Establishment of a Nomogram-Based Prognostic Model (LASSO-COX Regression) for Predicting Progression-Free Survival of Primary Non-Small Cell Lung Cancer Patients Treated with Adjuvant Chinese Herbal Medicines Therapy: A Retrospective Study of Case Series

Abstract

Nowadays, Jin-Fu-Kang Oral Liquid, one of Chinese herbal medicines (CHMs) preparations, has been widely used as an adjuvant therapy for primary Non-Small Cell Lung Cancer (PNSCLC) patients with the syndrome of deficiency of both qi and yin (Qi-Yin deficiency pattern) based on Traditional Chinese Medicine (TCM) theory. However, we found insufficient evidence of whether long-term CHM treatment could prolong Progression-Free Survival (PFS) of PNSCLC patients by quantitative measurement. Thus, we established a nomograph-based prognostic model for predicting PNSCLC patients' PFS, involved in Qi-Yin deficiency pattern and be treated with supplementary formula for Jin-Fu-Kang Oral Liquid over 6 months, by using their electronic medical records, in order to provide feasibly designed research for preliminarily judging and evaluating PNSCLC patients' individualization prognosis with long-term CHM-treatment in theoretical epidemiology. In our retrospective study, a series of 197 PNSCLC cases were enrolled and divided into 2 datasets at the ratio of 5:4 by Kennard-Stone algorithm, as a result of 109 in training dataset and 88 in validation dataset. Besides, TNM stage, operation history, sIL-2R, CA724 were considered as 4 highly-correlated predictors for modeling based on LASSO-Cox regression. Additionally, we respectively used training dataset and validation dataset for establishment including internal validation and external validation, and the prediction performance of model was measured by concordance index (C-index), integrated discrimination improvement (IDI), and net reclassification indices (NRI). Moreover, we found that the model containing clinical characteristics and bio-features presented the best performance by pairwise comparison. Next, the result of sensitivity analysis proved its stability. Then, for preliminarily examination of its discriminative power, all eligible cases were divided into high-risk or low-risk progression by the cut-off value of 57, in the light of predicted Nomogram scores. Ultimately, a completed TRIPOD checklist was used for self-assessment of normativity and integrity in modeling. In conclusion, our model might offer crude probability of uncertainly individualized PFS with long-term CHMs therapy in the real-world setting, which could discern the individuals implicated with worse prognosis from the better ones. Nevertheless, our findings were prone to unmeasured bias caused by confounding factors, owing to retrospective cases series.