THE TECHNOLOGY

INFECTION INFORMATION
– Use any available infection-relevant information about the patient

The family of TREAT products takes every available piece of infection-relevant information into account when giving advice and recommendation on the clinical workup. The stochastic models use this information for risk assessment (estimating the probability of bacteremia and/or mortality), diagnosis, and for determining the likely causative pathogens.

For TREAT Steward™, the infection-relevant information is not limited to the test results and examination of the individual patient, it also takes into account physiological factors influencing antimicrobial coverage of pathogens at different infection sites, i.e.:

• In vitro coverage for antimicrobials estimated for the local hospital based on an isolate database
• Modification of coverage to account for static in vivo effects, e.g. decreased penetration due to the blood-brain barrier
• Bioavailability of the antimicrobial agent given, e.g. oral absorption through the gastrointestinal tract

Other important physiological and biological mechanisms accounted for include renal toxicity, allergies, risk-factors, local prevalence of individual pathogens and the effect of previous antimicrobial treatment. TREAT Steward™ takes all of this information, and more, into account when making its probability calculations. The influence of all parameters included in the TREAT Steward™ reflects a tremendous system which has taken more than 70 man-years to build.

INTEGRATION
– "Grab what you can” principle

The family of TREAT products integrate seamlessly into a range of existing hospital IT systems, providing an elegant solution to display all infection-relevant information for clinicians, specialists, laboratory technicians and management alike. Integration is typically facilitated through web services or HL7 standards. In current installations, the data model retrieves and understands information concerning:

• Patient demographics; name, age, gender, department and admission dates
• Vital signs obtained by the nurses; temperature, heart rate, respiratory rate and blood pressure
• Patient background and infection risk factors; catheters, functional capacity and pacemakers
• Clinical chemistry laboratory test results; hematology, biochemistry values and blood gases
• Test results of microbiology samples; PCR, microscopy, blood- and local cultures
• Previous and current medicine and allergies

Furthermore, TREAT products

• Integrate into the existing workflow and can be activated from within the hospital’s electronic health record
• Provide alerts with early warning scores and monitoring of new septic patients
• Use single sign-on and integrates into hospital user password safety procedures
• Export infection notes and summaries to the electronic health record system

WEB TECHNOLOGY
– Access TREAT from anywhere at the hospital or laboratory facility

The family of TREAT products uses standard web technology, avoiding the need to install workstations or programs specifically for using the products. Any device with an internet browser that is connected to the hospital network can be used. The solutions can be fully built into existing electronic health record systems, making single sign-on possible – the user will not even recognize that they are using an external third party software solution.

COST-BENEFIT
- TREAT Steward™ performs a cost-benefit analysis for each patient

The cost-benefit analysis performed by TREAT Steward™ balances the costs of antimicrobial treatment (antimicrobial costs, side-effects and ecological costs) with the benefits of the treatment: the life-years and bed-days saved by the treatment. The calculation of the benefit is based on the reduction in mortality (saved life-years) and the shorter time to recovery (saved bed-days) based on the probability that the infection is covered by the selected treatment. The individual cost-benefit analysis makes TREAT Steward™ a perfect tool for personalized treatment.

All of the costs are updated during the calibration process to fit local preferences. The ecological costs are adjusted individually to reflect the hospital’s antimicrobial policy. A treatment considered likely to increase bacterial resistance substantially, should be given a higher ecological cost than a treatment considered less likely to cause resistance.

CAUSAL PROBABILISTIC NETWORK
The secret lies within probabilities and odds ratios

The decision engines behind the family of TREAT products utilize Causal Probabilistic Network (CPN) technology, also called Bayesian networks. A CPN is a probabilistic graphical model that for example can represent the probabilistic relationships between diseases and symptoms. It is an advanced statistical technique that supports decision-making in complex domains based on partial, uncertain or unknown information. The basic units of a CPN, the nodes, are variables represented as probabilities. The full model of TREAT Steward™ is a comprehensive network of more than 6000 nodes. These nodes express, for example, physiological/biological concepts and mechanisms, and along with the causal links between them represent the incorporation of evidence-based knowledge gained from hundreds of studies. The model of TREAT Steward™ represents more than 70 man-years of development work. The risk assessment model in TREAT Lab™ is effectively a sub-model of that used in TREAT Steward™, where the focus is purely on sepsis presentation.

The figure illustrates the concept of the model. In this case, the probability of an E. coli urinary tract infection (E_Coli_UTI) is affected by three risk factors; place of infection acquisition, gender and the presence of a urinary catheter. These three risk factors represent knowledge of increased probability of an E. coli UTI infection.

Variables may be ‘parent’ or ‘founding’ nodes containing prior probabilities that are not dependent on other variables (e.g. gender). However, most variables are ‘child’ nodes containing conditional probabilities that are dependent on other variables in the CPN (e.g. the probability of “E. coli urinary tract infection” is dependent on the parent variables “gender”, “presence of catheter” and “place of acquisition”). The relationship between nodes in the CPN reflects causality (e.g. urinary tract infection causes leukocyturia).

INTELLIGENT ADVICE
- Visualizing the core of TREAT Steward™ decision support

The main goal for the family of TREAT products is to bring decision support to the critical decision point for the care of the infectious patient.

For TREAT Steward™ this means that it provides access to the severity of a possible sepsis and provides the probability of no-, mild-, moderate-, severe-, and critical sepsis: a gradation based on infection-related mortality. Moreover, TREAT Steward™ suggests the most likely diagnosis, choosing between more than 50 different infections, and provides advice showing the most probable pathogen(s) and the associated expected in vivo coverage of different local hospital treatments. As a clinician, you can interactively study and familiarize yourself with the in vivo coverage of various antimicrobials by clicking different treatments in the “recommended treatments” panel.

SITE SPECIFIC CALIBRATION
– Fitting the model to local preferences is one of the keys to success

There are several factors related to infectious diseases that vary significantly from hospital to hospital. To account for these differences, TREAT Steward™ is recalibrated any time it is transferred to a new environment; whether this is at a new hospital or even a new department at the same hospital. The calibration procedure for TREAT Steward™ involves adjusting the probabilities of the factors dependent on the local conditions that are critical to the optimal performance of the system.

The most important calibration task for TREAT Steward™ is fitting the system to the local patterns of pathogen susceptibility to antimicrobials. However, the following calibration tasks are among others also necessary:

• Antimicrobial-associated costs
• Direct costs
• Ecological costs
• Side-effects

• Local preferences
• Standard treatments and policies
• Antimicrobials used for susceptibility testing

• Local conditions
• Prevalence of risk factors
• Basic hospital costs
• Prevalence of pathogens
• Patterns of pathogen susceptibility
• Distribution of contaminants
• In vivo modification of susceptibilities

Following the calibration procedure, the complete system is fine-tuned and validated. During this quality assurance phase, the performance of the model is validated and brought into accordance with local hospital preferences and guidelines. A number of selected standard patients go through an intensive review with respect to the performance of the model’s assessment of sepsis severity, diagnosis and pathogen predictions. Additionally, TREAT Steward’s recommended treatments are reviewed and the coverage and costs analyzed.