Preliminary Estimates of New Invasive Cancers Diagnosed in 2023 in the United States

Abstract

Objective:

Incidence of new cancer cases is tracked by population-based cancer registries. Collecting complete information from initial diagnosis through first course of treatment can take up to 2 years. In June 2025, the United States (US) Centers for Disease Control and Prevention (CDC) released reports of cancer incidence for cases diagnosed through 2022, the most recent diagnosis year for which complete cancer incidence data were available. Having more timely data may be useful in planning cancer prevention and control efforts, and preliminary incidence estimates may help fill that gap.

Materials and methods:

We estimated preliminary rates and counts of new cancer cases in 2023 using reported US Cancer Statistics data for 2000 to 2019 and age-period-cohort models to account for changes in population size, age distribution, and cancer risk. Estimates were calculated for all cancers combined and 23 common cancers by sex (male and female). As a validation of the model, we compared observed and modeled rates for 2022.

Results:

In 2023, an estimated 1,967,906 new cancer cases were diagnosed in the US, with an age-adjusted preliminary rate of 462 cases per 100,000 standard population. In a validation analysis, observed and modeled rates for 2022 were similar (0.5% percentage difference, on average), indicating that the age-period-cohort models performed well.

Conclusion:

Cancer registries are currently implementing data modernization strategies to improve timeliness of data. In the interim, preliminary incidence estimates can fill in gaps between cancer occurrence and cancer reporting. These early cancer rates and counts may be useful in guiding research and helping public health care professionals focus on areas of concern for cancer prevention and control efforts.

Introduction

Accurate and timely counts and rates of cancer incidence can help support public health actions, better resource planning, and informed policy decisions.¹ Population-based cancer registries contribute to this effort by collecting information about new cancer cases from multiple healthcare facilities and laboratories.¹ A complete case record includes data from initial diagnosis through the first course of treatment, and collection and consolidation of information from different sources can take up to 2 years.² As a result, there is a lag from when a cancer case is diagnosed to when it is included in reports.² For example, in June 2025, the Centers for Disease Control and Prevention (CDC) released reports of cancer incidence for cases diagnosed through December 2022, the most recent diagnosis year for which complete cancer incidence data were available.³ To improve reporting timeliness, cancer registries across the US began implementing data modernization strategies recommended by CDC.² In the interim, preliminary estimates may help fill the gap between cancer occurrence and cancer reporting.⁴

The objective of our study was to estimate age-adjusted incidence rates and counts in 2023 for all cancers combined and for 23 common cancers. We used models incorporating historical cancer incidence data, current population estimates, and population projections. These models also accounted for changes in population size, age distribution, and cancer risk.

Methods and Data Sources

We obtained 20 years of historical US cancer statistics incidence data from 2000 to 2019.³ The data are from registries that participate in CDC's National Program of Cancer Registries (collected under OMB contract # 0920-0469) and the National Cancer Institute's (NCI) Surveillance, Epidemiology, and End Results Program.³ The combined data from the 2 programs cover approximately 99% of the US population. All data must meet high-quality standards to be published in the US cancer statistics. While data from 2020–2022 were available in the US cancer statistics dataset, incidence data for this analysis were restricted to a 20-year period ending in 2019 because the analytic method we used requires calendar periods to be 5 years.^5,6 The 5-year calendar period containing the year 2020 would have been skewed because of low diagnoses resulting from disruptions to healthcare use during the onset of the COVID-19 pandemic.⁷

We included all invasive cancers in our analysis. Primary anatomic site and histology data were coded according to the International Classification of Diseases for Oncology, Third Edition (ICD-O-3).^8,9 Cases were then categorized according to Site Recode ICD-O-3/WHO 2008¹⁰ into 23 common cancer types (Table 1). All remaining cancer types were combined into a single category; this category was included to calculate all cancer types combined but is not presented in tables or figures.

Table 1.

Preliminary ^a Number and Age-Adjusted Rate ^b for New Invasive Cancer Diagnoses by Cancer Type and Sex, United States, 2023

	Male and Female		Female		Male
Cancer Type	Count	Rate	Count	Rate	Count	Rate
All cancer sites combined	1,967,906	462.2	970,903	443.3	997,003	491.8
Brain and other nervous system	24,676	6.4	10,842	5.4	13,834	7.5
Cervix	13,915		13,915	7.8
Colon and rectum	157,435	38.1	73,744	33.9	83,691	42.7
Corpus and uterus, NOS	68,468		68,468	30.5
Esophagus	20,216	4.5	4,327	1.8	15,889	7.7
Female breast	291,687		291,687	136.7
Hodgkin lymphoma	8,717	2.5	3,962	2.3	4,755	2.7
Kidney and renal pelvis	78,229	18.7	28,363	13.0	49,866	25.1
Larynx	12,054	2.7	2,526	1.1	9,528	4.6
Leukemias	60,582	14.8	24,988	11.6	35,594	18.7
Liver and intrahepatic bile duct	42,077	9.2	12,314	5.1	29,763	13.9
Lung and bronchus	233,709	51.8	117,887	48.5	115,822	56.2
Melanomas of the skin	105,329	25.3	43,541	20.7	61,788	31.6
Myeloma	34,271	7.8	15,280	6.5	18,991	9.4
Non-Hodgkin lymphoma	80,653	19.1	36,125	16.0	44,528	22.9
Oral cavity and pharynx	54,320	12.6	15,461	6.9	38,859	19.0
Ovary	21,717		21,717	10.2
Pancreas	63,609	14.5	30,743	13.0	32,866	16.3
Prostate	251,370				251,370	114.5
Stomach	28,290	6.6	11,081	5.0	17,209	8.7
Testis	9,614				9,614	6.0
Thyroid	48,895	13.5	35,257	19.5	13,638	7.4
Urinary bladder	83,290	18.9	19,655	8.2	63,635	32.2

NOS not otherwise specified.

^aPreliminary incidence rates and counts for 2023 are for the 50 states and the District of Columbia and are based on age-period-cohort models using cases reported from 2000 to 2019 by cancer registries meeting U.S. Cancer Statistics data quality criteria, covering approximately 99% of the US population.

^bEstimated rates are the number of cases per 100,000 people and are age-adjusted to the 2000 US standard population (18 age groups - Census P25-1130).

Incidence data were obtained by calendar year (2000–2019), age (18 age groups: 0 to 4 years, 5 to 9 years, 10 to 14 years, 15 to 19, and so on through 85 years and older), sex (male and female), and cancer type (23 common cancers and all remaining cancer types combined).

We obtained current population estimates from the NCI-modified population estimates files for 2000 to 2023 by year, sex, and age.¹¹ We obtained population projections for 2024 to 2029 from the US Census Bureau population projections files by year, sex, and age.¹² Population estimates and projections were used as denominators in rate calculations.

Statistical Analysis

To account for changes in population size, age distribution, and cancer risk, we used Nordpred functions available in R software¹³ to create age-period-cohort regression models.^5,6 Separate models were fit for each cancer site (23 common cancers and all remaining cancer types combined) by sex (male and female) with data aggregated into 4 calendar periods of 5 years each (2000–2004, 2005–2009, 2010–2014, 2015–2019): R_ap = (A_a + D•p + P_p + C_c)⁵ in which the dependent variable R_ap is the incidence rate in age group a in calendar period p; A_a is the age component for age group a; D is the net drift parameter (the common linear effect of both calendar period and birth cohort); P_p is the nonlinear period component of period p; and C_c is the nonlinear cohort component of cohort c. A net drift of 0 indicates that the trend for age-adjusted rates remained stable over time. A positive net drift indicates that age-adjusted rates increased over time, and a negative net drift indicates that age-adjusted rates decreased over time. We included 2 projection periods of 5 years each (2020–2024 and 2025–2029). We based projections on long-term trend data unless there was statistically significant curvature (P<.05) in the trend over time, in which case the linear drift component was based on the most recent 10-year period. The starting age group for each cancer type and sex group contained at least 10 cases. For age groups below this limit, projected rates were estimated using the average rates in the last 2 calendar periods. Estimates for all cancers combined were obtained by summing across cancer types. Estimates for both sexes were obtained by summing estimates for males and females.

We used 2 modifications that have been shown empirically to improve predictions.⁶ First, to offset exponential increases or decreases in incidence rates, we used the Power-5 link function. Second, under an assumption that trends are not likely to continue indefinitely,^5,6 the drift component D was reduced by 25% in the second projection period for most cancer types; this is the default reduction in the Nordpred function.¹³

After we assessed initial results, we modified the net drift parameter D for several cancers. First, we flagged cancer types for which predictions were not very accurate (more than a 10% difference in counts or rates; see methods for validation analysis below). These cancer types included esophagus, kidney, ovary, prostate, liver, and stomach. We examined year-specific data, including the most recent years (2020–2022) that were not part of the input dataset. We found that trends had leveled off substantially in recent years for these cancer types. Therefore, we increased the cut in the trend through the net drift parameter. We used the reduction values (0%, 25%, 50%, 75%, or 100%) recommended in the Nordpred function.¹³ For cancers of the esophagus, kidney, and ovary, we reduced the net drift parameter D by 50% for the first projection period and 75% for the second projection period. For cancers of the prostate, liver, and stomach, we reduced the net drift parameter D by 100% for both projection periods. These modifications resulted in better predictions.

Preliminary estimates for rates and counts were calculated by cancer type (all cancers combined and 23 common cancers) and by sex (both sexes, male, and female). Annual age-specific rates for each cancer type and sex were calculated by taking a weighted average of the age-specific rates of the current, prior, and subsequent 5-year periods; weights were determined using constants for the position of the year in the current 5-year period. For example, constants for the age-specific rates for 2022 were 5 (weight=1) for the current period (2020–2024) and 0 (weight=0) for both the prior period (2015–2019) and the subsequent period (2025–2029). Constants for the preliminary age-specific rates for 2023 were 4 (weight=.8) for the current period (2020–2024), 0 (weight=0) for the prior period (2015–2019), and 1 (weight=.2) for the subsequent period (2025–2029). The preliminary age-specific case counts by cancer type and sex were calculated by multiplying age-specific rates by age-specific population denominators. The preliminary age-specific, cancer-specific case counts for both sexes combined were calculated by summing across male and female. The preliminary age-adjusted incidence rates by cancer type and sex were age-adjusted to the 2000 US standard population using 18 age groups.¹⁴

As a validation analysis, for cases diagnosed in 2022, we compared modeled counts and age-adjusted rates for each cancer type by sex with their observed counts and age-adjusted rates. Observed counts and age-adjusted rates were adjusted for delayed reporting.¹⁵ We calculated the percentage difference between the modeled and observed estimates, with negative values indicating that the modeled estimate was less than observed and positive values indicating that the modeled estimate was more than observed. We flagged cells that had more than a 10% difference in counts or rates. We calculated the average percentage difference in counts and rates for all combinations of cancer type and sex as the sum of the percentage differences divided by 60 (the number of combinations).

Analyses were conducted using SEER*Stat software version 9.0.41.4 (National Cancer Institute), SAS version 9.4 (SAS Institute Inc), and R software version 4.5.0 (The R Project for Statistical Computing). This activity was reviewed by CDC, deemed not research, and conducted so as to be consistent with appliable federal law and CDC policy.

Results

In 2023, in the United States, an estimated 1,967,906 new cancer cases were diagnosed, with an age-adjusted preliminary rate of 462 cases per 100,000 standard population. Among females, an estimated 970,903 new cancer cases were diagnosed, with an age-adjusted preliminary rate of 443 cases per 100,000 standard population. Among males, an estimated 997,003 new cancer cases were diagnosed, with an age-adjusted preliminary rate of 492 cases per 100,000 standard population.

The five most common cancers expected to occur in 2023 were female breast (291,687 cases), prostate (251,370 cases), lung and bronchus (233,709 cases), colon and rectum (157,435 cases), and melanoma (105,329 cases). These five cancers together accounted for 53% of all cancers (1,039,530 cases).

In a validation analysis, the modeled rates for 2022 were very close to the observed rates, with a mean difference of 0.5% across all sex and cancer type combinations. For all cancers combined, the modeled rate was 0.77% lower than the observed rate, with a larger difference for males (-1.43%) than females (-0.30%) (Figure 1). Comparisons of counts were similar to those of rates. Among the 23 cancer types (4 female specific, 2 male specific and 17 for both sexes combined) shown in Figure 2, the percentage difference varied by cancer type and was highest for prostate cancer (-12.45%) and otherwise ranged from -4.53% for stomach cancer for both sexes combined to 6.80% for cancers of the corpus and uterus (Figure 2). Compared to observed rates, modeled rates were >5% higher for corpus and uterus; between 2% to 5% higher for liver and intrahepatic bile duct, non-Hodgkin lymphoma, thyroid, lung and bronchus, kidney and renal pelvis, cervix, and pancreas; <2% higher for urinary bladder, myeloma, brain and other nervous system, larynx, melanomas of the skin, testis, oral cavity and pharynx; <2% lower for colon and rectum, female breast, and ovary; and between 2% to 5% lower for Hodgkin lymphoma, esophagus, leukemias, and stomach. Besides prostate cancer, the percentage difference between modeled and observed rates was >10% only for stomach cancer among females (-11.4%; data not shown).

Figure 1.

Percentage Difference Between US Modeled and Observed Counts and Age-Adjusted Incidence Rates for All Invasive Cancer Cases Diagnosed in 2022 by Measure and Sex

The gray bars represent the magnitude of the percentage difference between the modeled and observed estimates, with negative values indicating the modeled estimate was less than observed and positive values indicating the modeled estimate was more than observed. Observed counts and rates were based on new invasive cancer cases diagnosed in 2022 and reported by cancer registries meeting US Cancer Statistics data quality criteria covering approximately 100% of the US population. Modeled counts rates were based on age-period-cohort models using cases reported from 2000 to 2019 by cancer registries meeting US Cancer Statistics data quality criteria, covering approximately 99% of the US population. Rates are the number of cases per 100,000 people and are age-adjusted to the 2000 US standard population (18 age groups—Census P25-1130). Observed counts and rates were also adjusted for reporting delay (https://surveillance.cancer.gov/delay/).

Figure 2.

Percentage Difference Between US Modeled and Observed Age-Adjusted Incidence Rates for Invasive Cancer Cases Diagnosed in 2022 by Cancer Type

The gray bars represent the magnitude of the percentage difference between the modeled rate and the observed rate, with negative values indicating that the modeled rate was less than the observed rate and positive values indicating that the modeled rate was more than the observed rate. Observed rates were based on new invasive cancer cases diagnosed in 2022 and reported by cancer registries meeting US Cancer Statistics data quality criteria covering approximately 100% of the US population. Modeled rates were based on age-period-cohort models using cases reported from 2000 to 2019 by cancer registries meeting US Cancer Statistics data quality criteria covering approximately 99% of the US population. Rates are the number of cases per 100,000 people and are age-adjusted to the 2000 US standard population (18 age groups—Census P25-1130). Observed rates were also adjusted for reporting delay (https://surveillance.cancer.gov/delay/).

Discussion

Using historical cancer incidence data and age-period-cohort models to account for changes in population size, age distribution, and cancer risk, we estimated preliminary age-adjusted incidence rates and counts for 2023 for all cancers combined and for 23 common cancers. We estimate that 1.97 million new cancer cases were diagnosed in 2023. Female breast, prostate, lung and bronchus, and colon and rectum cancers and melanoma accounted for more than half of these cases. These preliminary data can be used to support public health actions, resource planning, and informed policy decisions.

Our validation analysis indicated that these age-period-cohort models performed well for all cancers combined and for most cancers. The models resulted in higher than expected estimates for several cancers, which could be due to rates not increasing at the same pace or rates decreasing at a more rapid pace. For example, we estimated a higher rate of liver and intrahepatic bile duct cancer than observed; incidence of this cancer has begun to decline after years of increase, due in part to changes in exposure to risk factors, such as hepatitis B and C viruses.¹⁶ The models resulted in lower than expected estimates for several cancers, particularly prostate cancer and stomach cancer among females; this could be due to rates not decreasing at the same pace or a change in direction of trends. For example, stomach cancer incidence has risen sharply recently due to a change in the World Health Organization's classification of gastrointestinal stromal tumors,¹⁶ but this recent and abrupt change in trend was not adequately captured by our model. We also found that fluctuations in prostate cancer incidence, due in part to changes in the use of the prostate antigen test,¹⁷ were not captured well by our model. Additionally, we used observed rates as input in our models; using delay-adjusted rates may improve fit for some models, such as for leukemia, which tends to have a large reporting delay.

Our analysis had several other limitations. First, estimated counts and rates are model-based projections. Estimates can be useful for planning purposes, but observed counts and rates reported by population-based cancer registries are still considered the gold standard. Second, for reasons described in the methods, we chose to truncate the last year of observed data used in our age-period-cohort models to 2019. Even though more recent years were not included in the model, results from the validation analysis indicated that the models still performed well. In future analyses, we plan to incorporate more recent years of diagnosis.

Our study had several strengths, including the use of national, high-quality, population-based cancer registry data. Additionally, we used age-period-cohort models, which have been validated in studies using long-term cancer incidence data.⁶ A similar approach could be used by central cancer registries, keeping in mind that sufficient case counts by strata such as sex and cancer type are needed to reliably fit models.

Conclusion

US cancer registries are currently implementing data modernization strategies to improve timeliness of data. In the interim, preliminary estimates can fill in gaps between cancer occurrence and cancer reporting. These early cancer rates and counts may be useful in guiding research and helping public health care professionals focus on areas of concern for cancer prevention and control efforts.