Early Deaths in Childhood Cancer

Merlin R. Hamre, MD, MPH,1* James Williams, MD,1 Paul Chuba, PhD, MD,2

Kanta Bhambhani, MB, BS,1 Y. Ravindranath, MB, BS,1 andRichard K. Severson, PhD3

Background. Deaths prior to or shortly afterthe diagnosis of childhood cancer may reflectinadequacies in detection and appropriate re-ferral for care. This study was performed to de-termine the extent of and factors associatedwith early death in childhood cancer. Proce-dure. Patients with of primary cancer, aged <20years at diagnosis, were identified from theSEER data (n = 23,470) from 1973 to 1995.Early deaths were defined as cases identified by1) death certificate, 2) autopsy report, or 3)death within 1 month of initial diagnoses (n =481). Cause of death was determined by ICD-8and -9 codes. Age at diagnosis, year of diagno-sis, morphology, site of disease, race, and gen-der were evaluated for association with earlydeath. Results. Age <1 year at diagnosis (6.2%early deaths), being diagnosed earlier in the ob-servation period, and a diagnosis of a brain tu-mor, neuroblastoma, leukemia, or liver tumorwere associated with increased early death.

Gender and race were not associated with earlydeath. Among the cases for whom the malig-nant diagnosis was made at the time of death (n= 119), the cause of death was nonmalignantfor 36. For 22 of these cases the malignancywas an incidental finding and appeared not tocontribute directly to the cause of death.Among these patients, 11 had neuroblastoma, 9being <1 year of age. Conclusions. A decreasein the proportion of early deaths associatedwith childhood cancer has occurred during thepast 2 decades. This decrease may reflect ear-lier diagnosis or improved imaging capabilities,surgical techniques, medical therapy, and sup-portive care. Awareness among pediatricians,general practitioners, and emergency physi-cians is warranted, with a focus on high-riskgroups for early detection among childhoodcancer patients. Med. Pediatr. Oncol. 34:343–347, 2000. © 2000 Wiley-Liss, Inc.

Key words: mortality; neoplasms; leukemia; pediatrics; SEER Program

INTRODUCTION

Significant gains have been achieved over the pastseveral decades in the treatment of childhood cancer.With timely diagnosis and appropriate therapy, the ma-jority of children with cancer can expect to become long-term survivors. Current 5 year life expectancies for pa-tients with Wilms tumor, Hodgkin disease, acutelymphoblastic leukemia, non-Hodgkin lymphoma, andbrain tumors are 92%, 92%, 77%, 71%, and 64%, re-spectively (SEER data [1] for patients diagnosed 1988–1992). Nonetheless some children are not diagnosed and/or referred in a timely manner, resulting in death prior tothe initiation of therapy or at a time when such therapy isineffective.

Although delays in referral have been examined forchildhood cancer [2–4], early deaths from childhood can-cer have been limited to case reports or series [5]. Al-though these data are interesting, such deaths may not begenerally reported as part of clinical series insofar aspatients may have died prior to the time when they wouldhave been consider for enrollment. In the present studywe examined early deaths from childhood cancer usingdata from the National Cancer Institute’s SurveillanceEpidemiology and End Results (SEER) Program. Thisallowed us to evaluate a population-based sample of the

United States of patients diagnosed with cancer. We in-vestigated cause of death and factors potentially associ-ated with early death in patients with childhood cancer.

MATERIALS AND METHODS

The SEER Program is based on a series of population-based registries, which collect information on all newoccurrences of malignant disease for defined geographicareas of the United States, which constitute approxi-mately 14% of the country’s population. For thecurrent study, cases of primary neoplasms (second andsubsequent malignancies excluded) occurring at <20years of age were identified in the SEER data [1]. Casesof carcinoma in situ were excluded. “Early death” was

1Department of Pediatrics, Barbara Karmanos Cancer Institute, WayneState University, Detroit, Michigan2Department of Radiation Oncology, Barbara Karmanos Cancer Insti-tute, Wayne State University, Detroit, Michigan3Department of Family Medicine, Barbara Karmanos Cancer Institute,Wayne State University, Detroit, Michigan

*Correspondence to: Merlin R. Hamre, MD, MPH, Division of He-matology/Oncology, Children’s Hospital of Michigan, 3901 BeaubienBlvd., Detroit, MI 48201-2196. E-mail: mhamre@med.wayne.edu

Received 8 September 1999; Accepted 12 January 2000

Medical and Pediatric Oncology 34:343–347 (2000)

© 2000 Wiley-Liss, Inc.

defined as cases in which the diagnosis of malignancywas made at the time of death (autopsy reports or deathcertificates were the primary source of information forthe registries) or patients who died within 1 month frominitial diagnosis. Based on these criteria a total of 23,470patients with childhood cancer were identified, of whom481 experienced an early death.

Cause of death was determined by the recorded ICD-8and -9 codes for death and classified as resulting fromprimary cancer or other diagnoses. Cases for whom thecause of death could not be ascertained from death cer-tificates (SEER ICD codes 7777 or 7797) were classifiedas unknown. Age at diagnosis was grouped as <1, 1–4,5–9, 10–14, or 15–19 years of age. Race was classified asAfrican American, Caucasian, Hispanic, Asian/PacificIslander, Native American, and other. Morphology wasclassified on the basis of ICD-O codes as leukemia(9800–9941), lymphoma (including Hodgkin disease,9590–9723), brain tumors (9362, 9380–9481), neuro-blastoma (including ganglioneuroblastoma, 9490, 9500),hepatoblastoma/hepatocellular carcinoma (8170–8180,8970), Wilms tumor (8960), germ cell tumors (9060–9091), sarcoma (8800–8951, 8963, 8964, 8971–9044,9120–9340), carcinoma (8971–9044, 9120–9340), un-known (8000–8004), and other. Primary site of diseasewas also classified per the ICD-O codes as liver (220),skeleton (400–419), bone marrow (421), kidney (649,659), central nervous system (700–721, 753), adrenalgland (740–749), lymph nodes (770–779), unknown(809), and other.

Logistic regression was used to assess the potentialrelation between independent variables (such as age,race, and histology) and early death (yes/no). Each nomi-nal level variable, such as histology and race, was cat-egorized as a set of dummy variables, and its statisticalsignificance was assessed by a likelihood ratio test be-tween a model with none of the dummy variables in-cluded and a model with all of the dummy variablesincluded. Continuous-value variables, such as age at di-agnosis and calendar year diagnosed, were categorized asa set of dummy variables and then included as a set intologistic regression models. Interaction terms betweenvariables were included and evaluated for statistical sig-nificance prior to interpreting models with main-effectsterms only. The fit for each model was assessed by thelog likelihood method. Differences in proportions ofearly deaths across grouped years of diagnosis (for pa-tients with leukemias, CNS rumors, and liver tumors)were assessed byx2 tests for trend. AP value of <0.05was considered to be statistically significant.

RESULTS

In Table I, characteristics of patients with early deathsare compared to those of all other cancer patients sur-

viving beyond 1 month from diagnosis. Age, years ofdiagnosis, histology, and primary site all proved to besignificant on univariate as well as multivariate analysis(P < 0.0001). Race and sex were not associated withearly death. However, when the multivariate analysis

TABLE I. Comparison of Childhood Cancer PatientsExperiencing Early Death to Those Living Beyond the FirstMonth From Diagnosis*

Surviving$1 monthNo. (%)

Earlydeath

No. (%)P

multivariate

n 22,989 481Age at diagnosis (years) <0.0001

<1 1,617 (7.0) 107 (22.2)1–4 5,421 (23.6) 116 (24.1)5–9 3,861 (16.8) 86 (17.9)10–14 4,410 (19.2) 78 (16.2)15–19 7,680 (33.4) 94 (19.5)

Years diagnosed <0.00011973–1977 4,524 (19.7) 143 (29.7)1978–1982 4,832 (21.0) 104 (21.6)1983–1987 5,020 (21.8) 106 (22.0)1988–1992 5,333 (23.2) 79 (16.4)1993–1995 3,280 (14.3) 49 (10.2)

Histology <0.0001Leukemia 5,564 (24.2) 143 (29.7)Lymphoma 3,737 (16.3) 56 (11.6)Brain tumors 3,747 (16.3) 127 (26.4)Neuroblastoma 1,224 (5.3) 40 (8.3)Hepatoblastoma/

hepatocellularcarcinoma 206 (0.9) 23 (4.8)

Wilms tumor 930 (4.0) 12 (2.5)Germ cell tumors 1,261 (5.5) 21 (4.4)Sarcomas 2,819 (12.3) 17 (3.5)Carcinomas 1,745 (7.6) 16 (3.3)Unknown 181 (0.8) 16 (3.3)Other 1,575 (6.9) 10 (2.1)

Primary site <0.0001Liver 246 (1.1) 26 (5.4)Skeleton 1,351 (5.9) 4 (0.8)Bone marrow 5,583 (24.3) 143 (29.7)Kidney 1,057 (4.6) 16 (3.3)Central nervous

system 3,954 (17.2) 151 (31.4)Adrenal gland 445 (1.9) 24 (5.0)Lymph nodes 3,247 (14.1) 43 (8.9)Unknown 132 (0.6) 13 (2.7)Other 6,974 (30.3) 61 (12.7)

SexMale 12,307 (53.5) 247 (51.4) n.s.Female 10,682 (46.5) 234 (48.6)

RaceWhite non-Hispanic 17,351 (75.5) 342 (71.1) n.s.African American 2,272 (9.9) 60 (12.5)American Indian 193 (0.8) 4 (0.8)Asian/Pacific Islander 915 (4.0) 24 (5.0)Other 502 (2.2) 10 (2.1)Unknown 243 (1.1) 2 (0.4)White Hispanic 1,513 (6.6) 39 (8.1)

*P < 0.0001 for whole model compared to model omitting regressoreffects except for the intercept parameters.

344 Hamre et al.

was performed comparing only African Americans andCaucasians, African Americans had a significantly (P 40.02) greater risk of early death. The same analysis com-paring the other racial groups to Caucasians showed racenot to be associated with early death. Overall early deathwas associated with a younger age at diagnosis; diagno-sis during the beginning of the surveillance period; hav-ing a diagnosis of brain tumor, leukemia, hepatoblas-toma/hepatocellular carcinoma, and neuroblastoma; andprimary sites of liver, bone marrow, CNS, or adrenalglands.

Evaluation of age at diagnosis showed that earlydeaths accounted for 6.2% of cases diagnosed at <1 yearof age and decreased for older cases: 2.1%, 2.2%, 1.7%,and 1.2% for 1–4, 5–9, 10–14, and 15–19 years of age atdiagnosis, respectively. The proportion of early child-hood cancer deaths was examined by year of diagnosis.A clear decrease in the percentage of early deaths couldbe observed: 3.1%, 2.1%, 2.1%, 1.5%, and 1.5% for theyears 1973–1977, 1978–1982, 1983–1987, 1988–1992,and 1988–1995, respectively. Most of this decrease wasaccounted for by patients who died within 1 month ofdiagnosis rather than in cases who were identified byautopsy reports or death certificates only (data notshown).

ICD codes were examined for cases in which the di-agnosis was first reported to the SEER registries viadeath certificates or autopsy (n4 119). For 79 of thesecases the cause of death was attributed to the neoplasm;for the remaining 36 the cause of death was nonmalig-

nant, and for 6 cases the cause of death was unknown. Ofthe 36 deaths listed as nonmalignant, 3 were categorizedas conditions associated with malignancies (immune de-ficiencies), in 11 the cause of death could be directlyattributed to the malignancy, and in 22 the malignancywas an incidental finding at death. Eleven (50%) of themalignancies classified as incidental were neuroblastomaand 9 of these 11 deaths occurred at <1 year of age. Theremaining incidental findings of malignancies were 4brain tumors, 2 thyroid cancers, and 1 each of retinoblas-toma, Wilms tumor, Hodgkin disease, thymus carci-noma, and unspecified.

Considering the 6 cases in which the cause of deathwas unknown, the associated malignancies were neuro-blastoma (2), brain tumors (2), acute myeloid leukemia(1), and Hodgkin disease (1). Four of these patients were<1 year of age at diagnosis.

Proportions of early deaths were also examined indi-vidually for cases of CNS neoplasms, liver tumors, andleukemias (Fig. 1). These proportions showed two dif-ferent patterns with time. In leukemia patients, the per-centage of early deaths declined after the 1973–1977period and plateaued during subsequent years. CNS neo-plasms declined steadily in incidence and liver tumorsshowed a decline after the 1983–1987 period.

DISCUSSION

The SEER data as analyzed here allow for the enu-meration of early deaths, trends over time, and cause of

Fig. 1. Percentage of all occurrences of leukemia, CNS tumors, and liver tumors by years of diagnosis (P < 0.0001,P < 0.02,P < 0.01,respectively).

Early Deaths in Childhood Cancer 345

death. Limitations of SEER data are that they do notallow for a complete assessment of factors contributingto the mortality of these patients. Such factors includeperioperative complications, delay in diagnosis, sepsis,metabolic derangement, and toxicities of chemotherapy,among others. Consequently, we have been unable toevaluate factors, such as improvements in surgical prac-tice, supportive care, or timely referral, which may havehad an impact on early death in childhood cancer.

We found that infants <1 year of age were proportion-ally more represented among the cases of early deathfrom childhood cancer. This finding was not unexpected,for several reasons. The presenting signs and symptomsof cancer for this age group may be nonspecific [6].Consequently, infants may be at increased risk for deathprior to referral for appropriate therapy, or referral maytake place at a time when existing interventions are in-effective. Neoplastic diseases occurring in infants may bemore virulent, and infants are inherently less capable ofwithstanding aggressive and long-term therapies.

Leukemia, CNS tumors, neuroblastoma, and liver tu-mors accounted for the majority of early deaths. Specificaspects of these cancers and therapeutic developmentsthat may influence the early mortality associated withthese diagnoses are discussed below.

Leukemia diagnosis represented the largest cause ofearly death. Certainly this reflects the relatively largeincidence of leukemia in childhood. The greatest magni-tude of decrease in mortality in the time period from1973 to 1995 was in the number of deaths taking placewithin 1 month of diagnosis. This may be explained inpart by better supportive care, such as early empiric an-tibiotic therapy, antifungal therapy, nutritional and bloodproduct support, and aggressive induction regimens. Oneof the major causes of death in leukemia patient autopsieswas infection, in particular neutropenic typhlitis [7,8].Also, fungal infections remain one of the most seriousinfections in leukemia patients, especially for acute my-eloblastic leukemia (AML). The incidence of mycoticinfections contributing to leukemia mortality (not diag-nosed prior to autopsy) has decreased over time [9]. Nev-ertheless, fungemia remains a common cause of deathduring induction therapy [10]. Coagulopathy and hyper-leukocytosis are also risk factors for early death in AMLpatients [5,11,12]. It has been convincingly shown thatthe empiric use of broad-spectrum antibiotics for febrileneutropenic episodes and greater awareness of fungalinfections and antifungal therapy have contributed to im-proved leukemia patient survival [13,14].

The reported incidence of pediatric CNS malignancieshas increased 35% between 1973 and 1994 [15]. Smithand coworkers [16] have proposed a ‘jump model’ usinga step function for which a constant lower incidence inthe years before 1984–1985 preceded a constant higherrate afterward. A substantial amount of this increase was

thought to be due to an increase in the reporting of low-grade gliomas to the SEER registry, without a parallelincrease in mortality secondary to childhood brain tu-mors. Thus a plausible explanation for increased inci-dence was that the increased incidence resulted fromchanges in detection and/or reports of childhood brainneoplasms. Magnetic resonance imaging became ubiqui-tous at tertiary pediatric centers in the mid-1980s, andthis is likely to have increased the rate of detection. Withregard to the current data, a proportion of the decrease inthe percentage of early deaths from childhood cancercould be attributed to an increased reporting of malignantbrain tumors, which have a more favorable prognosis. Inaddition to the advancements in imaging, significant im-provements in neurosurgical techniques have occurredduring the past 2 decades, including stereotactic surgery[17], CUSA [18], and so on, as well as increased referralto pediatric neurosurgeons [19]. This may be associatedwith decreased operative morbidity and decreased earlydeaths observed in these patients.

Fifty percent of malignancies classified as incidentalfindings at death were accounted for by neuroblastoma.These incidental findings of neuroblastoma are not sur-prising in that neuroblastoma at <1 year of age has beendocumented to regress spontaneously [20,21]. As hasbeen shown in trials of screening for neuroblastoma inthis age group, aggressive efforts at identifying thesepatients are not warranted [22,23].

Interestingly, hepatoblastoma and hepatocellular car-cinoma were the most overrepresented causes of earlydeath, compared to its childhood incidence. Long-termsurvival for hepatoblastoma and hepatocellular carci-noma in childhood is closely related to a complete sur-gical resection [24]. Surgical attempts at complete resec-tion may be associated with great morbidity fromhemorrhage, and massive blood transfusions may be re-quired [25]. The past 20 years have provided advances,which may reduce perioperative death. It is now realizedthat preoperative chemotherapy can make total resectionof hepatoblastomas more feasible and reduce periopera-tive morbidity and mortality [26,27]. Specific surgicaldevelopments, including the use of intraoperative ultra-sound to assist in determining vascular relationships andextent of tumor [28], preoperative embolization in se-lected cases [29], and use of hypothermia with circula-tion arrest to remove vascular extensions [30], might alsohave contributed to the decreasing early mortality inthese patients.

CONCLUSIONS

Early deaths from childhood cancer have decreasedfrom 3% of cases in 1973–1977 to 1.5% of cases in1993–1995. Most of this improvement in early mortalityoccurred in patients dying within 1 month of diagnosis,with only minimal improvement in the number of au-

346 Hamre et al.

topsy or death certificate diagnoses. This most likely re-flects earlier diagnosis, enhanced imaging capabilities,and improved surgical techniques, medical therapy, andsupportive care. In comparison to cancer incidence rates,leukemia, CNS tumor, neuroblastoma, and hepatoblas-toma patients as well as the very young had the highestproportion of early deaths, whereas sarcoma and carci-noma patients had the lowest. A major limitation of thecurrent study is the lack of complete information regard-ing the cause of death. Still, awareness among pediatri-cians, general practitioners, and emergency physicians iswarranted, with a focus on high-risk groups for earlydeath among childhood cancer patients.

REFERENCES

1. Surveillance, Epidemiology, and End Results (SEER) Programpublic use CD-ROM (1973–1995), National Cancer Institute,DCPC, Surveillance Program, Cancer Statistics Branch, releasedApril 1998, based on August 1997 submission.

2. Edgeworth J, Bullock P, Bailey A, et al. Why are brain tumors stillbeing missed? Arch Dis Child 1996;74:148–151.

3. Pollock BH, Krischer JP, Vietti TJ. Interval between symptomonset and diagnosis of pediatric solid tumors. J Pediatr 1991;119:725–732.

4. Flores LE, Williams DL, Bell BA, et al. Delay in the diagnosis ofpediatric brain tumors. Am J Dis Child 1986;40:684–686.

5. Creutzig U, Ritter J, Budde M, et al. Early deaths due to hemor-rhage and leukostasis in childhood acute myelogenous leukemia.Associations with hyperleukocytosis and acute monocytic leuke-mia. Cancer 1987;60:3071–3079.

6. Steuber CP, Nesbit ME. Clinical assessment and differential di-agnosis of the child with suspected cancer. In: Pizzo PA, PoplockDG, eds. Principles and practice of pediatric oncology. Philadel-phia: Lippincott-Raven; 1997. p 273–287.

7. Katz JA, Wagner ML, Gresik MV, et al. Typhlitis. An 18-yearexperience and postmortem review. Cancer 1990;65:1041–1047.

8. Moir DH, Bale PM. Necropsy findings in childhood leukaemia,emphasizing neutropenic enterocolitis and cerebral calcification.Pathology 1976;8:247–258.

9. Nigro JF, Gresik MV, Fernbach DJ. Value of the postmortemexamination in a pediatric population with leukemia. J Pediatr1990;116:350–114.

10. Anderlini P, Luna M, Kantarjian HM, et al. Causes of initialremission induction failure in patients with acute myeloid leuke-mia and myelodysplastic syndromes. Leukemia 1996;10:600–608.

11. Ventura GJ, Hester JP, Smith TL, Keating MJ. Acute myeloblasticleukemia with hyperleukocytosis: risk factors for early mortalityin induction. Am J Hematol 1988;27:34–37.

12. Ribeiro RC, Pui CH. The clinical and biological correlates ofcoagulopathy in children with acute leukemia. J Clin Oncol 1986;4:1212–1218.

13. Love LJ, Schimpff SC, Schiffer CA, Wiernik PH. Improved prog-nosis for granulocytopenic patients with gram-negative bacter-emia. Am J Med 1980;68:683–688.

14. Anonymous. Empiric antifungal therapy in febrile granulocytope-nic patients. EORTC International Antimicrobial Therapy Coop-erative Group. Am J Med 1989;86:668–672.

15. Ries LA, Kosaary CL, Hankey BF, et al., eds. SEER cancer sta-tistics review, 1973–1994. Bethesda (MD): National Institutes ofHealth, National Cancer Institute; 1997. Report No. DHHS PublNo. (NIH)07-2789.

16. Smith MA, Freidlin B, Ries LA, Simon R. Trends in reportedincidence of primary malignant brain tumors in children in theUnited States. JNCI 1998;90:1269–1277.

17. Regis J, Bouillot P, Rouby-Volot F, et al. Pineal region tumors andthe role of stereotactic biopsy: review of the mortality, morbidity,and diagnostic rates in 370 cases. Neurosurgery 1996;39:907–912.

18. Roux FX, Nataf F, Pinaudeau M, et al. Intraspinal meningiomas:review of 54 cases with discussion of poor prognostic factors andmodern therapeutic management. Surg Neurol 1996;46:458–463.

19. Heideman RL, Packer RJ, Albright LA, et al. Tumors of thecentral nervous system. In: Pizzo PA, Poplock DG, eds. Principlesand practice of pediatric oncology. Philadelphia: Lippincott-Raven; 1997. p 642.

20. Haas D, Ablin AR, Miller C, et al. Complete pathologic matura-tion and regression of stage IVS neuroblastoma without treatment.Cancer 1988;62:818–825.

21. Yamamoto K, Hanada R, Kikuchi A, et al. Spontaneous regressionof localized neuroblastoma detected by mass screening. J ClinOncol 1998;16:1265–1269.

22. Woods WG, Tuchman M, Robison LL, et al. A population-basedstudy of the usefulness of screening for neuroblastoma. Lancet1996;348:1682–347.

23. Bessho F, Hashizume K, Nakajo T, Kamoshita S. Mass screeningin Japan increased the detection of infants with neuroblastomawithout a decreases in cases in older children. J Pediatr 1991;119:237–141.

24. von Schweinitz D, Hecker H, Harms D, et al. Complete resectionbefore development of drug resistance is essential for survivalfrom advanced hepatoblastoma—a report from the German Co-operative Pediatric Liver Tumor Study HB-89. J Pediatr Surg1995;30:845–852.

25. Estrin JA, Belani KG, Karnavas AG, et al. A new approach tomassive blood transfusion during pediatric liver resection. Surgery1986;99:664–670.

26. King DR, Ortega J, Campbell J, et al. The surgical management ofchildren with incompletely resected hepatic cancer is facilitated byintensive chemotherapy. J Pediatr Surg 1991;26:1074–1080.

27. Pierro A, Langevin AM, Filler RM, et al. Preoperative chemo-therapy in ‘unresectable’ hepatoblastoma. J Pediatr Surg 1989;24:24–28.

28. Thomas BL, Krummel TM, Parker GA, et al. Use of intraoperativeultrasound during hepatic resection in pediatric patients. J PediatrSurg 1989;24:690–692.

29. Kitahara S, Makuuchi M, Ishizone S, et al. Successful left triseg-mentectomy for ruptured hepatoblastoma using intraoperativetransarterial embolization. J Pediatr Surg 1995;30:1709–1712.

30. Chang JH, Janik JS, Burrington JD, et al. Extensive tumor resec-tion under deep hypothermia and circulatory arrest. J Pediatr Surg1988;23:254–258.

Early Deaths in Childhood Cancer 347