Observational multi-centre, prospective study to characterize novel pathogen-and host-related factors in hospitalized patients with lower respiratory tract infections and/or sepsis - the “TAILORED-Treatment” study

Antibiotics are the most prescribed class of drugs worldwide [1]. However, 30–50% of antibiotics are estimated to be prescribed inappropriately, making antibiotics also the most misused drug class [1‐5]. Antibiotic overuse, i.e. prescribing antibiotics to treat a non-bacterial disease, is a serious problem. For example, in the US over 80 million antibiotic prescriptions are given annually for viral infections in the outpatient setting [6]. This may cause the emergence of adverse events (AEs) such as allergic reactions and antibiotic-associated diarrhoea [7]. Importantly, antibiotic overuse has been linked to the emergence of resistant strains of bacteria, by the Center for Disease Control considered as “one of the world’s most pressing health problems in the 21^st century” [8, 9]. Annually, 23,000 deaths due to resistant bacteria are expected in the US alone [8, 10]. Resistant bacteria are projected to cause over 10 M annual deaths worldwide by 2050, surpassing cancer as the leading cause of death [8, 11]. A second type of antibiotic misuse is underuse, i.e. delayed or no antibiotic treatment in case of a bacterial disease, from which a patient could have benefited [12‐17]. Although reducing the risk of antibiotic-related AEs [18, 19], underuse of antibiotics may lead to a prolonged disease duration and an increased rate of complications that could have been avoided with early antimicrobial treatment [19‐21]. For example, up to 15–40% of adult patients hospitalized for bacterial pneumonia in the US receive delayed or no antibiotic treatment [14, 15]. Finally, applying the wrong antibiotic spectrum to treat a bacterial disease is the third type of antibiotic misuse. For example, administering third generation cephalosporins instead of macrolides to treat a respiratory infection caused by atypical bacteria.

Antibiotic overuse has also several health economics consequences. For instance, the annual cost of unnecessary antibiotic prescriptions for adult respiratory infections in the US is estimated to be $1.1 billion [22]. In addition, there are also indirect costs, including the costs of treating preventable antibiotic-related AEs and the costs of prolonged hospital stay as a result of AEs. Lastly, are the costs related to the emergence of antibiotic-resistant bacterial strains. The cost of treating patients with antibiotic-resistant bacteria is estimated at $16–26 billion annually in the US [23‐25] and over €1.5 billion in the EU [26]. The health economics consequences of underuse of antibiotics include the costs of treating preventable complications and a prolonged disease duration resulting from a delayed or no antibiotic treatment [19‐21].

Respiratory tract infections are one of the most common causes of hospital and outpatient visits in the EU, comprising 1 out of 3 admissions annually [9]. In patients with mild respiratory tract infections, over prescription of antibiotics in general and prescription of broad spectrum antibiotics are large problems. Broad spectrum antibiotics account for approximately half of all antibiotic prescribing for children with acute RTI [27], with the above mentioned consequences [8]. Sepsis-related hospital admissions are less frequent but represent a significant burden to healthcare systems in terms of adverse patient outcomes and the need for rapid, but effective, intervention strategies [28]. In septic patients the key clinical challenge is to ensure that the correct antimicrobial treatment is administered, and that this therapy is administered as soon as possible [29].

Although the current diagnostic tools for facilitating appropriate use of antibiotics (such as culture-, PCR-, and immunoassay-based) may be valuable in some clinical situations, they have major limitations (Fig. 1). First, existing diagnostic tools often require hours to days to provide information, whereas physicians need to decide whether to prescribe antibiotics within shorter time periods. Second, available diagnostic technologies usually require direct sampling of the pathogen. Such sampling is often not feasible if the infection site is inaccessible (e.g., pneumonia, sinusitis, middle-ear infection, etc.). A third limitation is that available technologies usually search for the presence of specific bacteria or viruses. However, many types of bacteria and viruses could be present as part of the natural flora without causing an infection. For example, the important respiratory tract pathogenic bacteria S. pneumoniae is also part of the upper respiratory tract natural flora in up to 68% and 15% of healthy children and adults, respectively [30] and respiratory viruses were detected in 35.4% of a-symptomatic children [31]. Finally, diagnostic solutions that were developed for detection of a specific pathogenic strain, often fail to detect the constantly evolving and emerging strains of the same family of pathogens, for example rapid Influenza tests showed low sensitivity, 45%, during the 2009 H1N1 influenza A viral outbreak [32]. These limitations lead to a ‘diagnostic gap’, which in turn often leads physicians to either over-prescribe (e.g., ‘just-in-case’ approach) or under-prescribe (e.g., ‘watchful waiting’ approach) antibiotics [18‐20], both of which adversely influence patient care and health economics.

The health and health economics consequences of antibiotic misuse highlight the need for a diagnostic tool that would help physicians to use antibiotics appropriately. Ideally, this diagnostic tool should provide a diagnosis that can accurately and rapidly differentiate between bacterial and viral infections. However, as previously described, current diagnostic tools are often inadequate (Fig. 1). Therefore, the TAILORED-Treatment consortium was established to develop new tools aimed at increasing the effectiveness of antibiotic therapy, reduce adverse events, and limit the emergence of antimicrobial resistance in children and adults. In reality, targeted antimicrobial therapy can most effectively be achieved by utilizing personalized data to facilitate a tailored and optimized approach to individual patient treatment. This can best be achieved by utilizing knowledge gained from both host-centred and pathogen-centred parameters during health and disease. Unfortunately, these parameters have traditionally, tended to be measured independently, and used on an ad hoc basis without careful integration for the best treatment of the patient. However, recent advances in the development of high-throughput and sensitive molecular-based technologies, improved databases, and bioinformatics analysis tools, rapidly enabling that the goal of personalized medicine and treatment can be reached. Unfortunately, however, there currently exists a ‘technological gap’ between recent state-of-the-art methodologies (for example with respect to gaining new insights into novel host-pathogen interactions) and laboratory-to-bedside results to benefit patients, physicians and society as a whole [33]. The observational TAILORED-Treatment project is designed to bridge this technological gap in order to generate novel insights and innovations that are readily exploitable in the field of personalized medicine and infectious diseases.

In this prospective, multi-centre, observational study, an expected total number of 1200 paediatric and adult patients aged 1 month and older with suspected LRTI and/or sepsis or a non-infectious disease are recruited, in order to develop a diagnostic model to differentiate between bacterial and viral infections. The study will generate a model for patients with LRTI and a separate model for patients with sepsis. The prediction of the model obtained from each patient will be compared with a consensus diagnosis by a panel of three experts in order to determine its accuracy (Fig. 2). The study is a non-interventional study, and therefore, the diagnostic prediction of the model will not be used for any clinical or diagnostics decision making. Local data monitor plans were used and executed by local, independent monitors.

Patients aged 1 month and older that attend the Emergency Department (ED) and hospital wards of seven Dutch and Israel medical centres (secondary and tertiary) due to a suspected LRTI and/or sepsis that agree (or their legal guardians agrees) to sign an informed consent and meet the inclusion criteria and none of the exclusion criteria (Table 1), are eligible to participate in the study. Informed consent is asked by the study team (i.e. research nurses or medical doctor). Sepsis criteria are defined according to published criteria [34]. Patients receive care as usual. The non-infectious disease group include patients with clinical signs of a non-infectious disease.

Currently, no single reference standard exists for differentiating between bacterial and viral aetiology in an acute infectious disease. Therefore, we follow the Standards for Reporting of Diagnostic Accuracy (STARD) recommendation [39], and create a highly rigorous reference standard based on expert panel adjudication (Fig. 2). The eCRF of each patient is available to a panel of three independent physicians that are affiliated to the country of recruitment. Based on the information included in the eCRF, each member of the panel assigns one of the following diagnostic labels to each one of the patients: (i) bacterial infection; (ii) viral infection; (iii) mixed infection (i.e., bacterial and viral co-infection); (iv) non-infectious disease; or (v) undetermined. Of note, each expert is blinded to the research results and to the labels of his peers on the expert panel. For this purpose, mixed infection (bacterial and viral co-infection) is considered as bacterial infections, as they often elicit similar patient management. A final diagnosis is determined based on consensus agreement between all three experts (Fig. 2). The final diagnosis is used for the purpose of calculating the accuracy of the newly developed diagnostic tools. Cases with only majority agreement (2 out of 3 experts) are analysed as a separate sub-group. The remaining cases are labelled as undetermined. Specifically, this label is assigned in cases where each of the experts assigns a different diagnosis or in cases where 2 or more of the experts assign an “undetermined” diagnosis. These patients are used for data analysis and are expected to comprise roughly 10% of all patients.

The study is conducted in two consecutive stages. Stage A is used to collect molecular, biochemical, microbiological and clinical data on a group of approximately 900 patients. Collected data is used to develop unique diagnostic signatures able to distinguish between bacterial and viral aetiologies, a mass spectrometry based rapid detection technique for identification of microbial pathogens and antimicrobial resistances and a personalized treatment algorithm for tailored antimicrobial therapy that integrates clinical, molecular and biochemical data. In stage B, a second, independent cohort of approximately 300 patients is recruited for the purpose of testing and validating the performances of the multi-parametric model.

The following power analysis aims to achieve the primary study objective of developing a new diagnostic model for classifying patients with viral and bacterial aetiologist and treatment algorithms to guide antimicrobial prescribing.

The multi-parametric model integrates various host-related and pathogen-related parameters into a single score that determines the likelihood of bacterial and viral aetiology (Fig. 3). The score is computed using a multinomial logistic regression formula that is developed during Stage A of the study. Using predetermined cut-offs, each of the patients is classified as bacterial, viral, or inconclusive (i.e., a marginal immune response that is neither clearly bacterial nor viral). The model diagnosis will be compared with the consensus diagnosis in order to determine its accuracy. The diagnostic accuracy is quantified by computing the following measures: sensitivity, specificity, positive predictive value, negative predictive value, total accuracy, positive likelihood ratio, negative likelihood ratio and diagnostic odds ratio. The area under the receiver operation curve is computed to perform cut-off independent comparisons of different diagnostic methods.