The dramatic advances that have been made in the treatment of childhood cancer over the past several decades have created a large and growing population of long-term survivors.1 Although many childhood cancers now can be cured, the aggressive therapies used are associated with increased risks of a variety of adverse health effects, including new primary cancers.2, 3, 4, 5, 6, 7 Opportunities for addressing second cancer risks in large populations of childhood cancer survivors followed over many years are uncommon. Rarely can individual medical centers achieve adequate sample sizes or follow-up, necessitating costly multicenter studies or the use of large, population-based cancer registries. Population-based studies have advantages over most clinic-based studies in that study populations are unaffected by referral patterns, and completeness of follow-up is less likely to be related to the occurrence of a new cancer. We used data from the National Cancer Institute's Surveillance, Epidemiology and End-Results (SEER) population-based cancer registry in the U.S. to assess the risk of new primary cancers among persons diagnosed with childhood cancer between 1973 and 2002.
The objectives of our study were to quantify risks for developing new malignancies among childhood cancer survivors, identify links between particular types of first and subsequent cancer, and evaluate the possible role of treatment. A cohort of 25,965 2-month survivors of childhood cancer diagnosed in the U.S. during 1973–2002 was identified and followed through SEER cancer registries. Observed-to-expected ratios (O/E) were calculated, and Poisson regression was used to compare risks among treatment groups. Childhood cancer survivors were at nearly 6-fold risk of developing a new cancer relative to the general population (O/E = 5.9, 95% CI: 5.4–6.5). Most common were subsequent primary cancers of the female breast, central nervous system, bone, thyroid gland and soft tissue, as well as cutaneous melanoma and acute non-lymphocytic leukemia (ANLL). The greatest risks of subsequent cancers occurred among patients diagnosed previously with Hodgkin lymphoma (HL), Ewing sarcoma, primitive neuroectodermal tumor, or retinoblastoma. Risk of subsequent solid cancers was higher among persons whose initial treatment for childhood cancer included radiotherapy, whereas the excess of subsequent ANLL was strongly related to chemotherapy. The O/E for subsequent ANLL increased with increasing calendar year of initial cancer diagnosis among survivors of cancers other than HL, most likely due to increasing use of leukemogenic drugs for solid cancers and non-Hodgkin lymphoma. Childhood cancer survivors are at markedly increased risk of developing a variety of new cancers relative to the general population, but the magnitude of excess risk and specific types of second cancer vary widely by type of first cancer. © 2007 Wiley-Liss, Inc.
Material and methods
The study included persons diagnosed with cancer before the age of 18 years and between calendar years 1973 and 2002 that were reported to 1 of 9 SEER statewide or regional population-based cancer registries.8, 9 The study population was followed through the SEER program, which relies on multiple methods for updating vital status and identifying new primary cancers.10 The SEER program collects data about host and tumor characteristics (age at diagnosis, sex, race, tumor histopathology and location, survival, etc.) and initial cancer treatment (yes/no for surgery, radiotherapy, chemotherapy, hormonal therapy or combinations of these treatments).
Data analysis was restricted to persons who survived a minimum of 2 months after the date of initial cancer diagnosis. Person–years (PYR) of observation were cumulated from that date until the earliest of date of death, date last known alive or end of study (December 31, 2002). The O/E ratio was calculated as the ratio of the observed number of subsequent cancers (O) to the expected number (E). The expected number was obtained by applying the age-, sex-, race- and calendar year-specific SEER incidence rates to the population of survivors. Two-sided 95% confidence intervals (CI) for O/E were calculated assuming the observed number of cancers followed a Poisson distribution. The excess absolute risk (EAR) was calculated as the observed minus expected number of cancers per 10,000 PYR. Results were calculated for all types of childhood cancer combined and separately by type of first cancer, based on International Classification of Childhood Cancer (ICCC) codes.11 Cumulative incidence of second cancer, adjusted for competing risks, was estimated according to the method of Gooley et al.12 Risks for various subgroups of children were compared directly for all invasive solid tumors and acute non-lymphocytic leukemia (ANLL) via multivariate analyses, using Poisson regression methods for grouped survival data [SAS (version 8) routine GENMOD; SAS Institute, Cary, NC], incorporating external reference rates from the SEER registries.13 Data were adjusted for type of first primary cancer, time since initial cancer diagnosis and age at initial diagnosis. Previous research has indicated that treatment-related ANLL is most strongly linked to chemotherapy and increased risk tends to appear 1–5 years after initial therapy. In contrast, the risk of second solid cancers is more typically related to radiotherapy with a longer latent period of 5–10 years. Thus, our analyses of calendar year effects of therapy assumed a minimum latency period of 1 year for ANLL and 5 years for solid cancers.
A total of 25,965 children who survived 2 or more months were observed for an average of 8.9 years (median, 6.3 years; range, 2 months–30.0 years). Follow-up to death or close of study was 89.1% complete. There were 14,626, 5-year survivors, 9,873, 10-year survivors and 3,567, 20-year survivors. The maximum age at the end of follow-up was 47 years; 2,834 individuals (10.9%) were followed past the age of 30 and 696 (2.7%) were followed into their 40s. The proportion receiving initial radiotherapy decreased steadily from 56% in 1973–1979 to 28% in 1995–2002. The trend for initial chemotherapy was in the opposite direction, increasing from 60 to 74% (data not shown). The relative decrease over time in the use of radiotherapy was greatest for non-Hodgkin lymphoma (NHL) and leukemia, followed by and Hodgkin lymphoma (HL) and solid cancers. The relative increase in use of chemotherapy was greatest for HL and solid cancers.
During the period of follow-up, 433 new primary cancers were diagnosed in 400 individuals (Table I). This represents a nearly 6-fold increased incidence relative to the general population (O/E = 5.9, 95% CI = 5.4–6.5, EAR = 15.6 per 10,000 PYR) (Table I). Twenty-four childhood cancer patients developed 2 new malignancies, and 4 developed 3 or more new malignancies. Risks were similar for females (O/E = 5.9, O = 226, EAR = 16.9) and males (O/E = 6.0, O = 207, EAR = 14.4), but were somewhat higher for blacks (O/E = 8.9,O= 45, EAR = 18.4) than whites (O/E = 5.5,O= 351, EAR = 15.0). The largest relative and EAR s for all types of subsequent cancer combined were seen among patients with initial diagnoses of primitive neuroectodermal tumor (PNET), retinoblastoma, Ewing sarcoma and HL.
|Histopathologic type1||Number2||Median||Initial therapy (%)||Totals|
|Total||Female||Male||Age at DX||PYR||Surgery||RT||CT||O||E||O/E||EAR|
|All first cancers||25965||11922||14043||8.2||6.3||49.9||37.1||65.4||433||73.26||5.9*||15.6|
|All CNS cancer||4806||2146||2660||7.6||5.0||75.2||55.5||29.4||69||10.90||6.3*||15.3|
|PNET, brain and CNS||945||353||592||6.4||3.7||93.0||81.8||62.4||22||1.51||14.6*||32.1|
|All bone cancer||1466||629||837||13.7||4.3||68.9||26.9||83.2||31||4.28||7.2*||23.8|
|All soft tissue sarcoma||1909||868||1041||10.3||6.5||76.4||45.6||60.8||33||5.99||5.5*||15.4|
|Germ cell tumors5||1296||624||672||14.5||8.0||85.7||22.5||54.9||21||5.38||3.9*||11.9|
For all new cancers combined, the O/E increased from 6-fold during the first year after initial cancer diagnosis to nearly 8-fold for the next 9 years, and then decreased to 4-fold among 20+ year survivors (Table II). In contrast, the EAR increased with time, from 8.3 excess cancers per 10,000 PYR in the first year of follow-up to 31.4 among persons surviving more than 20 years (data not shown). The cumulative incidence of developing a second cancer at 25 years was 3.6% (95% CI = 3.2–4.1%) for all first primary cancers combined and 11.9% for HL (95% CI = 9.5–14.6%) (Fig. 1).
|Subsequent cancer risks years after first primary cancer diagnosis|
|0.16–<1 yr||1–4 years||5–9 years||10–14 years||15–19 years||20+ years||Total|
|All subsequent cancers||20||6.1*||92||7.7*||103||7.6*||81||5.5*||74||5.1*||63||4.1*||433||73.26||5.9*||15.6|
|All solid cancers||17||8.8*||55||7.6*||83||9.2*||72||6.6*||71||6.2*||55||4.2*||353||53.77||6.6*||12.9|
|Buccal cavity, pharynx||0||0||2||10.7*||4||15.7*||6||20.9*||4||14.5*||2||5.9||18||1.38||13.0*||0.7|
|Nose, nasal cavity, ear||0||0||0||0||2||52.9*||0||0||0||0||0||0||2||0.17||11.6*||0.1|
|Lung and bronchus||0||0||1||20.6||1||12.3||2||15.7*||1||4.7||2||4.3||7||0.94||7.4*||0.3|
|Melanoma of the skin||5||54.4*||4||6.6*||7||5.4*||6||3.4*||5||2.7||3||1.6||301||7.55||4.0*||1.0|
|Eye and orbit||2||21.0*||1||5.0||1||19.7||1||25.6||0||0||1||27.4||6||0.46||13.1*||0.2|
|Brain and other CNS||4||6.8*||9||4.6*||17||10.7*||10||9.2*||5||6.9*||6||11.0*||51||6.49||7.9*||1.9|
|Bone and joints||1||6.2||12||18.6*||12||19.0*||8||18.2*||4||18.2*||4||39.2*||41||2.20||18.6*||1.7|
|Person–years at risk||20,096||70,503||60,450||40,346||24,582||15,166||231,142|
The most common types of subsequent primary cancer were cancers of the female breast, central nervous system (CNS), bone, thyroid gland and soft tissue, as well as cutaneous melanoma and ANLL (Table II). Significantly high O/Es also were noted for new malignancies of the buccal cavity (especially salivary gland), most digestive and respiratory tract sites, uterine corpus, prostate, testis, kidney and eye, as well as acute lymphocytic leukemia (ALL), chronic myeloid leukemia (CML), NHL and mesothelioma. Elevated O/E ratios for some new primary cancers (including ANLL, melanoma and cancer of the brain, bone, soft tissue and testis) were apparent within the first 5 years after diagnosis of the first cancer. For other cancers, including those of the buccal cavity, stomach, female breast, urinary tract and soft tissues, the relative risk appeared greater for longer-term survivors. In particular, there was a late-emerging 20 to 30-fold excess of stomach cancer. For subsequent ANLL, both the O/E and EAR peaked between 1 and 5 years after the first cancer.
In multivariate analyses using Poisson regression modeling, childhood cancer patients whose initial treatment included radiotherapy or chemotherapy or both were at higher risk of developing a new solid cancer than those not so treated, whereas ANLL was strongly associated with chemotherapy (Table III). Subsequent cancer sites showing the greatest difference in O/E by radiation among 5-year survivors included cancers of the female breast, brain and other CNS, bone and soft tissue, thyroid gland, stomach, pancreas and lung and bronchus (Table IV).
|Initial treatment||O||O/E||RR1||95% CI1|
|Subsequent solid cancers (≥5 year survivors)|
|No RT, no CT||34||2.6||1.0||Reference|
|RT, no CT||82||8.0||2.8||1.8–4.3|
|CT, no RT||64||6.5||2.1||1.4–3.3|
|RT and CT||93||9.7||3.2||2.1–4.9|
|Subsequent ANLL (≥1 year survivors)|
|No RT, no CT||1||2.1||1.0||Reference|
|RT, no CT||2||5.9||2.5||0.2–27.7|
|CT, no RT||20||33.5||13.9||1.8–110|
|RT and CT||11||24.0||10.5||1.3–84.3|
|Type of cancer||Any radiation||No radiation|
|All subsequent cancers||198||7.7*||28.7||113||3.8*||11.0|
|All solid cancers||175||8.8*||25.8||98||4.3*||9.9|
|Lung and bronchus||3||17.0*||0.9||1||5.4||0.2|
|Melanoma of the skin||8||2.6*||0.8||13||3.8*||1.3|
|Brain and other CNS||27||16.1*||4.2||10||4.7*||1.0|
|Bones and joints||20||33.2*||3.2||8||10.7*||1.0|
Risks of new primary cancers by type of initial cancer are shown in Table V. HL was the initial cancer associated with 111 (25.6%) of the 433 new primary cancers. There was an increasing trend in risk over time since HL diagnosis, with the EAR rising to over 90 excess cancers per 10,000 PYR among persons surviving 15 years or more (p < 0.0001; data not shown). Solid cancer risks were higher among HL patients who received radiotherapy for their initial cancer (O/E = 10.5) than for those who did not (O/E = 7.4). Relative risks were higher among females (O/E = 11.8) than males (O/E = 7.1), mainly due to a notably high risk of breast cancer. In addition, significantly elevated risks of more than 9-fold were seen for subsequent cancers of the salivary gland, stomach, pancreas, lung, uterine corpus, bone, soft tissue and thyroid gland, as well as for ANLL and NHL. Risk of ANLL remained elevated through the first 15 years of follow-up (data not shown).
|Subsequent cancer site or type|
|Buccal cavity and pharynx||Digestive system||Lung and bronchus||Female breast||Female genital system||Melanoma of skin||Brain and CNS||Thyroid gland||Urinary system||Bones and joints||Soft tissue||NHL||ANLL||All leukemia||All cancers|
|Other soft tissue||2||49.7*||1||9.4||2||62.8*||1||4.7||0||0.0||2||9.2*||0||0.0||0||0.0||0||0.0||2||39.5*||0||0.0||0||0.0||1||20.6||1||7.4||11||5.6*|
|Germ cell cancer||1||9.3||2||6.8||0||0.0||3||5.2*||2||4.3||2||3.4||1||2.8||1||2.0||0||0.0||0||0.0||0||0.0||0||0.0||0||0.0||0||0.0||214||3.9*|
Risks of new primary cancers following types of first cancer other than HL also are shown in Table V. Of note, statistically significant increases were seen for: (i) ANLL following ALL, NHL, Ewing sarcoma, osteosarcoma, rhabdomyosarcoma and Wilms tumor; (ii) breast cancer following NHL, Ewing sarcoma, osteosarcoma, rhabdomyosarcoma and germ cell tumors; (iii) CNS cancer following ALL or CNS cancer; (iv) bone or soft tissue sarcoma following ALL, CNS cancer, bone or soft tissue sarcoma, retinoblastoma or Wilms tumor; (v) thyroid cancer following ALL, CNS cancer or neuroblastoma; and (vi) buccal cavity and pharynx following ALL and soft tissue sarcomas. CML was increased among survivors of childhood astrocytoma, and ALL was elevated significantly among patients with an initial diagnosis of retinoblastoma or PNET. Among retinoblastoma patients, second cancer risks were greater among those who received radiation as part of their initial treatment (O = 12, O/E = 44.1) than among those who did not receive radiotherapy (O = 3, O/E = 4.2). The risk of subsequent cancer among males with initial germ cell cancers was due to 7 cases of contralateral testicular cancer.
Analysis of subsequent cancer risk by calendar year of initial cancer diagnosis showed an increasing relative risk of ANLL with increasing calendar year for the latency interval of 1–<5 years (p for trend = 0.005) and for all follow-up intervals combined (p for trend < 0.0001), Table VI). The O/E approached 100-fold for persons with an initial diagnosis in 1995–2002. Seven of the 13 cases in this category developed in persons with an initial bone or soft tissue sarcoma (O/E = 388; 95% CI: 156–799), and 4 had an initial diagnosis of NHL (O/E = 587; 95% CI: 160–1503); none occurred among former HL patients. The initial course of therapy for 12 of the 13 ANLL cases in the 1995–2002 period included chemotherapy. The O/E for new solid cancers followed a near constant O/E pattern with increasing calendar year of diagnosis for the 5–9 and 10–14 year follow-up intervals. However, among 15+ year survivors a significant trend in risk was detected over the calendar year periods 1973–1994 for both the relative risk (p for trend in O/E < 0.0001) and excess risk (p for trend in EAR = 0.012). Follow-up beyond 15 years was not possible for children diagnosed in the 1990–1994 and 1995–2002 calendar time periods.
|Subsequent primary cancer||First primary cancer||Calendar year||Years after first primary cancer diagnosis|
|5–<10 years||10–< 15 years||≥15 years||Total|
|Other first primary sites||1973–1979||17||9.4*||14||4.4*||46||3.1*||77||20.02||3.8*||11.7|
|1–<5 years||5–<10 years||≥10 years||Total|
|Other first primary sites||1973–1979||1||10.4||0||0.0||0||0.0||1||0.58||1.7||0.1|
Because of the population-based design and large size of SEER, the results presented here can be presumed to be broadly representative of the experience of survivors of childhood cancer in the U.S. as a whole during this era. There was no selection of the study population based on patients having been referred to a major cancer center or having survived for several years, and the study population included patients with all types of childhood cancer. Follow-up coverage allowed for estimation of short-term and long-term risks. To our knowledge, this is the largest published study to date in terms of the numbers of long-term (5+ year) survivors and total number of new primary cancers. The relatively large numbers with long-term follow-up enabled us to assess second cancer risks among survivors as they age into their 30s and 40s and to examine associations between particular types of first and second cancer within the same population. Importantly, the SEER data base includes recent calendar years of diagnosis, which allowed us to evaluate risk associated with newer treatments.
Our finding of an ∼6-fold increased risk of developing a new primary cancer in childhood cancer survivors relative to the general population is similar to risks reported for 3 other large cohorts from North America,3 Great Britain7 and Nordic countries.14 The Childhood Cancer Survivor Study includes over 13,000 children from 25 medical institutions in the U.S. and Canada who were diagnosed with cancers between 1970 and 1986 at ages less than 21 years and surviving at least 5 years.3 The British cohort7 included over 16,000, 3-year survivors diagnosed under age 20 years during 1926–1987, and the Nordic childhood cancer cohort included over 30,000 children ages less than 15 years diagnosed during 1943–1987. Therefore, to our knowledge, our investigation is the only large cohort study to report on subsequent cancer risk occurring among childhood survivors receiving more recent treatment regimens during the 1988–2002 calendar year period. Our results confirm the findings of these and other childhood cancer studies of high risks of second primary cancers among survivors of childhood and adolescent HL,5, 15-21 including an early excess of ANLL and a later emerging excess of a variety of solid cancers, most notably breast cancer. However, the present study has important new findings as well, particularly for initial cancers other than HD and new results by calendar year periods, which we focus on below.
We observed significantly elevated risks of acute leukemia following many types of first childhood cancers, including ALL, HL, NHL, CNS cancer, Wilms tumor, bone and soft tissue sarcoma and retinoblastoma, with a strong relationship to initial chemotherapy. For HL, this is probably due to MOPP treatments,15, 22 as more than half of the cases of ANLL after HL occurred among persons with initial cancers diagnosed in the 1970s. Although the use of chemotherapy in the treatment of HL increased considerably over the time period covered by the study, the incidence of subsequent ANLL did not, reflecting the switch to less leukemogenic chemotherapy regimens to treat HL. Of some concern is the apparent trend towards increased risk of ANLL after an initial bone or soft tissue sarcoma or NHL diagnosed during more recent years, which may reflect the increased use of epipodophyllotoxins and alkylating agents23, 24, 25, 26 to treat these cancers in more recent years.
The increased risks seen in the current data set for adult-onset tumors, such as breast, stomach, pancreas, esophagus, oral, lung, kidney and bladder cancers, may portend increasing absolute risks in the future. Whereas the overall relative risk for solid cancers decreased with increasing follow-up time, the EAR of new solid cancers increased with time. This finding is consistent with previous studies7, 14, 27, 28, 29 but is presented here with large numbers in a population-based cohort.
Chemotherapy likely was the most important cause of subsequent ANLL, but radiation treatment appears to have been more important in the excess incidence of solid cancers. Previous studies have documented excesses of cancers of the breast, thyroid gland, CNS, salivary gland, bone and stomach associated with radiotherapy during childhood.2, 16, 17, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 The frequent use of extended field radiotherapy and combined modality treatments almost certainly contributed to the large EARs that were seen following Ewing sarcoma, HL and PNET. Patients diagnosed with HL or Ewing sarcoma also tended to be older and more likely to have attained ages at which background cancer incidence is greater. The high relative risk of a second cancer in patients with retinoblastoma reflects the combination of genetic predisposition to multiple cancers, sensitivity to radiogenic cancer and very low background cancer risk in the general population during the early years of life.39, 49, 50, 51, 52, 53 The increased risk of ALL after retinoblastoma has not been reported previously in an analytic study. Although excess solid cancers as a group appear to be more strongly associated with radiotherapy than chemotherapy, previous studies of childhood cancer survivors have linked alkylating agents with second primary bone and soft tissue sarcoma.30, 35, 37, 42, 54
Although HL survivors accounted for, by far, the largest number of breast cancer cases in the present study, substantially increased relative risks of breast cancer were seen among survivors of other childhood cancers as well, including NHL, Ewing sarcoma, osteosarcoma, rhabdomyosarcoma and germ cell tumors. With increases in the numbers of survivors of these cancers reaching ages at which the background incidence rate of breast cancer increases sharply, second primary breast cancer may become an important late effect of treatment for these groups as well.
An important advantage of the SEER registry database is the ability to study a very large number of patients in a population-based setting; however, several limitations of our study also should be noted. Our estimates of second cancer risk are likely to be conservative, because of the under-ascertainment of new cancers among long-term survivors who migrated outside of the SEER catchment area where the first cancer was diagnosed. However, our overall estimate of a nearly 5-fold risk among 5-year survivors is similar to the 6-fold risk cited in the Childhood Cancer Survivor Study,3 which suggests that these effects are not likely to be large. It is reassuring that these investigations with different study designs yielded similar overall results. We also are limited by the lack of complete, detailed treatment information. As with all such studies, the risks for long-term survivors focus on treatments used in the past, and treatments for childhood cancer have changed considerably over the last 30 years.55, 56, 57, 58, 59
Childhood cancer survival has improved considerably since the early 1970s, and effective treatment for the initial cancer remains of paramount concern. Nonetheless, comparatively little information exists about treatment-related second cancer risks among long-term survivors of childhood cancer, and continued follow-up of existing large study cohorts is critical to determine how the enormous relative risks seen through adolescence and early adulthood carry through into older ages, when incidence rates of carcinomas increase dramatically.60, 61, 62 It will be of interest to determine whether the decreased use of radiation in the treatment of childhood cancers is accompanied by a reduced incidence of radiation-related second cancers.
We thank Mr. Nathan Appel, Mr. Eric Berger and Ms. Shannon Merkle of Information Management Services, Inc. for programming support, Mr. Ankur Saini for assistance with graphics and Ms. Denise Duong for clerical support. We also acknowledge the work of the SEER program in collecting the data that made our study possible.