Projection of time-to-event distributions is necessary to obtain accurate estimation of life expectancy, or prediction of event times for economic models.  Parametric survival analysis techniques are typically used, and can represent a broad range of shapes.  In some cases, however, the best distributional fit fails to capture the variation in hazards over the entire time span, or it provides acceptable fit to the data but yields clinically implausible projections (e.g., constant hazard of death).  More flexible techniques, like piecewise exponential models, can overcome these issues but remain generally underused.  In piecewise models, the time axis is divided into contiguous segments with a common parametric distribution assumed within each segment, but values of the parameters are allowed to vary.  In addition to greater flexibility, this framework allows inclusion of time-dependent predictors and/or time-dependent effects.  Two important considerations are the number and placement of divisions on the time axis, and the choice of the common distribution.  Examination of the cumulative and log-cumulative hazards plots can assist with these issues.  For instance, the number/placement of divisions for a piecewise-exponential model could be determined visually such that the points within each division of the cumulative hazard plot follow a linear pattern.  The same can be done with log-cumulative hazard function for a piecewise-Weibull model.  Although piecewise-exponential models can be made progressively more flexible by increasing the number of segments to capture even very complex patterns, the assumption of a constant hazard for the last segment can be limiting for projection.  Thus, models based on Weibull distributions may be more appropriate, and possibly achieve similar fit with fewer segments.  The subjectivity involved in these decisions can be minimized by using numeric optimizing strategies (e.g., grid search for placement of divisions) and use of fit statistics to select distributions.