Introduction

The increasing resistance of microorganisms to antimicrobials poses a significant global public health threat, with profound implications for morbidity, mortality, and healthcare costs.1 It is estimated that up to 70% of pathogens causing healthcare-associated infections exhibit resistance to one or more antimicrobial agents.2 Observational studies and mathematical models suggest that reducing antimicrobial consumption, both at populational and individual levels, is a crucial measure for controlling the emergence of multidrug-resistant microorganisms.3 This can be achieved through more judicious prescribing practices and shorter durations of antibiotic therapy.4 5

Numerous clinical trials and systematic reviews conducted over the past two decades have demonstrated the efficacy and safety of shorter antibiotic treatments compared with previous standard practices.6 7 These studies have encompassed various clinical settings, including ambulatory care, emergency departments and critically ill patients, with infections acquired in community or healthcare settings, involving different infection sites such as respiratory, abdominal, urinary tract, bloodstream, and others.8–13

In this context, using biomarkers to guide the duration of antimicrobial therapy has emerged as a promising approach.14 15 Among the biomarkers currently employed in clinical practice, inflammatory biomarkers, particularly procalcitonin (PCT) and C-reactive protein (CRP), are the most widely used in this scenario.15 16 Algorithms using PCT to guide antibiotic therapy have demonstrated reductions in antimicrobial consumption without significant differences in safety outcomes between control and intervention groups, particularly in intensive care patients.17–24 However, the high cost and limited availability of PCT raise concerns regarding the cost-effectiveness and applicability of these algorithms, particularly in low- and middle-income countries.

On the other hand, CRP has been investigated as a lower cost and more accessible alternative for guiding antibiotic therapy, despite its lower specificity in diagnosing infections (40–67%).16 Its role as a prognostic marker has been well-established in various studies,25–30 and recent research has demonstrated its potential utility in guiding antibiotic therapy, either in hospitalised patients31 or in an outpatient-setting.32 33 Results of a randomised clinical trial conducted at two teaching hospitals by Oliveira and colleagues suggest that CRP is as valuable as PCT in reducing antimicrobial consumption without apparent harm in a population of critically ill patients.34 In another study, CRP-based therapy significantly reduced the duration of antibiotic therapy in septic patients, compared with patients treated based on best current practices.35 A recent multicentre study performed by Dach and colleagues demonstrated the non-inferiority of a CRP-guided algorithm compared with fixed regimens of 7 and 14 days of treatment for uncomplicated bacteraemia caused by gram-negative bacilli.36

Despite these promising findings, further studies are warranted to assess the real-world impact of CRP-guided antibiotic treatment strategies in reducing antibiotic consumption, the duration of therapy, cost-effectiveness and safety. It is also essential to consider the development of algorithms that incorporate use of clinical variables in the decision-making process for antibiotic therapy, in addition to applying biomarker-based criteria. Furthermore, decision algorithms involving biomarkers proposed in recently published studies are still far from daily hospital medical practice. In order to enhance healthcare teams’ adherence to these algorithms, developing digital tools that facilitate widespread and reproducible usage would be highly beneficial.37 38

This study aims to evaluate the effectiveness and safety of a novel algorithm implemented through a mobile-accessible digital Clinical Decision Support System (CDSS) developed specifically for this project (see online supplemental figure 1). The algorithm combines serum levels of CRP with relevant clinical variables from individual patients to guide antibiotic therapy for adults in a hospital setting. The primary objective will be to assess the algorithm’s efficacy in reducing the duration of treatment while ensuring patient safety by monitoring adverse events, in comparison to antimicrobial therapy based on the best available evidence.

Methods and analysisTrial design

This study is designed as a prospective, single-centre, open-label, randomised controlled trial with parallel group assignment. The participants will be individuals with either a clinical suspicion or microbiological confirmation of bacterial infection, all of whom must have started antibiotic therapy within the past 72 hours. The primary outcome measure will be the duration of antibiotic treatment for the initial infectious episode. This trial aims to demonstrate superiority (figure 1).

Figure 1

Study protocol. CRP, C-reactive protein; ICF, informed consent Form.

Participants

The participants will be selected from patients admitted to the emergency department and the internal medicine ward of Hospital das Clínicas of the Universidade Federal de Minas Gerais (HC-UFMG), a tertiary university hospital located in Belo Horizonte, Minas Gerais, Brazil. Currently, these two departments have approximately 70–80 beds. The patient profile in the unit is characterised by its high complexity, with the most prevalent conditions including chronic cardiovascular diseases, chronic pulmonary diseases, advanced chronic liver disease, rheumatological conditions, solid and haematological neoplasms, acute infectious diseases and post-operative patients.

Eligible participants will include individuals over 18 years of age who have a clinical suspicion or microbiological confirmation of bacterial infection, with initiation of antibiotic therapy within the last 72 hours as determined by the attending physician team. The inclusion criteria will encompass patients with clinical suspicion of bacterial infection, recognising that a significant proportion of hospital-acquired infections lack microbiological confirmation.39 All participants will be presented with an informed consent form (ICF) in native language, and if the patient is unable to sign, their legal representative or accompanying person will be asked to sign on their behalf (see online supplemental informed consent form).

Exclusion criteria include: seriously immunocompromised patients (eg, patients infected with HIV with a CD4 count <200 cells/mm3; neutropenic patients with a neutrophil count <500 cells/mm3; solid organ or bone marrow transplant recipients; patients who received chemotherapy within the last 14 days and have a high risk of febrile neutropenia, as defined by the attending oncology team; patients using any kind of immunosuppressant medication; patients who have been on corticosteroids at a dose greater than 0.5 mg/kg of prednisone or equivalent for the past 30 days or have received pulse therapy with these drugs within the last 14 days and patients with primary immunodeficiency; patients diagnosed with acute liver failure or other conditions that clearly impair humoral, cellular or mixed immune defenses); conditions requiring prolonged antibiotic therapy identified prior to randomisation (eg, infective endocarditis, necrotising pneumonia, deep abscesses, osteomyelitis, complicated soft tissue infections, Staphylococcus aureus bacteraemia and others); patients that are expected to be discharged from the hospital within 72 hours of inclusion; patients in end-of-life care and patients with a life expectancy of less than 24 hours.

Randomization

Eligible patients will be randomly assigned to either the intervention or control group using block randomisation with a block size of four and a 1:1 allocation ratio. The randomisation sequence will be generated by the digital tool developed by the research team. This process will occur when the research team or the attending physician inputs the patient’s data into the mobile application, which will automatically assign the patient to a group and provide the corresponding recommendations. Randomisation will take place within 72 hours of initiating antibiotic therapy. Due to the nature of the intervention, both the principal investigator and the attending physician will be aware of each participant’s group assignment (open-label design).

Digital clinical decision support system

Digital technology usage emerges as an auspicious tool for the healthcare sector, enabling the optimisation of tasks that traditional systems were incapable of and expanding accessibility to various services. According to the WHO, digital health interventions may be classified according to their primary objective.40 Thus, there are systems designed to improve clients’ experience in healthcare, to facilitate logistics and management of systems, to optimise data storage and aggregation and, finally, to those aimed at healthcare providers, among which our digital application fits as a CDSS.

The digital CDSS was developed by the research team in collaboration with the Centro de Telessaúde at the Hospital das Clínicas, Federal University of Minas Gerais (CTS-UFMG). It will be available as a mobile application for download by the attending physicians.

The backend was implemented in Java 21 using the Spring Boot framework to create RESTful Application Programming Interfaces for communication between the frontend and backend. The frontend was built using the Angular UI framework V.18. The application uses PostgreSQL, a relational database, as its database. Push notifications were implemented using the Apple Push Notification Service and Firebase Cloud Messaging platforms. These technologies enable real-time message delivery to iOS and Android devices.

The app’s functionality is based on an algorithm created by the researchers, which uses clinical variables and CRP levels to suggest the appropriate duration of antibiotic treatment (figure 2). The app also contains the recommendations for the control group (figure 3). Randomisation between groups will be performed through a sequence generated by the app. Thus, all treatment recommendations will be provided through this tool, with positive reinforcement from the research team as needed (see online supplemental figure 1).

Figure 2

Intervention group algorithm. Peak CRP value was defined as the highest value within the first 72 hours of antibiotic therapy. Septic shock, defined as persisting hypotension requiring vasopressors to maintain mean arterial pressure ≥65 mm Hg and having a serum lactate level >2 mmol/L (18 mg/dL) despite adequate volume resuscitation.41 CRP, C-reactive protein; SOFA score, Sequential Organ Failure Assessment score.

Figure 3

Recommendations for control group. Recommended time may vary according to the antimicrobial agent. Patients should show no signs of a persistent infectious focus.

Study groupsIntervention group

For patients in the intervention group, attending physicians will be encouraged to follow the algorithm integrated into the digital CDSS, developed by the research team. This algorithm uses clinical variables and serum CRP levels to guide the duration of antibiotic therapy. The peak CRP is defined as the highest value recorded within the first 72 hours of treatment (figure 2).

Antibiotic therapy discontinuation will be encouraged under the following conditions:

If the peak CRP is below 100 mg/L: consider stopping antibiotics when CRP falls below 35 mg/L, with a minimum treatment duration of 3 days.

If the peak CRP is above 100 mg/L or the patient meets criteria for sepsis or septic shock: consider stopping antibiotics when CRP has decreased by 50%, after a minimum of 5 days.

If the patient does not meet the CRP criteria: antibiotic discontinuation will be recommended after 5–7 days, provided there is clinical improvement.

Before discontinuing antibiotic therapy, physicians should confirm that the patient is clinically improving, with no signs of a persistent infectious focus. Additionally, they will be encouraged to verify that the Sequential Organ Failure Assessment (SOFA) score is stable or decreasing. These factors will assist in determining the appropriateness of stopping antibiotics.

The intervention will be advisory rather than prescriptive. While the research team will provide recommendations, the choice of antimicrobial regimen and the final decision on treatment duration will remain the responsibility of the attending medical team.

Control group

For patients in the control group, the attending physician will be encouraged to determine the duration of antimicrobial therapy based on the best available evidence, considering the most likely infectious focus and the patient’s clinical response. These recommendations will be guided by international society guidelines and established best practices for antibiotic therapy (figure 3). For infectious conditions not covered in figure 3, the research team will collaborate with the attending team to discuss appropriate treatment on a case-by-case basis.

Additionally, it is recommended that CRP monitoring in the control group be discontinued after 72 hours of antibiotic therapy, as this period is considered sufficient for using CRP as a biomarker to assist in diagnosing the infectious condition.

As with the intervention group, the attending medical team retains full responsibility for all decisions regarding antimicrobial therapy, regardless of the recommendations provided by the research team.

OutcomesPrimary outcome

Duration of antibiotic therapy for the index infectious episode (the one that prompted inclusion in the study), measured in days.

Secondary outcomes

Efficacy outcomes

Total antimicrobial exposure, defined as the number of days of antibiotic exposure considering all therapeutic cycles during patient follow-up, measured in antimicrobial days/days of follow-up × 1000. Antimicrobial days=total number of days antibiotics were given. If a patient is on two antibiotics, both are counted (ie, the number of days for both drugs are added together). Days of follow-up=sum of days of follow-up in the study.

Antibiotic-free days, defined as the ratio of the number of days without antibiotic use to the total number of follow-up days, adjusted by a denominator of 100 using a simple proportion.

Length of hospital stay in days.

Estimation of the cost of antimicrobial therapy in each group of the study.

Adherence rate to the proposed algorithm in both the intervention and control groups. The research team will closely monitor the daily progress of patients in each group and will actively encourage adherence to the protocol associated with the respective group. For the intervention group, this may involve explaining the benefits of following the new algorithm, while for the control group, it may focus on adherence to the standard care. Any deviations from the algorithm will be noted for analysis.

Safety outcomes:

Hospital mortality from all causes.

Therapeutic failure, defined as the persistence or recurrence of signs and symptoms from the same infectious focus leading to resumption of antibiotic therapy within 48 hours after discontinuation.

Reinfection rate, defined as a new episode of infection from a distinct focus or isolation of new microorganisms and occurring more than 48 hours after discontinuing targeted antibiotic therapy for the initial infectious condition.

Subsequent infections caused by multidrug-resistant microorganisms, measured by the ratio of culture isolations per 100 patients.

Rate of Clostridioides difficile infection.

Follow-up

Patients will be followed up until hospital discharge, death, or 90 days, whichever occurs first.

Safety issues

Patients with non-fermenting gram-negative bacilli identified in cultures from any specimen should receive at least 7 full days of antibiotic therapy.

Patients with positive blood cultures should receive a minimum of 5 full days of antibiotic therapy.

Statistical methods

The sample size calculation was based on the study’s primary outcome and data on the difference in means in antibiotic therapy duration from previous research by the team, comparing CRP-guided and non-CRP-guided algorithms (figure 4). The analysis for comparing two independent means using a t-test was conducted with G*Power 3.1.9.7 software. A significance level of 5% was applied. The initial required sample size was determined to be 200 patients. Any losses to follow-up will be compensated by including replacement participants.

Figure 4

Data used for sample size calculation. CRP, C-reactive protein.

To achieve adequate participant enrolment and meet the target sample size for this trial, we will implement effective communication strategies. We will provide transparent and comprehensive information about the trial to potential participants, enabling them to make informed decisions. This will include details about the trial’s purpose, procedures, potential risks and benefits, ensuring that participants fully understand what their involvement entails.

Categorical variables will be presented as percentages, while continuous variables will be expressed as measures of central tendency. The primary outcome and secondary outcomes will be analysed based on the intention-to-treat principle. Associations between variables will be assessed using the χ2 or Fisher’s exact test, as appropriate. Comparisons of means or medians will be conducted using Student’s t-test or the Mann-Whitney U test, following the Shapiro-Wilk test for normality.

We intend to adjust the analysis using patients’ severity scores: SOFA score41 and the Charlson Comorbidity Index42; age and infection classification (infection vs sepsis/septic shock); confirmed versus unconfirmed microbiology; and adherence versus non-adherence to the study protocol.

The primary outcome results will be reported as the median difference between the two treatment groups. Other results from secondary outcomes will be presented both unadjusted and with multivariable analysis, using relative risks and their corresponding 95% confidence intervals.

Prespecified subgroup analyses for the primary outcome will be performed based on age, the origin of infection (community-acquired or nosocomial), SOFA score, comorbidities assessed by the Charlson Comorbidity Index, infection classification (infection vs sepsis/septic shock), infectious focus, and the appropriateness of the initial empirical therapy.

The main analyses will be performed on an intention-to-treat basis. We will further perform a per protocol analysis restricted to the group of patients in whom the protocol was strictly followed. In the end, protocol over-ruling will be considered for all participants in whom the duration of antibiotic therapy was longer or shorter than initially specified by the study protocol, including both the intervention and control groups.

A survival analysis to compare the duration of antibiotic therapy between the two study groups will be performed using a Cox regression model, adjusted for disease severity, infection focus and underlying health conditions.

The collected data will be used to build a database for future research.

Ethics and dissemination

Only patients who agree to participate in the study after reading the ICF, provided in native language, will be included. If the patient cannot sign, their legal representative or accompanying person must sign on their behalf. The ICF provides detailed information about the study, its purpose, procedures, potential risks and benefits, confidentiality and the voluntary nature of participation (see online supplemental informed consent form).

This project was submitted for consideration to the Research Ethics Committee of the Federal University of Minas Gerais (COEP-UFMG) and received approval (approval Number: 5.905.290).

This trial was registered on ClinicalTrials.gov (NCT05841875). The trial protocol and additional details can be accessed at https://clinicaltrials.gov/ct2/show/NCT05841875.

The project will generate clinical and laboratory data from approximately 200 patients, extracted from electronic medical records via the Aplicativo de Gestão para Hospitais Universitários (AGHU, 2009) and laboratory results from the MatrixNet system (MATRIXSAUDE, 2020). In addition, three to five serum aliquots of 0.5 mL each will be collected from each patient after informed consent is obtained. These aliquots will be stored for potential future analyses by the research team with three to serum aliquots (0.5 mL each) per patient.

Clinical and laboratory data will be securely stored using the Research Electronic Data Capture platform (Vanderbilt University, Nashville, USA, 2004). The serum samples will be preserved in a −80°C freezer in a biorepository that complies with the Institutional Biorepository Regulations. The biorepository will be housed at the School of Medicine, Federal University of Minas Gerais (UFMG), located at Av. Prof. Alfredo Balena, 190, Belo Horizonte, MG, CEP 30 130–100. These data and samples will be preserved for potential use in future research projects by the group.

Access to the scientific data will be limited to the research team via a permission-based system that requires login credentials. The serum samples stored in the biorepository will be similarly restricted, with access controlled in compliance with the Institutional Biorepository Regulations of the hosting institution. All data will be de-identified prior to sharing or distribution. A certificate of confidentiality will be included as part of the informed consent process to further safeguard participants’ privacy and rights (see online supplemental data management and sharing plan).