Development and validation of a prediction model using molecular marker for long‐term survival in unresectable stage III non‐small cell lung cancer treated with chemoradiotherapy

INTRODUCTION

Lung cancer is the leading cause of cancer death worldwide, with non-small cell lung cancer (NSCLC) accounting for 85% of all cases.1 Approximately one third of patients with NSCLC have locally advanced disease at initial diagnosis.2 Definitive concurrent chemoradiotherapy (CRT) has been the backbone therapy for unresectable and medically inoperable stage III NSCLC and 15%–32% patients receiving CRT have been reported to survive at 5 years.3-5 Recently, the PACIFIC trial demonstrated durvalumab (Imfinzi, AstraZeneca) as consolidation therapy significantly improved the survival of patients who had no progression after CRT with 5-year overall survival (OS) of 42.9%.6, 7 Durvalumab was then licensed and became the new standard of care for patients in this disease setting.8

Due to the prominent heterogeneity of locally advanced NSCLC (LANSCLC), survival of patients varied widely and whether all the patients were suitable for consolidated immunotherapy remained unclear.9-11 Therefore, predicting survival and identifying patients at low or high risk of death after CRT were essential for individualized treatment and enhanced immunotherapy decisions. The American Joint Committee on Cancer (AJCC) TNM staging system was the gold standard for the survival risk classification, but was initially developed to evaluate operability rather than outcome after CRT. For prediction and risk stratification in LANSCLC patients, the solely TNM-based method might be more inaccurate. It was previously reported that other factors such as sex, histology and hematological indicators significantly impact on individual survival.12-14 Also, the prognostic value of epidermal growth factor receptor (EGFR) mutations in adenocarcinoma was increasingly being understood, which led to further molecular heterogeneity.15

Therefore, a dedicated prediction model integrating multiple factors for unresectable or inoperable stage III NSCLC patients was urgently needed. A nomogram has been acknowledged as a reliable tool with multivariate visualization to predict the prognosis of patients with malignancies.16, 17 To date, limited attempts to develop prognostic models for LANSCLC have been reported.14, 18 In this study, we aimed to build and validate a new nomogram incorporating clinical, treatment-related and molecular features of EGFR mutation to predict the 3- and 5-year OS, by exploring prognostic factors in a large population of LANSCLC patients treated with CRT. An independent cohort from the prospective clinical trial (NCT01494558) was used for external validation.19 In addition, based on the model, the cutoff values were determined to stratify patients into different risk subgroups according to the outcome.

METHODS Study cohort

This study was conducted with the approval of our institutional review board. Consecutive patients who received definitive CRT in our institution between January 1st, 2013 and December 31st, 2017 were retrospectively reviewed. As in the PACIFIC trial, consolidative durvalumab was administered for unresectable, stage III NSCLC patients without disease progression after concurrent CRT and the ongoing new series of trials also enrolled patients receiving sequential CRT.6, 7 The inclusion criteria were designed as follows: (1) patients aged 18 years or older, (2) initially diagnosed with stage III NSCLC by pathology and radiography, (3) unresectable or medically inoperable, (4) received concurrent or sequential chemotherapy, (5) completed a total radiation dose ≥50 Gy with intensity-modulated radiotherapy (IMRT) technique, and (6) received regular follow-up with thoracic and abdominal computed tomography (CT), brain magnetic resonance imaging (MRI) and bone emission computed tomography (ECT) or positron emission-computed tomography (PET). The exclusion criteria included: (1) patients who progressed or died during chemoradiotherapy, (2) were diagnosed with a second primary cancer, and (3) had incomplete clinical information. Tumor staging was evaluated according to the AJCC eighth edition TNM classification and staging system by two investigators retrospectively.

The patients included in the study were randomly stratified (2:1) into the training and testing groups. To examine the generalizability of the model, an independent external cohort from a prospective, randomized phase III trial (NCT01494558) was used for validation. Participants from this trial were diagnosed as unresectable or inoperable stage III NSCLC and treated with definitive CRT (thoracic radiotherapy of 60–66 Gy and platinum-based chemotherapy)19 and only patients meeting the inclusion criteria and with sufficient clinical data to score all factors in the established nomogram were included.

Data collection

Medical records were reviewed to obtain patient, tumor and treatment-related information and a standardized data form including all the factors was created to collect the data. Continuous factors were listed with the median and range, whereas categorical factors were summarized by the frequency and proportion. Patient-related factors included: sex, age, Karnofsky performance status (KPS) score, smoking history, pretreatment peripheral hematological indicators as neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR) and systemic immune-inflammation index (SII, calculated as platelet counts × neutrophil counts/lymphocyte counts). Tumor-related factors contained histology (including EGFR mutation status in nonsquamous NSCLC), tumor size (maximum diameter), clinical TNM stage, laterality and location (evaluated based on the lobe of the lungs). Regarding treatment-related factors, gross tumor volume (GTV), planning target volume (PTV), radiation dose, sequence of CRT, chemotherapy regimen and radiation pneumonitis (RP) grade were entailed.

According to treatment strategy, the GTV of radiotherapy (RT) included the primary disease as well as any involved regional lymph nodes, which were defined as those with a short-axis diameter of at least 1 cm on the CT scan, or with high fluorodeoxyglucose (FDG) uptake on PET-CT scan. The clinical target volume (CTV) was generated by expanding the GTV with 0.6–0.8 cm, as well as ipsilateral hilum and mediastinal nodal stations involved. The PTV was created by a uniform 0.5 cm expansion around the CTV. The median prescribed dose to PTV was 60 Gy in 30 fractions and ranged from 50 to 72 Gy in 25 to 35 fractions, median biologically equivalent dose (alpha/beta ratio 10 Gy, BED10) of which was 72 Gy. RT was given with 6-MV X-rays by linear accelerators and all patients received conventionally fractionated radiotherapy with one fraction per day and five fractions per week. Weekly cone beam computed tomography (CBCT) was acquired for registration throughout the course of radiotherapy. Chemotherapy of platinum-based double agents was administered every 3 weeks and the dominant regimens included etoposide, paclitaxel or pemetrexed combined with cisplatin or carboplatin. Follow-up data were collected by the medical records and imaging examinations as previously described. Telephone calls, medical insurance records and death certificates were also required. Overall survival was defined as the time from the date of primary treatment to the date of death.

Model construction and validation

Cox regression analyses were applied to select prognostic factors in the training group. Variables achieving p-values less than 0.1 by univariate analyses were entered into the multivariate analyses. The final model factors were selected using a backward stepdown process, with the Akaike information criterion as a stopping rule. Based on the results of multivariate analyses, the nomogram was created with Cox proportional hazards model to give the 3- and 5-year OS.

The evaluation of the nomogram comprised the assessment of discrimination and accuracy. Discrimination was calculated with a concordance index (C-index). The C-index value of 0.5 indicated a random probability and 1.0 indicated a perfect ability to discriminate outcome. Model accuracy was assessed by the calibration plot. The calibration slope and intercept could measure the agreement between predicted and observed outcomes and a perfect calibration plot would show a 45 upwards line. The internal validation was carried out in the training group with bootstrap resampling (1000 resamples) used. The external validation was implemented in the testing set and the external testing cohort from the prospective trial (NCT01494558). Cox regression analysis, conducted using each patient's total score as an independent factor, was used to evaluate the C-index and calibration plots. Comparisons between the model and the eighth edition AJCC TNM staging system were performed with integrated discrimination improvement (IDI) to quantify the difference on performance.20

Risk group stratification

In addition to comparing the C-index numerically, we sought to examine the risk discrimination ability of the model beyond traditional AJCC-TNM staging. By the X-tile analysis (Yale University, New Haven, CT, USA) on the model total scores of patients in the training group (from the highest to the lowest), cutoff values were determined to classify the patients into different risk groups.21 The cutoffs were then adopted to the testing group and external testing cohort. The Kaplan–Meier survival curves stratified by the risk level and TNM staging were delineated respectively.

Statistical analysis

Comparisons of the baseline parameters between the training and testing groups were conducted by Chi-square test or Mann–Whitney U test. Survival curves were estimated with the Kaplan–Meier method and compared with a log-rank test. All tests were two-sided, and p < 0.05 was defined as a statistically significant result. Statistical analysis was performed by SPSS software (version 25.0) and R (version 4.0.4) via R Studio software (version 1.4.1106). R packages “survival”, “time-ROC”, “rms”, and “shiny” were used. This study followed the TRIPOD statement.22

RESULTS Patient characteristics and survival

A total of 758 LANSCLC patients were treated with CRT from January 1st, 2013 to December 31st, 2017 in our institution and 533 patients were ultimately eligible for analysis based on the inclusion and exclusion criteria. In the whole population, there were 91 (17.1%) females and 442 (82.9%) males with the median age of 60 (range: 23–81). The majority of patients were smoker (76.9%) and had high performance score of KPS ≥ 80 (97.4%). Concerning histology, squamous cell carcinoma (SCC) was diagnosed in 324 (60.8%) patients and among 184 (34.5%) nonsquamous NSCLC patients, 38 (20.7%) patients carried mutant EGFR. The median tumor size was 4.4 (range: 0.9–13.4) cm. A total of 127 (23.8%) patients were classified as IIIA stage, whilst 121 (22.7%) were with the IIIC disease. Stratified by a 2:1 ratio, 356 patients were assigned to the training group and 177 to the testing group. The baseline characteristics of patients in the training and testing groups are shown in Table 1. Apart from PLR, no factor presented significant difference between the two groups (p > 0.05).

TABLE 1. The included characteristics of the training and testing sets Characteristic n (%) Training set (n = 356) Testing set (n = 177) p-value Patient characteristics Sex 0.302 Male 291 (81.7) 151 (85.3) Female 65 (18.3) 26 (14.7) Age (median, year) 60 (23–81) 60 (24–77) 0.415 KPS 0.166 70 10 (2.8) 4 (2.3) ≥80 346 (97.2) 173 (97.7) Smoking history 0.290 Non-smoker 87 (24.4) 36 (20.3) Smoker 269 (75.6) 141 (79.7) NLR (median) 2.2 (0.4–14.9) 2.3 (0.3–41.6) 0.949 PLR (median) 124.5 (26.6–377.4) 121.4 (25.3–937.8) 0.018 SII (median) 507.6 (56.7–5735.0) 534.2 (24.5–14452.1) 0.757 Tumor characteristics Histology 0.580 SCC 215 (60.4) 109 (61.6) NS EGFR mut− 54 (15.2) 28 (15.8) NS EGFR mut+ 27 (7.6) 11 (3.1) NS EGFR unknown 40 (11.2) 24 (6.7) NOS 20 (5.6) 5 (2.8) T stage 0.408 T1 35 (9.9) 20 (11.3) T2 99 (27.8) 41 (23.2) T3 77 (21.6) 48 (27.1) T4 145 (40.7) 68 (38.4) N stage 0.868 N0 8 (2.2) 3 (1.7) N1 26 (7.4) 10 (5.6) N2 166 (46.6) 84 (47.5) N3 156 (43.8) 80 (45.2) TNM stage 0.111 IIIA 84 (23.6) 43 (24.3) IIIB 200 (56.2) 85 (48.0) IIIC 72 (20.2) 49 (27.7) Laterality 0.763 Left 152 (42.7) 78 (44.1) Right 204 (57.3) 99 (55.9) Location 0.766 Upper/middle lobe 245 (68.8) 117 (66.1) Lower lobe 100 (28.1) 53 (29.9) Undefined 11 (3.1) 7 (4.0) Tumor size (median, cm) 4.4 (0.9–13.4) 4.5 (1.0–10.5) 0.681 Treatment characteristics GTV (ml) 80.7 (3.3–640.5) 69.2 (3.41–668.3) 0.190 PTV (ml) 429.6 (17.1–1195.3) 450.6 (51.4–1317.1) 0.550 RT dose (median, Gy) 60.0 (50.0–72.0) 60 (50.0–70.0) 0.515 CT sequence 0.213 Sequential 161 (45.2) 70 (39.5) Concurrent 195 (54.8) 107 (60.5) CT regimen 0.358 Etoposide-platinum 226 (63.5) 108 (61.0) Paclitaxel-platinum 98 (27.5) 46 (26.0) Pemetrexed-platinum- 32 (9.0) 23 (13.0) Radiation pneumonitis 0.240 ≤2 grade 332 (93.3) 170 (96.0) >2 grade 24 (6.7) 7 (4.0) Abbreviations: CT, chemotherapy; GTV, gross tumor volume; KPS, Karnofsky performance score; Mut, mutation; NLR, neutrophil-to-lymphocyte ratio; NOS, not otherwise specified; NS, nonsquamous; PLR, platelet-to-lymphocyte ratio; PTV, planning target volume; RT, radiotherapy; SII, systemic immune-inflammation index; SCC, squamous cell carcinoma.

All 533 patients included in the study had survival data and the Kaplan–Meier curve of the overall population was shown in Supplementary Figure S1. There were 298 events (deaths) over a median follow-up time of 39.6 (range: 4.9–80.8) months and the median survival was 30.6 (95% CI: 26.6–34.6) months. The 3- and 5-year OS for the enrolled patients was 44.2% and 29.6%, respectively.

Independent prognostic factors

According to the univariate analysis of training group, factors such as female (vs. male, p < 0.001), KPS ≥ 80 (vs. 70, p = 0.008) and non-smoker (vs. smoker, p = 0.011) were associated with better prognosis. Among all the histological types, nonsquamous NSCLC with mutant EGFR showed the survival superiority, followed by NOS, nonsquamous NSCLC without EGFR mutations and SCC. Clinical T and N component stage presented no significant correlation with OS with p value > 0.05, but the overall clinical TNM stage was an independent factors influencing OS, for patients diagnosed as IIIC stage had significant shorter survival in comparison with IIIA stage (HR = 1.642, 95% CI: 1.090–2.475, p = 0.018). In addition, metrical tumor size was a significant parameter for OS (p = 0.007), whilst tumor laterality and location were excluded (p > 0.05). With respect to treatment-related factors, PTV volume, RT dose, RP grade and chemotherapy sequence were associated with OS with p-values of <0.001, 0.044, 0.017 and <0.001, yet all pretreatment hematological inflammatory indices demonstrated no statistically significant correlation (p > 0.05).

All factors with p < 0.1 in univariate analyses were entered into Cox multivariate analyses. Sex, histology, PTV volume, chemotherapy sequence and RP grade retained independent significant factors in the multivariate analyses. The results of univariate and multivariate analyses for OS are listed in Table 2. The Kaplan–Meier curves stratified by these factors and the corresponding p-values are presented in Supplementary Figure S2.

TABLE 2. Univariate and multivariate analyses of the included characteristics for overall survival Characteristic Univariate analysis Multivariate analysis HR (95% CI) p-value HR (95% CI) p-value Patient characteristics Sex Male 1 - 1 - Female 0.460 (0.305–0.695) <0.001 0.451 (0.282–0.722) 0.001 Age (median, year) 1.008 (0.992–1.024) 0.351 KPS 70 1 - 1 - ≥80 0.356 (0.167–0.761) 0.008 1.866 (0.816–4.271) 0.139 Smoking history Non-smoker 1 - 1 - Smoker 1.589 (1.111–2.271) 0.011 1.196 (0.803–1.781)0. 0.380 NLR 1.044 (0.968–1.125) 0.263 PLR 1.001 (0.998–1.003) 0.661 SII 1.000 (1.000–1.000) 0.148 Tumor characteristics Histology SCC 1 - 1 - NS EGFR mut− 0.562 (0.364–0.867) 0.009 0.618 (0.396–0.963) 0.034 NS EGFR mut+ 0.298 (0.139–0.637) 0.002 0.371 (0.172–0.800) 0.011 NS EGFR unknown 1.115 (0.733–1.695) 0.611 1.582 (0.973–2.571) 0.064 NOS 0.479 (0.235–0.979) 0.044 0.598 (0.290–1.236) 0.165 T stage T1 1 - T2 0.908 (0.530–1.558) 0.727 T3 1.343 (0.775–2.326) 0.293 T4 1.360 (0.821–2.253) 0.232 N stage N0 1 - N1 0.772 (0.281–2.126) 0.617 N2 1.035 (0.421–2.544) 0.940 N3 0.970 (0.394–2.390) 0.947 TNM stage IIIA 1 - 1 - IIIB 1.105 (0.784–1.557) 0.569 1.059 (0.718–1.562) 0.774 IIIC 1.642 (1.090–2.475) 0.018 1.569 (0.989–2.489) 0.056 Laterality Left 1 - Right 0.915 (0.694–1.206) 0.528 Location Upper/middle lobe 1 - Lower lobe 1.161 (0.853–1.580) 0.342 Undefined 1.107 (0.489–2.509) 0.807 Tumor size 1.097 (1.026–1.174) 0.007 1.020 (0.942–1.104) 0.632 Treatment characteristics GTV 1.000 (0.999–1.002) 0.814 PTV 1.002 (1.001–1.002) <0.001 1.001 (1.001–1.002) 0.001 RT dose 0.956 (0.915–0.999) 0.044 0.955 (0.910–1.002) 0.062 CT sequence Sequential 1 - 1 - Concurrent 0.714 (0.541–0.941) 0.017 0.594 (0.436–0.809) 0.001 CT regimen Etoposide-platinum 1 - Paclitaxel-platinum 1.390 (0.896–2.157) 0.142 Pemetrexed-platinum- 0.532 (0.195–1.456) 0.219 Radiation pneumonitis ≤2 grade 1 - 1 - >2 grade 2.798 (1.715–4.563) <0.001 3.319 (1.989–5.536) <0.001 Abbreviations: CI, confidence interval; CT, chemotherapy; GTV, gross tumor volume; HR, hazard ratio; KPS, Karnofsky performance score; Mut, mutation; NLR, neutrophil-to-lymphocyte ratio; NOS, not otherwise specified; NS, nonsquamous; PLR, platelet-to-lymphocyte ratio; PTV, planning target volume; RT, radiotherapy; SII, systemic immune-inflammation index; SCC, squamous cell carcinoma.

留言 (0)

沒有登入
gif