US cancer deaths prevented due to survival improvements stratified by extent of disease, 2010-2019

Abstract

Background

Progress against cancer mortality has been driven by primary prevention, early detection, and cancer treatment. We estimated the number of cancer deaths that were avoided due to stage-specific improvements in cancer survival among patients diagnosed in 2010-2019 followed through 2020.

Methods

We used cancer incidence data from 17 Surveillance, Epidemiology, and End Results (SEER) cancer registries during 2004-2019. We estimated the number of cancer deaths prevented due to cancer- and stage-specific survival improvements (based on SEER summary stage) as the observed minus expected number of cancer deaths through 2020. We calculated the expected number of cancer deaths from estimated cumulative incidence models setting the calendar year effect to 2009.

Results

During 2010-2019, there were 3 310 270 incident cancers and 966 733 cancer deaths through 2020 in SEER-17. Improvements in stage-specific cancer survival resulted in a 4.7% (95% CI = −5.3% to −4.2%) decline in cancer deaths in females (22,874 fewer deaths) and a 4.4% (95% CI = −4.9% to −3.9%) decline in males (23 198 fewer deaths) in SEER-17 regions, corresponding to approximately 173 900 fewer cancer deaths in the full US population. The largest absolute declines were for lung and liver cancers, whereas the largest relative declines were observed for melanoma and leukemia. Cancer deaths prevented were not statistically significant for colorectal or prostate cancers. All statistical tests were 2-sided.

Conclusions

Stage-specific survival gains, reflecting treatment advances and improved access to cancer treatment from 2010 to 2019, resulted in an estimated 173 900 fewer cancer deaths among US cancer patients diagnosed during this time period.

In 2024, there were an estimated 2 million new cancer diagnoses and 612 000 cancer deaths in the United States.¹ There has been tremendous progress against cancer mortality over the last 2 decades, with age-standardized cancer death rates declining 27% from 2000 to 2019,² driven by a combination of primary prevention efforts to reduce cancer incidence, early detection to improve prognosis, and improvements in cancer treatments to increase survival.^2-4 A recent study estimated that 5.94 million breast, cervical, colorectal, and prostate cancer deaths in the United States were prevented during 1975-2020 thanks to prevention, screening, and treatment efforts: 3.39 million from tobacco control, 1.36 million from other cancer prevention and screening interventions, and 1.20 million from cancer treatment.⁴

Five-year survival rates increased from 64% for cancers diagnosed in 2000 to 68% in 2014.² Some of this benefit is due to effective cancer screening, as well as lead-time bias induced by earlier diagnosis of certain cancers.¹ Survival improvements are also reflective of advances in cancer treatment, including notable improvements in survival due to treatment advances for non-small cell lung cancer (NSCLC), non-Hodgkin lymphoma (NHL), chronic myeloid leukemia, breast cancer, and late-stage melanoma.^5,6

We estimated the number of cancer deaths that were avoided due to improvements in survival stratified by extent of disease (captured by summary stage) among cancers diagnosed during 2010-2019 and followed through 2020 for 8 cancer types and all cancers combined. As stage-specific survival gains are largely due to cancer therapies, these estimates approximate the number of cancer deaths avoided due to treatment advances and increases in receipt of treatment.

Methods

Study population

Cancer incidence data were ascertained from 17 population-based cancer registries in the Surveillance, Epidemiology, and End Results (SEER) Program, representing 26.5% of the US population. This analysis focused on first primary cancer cases diagnosed during 2004-2019. Data from 2020 to 2021 were not included due to documented declines in cancer incidence during 2020 due to the COVID-19 pandemic.⁷ We separately examined the 8 leading causes of cancer death in the United States, defined based on ICD-10 codes⁸—lung and bronchus cancer (ie, “lung”), colorectal cancer, pancreatic cancer, female breast cancer (ie, “breast”), prostate cancer, liver and intrahepatic bile duct (ie, “liver”) cancer, leukemia, and NHL.⁹ Melanoma was also included given large recent reductions in mortality. All other cancer types were combined. Analyses were carried out stratified by SEER combined summary stage (localized, regional, distant, unknown stage). All leukemias were classified as distant stage. Summary stage is derived from the best available data on clinical and pathological extent of disease.¹⁰ We excluded cases diagnosed on death certificate only, and in situ cancers. Cancer registry information on sex, race, and ethnicity is abstracted from medical records. Institutional Review Board approval was not required for this analysis.

Statistical analysis

Relative survival estimates

We estimated 2- and 5-year relative survival by cancer type and stage for cancers diagnosed during 2009-2018 and 2009-2015, respectively, followed until 2020. Analyses were carried out using SEER*Stat version 8.4.3. Annual percent changes (APCs) in 2- and 5-year relative survival were estimated with Joinpoint regression.¹¹ Joinpoint regression identifies calendar years where statistically significant changes in the trends have occurred.

Estimating the prevented number of cancer deaths

Figure S1 presents an overview of the statistical methods. The goal of this analysis was to compare the number of observed cancer deaths that occurred among people diagnosed during 2010-2019 and followed through 2020 with the expected number of cancer deaths that would have occurred in each subsequent year if stage-specific mortality had not changed after 2009. This approach approximates the impact of survival improvements on the reduction in cancer deaths among people diagnosed with cancer during 2010-2019 by controlling for changes in incidence and summary stage distribution.

Specifically, we let $D_i=1$ if person i dies from cancer within $\tau$ years of diagnosis, and $D_i=0$ otherwise. Then $D_i$ has a binomial distribution with $P\left(D_i=1\right|{\tau ,\mathit{X}}_i),$ where ${\mathit{X}}_i$ denotes a vector of personal and clinical characteristics of each individual. For a population of N individuals, the expected number of cancer deaths (CD) within τ years from diagnosis is then given by

$$\widehat{CD}={\sum }_{\mathrm{i}}^{\mathrm{N}}P\left(D_i=1\right|{\tau ,\mathit{X}}_i).\mathrm{ }$$ (1)

We modeled $P\left(D_i=1\right|{\mathit{X}}_i) \mathrm{using} \mathrm{t}$he absolute risk, or cumulative incidence,¹² of dying from cancer (cause 1) within $\tau$ years of diagnosis for a person with predictors ${\mathit{X}}_i$:

$$r\left(\mathrm{\tau }\right|\mathit{X})={\int }_0^{\mathrm{\tau }}{\mathrm{\lambda }}^1\left(u\right|\mathit{X})\exp \left[-{\int }_0^u\{{\mathrm{\lambda }}^1\left(s\right|\mathit{X})+{\mathrm{\lambda }}^2(s\left|\mathit{X}\right)\}\mathrm{d}s\right]\mathrm{d}u.\mathrm{ }$$ (2)

In Equation 2, ${\lambda }^1\left(u\right|X)$ is the cause-specific hazard for the main cause of interest, death from cancer, and ${\lambda }^2\left(s\right|X)$is the hazard for competing events (ie, all other deaths), which is also a function of the predictors $\mathit{X}\mathit{ }$or a subset of $\mathit{X}$. Each hazard was modeled as the product of a nonparametric baseline hazard function and a relative risk term:

$${\mathrm{ }\mathrm{\lambda }}^k\left(t\right|X)={\mathrm{\lambda }}_0^k(t)\exp \left({\mathit{X}}^{\mathit{T}}{\mathbf{\beta }}_{\mathit{k}}\right),k=\mathrm{1,2}.$$ (3)

To estimate $r\left(\tau \right|\mathit{X})$ in (1), we first estimate the relative risk parameters in (3) using two Cox proportional hazards regression models, one for cancer death and one for competing deaths based on data from people with cancer diagnosed in SEER during 2004-2019. We fit separate models by sex for each cancer type, additionally stratified by cancer stage (to limit the impact of screening-related stage shifts). The baseline hazards ${\lambda }_0^k\left(t\right), k=\mathrm{1,2} \mathrm{in}\mathrm{ }\left(3\right)$were estimated using the Breslow estimator. We used the following predictors in each hazard model: age at cancer diagnosis (spline), year of cancer diagnosis (single years), and race and ethnicity (Hispanic, non-Hispanic Black, non-Hispanic White, other [ie, Asian, Pacific Islander, American Indian and Alaska Native]/unknown/multiple). We adjusted for race and ethnicity, because there are established differences in cancer survival across racial and ethnic groups in the United States, potentially reflecting delayed diagnoses, access to care, and other factors.¹³ We assessed the proportionality assumption for the variables used in the Cox models by visual inspection of the hazards plots and found no evidence of strong violations.

We then predicted the $\tau -$year risk for patients in the cohort diagnosed in ≥2010, keeping all predictors fixed, but setting year of diagnosis to 2009. The projection period $\tau$ was defined as the difference between 2020 and diagnosis year (eg, for a person diagnosed in 2010, $\tau$ =10 years). We compared the expected number of cancer deaths, $\widehat{CD}$ in (1) for patients diagnosed in 2010 or later with the observed numbers, O, for that cohort and calculated the percent change in deaths as (O-$\widehat{CD})/\widehat{CD}$)). Total reductions in cancer deaths among SEER-17 cancer patients were estimated by summing sex- and stage-specific estimates of the number of cancer deaths prevented across cancer types. The national number of cancer deaths averted was estimated by dividing estimates by the fraction of the population in the SEER-17 catchment area (0.265).¹⁴ All statistical tests were 2-sided and bootstrap-based confidence intervals (CIs) for all quantities were computed based on resampling individuals within each cancer, sex, and stage category (P < .05 considered to be statistically significant). All calculations were conducted in SAS 9.4 and R.

Results

During 2010-2019, 1 660 589 cancers occurred among females and 1 649 681 cancers among males living in 17 regions of the United States. Sixty-seven percent of cancer patients were White, 12.5% Hispanic, 10.5% Black, 7.7% Asian American or Pacific Islander, and 0.6% American Indian or Alaska Native. The most common incident cancers were female breast (n = 517 000), prostate (n = 474 411), and lung cancer (n = 359 260; Table S1). Stage distributions differed by cancer type (eg, 55.2% of lung cancers and 51.7% of pancreatic cancers were distant stage, compared with only 4.0% of melanomas and 5.9% of female breast cancers). The fraction of cancer patients who had any cancer treatment recorded in the registry data remained relatively constant (78.9% in 2009 to 77.7% in 2016-2019).

Trends in relative survival, 2009-2019

For all cancers combined, 2-year (2013-2018: APC = 0.15%/year, P = .1) and 5-year relative survival estimates (2013-2015: APC = 0.27%/year; P = .2) for local stage disease remained stable in recent years, whereas 2-year and 5-year relative survival estimates for regional (2-year, 2014-2018: APC = 0.78%/year and 5-year, 2013-2015: APC = 1.1%/year; P < .05), distant (2-year, 2014-2018: APC = 1.8%/year and 5-year, 2013-2015: APC = 2.6%/year; P < .05), and unknown stage cancers (2-year, 2013-2018: APC = 4.4%/year; 5-year 2012-2015: APC = 2.0%/year; P < .05) improved statistically significantly in the most recent time periods (Table S2).

Association between cancer diagnosis year and cancer mortality

There were statistically significantly decreasing trends in HRs with increasing calendar year of diagnosis (P_trend < .05), representing improved survival for most combinations of cancer type and stage (Figure S2). For example, when diagnosis year 2019 was compared with 2009, the risk of death was 34% (HR = 0.66; 95% CI = 0.56 to 0.78) lower for males with distant stage melanoma, and 26% lower (HR = 0.74; 95% CI = 0.64 to 0.86) for females with localized stage liver cancer.

SEER-17 estimates of changes in cancer deaths due to stage-specific survival improvements

Among cancer patients diagnosed during 2010-2019 in the 17 SEER regions, there were 460 130 cancer deaths among females and 506 603 among males through 2020. Improvements in stage-specific cancer survival during this period resulted in a 4.7% (95% CI = −5.3% to −4.2%) decline in cancer deaths in females (22 874 fewer deaths; 95% CI = −25 562 to −20 256) and a 4.4% (95% CI = −4.9% to −3.9%) decline in males (23 198 fewer deaths; 95% CI = −26 193 to −20 369; Figure 1, Table S3). The largest declines were for distant stage disease (19 188 fewer deaths, 95% CI = −21,559 to −16,612), followed by regional (−12 300, 95% CI = −14 154 to −10 131), localized (−10 664, 95% CI = −12 876 to −8720), and cancers of unknown stage (−3920, 95% CI = −5250 to −2900; Figure 2, Table S4).

Figure 1.

Reduction in cancer deaths due to improvements in survival among cancers diagnosed in 2010-2019 in SEER-17 by sex, cancer, and SEER summary stage. Bars represent the difference between the number of observed and expected cancer deaths by cancer type and stage. Lines represent 95% confidence intervals. All leukemias were classified as distant stage.

Figure 2.

Percent change in cancer deaths due to improvements in survival among cancers diagnosed in 2010-2019 in SEER-17 by sex and cancer. Bars represent the percent change in the number of cancer deaths calculated as the number of observed cancer deaths minus the number of expected cancer deaths, divided by the number of expected cancer deaths, and lines represent 95% confidence intervals.

Among males in SEER-17, the largest absolute declines in cancer deaths were observed for lung cancer (6081 fewer deaths than expected [95% CI = −7070 to −5077]; 46% of deaths avoided were among those diagnosed at distant stage) and liver cancer (2597 fewer deaths; 95% CI = −3242 to −1897); 55% of deaths avoided were among those at localized stage), and the largest relative declines were observed for melanoma (−13.9%; 95% CI = −18.7% to −11.4%) and leukemia (−8.8%; 95% CI = −11.8% to −5.3%; Figures 1 and 2, Table S3). Among women, the largest absolute declines in cancer deaths were observed for lung cancer (7701 fewer deaths [95% CI = −8663 to −6743]; 45% of deaths avoided were distant stage) and breast cancer (1535 fewer deaths [95% CI = −2610 to −438], 49% of deaths avoided were distant stage); however, the statistically significant declines in females with breast cancer were restricted to distant stage. The largest relative declines among females were observed for melanoma (−15.6%, 95% CI = −22.3% to −10.8%) and leukemia (−10.4%; 95% CI = −13.8% to −6.5%). The number of cancer deaths prevented was not statistically significant for those diagnosed with colorectal or prostate cancer.

US estimates of changes in cancer deaths due to stage-specific survival improvements

When weighted to the full US population, approximately 173 900 fewer cancer deaths (86 300 in women; 87 500 in men) occurred among individuals diagnosed with cancer during 2010-2019 (Figure 3). Among men, 26.2% of the reduction occurred among people with lung cancer, 11.2% liver cancer, and 37.3% all other cancers not specifically analyzed. Among women, 33.7% of the reduction occurred among people with lung cancer, 6.7% breast cancer, and 33.7% all remaining cancers.

Figure 3.

Estimated reduction in cancer deaths nationally due to improvements in survival among cancers diagnosed in 2010-2019 by sex and cancer. Bars represent the difference between the number of observed and expected cancer deaths by cancer type, and lines represent 95% confidence intervals. Estimates were upweighted from SEER-17 estimate by dividing by .265, as SEER-17 represents 26.5% of the US population.

Discussion

We estimated that improvements in stage-specific survival resulted in nearly 174 000 fewer cancer deaths in the United States among individuals diagnosed with cancer during 2010-2019 than would have been expected based on 2009 survival rates. There was a 4.7% reduction in the number of cancer deaths among women, and a 4.4% reduction in men. The risk of cancer death declined with year of diagnosis for most combinations of stage at diagnosis and cancer type. The largest relative reductions in cancer deaths were observed for melanoma and the largest absolute declines for lung cancer. These prevented cancer deaths can be attributed, at least in part, to improvements in cancer treatment and expanded receipt of treatment over this relatively short period of time. No statistically significant reductions in stage-specific cancer deaths were observed for colorectal or prostate cancers.

US cancer mortality rates declined 14% from 2010 to 2019,¹⁵ reflecting a combination of decreased cancer incidence for some cancer types, mainly lung and colorectal cancer, and increased survival for some cancer types. Here, we estimated the independent contribution of advances in cancer treatment to declines in cancer mortality. For some cancer types, we estimated large declines in cancer deaths among those diagnosed during 2010-2019, including leukemia, NHL, melanoma, and cancers of the liver and lung. Notably, increases in stage-specific lung cancer survival resulted in an estimated 52 000 fewer cancer deaths nationally among individual diagnosed during 2010-2019, 30% of the total decline. Non-small cell lung cancer death rates declined more than twice as fast as incidence rates during 2013-2016, indicating survival improvements driven by targeted therapies (for example, against EGFR and ALK oncogenes).⁶ Immunotherapy to treat metastatic NSCLC was approved by the FDA in 2015, and may have contributed to improved survival in more recent years. Although these therapies are used to treat metastatic disease, we also observed a 15.5% decline in cancer deaths among people diagnosed with localized stage lung cancer, which may be due to advances in surgery or radiotherapy.¹⁶

We estimated a 15% decline in cancer deaths among people diagnosed with melanoma during 2010-2019. This has been largely attributed to notable treatment advancements since 2011,¹⁷ including targeted treatments and multiple immune checkpoint inhibitors, resulting in a 6% per year decline in melanoma mortality during 2013-2017.¹⁸ Earlier detection due to increased surveillance could also have contributed to reductions in mortality, with further downstaging even within localized stage disease.¹⁹ Notable treatment-related survival improvements have also been reported for certain types of lymphomas and leukemias,^20,21 although these malignancies are composed of several specific cancer types with different treatment strategies and prognoses, and shifts in subtypes may also contribute.

For most cancer types examined, the largest absolute declines in cancer deaths were observed among those with distant stage disease; however, for liver cancer, the declines among those with localized stage disease were the largest, perhaps reflecting improvements in treatment allocation algorithms and surgical resection or detection of smaller localized tumors, allowing more definitive treatment.²² Transplantation significantly improves survival of localized stage liver cancer patients; however, the insufficient availability of donated organs has limited the impact on liver cancer mortality rates at a population level.²³

Although the second largest absolute decline in cancer deaths among females occurred in those diagnosed with breast cancer, the relative decline was modest (2.6%) and restricted to those with distant stage disease. There have been notable declines in national breast cancer mortality rates in the United States due to a combination of early detection and treatment, with treatment estimated to have accounted for three-quarters of the improvement.^2,4 Prior work found that advances in adjuvant treatment contributed more to mortality declines than screening for ER-positive breast cancers, with more equal contributions for ER-negative breast cancers in 2000.²⁴ Continued improvements in chemotherapy and targeted therapies have likely contributed to the improved survival outcomes for patients with both early and later stage breast cancer.²⁵ Many of the recent approvals are likely to continue to improve survival in breast cancer patients.

Colorectal and prostate cancers are leading causes of cancer incidence and mortality in the United States. Unfortunately, we estimated no statistically significant reduction in cancer deaths due to stage-specific survival improvements.⁹ Although we did estimate a 3.2% reduction in cancer deaths among pancreatic cancer patients, there have not been major therapeutic advances over the past 2 decades, although targeting mutant KRAS, currently in clinical trials, may have the potential to improve the outlook for this tumor.^2,26 It is possible that these survival gains have been due to earlier diagnosis due to incidental detection and increased use of imaging in the absence of population-wide screening. Although our analysis accounts for staging shifts over time, downstaging within localized, regional, and distant stage disease could also have influenced our estimates. Given the large burden of colorectal, prostate, and pancreatic cancers in the US population, treatment advances with widespread access and uptake could have a large impact on cancer mortality rates.

It is important to note that our study aimed to estimate the impact of treatment improvements on cancer mortality over this recent 10-year time period. These estimates do not capture the impact of previous treatment advances, which have resulted in exceptional survival rates for many common cancers. For example, in 2009, the 5-year-relative survival for breast cancer was 99% for localized stage and 86% for regional stage disease, 91% for localized stage, and 71% for regional stage for colorectal cancer and 100% for both localized and regional stage for prostate cancer. In addition, some of the recent treatment improvements have focused on modifying or omitting treatment to reduce side-effects, which are important advances that do not translate to temporal improvements in cause-specific survival.

Increased access to standard of care treatment could result in declines in cancer mortality that are also due to the broader use of existing treatments with demonstrated benefit. For example, in a recent analysis of the National Cancer Database, increases in the use of chemotherapy to treat pancreatic cancer and neoadjuvant systemic therapy to treat breast cancer were reported from 2010 to 2019.²⁷

The current study aimed to isolate the impact of treatment advances on the number of cancer deaths prevented, by removing the impact of cancer prevention and reducing the impact of early detection from our estimates. Cancer prevention and early detection have had a large impact on favorable trends in cancer mortality in the United States. For example, there has been a sustained decline in the prevalence of cigarette smoking, which resulted in large decreases in lung cancer incidence and mortality,²⁸ as well as in other smoking-related cancers. There is also evidence that smoking cessation improves prognosis, although the proportion of cancer patients who quit smoking remains low.²⁹ Early detection of cancer through screening has also affected cancer mortality rates in the United States. Screening for colorectal cancer and removal of precancerous polyps has resulted in national declines in colorectal cancer mortality.³⁰ There are also population-based screening tools and recommendations for other cancer types, including breast, lung, prostate, and cervix.^31-34 Increasing use of screening would accelerate mortality reductions, particularly for lung cancer, where uptake remains very low, and colorectal cancer, where even modest increases in uptake could substantially reduce mortality.^35,36 To our knowledge, there have not been comparable analyses that examined the impact of recent improvements in prevention and screening on cancer mortality or whether noteworthy mortality reductions have mainly been a continuation of practices developed before 2010.

Given simultaneous progress in prevention, early detection, and treatment for many cancer types, it is difficult to completely disentangle the impact of advances in cancer treatment specifically on cancer mortality. Here, we used stage-specific survival improvements as a proxy for treatment benefits. Our counterfactual approach completely removes the impact of changes in cancer incidence on mortality rates by comparing observed data with an alternate scenario where the number of incident cancer cases is the same, but no stage-specific survival improvements (as measured by summary stage) occurred after 2009. Focusing on stage-specific estimates reduces the impact of screening- or detection-induced stage shifts; however, a main limitation of our analysis is that we could not account for downstaging within each broad stage grouping, which could also contribute to apparent improvements in stage-specific survival. Clinical staging systems are more granular and offer better differentiation in tumor characteristics and prognosis, but they are cancer-specific and frequently updated, precluding their use in this analysis. Thus, we used SEER summary stage in this analysis, as definitions have been relatively stable over time. In addition, when we upweighted our results to the full US population, we made the assumption that the roughly one-quarter of the population included in SEER-17 is representative of people with cancer nationwide, although we know that the SEER program oversamples certain demographic groups.¹⁴

Further decreases in US cancer mortality rates will require sustained progress and further investments in cancer prevention, screening, and treatment. Here, we estimated that there were 174 000 fewer cancer deaths due to advances in cancer therapies among people diagnosed with cancer during 2010-2019. This reduction in deaths reflects improvements in treatment occurring after 2009 and thus are in addition to the many deaths averted as a result of treatment approaches developed in prior years. This is a substantial number of lives saved; however, accelerated therapeutic advances and increases in access to treatment should remain a priority, particularly for cancers with poor survival, and those where there has been little progress in prevention and early detection.

Author contributions

Meredith Shiels (Conceptualization, Methodology, Supervision, Visualization, Writing—original draft), Neal Freedman (Conceptualization, Methodology, Writing—review & editing), Anika Haque (Formal analysis, Writing—review & editing), Amy Berrington de Gonzalez (Writing—review & editing), Stanley Lipkowitz (Writing—review & editing), Douglas Lowy (Conceptualization, Writing—review & editing), and Ruth Pfeiffer (Conceptualization, Formal analysis, Methodology, Writing—review & editing)

Supplementary material

Supplementary material is available at JNCI: Journal of the National Cancer Institute online.

Funding

This work was funded by the Intramural Research Program of the National Cancer Institute and the Institute of Cancer Research.

Conflicts of interest

The authors have no conflicts of interest to declare.

Data availability

Data available upon request from the National Cancer Institute’s Division of Cancer Control and Population Sciences.