A Phase II Study on Allogeneic Stem Cell Transplantation in Patients With Active Acute Leukemia

The overall survival at 2 years (from time of enrolment) of all patients enrolled in to the study (either transplanted or not).

This is the simplest outcome, defined as the probability of survival irrespective of disease state at any point in time. Patients alive at their last follow-up are censored. It is analyzed by the Kaplan-Meier method, Log-Rank Test and parametric or semiparametric survival models.

DFS is defined as the probability of being alive free of disease at any point in time. Thus, death or disease relapse are treated as events (1). Patients alive and free of disease at their last follow-up are censored. The statistical methods for the analysis of DFS are the same as for OS (Kaplan-Meier curve, Log-Rank Test and survival models).

RI is defined as the probability of having had a relapse before time t. Death without experiencing a relapse is a competing event. The correct method of analysis is therefore the estimation of the Cumulative Incidence curve, comparable by the Gray Test and, for the multivariate analysis, the application of the proportional hazard model for the sub-distribution of competing risks, by Fine and Gray. In studying relapse, sometimes the interest is not only in the estimation of the cumulative incidence curve, but also in the estimation of the hazard ratios for comparing groups of patients. It is therefore common to apply also a survival (Cox or parametric) model considering relapse as an event and death without relapse as a censoring (the response time is given by the minimum between time to relapse and time to death without relapse; as usual, a patient who is alive and free of relapse is also censored).

It is defined as the probability of dying without previous occurrence of a relapse, which is a competing event. The same indications as for the analysis of RI apply.

It is defined as the probability of being alive with no indication of disease progression (relapse is considered as progression for patients in CR). It is analyzed by KaplanMeier curve, Log-Rank Test and parametric or non-parametric survival models.

The day of engraftment is defined as the first day of 3 consecutive days with a persistent blood cell count above a predefined level:

WBC 1 x 109/l PMN 0.5 x 109/l Platelets 50 x109 /l or 20x 10 9 /l Death without recovery is a competing event, while no engraftment at the last followup is to be considered as a censored observation. Relapse or disease progression could be considered (depending on the disease being studied) as further competing events: this must be discussed with the responsible physician.

The available information in the EBMT data regard the date of onset and the maximum grade of aGvHD. It is therefore possible to estimate the probability of aGvHD in a competing risks setting (death is a competing event; whether relapse/progression is a competing event must be discussed with the physician). By definition, patients alive (relapse/progression-free) at day 100 without having experienced aGvHD are censored. If the dates of onset are missing for the majority of patients, the analysis can focus only on the occurrence of aGvHD, which is analyzed by a logistic regression model. This method would however be incorrect if there is a (non negligible) percentage of censored observations or if competing events occurred before day 100.

When possible, if information on the date of 1°occurrence of cGvHD is available, it should be analyzed as a time-to-event outcome, being death (and possibly relapse/progression) the competing event; data are censored for patients alive (relapse/progression-free) without episodes of cGvHD at last follow-up. Since cGvHD is defined only for patients surviving at least 100 days, the survival model should consider a left truncation at 100 days; alternatively, the time of occurrence of cGvHD must be computed from 100 days. If information on the timing of cGvHD is not available, the outcome considered is the occurrence, and the statistical model to be used is the logistic regression. Only patients surviving at least 100 days are considered to be at risk of developing cGvHD, therefore the analysis must be restricted to these patients. This analysis is of course not satisfactory because it does not take into account the occurrence of death and censoring.