Statistical Analysis Plan Study Notes: Direct-from-Blood Bacterial Testing Trial (NCT06069206)

Clinical Trial Identification and Protocol Overview

• Trial Title: Effect of Direct-from-blood Bacterial Testing on Antibiotic Administration and Clinical Outcomes: A Learning Healthcare Pragmatic Randomized Trial
• NCT Number: NCT06069206
• Principal Investigators (PIs): David Gaston, MD, PhD, and Romney Humphries, PhD, D(ADBMM), M(ASCP)
• Statistical Analysis Plan (SAP) Version: Version 2.0
• SAP Version Date: August 26, 2024
• ClinicalTrials.gov Upload Date: January 26, 2026
• Version Change History:
  • Version 1.0 (November 15, 2023): Original SAP version.
  • Version 1.1 (August 26, 2024):
  • Updated intervention group procedures reflecting protocol amendment approved on January 24, 2024.
  • Updated exclusion criteria reflecting protocol amendment approved on February 6, 2024.
  • Updated sample size considerations, main and additional analyses of the primary outcome, exploratory outcome and effect modification analyses, and missing data handling clarifications to align with manuscript submission to BMJ Open on June 20, 2024.

Introduction and Clinical Rationale

• Burden of Disease: Bacterial bloodstream infections are common, severe, and life-threatening clinical events.
• Limitations of Standard Diagnostics: Standard blood cultures typically require $24$\text{ to }$72\,\text{hours}$ to yield actionable results and exhibit imperfect diagnostic sensitivity.
• Emerging Diagnostic Technologies: Direct-from-blood rapid diagnostic tests identify bacterial pathogens directly from whole blood specimens without requiring culture growth, serving as an adjunct to standard blood cultures for early therapeutic optimization.
• Empiric Antimicrobial Therapy: Early intravenous (IV) antimicrobial administration is essential for adult patients presenting with suspected sepsis or bloodstream infection. Guidelines recommend empiric IV vancomycin for coverage of methicillin-resistant Staphylococcus aureus (MRSA).
• Therapeutic Challenges with Vancomycin: Vancomycin features a narrow therapeutic range and carries a substantial risk of toxicity, necessitating close therapeutic drug monitoring.
• Institutional Rationale: Vanderbilt University Medical Center aims to evaluate how health-system-wide deployment of direct-from-blood bacterial pathogen testing impacts antimicrobial administration, antimicrobial stewardship, clinical outcomes, health system resources, and overall care costs.

Study Design and Patient Population

• Study Design: Single-center, parallel-group, non-blinded, pragmatic, randomized clinical trial comparing standard blood culture management (usual care) against direct-from-blood pathogen testing plus blood culture (intervention group).
• Target Study Population Size: Total of $500$ adult patients ($250$ patients randomized to each arm).
• Setting: Emergency Department (ED) at Vanderbilt University Hospital.
• Inclusion Criteria:
  • Patient physical location in the Emergency Department at Vanderbilt University Hospital.
  • Time from presentation to the Emergency Department is $\le 12\,\text{hours}$.
  • Age $\ge 18\,\text{years}$.
  • Treating clinician has ordered blood cultures.
  • Treating clinician has ordered intravenous vancomycin.
• Exclusion Criteria:
  • Known status as a prisoner.
  • Known pregnancy.
  • History of receiving $> 2\,\text{doses}$ of IV vancomycin since ED presentation.
  • Documented positive bacterial culture within the preceding $7\,\text{days}$.
  • Known infection requiring $\ge 7\,\text{days}$ of IV vancomycin treatment regardless of diagnostic test results (e.g., severe skin and soft tissue infections).
• Randomization Procedure:
  • Patients meeting eligibility criteria are randomized in a $1:1$ allocation ratio.
  • Sequence generation is automated via computerized randomization within EPIC software functionality.
  • Method: Simple randomization without stratification, permuted block sizing, or minimization.
• Sample Size and Power Calculations:
  • Planned follow-up for primary outcome: $14\,\text{days}$ post-randomization.
  • Historical baseline control data indicate a median time to the primary outcome of approximately $48\,\text{hours}$.
  • Statistical Power: Assuming true median times to primary outcome of $48\,\text{hours}$ (control) versus $36\,\text{hours}$ (intervention), a sample size of $500$ achieves a power of $0.895$ to reject the null hypothesis of equal survival distribution curves.
  • Type I Error Rate: Specified at $\alpha = 0.05$ (two-sided).

Study Interventions and Clinical Workflows

• Blood Culture Group (Usual Care):
  • Blood culture collection, processing, and reporting follow standard clinical laboratory protocols at Vanderbilt.
  • Test updates are uploaded intermittently into the Electronic Health Record (EHR).
  • Negative culture updates post to EHR for routine provider review.
  • Gram-positive or gram-negative detection triggers performance of the ePlex/Roche rapid blood culture identification test, which covers $95\%$ of common bloodstream pathogens.
  • ePlex/Roche test results trigger automated EHR documentation and direct electronic notifications/alerts to the clinical team.
  • Infectious Disease fellows receive automated alerts for specific critical pathogens, including Staphylococcus aureus, Enterococcus species, and Staphylococcus lugdunensis.
  • Blood cultures are subcultured onto solid growth media for formal organism identification and antimicrobial susceptibility testing (AST).
  • All results post to the patient's MyHealth at Vanderbilt portal.
• Direct-From-Blood Test Group (Intervention Group):
  • Direct-from-blood testing is performed concurrently with standard blood cultures.
  • Blood collection staff obtain an additional $4\,\text{mL}$ sample dedicated to direct-from-blood assay during the blood culture draw.
  • Sample re-attempts: If an initial sample is inadequate (e.g., incorrect collection tube type, insufficient blood volume, missed order, or cancelled laboratory order), re-collection attempts are permitted up to $12\,\text{hours}$ post-randomization.
  • Diagnostic Platform: T2Bacteria Panel, processed in the Vanderbilt clinical laboratory.
  • Reporting and Notification Channels:
  1. EHR Microbiology Section: Results post directly into the standard microbiology section alongside blood culture data, complete with test performance specifications and diagnostic interpretation guidelines. Results are viewable by patients in MyHealth at Vanderbilt.
  2. Text Page Alerts: Automated direct text pages are transmitted to primary clinical care team providers at the exact moment results populate the EHR. Text notifications report status for all $5$ target pathogens on the T2Bacteria Panel and provide a link to result interpretation guidelines.
  3. Interruptive Best Practice Alerts (BPA): An automated BPA triggers at $7:00\,\text{AM}$ for patients with negative direct-from-blood results for S. aureus who maintain an active order for IV vancomycin. The alert prompts the provider to document a specific clinical rationale for continuation (e.g., suspected/confirmed non-MRSA indication, critical illness, severe alternative antibiotic allergy, surgical prophylaxis, or other). The BPA is non-binding and does not restrict clinical autonomy or trigger automated order cancellations.
  4. Pharmacokinetics Integration: Direct-from-blood assay results are provided to the consulting Pharmacokinetics (PK) service to inform clinical dosing adjustments and team recommendations.
  • Decision-Making Framework: Direct-from-blood test results are supplemental; diagnostic and therapeutic management decisions rely on overall clinical context (demographics, comorbidities, acute status, vital signs, concurrent labs, and therapy).

Co-Intervention: Pharmacokinetics Service Protocol

• Scope: Standard-of-care IV vancomycin management by the inpatient Pharmacokinetics service applies equally across both trial arms.
• Standard Clinical Workflow:
  1. Review patient clinical data essential for antibiotic dosing and safety monitoring.
  2. Verify or place orders for baseline serum creatinine levels.
  3. Calculate and submit pharmacokinetic-based vancomycin dosing orders.
  4. Order serum vancomycin drug levels (peak, trough, or random levels) as indicated.
  5. Perform ongoing clinical follow-up, serum level monitoring, and dose adjustments.
  6. Enter detailed daily PK consultation notes in the EHR until service discontinuation.
  7. Notify attending care teams immediately for specific clinical triggers:
  • Development of drug adverse reactions or signs of hypersensitivity.
  • Systemic drug concentration measuring $2\times$ above upper limit or significantly below lower therapeutic boundary.
  • Serum creatinine increase of $\ge 2\times$ relative to baseline value.

Study Endpoints and Process Measures

• Primary Endpoint:
  • Time to Last Dose of Intravenous Vancomycin: Measured in hours from randomization to the recorded start time of the final dose of IV vancomycin administered within $14\,\text{days}$ post-randomization.
• Secondary Endpoint:
  • Time to Last Dose of Systemic Anti-Pseudomonal Beta-Lactam Antibiotic: Measured in hours from randomization to the start time of the final dose of systemic anti-pseudomonal beta-lactam received within $14\,\text{days}$ post-randomization.
• Exploratory Antimicrobial Stewardship Outcomes:
  1. Cumulative total number of IV vancomycin doses received between randomization and Day $14$.
  2. Total number of days receiving gram-positive antibacterial therapy within $14\,\text{days}$.
  3. Total number of days receiving gram-negative antibacterial therapy within $14\,\text{days}$.
  4. Time to initiation of effective antibacterial therapy for bloodstream pathogens identified on baseline blood cultures (collected $\le 12\,\text{hours}$ before or at randomization), evaluated at Day $14$.
  5. Incidence rate of Clostridioides difficile infection occurring between randomization and hospital discharge or Day $28$ (whichever occurs first).
• Exploratory Safety Outcomes:
  1. Proportion of patients experiencing an antimicrobial-related allergic reaction within $14\,\text{days}$.
  2. Proportion of patients with IV vancomycin discontinued within $72\,\text{hours}$ of randomization who had a culture positive for MRSA within $\pm 24\,\text{hours}$ of randomization.
  3. Proportion of patients with all anti-staphylococcal therapy discontinued within $72\,\text{hours}$ of randomization who had a culture positive for S. aureus within $\pm 24\,\text{hours}$ of randomization.
  4. Proportion of patients with all anti-pseudomonal therapy discontinued within $72\,\text{hours}$ of randomization who had a culture positive for Pseudomonas aeruginosa within $\pm 24\,\text{hours}$ of randomization.
• Exploratory Clinical Outcomes:
  1. Maximum Acute Kidney Injury (AKI) stage based on KDIGO creatinine criteria within $14\,\text{days}$.
  2. Proportion of patients requiring renal replacement therapy (RRT) within $14\,\text{days}$.
  3. Nadir platelet count documented within $14\,\text{days}$.
  4. Hospital-free days to Day $28$ (alive and out of hospital; outcome censored at index discharge).
  5. ICU-free days to Day $28$ (alive and out of intensive care unit; outcome censored at index discharge).
  6. All-cause in-hospital mortality assessed at Day $28$.
• Process Measures:
  1. Time from randomization to initial detection of blood bacterial pathogen (via direct-from-blood test or blood culture).
  2. Time from randomization to direct-from-blood test result turnaround (intervention arm).
  3. Time from randomization to blood culture result turnaround (both arms).
  4. Concordance rate between direct-from-blood test results and formal blood culture results (intervention arm).
  5. Rate of non-vancomycin antimicrobial administration for MRSA bacteremia within $14\,\text{days}$.
  6. Total number of Infectious Disease specialty consultations within $14\,\text{days}$.
  7. Total count of supratherapeutic vancomycin serum levels within $14\,\text{days}$.
  8. Proportion of trial patients managed by the Pharmacokinetics consult service.

Statistical Methodology and Datasets

• Analysis Population: Intention-to-Treat (ITT) population, analyzing all randomized participants according to their assigned group regardless of protocol adherence or specimen processing success. No per-protocol exclusions are planned.
• Analytical Software: Executed using reproducible statistical code in R (R Foundation for Statistical Computing, Vienna, Austria).
• Significance Testing: Two-sided significance threshold defined as $P < 0.05$ for the primary endpoint. No adjustments for multiplicity will be applied given the single primary outcome structure.
• Secondary and Exploratory Reporting: Results emphasize effect estimates (hazard ratios, odds ratios, median differences) and $95\%$ confidence intervals rather than reliance on nominal P-values.
• Descriptive Summaries:
  • Categorical data: Frequencies and percentages ($n, \%$).
  • Continuous data: Means and standard deviations (mean $\pm$ SD) or medians and interquartile ranges (IQR).
  • Missing data frequencies reported for all baseline covariates and study metrics.
  • Data distributions illustrated using box plots, violin plots, or histograms.

Primary and Secondary Endpoint Analyses

• Main Primary Outcome Analysis:
  • Unadjusted ITT comparison using a Cox proportional hazards model.
  • Dependent Variable: Time to last dose of IV vancomycin within $14\,\text{days}$.
  • Independent Variable: Randomized treatment arm assignment.
  • Discharge Assumption: Patients discharged alive on or prior to Day $14$ are assumed to receive no further inpatient IV vancomycin post-discharge.
  • Effect Metric: Hazard Ratio (HR) with $95\%$ confidence interval.
• Sensitivity Analyses for Primary Outcome:
  1. Fixed Penalty for Death: Patients dying on or before Day $14$ post-randomization are assigned an arbitrary time-to-event value of $15\,\text{days}$ (worse than the maximum achievable outcome for surviving patients) to prevent mortality from masking antibiotic exposure differences.
  2. Fine-Gray Competing Risk Model: Subdistribution hazard model accounting for death as a competing risk. Discontinuation while alive requires surviving $\ge 48\,\text{hours}$ after the last vancomycin dose; death within $48\,\text{hours}$ of the last dose is classified as death prior to discontinuation.
  3. Proportional Odds Model: Ordinal logistic regression where duration is categorized into integer days ($1\,\text{to }14\,\text{days}$), with death $\le 14\,\text{days}$ assigned an ordinal rank of 15$.\n 4. **Subgroup Sensitivity Analyses**:\n * Analysis restricted to patients with successful blood culture collection within \pm 12\,\text{hours} of randomization.\n * Analysis restricted to patients receiving \ge 1\,\text{dose}$of IV vancomycin within$\pm 12\,\text{hours} of randomization.\n* **Modified Intention-to-Treat (mITT) Analysis**:\n * Primary analysis repeated exclusively among participants whose blood cultures yielded final results and, in the intervention arm, whose direct-from-blood assay yielded a final result.\n* **Secondary Endpoint Analysis**:\n * Time to last dose of systemic anti-pseudomonal beta-lactam evaluated using the exact Cox proportional hazards modeling structure described for the primary endpoint.\n\n# Exploratory Endpoint Analyses\n\n* **Categorical Variables**:\n * Unadjusted comparisons: Chi-square test.\n * Adjusted comparisons: Binary or multinomial logistic regression models.\n* **Continuous and Ordinal Variables**:\n * Unadjusted comparisons: Wilcoxon rank-sum test.\n * Adjusted comparisons: Proportional odds logistic regression models.\n* **Time-to-Event Variables**:\n * Unadjusted and adjusted comparisons: Cox proportional hazards regression models.\n* **Reporting Standard**: Point estimates accompanied by 95\% confidence intervals without adjusting for multiplicity.\n\n# Effect Modification Analyses\n\n* **Statistical Interaction Modeling**: Formal test of interaction within a Cox proportional hazards model containing main fixed effects for trial group, baseline modifier, and the group-by-modifier interaction term.\n* **Modeling Continuous Modifiers**: Continuous modifier variables evaluated using restricted cubic splines (3\text{ to }5 knots) to account for potential non-linear interactions; variables may be dichotomized for graphical presentation.\n* **ICEMAN Framework Prespecified Effect Modifiers and Hypotheses**:\n 1. **End-Stage Kidney Disease (ESKD) on RRT (Yes/No)**: Hypothesized *no* effect modification on the primary endpoint.\n 2. **Sepsis Status (Yes/No meeting Sepsis-3 criteria)**: Hypothesized *no* effect modification on the primary endpoint.\n 3. **Baseline Illness Severity (SOFA Score)**: Hypothesized *no* effect modification on the primary endpoint.\n 4. **Suspected Source of Infection (Pulmonary, Intra-Abdominal, Genitourinary, Other, Unknown)**: Hypothesized *presence* of effect modification, expecting a significantly larger reduction in vancomycin duration in the intervention arm among patients with non-pulmonary infection sources relative to pulmonary sources.\n 5. **Transplant Status (Yes/No for solid organ or stem cell transplant)**: Hypothesized *no* effect modification on the primary endpoint.\n 6. **Neutropenia (Yes/No for Absolute Neutrophil Count < 1500\,\text{cells}/\mu\text{L}$$)**: Hypothesized *presence* of effect modification, expecting a significantly larger reduction in vancomycin duration in the intervention arm among non-neutropenic patients compared to neutropenic patients.

Missing Data Handling Strategy

• Loss to Follow-up: Expected to be zero prior to primary outcome determination, as outcome ascertainment is bounded entirely within the index hospital admission.
• Outcome Data Imputation: No imputation will be performed for missing primary, secondary, or exploratory outcome metrics.
• Baseline Covariate Imputation: Missing values among baseline covariates in adjusted multivariable models will be handled using multiple imputation techniques.