Bayesian Cure Rate Modeling of Local Tumor Control: Evaluation in Stereotactic Body Radiotherapy for Pulmonary Metastases

Publication date: Available online 15 December 2015
Source:International Journal of Radiation Oncology*Biology*Physics
Author(s): Rainer J. Klement, Michael Allgäuer, Nicolaus Andratschke, Oliver Blanck, Judit Boda-Heggemann, Karin Dieckmann, Marciana Duma, Iris Ernst, Michael Flentje, Ute Ganswindt, Peter Hass, Christoph Henkenberens, Detlef Imhoff, Henning K. Kahl, Robert Krempien, Fabian Lohaus, Ursula Nestle, Meinhard Nevinny-Stickel, Cordula Petersen, Vanessa Schmitt, Sabine Semrau, Florian Sterzing, Jan Streblow, Thomas G. Wendt, Andrea Wittig, Matthias Guckenberger
BackgroundThe majority of radiobiological models about prediction of tumor control probability (TCP) does not account for the fact that many events could remain unobserved because of censoring.Methods and MaterialsWe applied two fundamental Bayesian cure rate models to a sample of 770 pulmonary metastasis treated with stereotactic body radiotherapy at German, Austrian and Swiss institutions: (i) the model developed by Chen, Ibrahim and Sinha (the CIS99 model); (ii) a mixture model similar to the classic model of Berkson and Gage (the BG model). In the CIS99 model the number of clonogens surviving the radiation treatment follows a Poisson distribution while in the BG model only one dominant recurrence-competent tissue mass may remain. The dose delivered to the isocenter, tumor size and location, sex, age and pre-treatment chemotherapy were used as covariates for regression.ResultsMean follow-up time was 15.5 months (range 0.1-125). Tumor recurrence occurred in 11.6% of the metastases. Delivered dose, female sex, peripheral tumor location and having received no chemotherapy before RT were associated with higher TCP in all models. Parameter estimates of the CIS99 were consistent with the classical Cox proportional hazards model. The dose required to achieve 90% tumor control after 15.5 months was 146 (114-188) Gy10 in the CIS99 and 133 (101-164) Gy10 in the BG model; however, the BG model predicted lower tumor control at long (≳20 months) follow-up times and gave a suboptimal fit to the data compared to the CIS99 model.ConclusionsBiologically motivated cure rate models allow adding the time component into TCP modeling without being restricted to the follow-up period which is the case for the Cox model. In practice, application of such models to the clinical setting could allow for adaption of treatment doses depending on whether local control should be achieved in the short or longer term.

Teaser

We apply two Bayesian cure rate models to a sample of 770 pulmonary metastases treated with stereotactic body radiotherapy in order to predict tumor control probability while accounting for censored observations. Both models yielded stable posterior model parameter estimates, indicating their identifiability. In practice, application of such models to the clinical setting could allow for adaption of treatment doses depending on whether local control should be achieved in the short or longer term.


from Cancer via ola Kala on Inoreader http://ift.tt/1YkmQxy
via IFTTT