Brought to you by

Reply to: Prediction of survival by texture-based automated quantitative assessment of regional disease patterns on CT in idiopathic pulmonary fibrosis: methodological issues:

by Sang Min Lee, MD and Joon Beom Seo, MD (seojb@amc.seoul.kr)

Lee, S.M., Seo, J.B., Oh, S.Y. et al. Eur Radiol (2018) 28: 1293. https://doi.org/10.1007/s00330-017-5028-0

We thank the authors of this letter for their interest in our manuscript “Prediction of survival by texture-based automated quantitative assessment of regional disease patterns on CT in idiopathic pulmonary fibrosis” [1] and for their valuable comments. Below, we would like to briefly respond to the addressed questions.

To the best of our knowledge, a single best P-value threshold for the inclusion of variables in multivariate analysis is not clearly defined. Although we selected variables with a P-value less than 0.05 as the threshold value, we agree with the opinion that the threshold of less than 0.2 could be more appropriate for the subsequent multivariate analysis. As suggested, we performed another statistical analysis using the threshold P-value of 0.2. Interval change in TLC and interval change in SpO2 were additionally included (P = 0.175 and 0.173, respectively). In multivariate analysis with the backward elimination method, interval change in TLC exhibited a P-value of 0.875 in the first step and SpO2 showed a P-value of 0.646 in the third step. As a result, both variables were eliminated during the statistical process and the final results remained the same.

The second question concerns the verification of the proportional hazard assumption.
We agree with the comment and tested the proportional hazard assumption using the Goodness-of-fit (P > 0.05; range, 0.119 ─ 0.960). Based on this test result, the validity of the Cox proportional regression model was verified.

Lastly, the authors mentioned that there was no internal or external validation for our prediction model. We understand their concern and agree that without validation process our results should be interpreted with certain degree of caution. However, the primary aim of our study was to investigate whether automated quantification of regional disease patterns can predict survival of patients with idiopathic pulmonary fibrosis (IPF) and our results revealed a significant association between the CT parameters and survival in patients with IPF. There has been some overlap in the use the term “prediction” with “association” in the literature [2]. Furthermore, we have already stated in the discussion that the lack of external validation as a limitation in our study and that further validation may be necessary to confirm our results.