost, and pharmacokinetic and pharmacodynamic characteristics. In parallel, we used the database of a National Antimicrobial Resistance Surveillance System (BR-GLASS) to observe the microorganisms most commonly found in infections, particularly assessing antimicrobial resistance and infection site. This search compiled local and national epidemiological data on bacterial infections and quantified the best antimicrobial therapy options according to the sensitivity profile. This dataset served to build the machine learning algorithm. Validation was performed by an expert panel and confronted with three possible options suggested by the app for each clinical case developed by the team. App usability tests were also conducted with software designers, developers, and potential users. So far, the app has proven to be easy to use with quick results (less than two minutes per patient for data entry and results).
Why is it innovative
The system's innovation is based on correlating numerous variables on different bacterial infections, collecting data from the hospital's laboratory information system and providing physicians with a ranking of the best antibiotics for use in each case to support the clinical decision on the best treatment option.
Enlargement of the BR-GLASS database (currently containing data from five hospitals but expected to encompass 20 hospitals by 2023) and improvements in automated data feeding will enable the app to update data rapidly and dynamically, otherwise impossible for clinicians. Empirical antimicrobial prescription will thus no longer be based on isolated observations of a single prescribing professional or data from the literature but on evidence from a robust nationwide database that is updated in real time.
Problem that solves
Antimicrobial resistance has spread worldwide in recent years, and it is estimated that by 2050 there may be 10 million deaths in the world, at a cost exceeding US$ 100 trillion. In 2019, studies pointed to 4.5 million deaths related to AMR, and these numbers must have risen even more with the indiscriminate use of antimicrobials during the COVID-19 pandemic. Deciding which antimicrobial to prescribe is often difficult for clinicians, either empirically or using laboratory data. This project proposes the development of an app that uses an algorithm to assist the clinician's decision in the face of a bacterial infection, whether moderate or severe, in hospital or in the community, thus mitigating selective pressure from antimicrobial misprescription.
Implications for the brazilian health system
The system predicts the best antimicrobial for use in each clinical case. Although the tool is still not available in Brazil, it could mitigate AMR caused by misprescription and indiscriminate use of antibiotics while reducing costs for the national public health system (SUS). This kind of support is paramount in remote units or small communities that lack medical specialists or laboratories capable of accurately identifying the infectious agent and its degree of sensitivity to antimicrobials.
The project´s second phase requires the inclusion of more infection sites, adding sterile fluids (e.g., CSF), endocarditis, intraabdominal infection, and respiratory infections. It also requires other improvements such as the system´s clinical validation in a large hospital and integration with hospital information systems to optimize automated completion of patient and microbiological data, enabling the recording of consultations on the app for future review and critical analysis. Meanwhile, the expansion of participating hospitals in the BR-GLASS System will ensure a robust and reliable database.
The next steps will be funded partially by a comprehensive AMR project approved through an international call for proposals by the U.S. Centers for Disease Control and Prevention (CDC), called Strengthening the Brazilian Antimicrobial Resistance Surveillance System, under grant number CK21-2104. The project´s various activities feature expansion of the BR-GLASS System to 55 hospitals from all states of Brazil and updating and maintenance of the SMART-CDSS App activities. The project is led by CGLAB/DAEVS/SVSA/MS and LAPIH/FIOCRUZ, in partnership with LACEN/PR, and is scheduled to last five years (2022-2026).
- Pillonetto M, Kalil A, Becker G, Giamberardino AL, Teixeira B, Bergamo R, Madeira H, Dias V, Miorando, Giglio R. (2020). SMART-CDSS An artificial intelligence system for antimicrobial prescription support. Gates Open Res, 4. https://doi.org/10.21955/gatesopenres.1116684.1) - 10/2023
- Pillonetto M, et al. (2021). An Artificial Intelligence System for Antimicrobial Prescription Support. In: 31 European Congress On Clinical Microbiology and Infectious Diseases - 09/2021