A dynamic prognostic model to predict survival in post-polycythemia vera myelofibrosis

Polycythemia vera (PV) is a chronic myeloproliferative disorder with a propensity to develop myelofibrosis, a condition named post polycythemia vera myelofibrosis (post-PV MF). Survival and prognostic factors after transition to MF remain to be defined. We studied 68 patients with post-PV MF to define survival and prognostic factors for survival at diagnosis of post-PV MF. We also developed a dynamic prognostic model to predict survival at any time from diagnosis of post-PV MF. The median interval between the diagnosis of PV and that of post-PV MF was 13 years (range, 4–29.6 years). Patients with post-PV MF were observed for 181 person-years of follow-up. At diagnosis of post-PV MF, 43 (63%) of 68 patients had less than 65 years. During the follow-up, the incidence of thrombosis was 42 × 1000 person-years (95% CI: 19–93.5) and the incidence of leukemia was 50.3 × 1000 person-years (95% CI: 26–115). The median survival was 5.7 years. Multivariable Cox proportional hazard regression including age, hemoglobin value, platelet count, leukocyte count, and spleen size, showed that hemoglobin < 10 g/dL (P < .001) and platelet count < 100 × 109/L (P= .026) were independent risk factors for survival. We stratified patients at diagnosis of post-PV MF, according to these factors, obtaining two risk groups with significantly different survival (P = .003): low risk (Hb > 10 g/dL and platelet count > 100 × 109/L) with a median survival of 7 years, and high risk (Hb < 10 g/dL or platelet count < 100 × 109/L) with a median survival of 2 years. The prognostic model retained significance after adjustment for age in a multivariable Cox proportional hazard regression (HR: 4.3, 95% CI: 1.6–11.4; P= .003). To assess whether this prognostic model may predict survival at any time from diagnosis of post-PV MF, we evaluated in a time-dependent analysis 64 patients who had longitudinal blood cell counts during follow-up. As first step, we evaluated univariate survival analysis with hemoglobin value < 10 g/dL and platelet count < 100 ×109/L as time-dependent covariates. Both time-dependent parameters affected survival (HR for hemoglobin 5.8, 95% CI: 2.2–15.2, P < 0.001; HR for platelets 4.5, 95% CI: 1.67-12, P=.002). As second step, we evaluated the prognostic model assessed at diagnosis as time-dependent covariate, to define whether the acquisition of one risk factor during follow-up may affect survival. The HR was 7.5 (95% CI: 2.4-23.4; P < .001). The time-dependent prognostic model retained statistical significance after adjustment for age (P < .001). In conclusion, in patients developing post-PV myelofibrosis, a prognostic model based on hemoglobin level < 10 g/dL and platelet count < 100 × 109/L may predict survival at diagnosis of post-PV MF and at any time thereafter.