Late deaths and survival after childhood cancer: implications for cureBMJ 1994; 309 doi: http://dx.doi.org/10.1136/bmj.309.6948.162 (Published 16 July 1994) Cite this as: BMJ 1994;309:162
- C M Robertson,
- M M Hawkins,
- J E Kingston
- University of Oxford, Childhood Cancer Research Group, Oxford OX2 6HJ
- St Bartholomew's Hospital, London EC1
- Correspondence to: Dr Hawkins.
- Accepted 27 April 1994
Objectives: To investigate causes of death and survival in subjects who had survived at least five years after diagnosis of childhood cancer; to compare observed mortality with that expected in the general population; and to compare results with a corresponding cohort diagnosed earlier.
Design : Retrospective cohort study.
Setting : Population based National Register of Childhood Tumours.
Subjects : 9080 five year survivors of childhood cancer diagnosed in Britain during 1971-85, of whom 793 had died. Comparison with corresponding cohort diagnosed during 1940-70.
Main outcome measures : Cause of death established from all available sources of information (including hospital and general practitioner records and postmortem reports) and underlying cause of death coded on death certificate.
Results : Of the 781 deaths for which sufficient information was available, death was attributed to recurrent tumour in 578 (74%) cases, treatment related effect in 121 (15%), second primary tumour in 52 (7%), and other causes in 30 (4%). Comparison of observed mortality with that expected in the general population indicated a fourfold excess of deaths from non-neoplasic causes. The risk of dying of recurrent tumour in the next 10 years after surviving five years from diagnosis during 1940-70 and 1971-85 fell from 12% to 8%. The risk of dying from a treatment related effect increased slightly from 1% to 2%.
Conclusion : Improvements in five year survival after childhood cancer have been accompanied by a reduction in risk of dying from recurrent tumour during the subsequent 10 years and by a slight increase in risk of dying from treatment related effects. The results provide information relevant to decisions concerning balance between effective treatments and their potentially harmful effects.
During recent decades survival to five years after diagnosis of childhood cancer has greatly improved, but little is known about subsequent survival
We studied over 9000 patients who had survived five years after a diagnosis of childhood neoplasm in 1971-85 and compared them with over 4000 patients who had survived at least five years after diagnosis in 1940- 70
There was considerable improvement in survival to five years and, in the subsequent 10 years, a substantial reduction in the risk of death from recurrent cancer and slightly increased risk of death from an adverse effect of treatment
Recurrent tumour remained the largest single cause of death among the five year survivors, and mortality from non-neoplastic causes was four times higher than in the general population
Although increased mortality from treatment is of concern, the risk is greatly outweighed by the danger of recurrent cancer; therefore the primary aim should remain the development of treatments that reduce the risk of recurrent cancer
Survival after childhood cancer has dramatically improved over recent decades, with many diagnostic groups now showing a five year survival rate of at least 60%.1 This has resulted from increasing use of intensive chemotherapy combined with other modalities of treatment, improved generalised supportive management, and increased centralisation of care. With increasing numbers of long term survivors, the long term effects of treatment for childhood malignancy must be carefully monitored because these will become increasingly important in determining future treatment protocols.
Patients who survive at least five years after childhood cancer experience an excess of deaths compared with the general population.2 The Childhood Cancer Research Group examined cause of death after five year survival in a cohort of children treated for cancer before 1971 and found excess mortality from certain causes and preventable deaths.3 Children treated more recently would probably show different patterns of mortality because of the improvements in survival and modern treatment regimens, in which chemotherapy is used more extensively. A recent study of causes of death in all patients (not just five year survivors) who had been treated for non-Hodgkin's lymphoma showed that there was considerable mortality related to treatment, of which a substantial proportion was related to chemotherapy.4
The objectives of this study were, firstly, to determine the causes of deaths of children treated for cancer during 1971-85 who had survived at least five years and to relate the causes of death to type of tumour and treatment; secondly, to compare the observed mortality from specific causes with that expected in the general population to identify any departure from the expected pattern of mortality; and thirdly, to compare these results with those from our previous study of late deaths3 to detect any differences in the pattern of late mortality after treatment in 1940-70 and 1971-85. By investigating the proportion of patients dying of recurrent tumour at specified times after five year survival, we hoped to clarify the extent to which different childhood neoplasms have proved curable in the two periods of diagnosis.
The National Register of Childhood Tumours, maintained by the Childhood Cancer Research Group in Oxford, has been routinely notified of tumours occurring in children aged under 15 years since 1962 through the national cancer registration scheme operating in Britain. This provides, within the limits of completeness of registration, a population based series of childhood cancer cases. From this series we selected patients who had cancer diagnosed between 1971 and 1985 and who had survived at least five years after diagnosis. All diagnoses from the Birch and Marsden classification5 apart from Langerhan cell histiocytosis were included. The cohort was followed up until the end of 1990; deaths occurring after this point were ignored. This period of diagnosis was studied as it followed immediately after our previous study of children diagnosed between 1940 and 1970.3
For deaths occurring more than five years after diagnosis we tried to obtain details of the circumstances leading to death and to establish the precise underlying cause of death. The main sources of information were the files held at the National Register of Childhood Tumours and included death certificates, cancer registrations, United Kingdom Children's Cancer Study Group registrations and follow up forms, and abstractions from general practitioners' notes and hospitals' notes. When the information in the files was inadequate we obtained further information from hospital notes, postmortem reports, and in a few cases by direct questions to the consultant responsible for the care of the relevant child.
For each death we assigned the underlying cause, based on all available information, to one of the following broad categories: recurrent tumour, treatment related cause (other than second primary tumour), second primary tumour, accident or trauma, medical condition not related to the tumour or its treatment, or insufficient information. In addition, we recorded the International Classification of Diseases, eight and ninth revisions,6, 7 code for the underlying cause of death on the death certificate, as coded by the Office of Population Censuses and Surveys for England and Wales and by the General Register Office for Scotland.
We adopted certain conventions in assigning deaths to the above categories. Bronchopneumonia was frequently given as the cause of death for patients who had gradually deteriorated with recurrent tumour; these deaths were all coded as recurrent tumour deaths rather than infection. Treatment related deaths were defined as a death where treatment was considered to be the major contributing factor to the death but not necessarily the only factor: if a child died from infection while receiving chemotherapy or within six months of completing treatment or within a year of a bone marrow transplantation the death was coded as treatment related. When death occurred from any cause and there was insufficient evidence to determine whether the patient had active disease at the time the death was coded to insufficient information.
We examined survival to each particular event of interest (death from recurrent tumour or a treatment related cause) with standard life table techniques.8 When analysing survival to a particular cause of death we regarded all other causes of death as losses to follow up. Mortality from specific causes of death among survivors was compared with that expected in the general population of England and Wales (Office of Population Censuses and Surveys, unpublished report). Expected numbers were estimated by accumulating the person years at risk within strata (defined by sex, age, and calendar period) and multiplying the total person years by the corresponding population mortality. Observed and expected numbers of deaths were compared with exact Poisson methods.9, 10 For this comparison it was necessary to rely on death certificates alone for information on observed deaths because the general population mortality was based on death certificates.
To provide an accurate comparison we carried out new parallel analyses of the two cohorts of five year survivors diagnosed in 1940-70 and 1971-85. We used life table methods to estimate the percentage of patients who survived to five years after diagnosis during 1962-70 and 1971-85; only records relating to patients surviving at least three years were available for patients diagnosed before 1962. We also used life table analyses to estimate the percentage of five year survivors diagnosed in 1940-70 and 1971-85 who died of recurrent tumour or a treatment related cause within the subsequent 10 years.
Of the 9080 five year survivors of childhood neoplasms diagnosed in 1971-85 who were eligible for the analysis, 793 had died by 1990. Death certificates were obtained for 789 of the deaths, hospital notes for 334, postmortem reports for 90, and general practitioner notes for 533. Table I shows that there was insufficient information available to code the cause of 12 deaths; of the remaining 781 deaths, 578 (74%) were attributed to recurrent tumour and 52 (7%) to second primary tumour. There was considerable variation among the different tumour groups in the proportion of deaths due to a particular cause. Among those who had had a diagnosis of acute lymphoblastic leukaemia, tumours of the central nervous system, malignant bone tumours, or soft tissue sarcomas, over 75% of deaths were due to recurrent tumour. In contrast only 31% of deaths after non-Hodgkin's lymphoma were due to recurrent tumour, while 38% resulted from second primary tumours and 31% from a medical condition related to treatment. Other groups for which more than 20% of deaths were attributable to a medical condition related to treatment were those who had had a diagnosis of acute non-lymphoblastic leukaemia, neuroblastoma, or Wilms's tumour.
Table II shows the causes of death during follow up. The most common causes of death during 5-9 years after diagnosis of childhood neoplasm was recurrent tumour, except among survivors of non-Hodgkin's lymphoma, for whom the number of deaths from second primary tumour equalled the number of deaths from recurrent tumour. The most common cause of death among 10 year survivors was still recurrent tumour for all tumours, except for survivors of non-Hodgkin's lymphoma, among whom there were more deaths from second primary tumour, and of Wilms's tumour, for whom treatment related deaths were most common. The proportion of deaths due to causes other than recurrent tumour increased from 145/632 (23%) during 5-9 years after diagnosis to 41/123 (33%) during 10-14 years and 10/19 (53%) during 15-19 years. The probability of death from recurrent tumour decreased from 6.9% (487/7088) during 5- 9 years after diagnosis to 2.3% (82/3622) during 10-14 years and to 0.8% (9/1075) during 15-19 years. These probabilities are approximations calculated with the mean number of survivors in each period as the denominator.
Of the 203 deaths from causes other than recurrent tumour (table I), we obtained hospital notes or postmortem reports for 183 (90%). Of the 781 deaths to which we could attribute a cause, 20 (3%) were due to medical conditions not directly related to the tumour or its treatment while 121 (15%) were due to medical conditions related to treatment. These 121 treatment related deaths were subcategorised: infections (54), respiratory (13), cardiovascular (11), central nervous system (11), graft versus host disease (7), operative (7), renal (5), gastrointestinal (including hepatic) (5), metabolic or endocrine (3), specific drug toxicity (3), and haemorrhage (2). (Details of these deaths will be given in a separate paper.) Of particular interest were 10 of the 13 respiratory deaths, which were caused by pulmonary fibrosis that occurred after treatment with carmustine, busulphan, or thoracic irradiation; seven of the 11 cardiovascular deaths, which followed treatment with adriamycin; and a further three cardiac deaths which followed thoracic irradiation.
Table III is based on life table estimates and shows the percentage of patients surviving at least five years after diagnosis of childhood neoplasm during 1962-70 and 1971-85 and the percentage of five year survivors with cancers diagnosed in 1940-70 and 1971-85 who subsequently died of recurrent tumour or a treatment related cause in the next 10 years. The proportion of patients surviving five years after diagnosis increased from 26% for those with a diagnosis during 1962-70 to 50% for those with diagnosis during 1971-85. Of those who survived five years, the proportion dying of recurrent tumour in the next 10 years decreased from 11.5% to 8.3% and the proportion dying from a treatment related cause increased slightly from 0.8% to 1.9%. There was considerable variation in these proportions between the different diagnostic groups. Survivors of acute lymphoblastic leukaemia and of Hodgkin's disease showed the greatest reduction in deaths from recurrent tumour. For acute lymphoblastic leukaemia, this reduction is probably explained by the introduction during 1971-85 of prophylaxis against central nervous system disease, which accounted for many of the deaths from recurrent disease in the earlier period. The reduction in Hodgkin's disease is probably related to the more extensive use of chemotherapy during 1971-85. Among the five year survivors of acute non-lymphoblastic leukaemia, Hodgkin's disease, non-Hodgkin's lymphoma, retinoblastoma, and Wilms's tumour diagnosed in 1971-85, no more than 5% died of recurrent tumour during the next 10 years. However, more than 10% of the five year survivors of acute lymphoblastic leukaemia, central nervous system tumours (except for astrocytoma), and Ewing's tumour died from recurrent tumour during the next 10 years.
Table IV is based on information from death certificates only and compares the observed and expected number of deaths from all causes other than neoplasm among five year survivors. The number of non- neoplastic deaths observed was four times higher than expected. The number of observed deaths from infections was 25 times higher than expected and all but one death occurred among survivors of acute lymphoblastic leukaemia. There was a 20-fold excess of respiratory deaths and a fivefold excess of cardiovascular deaths. When all diagnostic groups were considered together there was no evidence of an excess of deaths from suicide or non-medical accidents. Survivors of acute lymphoblastic leukaemia experienced four times the expected number of deaths: in particular there were 83 times the expected number of deaths from infection and 19 times the expected number of deaths from respiratory causes. Survivors of central nervous system tumours experienced five times the expected number of deaths; in particular, there were 36 times the expected number of respiratory deaths. The other excesses seen in table IV arose from a variety of different causes, and no systematic pattern emerged.
Comparison of patients with childhood cancer diagnosed in 1962-70 and 1971-85 confirms that there have been great improvements in survival: five year survival after all childhood neoplasms increased from 26% in the first period to 50% in the second. Of those surviving at least five years, the proportion dying of recurrent tumour during the subsequent 10 years decreased from 12% after a diagnosis in 1940-70 to 8% after a diagnosis in 1971-85, while the proportion dying of a treatment related cause increased only slightly, from 1% to 2%. Follow up beyond 15 years from diagnosis is needed to determine whether more recently introduced chemotherapy is associated with adverse outcomes in the longer term.
To our knowledge the only other published study of late deaths in children treated for cancer during the era of modern treatment involving chemotherapy, which began in Britain around 1970, is a relatively small clinical series.11 A study of late deaths after treatment during 1945- 74 from America12 showed a similar excess mortality to that identified among British five year survivors diagnosed during 1940- 70.3 We have not described deaths from second primary tumours in detail because these will be discussed in a subsequent paper; the patterns and incidence of second primary tumours among survivors of childhood cancer are described elsewhere.13, 14
Treatment related deaths
Some might consider that our definition of treatment related death was too broad. In particular, infective deaths that occurred while patients were still receiving treatment might be regarded as deaths from recurrent tumour since this was the reason for their treatment. Despite our broad definition of treatment related death, however, the risk of such deaths among five year survivors was still small, particularly when compared with the risk of death from recurrent tumour.
Of the 121 treatment related deaths, at least 96 occurred in patients who had undergone treatment for recurrent or persistent disease, and the deaths were therefore probably unavoidable. Some of the remaining 25 deaths may have been avoidable, including six deaths from doxorubicin cardiomyopathy, six from pulmonary fibrosis, and four from other radiation effects. It is now well recognised that late manifestations of anthracycline cardiotoxicity occur.15, 16 The figures in this study are likely to be an underestimate of the mortality from this cause as the follow up was relatively short, particularly for the more recent diagnoses. Pulmonary fibrosis continues to be a cause of treatment related mortality, with chemotherapy now emerging as an associated factor as well as radiotherapy. Four of the patients who died from pulmonary fibrosis after treatment with carmustine have previously been reported.17 Of the six deaths (including two patients who had undergone treatment for recurrent disease) from other radiation effects, four were related to effects on the cardiovascular system and two were related to intra-abdominal fibrosis. Deaths from infection accounted for a large proportion of the treatment related deaths. Although most of these deaths occurred in children who were still being treated or who had recently been treated for relapse, some of these deaths were possibly avoidable. In particular, the deaths from measles, varicella, pneumocystis, and overwhelming sepsis after splenectomy might have been avoided.
It is of concern that the risk of treatment related deaths increased, but it is important to note that the greatest risk of death after five year survival remained recurrent tumour, which accounted for 77% of the deaths during five to nine years after diagnosis, 67% during 10-14 years, and 47% during 15-19 years. Therefore, although it is important to reduce avoidable treatment related mortality, the primary aim should remain the development of treatments that reduce the number of deaths from recurrent tumour among five year survivors.
Limitations of study - This study provided information on the causes of late death after particular types of childhood cancer in Britain. However, most of the cases recorded in the National Register of Childhood Tumours include only crude information on treatment. In particular, there is no record of the specific drugs used in treatment. Thus, we were unable to estimate the risk of specific causes of death in relation to type of treatment received. In particular, although the risk of death from anthracycline related cardiomyopathy within 10 years was small for the whole cohort of five year survivors, we have no means of identifying the subgroups of patients exposed to anthracyclines. In addition, most of the cases recorded in the register have no details of history of relapse or recurrence, and it is therefore not possible to examine the risk of particular causes of death among survivors who do not have a relapse.
Our studies, being mainly population based, are unique in providing information on the long term prospects of cure and the risks of dying from particular causes for all patients treated for childhood cancer in Britain. It is well known that there have been great improvements in five year survival after childhood cancer, but the question of what happens thereafter has rarely been investigated. Our results are encouraging in that the substantial improvement in five year survival was not accompanied by a substantially increased subsequent risk of death from recurrent tumour or a treatment related cause. Continued monitoring of these survivors of childhood cancer is important as unforeseen excesses of mortality might emerge. Such studies of mortality reveal the fatal long term risks of childhood cancer and its treatment and such risks are becoming increasingly important in guiding decisions about future treatments.
We thank the consultants who provided much of the information on which this paper is based; the United Kingdom Childrens Cancer Study Group for its support; the consultants who made comments on early drafts of this paper; Michael Potok, Elizabeth Mowat, Maureen Allen, Pat Brownbill, Dave Winter, Hazel Burton, and other members of the Childhood Cancer Research Group for their help with this work; and our colleagues Charles Stiller and Gerald Draper for their support and advice. Our long term follow up study is supported by the Cancer Research Campaign and Leukaemia Research Fund. The Childhood Cancer Research Group is supported by the Department of Health and the Scottish Home and Health Department. This work would have been impossible without the help of the Office of Population Censuses and Surveys (particularly the NHS central register), the Information Services Division of the Common Services Agency of the Scottish Health Service, the Registrar General for Scotland, and the regional cancer registries.