Conditional Survival Among Patients With Carcinoma of the Lung* Ray M. Merrill, PhD, MPH; Donald Earl Henson, MD; and Michael Barnes, PhD Objective: One- and 5-year probabilities of survival or death change once a patient has already survived > 1 year after diagnosis. The current paper reports these probabilities for lung cancer patients according to histologic subtype, stage, and age at diagnosis. Methods: Cumulative observed survival rates were calculated and compared among 95,283 patients with histologically confirmed lung cancer (diagnosed from 1983 to 1992 and followed through 1995) by the life-table method using population-based tumor registries participating in the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute. On the basis of the cumulative survival estimates, we derived the probability of death in the next year, conditioned on having already survived to the start of the year (annual hazards), and the probability of survival conditioned on having already survived > 1 year (conditional survival). These rates were reported according to histologic subtype, stage, and age groups. Results: At the time of diagnosis, annual hazard rates differ greatly among histologic subtypes. However, by 5 years after diagnosis, the rates become similar. Bronchioloalveolar carcinoma displays the lowest annual hazards and small-cell carcinoma displays the highest annual hazards. Stage-age subcategories within histologic subtypes continue to show large differences in annual hazard rates. Five-year conditional survival probabilities are also reported, providing survival information that is consistent to that obtained from the annual hazards. Conclusions: One- and 5-year prognosis for lung cancer patients is influenced by years already survived and histology, stage, and age at diagnosis. Annual hazards and conditional survival provides useful and more relevant information than conventional survival estimates for patients and their physicians. These statistics can be directly obtained from cumulative survival estimates and should be more widely reported. (CHEST 1999; 116:697 703) Key words: annual hazards; conditional survival; lung neoplasms; observed survival; outcome Abbreviations: NCI National Cancer Institute; SEER Surveillance Epidemiology and End Results The projection of short- and long-term survival for cancer patients is typically made at the time of diagnosis and is referred to as cumulative survival. For editorial comment see page 593 However, patients who have survived 1 year, 2 years, or longer after diagnosis have a different probability of surviving 1 year. 1 3 Probability estimates of *From the Department of Health Science (Drs. Merrill and Barnes), College of Health and Human Performance, Brigham Young University, Provo, UT; Division of Epidemiology (Dr. Merrill), Department of Family and Preventive Medicine, University of Utah College of Medicine, Salt Lake City, UT; and Cancer Biomarkers Research Group (Dr. Henson), Division of Cancer Prevention, National Cancer Institute, Rockville, MD. Manuscript received December 2, 1998; revision accepted April 6, 1999. Correspondence to: Ray M. Merrill, PhD, MPH, Assistant Professor of Biostatistics and Epidemiology, Department of Health Science, Brigham Young University, 213 RB, Provo, UT 84602; e-mail: Ray_Merrill@byu.edu survival or death that are conditioned on having already survived a given period of time can provide patients and their physicians with more relevant information for directing personal health-related decisions. In this report, such estimates are presented for 95,283 patients with invasive carcinoma of the lung and bronchus (hereafter referred to as lung). Lung cancer cases were obtained from the Surveillance, Epidemiology, and End Results (SEER) Program of the National Cancer Institute (NCI), 4 diagnosed from 1983 to 1992, and followed through 1995 for vital status and cause of death. Emphasis is placed on conditional estimates for histologic subtypes, stage, and age at diagnosis, which are important prognostic factors for lung cancer. 5 7 A primary aim of this report is to illustrate the simple method by which conditional survival estimates are derived from conventionally reported cumulative survival estimates and to encourage their consideration. Their ability to identify unique bene- CHEST / 116 / 3/ SEPTEMBER, 1999 697
fits of early lung cancer diagnosis and the extended effects of tobacco smoking, even among patients who have already survived several years, is presented. Materials and Methods The data used in this analysis were obtained from the nine tumor registries participating in the SEER Program of the NCI. 4 The NCI contracts with medically oriented nonprofit universities and state health departments to obtain data on cancer patients diagnosed in the SEER catchment areas. These areas cover five states (Connecticut, Iowa, New Mexico, Utah, and Hawaii) and four metropolitan areas (Detroit, MI; Atlanta, GA; San Francisco-Oakland, CA; and Seattle-Puget Sound, WA), and represent about 10% of the US population. The SEER Program is the primary source for national estimates of cancer incidence and survival. The tumor registries in the SEER Program routinely abstract records of all cancer patients in hospitals, clinics, and nursing homes; from private pathology laboratories and radiotherapy units; and from death certificates. Data collected by the SEER registries include demographic information, tumor characteristics, morphology, diagnostic information, extent of disease, and, since 1983, the first course of cancer-directed therapy initiated within 4 months after diagnosis. Vital status and cause of death are obtained by active follow-up of cancer cases. Lung tumors are identified using the International Classification of Diseases for Oncology codes C34.0 to C34.9. 8 There were 112,950 cases of invasive adenocarcinoma (with subcategories listed in Table 1), squamous-cell, small-cell, and large-cell carcinomas diagnosed in the nine SEER tumor registries from 1983 through 1992. Cases were followed through December 31, 1995 for vital status. Classification of lung tumors by histologic subtypes used in this analysis is given in Table 1. This classification is the same as reported by the SEER Program, 5 with a few minor exceptions represented in the table. The four conventional stage categories recorded by SEER were adopted because they have been consistently defined over the study period. 9 Ten years of diagnosed cases and 13 years of follow-up provided a sufficient number of cases to compute survival estimates among select subgroups of the population (eg, histology-stage combinations). The four terms used to describe disease stage are: localized (an invasive neoplasm confined to the organ of origin); regional (a Table 1 Classification of Invasive Adenocarcinoma and Squamous-Cell, Small-Cell, and Large-Cell Lung Carcinomas (ICD-O)* Adenocarcinoma Adenocarcinoma NOS and other specific adenocarcinomas (M- 8140 to M-8143, M-8290, M-8310, M-8320, M-8480, M-8490, M-8550, M-8571) Bronchioloalveolar carcinoma (M-8250, M-8251) Papillary adenocarcinoma (M-8050, M-8260) Adenosquamous carcinoma (M-8560) Bronchial gland carcinoma Adenoid cystic carcinoma (M-8200) Mucoepidermoid carcinoma (M-8430) Squamous-cell carcinoma (M-8032, M-8052, M-8070 to M-8076) Small-cell carcinoma (M-8034, M-8041 to M-8045, M-8246) Large-cell carcinoma (M-8012, M-8030, M-8031) *ICD-O International classification of diseases for oncology; NOS not otherwise specified. neoplasm that has extended beyond the limits of the organ of origin directly into surrounding organs or tissues, into regional lymph nodes, or both); distant (a neoplasm that has spread to parts of the body remote from the primary tumor); and unstaged (a neoplasm for which there is not sufficient information to assign a stage). 8 The SEER Program Comparative Staging Guide for Cancer provides a description of how the SEER stage groupings (localized, regional, distant, and unstaged) compare to the American Joint Committee on Cancer s TNM stage groupings. 9 Localized disease reflects TX, T0, T1, or T2 with unstated or no lymph node involvement. This corresponds to Stages IA and IB. Regional disease reflects TX, T0, T1, or T2 with regional lymph involvement, or T3 or T4 with regional, unstated, or no lymph node involvement. These correspond to Stages IIA, IIB, IIIA, and IIIB. Distant disease reflects M1 (Stage IV). Specific exclusions were as follows: 15,121 cases in which lung cancer was not the first primary malignant tumor, 814 cases identified through autopsy or death certificate, 659 cases of unknown race, 81 cases with unknown survival time, and 905 cases not microscopically confirmed. Bronchial gland carcinoma cases were also excluded because of small numbers (ie, 87). Thus, 95,283 cases of invasive adenocarcinoma, squamous-cell, smallcell, and large-cell carcinomas remained for analysis. Carcinoma in situ cases were not considered because of insufficient numbers to compute conditional estimates. Observed survival probabilities were calculated by the lifetable method 10 using the SEER Survival System (SEER*Stat for Windows 95/NT, version 1.1; National Cancer Institute; Rockville, MD). 11 Projected observed survival estimates that use the time of diagnosis as a reference are referred to as cumulative survival estimates. On the basis of such estimates, we derived survival probabilities conditioned on having already survived 1 year (conditional survival), as well as the probability of death in the next year conditioned on having survived to the start of the year (annual hazards). To illustrate how we obtain conditional survival estimates from the cumulative survival estimates, suppose we are interested in the population s 5-year lung cancer survival probability conditioned on already having survived 5 years. The estimate is obtained by dividing the cumulative survival at 10 years by the cumulative survival at 5 years. The 1-year lung-cancer survival estimates conditioned on already having survived 5 years after diagnosis are derived by dividing the cumulative survival estimates at 6 years by the cumulative survival estimates at 5 years. Subtracting this survival probability from 1 gives the probability of dying in the year conditioned on having already survived 5 years. This is referred to as an annual hazard. Both estimates of conditional survival and annual hazards are presented in this report to provide insight on the effect years already survived have on the prognosis for lung cancer patients. Results Lung cancer cases, estimated annual hazards, and estimated cumulative observed survival probabilities are presented for histologically confirmed cases of invasive adenocarcinoma and squamous-cell, smallcell, and large-cell carcinomas of the lung (Table 2). Patients with bronchioloalveolar carcinoma have the highest survival, as well as the lowest annual hazards. Patients with small-cell carcinoma experience the lowest survival rates and also have the highest annual hazards. Large differences in annual hazards appear across the histologic subtypes when conditioned on 698 Clinical Investigations
Table 2 Lung Cancer Cases, Estimated Conditional Probabilities of Death (Annual Hazards), and Estimated Cumulative Observed Survival by Year and Histology* No. of Years After Diagnosis Variables 0 1 2 3 4 5 6 7 8 9 10 Adenocarcinoma NOS and other specific adenocarcinomas lx 29,872 12,585 7,833 5,933 4,327 3,250 2,479 1,814 1,334 945 630 qx 57.74 37.48 23.94 17.39 13.90 11.02 11.38 10.59 9.60 9.90 7.30 Sx 42.26 26.42 20.09 16.60 14.29 12.72 11.27 10.08 9.11 8.21 7.60 Bronchioloalveolar carcinoma lx 2,684 1,890 1,532 1,320 1,032 793 625 475 363 269 184 qx 29.42 18.85 13.47 10.70 11.08 9.79 10.51 8.68 7.00 9.70 9.80 Sx 70.58 57.28 49.56 44.25 39.35 35.50 31.77 29.01 26.97 24.37 22.00 Papillary adenocarcinoma lx 688 397 265 198 140 107 80 59 43 31 18 qx 42.18 33.25 24.62 22.11 16.36 13.93 8.28 12.84 Sx 57.82 38.59 29.09 22.66 18.95 16.31 14.96 13.04 11.09 9.81 8.50 Adenosquamous carcinoma lx 1,516 762 493 377 289 222 174 133 99 66 38 qx 49.64 34.65 23.35 15.00 12.13 11.88 11.66 12.15 15.60 Sx 50.36 32.91 25.23 21.44 18.84 16.60 14.67 12.89 10.87 8.79 8.20 Squamous-cell carcinoma lx 29,931 12,942 7,544 5,429 3,970 2,982 2,247 1,667 1,220 858 567 qx 56.63 41.55 27.76 18.67 14.98 14.11 13.09 12.23 13.10 11.00 10.10 Sx 43.37 25.35 18.31 14.89 12.66 10.87 9.45 8.30 7.21 6.41 5.80 Small-cell carcinoma lx 20,172 6,956 2,332 1,416 936 672 503 360 242 159 98 qx 65.41 66.34 38.76 24.48 14.21 14.96 13.81 13.91 16.90 13.80 20.50 Sx 34.59 11.64 7.13 5.39 4.62 3.93 3.39 2.92 2.42 2.09 1.70 Large-cell carcinoma lx 10,420 3,489 1,966 1,440 1,046 799 619 489 369 259 172 qx 66.41 43.45 26.40 18.74 14.20 10.05 10.60 9.97 10.30 10.60 11.90 Sx 33.59 18.99 13.98 11.36 9.75 8.77 7.84 7.06 6.33 5.66 5.00 *Data source: SEER, 1983 to 1992 diagnosis period with follow-up through 1995. lx No. of patients alive at the start of the year; qx estimated probability of dying in the year given alive at the start; Sx estimated cumulative probability of survival to the end of the year. Estimate not reported because the standard error of the survival probability was 5%. years 0 through 4. The trends in annual hazard rates decrease during this period of time. However, trends in the annual hazards become more similar and flat when conditioned on the years 5 through 10. Another perspective of the prognosis for lung cancer patients is provided in Figure 1. In this figure, 5-year lung cancer conditional survival is presented by histologic subtype. The estimated 5-year survival probabilities increase and then level off when conditioned on successive years already survived. Although no 5-year survival probabilities are greater than 63%, survival consistently improves with years already survived, with one exception: For patients with adenosquamous carcinoma, the conditional survival estimates begin to fall when conditioned on having already survived 4 years. Although only estimates are reported in the figure with standard errors less than 5%, large confidence intervals for the last two reported estimates of adenosquamous carcinoma and the last reported estimate for papillary adenocarcinoma indicate that the slopes may not be real. Patients with bronchioloalveolar carcinoma have the best 5-year conditional survival, whereas patients with small-cell carcinoma have the worst 5-year conditional survival. Table 3 presents the number of patients alive at the beginning of the fifth year after diagnosis, the annual hazard, and the cumulative survival through the fifth year by histology-stage combinations. Cumulative survival is directly associated with stage at diagnosis. Cumulative survival to 5 years after diagnosis for patients with localized disease is two to four times that of patients with regional disease across histologic subtypes, and the annual hazards tend to increase with more advanced stage. Almost everyone diagnosed with distant-stage disease is dead by 5 years after diagnosis, and the annual hazards are similar across histologic subtypes. Very poor cumulative survival across histologic subtypes is also shown for patients diagnosed with unstaged disease. Among patients diagnosed with localized or regional disease, the number of patients living to 5 years after diagnosis was adequate for further exploration of the data by age. We originally chose age CHEST / 116 / 3/ SEPTEMBER, 1999 699
Figure 1. Five-year lung cancer observed survival conditioned on already having survived up to 5 years after diagnosis by histology. Data source: SEER cases diagnosed between 1983 and 1992 and followed through 1995. Estimates with standard errors 5% were not plotted. NOS not otherwise specified. groups 0 to 49, 50 to 69, and 70 years, but similar results and small numbers led us to regroup the ages as0to69and 70 years. Within histologic-stage categories, age further has a large effect on the cumulative survival and conditional hazard estimates (Table 4). The superior outcome among patients diagnosed with bronchioloalveolar carcinoma and the poorest outcome among patients diagnosed with small-cell carcinoma are maintained. Figure 2 provides 5-year conditional survival estimates for the histologic-stage-age subgroups. This figure illustrates that histology has a larger influence on conditional survival estimates among locally diagnosed cases than among regionally diagnosed cases. Large influences of stage and age are seen on the conditional survival estimates. Patients with bronchioloalveolar carcinoma continue to show the best survival, and patients with small-cell carcinoma continue to show the worst survival, with an exception for regionally diagnosed cases, aged 0 to 69 years. In this group, patients with large-cell carcinoma display the best survival, and patients with non large-cell carcinoma display similar, albeit lower, survival when conditioned on having already survived 3, 4, or 5 years after diagnosis. 700 Clinical Investigations
Table 3 Lung Cancer Cases, Estimated Conditional Probabilities of Death (Annual Hazards), and Estimated Cumulative Observed Survival at 5 Yrs After Diagnosis by Histology and Stage* Histology Variables AD BA PA AS SQ SC LC Localized lx 1,910 571 75 131 1,586 148 320 qx 9.0 7.6 10.0 8.8 13.3 10.4 10.5 Sx 43.8 56.8 49.1 40.9 29.3 12.3 28.8 Regional lx 1,054 176 25 76 1,110 202 222 qx 12.2 15.0 13.7 14.6 11.6 9.2 Sx 14.7 27.3 12.6 16.8 11.2 6.2 10.1 Distant lx 135 18 4 6 108 87 28 qx 16.7 16.2 17.5 Sx 1.2 3.9 2.2 1.1 1.4 1.1 0.9 Unstaged lx 151 28 3 9 178 66 49 qx 22.6 17.6 23.5 14.6 Sx 5.9 14.0 5.2 7.6 4.5 3.5 4.8 *Data source: SEER, 1983 to 1992 diagnosis period with follow-up through 1995. AD adenocarcinoma not otherwise specified and other specific adenocarcinomas; BA bronchioloalveolar carcinoma; PA papillary adenocarcinoma; AS adenosquamous carcinoma; BG bronchial gland carcinoma; SQ squamous-cell carcinoma; SC small-cell carcinoma; LC large-cell carcinoma. See Table 2 for other abbreviations. Estimate not reported because the standard error of the survival probability was 5%. Discussion The SEER Program began reporting national estimates of cancer incidence and survival in 1973. With that data now available through 1995, SEER provides a valuable source of information to compute conditional survival. The data are population-based, so the lung cancer survival estimates in this study reflect the average lung cancer survival in the population. The results are made more specific by reporting them by histology, stage, and age categories assigned at diagnosis. Further breakdowns of the data were not performed because of small numbers. Table 4 Lung Cancer Cases, Estimated Conditional Probabilities of Death, and Estimated Cumulative Probabilities of Observed Survival at 5 Yrs After Diagnosis by Histology, Stage, and Age* Histology Variables AD BA PA AS SQ SC LC Local Age0to69yr lx 1,504 406 62 97 1,091 147 318 qx 8.3 6.0 9.5 9.6 11.9 15.4 7.0 Sx 49.9 62.7 60.1 45.9 36.9 17.6 38.8 Age 70 yr lx 406 165 19 34 495 38 84 qx 30.5 11.5 6.7 16.5 5.6 9.3 Sx 11.7 46.2 32.3 20.0 8.3 20.3 Regional Age0to69yr lx 859 123 20 53 829 217 234 qx 10.4 12.8 12.9 14.0 7.8 Sx 17.5 30.8 15.4 17.9 13.7 8.5 13.9 Age 70 yr lx 195 53 5 23 281 44 50 qx 19.8 19.4 Sx 8.3 21.2 14.4 7.2 3.6 5.5 *Data source: SEER, 1983 to 1992 diagnosis period with follow-up through 1995. See Tables 2 and 3 for abbreviations. Estimate not reported because the standard error of the survival probability was 5%. CHEST / 116 / 3/ SEPTEMBER, 1999 701
Figure 2. Five-year lung cancer observed survival conditioned on already having survived up to 5 years after diagnosis by histology-stage-age categories. Data source: SEER cases diagnosed between 1983 and 1992 and followed through 1995. Estimates with standard errors 5% were not plotted. See Figure 1 for abbreviation. Observed survival probabilities are reported because, to a patient with lung cancer, these probabilities are likely to be more meaningful than relative survival probabilities. Tobacco smoking is the primary risk factor for lung cancer, 12 13 and it has been associated with the major histologic types. 6 It has been hypothesized that the strength of the association between tobacco smoking and histologic subtype is based on location; ie, more central lesions, such as squamous and small-cell carcinomas, should have a stronger association than more peripheral lesions, such as adeno- 702 Clinical Investigations
carcinoma and large-cell carcinoma. 6 This is supported by the fact that lung cancer in nonsmokers tends to be adenocarcinoma. 14 Previous studies have identified the worst survival among patients with small-cell carcinomas and the best survival among patients with adenocarcinoma across stage categories. 5 Our findings show that this differential survival is maintained when conditioned on years already survived since diagnosis. The benefits of early-stage diagnosis are uniquely illustrated by this report. For example, among squamous-cell lung cancer patients with localized, regional, distant, and unstaged disease, the 2-year cumulative survival rates from the time of diagnosis are 45%, 19%, 3%, and 11%, respectively. Among patients having already survived 2 years, the corresponding 5-year relative survival probability becomes 47%, 34%, 17%, and 20%, respectively. Hence, even among cases with an initial survival of 2 years after diagnosis, the probability of surviving another 5 years continues to be influenced by their stage of disease at diagnosis. Little or no change in lung cancer survival occurred during the study period. 4 5 This suggests minimal progress or change in treatment patterns during this time, although newer treatments now in development show promise for improving lung cancer survival. 15 16 Among the local or regional nonsmall-cell lung cancer (adenocarcinoma, squamous-cell carcinoma, and large-cell carcinoma) cases, the proportion treated with surgery alone was 36% from 1983 to 1987 and 41% from 1988 to 1992. The proportion of cases treated with radiation remained very close to 50% through the study period. The proportion of patients receiving both surgery and radiation remained close to 16%. For local/ regional small-cell lung cancer, the proportion of cases treated with surgery alone was 7% from 1983 to 1987, but increased to 10% from 1988 to 1992. The proportion of patients treated with radiation alone remained very close to 53%, and the proportion of patients receiving both surgery and radiation remained close to 3%. Hence, differences in the levels of the conditional survival probabilities are not likely to be confounded by changes in treatment, at least in the case of nonsmall-cell disease. For smallcell disease, changes in chemotherapy may or may not have occurred. Chemotherapy was not considered here because it is systematically underreported by the SEER Program (Lynn A. Ries, M.S., Cancer Statistics Branch, NCI; personal communication, October 6, 1998). The estimates of annual hazards and conditional survival probabilities given in this report are intended to provide more meaningful information than cumulative survival for patients who have already survived an initial period. Physicians and others who deal with these patients will find this information useful for both counseling patients and understanding the disease process. Further, the reported estimates may provide insight into the value of early detection of lung cancer. For these reasons, we recommend that estimates of annual hazards and cumulative survival be more widely derived and reported. References 1 Henson DE, Ries LA. On the estimation of survival. Semin Surg Oncol 1994; 10:1 5 2 Henson DE, Ries LA, Carriaga MT. Conditional survival of 56,268 patients with breast cancer. Cancer 1995; 76:237 242 3 Merrill RM, Henson DE, Ries LA. Conditional survival estimates in 34,963 patients with invasive carcinoma of the colon. Dis Colon Rectum 1998; 41:1097 1106 4 Ries LA, Kosary CL, Hankey BF, et al, eds. SEER cancer statistics review, 1973 1994. Bethesda, MD: National Cancer Institute, National Institutes of Health, 1997; Publication No. 97 2789 5 Travis WD, Travis LB, Devesa SS. Lung cancer. Cancer 1995; 75:191 202 6 Morabia A, Wynder EL. Cigarette smoking and lung cancer cell types. Cancer 1991; 68:2074 2078 7 Ries, LA. Influence of extent of disease, histology, and demographic factors on lung cancer survival in the SEER population-based data. Semin Surg Oncol 1994; 10:21 30 8 International classification of diseases for oncology. 2nd ed. Geneva, Switzerland: World Health Organization, 1990 9 SEER program: comparative staging guide for cancer. Version 1.1. Bethesda, MD: National Institutes of Health, 1993; Publication No. 93 3640 10 Ederer F, Axtell LM, Cutler SJ. The relative survival rate: a statistical methodology. Natl Cancer Inst Monog 1961; 6:101 121 11 National Cancer Institute. SEER cancer incidence public-use database 1973 1995. Rockville, MD: National Cancer Institute, 1998 12 Reducing the health consequences of smoking: 25 years of progress; a report of the Surgeon General. Washington, DC: US Department of Health and Human Services, 1989; DHHS Publication No. (CDC) 89 8411 13 Doll R, Peto R. The causes of cancer in quantitative estimates of avoidable risks of cancer in the United States today. J Natl Cancer Inst 1981; 66:1191 1308 14 McDuffie HH, Klaassen DJ, Dosman JA. Determinants of cell type in patients with cancer of the lungs. Chest 1990; 98:1187 1193 15 Birchard K. New treatments reported for lung cancer. Lancet 1997; 350:568 16 Zou Y, Zong G, Ling YH, et al. Effective treatment of early endobronchial cancer with regional administration of liposome-p53 complexes. J Natl Cancer Inst 1998; 90:1130 1137 CHEST / 116 / 3/ SEPTEMBER, 1999 703