COVID19-hematological Malignancies: the Italian Hematology Alliance

This is a retrospective/prospective, cohort, non-interventional observational study. An informed consensus for the participation is available. In this section we provide informations on sample size and statistical analysis. In Italy, the projected estimate of complete HM prevalence at Jan 1, 2020 has been established as 48,254 cases for Hodgkin lymphoma, 110.715 cases for non Hodgkin Lymphomas, 67,301 for leukemias, and 25,066 for multiple myeloma (Guzzinati et al, BMC Cancer 2018). The Italian Dipartimento della Protezione Civile website reported (March 23, 2020) that 63,927 cases are currently infected with SARS-CoV-2. No formal sample size calculation was made for this project but, on the basis of data available to date, considering the prevalence of hematological patients in Italy (0.4%) and assuming that these patients have the same risk of contracting COVID-19 as the general population, we supposed to enroll at least 250 patients (at March 24, 2020). Statistical analyses All data collected will be summarized using appropriate descriptive statistics: absolute and relative frequencies for discrete variables; mean, standard deviation, median and interquartile range for continuous ones. To identify factors significantly associated with composite endpoint, log-binomial regression will be used for modelling risk ratio together with 95% confidence interval estimated. The least absolute shrinkage and selection operator (LASSO) method will be applied for selecting the factors able to independently predict primary end-point. LASSO selects variables correlates to the measured outcome by shrinking coefficients weights, down to zero for the ones not correlated to outcome. In addition, machine learning techniques will be used for validating results from LASSO. A weight will be assigned to each coefficient of the selected predictors and weights will be summed to produce a total aggregate score. Predictive performance will be assessed through discrimination and calibration. Discrimination indicates how well the model can distinguish individuals with the outcome from those without the outcome. Two, the net reclassification improvement (NRI) will be calculated for assessing the 'net' number of individuals correctly reclassified using "the new model" over a comparator index \[i.e., CCI (Charlson Comorbidity Score) or MCS (Multisource Comorbidity Score), or HM-disease specific\]. Calibration ascertains the concordance between the model's predictions and observed outcomes, which we evaluated using a calibration plot. Cartographic and geostatistical methods will be used to exploring the spatial patterns of disease. An Exploratory Spatial Data Analysis (ESDA) and the Kriging method will be also applied to describe and model spatial (geographical) pattern.