The Lyman and a current parallel model: Are they equivalent in predicting radiation induced lung toxicity

Estimation of the probability of radiation pneumonitis (NTCP) is useful in evaluating 3DCRT of lung cancer and in developing criteria for dose escalation. The Lyman equation is often used and a biology-based four-parameter parallel architecture model has recently been proposed. In both models, calculated NTCP increases monotonically with a two-parameter auxiliary function - the effective dose (d_eff) in the Lyman equation and the fractional damage (f_dam) in the parallel model. Scatter graphs generated from clinical treatment plans show a roughly linear relationship f_dam and d_eff. It has been suggested that the two models are equivalent for practical use; i.e. merit descriptors derived from lung DVHs using either model have the same predictive ability for toxicity probability. However, based on calculations performed for partial organ irradiation, we show that the predictions of the two models may differ. Specifically, as the prescription dose increases, the slope of f_dam vs d_eff decreases. To validate this hypothesis with clinical data, we calculated d_eff and f_dam from the DVHs of 89 lung cancer patients treated with 3DCRT to prescription doses ranging from 45 Gy to 81 Gy. Multivariate analysis confirms that f_dam is significantly correlated with both d_eff and prescription dose. Given two patients with the same d_eff, the one treated to higher prescription dose tends to have smaller f_dam. Consequently, plans may be acceptable by one model criterion but not the other. Which model best predicts NTCP requires future follow-up data and correlation with clinical outcome.