Development and validation of a preoperative prognostic index independent of TNM stage in resected non-small cell lung cancer

Patients diagnosed with NSCLC who received surgical treatment at Ohara Memorial Kurashiki Central Hospital between January 2007 and December 2011 (cohort 1) and the Tazuke Kofukai Medical Research Institute, Kitano Hospital between January 2007 and December 2012 (cohort 2) were retrospectively reviewed. All patients met the following criteria: pathological confirmation of NSCLC; no preoperative treatment; complete resections, more radical than or equal to segmentectomy with lymph node dissections or samplings; no microscopic residual tumor; no evidence of active infection such as pneumonia prior to surgery; evaluation of ILD with a HRCT scan, which was performed at diagnosis of lung cancer and was available for review; no history of acute exacerbation of chronic obstructive pulmonary disease or ILD within a month prior to surgery; and availability of laboratory data, including NLR, CEA and CYFRA 21-1, and follow-up information. Chronic obstructive pulmonary disease was diagnosed on the basis of the Global Initiative for Chronic Obstructive Lung Disease [21]. ILD was diagnosed only by two pulmonologists, based on medical history, physical examination, and abnormalities compatible with bilateral lung fibrosis on HRCT, according to the guidelines of the American Thoracic Society in 2011 [22]. The cases were categorized into two groups according to their radiologic appearance on HRCT: (1) usual interstitial pneumonia (UIP) pattern: characterized by the presence of basal-dominant reticular opacities and predominantly basal and subpleural distribution of honeycomb lesions, with multiple equal-sized cystic lesions of 2 to 10 mm diameter with a thick wall; and (2) non-UIP pattern: characterized by the presence of basal-predominant ground glass opacities and infiltrative shadows inconsistent with UIP patterns. We classified histology of lung cancer according to the World Health Organization guidelines [23]. Lung cancer pathological stages were based on the seventh edition of TNM classification of malignant tumors [2]. The study protocol was approved by the ethical committee of the Tazuke Kofukai Medical Research Institute, Kitano Hospital and Ohara Memorial Kurashiki Central Hospital in accordance with the Declaration of Helsinki.

The following clinical characteristics were reviewed from the available clinical records: age, sex, smoking history, resected side, surgical procedure, pathological stage, pathological tumor status, pathological lymph node status, histology of lung cancer, neutrophil counts, lymphocyte counts, NLR, CEA levels, CYFRA 21-1 levels, HRCT findings, percent vital capacity (%VC), percent forced expiratory volume in the first second (FEV1%), and causes of death. DFS was measured from the date of surgery until the date of recurrence or death, or until the date the patient was last known to be disease free. OS was measured from the date of surgery until the date of death from any cause or until the date on which the patient was last known to be alive.

We estimated DFS and OS employing the Kaplan–Meier analysis [24]. Differences between survival curves were tested for statistical significance using the two-tailed log-rank test. We used the method of Holm to account for multiple testing [25]. Univariate and multivariate prognostic analyses were performed for DFS and OS outcomes using the Cox proportional hazards model. To devise a prognostic index, a multivariate proportional hazards model was employed to derive a final model of the variables that had a significant independent relationship with survival in cohort 1. We classified patients into the three risk groups: low-risk (5-year OS rate ≥ 80%), intermediate-risk (50% ≤5-year OS rate < 80%), and low-risk (5-year OS rate < 50%). Then, the devised prognostic index was validated in the data set of cohort 2. Receiver operating characteristic (ROC) curve analysis was used to determine the optimal cut-off values for NLR level, devised prognostic index, pathological stage; values with maximum joint sensitivity and specificity were selected. The DeLong’s test was used for comparison of the areas under the receiver operating characteristic curve (AUC). All statistical analyses were performed using R version 2.13.1 statistical software (R Foundation for Statistical Computing, Vienna, Austria). All P values are two-sided, and P values less than 0.05 were considered statistically significant.

A total of 604 patients in cohort 1 and 333 patients in cohort 2 were enrolled in this study. The clinicopathological characteristics of patients are shown in Table 1. Significant differences between the two cohorts were found in sex, FEV1%, %VC, surgical procedures, clinical stage, clinical N factor, pathological T factor, pathological N factor, ILD, and postoperative chemotherapy. The median follow-up duration was 57.0 months (range: 0.20–99.3 months) in cohort 1 and 48.3 months (range: 0.9–103.7 months) in cohort 2. The 5-year DFS rate and OS rate of patients in cohort 1 were 63.2% (pathological stage I, 72.1%; pathological stage II, 40.8%; pathological stage IIIA, 28.3%) and 76.6% (pathological stage I, 81.5%; pathological stage II, 63.3%; pathological stage IIIA, 58.1%), respectively. The 5-year DFS rate and OS rate of patients in cohort 2 were 64.6% (pathological stage I, 76.4%; pathological stage II, 39.9%; pathological stage IIIA, 30.2%) and 76.6% (pathological stage I, 84.6%; pathological stage II, 63.5%; pathological stage IIIA, 49.4%), respectively.

A ROC curve for NLR was employed to determine the cut-off value in the data set of cohort 1. The AUC for NLR was 0.63 (95% confidence interval [CI], 0.58–0.68) (Additional file 1: Figure S1). A NLR of 2.1 corresponded to the maximum joint sensitivity and specificity on the ROC curve (72.8% sensitivity and 51.7% specificity).

Univariate analysis showed ten significant risk factors for the OS in cohort 1 (Table 2). The multivariate analysis identified seven prognostic factors in cohort 1: age ≥ 70 years, ever-smokers, NLR ≥2.1, CYFRA 21-1 above the normal limit, non-UIP, and UIP.

To develop the prognostic index, we examined two prognostic indexes (prognostic index 1 and 2) composed of the seven independent factors identified in the multivariate analysis: age ≥ 70 years, ever-smokers, %VC <80%, NLR ≥2.1, CYFRA 21-1 above the normal limit, non-UIP, and UIP. In the prognostic index 1, we allocated four points for UIP, two points for ever-smokers and non-UIP, and one point for the other four factors according to the hazard ratios (HRs) found in the multivariate analysis, while in the prognostic index 2 we allocated two points for UIP and one point for the other six factors for the ease of calculation. The comparisons of the AUCs for the ROC curves revealed no significant differences between the prognostic indexes in cohort 1 (0.78 [95% CI, 0.74–0.82; the prognostic index 1] vs. 0.77 [95% CI, 0.73–0.81; the prognostic index 2]; P = 0.174, Additional file 2: Figure S2). Therefore, taking the ease of calculation in clinical practices into consideration, we adopted the prognostic index 2 (Additional file 3: Table S1). Five-year OS rates according to risk scores of the prognostic index are shown in Additional file 3: Table S2. The AUC for the prognostic index in cohort 2 was 0.76 (95% CI, 0.70–0.83).

Univariate Cox regression analyses showed that risk scores of the prognostic index, clinical T factor, clinical N factor were significantly associated with OS (risk scores of the prognostic index: HR 2.18; 95% CI, 1.92–2.48; P < 0.001, clinical T factor: HR 1.95; 95% CI, 1.58–2.41; P < 0.001, clinical N factor: HR 1.70; 95% CI, 1.33–2.18; P < 0.001; Additional file 3: Table S3). Multivariate cox regression analyses demonstrated that risk scores of the prognostic index were significantly related with OS (HR 2.03; 95% CI, 1.78–2.32; P < 0.001), independent of clinical T factor (HR 1.41; 95% CI, 1.10–1.80; P < 0.001), and clinical N factor (HR 1.54; 95% CI, 1.20–1.99; P < 0.001).

The AUC for risk scores of the prognostic index was 0.77 (95% CI, 0.73–0.81). The comparisons of the AUC in both cohorts revealed that the prognostic model presented a better diagnostic performance compared with the clinical staging system (0.77 vs. 0.62; P < 0.001; Fig. 1a), or with the pathological staging system (0.77 vs. 0.62; P < 0.001; Fig. 1b).

DFS and OS were evaluated according to the three risk groups in cohort 1 with Kaplan–Meier analysis (Fig. 2). Patients were then stratified according to the following three risk groups: low-risk (no points or one point, 223 patients [36.9%]), intermediate-risk (two or three risk points, 326 patients [54.0%]), and high-risk (more than three points, 55 patients [9.1%]). In cohort 1, the 5-year DFS rate resulted in 77.8%, 58.8%, and 22.6% for patients at low risk, intermediate risk, and high risk, and the 5-year OS rate of 95.2%, 70.4%, and 28.9% in each risk category, respectively. In the analysis of DFS and OS of all patients included in cohort 1, significant differences in outcomes among the three groups were seen (DFS: P < 0.001; intermediate vs. low; P < 0.001, high vs. low; P < 0.001, high vs. intermediate; P < 0.001, Fig. 2a) (OS: P < 0.001; intermediate vs. low; P < 0.001, high vs. low; P < 0.001, high vs. intermediate; P < 0.001, Fig. 2B). In addition, subgroup analyses were performed (pathological stage I [DFS, Fig. 2c; OS, Fig. 2d], pathological stage II and IIIA [DFS, Fig. 2e; OS, Fig. 2f], and adenocarcinoma [DFS, Fig. 2g; OS, Fig. 2h]).

Next, survivals were evaluated according to the three risk groups in cohort 2 (Fig. 3). In cohort 2, the 5-year DFS rate resulted in 78.3%, 61.8%, and 39.5% for patients at low risk (137 patients [41.1%]), intermediate risk (136 patients [40.8%]), and high risk (60 patients [18.0%]), and the 5-year OS rate of 87.8%, 78.4%, and 47.3% in each risk category, respectively. In the analysis of DFS and OS of all patients included in cohort 2, significant differences in outcomes among the three groups were seen (DFS: P < 0.001; intermediate vs. low; P = 0.005, high vs. low; P < 0.001, high vs. intermediate; P = 0.002, Fig. 3a) (OS: P < 0.001; intermediate vs. low; P = 0.015, high vs. low; P < 0.001, high vs. intermediate; P < 0.001, Fig. 3b). Subgroup analyses were performed, too (pathological stage I [DFS, Fig. 3c; OS, Fig. 3d], pathological stage II and IIIA [DFS, Fig. 3e; OS, Fig. 3f], and adenocarcinoma [DFS, Fig. 3g; OS, Fig. 3h]). All survival data, excluding DFS in pathological stage II and IIIA, were significantly stratified according to the three risk groups in both cohorts.

HRs of intermediate-risk group or high-risk compared to low-risk group were evaluated in cohort 1 (Additional file 3: Table S4) (DFS: intermediate vs. low; 2.13, 95% CI, 1.54–2.94; P < 0.001, high vs. low; 5.24, 95% CI, 3.41–8.03; P < 0.001, OS: intermediate vs. low; 6.20, 95% CI, 3.55–10.8; P < 0.001, high vs. low; 20.6, 95% CI, 10.9–38.7; P < 0.001). HRs adjusted for pathological stage and histology were evaluated in cohort 1 (DFS: intermediate vs. low; 2.04, 95% CI, 1.46–2.83; P < 0.001, high vs. low; 3.88, 95% CI, 2.38–6.34; P < 0.001, OS: intermediate vs. low; 5.69, 95% CI, 3.22–10.0; P < 0.001, high vs. low; 13.3, 95% CI, 6.68–26.4; P < 0.001).

In addition, HRs of intermediate-risk group or high-risk compared to low-risk group were also evaluated in cohort 2 (DFS: intermediate vs. low; 1.90, 95% CI, 1.19–3.02; P = 0.007, high vs. low; 3.90, 95% CI, 2.38–6.38; P < 0.001, OS: intermediate vs. low; 2.16, 95% CI, 1.13–4.13; P = 0.021, high vs. low; 7.71, 95% CI, 4.12–14.4; P < 0.001). HRs adjusted for pathological stage and histology were evaluated in cohort 2 (DFS: intermediate vs. low; 2.02, 95% CI, 1.26–3.24; P = 0.004, high vs. low; 4.92, 95% CI, 2.71–8.93; P < 0.001, OS: intermediate vs. low; 2.26, 95% CI, 1.16–4.41; P = 0.017, high vs. low; 11.3, 95% CI, 5.32–23.8; P < 0.001).