Abstract
To estimate risk of recurrence for women diagnosed with nonmetastatic breast cancer considering the risks of other causes mortality.
We extend a method based on the diagnosis–metastasis–death pathway to include risks of other causes mortality. We estimate three probabilities as cumulative incidence of: (i) being alive and recurrence-free, (ii) death for other causes before a recurrence, and (iii) recurrence. We apply the method to female breast cancer relative survival from the Surveillance, Epidemiology, and End Results Program registries (2000–2018) data.
The cumulative incidence of recurrence shows a higher increase with more advanced cancer stage and is less influenced by age at diagnosis. At 5 years from diagnosis, the cumulative incidence of recurrence is less than 3% for those diagnosed with stage I, 10% to 13% for those diagnosed with stage II, and 37% to 47% for those diagnosed with stage III breast cancer. The estimates of recurrence considering versus ignoring the risks of dying from other causes were generally consistent, except for older women with more advanced stage, and longer time since diagnosis. In these groups, the net probability of recurrence, excluding the risks of dying from other causes, were overestimated.
For patients with cancer who are older or long-term survivors, it is important to include the risks of other cause mortality as the crude cumulative incidence of recurrence is a more appropriate measure.
These estimates are important in clinical decision making, as higher competing mortality may preclude the benefits of aggressive treatments.
Introduction
Improvements in breast cancer care have prolonged patient survival, but many patients still develop metastatic recurrence (1–3), the most severe form of the disease. While population-based cancer registries are an important data source for understanding cancer incidence, mortality, and survival, they do not routinely follow patients to collect intermediate outcomes such as disease recurrence, progression, or subsequent treatment. There has been increasing interest in understanding population-based risks of recurrence or metastatic disease (4), which can provide a more comprehensive characterization of the disease course, better quantify the cancer burden, and aid in clinical decisions (5).
Recently, a method has been developed to estimate the risk of metastatic recurrence (6), using stage-specific cancer registry survival data with published data on survival from recurrence (1). Estimates from this method have shown that metastatic recurrence risk was higher for women diagnosed at older ages, earlier year of diagnosis, more advanced stage, and with hormone receptor (HR)-negative tumors. For women ages 60 to 74, the estimated percent whose cancer recurred within 5 years were 5.3% and 21.6% for women diagnosed in 2000 to 2013 with local and regional breast cancer, respectively.
These estimates reflected the risk of recurrence while excluding the chances of dying of other causes, i.e., net survival (7). Net survival statistics are the best measures to represent the net effect of cancer diagnosis on survival or intermediate outcomes, removing the effects of death from other causes. They are relevant in cancer control, public health, and comparative studies. However, they are less useful for predicting a patient's outcomes, because cancer survivors are at risk of dying of other causes which may preclude recurrence (8). Failing to account for risks of dying from other causes may lead to an overestimation of the risk of experiencing recurrence. The higher the risk of dying of other causes, the more pronounced the overestimation is. Survival measures that include the risks of dying of other causes are known as competing risk survival, crude survival, crude probability of death, and cumulative incidence (7–9).
This study's objective is to extend the previous method for estimating risk of experiencing a metastatic recurrence by including the risks of dying from competing causes. Similar to the previous method (6), we use the term recurrence to encompass both a metastatic recurrence, i.e., the detection of metastasis after being disease-free following treatment, and progression to metastasis after initially being diagnosed with early-stage disease, but never being disease free. Although in some settings it is important to distinguish between recurrence and progression (2), the focus of this paper is to be inclusive of all patients that are diagnosed with metastatic disease after an early-stage disease diagnosis (1, 2). Considering the risks of dying of other causes, we estimate at any point in time after diagnosis the cumulative probabilities of experiencing a recurrence and of dying of non-cancer causes before a recurrence. By comparing the estimates with and without accounting for the risks of dying from competing causes, we aim to quantify the impact of including non-cancer mortality on the estimation of recurrence risk. We apply these methods to breast cancer survival data from the Surveillance, Epidemiology, and End Results (SEER) Program registries.
Materials and Methods
Overview and notation
The method employed in this paper is an extension of the method presented in a previous study (6) and is based on the diagnosis–metastasis–death pathway (Fig. 1). Similarly, we define recurrence as the detection of metastatic disease after a diagnosis of early-stage nonmetastatic breast cancer, regardless of whether it is progression or recurrence (6). To estimate the “latent” proportions of patients who are cured versus non-cured at diagnosis, we use mixture cure survival models. In this analysis, both the cured and non-cured groups are at risk of dying from other causes, but only the non-cured group can die from the diagnosed cancer (Fig. 1).
Cancer diagnosis–recurrence–death pathway including the risks of dying of other causes. At diagnosis, patients with cancer can die of cancer or other causes where hCA and hOC, are the net hazards of death from cancer and from other causes. The net cancer hazard hCA is estimated from cancer registry relative survival data. Mixture cure survival model is used to estimate the latent proportion c of patients not at risk of dying of their cancer (cured) and survival (with hazard hU) for the (1-c) proportion not cured, SCA(t) = c + (1-c) SU(t). Under independence, the survival time for those not cured is the sum of the survival time from diagnosis to recurrence and the time from recurrence to cancer death, hU= h1 + h2.
Cancer diagnosis–recurrence–death pathway including the risks of dying of other causes. At diagnosis, patients with cancer can die of cancer or other causes where hCA and hOC, are the net hazards of death from cancer and from other causes. The net cancer hazard hCA is estimated from cancer registry relative survival data. Mixture cure survival model is used to estimate the latent proportion c of patients not at risk of dying of their cancer (cured) and survival (with hazard hU) for the (1-c) proportion not cured, SCA(t) = c + (1-c) SU(t). Under independence, the survival time for those not cured is the sum of the survival time from diagnosis to recurrence and the time from recurrence to cancer death, hU= h1 + h2.
We assume that patients in the non-cured group who die of cancer have experienced a recurrence before. However, it is possible that some non-cured patients may die from other causes before experiencing a recurrence. Once a recurrence is experienced, patients may die from cancer or other unrelated causes.
In population-based survival analysis, |${S}_{OC}$| is estimated from the general population life tables assuming that cancer mortality is small compared with other-cause mortality (10). Net cancer survival |${S}_{CA}$| is estimated using relative survival (Ederer II; ref. 11), or the Pohar-Perme net survival estimator (12). Both methods use life tables stratified by age, sex, calendar year, race and geography (13) compared with overall cancer patients’ survival, enabling estimation of the excess mortality associated with cancer and “net” cancer survival. In the application to breast cancer data, we use Ederer II relative survival to estimate |${S}_{CA}$|.
Time from diagnosis to recurrence and from recurrence to cancer death
Recurrence-free survival and other measures under competing causes of death
For patients diagnosed with nonmetastatic disease, experiencing a recurrence, and dying of other causes are considered competing events. Our objective is to estimate the cumulative incidence or cumulative probabilities of experiencing each event in the presence of the risk of the other event, as proposed by Fine and Gray (16, 17). Using the estimates provided by expressions (E1), (E2), and (E3) at any given time t since diagnosis, we aim to estimate the cumulative incidence or cumulative probabilities of:
- A)Being alive and recurrence-free (recurrence-free survival)
- B)Dying of other causes before a recurrence
- C)Recurrence (irrespective of life status after)
In addition, we can estimate the probabilities of:
C3) Recurrence and being dead of other causes
Formula derivations and details, including calculation of the variables for the cumulative incidence of recurrence C) are provided in Supplementary Materials.
Comparison with estimates excluding the risks of dying of other causes (net)
We compared the cumulative incidence of recurrence C, with the cumulative probability of recurrence excluding the risks of dying of other causes (net; ref. 6) which is calculated as |$c + (1 - c)\,{S}_1(t)$|.
Estimation: application to breast cancer
The application of the method consists of four steps. We first calculate relative survival for patients diagnosed with a first primary malignant female breast cancer between 2000 to 2018 in SEER-17 (November 2021 data submission) and follow-up information through December 2019 using the SEER*Stat software. Patients were grouped by age at diagnosis (15–54, 55–64, 65–74, 75+) and by stage at diagnosis using the American Joint Committee on Cancer 6th ed. stage groupings (I, II, III, and IV), (https://seer.cancer.gov/seerstat/variables/seer/ajcc-stage/6th/). We excluded patients diagnosed through death certificate or autopsy, and those with zero months of survival. We included only patients whose breast cancer was their first primary cancer because of uncertainties in associating recurrence with breast cancer or the first tumor. The SEER*Stat relative survival also provides other causes (expected) survival which is calculated by matching each cancer patient to the specific age, year, sex, race/ethnicity, and area level socio-economic status life table (13).
In step 2 we fit a log-logistic cure mixture survival (E2) to relative survival using the CanSurv Version 1.3 software (available at https://surveillance.cancer.gov/cansurv/download) to estimate the cure fraction |$\hat{c}$| and the survival for the uncured patients |${\hat{S}}_U$|, for each stage (I, II, and III) and age at diagnosis group combinations (15). The third step consists of uploading the relative survival data together with the output from CanSurv into the NCI RecurRisk web tool (available at https://analysistools.cancer.gov/) to estimate the recurrence-free survival |${S}_1$| from |${\hat{S}}_U$| and survival from metastasis |${\hat{S}}_2$|, as described in section 2.2. The expected survival used in relative survival is used to estimate other causes mortality and is matched to each cancer patient by age, calendar year, sex, race/ethnicity, and area level socio-economic status (13). In the fourth step, we use the equations in this paper to estimate the probabilities defined above.
Data availability
The SEER data generated in this study are publicly available at https://seer.cancer.gov/data/.
Results
The study includes 740,052 women diagnosed with breast cancer stages I–III at ages 15 years or older between 2010 to 2018. Women diagnosed with stage IV breast cancer are not included in the study cohort; however, their survival time was used to estimate the survival from recurrence, as specified in equation E4.
Cumulative incidence of recurrence versus dying of other causes before recurrence
The cumulative incidence of recurrence shows a higher increase with more advanced stage, while it is less influenced by age at diagnosis. The cumulative incidence of recurrence at 5 years from diagnosis range from <3%, 10% to 13%, and 37% to 47% for patients diagnosed with stages I, II, and III, respectively, depending on the age group (Table 1). The cumulative incidence of death from other causes before a recurrence increases with age and time since diagnosis (Fig. 2; Supplementary Materials Table S1). For most patients diagnosed with stages II and III breast cancer, the cumulative incidence of recurrence exceeds the cumulative incidence of dying from other causes. An exception is observed for older women (ages 75+) diagnosed with stage II breast cancer, where their chances of experiencing a recurrence versus dying of other causes before a recurrence are 12.6% versus 28.2% at 5 years from diagnosis, respectively. The percent of women alive and in recurrence 5 years after diagnosis are highest for women diagnosed with stage III breast cancer being 14.9%, 12.2%, 11.0%, and 6.8% at ages 15 to 54, 55 to 64, 65 to 74, and 75+, respectively (Table 1).
Estimates of the percent of women (A) alive and recurrence-free, (B) dying of other causes (OC) before advancing to metastatic recurrence, and (C) experienced metastatic recurrence (with standard errors) within 5 years from diagnosis, by stage and age at diagnosis. Estimates of the percent who advanced to recurrence are also portioned in women who is alive and those who die of any cause. The last column displays net estimates of the percent of women advancing to metastatic recurrence, excluding the competing risks of dying of other causes (net). Crude estimates consider risks of dying of other causes while net do not.
| Crude | Net cumulative probability of experieenced a recurrence | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cumulative probability of experienced recurrence | |||||||||||
| Stage | Age | No. alive at diagnosis | Years since diagnosis | Alive & recurrence-free (A) | Died of OC before recurrence (B) | Alive | Died | Total* (C) | Total Std. Err. | Est. | Std. Err. |
| Stage I | 15–54 | 112,764 | 5 | 96.1% | 1.4% | 1.4% | 1.1% | 2.5% | 0.10% | 2.5% | 0.10% |
| 55–64 | 98,988 | 5 | 95.4% | 3.8% | 0.4% | 0.4% | 0.8% | 0.09% | 0.8% | 0.09% | |
| 65–74 | 91,557 | 5 | 91.0% | 8.7% | 0.2% | 0.1% | 0.3% | 0.10% | 0.3% | 0.11% | |
| 75+ | 64,468 | 5 | 71.8% | 28.1% | 0.1% | 0.1% | 0.1% | 0.13% | 0.2% | 0.16% | |
| Stage II | 15–54 | 113,222 | 5 | 86.3% | 1.3% | 5.8% | 6.7% | 12.5% | 0.24% | 12.5% | 0.24% |
| 55–64 | 67,037 | 5 | 86.2% | 3.6% | 4.2% | 6.0% | 10.2% | 0.24% | 10.4% | 0.24% | |
| 65–74 | 51,137 | 5 | 81.5% | 8.2% | 3.9% | 6.4% | 10.3% | 0.28% | 10.7% | 0.29% | |
| 75+ | 43,028 | 5 | 59.2% | 28.2% | 3.1% | 9.6% | 12.6% | 0.45% | 14.7% | 0.50% | |
| Stage III | 15–54 | 43,677 | 5 | 60.1% | 1.0% | 14.9% | 24.1% | 39.0% | 0.60% | 39.1% | 0.60% |
| 55–64 | 23,611 | 5 | 60.3% | 2.8% | 12.2% | 24.7% | 36.9% | 0.61% | 37.3% | 0.61% | |
| 65–74 | 15,918 | 5 | 55.5% | 6.4% | 11.0% | 27.1% | 38.1% | 0.68% | 39.2% | 0.69% | |
| 75+ | 14,645 | 5 | 32.9% | 19.6% | 6.8% | 40.7% | 47.5% | 0.83% | 52.4% | 0.82% | |
| Crude | Net cumulative probability of experieenced a recurrence | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Cumulative probability of experienced recurrence | |||||||||||
| Stage | Age | No. alive at diagnosis | Years since diagnosis | Alive & recurrence-free (A) | Died of OC before recurrence (B) | Alive | Died | Total* (C) | Total Std. Err. | Est. | Std. Err. |
| Stage I | 15–54 | 112,764 | 5 | 96.1% | 1.4% | 1.4% | 1.1% | 2.5% | 0.10% | 2.5% | 0.10% |
| 55–64 | 98,988 | 5 | 95.4% | 3.8% | 0.4% | 0.4% | 0.8% | 0.09% | 0.8% | 0.09% | |
| 65–74 | 91,557 | 5 | 91.0% | 8.7% | 0.2% | 0.1% | 0.3% | 0.10% | 0.3% | 0.11% | |
| 75+ | 64,468 | 5 | 71.8% | 28.1% | 0.1% | 0.1% | 0.1% | 0.13% | 0.2% | 0.16% | |
| Stage II | 15–54 | 113,222 | 5 | 86.3% | 1.3% | 5.8% | 6.7% | 12.5% | 0.24% | 12.5% | 0.24% |
| 55–64 | 67,037 | 5 | 86.2% | 3.6% | 4.2% | 6.0% | 10.2% | 0.24% | 10.4% | 0.24% | |
| 65–74 | 51,137 | 5 | 81.5% | 8.2% | 3.9% | 6.4% | 10.3% | 0.28% | 10.7% | 0.29% | |
| 75+ | 43,028 | 5 | 59.2% | 28.2% | 3.1% | 9.6% | 12.6% | 0.45% | 14.7% | 0.50% | |
| Stage III | 15–54 | 43,677 | 5 | 60.1% | 1.0% | 14.9% | 24.1% | 39.0% | 0.60% | 39.1% | 0.60% |
| 55–64 | 23,611 | 5 | 60.3% | 2.8% | 12.2% | 24.7% | 36.9% | 0.61% | 37.3% | 0.61% | |
| 65–74 | 15,918 | 5 | 55.5% | 6.4% | 11.0% | 27.1% | 38.1% | 0.68% | 39.2% | 0.69% | |
| 75+ | 14,645 | 5 | 32.9% | 19.6% | 6.8% | 40.7% | 47.5% | 0.83% | 52.4% | 0.82% | |
Note: The results used the analytical deconvolution method and the log-logistic cure mixture model. Survival from recurrence used an adjustment of r = 1.35 compared with de novo distant-stage breast cancer.
*The sum of the percent of those who are alive and dead after advancing to recurrence may not equal the total who advanced to recurrence because of rounding.
Estimated cumulative percentages of patients who advanced to recurrence and who died of other causes before advancing to recurrence by age and stage at diagnosis.
Estimated cumulative percentages of patients who advanced to recurrence and who died of other causes before advancing to recurrence by age and stage at diagnosis.
Conditional probabilities of experiencing recurrence
Table 2 displays the probabilities of experiencing a recurrence in the next 5 years, given that patients are alive and recurrence-free at 0, 5, and 10 years from diagnosis. Notably, for stages II and III, the percentages experiencing a recurrence in the next 5 years decrease with survival time; they are lower among 5-year and 10-year survivors compared with those just diagnosed. For example, among women diagnosed with stage III breast cancer at ages 55 to 64, the percentage advancing to recurrence in the next 5 years is 36.9% for those just diagnosed, 12.9% for 5-year survivors, and 7.6% for 10-year survivors. The clear exception to this trend is women diagnosed with stage I breast cancer, where the conditional probabilities of recurrence are small and mostly remain similar or slightly increase with survival time.
Conditional probabilities of being alive and metastatic (MR) recurrence-free, advancing to MR before dying of other causes, dying of other causes before advancing to MR in the presence of competing causes of death in an interval given being alive and recurrence-free at the beginning of interval. Respective conditional probabilities of advancing MR in the absence of other causes of death (net) are provided in the last column.
| Conditional crude | Conditional net | |||||
|---|---|---|---|---|---|---|
| Stage | Age | Interval from diagnosis in years | Alive & recurrence-free | Died of OC before recurrence | Advanced to recurrence (before dying of OC) | Advanced to recurrence |
| Stage I | 15–54 | 0 to 5 | 96.1% | 1.4% | 2.5% | 2.5% |
| 5 to 10 | 96.4% | 2.1% | 1.6% | 1.6% | ||
| 10 to 15 | 96.2% | 3.0% | 0.7% | 0.7% | ||
| 55–64 | 0 to 5 | 95.4% | 3.8% | 0.8% | 0.8% | |
| 5 to 10 | 93.2% | 5.5% | 1.3% | 1.4% | ||
| 10 to 15 | 90.1% | 8.1% | 1.8% | 1.8% | ||
| 65–74 | 0 to 5 | 91.0% | 8.7% | 0.3% | 0.3% | |
| 5 to 10 | 84.9% | 13.5% | 1.7% | 1.8% | ||
| 10 to 15 | 74.7% | 21.3% | 4.0% | 4.5% | ||
| 75+ | 0 to 5 | 71.8% | 28.1% | 0.1% | 0.2% | |
| 5 to 10 | 60.0% | 38.6% | 1.5% | 2.0% | ||
| 10 to 15 | 44.8% | 49.9% | 5.3% | 7.7% | ||
| Stage II | 15–54 | 0 to 5 | 86.3% | 1.3% | 12.5% | 12.5% |
| 5 to 10 | 93.6% | 1.9% | 4.4% | 4.5% | ||
| 10 to 15 | 95.0% | 2.9% | 2.2% | 2.2% | ||
| 55–64 | 0 to 5 | 86.2% | 3.6% | 10.2% | 10.4% | |
| 5 to 10 | 89.0% | 5.3% | 5.7% | 5.8% | ||
| 10 to 15 | 88.5% | 7.9% | 3.5% | 3.7% | ||
| 65–74 | 0 to 5 | 81.5% | 8.2% | 10.3% | 10.7% | |
| 5 to 10 | 78.4% | 12.8% | 8.8% | 9.4% | ||
| 10 to 15 | 71.7% | 20.4% | 7.9% | 8.8% | ||
| 75+ | 0 to 5 | 59.2% | 28.2% | 12.6% | 14.7% | |
| 5 to 10 | 53.1% | 37.2% | 9.7% | 12.1% | ||
| 10 to 15 | 43.8% | 49.3% | 6.9% | 9.3% | ||
| Stage III | 15–54 | 0 to 5 | 60.1% | 1.0% | 39.0% | 39.1% |
| 5 to 10 | 90.8% | 1.9% | 7.3% | 7.4% | ||
| 10 to 15 | 92.9% | 2.8% | 4.3% | 4.4% | ||
| 55–64 | 0 to 5 | 60.3% | 2.8% | 36.9% | 37.3% | |
| 5 to 10 | 82.1% | 5.0% | 12.9% | 13.2% | ||
| 10 to 15 | 84.7% | 7.7% | 7.6% | 7.9% | ||
| 65–74 | 0 to 5 | 55.5% | 6.4% | 38.1% | 39.2% | |
| 5 to 10 | 69.6% | 11.9% | 18.5% | 19.6% | ||
| 10 to 15 | 67.6% | 19.4% | 13.0% | 14.3% | ||
| 75+ | 0 to 5 | 32.9% | 19.6% | 47.5% | 52.4% | |
| 5 to 10 | 43.6% | 32.5% | 23.9% | 29.0% | ||
| 10 to 15 | 38.6% | 44.3% | 17.2% | 22.6% | ||
| Conditional crude | Conditional net | |||||
|---|---|---|---|---|---|---|
| Stage | Age | Interval from diagnosis in years | Alive & recurrence-free | Died of OC before recurrence | Advanced to recurrence (before dying of OC) | Advanced to recurrence |
| Stage I | 15–54 | 0 to 5 | 96.1% | 1.4% | 2.5% | 2.5% |
| 5 to 10 | 96.4% | 2.1% | 1.6% | 1.6% | ||
| 10 to 15 | 96.2% | 3.0% | 0.7% | 0.7% | ||
| 55–64 | 0 to 5 | 95.4% | 3.8% | 0.8% | 0.8% | |
| 5 to 10 | 93.2% | 5.5% | 1.3% | 1.4% | ||
| 10 to 15 | 90.1% | 8.1% | 1.8% | 1.8% | ||
| 65–74 | 0 to 5 | 91.0% | 8.7% | 0.3% | 0.3% | |
| 5 to 10 | 84.9% | 13.5% | 1.7% | 1.8% | ||
| 10 to 15 | 74.7% | 21.3% | 4.0% | 4.5% | ||
| 75+ | 0 to 5 | 71.8% | 28.1% | 0.1% | 0.2% | |
| 5 to 10 | 60.0% | 38.6% | 1.5% | 2.0% | ||
| 10 to 15 | 44.8% | 49.9% | 5.3% | 7.7% | ||
| Stage II | 15–54 | 0 to 5 | 86.3% | 1.3% | 12.5% | 12.5% |
| 5 to 10 | 93.6% | 1.9% | 4.4% | 4.5% | ||
| 10 to 15 | 95.0% | 2.9% | 2.2% | 2.2% | ||
| 55–64 | 0 to 5 | 86.2% | 3.6% | 10.2% | 10.4% | |
| 5 to 10 | 89.0% | 5.3% | 5.7% | 5.8% | ||
| 10 to 15 | 88.5% | 7.9% | 3.5% | 3.7% | ||
| 65–74 | 0 to 5 | 81.5% | 8.2% | 10.3% | 10.7% | |
| 5 to 10 | 78.4% | 12.8% | 8.8% | 9.4% | ||
| 10 to 15 | 71.7% | 20.4% | 7.9% | 8.8% | ||
| 75+ | 0 to 5 | 59.2% | 28.2% | 12.6% | 14.7% | |
| 5 to 10 | 53.1% | 37.2% | 9.7% | 12.1% | ||
| 10 to 15 | 43.8% | 49.3% | 6.9% | 9.3% | ||
| Stage III | 15–54 | 0 to 5 | 60.1% | 1.0% | 39.0% | 39.1% |
| 5 to 10 | 90.8% | 1.9% | 7.3% | 7.4% | ||
| 10 to 15 | 92.9% | 2.8% | 4.3% | 4.4% | ||
| 55–64 | 0 to 5 | 60.3% | 2.8% | 36.9% | 37.3% | |
| 5 to 10 | 82.1% | 5.0% | 12.9% | 13.2% | ||
| 10 to 15 | 84.7% | 7.7% | 7.6% | 7.9% | ||
| 65–74 | 0 to 5 | 55.5% | 6.4% | 38.1% | 39.2% | |
| 5 to 10 | 69.6% | 11.9% | 18.5% | 19.6% | ||
| 10 to 15 | 67.6% | 19.4% | 13.0% | 14.3% | ||
| 75+ | 0 to 5 | 32.9% | 19.6% | 47.5% | 52.4% | |
| 5 to 10 | 43.6% | 32.5% | 23.9% | 29.0% | ||
| 10 to 15 | 38.6% | 44.3% | 17.2% | 22.6% | ||
Note. The results used the analytical deconvolution method and the log-logistic cure mixture model. Survival from recurrence used an adjustment of HR r = 1.35 compared with de novo distant-stage breast cancer.
Crude versus net comparison
Figure 3 illustrates the difference between the net and crude cumulative probabilities of recurrence, by time since diagnosis, stage, and age. The figure shows that the estimates are generally quite similar (represented by the white areas). However, in specific scenarios, the net probabilities of recurrence, are higher and tend to overestimate the estimated risks of recurrence (red areas). These scenarios include women diagnosed at older ages, more advanced stages, and longer time since diagnosis. When comparing the net and crude conditional percentages of experiencing a recurrence in the next 5 years differences are smaller and only observed for women diagnosed at ages 75+ years and stages II or III breast cancer (Fig. 3).
Absolute difference between crude and net cumulative incidence of recurrence by years since diagnosis. Absolute difference between conditional crude and net cumulative incidence of recurrence within the next 5 years for patients' recurrence-free 5 or 10 years from diagnosis.
Absolute difference between crude and net cumulative incidence of recurrence by years since diagnosis. Absolute difference between conditional crude and net cumulative incidence of recurrence within the next 5 years for patients' recurrence-free 5 or 10 years from diagnosis.
Discussion
Our study extends previous estimates of the risk of metastatic recurrence by accounting for the risks of dying from other causes. Our analysis of breast cancer suggests that the cumulative incidence of recurrence varies significantly depending on the cancer stage, but less so based on age. For instance, among women aged 15 to 54 with stages I, II, and III breast cancer, we estimate that 2.5%, 12.5%, and 39%, respectively, will experience recurrence within five years of diagnosis. The corresponding figures for women aged 75 or over are 0.1%, 12.6%, and 47.5%. By accounting for the risks of dying from other causes, the effect of age on the estimates is reduced, and the differences between age groups become less pronounced, compared with the net estimates.
The cumulative incidence of recurrence, considering (crude) versus excluding (net) the risks of dying from other causes, were generally similar. However, there were exceptions for older women, more advanced stage, and longer time since diagnosis. Failing to consider the risks associated with mortality from other causes resulted in an overestimation of the probability of recurrence in these cases. Therefore, it is essential to incorporate the possibility of dying from other causes when estimating the probability of disease progression, especially for older patients who face a higher risk of dying due to alternative health issues. These types of measures reflect the true progression-survival experience of patients with cancer and play a crucial role in clinical decision-making, as higher risks of dying from competing causes may preclude the benefits of aggressive treatment strategies.
Our study has some limitations. We focused on risk groups categorized by stage and age at diagnosis and did not include other important prognostic factors such as HR status and HER2 status. Although HR status has been available in the SEER data since 2000, the collection of HER2 status began in 2010, which limited our ability to estimate long-term outcomes for these factors. Our decision to include only age and stage was made to illustrate the method and provide estimates for the most critical survival predictors. In addition, we did not control for treatment and/or comorbidity in the analysis. Adjuvant treatment information in the SEER data is incomplete, and comorbidity data is unavailable. Consequently, our estimates reflect the average treatment patterns and comorbidities within a specific stage and age group. For future projects, we plan to improve our estimates by incorporating more detailed cancer profiles and patient characteristics, potentially including comorbidity measures from SEER-Medicare data (18, 19). We assumed that women with recurrent metastatic breast cancer had a 1.35 times higher risk of death compared with women diagnosed with de novo metastatic breast cancer, based on findings from a single institution study (1). However, it is worth noting that previous estimates of the net recurrence risk have demonstrated robustness across a range of values from r = 1.0 to r = 1.7 (6). This indicates that the crude estimates would likely yield similar results as well. The focus of this paper is on metastasis after a diagnosis with early-stage disease, regardless of whether it is a progression or recurrence. Clinically distinguishing between recurrence and progression can be challenging, as a patient may appear to be disease-free, even though undetected microscopic metastasis may be present after treatment. Inclusion of both progression and recurrence cases aims to be inclusive and encompass all patients experiencing metastatic disease after the diagnosis of early-stage disease.
Our study has strengths. We used the SEER data, which reflect a large representative population of women diagnosed with breast cancer in the U.S. Previous research (6) has shown the method to be very robust to various assumptions, e.g., the parametric form of the mixture cure survival and the independence assumption between the time from diagnosis to recurrence and from recurrence to cancer death. The methods excluding the risks of dying of other causes validated well against cancer registry collected recurrence data (20). In addition, the conditional estimates we provided are highly useful for survivors who are recurrence-free and wish to know their likelihood of remaining recurrence-free in the future.
In conclusion, there is an urgent need for population-representative estimates of cancer recurrence, given the current limited data available. Our study aimed to address this issue by providing population-based estimates of the cumulative incidence of recurrence, including the risks of dying from other causes. This approach offers a more accurate representation of the survival experience of patients with cancer. The study showed that while the net and crude estimates were generally similar, the inclusion of the risks of dying from other causes led to lower and more accurate estimates of the cumulative incidence of recurrence in some cases, especially among older patients diagnosed with more advanced stages. Moreover, the methodology used in this study has the potential to be applied to other cancer types, bridging data gaps, providing insight into future data collection efforts, and enhancing our understanding of the cancer recurrence burden in the population.
Authors' Disclosures
Z. Zou reports personal fees from NCI during the conduct of the study; personal fees from NCI outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
A.B. Mariotto: Conceptualization, formal analysis, methodology, writing–original draft, writing–review and editing. L. Botta: Conceptualization, formal analysis, methodology, writing–original draft, writing–review and editing. A. Bernasconi: Conceptualization, formal analysis, writing–original draft, writing–review and editing. Z. Zou: Software, formal analysis, methodology, writing–review and editing. G. Gatta: Conceptualization, writing–original draft, writing–review and editing. R. Capocaccia: Conceptualization, formal analysis, methodology, writing–original draft, writing–review and editing.
The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
Note: Supplementary data for this article are available at Cancer Epidemiology, Biomarkers & Prevention Online (http://cebp.aacrjournals.org/).
RSS Feeds