Anticoagulation versus Antiplatelet Therapy After Cancer-Associated Ischemic Stroke: A Net-Benefit Threshold Analysis

Technical Summary: Anticoagulation versus Antiplatelet Therapy After Cancer-Associated Ischemic Stroke

Problem Statement
Cancer-associated ischemic stroke presents a complex clinical dilemma: patients face high risks of recurrent thromboembolism and death, yet the optimal antithrombotic strategy for secondary prevention remains uncertain. While existing syntheses and network meta-analyses provide treatment rankings, they fail to address the specific "threshold question" driving individual clinical decisions. Specifically, current evidence does not define the precise trade-off between recurrence reduction and bleeding risk at which anticoagulation (AC) becomes net-beneficial compared to antiplatelet therapy (AP). Furthermore, available randomized data in cancer-relevant populations are often small, indirect, or underpowered, leaving a gap in determining whether the potential benefit of AC outweighs its hemorrhagic risks given the high competing mortality of active cancer.

Methodology
The authors developed a short-horizon, mortality-adjusted net-benefit threshold model to evaluate the conditions under which AC provides greater clinical net benefit than AP. The study adhered to ISPOR–SMDM Modeling Good Research Practices and excluded cost-effectiveness analysis, focusing solely on utility-weighted clinical net benefit (Quality-Adjusted Life Years, QALYs).

• Model Structure: A decision tree with a 6-month time horizon was constructed to reflect the limited window for secondary prevention benefit in patients with active cancer due to competing mortality. Outcomes included no event, recurrent ischemic stroke/systemic embolism, major bleeding (including intracranial hemorrhage), and death.
• Threshold Variable Approach: Unlike traditional meta-analyses that estimate a fixed pooled effect, the model treated the relative risk (RR) of recurrence for AC versus AP as a threshold variable. The goal was to identify the "break-even" point where the incremental net benefit (INB) equals zero.
• Parameterization:
  • Recurrence: Three priors were used for probabilistic sensitivity analysis (PSA): a legacy broad comparison prior and two "evidence-informed" lognormal priors derived from randomized head-to-head estimates (NAVIGATE-ESUS cancer subgroup and the main ARCADIA trial).
  • Bleeding: Excess major bleeding risk was modeled as a wide triangular scenario range (0.00–0.08) rather than a fixed point estimate, acknowledging sparse arm-level data.
  • Mortality: A 6-month competing mortality rate was interpolated (base case 30%) based on cohort data, treated as non-differential between strategies in the base case.
  • Utilities: Health states were weighted using mRS-based utilities (UK EQ-5D-3L value set), with pragmatic scenario inputs for event disutilities.
• Analysis: Deterministic sensitivity analysis generated a two-way threshold surface mapping the preference boundary across recurrence RR and excess bleeding risk. Probabilistic sensitivity analysis (10,000 simulations) calculated the probability of AC being favored under different priors. Expected Value of Perfect Information (EVPI) was computed to quantify residual decision uncertainty.

Key Contributions

1. Decision-Analytic Framework: The study introduces a specific threshold model that integrates recurrence prevention, hemorrhagic harm, and limited survival into a single decision-relevant metric, moving beyond simple treatment rankings.
2. Evidence Positioning: The authors systematically positioned available randomized head-to-head estimates (NAVIGATE-ESUS, ARCADIA) against the calculated break-even threshold, highlighting the indirectness and imprecision of current data regarding active cancer hypercoagulability.
3. Quantification of Uncertainty: By treating the treatment effect as a variable rather than a fixed estimate, the study quantifies the probability that AC is favored under different evidence assumptions, demonstrating how the inclusion of evidence-informed priors shifts the decision landscape.

Results

• Break-Even Threshold: Under base-case assumptions (2% excess major bleeding), anticoagulation becomes net-beneficial only if it reduces the relative risk of recurrent stroke to below 0.825.
• Probability of Favorability:
  • Under the legacy broad prior (Uniform 0.40–1.10), the probability that AC was favored was 0.44.
  • Under the ARCADIA-informed neutral prior (HR 1.00, 95% CI 0.64–1.55), this probability dropped to 0.12.
  • Under the NAVIGATE-cancer-informed skeptical prior (HR 1.43, 95% CI 0.71–2.87), the probability fell to 0.04.
• Evidence Alignment: The available randomized estimates (NAVIGATE-ESUS cancer subgroup HR 1.43; ARCADIA main trial HR 1.00) lay predominantly at or above the break-even threshold of 0.825. Their confidence intervals crossed both the break-even point and the null (RR=1.00). The only estimate below the break-even was an underpowered, crude estimate from a cancer-history subgroup (≈0.69) lacking a formal interval.
• Value of Information: The EVPI was small, particularly under evidence-informed priors, indicating that the remaining uncertainty stems from the indirectness and imprecision of existing evidence rather than a high potential for a large net benefit from a new trial.

Significance and Claims
The paper claims that current randomized evidence does not demonstrate a recurrence reduction sufficient to make anticoagulation net-beneficial for cancer-associated ischemic stroke under base-case harm assumptions. The authors emphasize that the model does not prove AC is inferior, but rather that the existing data cannot confirm it is superior.

The significance of the work lies in its reframing of the clinical question: it shifts the focus from "which treatment is better on average" to "what specific reduction in recurrence is required to justify the bleeding risk." The authors conclude that the remaining uncertainty reflects the limitations of indirect and imprecise evidence, thereby supporting the rationale for a future confirmatory randomized trial rather than an immediate change in clinical practice. The study serves as a structured description of decision uncertainty and a design target for future research, explicitly stating it is hypothesis-generating and decision-structuring, not a clinical guideline.