Identifying Patients with Erythrocytosis Not Requiring JAK2 Testing: External Validation and Decision Analysis of the Jakpot Prediction Rule

Background: Diagnostic workup of erythrocytosis should consider JAK2 testing for polycythemia vera (PV), a myeloproliferative neoplasm (MPN) associated with increased risk of stroke, myocardial infarction, and thrombosis when untreated. However, universal testing would incur tremendous cost. The JAKPOT clinical prediction rule (JGIM 2023) was developed to determine which patients with erythrocytosis can safely forgo JAK2 testing. Patients who met 0/3 criteria (red blood cell count >6.45x1012/L, platelets >350x109/L, and neutrophils >6.2x109/L) were JAKPOT “low-risk” with a test sensitivity of 94.7%. In this study, we 1) externally validated JAKPOT in two independent cohorts and 2) employed decision analysis to derive minimum cost-effective JAKPOT test characteristics needed to safely forgo JAK2 testing in patients with erythrocytosis. Methods: Demographics and laboratory data for all available patients with JAK2 testing were extracted from the Beth Israel Deaconess Medical Center (BIDMC) and the Veteran Affairs (VA) data warehouses. Patients with a known hematologic malignancy were excluded. In the BIDMC cohort, male patients with hemoglobin ≥16.5 g/dL and female patients with hemoglobin ≥16 g/dL were included. In the VA cohort, due to strong male preponderance, patients with Hemoglobin ≥16.5 g/dL were included. Only patients with all three JAKPOT variables available on the day of JAK2 testing were included to minimize confounding. Sensitivity (Sn), specificity (Sp), and negative predictive value (NPV) were calculated for each cohort. We then developed a Markov cohort model to evaluate the cost-effectiveness of the JAKPOT prediction tool (JAKPOT strategy) versus status quo (JAK2 test all) for individuals with erythrocytosis at median age 60. The model was parameterized with 1) international, prospective PV natural history data (4393 person-years), 2) randomized trial data on treated (Hct <45%) vs untreated (Hct >45%) PV patients, 3) PV-specific health resource utilization and utilities, and 4) long-term impacts on quality of life and health resource utilization for PV patients with thrombotic or non-fatal cardiovascular events (ischemic stroke, myocardial infarction, venous thromboembolism). We presumed JAKPOT-low patients with persistent erythrocytosis would be clinically reevaluated at 1 year (or sooner). The primary outcome was the incremental net monetary benefit (iNMB) for a 60 to 61-year-old cohort of Americans with erythrocytosis (~88,000). The secondary outcome was a threshold minimum sensitivity for the cost-effective application of JAKPOT. We performed deterministic and probabilistic sensitivity analyses (PSA; 10,000 Monte Carlo iterations) to evaluate the effect of all parameters individually on model outcomes, as well as in all parameters simultaneously. This analysis is reported from the US health system perspective at a lifetime time-horizon and a willingness-to-pay threshold of $100,000/QALY. Results: 52,865 patients with JAK2 testing were identified across the BIDMC (2038) and VA cohorts (50,827), with 333 and 8213 remaining after exclusion criteria were applied. The median ages were 63 and 64 with 76% and 98% males, and there was a JAK2 positivity rate of 14.4% and 6.6% in each cohort, respectively. JAKPOT test characteristics in the BIDMC cohort were Sn 0.94 (0.83-0.98), Sp 0.68 (0.63-0.74), NPV 0.98 (0.96-0.99). JAKPOT test characteristics in the VA cohort were Sn 0.81 (0.78-0.84), Sp 0.68 (0.67-0.69), NPV 0.98 (0.976-0.984). In the base-case, the status quo strategy accrued more population lifetime QALYs (264 more) and cost $14 million more than the JAKPOT strategy. The iNMB favored status quo over the JAKPOT strategy by $12.3 million [95% credible interval $3.5-$22.9 million], with status quo favored in 100% of 10,000 Monte Carlo iterations. The JAKPOT rule threshold minimum sensitivity (i.e., for the JAKPOT strategy to be cost-effective) was 97%. Conclusion: Despite high NPV in the studied cohorts, JAKPOT is a cost-ineffective clinical prediction tool due to insufficient test sensitivity across derivation and external validation cohorts. Given that JAK2 is an expensive test with near perfect Sn and Sp, a first-step clinical prediction rule in the rule out of PV needs improved sensitivity than what is currently available (i.e., to sufficiently abrogate increased patient risk-exposure time to fatal and non-fatal thromboembolic events).