ELCC 2017 News: Development and Validation of Outcome Prediction Models Aids Effective Use of Radiotherapy in NSCLC

Patient data used for evaluation of outcome prediction model

Validation using patient data of an outcome prediction model for the development of dyspnoea subsequent to radiotherapy revealed that the prognostic factors in the model did not adequately predict for delta toxicity endpoints, according to results reported on 6 May, 2017 at the European Lung Cancer Conference (ELCC), in Geneva, Switzerland. The model was developed to identify patients at risk of dyspnoea, or breathing difficulty, after radiotherapy and the dataset comprised patients with non-small cell lung cancer (NSCLC) who underwent radiotherapy.

Dr. Gilles Defraene, Department of Oncology, KU Leuven - University of Leuven, in Leuven, Belgium, and colleagues evaluated the prognostic factors for lung toxicity contained within an outcome prediction model published by Appelt et al. (Acta Oncol 2014) that was identified in a literature search. The model is based on a review of radiation pneumonitis reports and retained the most important predictors for its development: Mean lung dose (MLD) as dosimetric factor, and 6 other factors that influence the susceptibility of a patient to this condition, including pre-existing pulmonary comorbidity, age >63 years, mid/inferior tumour location, and sequential chemotherapy as risk factors. The model identified current smoking and smoking history as protective factors.

The team retrospectively assessed a dataset of 109 patients with NSCLC who were treated at the MAASTRO Clinic using 1.8 Gy fraction doses administered in two fractions per day up to 79.2 Gy. All treatments had been performed using 3D-conformal radiotherapy techniques to be consistent with the study that was the basis of the model. The required parameters were retrospectively collected together with the dyspnoea endpoint, according to common toxicity criteria (CTC 3.0) scoring at baseline and at 6 months after radiotherapy.

Current smoking and pulmonary co-morbidities were prognostic of post radiotherapy dyspnoea but not for change in dyspnoea from baseline

Within 6 months after radiotherapy, 19.3% of patients had developed dyspnoea ≥2 per CTC scoring. Using logistic regression modelling on the dataset, current smoking and pulmonary comorbidity were confirmed as prognostic factors for this dyspnoea, the odds ratios (OR) were 0.28 for current smoking (p = 0.02), and OR 2.95 for presence of a pulmonary co-morbidity (p=0.02).

The OR for tumour location was outside of the reported 95% confidence interval (CI). The dosimetric factor in the published model, MLD, did not associate with outcome in any of the models employed in this study.

ELCC 2017 Abstract 74PD

Distribution of raw dyspnoea scores 6 months post RT with clearly lower scores in smokers (left). Changes in dyspnoea score from baseline (e.g. G2 at baseline which persists after RT is previously scored as G2 toxicity but now scored as G0 change) show similar normal distribution around zero, both for smokers and non-smokers (right). A ‘toxicity change’ approach could thus rule out a protective effect of smoking.
Credit: Dr. Gilles Defraene

An evaluation of the change in dyspnoea with respect to the baseline score (delta dyspnoea ≥1) which had a prevalence of 18.6%, revealed that current smoking and pulmonary co-morbidity were no longer significant prognostic factors, OR 0.56 (p = 0.27) and OR 0.47 (p = 0.21), respectively.

The reason for this different result is that both factors associated strongly with the baseline dyspnoea status. According to the authors, worse baseline dyspnoea is often a manifestation of existing comorbidities and may affect the probability of smoking cessation.

Conclusions

The authors underscored that validated outcome prediction models with high discriminative power are important for cost-effective use of proton therapy for locally-advanced NSCLC, stressed the importance of including the consideration of delta toxicity, or change from baseline, in the development of meaningful prognostic models for radiotherapy outcome.

Discussant point: Suresh Senan, Department of Radiation Oncology, Cancer Center Amsterdam, VUMC, Amsterdam, The Netherlands, discussed the study results in the poster discussion session and elaborated upon prediction of radiation pneumonitis from the aspects of baseline risk assessment, choice of treatment strategy and identification of patients to study new radiation techniques.

Disclosure

This project received funding from the EU under grant agreement No 601826 (REQUITE).

Reference

74PD - G. Dafraene, et al. Radiation-induced lung toxicity prediction modeling in NSCLC: importance of baseline toxicity scoring