Radiotranscriptomics signature-based predictive nomograms for radiotherapy response in patients with nonsmall cell lung cancer: Combination and association of CT features and serum miRNAs levels

Abstract

Purpose: We aimed to establish radiotranscriptomics signatures based on serum miRNA levels and computed tomography (CT) texture features and develop nomogram models for predicting radiotherapy response in patients with nonsmall cell lung cancer (NSCLC).

Methods: We first used established radioresistant NSCLC cell lines for miRNA selection. At the same time, patients (103 for training set and 71 for validation set) with NSCLC were enrolled. Their pretreatment contrast-enhanced CT texture features were extracted and their serum miRNA levels were obtained. Then, radiotranscriptomics feature selection was implemented with the least absolute shrinkage and selection operator (LASSO), and signatures were generated by logistic or Cox regression for objective response rate (ORR), overall survival (OS), and progression-free survival (PFS). Afterward, radiotranscriptomics signature-based nomograms were constructed and assessed for clinical use.

Results: Four miRNAs and 22 reproducible contrast-enhanced CT features were used for radiotranscriptomics feature selection and we generated ORR-, OS-, and PFS- related radiotranscriptomics signatures. In patients with NSCLC who received radiotherapy, the radiotranscriptomics signatures were independently associated with ORR, OS, and PFS in both the training (OR: 2.94, P < .001; HR: 2.90, P < .001; HR: 3.58, P = .001) and validation set (OR: 2.94, P = .026; HR: 2.14, P = .004; HR: 2.64, P = .016). We also obtained a satisfactory nomogram for ORR. The C-index values for the ORR nomogram were 0.86 [95% confidence interval (CI), 0.75 to 0.92] in the training set and 0.81 (95% CI, 0.69 to 0.89) in the validation set. The calibration-in-the-large and calibration slope performed well. Decision curve analysis indicated a satisfactory net benefit.

Conclusions: The radiotranscriptomics signature could be an independent biomarker for evaluating radiotherapeutic responses in patients with NSCLC. The radiotranscriptomics signature-based nomogram could be used to predict patients' ORR, which would represent progress in individualized medicine.

Keywords: CT texture features; miRNAs; nomogram; nonsmall cell lung cancer; radiotherapy response.

Figure 1

Flow diagram in this study. CT, computerized tomography; NSCLC, nonsmall cell lung cancer; PFS, progression‐free survival; ROI, region of interest; OS, overall survival

Figure 2

MiRNAs increase or decrease NSCLC cells’ radiosensitivity in vitro. n = 3 per group. mim1 = miR‐1290 mimics, mim2 = miR‐2861 mimics, inhib1 = miR‐25‐5p inhibitor, inhib2 = miR‐92a‐1‐5p inhibitor

Figure 3

Radiotranscriptomics features selection and validation for objective response rate (ORR). A, The least absolute shrinkage and selection operator (LASSO) coefficient profiles of the 22 most stable texture features and 4 miRNAs expression levels. B, Tuning parameter (λ) selection in the LASSO model used 10‐fold cross‐validation via minimum criteria. A λ value of 0.044 was chosen according to 10‐fold cross‐validation. C, The radiotranscriptomics scores were calculated and classified into radiosensitive and radioresistant groups in both training and validation sets. Radsen = radiosensitive patients, Radres = radioresistant patients. D, Receiver operating characteristic (ROC) curve analysis of radiotranscriptomics scores in both training and validation sets. AUC, area under curve; CI, confidence interval

Figure 4

Radiotranscriptomics features selection and validation for overall survival (OS). A, The LASSO coefficient profiles of the 22 most stable texture features and 4 miRNAs expression levels. B, Tuning parameter (λ) selection in the LASSO model used 10‐fold cross‐validation via minimum criteria. A λ value of 0.135 was chosen according to 10‐fold cross‐validation. C,D, Kaplan‐Meier curve analysis of OS based on radiotranscriptomics score in both training and validation sets. The green dots indicated censored observations

Figure 5

Radiotranscriptomics based nomogram development, assessment, and clinical use for objective response rate (ORR). A, The nomogram was developed in the training set incorporating age (1: ≤60, 2:> 60), sex (1: Women, 2: Men), differentiation (1: Well and Moderate, 2: Poor), T stage (1: T1 and T2, 2: T3 and T4), N stage (1: N0 and N1,2: N2 and N3), M stage(1:M0, 2:M1), chemotherapy (1: Nonplatinum drugs, 2: Platinum drugs only or both nonplatinum and platinum drugs), and ORR score. C,D, Calibration curves of the nomogram in both training and validation sets. D, Decision curve analysis for the nomogram model in both training and validation sets