Abstract
Background
Small‐cell lung cancer (SCLC) is one of the most aggressive types of lung cancer. The prognosis for SCLC patients depends on many factors. The intent of this study was to construct a nomogram model to predict mortality for extensive‐stage SCLC.
Methods
Original data was collected from the Surveillance, Epidemiology, and End Results (SEER) database of the National Cancer Institute in the United States. A nomogram prognostic model was constructed to predict death probability for extensive‐stage SCLC.
Results
A total of 16 554 extensive‐stage SCLC patients from 2004 to 2014 in the SEER database were included in this study. Gender, race, age, TNM staging (including tumor extent, nodal status, and metastasis), and treatment (surgery, chemotherapy, and radiotherapy) were identified as independent predictors for lung cancer‐specific death for extensive‐stage SCLC patients. A nomogram model was constructed based on multivariate models for lung cancer related death and other cause related death. Performance of the two models was validated by calibration and discrimination, with C‐index values of 0.714 and 0.638, respectively.
Conclusion
A prognostic nomogram model was established to predict death probability for extensive‐stage SCLC. This validated prognostic model may be beneficial for treatment strategy choice and survival prediction.
Keywords: Small cell lung cancer, nomogram, mortality
Introduction
Small‐cell lung cancer (SCLC) is one of the most lethal and aggressive types of lung cancer, with the majority of patients first diagnosed as extensive disease.1 Platinum‐based chemotherapy has remained the first line standard treatment for extensive‐stage SCLC for decades.2, 3 However, despite the initial sensitive response to chemotherapy, most patients relapse after three months of treatment, or even show signs of cancer progression during treatment. To date, targeted therapy and immunotherapy have only made modest improvements in SCLC patients’ prognosis. The overall survival (OS) for extensive‐stage SCLC is poor and the prognosis for these patients is highly variable.
The American Joint Committee on Cancer (AJCC) TNM classification and Veterans Administration Lung Study Group (VALSG) system are commonly used for the prognosis of small cell lung cancer patients. AJCC TNM classification stratifies the disease by primary tumor (T stage), lymph node involvement (N stage), and distant metastasis (M stage)4, while VALSG divides tumors into limited disease (LD) and extensive disease (ED). However, even for patients at the ED stage, their survival varies dramatically. Previous studies have indicated that clinical pathological characteristics might be related to the survival of SCLC patients.5, 6, 7 However, the predictive value of these clinical factors remains unclear.
A nomogram is a mathematical model to predict certain endpoints, such as disease progression or death8, based on several parameters. Several studies have constructed nomogram models to stratify the survival probability of SCLC.9, 10, 11, 12 However, the prognosis and treatment strategy for limited and extensive‐stage SCLC patients are different. The aims of the current study were to construct and validate a nomogram model utilizing SEER database data involving clinical‐pathological characteristics, and to predict survival in extensive‐stage SCLC patients using this model. Compared to published models, our nomogram model had the following advantages: (i) the model was developed using a relatively large cohort; (ii) the model was developed to focus uniquely on extensive‐stage SCLC; and (iii) the model used the more accurate AJCC TNM staging criteria to classify SCLC patients.
Methods
Study patients
We utilized the SEER program (https://seer.cancer.gov/seerstat, 23 March 2017) of the National Cancer Institute to select the population for this study. Patients diagnosed between 2004 and 2014 with SCLC as a first primary malignancy were selected for the study (SEER datasets opened 2004–2014 period data for researchers are freely available). Patients with the following International Classification of Diseases for Oncology, Third Edition (ICD‐O‐3), histology codes were included in the study population: 8002/3: malignant tumor, small cell type; 8041/3: small cell carcinoma, NOS; 8042/3: oat cell carcinoma; 8043/3: small cell carcinoma, fusiform cell; 8044/3: small cell carcinoma, intermediate cell; 8045/3: combined small cell carcinoma; the ICD‐O‐3 site codes: lung and bronchus including main bronchus (C34.0); upper lobe, lung (C34.1); middle lobe, lung (C34.2); lower lobe, lung (C34.3); overlapping lesion of lung (C34.8); and lung, NOS (C34.9).
Extensive‐stage SCLC patients (AJCC staging IV) were selected for further analysis. Cancers diagnosed by autopsy or as the result of death certificates only were excluded. Other exclusion criteria for this study included collaborative stage (CS), tumor size code of 990 (microscope), or 996–999 (unknown); CS tumor extent code of 950, 980, or 999 (unknown extent); CS metastasis code 75 (stated as M1, NOS) or 99 (unknown); lymph node involvement (unknown or lymph node involvement, not otherwise specified).
The analysis of the SEER patients used de‐identified summary‐level data and required no ethical approval.
Statistical analysis
Cause‐specific mortality was the primary endpoint. Consistent with the COD code, we classified cause of death as cancer‐specific death and death from other causes. Cancer‐specific death was defined as those that met the following criteria: (i) Dead = attributable to this cancer dx; and (ii) COD TO SITE RECORD = lung and bronchus; death from other causes defined as cause specific death = dead of other cause. Covariates included in the prediction model were selected based on known clinically prognostic factors and availability in the SEER database and included: age, gender, race (white, black or others/unknown), marriage (married or other status), anatomic sites (upper, middle, lower, bronchus or others), primary tumor location (left or right), tumor size, extent of tumor (local or regional), nodal status (N0, N1, N2 or N3), metastasis (m1a or m1b), grading (well, moderately, poorly, undifferentiated or NOS), surgery (yes or no), chemotherapy (yes or no), and radiotherapy (yes or no). The cumulative incidence function (CIF) was used to describe the probability of death, and the CIF difference between category groups was determined by Gray's test.13 When CIF was evaluated, the age at diagnosis and tumor size were divided into three groups. The age at diagnosis was regrouped as less than 60, 60 to 75, and above 75 years. Tumor size was regrouped as less than 3 cm, 3–5 cm and above 5 cm.
Fine and Gray proportional hazards regression was performed to predict 3, 6 and 12 month probabilities of two competing mortality outcomes.14 The continuous variables were fitted and modeled using restricted cubic splines with three knots at the 10%, 50%, and 90% empirical quantiles, and interactions were not evaluated. We used a model selection technique based on Bayesian information criteria when establishing competing risks models to avoid overfitting. Discrimination and calibration were both measured to internally validate the performance of the nomogram. Discrimination was indicated by the Harrell C index, and was defined as the ability of the model to separate subject outcomes.15 With respect to competing risks, an ordered pair was considered as evaluable if the first patient experienced the event of interest at a time when the second patient was still at risk. The C‐index of the model ranged from 0.5 to 1.0 (0.5 represented random chance, while 1.0 represented a perfectly discriminating model). Calibration referred to the agreement between predictions and observed outcomes. Calibration was evaluated at 3, 6, and 12 months with a calibration plot in which we compared the final reduced model‐predicted probability of death with the observed cumulative incidence of death. The model‐predicted probabilities were averaged within quintiles defined by the magnitude of the predictions. The marginal cumulative incidence of death was calculated within each quintile of individuals using the method provided by Gray. Finally, a calibration curve was plotted representing the marginal estimate versus model average predictive probability. Predictions in a perfect calibration plot should fall on the 45‐degree diagonal line, which indicated the model predicted the actual survival very well. Bootstrapping with 1000 resamples was adopted in internal validation.
All statistical analyses were performed using R software (version 3.3.3; http://www.r-project.org). The R packages cmprsk, rms, and mstate were used for modeling and developing the nomogram. All reported significance levels were two‐sided, with statistical significance set at 0.05.
Results
Patient characteristics
We identified 18 027 eligible extensive‐stage SCLC patients from 2004 to 2014 in the SEER database. The flow chart for patient selection is in Figure 1, and the list of the patient demographics and clinical characteristics are displayed in Table 1. The majority of patients were Caucasian (87.1%) and over 60 years old (74.8%). Most received chemotherapy (69.4%) and did not undergo surgical procedures associated with their treatment (99.1%).
Figure 1.
Flow chart of study patients’ selection.
Table 1.
Cumulative incidence of death among patients with extensive‐stage SCLC
| Cancer‐specific death | Death from other causes | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Characteristics | N | % | Event | % | Six months | One year | Three years | P‐value | Six months | One year | Three years | P‐value |
| Total | 18 027 | 16 554 | 51.2 | 75.7 | 92.3 | 3.1 | 4.0 | 5.0 | ||||
| Demographic characteristics | ||||||||||||
| Age at diagnosis, years | <0.001 | <0.001 | ||||||||||
| <60 years | 4551 | 25.2 | 4102 | 24.8 | 41.1 | 70.6 | 92.5 | 1.8 | 2.8 | 3.7 | ||
| 60–75 years | 9785 | 54.3 | 8953 | 54.1 | 50.0 | 75.4 | 92.1 | 3.0 | 4.0 | 5.0 | ||
| >75 years | 3691 | 20.5 | 3499 | 21.1 | 67.2 | 82.8 | 92.4 | 4.7 | 5.6 | 6.2 | ||
| Gender | <0.001 | 0.014 | ||||||||||
| Female | 8453 | 46.9 | 7690 | 46.5 | 49.1 | 73.4 | 92.0 | 2.7 | 3.6 | 4.5 | ||
| Male | 9574 | 53.1 | 8864 | 53.5 | 53.2 | 77.7 | 92.5 | 3.4 | 4.5 | 5.3 | ||
| Race | <0.001 | <0.001 | ||||||||||
| White | 15 709 | 87.1 | 14 465 | 87.4 | 51.7 | 76.3 | 92.7 | 2.9 | 3.9 | 4.7 | ||
| Black | 1604 | 8.9 | 1459 | 8.8 | 48.6 | 72.9 | 90.7 | 3.6 | 4.4 | 6.1 | ||
| Others | 714 | 4.0 | 630 | 3.8 | 47.7 | 67.8 | 86.3 | 5.2 | 6.1 | 8.4 | ||
| Marriage | <0.001 | 0.010 | ||||||||||
| Married | 9134 | 50.7 | 8378 | 50.6 | 48.2 | 74.3 | 92.4 | 2.8 | 3.7 | 4.6 | ||
| Others | 8893 | 49.3 | 8176 | 49.4 | 54.3 | 77.2 | 92.1 | 3.3 | 4.4 | 5.4 | ||
| Disease characteristics | ||||||||||||
| Anatomic sites | 0.009 | 0.203 | ||||||||||
| Upper | 9369 | 52.0 | 8563 | 51.7 | 50.4 | 74.9 | 92.3 | 3.0 | 3.8 | 4.7 | ||
| Middle | 746 | 4.1 | 695 | 4.2 | 50.3 | 73.2 | 92.2 | 2.9 | 4.1 | 5.1 | ||
| Lower | 4011 | 22.2 | 3695 | 22.3 | 51.4 | 76.2 | 92.0 | 3.3 | 4.7 | 5.6 | ||
| Bronchus/others | 3901 | 21.6 | 3601 | 21.8 | 53.3 | 77.6 | 92.5 | 3.0 | 3.9 | 4.8 | ||
| Primary tumor location | 0.702 | 0.720 | ||||||||||
| Left‐sided | 7764 | 43.1 | 7131 | 43.1 | 51.2 | 75.9 | 92.4 | 3.1 | 4.1 | 4.9 | ||
| Right‐sided | 10 263 | 56.9 | 9423 | 56.9 | 51.3 | 75.5 | 92.2 | 3.1 | 4.0 | 5.0 | ||
| Tumor size (cm) | <0.001 | 0.319 | ||||||||||
| ≤3 cm | 4747 | 26.3 | 4351 | 26.3 | 48.2 | 72.4 | 91.1 | 3.3 | 4.2 | 5.3 | ||
| 3–5 cm | 5016 | 27.8 | 4634 | 28.0 | 51.8 | 76.3 | 92.3 | 3.1 | 4.2 | 5.0 | ||
| >5 cm | 8264 | 45.8 | 7569 | 45.7 | 52.6 | 77.2 | 92.9 | 2.9 | 3.8 | 4.7 | ||
| Tumor extension | <0.001 | 0.487 | ||||||||||
| Local | 7741 | 42.9 | 7023 | 42.4 | 48.6 | 73.9 | 91.5 | 3.1 | 4.1 | 5.1 | ||
| Regional | 7304 | 40.5 | 6599 | 39.9 | 51.1 | 75.7 | 92.4 | 3.0 | 3.9 | 4.8 | ||
| Distant | 2982 | 16.5 | 2932 | 17.7 | 58.4 | 80.2 | 93.7 | 3.3 | 4.2 | 4.8 | ||
| Nodal status | 0.002 | 0.366 | ||||||||||
| N0 | 2338 | 13.0 | 2136 | 12.9 | 52.1 | 72.2 | 90.4 | 3.2 | 3.8 | 4.9 | ||
| N1 | 1246 | 6.9 | 1148 | 6.9 | 50.1 | 73.7 | 90.7 | 3.3 | 4.5 | 5.8 | ||
| N2 | 10 695 | 59.3 | 9899 | 59.8 | 52.0 | 76.5 | 92.7 | 3.0 | 4.1 | 4.9 | ||
| N3 | 3748 | 20.8 | 3371 | 20.4 | 48.9 | 76.4 | 92.8 | 3.2 | 3.8 | 4.8 | ||
| Metastasis | <0.001 | <0.001 | ||||||||||
| M1a | 1514 | 8.4 | 1326 | 8.0 | 38.6 | 60.6 | 83.7 | 4.7 | 5.9 | 7.4 | ||
| M1b | 16 513 | 91.6 | 15 228 | 92.0 | 52.4 | 77.1 | 93.1 | 2.9 | 3.9 | 4.7 | ||
| Grading | 0.024 | 0.846 | ||||||||||
| Well or moderately | 82 | 0.5 | 74 | 0.4 | 37.0 | 65.1 | 89.3 | 3.7 | 3.7 | 3.7 | ||
| Poorly | 1464 | 8.1 | 1342 | 8.1 | 50.2 | 74.5 | 91.9 | 2.9 | 4.1 | 4.9 | ||
| Undifferentiated | 3399 | 18.9 | 3193 | 19.3 | 50.2 | 74.7 | 91.8 | 3.0 | 4.4 | 5.3 | ||
| NOS | 13 082 | 72.6 | 11 945 | 72.2 | 51.7 | 76.2 | 92.4 | 3.1 | 3.9 | 4.9 | ||
| Treatment characteristics | ||||||||||||
| Surgery | <0.001 | 0.566 | ||||||||||
| Yes | 171 | 0.9 | 141 | 0.9 | 35.8 | 57.8 | 81.1 | 2.4 | 3.7 | 5.2 | ||
| No | 17 856 | 99.1 | 16 413 | 99.1 | 51.4 | 75.9 | 92.4 | 3.1 | 4.0 | 5.0 | ||
| Chemotherapy | <0.001 | 0.015 | ||||||||||
| Yes | 12 513 | 69.4 | 11 199 | 67.7 | 37.1 | 69.4 | 91.6 | 2.2 | 3.5 | 4.7 | ||
| No | 5514 | 30.6 | 5355 | 32.3 | 83.4 | 90.0 | 93.9 | 4.9 | 5.3 | 5.4 | ||
| Radiotherapy | <0.001 | <0.001 | ||||||||||
| Yes | 7587 | 42.1 | 6752 | 40.8 | 40.3 | 68.4 | 91.1 | 2.0 | 3.0 | 4.3 | ||
| No | 10 440 | 57.9 | 9802 | 59.2 | 59.2 | 81.0 | 93.1 | 3.9 | 4.8 | 5.5 | ||
The median follow‐up time was 5 months (interquartile range 1–10 months). A total of 16 554 (91.8% of 18 027) patients died during the follow‐up period: 15683 (94.7% of 16 554) from cancer‐specific death and 871 (5.3% of 16 554) from causes other than their cancer. Of those who died of other causes, the most common causes were heart disease (28.0%), chronic obstructive pulmonary disease (COPD). (11.4%), cerebrovascular diseases (3.8%), accidents and adverse effects (3.3%).
Probability of death
The cumulative incidence function curves are plotted in Figure 2. 15 683 patients died from cause‐specific death and 871 from causes other than their cancer. The probabilities of cancer‐specific death and death from other causes were as follows: 6 months, 51.2% and 3.1%, respectively; one year, 75.7% and 4.0%, respectively; three years, 92.3%% and 5.0%, respectively. Based on univariate analysis, the probability of death increased with age (P < 0.001) for patients who died from lung cancer‐specific death and other causes. Male and Caucasian patients exhibited higher cumulative incidences of death compared with their counterparts (Table 1) for both groups. For patients who died from lung cancer‐specific deaths, anatomic sites, tumor size, nodal status, tumor extension, and metastasis staging and differentiation showed significant associations with the probability of death. Surgery (P < 0.001), chemotherapy (P < 0.001), and radiotherapy (P < 0.001) significantly lowered the cumulative incidence of death among patients who received these treatments. For patients who died from other causes, only metastasis staging was significantly associated with death probability. Chemotherapy (P = 0.015) and radiotherapy (P < 0.001) also decreased the cumulative incidence of death from other causes. The estimates of the crude cumulative incidence of cause‐specific and other causes of death by age at diagnosis, gender, race, marriage, anatomic sites, primary tumor location, tumor size, extent of tumor, nodal status, metastasis, grading, and treatment can be found in Table 1.
Figure 2.
Cumulative incidence estimates of death according to patient characteristics (solid line indicates cause‐specific death; dotted lineindicates other causes of death).
The results of competing risk model found that age at diagnosis, tumor size, gender, race, extent of tumor, nodal status, metastasis, surgery, chemotherapy, and radiotherapy could strongly predict cause‐specific death (Table 2). Extensive‐stage SCLC patients with advanced age, larger tumor size, distant and lymph node metastasis, and those both male and Caucasian had a higher probability of death as a result of lung cancer. Patients who underwent surgery, chemotherapy or radiotherapy had a lower rate of lung cancer‐specific mortality. For those patients who died from other causes, the tumor related factors such as tumor size, tumor extension, nodal status, and treatment (chemotherapy and surgery) were no longer associated with death from other causes.
Table 2.
Proportional subdistribution hazard models of probabilities of cancer‐specific death and death from other causes for patients with extensive‐stage SCLC
| Cancer‐specific death | Death from other causes | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Full model | Reduced model | Full model | Reduced model | |||||||||
| Characteristics | β | sdHR (95%CI) | P | β | sdHR (95%CI) | P | β | sdHR (95%CI) | P | β | sdHR (95%CI) | P |
| Age | 0.004 | 1.004 (1.001–1.008) | 0.014 | 0.004 | 1.004 (1.000–1.007) | 0.024 | 0.026 | 1.026 (1.009–1.044) | 0.003 | 0.027 | 1.027 (1.01–1.044) | 0.002 |
| Age’ | 0.005 | 1.005 (1.001–1.010) | 0.029 | 0.005 | 1.005 (1.001–1.010) | 0.021 | −0.013 | 0.987 (0.969–1.006) | 0.180 | −0.012 | 0.988 (0.97–1.007) | 0.210 |
| Tumor size | 0.034 | 1.034 (1.021–1.048) | <0.001 | 0.035 | 1.036 (1.022–1.049) | <0.001 | −0.020 | 0.981 (0.928–1.036) | 0.490 | –– | –– | –– |
| Tumor size’ | −0.025 | 0.975 (0.964–0.988) | <0.001 | −0.026 | 0.975 (0.963–0.987) | <0.001 | −0.001 | 0.999 (0.944–1.059) | 0.980 | –– | –– | –– |
| Male | 0.078 | 1.081 (1.045–1.118) | <0.001 | 0.071 | 1.074 (1.039–1.110) | <0.001 | 0.229 | 1.257 (1.092–1.447) | 0.002 | 0.219 | 1.245 (1.083–1.432) | 0.002 |
| Race | ||||||||||||
| Black | −0.111 | 0.895 (0.844–0.949) | <0.001 | −0.105 | 0.900 (0.849–0.955) | <0.001 | 0.249 | 1.283 (1.031–1.596) | 0.026 | 0.236 | 1.266 (1.018–1.574) | 0.034 |
| Others | −0.258 | 0.772 (0.703–0.849) | <0.001 | −0.260 | 0.771 (0.702–0.847) | <0.001 | 0.605 | 1.832 (1.408–2.383) | <0.001 | 0.590 | 1.804 (1.388–2.344) | <0.001 |
| Married | −0.035 | 0.965 (0.933–0.999) | 0.046 | –– | –– | –– | −0.201 | 0.818 (0.711–0.940) | 0.005 | −0.197 | 0.821 (0.715–0.943) | 0.005 |
| Anatomic sites | ||||||||||||
| Middle | −0.019 | 0.981 (0.898–1.071) | 0.670 | –– | –– | –– | 0.042 | 1.043 (0.745–1.459) | 0.810 | –– | –– | –– |
| Lower | −0.033 | 0.967 (0.926–1.010) | 0.130 | –– | –– | –– | 0.133 | 1.142 (0.970–1.346) | 0.110 | –– | –– | –– |
| Bronchus/others | 0.032 | 1.033 (0.989–1.079) | 0.140 | –– | –– | –– | 0.015 | 1.015 (0.852–1.209) | 0.870 | –– | –– | –– |
| Right‐sided | −0.006 | 0.994 (0.961–1.029) | 0.750 | –– | –– | –– | 0.010 | 1.010 (0.880–1.159) | 0.890 | –– | –– | –– |
| Tumor extension | ||||||||||||
| Regional | 0.050 | 1.051 (1.011–1.093) | 0.012 | 0.056 | 1.058 (1.018–1.099) | 0.004 | −0.020 | 0.981 (0.840–1.144) | 0.800 | –– | –– | –– |
| Distant | 0.140 | 1.150 (1.093–1.210) | <0.001 | 0.143 | 1.153 (1.096–1.214) | <0.001 | −0.018 | 0.982 (0.808–1.193) | 0.850 | –– | –– | –– |
| Nodal status | ||||||||||||
| N1 | 0.037 | 1.038 (0.959–1.123) | 0.360 | 0.036 | 1.036 (0.958–1.121) | 0.380 | 0.190 | 1.210 (0.903–1.621) | 0.200 | –– | –– | –– |
| N2 | 0.126 | 1.135 (1.076–1.197) | <0.001 | 0.126 | 1.134 (1.076–1.196) | <0.001 | 0.011 | 1.011 (0.825–1.240) | 0.910 | –– | –– | –– |
| N3 | 0.118 | 1.125 (1.058–1.196) | <0.001 | 0.119 | 1.126 (1.059–1.198) | <0.001 | −0.027 | 0.973 (0.766–1.237) | 0.820 | –– | –– | –– |
| Metastasis M1b | 0.414 | 1.513 (1.424–1.608) | <0.001 | 0.413 | 1.511 (1.423–1.606) | <0.001 | −0.487 | 0.614 (0.505–0.748) | <0.001 | −0.479 | 0.619 (0.509–0.753) | <0.001 |
| Grading | ||||||||||||
| Poorly | 0.243 | 1.275 (1.008–1.612) | 0.043 | –– | –– | –– | 0.342 | 1.407 (0.442–4.479) | 0.560 | –– | –– | –– |
| Undifferentiated | 0.246 | 1.279 (1.015–1.612) | 0.037 | –– | –– | –– | 0.441 | 1.555 (0.495–4.881) | 0.450 | –– | –– | –– |
| NOS | 0.270 | 1.310 (1.042–1.648) | 0.021 | –– | –– | –– | 0.341 | 1.407 (0.450–4.392) | 0.560 | –– | –– | –– |
| Surgery | −0.307 | 0.736 (0.626–0.864) | <0.001 | −0.329 | 0.720 (0.612–0.846) | <0.001 | 0.189 | 1.208 (0.651–2.244) | 0.550 | –– | –– | –– |
| Chemotherapy | −0.801 | 0.449 (0.429–0.470) | <0.001 | −0.802 | 0.448 (0.428–0.469) | <0.001 | −0.066 | 0.936 (0.807–1.085) | 0.380 | –– | –– | –– |
| Radiotherapy | −0.162 | 0.850 (0.824–0.877) | <0.001 | −0.164 | 0.848 (0.822–0.876) | <0.001 | −0.158 | 0.854 (0.742–0.983) | 0.028 | −0.170 | 0.844 (0.734–0.97) | 0.017 |
Age’ and Tumor size’ are constructed spline variables (when k = 3). A model selection technique based on the Bayesian information criteria was used.
Nomogram
A nomogram was constructed in order to produce a user‐friendly model that would allow users to estimate the risk for lung cancer specific death and death from other causes by entering a patient's clinicopathologic characteristics and treatment information. The nomogram based on Fine and Gray's model can be found in Figure 3. For example, consider a 60 year‐old Chinese man classified using the American Joint Committee on Cancer (AJCC) stage IV (T3N2M1a) SCLC, who only received chemotherapy. The probabilities of death from lung cancer in 3, 6, and 12 months predicted by the nomogram model were 0.14, 0.24 and 0.42, respectively.
Figure 3.
Nomogram for predicting 3, 6, and 12 month probabilities of (a) lung cancer death and (b) other causes death in Extensive‐stage patients.
Model performance was internally validated for discrimination and calibration. Discrimination, as measured by the 1000 resample bootstrap‐corrected C‐index, was 0.714 (95% CI, 0.712–0.716) for the cancer‐specific death and 0.638 (95% CI, 0.628–0.649) for death from other causes. The calibration plot (Figure 4) showed good agreement between predicted and observed outcomes.
Figure 4.
Calibration plot of the nomogram in the original cohort. The x‐axis represents the mean predicted probability of the conditional cumulative incidence model. The y‐axis represents observed cumulative incidence of death. The solid line represents equality between the predicted and observed probability.
Discussion
The treatment of extensive‐stage SCLC is relatively uniform. However, the prognosis for most patients is highly variable. Veterans Administration and AJCC TNM staging are currently used for the prognosis of SCLC; however, more accurate prognostic models integrating additional demographic and clinical parameters are needed. Several prognostic scoring systems have been developed for SCLC, including the Manchester Score16 and the Spain prognostic index.17 However, patients in both studies received treatment nearly 40 years ago (1979–1993). The intensive treatment regimens involved in both studies, including cyclophosphamide, etoposide, and methotrexate, are no longer used.
The nomogram has been considered a more accurate model to predict patient prognosis.15, 18 Our study developed a prognostic model for the prediction of lung cancer related death for extensive‐stage SCLC. Although several studies have identified clinical‐pathologic and treatment predictors for survival of SCLC.9, 10, 11, 12 to our knowledge, the current study is the first to develop a nomogram prediction model for extensive‐stage SCLC. Additionally, our study was based on a relatively large cohort of patients and involved various treatment strategies. This is the first nomogram model for SCLC based on the SEER database, which is one of the largest cancer‐related databases worldwide.
We identified age, gender, race, TNM staging (including tumor extent, nodal status, and metastasis), and treatment (surgery, chemotherapy and radiotherapy) as independent predictors for lung cancer‐specific death. Advanced age, male, and Caucasian ethnicity9, 10, 11, 12 have been reported as risk factors of SCLC survival. Chemotherapy also reduced the risk of death in previously published SCLC nomogram models.9, 10, 11, 12 Notably, our model demonstrated that radiotherapy and surgery also reduced the risk of death in our extensive‐stage SCLC cohort. Although previous clinical trials19 have found extensive‐stage SCLC patients do not benefit from radiotherapy and surgery, recent studies reported that thoracic radiotherapy improved survival and local control for extensive‐stage SCLC patients who responded to chemotherapy.20, 21, 22 Similarly, certain subgroups of extensive‐stage SCLC patients might still benefit from surgery, and this possibility requires further study.
We assessed the performance of a nomogram by calibration and discrimination. The C‐index indicated the predictive value of our nomogram. In this study, internal validation provided good discrimination power, and the unadjusted C‐index was 0.714. This nomogram performed similarly to, or better than, the published SCLC nomograms of Pan et al. (C‐index 0.68, 95% CI, 0.64–0.72)12, Wang et al. (C‐index 0.722 ± 0.004)11, Xiao et al. (C‐index 0.60; 95%CI, 0.55–0.65)9 and Xie et al. (C‐index 0.73).10 Therefore, the nomogram developed in our study possesses great potential to estimate death risk for extensive‐stage SCLC patients. This nomogram can be used in clinical practice to avoid over treatment. In randomized clinical trials, prognostic variables in this model can also be used for stratification.
There are several limitations of this study. Firstly, these models were validated by 1000 resamplings, and the use of an independent external cohort may be a better validation method. Secondly, we did not include certain known prognostic variables, such as weight loss, the level of neuron‐specific enolase (NSE), and level of lactate dehydrogenase (LDH). Thirdly, the TNM staging in this study utilized the sixth edition, rather than the currently used eighth edition TNM staging guildelines. Additionally, SCLC has a unique genome map,23 as demonstrated by next generation sequencing, and parameters such as gene mutation may be used to predict survival in future studies.
In conclusion, a novel predictive nomogram model based on a large database was constructed for prognosis of patients with extensive‐stage SCLC. This validated prognostic model may be beneficial for treatment strategy choice and survival prediction.
Disclosure
All grants were received from the government. All authors declare that they have no conflict of interest, agree to the submission, and declare no potential conflicts of interest.
Acknowledgments
This work was supported by the Science Foundation of Peking University Cancer Hospital(18‐02); Beijing Municipal Administration of Hospitals Incubating Program (PX2019038); Science Foundation of Peking University Cancer Hospital (2017–18).
Contributor Information
Jia Zhong, Email: zhong-jia817@163.com.
Ziping Wang, Email: wangzp2007@126.com.
References
- 1. Nicholson AG, Chansky K, Crowley J et al The International Association for the Study of Lung Cancer lung cancer staging project: Proposals for the revision of the clinical and pathologic staging of small cell lung cancer in the forthcoming eighth edition of the TNM classification for lung cancer. J Thorac Oncol 2016; 11 (3): 300–11. [DOI] [PubMed] [Google Scholar]
- 2. Rudin CM, Poirier JT. Small‐cell lung cancer in 2016: Shining light on novel targets and therapies. Nat Rev Clin Oncol 2017; 14 (2): 75–6. [DOI] [PubMed] [Google Scholar]
- 3. Alexandrov LB, Ju YS, Haase K et al Mutational signatures associated with tobacco smoking in human cancer. Science 2016; 354 (6312): 618–22. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Zarogoulidis K, Latsios D, Porpodis K et al New dilemmas in small‐cell lung cancer TNM clinical staging. OncoTargets Ther 2013; 6: 539–47. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Zhou T, He X, Fang W et al Pretreatment albumin/globulin ratio predicts the prognosis for small‐cell lung cancer. Medicine (Baltimore) 2016; 95 (12): e3097. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Takeuchi H, Ikeuchi S, Kitagawa Y et al Pretreatment plasma fibrinogen level correlates with tumor progression and metastasis in patients with squamous cell carcinoma of the esophagus. J Gastroenterol Hepatol 2007; 22 (12): 2222–7. [DOI] [PubMed] [Google Scholar]
- 7. Yin Y, Wang J, Wang X et al Prognostic value of the neutrophil to lymphocyte ratio in lung cancer: A meta‐analysis. Clinics (Sao Paulo) 2015; 70 (7): 524–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol 2008; 26 (8): 1364–70. [DOI] [PubMed] [Google Scholar]
- 9. Xiao HF, Zhang BH, Liao XZ et al Development and validation of two prognostic nomograms for predicting survival in patients with non‐small cell and small cell lung cancer. Oncotarget 2017; 8 (38): 64303–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Xie D, Marks R, Zhang M et al Nomograms predict overall survival for patients with small‐cell lung cancer incorporating pretreatment peripheral blood markers. J Thorac Oncol 2015; 10 (8): 1213–20. [DOI] [PubMed] [Google Scholar]
- 11. Wang S, Yang L, Ci B et al Development and validation of a Nomogram prognostic model for SCLC patients. J Thorac Oncol 2018; 13 (9): 1338–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Pan H, Shi X, Xiao D et al Nomogram prediction for the survival of the patients with small cell lung cancer. J Thorac Dis 2017; 9 (3): 507–18. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Gray RJ. A class of K‐sample tests for comparing the cumulative incidence of a competing risk. Ann Stat 1988; 16 (3): 1141–54. [Google Scholar]
- 14. Fine JP, Gray RJ et al A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999; 94 (446): 496–509. [Google Scholar]
- 15. Harrell FJ, Lee KL, Mark DB et al Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15 (4): 361–87. [DOI] [PubMed] [Google Scholar]
- 16. Cerny T, Anderson H, Bramwell V, Thatcher N, Blair V. Pretreatment prognostic factors and scoring system in 407 small‐cell lung cancer patients. Int J Cancer 1987; 39 (2): 146–9. [DOI] [PubMed] [Google Scholar]
- 17. Maestu I, Pastor M, Gómez‐Codina J et al Pretreatment prognostic factors for survival in small‐cell lung cancer: A new prognostic index and validation of three known prognostic indices on 341 patients. Ann Oncol 1997; 8 (6): 547–53. [DOI] [PubMed] [Google Scholar]
- 18. Karakiewicz PI, Briganti A, Chun FK et al Multi‐institutional validation of a new renal cancer‐specific survival nomogram. J Clin Oncol 2007; 25 (11): 1316–22. [DOI] [PubMed] [Google Scholar]
- 19. Lad T, Piantadosi S, Thomas P, Payne D, Ruckdeschel J, Giaccone G. A prospective randomized trial to determine the benefit of surgical resection of residual disease following response of small cell lung cancer to combination chemotherapy. Chest 1994; 106 (6 Suppl): 320S–3S. [DOI] [PubMed] [Google Scholar]
- 20. Slotman BJ, van Tinteren H, Praag JO et al Use of thoracic radiotherapy for extensive stage small‐cell lung cancer: A phase 3 randomised controlled trial. Lancet 2015; 385 (9962): 36–42. [DOI] [PubMed] [Google Scholar]
- 21. Giuliani ME, Atallah S et al Clinical outcomes of extensive stage small cell lung carcinoma patients treated with consolidative thoracic radiotherapy. Clin Lung Cancer 2011; 12 (6): 375–9. [DOI] [PubMed] [Google Scholar]
- 22. Yee D, Butts C, Reiman A et al Clinical trial of post‐chemotherapy consolidation thoracic radiotherapy for extensive‐stage small cell lung cancer. Radiother Oncol 2012; 102 (2): 234–8. [DOI] [PubMed] [Google Scholar]
- 23. Han JY, Lee YS, Kim BC et al Whole‐genome analysis of a patient with early‐stage small‐cell lung cancer. Pharmacogenomics J 2014; 14 (6): 503–8. [DOI] [PubMed] [Google Scholar]