Presented at the Fourth Annual Academic Surgical Congress, Fort Myers, Florida, February 2009.
Results from a population-based analysis
Article first published online: 24 FEB 2010
Copyright © 2010 American Cancer Society
Volume 116, Issue 9, pages 2234–2241, 1 May 2010
How to Cite
Xing, Y., Chang, G. J., Hu, C.-Y., Askew, R. L., Ross, M. I., Gershenwald, J. E., Lee, J. E., Mansfield, P. F., Lucci, A. and Cormier, J. N. (2010), Conditional survival estimates improve over time for patients with advanced melanoma. Cancer, 116: 2234–2241. doi: 10.1002/cncr.24966
We would like to acknowledge Debbie Dunaway for assistance with manuscript preparation.
- Issue published online: 20 APR 2010
- Article first published online: 24 FEB 2010
- Manuscript Accepted: 10 AUG 2009
- Manuscript Received: 6 AUG 2009
- conditional survival;
Conditional survival (CS) has emerged as a clinically relevant measure of prognosis for cancer survivors. The objective of this analysis was to provide melanoma-specific CS estimates to help clinicians promote more informed patient decision making.
Patients with melanoma and at least 5 years of follow-up were identified from the Surveillance Epidemiology and End Results registry (1988-2000). By using the methods of Kaplan and Meier, stage-specific, 5-year CS estimates were independently calculated for survivors for each year after diagnosis. Stage-specific multivariate Cox regression models including baseline survivor functions were used to calculate adjusted melanoma-specific CS for different subgroups of patients further stratified by age, gender, race, marital status, anatomic tumor location, and tumor histology.
Five-year CS estimates for patients with stage I disease remained constant at 97% annually, while for patients with stages II, III, and IV disease, 5-year CS estimates from time 0 (diagnosis) to 5 years improved from 72% to 86%, 51% to 87%, and 19% to 84%, respectively. Multivariate CS analysis revealed that differences in stages II through IV CS based on age, gender, and race decreased over time.
Five-year melanoma-specific CS estimates improve dramatically over time for survivors with advanced stages of disease. These prognostic data are critical to patients for both treatment and nontreatment related life decisions. Cancer 2010. © 2010 American Cancer Society.
In 2008, 62,480 Americans were diagnosed with melanoma, making it the sixth and seventh most common malignancy among men and women, respectively.1 The risk of developing metastatic melanoma and dying of disease is directly related to the stage of disease at presentation, which in turn is based on primary tumor thickness, ulceration, and the presence of metastatic nodal or distant disease.2 Traditional Kaplan-Meier 5-year survival estimates for melanoma patients presenting with American Joint Committee on Cancer (AJCC) stage I, II, III, and IV disease have been reported to be 93%, 68%, 45%, and 11%, respectively.3
It has become evident that many traditional survival estimates become less meaningful and less accurate for ongoing cancer survivors. As a result, conditional survival (CS), which represents the probability that a cancer patient survives additional years, given that the patient has already survived a given number of years, has emerged as a more clinically relevant measure of prognosis for patients who have survived several years beyond diagnosis and treatment. CS has been reported for patients with breast,4-6 lung,6-8 gastric,9 colon,6, 10 rectal,11 gallbladder,12, 13 ovarian,14 prostate,6, 15 head /neck,16 and brain cancers.17-19 However, CS estimates have not been published for patients with melanoma. The objective of this analysis was to determine melanoma-specific CS estimates for patients with various stages of disease to provide information that would be valuable for discussions between clinicians and patients with respect to both treatment and nontreatment related decisions.
MATERIALS AND METHODS
Data from the Surveillance Epidemiology and End Results (SEER) database program of the National Cancer Institute (released May 2008) was used in this study. During the inclusion period of our study (1988-2000), SEER comprised approximately 14% of the US population and includes patient demographics, tumor site and morphology, detailed pathologic information, as well as data on the first course of treatment (surgical and radiation therapy).
Patients with invasive cutaneous malignant melanoma diagnosed between 1988 and 2000 and with survival data available up to 2005 were identified from SEER (n = 67,410). The 1988 study date was chosen as the earliest study date for the cohort because it was the first year SEER coded pathological fields that would allow restaging according to the AJCC sixth edition,2 and 2000 was selected as the latest study date for the cohort, as it was the last year of diagnosis that would allow for at least 5 years of available follow-up data. All patients were manually restaged, using updated pathological information from SEER and the staging criteria of the AJCC sixth edition staging manual.2 The study cohort was limited to patients whose invasive melanoma was their first and only cancer diagnosis (n = 48,625). Subsequently, patients with incomplete pathological data were excluded, and the final study cohort comprised 40,520 patients with stage I to IV melanoma.
By using the traditional methods of Kaplan and Meier,20 disease-specific survival (DSS), overall survival (OS), and relative survival (the ratio of observed survival in a population to the expected survival rate) estimates at Year 1 through Year 10 were calculated for each stage-specific cohort. Separately, for those patients with a minimum of 5 years of follow-up data available, 5-year disease-specific CS, overall CS, and relative CS estimates were calculated for each stage-specific cohort, using previously described statistical methods.9 Conditional survival represents the probability that a cancer patient survived additional years, given that the patient has already survived a given number of years. For example, to compute the 5-year conditional survival for patients who have survived 1 year, the 6-year DSS is divided by the 1-year DSS. Accordingly, we calculated the stage-specific univariate 5-year CS for melanoma patients, given that the patients had already survived 1, 2, 3, 4, or 5 years.
To account for the potential covariate influences on the survival function, additional estimates were adjusted for age, gender, race, marital status, anatomic tumor location, and tumor histology. Specifically, the computation of the 1-year to 10-year DSS was completed using the adjusted survival functions for age, gender, race, anatomic tumor location, and tumor histology.21 To account for potential covariate influences, adjusted survival estimates and standard errors were obtained using the coefficient estimates from the final Cox model and the baseline survival function.21 For example, stage III CS for patients aged <50 years versus aged ≥50 years was calculated based on the adjusted survival functions for these 2 groups from the stage III cohort, using the baseline hazard incorporating the influence of gender, race, marital status, anatomic tumor location, and tumor histology simultaneously. The proportional hazards assumption was assessed graphically on the basis of Schoenfeld residuals.22 The test of slope23 was also performed to assess whether the CS improvements differed by categories of age, sex, race, histology, and location.
We used the following formula, a variation of the usual “Greenwood's formula”, to calculate the standard errors of CS estimates.17
Given m time intervals, dk is the number of deaths during interval k; rk is the number at risk for interval k; S(tj|ti) is the probability of survival past tj given survival past ti.
The majority of the statistical calculations were performed using Stata/MP (version 10.0; StataCorp, College Station, Tex). Prism version 5.02 (GraphPad Software, La Jolla, Calif) was used to examine the significance of the rates of change for conditional survival. All P values were 2-sided, and P values ≤0.05 were considered significant.
The median follow-up for the entire cohort (n = 40,520) was 9 years, and the majority of patients were from western states. Most of the patients were diagnosed with stage I (80%), followed by stage II (12.6%), stage III (4.8%), and stage IV (2.6%) disease. There was a larger proportion of older patients (≥50 years) in the cohorts with stage II and IV disease than in the cohorts with stage I and III disease (69.8% and 71.9% vs 51% and 52.4%, respectively, P < .001). About two thirds of the patients with stage IV disease were male, and there were more African American (2.5%) patients in the stage IV group than in the other 3 less advanced stage groups (0.4% for stage I and 1% for stages II and III). Nodular melanoma was more commonly noted in patients with stage II (33.5%) and stage III (24.4%) disease, while superficial spreading melanoma was the predominant histologic subtype in the stage I group (50.9%). About two thirds of patients with stage I and II disease received treatment in the form of wide local excision (WLE) and/or sentinel lymph node biopsy (SNB), while over 70% percent of the patients with stage III disease underwent lymph node dissection (LND). For those with stage IV disease, radiation therapy was more often used (33.9%) than in the other 3 stage-specific cohorts.
Figure 1A depicts the traditional DSS curves for patients in this cohort with stages I through IV melanoma. The 10-year DSS estimates were 94.3% (95% confidence interval [CI], 94.0-94.6) for stage I, 61.5% (95% CI, 59.9-63.1) for stage II, 44.1% (95% CI, 41.4-46.7) for stage III, and 16.1% (95% CI, 12.7-19.7) for stage IV. In terms of traditional OS, the 10-year survival estimates were 83.0% (95% CI, 82.5-83.5) for stage I, 45.0% (95% CI, 43.5-46.6) for stage II, 36.5% (95% CI, 34.0-39.0) for stage III, and 9.1% (95% CI, 7.0-11.6) for stage IV (Fig. 1B).
Stage-specific, 5-year disease-specific melanoma CS is presented as a bar graph in Figure 2A. For patients with stage I melanoma, 5-year CS estimates remained constant over time at 97%, reflecting the high likelihood of cure at the time of diagnosis. For patients with stage II melanoma, 5-year CS estimates increased over time, from 72% at time 0 (diagnosis) to 86% for patients surviving 5 years. With more advanced melanoma (stages III and IV), there were even greater changes in 5-year CS over time. For patients with stage III disease, 5-year CS estimates increased from 51% at time 0 to 87% for survivors at 5 years. Five-year CS estimates for patients with stage IV melanoma also improved dramatically for survivors over time, from 19% at time 0 to 84% for survivors at 5 years (Fig. 2A). Similar trends were found for 5-year overall CS (Fig. 2B). For example, 5-year CS estimates increased from 46% at time 0 to 80% for stage III survivors at 5 years.
The Cox proportional hazards model was used to identify prognostic factors associated with DSS for patients with stages I-IV melanoma (Table 1). For stage I, the model demonstrated that age, gender, race, marital status, anatomic tumor location, and tumor histology were all significant prognostic factors. For stages II and III, all aforementioned variables except race were significant, while for stage IV, only age, marital status, and anatomic tumor location were statistically significantly associated with DSS.
|Characteristic||Stage I n=32,430||Stage II n=5089||Stage III n=1963||Stage IV n=1038|
|HR (95% CI)||HR (95% CI)||HR (95% CI)||HR (95% CI)|
|≥50||1.60 (1.44-1.78)||1.56 (1.39-1.74)||1.60 (1.40-1.82)||1.31 (1.11-1.56)|
|Men||1.50 (1.34-1.67)||1.49 (1.33-1.67)||1.24 (1.07-1.42)||1.08 (0.92-1.27)|
|Non-white||0.37 (0.25-0.55)||0.83 (0.62-1.12)||0.92 (0.64-1.31)||0.74 (0.48-1.15)|
|Unmarried||1.29 (1.15-1.45)||1.16 (1.04-1.29)||1.24 (1.09-1.42)||1.25 (1.06-1.46)|
|Unknown||0.40 (0.34-0.47)||0.66 (0.55-0.80)||0.98 (0.69-1.40)||1.00 (0.67-1.49)|
|Nodular||4.32 (3.72-5.03)||1.11 (0.97-1.27)||1.36 (1.12-1.65)||1.09 (0.65-1.84)|
|Acral||2.48 (1.59-3.86)||1.28 (0.97-1.69)||1.47 (1.00-2.15)||1.06 (0.45-2.45)|
|Other/unknown||1.26 (1.13-1.40)||0.86 (0.75-0.98)||1.21 (1.02-1.45)||1.19 (0.75-1.87)|
|Upper extremity (ref)||1||1||1||1|
|Lower extremity||1.06 (0.89-1.26)||1.29 (1.10-1.52)||1.11 (0.89-1.38)||0.87 (0.58-1.29)|
|Trunk||1.54 (1.34-1.77)||1.46 (1.27-1.68)||1.13 (0.92-1.38)||1.37 (0.97-1.94)|
|Head/neck||1.68 (1.44-1.96)||1.30 (1.11-1.51)||1.36 (1.09-1.70)||1.09 (0.76-1.57)|
|Other||1.54 (0.82-2.89)||0.94 (0.46-1.90)||1.42 (1.07-1.89)||1.29 (0.95-1.75)|
To account for their covariate influences on the survival function, additional CS estimates were adjusted for age, gender, race, marital status, anatomic tumor location, and tumor histology. The adjusted CS was calculated for subgroups of patients further stratified by age. Figures 3A through 3D are generated from the adjusted survival function for these 2 age groups from each stage cohort, while controlling for the influence of above covariates. Overall, patients who were younger than 50 years of age had much higher estimated survival at the time of diagnosis compared with patients who were 50 years of age or older. The differences between melanoma-specific survival at the time of diagnosis were 1% (stage I), 10% (stage II), 16% (stage III), and 9% (stage IV). In terms of CS, younger patients tended to have better outcomes, and of note, these age-related differences in CS generally diminished over time for the stages II and III cohorts.
When stage-specific subgroups were stratified by gender, males were noted to have lower 5-year melanoma-specific survival than females at the time of diagnosis (stage I: 96% vs 97%, stage II: 69% vs 78%, stage III: 48% vs 56%, and stage IV: 17% vs 20%, respectively). The CS estimates for both genders increased over time for each year of follow-up; specifically, at Year 5, CS for males and females was 98% vs 98% for stage I, 83% vs 88% for stage II, 85% vs 88% for stage III, and 82% vs 83% for stage IV patients, respectively.
We also examined the impact of race, anatomic tumor location, and histology on stage-specific melanoma CS, and differing patterns emerged across disease stage groups. Because of the favorable prognosis for all patients with stage I melanoma, there were no major differences in CS among patients when stratified by race, tumor location, and histology. However, for the cohorts with stages II-IV disease, we observed that the effects of race, anatomic tumor location, and histology decreased over time for stage-specific melanoma CS over time.
For example, 5-year survival estimates for Caucasian patients with stage IV melanoma were lower than estimates for stage IV non-Caucasians by 9% at the time of diagnosis (18% vs 27%) and by 4% at year 5 (82% vs 86%), demonstrating a trend that race-related differences in CS decreased over time (P = .06). In terms of histologic subtype, acral lentiginous melanomas were associated with worse CS than the superficial and nodular groups, regardless of the length of follow-up. But this subgroup always had the largest gain in CS over the 5-year follow-up period for stages I-III patients (improvements of 3%, 16%, and 40% for stages I, II, and III, respectively). With respect to tumor location, patients with stages I-III disease who had primary tumors in the upper extremity were noted to have the most favorable CS, followed by the lower extremity, head/neck, and truncal groups, while patients with stage IV disease who had primary tumors in the lower extremity had the highest CS among different locations.
By testing whether the trends of the survival curves were different among various subgroups (ie, age, race, sex, histology, and tumor location); in other words, we found that sex and histology were determinates of CS improvement for patients with stage I disease, age, sex, and location for stage II, age, sex, and histology for patients with stage III disease (all P-values ≤.05), and that none of the variables were determinates of CS improvement for the cohort with stage IV disease.
Our analysis demonstrates that 5-year melanoma-specific CS improves dramatically over time for surviving patients with stage III and IV melanoma. For survivors with advanced (stage III and IV) melanoma, these estimates provide critical prognostic information; and for clinicians, it provides quantitative data for what is often qualitatively observed—a subset of patients continuing with biologically favorable disease and who survive beyond what would have been clinically predicted at the time of diagnosis. It is these unanticipated long-term survivors who provide the ongoing impetus for clinicians to continue to pursue aggressive treatment approaches despite having advanced melanoma at diagnosis.
The results of this CS analysis are consistent with other studies in which similar trends in 5-year CS have been reported in patients with advanced stages of lung,6-8 gastric,9 colon,6, 10 reactal,11 gallbladder,12, 13 ovarian,14 head/neck,16 and brain cancers.17-19 However, the patterns for CS in this cohort differed from those found in patients with breast and prostate cancer as reported by Kato et al.6 Specifically, there was a less dramatic increase in median overall CS for patients with breast cancer who gained only an additional 4 months to 7 months in CS every year after diagnosis,6 while patients with advanced prostate cancer were noted to have CS estimates that plateaued over time. These disparate results may be explained in part by the multivariate models used and/or the various efficacy of adjuvant treatment in different malignancies. Our study demonstrated that for patients with stage III melanoma, although some adverse prognostic factors for survival are significant at the time of initial diagnosis, such as age and gender, these factors become less influential over time. This is in contrast to the findings of Wang et al, who reported that the effects of gender, age (≥65 years vs <65 years), and race (African American vs Caucasian) on survival persisted over time for patients with rectal cancer.11 Other studies have investigated the effects of tumor grade and histology on CS, and for patients with poor grade, serous, and undifferentiated epithelioid ovarian cancer, CS improved much more rapidly with time.14
Estimates for 5-year OS in this cohort at the time of diagnosis for this cohort of patients with stage I, II, III, and IV disease were 91%, 60%, 46%, and 13%, respectively, which are similar to estimates reported by the AJCC.3 Because there have been some questions related to the accuracy of cause of death data reported in SEER, Choi et al14 have recommended relative survival as a more appropriate outcome measure to avoid this potential problem.24 However, with respect to melanoma, the cause of melanoma-related death data field has been examined by Weinstock et al25 and noted to be 93% accurate. In addition, relative survival estimates were examined in this cohort and determined to be similar to the reported estimates of DSS (data not shown).
Conditional survival estimates provide critical quantitative information for patients and their families that may reduce unnecessary anxiety. Fear of cancer recurrence, which is distinct from general patient concerns for future health,26 has been well-documented as an increasingly distressful concern for survivors.27-30 Conditional survival estimates may also serve to alleviate the associated anxiety of cancer caregivers. A recent study by Mellon et al indicated that survivors and caregivers often influence each others' fears, and that caregivers often experience significantly more fear of recurrence than survivors.31 Furthermore, recognition of the needs of cancer caregivers has led to the recent development of adjunctive scales such as the Cancer Survivors' Partners Unmet Needs (CaSPUN) measure to augment those for cancer survivors like the Cancer Survivors' Unmet Needs (CaSun) measure.32, 33 In addition, CS estimates revised annually accounting for additional years of survival may impact the economic decisions of survivors and their families. A recent study of the Cancer Problems in Living Scale (CPILS) from the American Cancer Society's Studies of Cancer Survivors revealed that the economic and fiscal distress of cancer survivors is distinct from other forms of emotional distress.26 The concerns of cancer survivors relating to the domains of control, uncertainty, and the future have also been shown to be important factors in models predicting overall quality of life.34 It is likely that understanding the increased likelihood of continued survival over time will help alleviate anxiety in each of these domains. Finally, given that quality of life has been shown to be an independent predictor of survival in melanoma patients with advanced disease,35, 36 reduced anxiety, and emotional distress may impact survival outcomes.
There are several limitations associated with the current study. SEER currently comprises comprehensive demographic and pathological data from 14 population-based cancer registries. First, SEER does not provide information on comorbidities, social economics status, nor complete treatment-related information. Because the final multivariate model could not incorporate these variables, our CS estimates might be biased and potentially limited to SEER regions only. In addition, SEER does not include data pertaining to disease recurrence, thus limiting our ability to examine disease-free CS outcomes in this analysis. The pathological data provided by SEER also do not designate micrometastatic nodal disease, which is required to stratify patients with stage III disease (who are known to have widely varying traditional survival estimates) into more homogeneous groups based on substage (ie, IIIA, IIIB, and IIIC). Furthermore, the sample size for the stage IV cohort is comparatively small (n = 1038) and the small fraction of survivors in this group resulted in wider ranges of error for the 5-year CS estimates. As a means of highlighting this point, we have included error bars on our graphic representations (Fig. 3D).
Compared with traditional survival estimates, such as median survival, DSS, and OS from the time of diagnosis, CS estimates provide more relevant quantitative information for cancer survivors as their risk profiles change over time. For clinicians, this information should be taken into account when making ongoing treatment decisions and when considering post-treatment surveillance intervals after annual milestones. These data are also critical for cancer survivors and their families when planning nontreatment related life decisions.
CONFLICT OF INTEREST DISCLOSURES
This project was supported by grant number 1 R01 CA127328-01 (Cormier, PI) from National Cancer Institute, National Institutes of Health. The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
- 1American Cancer Society. Cancer facts & figures 2008. Available at: http://www.cancer.org/downloads/STT/2008CAFFfinalsecured.pdf
- 2American Joint Committee on Cancer. In: GreeneFL, PageDL, FlemingID, et al., eds. AJCC Cancer Staging Handbook. 6th ed. New York: Springer; 2002: 421.
- 11Conditional survival in rectal cancer: a SEER database analysis. Gastrointest Cancer Res. 2007; 1: 84-89., , , .
- 13Conditional survival in gallbladder carcinoma: results from the SEER 11 dataset. J Clin Oncol. 2006; 24: 4130., , , .
- 21Regression models and life tables. J R Statistical Soc. 1972; B34: 187-220..
- 23Biostatistical Analysis. 4th ed. New Jersey: Prentice Hall; 1999..