Early mortality in children with cancer in Denmark and Sweden: The role of social background in a setting with universal healthcare

Socioeconomic differences in overall survival from childhood cancer have been shown previously, but the underlying mechanisms remain unclear. We aimed to investigate if social inequalities were seen already for early mortality in settings with universal healthcare. From national registers, all children diagnosed with cancer at ages 0–19 years, during 1991–2014, in Sweden and Denmark, were identified, and information on parental social characteristics was collected. We estimated odds ratios (OR) and 95% confidence intervals (CI) of early mortality (death within 90 days after cancer diagnosis) by parental education, income, employment, cohabitation, and country of birth using logistic regression. For children with acute lymphoblastic leukaemia (ALL), clinical characteristics were obtained. Among 13,926 included children, 355 (2.5%) died within 90 days after diagnosis. Indications of higher early mortality were seen among the disadvantaged groups, with the most pronounced associations observed for maternal education (ORadj_Low_vs_High 1.65 [95% CI 1.22–2.23]) and income (ORadj_Q1(lowest)_vs_Q4(highest) 1.77 [1.25–2.49]). We found attenuated or null associations between social characteristics and later mortality (deaths occurring 1–5 years after cancer diagnosis). In children with ALL, the associations between social factors and early mortality remained unchanged when adjusting for potential mediation by clinical characteristics. In conclusion, this population‐based cohort study indicated differences in early mortality after childhood cancer by social background, also in countries with universal healthcare. Social differences occurring this early in the disease course requires further investigation, also regarding the timing of diagnosis.

were seen among the disadvantaged groups, with the most pronounced associations observed for maternal education (OR adj_Low_vs_High 1.65 [95% CI 1.22-2.23])and income (OR adj_Q1(lowest)_vs_Q4(highest) 1.77 [1.25-2.49]).We found attenuated or null associations between social characteristics and later mortality (deaths occurring 1-5 years after cancer diagnosis).In children with ALL, the associations between social factors and early mortality remained unchanged when adjusting for potential mediation by clinical characteristics.In conclusion, this population-based cohort study indicated differences in early mortality after childhood cancer by social background, also in countries with universal healthcare.Social differences occurring this early in the disease course requires further investigation, also regarding the timing of diagnosis.

What's new?
In high-income countries in Europe, childhood cancer survival is suspected of being linked to socioeconomic status, despite mandates for equal access to healthcare.Little is known, however, about the impact of socioeconomic differences on early mortality in settings with universal healthcare.This study analyzed mortality following childhood cancer diagnosis in Sweden and Denmark.Early mortality was associated with socioeconomic status, wherein socially disadvantaged groups were at increased risk of death within three months of diagnosis.This was most clearly seen for maternal income and education.The relevance of socioeconomic differences on timing of childhood cancer diagnosis and survival warrants further investigation.

| INTRODUCTION
4][5] The use of standardized treatment protocols with better risk grouping, clinical trial participation, improved supportive care, and cross-national collaborations among paediatric oncologists are examples of reasons underlying this development. 1,2,6cioeconomic differences in overall survival from childhood cancer have previously been described, also for high income countries in Europe, where equal access to healthcare is presumed. 7,8British researchers observed that differences in survival from acute lymphoblastic leukaemia (ALL) by socioeconomic status emerged at the time when treatment management required parental/child adherence, that is from start of oral treatment in the outpatient setting, hypothesizing that socioeconomic differences in treatment adherence may explain their findings. 9However, studies from Sweden and Denmark have shown survival differences by parental education or family characteristics already within the first year after diagnosis. 10,11A recent study from the US investigated the role of area-based social measures in relation to early mortality in childhood cancer, and observed that children living in disadvantaged counties had a higher risk of dying during the first month after a haematological malignancy, compared to children living in more affluent areas. 12If there are social differences in mortality this early after the cancer being diagnosed, these differences are unlikely to be related to treatment adherence, but more likely to have other underlying explanations, which may be related to the timeliness of diagnosis, disease severity or treatment initiation.However, in contrast to the United States, access to cancer care services is assumed to be universal in the Nordic countries, and therefore it is of importance to assess whether similar associations are found in this setting.We hypothesize that social background could be associated with early mortality also in the Nordic countries, and that this is related to other underlying mechanisms than pure income barriers to treatment.
Using a population-and register-based cohort study including all children diagnosed with cancer in Denmark and Sweden, we aimed to determine whether social background was associated with early mortality in childhood cancer.We also sought to assess whether potential associations differed from corresponding associations for later mortality.Moreover, we examined the role of clinical characteristics for the associations between social background and early mortality, among children with ALL.

| Setting
Denmark and Sweden have a civil registration system with nationwide administrative registers and a unique personal identification number that enables individual linkage of information.Moreover, Denmark and Sweden have similar healthcare systems, paid through taxes and with universal access, as well as longstanding largely standardized diagnostic procedures and treatment protocols, 6,13 which made it appropriate to combine data from the two populations into one comprehensive register-based cohort study.

| Study population
We identified all incident cases of first cancers (including also non-malignant central nervous system [CNS] tumours) diagnosed in children aged 0-19 in the period 1991-2014 from the Danish and Swedish national cancer registers. 14Registration of all cancers diagnosed within the two countries is mandatory and does not depend on given treatment.We categorized cancer diagnoses according to the International Classification of Childhood Cancer (ICCC), 15 into leukaemia (group I: ALL [group Ia] and acute myeloid leukaemia [AML, group Ib] separately), lymphomas (group II), CNS tumours (group III) and other non-CNS solid tumours (all other malignant tumours).We excluded children who were not resident in the country at the time of diagnosis or emigrated within 1 year after diagnosis, children for which no biological parents could be identified, and children with the cancer pre-disposing syndromes Down syndrome, neurofibromatosis, and tuberous sclerosis (Figure 1).

| Outcome
Early mortality, our primary outcome, was defined as death from any cause within the first 90 days after the cancer diagnosis.Information on the date of death was obtained from the national causes of death registers, whereas the date of diagnosis was obtained from the national cancer registers.Date of diagnosis is defined as the date of the first examination/test underlying the diagnosis (Sweden), 16 and as the "admission date for the first contact during which the diagnosis was confirmed" (Denmark). 17In the Danish Cancer Registry, the exact date of diagnosis is available from 2004 onwards, before that only month of diagnosis was recorded. 14For the main analyses, the date of diagnosis for Danish children diagnosed prior to 2004 were set to the 15th of the month.For comparison with earlier studies, we also assessed mortality within the first 30 days after diagnosis as a secondary analysis including only children for which the exact date of diagnosis was available (n = 11,178; 80.3%).
To investigate if potential associations with early mortality differ from associations with later mortality, we included the outcome later mortality (defined as deaths occurring 1-5 years, 366-1825 days, after cancer diagnosis).To have a clear distinction between early and later mortality, deaths occurring 91-365 days after the diagnosis were not our main outcome.For transparency, these deaths were analysed separately as a secondary analysis.

| Parental social characteristics
We retrieved information about parental social characteristics for biological parents the year closest available before the child's diagnosis.
The information was retrieved from the Danish social registries (Education registers, The Income Statistics Register, The Integrated Database for Labour Market Research and Danish Civil Registration System), [18][19][20] and the Swedish Longitudinal integrated database for health insurance and labour market studies and Total Population Register. 21,22The social characteristics included parental education (lower secondary or less, upper secondary, postsecondary), employment (employed, unemployed [including outside of workforce]), country of birth (Nordic [Denmark, Sweden, Finland, Iceland, Norway], non-Nordic), cohabitation (single, cohabiting/married), and income (quartiles).Parental income was defined as each parent's individual disposable income, categorized in quartiles based on the sex-and calendar year specific income distribution of the entire population in Denmark and Sweden, respectively.

| Statistical analysis
We assessed the association between each of the parental social characteristics and early mortality by estimating odds ratios (OR) with 95% confidence intervals (CI), using logistic regression models.We fitted univariable and multivariable models, adjusted for sex, age at diagnosis (0, 1-4, 5-9, 10-14, 15-19 years old), year of diagnosis (categorized in 10-year intervals), and country of diagnosis, for each social factor separately.We conducted separate analyses for ALL, AML, lymphomas, CNS-tumours, and non-CNS solid tumours, as well as for country.We chose a logistic regression model instead

Among the 13,926 children diagnosed with cancer in Denmark and
Sweden during 1991-2014, 355 children (2.5%) died within 90 days after diagnosis (Table 1).Figure 2 shows the time trends of early and later mortality.The group of children dying early included a larger proportion of those diagnosed in the first year of life, in the earlier time period (1991-2000), and with AML and CNS-tumours, compared to the full cohort (Table 1).
Children of mothers with lower secondary or less education had a higher risk of early mortality compared to children of mothers with postsecondary education (OR adj 1.65, 95% CI 1.22-2.23).A similar pattern was seen for paternal education although somewhat less pronounced (OR adj 1.35, 95% CI 0.97-1.88),Table 2. Higher risks of early mortality were also seen among children of mothers in the lowest income quartile compared to children with high maternal income (OR adj 1.77, 95% CI 1.25-2.49),whereas such association was not evident for paternal income.Children of unemployed mothers were at higher risk of early mortality, compared to children of employed mothers (OR adj 1.29, 95% CI 1.01-1.64).Moreover, there was a tendency of increased odds of early mortality among children to parents born outside of the Nordic countries (Maternal: OR adj 1.19, 95% CI 0.86-1.63;Paternal: OR adj 1.25, 95% CI 0.92-1.70).In contrast to the results for early mortality, we found weaker or null associations between parental social characteristics and later mortality (deaths 366-1825 days after diagnosis), Table 2. Results were overall similar in Denmark and Sweden; associations for early mortality were seen especially between maternal education and income in both countries (Supplementary Table 1).The impact of paternal education on early mortality was somewhat more pronounced in Denmark than in Sweden.
The stratified analyses for early mortality by cancer types (Table 3) revealed overall similar results as to those seen in the main analysis although confidence intervals were wide.Differences in early mortality by parental education were particularly evident among children with CNS and non-CNS solid tumours.For maternal income, there was a pattern of higher early mortality in the lower income quartiles, particularly pronounced among children with ALL and non-CNS solid tumours (Table 3).
In total, 60/355 children dying early were diagnosed with ALL, and the proportion of early mortality among all children diagnosed with ALL was 2.3% (60/2590).Most of the children dying early from ALL died during the induction phase of the treatment, while children dying later most often died after a relapse (Supplementary Table 2).
We examined the associations between parental social factors and clinical characteristics at time of diagnosis among children with ALL (Supplementary Table 3); maternal income was associated with both immunophenotype and genotype ( p-value .014and <.001, respectively), but no clear gradients were seen.Statistically significant associations were also observed between paternal cohabitation and country of birth, and genotype ( p-value .016and .041,respectively), as well as for paternal education level and white blood cell count (p-value .045).To assess the potential mediation from clinical characteristics on the associations with early mortality, we further adjusted for these characteristics in the regression models, but this had no impact on the association between parental social factors and early mortality (Table 4).(Continues) In secondary analyses, we defined the outcome as early mortality within the first 30 days after diagnosis which occurred in 147 children (1.3%).For these analyses, Danish children diagnosed before 2004 were excluded since their exact date of diagnosis was not available.
These analyses showed similar patterns to those of the main analyses, with higher risks of early death especially seen among children of parents with lower education (Supplementary Table 4).We also analysed 686 deaths occurring 91-365 days after the diagnosis among 13,571 included children and observed attenuated associations with parental education compared to the main results of early mortality.However, increased risks of death within this period were observed for children whose fathers had low income or were unemployed, as well as for children of single mothers (Supplementary Table 5).

| DISCUSSION
In this unique population-based cohort study including virtually all children diagnosed with childhood cancer in Denmark and Sweden in Cohabitating parents were defined as parents living with a marital or non-marital partner, with the exception that in Sweden only parents living with a marital partner or a non-marital partner that he/she have common children with, were registered as cohabiting.Our findings of a higher early mortality among the disadvantaged groups are in line with the results from a US study, where several measures of social disadvantage were shown to be associated with early deaths from haematological malignancies and CNS-tumours in crude analyses. 12However, in multivariable analyses, the only analysed indicator of social disadvantage was neighbourhood income among children with haematological malignancies, and the higher risk of early death among children in disadvantaged income areas remained. 12We have not identified any studies from countries with universal healthcare investigating the association between social factors and early childhood cancer mortality.However, socioeconomic differences in overall mortality from childhood cancer has been observed also in such countries, 7,8 for example Finland, Switzerland and the United Kingdom. 9,23,24It has been hypothesized that the differences could be explained by lower adherence to complex treatment protocols among socially disadvantaged children, supported by results from a British study that observed survival differences in ALL when the treatment moved to outpatient care. 9However, a recent Danish study including 173 children with ALL, showed no differences in adherence to maintenance therapy between families of different socioeconomic position, but a difference in doctor's compliance to protocol recommendations measured by prescribed drug doses. 25Both family adherence and doctor's compliance would have an impact on survival later in the disease course, while our findings indicate that the survival differences are mainly evident in the period closest to the diagnosis.However, the underlying mechanisms might differ between settings and cancer types, and our cancer type specific analyses do not allow firm conclusions.
To assess whether the early differences in mortality by social background, could be explained by differences in more specific features of the disease, we assessed the associations between social and clinical characteristics in children with ALL.Although there were some statistically significant findings, there were no clear patterns and inclusion of these clinical characteristics in the regression models of early mortality did not appreciably change effect estimates.Two previous studies assessed the association between socioeconomic status or parental education, and white blood cell count among children with leukaemia, and observed no associations, 26,27 although children of single parents were more likely to have a higher white blood cell count. 27e first symptoms of childhood cancer are often few and unspecific, 28 and an increased utilization of primary care has been documented already several months before the diagnosis. 29The increase in use of primary care seems to be modified by social background with somewhat higher rates of additional contacts among disadvantaged groups. 29This may suggest that disadvantaged families experience more difficulties in the trajectory to a cancer diagnosis which may result in a later defined date of diagnosis, although such conclusions are speculative, and its impact on prognosis and early mortality is unclear.
We included several measures of social background in this study since they might have different associations with cancer survival. 30  We included individual level information of both biological parents, while area-based measures of disadvantage have been used previously. 12When area-based measures are used as a proxies for individual measures, it might misclassify individuals leading to underestimation of potential associations. 31However, the social context of the neighbourhood may in itself affect health, 31 and potentially early mortality in childhood cancer, which is an aspect that remains to be investigated in the Nordic context.
The infrastructure in Denmark and Sweden with long-standing high-quality national population-and health registers, provides an ideal setting for investigating socioeconomic differences in relation to the risk of early childhood cancer mortality.Information on parental social characteristics was gathered before the cancer diagnosis and did not depend on study participation and self-reporting.Moreover, registration in the national cancer registers is mandatory and independent on both treatment and survival, meaning we were likely to capture virtually all cases.We observed that 2.5% of the children diagnosed with cancer died within 3 months after diagnosis, and 1.3% within the very first month after diagnosis.These proportions are comparable to estimates of deaths within the first month after diagnosis from other population-based registers in high-income countries, 12,32,33 although direct comparisons are challenging because of differences in the definition of date of diagnosis. 32Due to the difficulty of setting a date of diagnosis in diseases with no clear onset, and the potential differences and temporal changes in practice at clinics and registry offices, we focused on early deaths occurring within 90 days after diagnosis, instead of 1 month. 12This is also the reason why we chose to use a logistic regression model instead of a timeto-event approach.Most of the early deaths occurred within one and a half month after diagnosis, and the 3 months cut point for early deaths is unlikely to be sensitive to the uncertainties in determining date of diagnosis.
A few other limitations with the current study should be mentioned.First, statistical power is limited in some analyses.Early deaths in childhood cancer are rare and even though we included nationwide information from two countries over 24-years, the cancer type specific analyses included few cases and chance cannot be ruled out as an explanation for some associations.Second, we only

K
E Y W O R D S childhood cancer, cohort, early mortality, register-based study, socioeconomic factors

For
children diagnosed with ALL (n = 2590), we obtained clinical information from the Nordic Society of Paediatric Haematology and Oncology (NOPHO) database on immunophenotype (B-precursor, T-cell), genotype (HeH/ETV6-RUNX1: "favourable genotype" [high hyperdiploidy-HeH or ETV6-RUNX1], other; available from Children first time diagnosed with cancer and registered in the Swedish or Danish National Cancer Register during 1991-2014, aged 0-19 years (including malignant tumours and non-malignant CNS tumours), n = 14,560 In the country at diagnosis, n = 14,497 No emigration within the first year after diagnosis, n = 14,473 No biological parents identified, n = 114 Emigration within the first year after diagnosis, n = 24 Formally not in the country at diagnosis, n = 63 At least one biological parent identified, n = 14,359 Down syndrome, Neurofibromatosis, Tuberous sclerosis, n = 433 Final study population, n = 13,926 F I G U R E 1 Flow chart of inclusion and exclusion criteria for the study population.
1992 and onwards), and white blood cell count at diagnosis (<10, 10-50, >50 [Â10 9 /L]).Valid linkage was done for 2380 children (91.9%) after exclusion of 35 children with mature B-cell and bilinage ALL.For children dying after a diagnosis of ALL, we obtained information regarding in what phase the death occurred (during induction, in remission, after relapse, after a second malignant neoplasm).
of a time-to event model because the former is less sensitive to the uncertainty in defining the date of diagnosis; childhood cancer has no clear onset, and practices in setting the date of diagnosis differ between clinics (i.e., sometimes the date of pathology report instead of first examination), and time periods.The associations between parental social characteristics and later mortality (deaths 366-1825 days after diagnosis) were assessed among children surviving at least 1 year.These analyses were restricted to children diagnosed until 2010 with no emigration within 5 years after diagnosis to ensure full follow-up: in the analyses of earlier deaths, we censored only children that emigrated in the first year.To evaluate the potential mediating role of clinical characteristics among children with ALL, we first assessed the univariable associations between the social and clinical characteristics using Pearson's chi-squared test.In the next step, clinical characteristics with a statistically significant association to parental social factors ( p-value <.05) were included in the regression models of parental social characteristics and early mortality to determine whether they were potentially mediating the associations observed in the main analysis.In secondary analyses, we evaluated the outcomes early mortality within the first 30 days after diagnosis, and deaths occurring 91-365 days after the diagnosis, by fitting the same regression models as in the main analyses.Analyses were conducted in SAS version 9.4 (SAS Institute, Cary, NC, USA) and Stata version 16 (StataCorp LP, College Station, TX, USA).

T A B L E 1
Characteristics of children diagnosed with cancer 1991-2014 in Denmark and Sweden, by early and later mortality.
Early mortality (0-90 days) Later mortality (366-1825 days) Total mortality within 5 years after diagnosis Early and later mortality among children diagnosed 1991-2014 (2010) F I G U R E 2 Early mortality, later mortality and 5-year mortality among all children diagnosed with cancer 1991-2014 (2010 for 5-year mortality) in Denmark and Sweden, by year of diagnosis.1991-2014, we found that social background was associated with early mortality.Indications of a pattern with the disadvantaged groups being at highest risk of death within 3 months after cancer diagnosis were seen and the most pronounced associations were observed for maternal education and income.In contrast, we observed attenuated or null associations for later mortality, occurring 1-5 years after T A B L E 2 Early and later mortality by parental social characteristics, among children diagnosed with cancer 1991-2014 in Denmark and Sweden, odds ratios (OR) with 95% confidence intervals (CI).
only among children with ALL, clinical characteristics did not explain the associations between social background and early mortality.

7
excluded children with the most common types of cancer predisposing syndromes (Down syndrome, neurofibromatosis, and tuberous sclerosis) from the study.However, since other cancer predisposing syndromes are very rare, they are unlikely to have any impact on the associations between parental social characteristics and early mortality.Last, we only had information regarding clinical characteristics among children with ALL.Inclusion of further clinical information, such as tumour stage, for other cancer types would be beneficial for understanding the underlying mechanisms of social differences in early mortality.T A B L E 4 Early mortality (death within 90 days after diagnosis) among children diagnosed with ALL 1991-2014 in Denmark and Sweden and included in the NOPHO database, a odds ratios (OR) with 95% confidence intervals (CI).
Early mortality (death within 90 days after diagnosis) among children diagnosed with cancer 1991-2014 in Denmark and Sweden, odds ratios (OR) with 95% confidence intervals (CI), stratified by cancer type.
a Including children diagnosed up until 2010 with no emigration within 5 years after diagnosis.bIncluding children surviving at least 366 days.T A B L E 3 Each social factor is included in separate models and not mutually adjusted.Numbers don't add up to the total because of missing values.Abbreviations: ALL, acute lymphoblastic leukaemia; NA, not applicable; NOPHO, Nordic Society of Paediatric Haematology and Oncology.Information is obtained for 2380/2590 children identified in the NOPHO database for whom the date of diagnosis differed by less than 31 days compared to the national cancer registers.Children with mature B-cell and bi-linage ALL are excluded (n = 35).
a b Adjusted for sex, age at diagnosis, country, and time period of diagnosis.c Less than 5 deaths in at least one category.