Declining Cancer Incidence in the Elderly: Decreasing Diagnostic Intensity or Biology?

The incidence of most cancers increases with age but typically declines after age 75–85 (1–9). There is no exhaustive explanation for this pattern, but proposed hypotheses include a natural selection of aged “survivors” with less lifetime carcinogen exposure, genetic cancer protective traits, and/or a reduction in somatic mutation rate over age (5, 10–12). The age-incidence pattern is however inconsistent between anatomical sites, countries, and over time, indicating that a universal, biological explanation is unlikely (5–9, 13, 14). Malignancies requiring invasive and potentially harmful diagnostic procedures, such as cancers of the lung and pancreas, typically decline more steeply, while cancers that are straightforward to diagnose using less invasive techniques, such as skin cancer, tend to increase continuously over age (3–5, 15, 16). Moreover, cancer peak age is reported to increase over calendar time and is higher and/or less pronounced in countries where cancer surveillance is complemented by backtracking missing cancer registrations from cause of death certificates (5, 7, 9, 13). Autopsy studies in elderly indicate underdiagnosis and/or underreporting of cancer in the very old (17–20).

To delineate and understand the determinants of cancer incidence in the elderly is important, not only to optimize and allocate health care resources in an aging population, but perhaps more importantly to improve the understanding of carcinogenesis. Nevertheless, the declining cancer incidence in the elderly has attracted surprisingly little attention and never been thoroughly investigated. Consequently, we performed a large, nationwide cohort study to further characterize the true cancer age-incidence pattern for a range of malignancies.

We conducted a population-based cohort study using individual-level data from the nationwide Swedish Cancer Register held by the Swedish National Board of Health and Welfare. Reporting is mandatory by law for both pathology and clinical departments, ensuring a high national coverage of approximately 95% (21, 22). In addition to clinical and demographic data, the Swedish Cancer Register contains information on whether the recorded cancer was detected “unexpectedly” or confirmed “expectedly” at autopsy (22). Using the national registration number, the Swedish Cancer Register is linked to the Swedish Cause of Death and Total Population Registers, adding dates of death, immigration, and emigration, thereby ensuring an unbiased complete follow-up. Aggregated age- and sex-specific background population and mortality counts including information on autopsies, were retrieved from Statistics Sweden and the Swedish National Board of Health and Welfare, respectively.

From the Swedish Cancer Register, we extracted all cases of pancreatic, lung, non-meningioma brain, anorectal, urinary bladder, breast, and prostate cancer, as well as all cases of non–Hodgkin lymphoma (NHL), and malignant melanoma of the skin, diagnosed in year 1970 to 2014, at age 20–99. We classified all cancers according to the 7th edition of the International Classification of Diseases (ICD-7) and, when relevant, the World Health Organization Histological Classification of Neoplasms (CANC/24.1; Supplementary Table S1; refs. 23, 24). To avoid a systematic exclusion of cancers in the elderly, we allowed for second primary tumors from different anatomic sites without restrictions (25, 26). To avoid duplicate registrations, cancers of the same origin were included if at least 2 years had passed between diagnoses in lung, anorectal, urinary bladder, NHL, melanoma skin, and breast cancer, meaning that multiple cases could be recorded for the same individual for these sites. We only included the first registration of pancreatic, non-meningioma brain, and prostate cancer. This decision was based on the clinical notion of cancer risk in the remaining and/or contralateral organ or tissue type after curative therapy in the first group of cancers while a succeeding cancer diagnosis in the second group more likely constitutes a local recurrence or duplicate registration. As an attempt to circumvent the effects of a likely lower diagnostic intensity in the elderly, we did not systematically exclude non pathology-confirmed cancers (3). For the same reason, we included cases with registered date of diagnosis up to 28 days after the date of death, recognizing that this likely reflects delay between death, autopsy, and final histopathologic confirmation.

In addition to grouping based on anatomic site and histology, cancers were also categorized on the basis of perceived diagnostic invasiveness. Here, pancreatic, lung, and non-meningioma brain tumors were classified as requiring “high invasiveness” diagnostic procedures; anorectal, urinary bladder, and NHL neoplasms as “medium invasiveness”; and melanoma skin, breast, and prostate cancer as “low intensiveness.” The high-invasiveness diagnostic group is distinguished by the fact that histologic verification is connoted with a substantial risk of severe complications (e.g., intracranial or intraabdominal hemorrhage and pneumothorax) or extensive surgery (15, 16, 27). Tumors in the medium-invasiveness group are commonly accessible via endoscopy or biopsy of palpable lymph nodes (28, 29). Diagnosing malignant melanoma, breast, and prostate cancer is in most cases straightforward by cutaneous, breast, or transrectal biopsy, respectively (30, 31).

All analyses were stratified by anatomical site. To highlight differences in “diagnostic invasiveness” between cancer sites, we estimated the numbers and proportions of cancers that were pathology confirmed (i.e., morphologically verified by biopsy and/or cytology), diagnosed at autopsy (i.e., unexpected autopsy findings), and with a survival of less than 1 month from diagnosis.

Age was categorized into 5-year age groups due to low number of cancer cases in the very young and very old, and also low population counts in the very old. Incidence rates (IR) per 100,000 person-years were calculated as the number of cancer cases divided by background population counts in strata of age, calendar year, and for breast and prostate cancer also sex. The reported cancer IR and the proportion of pathology-confirmed cases were plotted by age. Cancer “peak age” was defined as the age group with the highest IR. To quantify the cancer decline in elderly, we used Poisson regression to estimate the relative incidence drop, that is, incidence rate ratio, including 95% confidence intervals (95% CI) comparing peak age with age group 95–99, adjusting for sex and calendar year.

To contrast the reported and the autopsy-detected cancer age-incidence patterns, we calculated the proportion of performed autopsies where cancer was unexpectedly detected. The number of unexpected cancer findings in autopsy was divided with autopsy counts in strata of age, calendar year, and (breast and prostate cancer) sex. Cancer “peak age” was defined as the age group with the highest proportion of cancers detected in autopsy. When peak age did not coincide with the oldest age group, we quantified the autopsy-detected cancer relative decline using Poisson regression to estimate incidence rate ratios, including sex and calendar year in the models. The proportion of performed autopsies where cancer was unexpectedly detected was plotted by age.

We calculated a projected cancer incidence by age, assuming that the proportion of unexpected cancers would be the same in deaths not undergoing autopsy. The number of unexpected autopsy findings was divided by the number of performed autopsies and then multiplied with the number of deaths in strata of age and calendar period, and in breast and prostate cancer also sex. The “total” incidence rate was then estimated using the sum of the reported number of cancer cases and the “projected” number of incident autopsy cancers as the numerator.

In all figures, point estimates were accompanied by smooth estimates, predicted using restricted cubic splines with 4 or 5 degrees of freedom, in Poisson regression (incidence rates) or linear regression models (proportions). Model fit was validated comparing spline-derived with age-specific point estimates graphically. The significance level was set to 0.05 and all tests of statistical significance were two-sided.

All statistical analyses were performed using Stata Intercooled 15.1 (StataCorp, RRID: SCR_012763).

The study was approved by the Ethical Review Board in Stockholm, Sweden 2012/1010-31/3.

Raw data generated by the Swedish National Board of Health and Welfare, are considered sensitive and therefore not publically available due to Swedish laws and regulations. Derived, aggregated data supporting the study findings are available upon request.

From the Swedish Cancer Register, we identified a total of 1,020,397 cases of pancreatic, lung, non-meningioma brain, anorectal, urinary bladder, NHL, melanoma skin, breast, and prostate cancer (Table 1). The proportion of pathology-confirmed cases were substantially lower in cancers requiring high-invasiveness diagnostic procedures, ranging from 54% (pancreatic) to 75% (lung), as compared with low-invasiveness where it ranged from 93% (prostate) to 99% (melanoma skin cancer). A total of 29,132 (2.9%) of all registered cancer cases constituted unexpected autopsy-detected diagnoses. The range comparing low- to high-invasiveness cancers was wide and only 44 (0.1%) registered melanoma skin cancers were detected at autopsy, as compared with 5,897 (12%) pancreatic tumors. The proportion of cancer cases with survival less than a month ranged from 0.9% (melanoma skin) to 34% (pancreatic).

Figure 1 presents the reported incidence rate (left y-axis) together with the proportion of pathology-confirmed cases (right y-axis) by cancer type, over age. As expected, there was a general pattern of declining incidence in high age in all cancers, except melanoma skin, and generally less so among sites requiring low-invasiveness diagnostics. The age-incidence pattern was similar stratified by sex (Supplementary Fig. S1). The proportion of pathology-confirmed cancers decreased continuously over age for all malignancies, except melanoma skin cancer.

In Fig. 2, the proportion of performed autopsies where cancer was unexpectedly detected is plotted by age at death. There is no longer a consistent decline in the elderly except for a slight drop in incident lung and non-meningioma brain cancer. Of note, only a small number (n = 44) of melanoma skin cancer was discovered unexpectedly at autopsy, limiting interpretations.

In Table 2, we present the cancer peak age, IR at peak age, IR at age 95–99, and the relative IR decline comparing peak age with the oldest age group (age 95–99), adjusting for sex and calendar year. Here we analyzed both all reported incident cancers, as well as unexpected autopsy-detected cancers, the latter analysis using the corresponding number of performed autopsies as denominator. In the first analysis, the peak age ranged from 65–69 (non-meningioma brain cancer) to 95–99 years (melanoma skin cancer), but was mostly in the 80–84 year range (Table 2; Fig. 1). Furthermore, as is illustrated in Fig. 1, peak age was generally lower and the incidence decline larger in cancers requiring high-invasiveness diagnostics. Assessing peak age among unexpected autopsy findings, peak age corresponded with the oldest age group, with three exceptions: lung cancer peak age 75–79, non-meningioma brain cancer peak age 70–74, and NHL peak age 90–94, where the estimated incidence decline was only statistically significant in the first two (Table 2; Fig. 2). With the exception of melanoma skin, where only 44 unexpected autopsy findings were observed, lung and non-meningioma brain cancer, the proportion of autopsies with an unexpected cancer finding increased continuously and clearly with age for all sites (Fig. 2). As reported previously, a higher proportion of deaths in men (31%) than women (21%) were examined by autopsy and autopsy rates declined sharply with age and calendar year of death (Supplementary Fig. S2; ref. 32).We used the proportion of unexpected cancers in autopsy to project cancer incidence rates in the general population, assuming that deaths not undergoing autopsy would have the same rate of unexpected findings as those who did. The resulting estimates are presented together with the reported in Fig. 3, revealing sharply and continuously increasing cancer incidence with age for all sites.

In this large, population-based cohort study using Swedish Cancer Register and autopsy data to study age-incidence patterns in a range of solid tumors, we confirm a consistent decline in registered cancer incidence in the elderly that completely disappears or is substantially reduced when investigating autopsy-detected incident cancers. Our findings contradict the widespread hypothesis of healthy, cancer-resistant elderly and rather support that the reported cancer decline in the very old is an artifact driven by a reduction in reporting, diagnostic intensity, and/or access to cancer screening resources in older age groups (5, 10–12, 17–20).

Although findings were clear and consistent with our hypothesis that the cancer decline in high age is mostly driven by gradually declining diagnostic intensity, some weaknesses in our analyses deserve attention. First, we did not have access to detailed data on how diagnostic activities varied across lifespan. Instead, we base our assessment on the perceived invasiveness of diagnostic measures which was strongly associated with the incidence decline at old age. We did however see a substantial decline in the extent to which cancers were pathology confirmed in the elderly which in turn was especially marked for those that require highly invasive diagnostics.

Second, the classification of diagnostic invasiveness was based on joint clinical experience rather than solid scientific evidence and was hence extensively discussed among the co-authors (15, 16, 27–31). There are certainly some overlaps, especially between the middle and low diagnostic intensity categories where, for example, we classified NHL as requiring medium and prostate cancer as requiring low diagnostic intensity (29, 31). There are clear exceptions, where both of these malignancies might require very invasive procedures for a correct diagnosis such as retroperitoneal or intracranial biopsy for NHL and transrectal biopsy for prostate cancer (29, 31). Still, both tumor types should on average be more accessible for biopsy than brain or lung tumors, an assumption that is corroborated by the striking decline in the fraction of the latter two tumors that are histologically confirmed in advanced age.

Third, results from analyses where we investigated the occurrence of unexpected autopsy findings must be interpreted with caution for multiple reasons. The likelihood to perform an autopsy is presumably related to the probability of finding a yet undetected malignancy, meaning that the underlying assumption of a similar frequency of undetected cancer in persons who did not undergo autopsy is probably not valid. Also, the proportion of deaths investigated by autopsy decreases sharply by age (Supplementary Fig. S1; ref. 32). Even so, unless the relation between unexpected tumor findings in autopsied and non-autopsied deaths changes dramatically with age, this bias seems unlikely to fully explain the pattern of monotonously increasing cancer incidence observed in Figs. 2 and 3. The estimates presented in Fig. 2 are most likely an underestimation of the true proportion of undiagnosed cancers in autopsy material and focus should be on trends rather than numeric values (17, 18, 20). We did not have access to data on cause of death and indication for autopsy and it is delicate to speculate on how accurately the operating pathologist would diagnose and report a malignant lesion if the cause of death is obviously not related to cancer.

Among the strengths of the study is the population-based design, with detailed data on an entire population over a 45-year period from prospectively maintained cancer and population registers. This ensures both excellent statistical power with the exception of autopsy detected melanomas, and a high degree of representativeness. Still, the data are observational in nature and were not collected for the purpose of studies such as this one. Hence, it is hard to disentangle effects of decreasing diagnostic intensity with increasing age from possible, inherent biological effects.

Our results corroborate previous reports of a cancer incidence decline in the elderly (1–9). Comparing the age-incidence pattern with other Nordic countries, it is noticeable that the decline in elderly is less pronounced and/or occurs later in life in Denmark and Finland where unrecorded cancers from death certificates are backtracked and included in the cancer registers (7, 13). A reduced proportion of histologically confirmed cancers in elderly have been described previously (3, 7). No one has however, to our knowledge, investigated this hypothesis, categorizing tumors according to diagnostic invasiveness. Scrutinizing published reports, the cancer peak preceding the decline in the very old, seems to be less pronounced in low-invasiveness tumors such as skin cancer compared with high-invasiveness tumors like lung and pancreatic cancer (3, 7). Multiple studies on cancer and aging present cancer incidence together with cancer-specific mortality over age and report the mortality peak to occur latter, plateau, or increase continuously over age (3, 7, 8, 13). We caution against using cancer-specific mortality as a proxy for incidence. First, this measure does not only reflect cancer incidence with delay, but also survival, supposedly decreased in elderly due to a reduced capacity to tolerate cancer-specific therapy. Second, cancer-specific mortality relies on correct registrations of cause of death which is questionable in population-based studies using administrative data. Committed autopsy studies support the notion of a higher rate of undiagnosed cancers in elderly in autopsy material (17, 18). The prevalence of metastases in autopsy material has however been reported to be lower among elderly (2, 17–19). This raises the question whether the increasing “total” cancer incidence rate, including projected cases from autopsy data, over age is driven by incidentalomas rather than clinically significant cancers (2, 17)? Studies reporting a more advanced cancer stage distribution in elderly rather indicates the opposite, strengthening the notion of a leveled out diagnostic intensity in elderly (9, 14).

Overall, the combination of data on how the fraction of cancers that are pathology-confirmed decreases with age, how cancer incidence peaks at an earlier age for tumors that require more invasive diagnostic procedures, and how the incidence of unexpected autopsy findings does not seem to peak, all point in the same direction. Therefore, our interpretation is that the declining cancer incidence in advanced age is mostly driven by gradually declining diagnostic intensity.

To summarize, with the exception of melanoma skin cancer, we observe a consistent cancer incidence decline in high age, which in turn seems most likely to be driven by a matching decrease in diagnostic intensity. Our conclusion is that in most instances the decrease in cancer incidence in the elderly is not the effect of an actual biological decline in cancer occurrence. Cancer alarm symptoms in healthy, aged individuals need to be investigated promptly according to current guidelines and with increasing numbers of elderly patients with cancer, resources must be allocated to health care planning and investments to prepare for the anticipated cancer pandemic. Future clinical cancer trials should focus to include elderly to set updated standards for cancer therapy in the predominant age group.

No disclsoures were reported.

C. Radkiewicz: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. J.J. Krönmark: Conceptualization, methodology, writing–original draft, writing–review and editing. H.-O. Adami: Conceptualization, validation, investigation, methodology, writing–original draft, writing–review and editing. G. Edgren: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing.

We did not obtain any specific funding to conduct this study. G. Edgren is supported by grants from the Swedish Research Council (2017-01954) and Region Stockholm (clinical research appointment). The funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the report.

Georg Klein for inspiration and support in formulating and drafting of the original hypothesis and student report on this enigma.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.