Our study including 727 adult patients with sepsis has been previously published [13], reporting detailed methodology and clinical outcomes. In summary, the study was a prospective, time-series evaluation of a bundled intervention consisting of ASP, PCT, and BCID. The evaluation included pre-intervention (342 patients) and intervention (385 patients) cohorts from two adult ICUs in Alberta, Canada. The study was approved by the Research Ethics Boards at the University of Alberta (Pro00101420) and the University of Calgary (REB17-2244), with a waiver of consent given the study was classified as a quality improvement initiative.

This cost-effectiveness analysis was pre-planned and conducted from the perspective of the Canadian public healthcare payer, responsible for healthcare expenses associated with the management of sepsis during the index admission and costs for subsequent readmissions within 90 days after index discharge. Our cost estimates include detailed patient-level direct costs such as hospital and ICU stays, antimicrobial utilization, treatment for nosocomial CDI, and PCT and BCID testing, as well as indirect overhead costs including physician and pharmacy services for ASP. All costs are reported in Canadian dollars adjusted to 2022 values using the Canadian Consumer Price Index (CPI).

Data were primarily obtained from health administrative databases, the Discharge Abstract Database (DAD), and medical record reviews, with supplementary information from published literature. The costs obtained represent patient-level specific direct costs, determined through the resource intensity weighting (RIW) approach. The RIW is a metric reflecting the expected resource consumption associated with an individual patient [14]. Its calculation incorporates various factors pertinent to the patient’s health status, including diagnoses, procedures, age, and other relevant characteristics. A patient assigned a higher RIW score is indicative of greater resource requirements, consequently resulting in a higher associated cost. The calculation of a patient’s cost involves multiplying their RIW score by the standard hospital stay cost (CSHS). The CSHS, in turn, indicates an average cost derived by dividing a hospital’s total inpatient expenses by the cumulative sum of RIW scores assigned to treated patients. The cost items involve a comprehensive range of hospital expenditures, including non-physician staff salaries, medical and surgical supplies, and indirect overhead costs such as administration and support services. All physician services in hospitals are publicly funded, therefore, patient-level specific costs were collected from the Practitioner Claims database by Alberta Health (AH) [15]. Only hospital-related costs were included following the index discharge date.

In our cost estimation, we factored in the occurrence of CDI and antimicrobial use. CDI incidence data were gathered through provincial Infection Control and Prevention databases. The cost of CDI treatment was derived from a previous study conducted in Alberta [16], which reported a hospital-acquired CDI cost of $9365 per patient (adjusted to the 2022 price). Antimicrobial costs were obtained from the Alberta Health Services formulary/inpatient drug cost database. The annual cost for ASP in the ICU was estimated at $114,354, comprising $40,000 for physician-related expenses and $74,344 for dedicated ASP pharmacists. Costs associated with implementing PCT and BCID were obtained from participating Alberta chemistry and microbiology laboratory managers. Regarding testing costs, PCT testing cost $27 per test while BCID testing incurred a fee of $150 per test.

We used eCritical Alberta, which is a provincial critical care database that prospectively captures patient-specific demographic, clinical, and outcome data on all patients admitted to an Alberta ICU. These data have been extensively validated [17]. We obtained patient-specific in-hospital and in-ICU mortality rates from eCritical Alberta. The primary outcome was cost per sepsis case. Secondary outcomes included readmission rates, Clostridioides difficile infections, mortality, and lengths of stay.

To estimate health utility, the EQ-5D was utilized for survivors of critical illness and patients facing antimicrobial-related adverse events, with data from the literature that calculated using the UK value set [18] (Table S.1 in Supplementary Material). Quality-adjusted life years (QALYs) were calculated using quality-of-life (QOL) health utility scores for ICU patients and the disutility in patients afflicted by CDI [17]. Patients who died during their hospital or ICU stay were assigned a QOL utility score of zero. As the model was conducted within a 90-day time horizon, the generated health outcome represented QALYs over the 90 days. Outcomes are reported as the incremental net monetary benefit (iNMB), a metric in cost-utility analysis that is used to quantify the economic impact of treatment options in monetary terms [19]. The NMB of a treatment strategy is calculated as ([benefit × WTP threshold value] − cost), where a WTP threshold of $50,000 is applied in the analysis. The iNMB is then the difference between the NMB of two strategies. A positive iNMB for one treatment over another suggests the treatment is favorable (that its benefits outweigh those of the alternative). Conversely, a negative iNMB suggests that the treatment is comparatively less economically favorable than the alternative.

We performed an economic evaluation using data obtained during the clinical study [13] to assess the cost-utility of the intervention compared with pre-intervention within a 90-day time horizon. As a comparator, we selected the pre-intervention group as it represents the existing standard practice in Alberta. No discount rate was applied to costs and effectiveness, as the study period was less than 1 year.

To evaluate cost-utility, we constructed a decision tree model. The flowchart depicted in Fig. 1 outlines the care path and associated resource utilization. Expected costs and outcomes were estimated from the resource use recorded for patients as they progress through the care pathway depicted.

Patient flowchart. The flowchart outlines the process for patients admitted to the ICU with confirmed or suspected sepsis. The intervention group includes ASP, PCT, and BCID testing. Surviving patients are investigated for adverse events (i.e., CDI) and associated costs. Decision points assess first and second hospital readmissions. The flowchart ends with two possible states: patient death in the hospital or no hospital readmission. The control group follows a similar flow, but lacks ASP, PCT, and BCID implementation. The probabilities at each decision point within the decision tree pathways are determined through statistical analysis of the study data. Detailed results are provided in Tables 1 and S.2. Additionally, point estimates of these probabilities are available in Table S.3. AE adverse event, ASP antimicrobial stewardship program, BCID blood culture identification, CDI Clostridioides difficile infection, died died in hospital, PCT procalcitonin, re-ad re-admission

The flowchart starts with patients admitted to the ICU with confirmed or suspected sepsis. The first decision node pertains to whether the patient dies in the ICU or during their hospital stay. If such an event occurs, the model ends. In addition, an investigation is undertaken regarding CDI, an adverse event (AE) resulting from antimicrobial usage, for those patients who survived their ICU or hospital stay. The costs incurred in managing CDI are calculated as part of this analysis. The flowchart also includes decision points to ascertain whether patients experience first and second readmissions to hospital.

The flowchart ends with two possible states: the patient’s death (in the ICU or hospital) or the patient not being readmitted to the hospital. Given the relatively low rate of multiple hospital readmissions within 90 days of discharge (Table S.1), the model does not distinguish between patients experiencing more than two readmissions. Consequently, the second readmission encompasses patients with two or more readmissions.

The flowchart for the pre-intervention group mirrors that of the intervention group, except that the pre-intervention group does not include the implementation of ASP, PCT, and BCID. The decision tree model was developed and analyzed using TreeAge Pro 2015 (TreeAge Software, Inc., Williamstown, MA, USA).

We utilized the study data to estimate parameter values for the decision model, incorporating costs, mortality rates, CDI incidence, and the number of readmissions within 90 days of discharge for the intervention and pre-intervention groups (Tables 1 and 2). Additionally, we analyzed patient distribution on the basis of hospital and ICU readmissions within the 90 days (Table S.2 in Supplementary Material).

We employed a Wilcoxon rank-sum nonparametric test to assess the cost difference between the groups, which is suitable for comparing mean differences in non-normally distributed data [20]. Furthermore, we presented the observed proportions of readmission, mortality, and CDI cases in a two-by-two table and calculated odds ratios. The difference in observed proportions was then assessed using the chi-squared test [21]. Statistical analyses were performed using R version 4.2.3 [22].

To assess differences in means, we employed Monte Carlo simulation, a widely used mathematical technique to account for uncertainty in the results. The cost and effectiveness data were repeatedly modeled in the simulation for 10,000 iterations, using predetermined probability distributions [23,24,25]. We used the gamma distribution for cost inputs and the beta distribution for probabilities and quality of life utilities. The outcomes are presented through the incremental cost-effectiveness plane scatterplot and cost-effectiveness acceptability curve (CEAC). The cost-effectiveness plane scatterplot illustrates the variability in the cost-utility outcomes, while the CEAC conveys the likelihood of the intervention and pre-intervention being cost-effective [26, 27].

We performed a one-way sensitivity analysis by varying mortality and CDI incidence within the ranges based on the 95% confidence intervals (CI) (Table 1) to assess the robustness of our findings.