Page 82 - 2019_12-Haematologica-web

A. Ferrari et al.
A
B
C
D
Figure 3. Outcomes incidence during follow up according to logistic Generalized Linear Mixed Model (GLMM) and comparison with negative-binomial model.
Dashed lines are 95% Confidence Interval (CI), observed frequencies are plotted in hollow circles of size proportional to sample size in person/years. ICC (Intracluster Correlation Coefficients) and P-values of Likelihood Ratio Tests of random slopes are reported. Thrombosis (A). Mortality (B). Myelofibrosis (C). Acute myeloid leukemia (D).
predictor of thrombosis risk in meta-regression.
A logistic model allows for incidence rates to change over time. To confirm that our results do not heavily depend on this assumption, we carried out a sensitivity analysis comparing the logistic GLMM to a negative bino- mial regression. In a negative binomial regression, yearly incidence is assumed constant over time. Results from the two models were fundamentally in agreement for throm- bosis and AML outcomes, whereas for MF and overall mortality, they started diverging after five years of follow up. This indicates that, for practical purposes, thrombosis incidence rate can be assumed to be constant over time, at
least up to a 10-year observation period.
Thrombosis incidence
Adjusted estimates for annual incidence of thrombosis are reported in Table 2, globally and stratified by median age and previous thrombosis. Average incidence rate was 3.3% persons/year, ranging from 1.9% at 60 years of age with no history of thrombosis to 6.8% at a median age of 80 years. Estimates increase with median age and are higher in presence of history of thrombosis, but the latter difference is not statistically significant. On the other hand, in a sub-analysis on arterial and venous thrombotic events, previous thrombosis was a highly significant (P<0.001) predictor of incidence of arterial thrombosis, but not of venous.
Hematologic transformations and mortality
Interestingly, incidence of MF and overall mortality increases steeply after five years of follow up according to the logistic GLMM. Estimates of myelofibrosis risk at a median age of 68 years are 0.9%, 5.0% and 33.7% at 1, 5 and 10 years respectively, whereas mortality under the same conditions was 2.4%, 12.6% and 56.2%, but these estimates increase or decrease with age at the start of fol- low up. Specifically, the odds of MF transformation increase on average 6% (95%CI: 1-11%) for each year of age, while those of mortality increase by 21% (95%CI: 9- 33%).
Acute myeloid leukemia evolution, on the other hand, showed a stable incidence over time. According to the negative binomial model, the annual rate of AML transfor- mation was 0.4%, although the logistic model suggests a slight tendency to increase after around eight years.
Bleeding
The number of major bleedings was considered too small for reliable inference. Based on 88 events over 1,485 patients, pooled incidence of bleeding was 1% per year, independently of follow-up duration or antithrombotic therapy, as shown by meta-regression. This estimate was quite consistent, since no evidence of study heterogeneity was found for this outcome, but the small sample size may have limited accurate detection of these effects.
2396
haematologica | 2019; 104(12)