Introduction

Cancer is the leading cause of death among children worldwide, and the recorded incidence is still on the rise [1, 2]. Approximately 416,500 new childhood cancer cases were diagnosed in 2017, which resulted in 11.5 million disability-adjusted life-years (DALYs) globally [3]. Moreover, childhood cancer is substantially underdiagnosed due to the limited data for low-income and middle-income countries [4]. The cause of childhood cancer is not well understood, and known risk factors only include some congenital genetic syndromes, prenatal and perinatal factors. However, only 5–10% of childhood cancers are suspected to be attributable to inherited syndromes [5]. Therefore, it is necessary to evaluate the effect of prenatal and perinatal factors on the development of childhood malignancy.

Diabetes during pregnancy conferred an elevated risk of adverse pregnancy outcomes, including caesarean section and fetal macrosomia [6]. Besides, children born to women with gestational diabetes mellitus (GDM) are at greater risk of obesity, type 2 diabetes and cardiovascular disease [7,8,9]. Additionally, epidemiological studies have extensively established the relationship between diabetes and cancer incidence in adults [10,11,12,13]. Thereby, there is an urgent need to assess whether maternal diabetes increases the incidence of childhood cancers in the offspring.

Several studies have assessed the risk of development of childhood malignancy in the children of diabetic mothers, but the results were not consistent. Therefore, a detailed meta-analysis of observational studies was performed to determine the association of childhood cancer with maternal diabetes.

Methods

Search strategy and study selection

We performed a comprehensive literature search, including the PubMed, Embase and Web of science databases from their inception to June 30, 2020, to identify studies that investigated the association between maternal diabetes and risk of childhood cancer in the offspring. The literature search strategy included the following key terms (Supplementary Table 1): “maternal diabetes”, “gestational diabetes mellitus”, “pre-gestational diabetes”, “pre-pregnancy diabetes”, “pre-existing diabetes” (PDM), “childhood cancer”, “childhood malignancies”, “pediatric neoplasms” and “pediatric cancers”. Besides, we manually searched for references cited in the original study and the review.

Eligible studies should meet the following inclusive criteria: (1) studies were cohort or case–control studies; (2) the exposure of interest was maternal diabetes including PDM and GDM, and the outcome of interest was the diagnosis of childhood malignancy in the offspring; (3) the sample size with maternal diabetes was available; (4) reported the effect estimates (ESs) (odds ratio (OR), relative risk (RR), standard incidence ratio (SIR), or hazard ratio (HR)) and their corresponding 95% confidence intervals (CIs) or gave sufficient data to compute them.

Data extraction and quality assessment

The following data were extracted independently from the included literature by two reviewers: first author, year of publication, country, study type (cohort or case–control study), study period, type of maternal diabetes (diabetes (any), PDM and GDM), type of childhood cancer, number of cases with maternal diabetes, number of cancer cases, and ascertainment of maternal diabetes and cancer, adjusted ESs and corresponding 95% CIs, as well as confounders for adjustment.

Newcastle–Ottawa Scale (NOS) guidelines include 3 quality parameters: four items for selection, two items for comparability and three items for outcomes, to assess the quality of studies included in our meta-analysis [14]. Studies scoring 7–10 were identified as high-quality, those with scores of 3–6 were considered moderate quality, and the others were of low quality.

Statistical analysis

The summary ORs and corresponding 95% CIs were calculated to examine the effect of maternal diabetes on the incidence of childhood cancer in the offspring. Since the risk of childhood cancer in the general population was very low, all ESs were interpreted as OR for simplicity. A fixed-effect model was used for outcomes with low heterogeneity (I2 < 25%) among included studies; otherwise, a random-effect model was used (I2 > 25%). Cochran Q and I2 statistics were used to evaluate statistical heterogeneity between the included studies. I2 values of < 25%, 25–75% and > 75% were defined as low, medium and high heterogeneity, respectively [15]. To investigate the sources of heterogeneity, we conducted subgroup analyses and meta-regression analyses according to maternal diabetes types (diabetes (any), PDM and GDM), study design (cohort and case–control studies), study location (Europe and other regions), number of cancer cases (≥ 3000 and < 3000), cancer diagnosis (medical records and cancer registry) and whether adjusted for birth weight and birth order (Yes and No). Sensitivity analyses were performed to evaluate the robustness of the results of our meta-analysis. Funnel plots were used to assess publication bias. Furthermore, Egger’s linear regression tests and Begg’s adjusted rank correlation test were conducted and a P value of < 0.05 indicated potential publication bias [16, 17]. All statistical analyses were performed using Stata, version 14.0 (Stata Corp, College Station, Texas).

Results

Study selection

Figure 1 displayed a flow diagram of the identification process of the eligible studies. A total of 527 citations were originally retrieved by searching PubMed, Embase and Web of Science, whereas four studies were manually identified by reviewing the reference list of the relevant literature. After removing duplicate articles, 477 citations reminded. Then, 432 articles were excluded after screening the titles and abstracts. Finally, we carefully reviewed the full text of the remaining 45 articles, 22 of which were enrolled in our meta-analysis [18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39].

Fig. 1
figure 1

PRISMA flow diagram for identification of relevant articles for the meta-analysis

Full size image

Study characteristics

The baseline characteristics of  the eligible studies were shown in Table 1. Of the 22 articles included in our meta-analysis, 16 were case–control studies [18, 19, 21,22,23,24,25,26,27, 29,30,31,32,33, 37, 39] including a total of 23,015 cancer cases and 3,974,960 controls, and the remaining six were cohort studies [20, 28, 34,35,36, 38], with sample sizes ranging from 8839 to 1,226,515. Most included studies (n = 15) focused on a single cancer site [18, 19, 21, 23,24,25,26,27, 29, 30, 32, 33, 35,36,37], and seven studies assessed the overall childhood cancer risk in offspring [20, 22, 28, 31, 34, 38, 39]. Three studies reported risk estimates for multiple types of cancer [31, 34, 39]. As for the exposure of interest, ten studies gave data on maternal diabetes and did not specify diabetes type [18, 19, 21,22,23, 27, 30, 32,33,34], while 11 reported the risk estimates for GDM women [20, 24,25,26, 28, 29, 31, 35, 37,38,39]. These studies were conducted in the following countries: the USA (n = 6), Sweden (n = 5), Greece (n = 2), Denmark (n = 2), Canada (n = 2), Italy (n = 1), Finland (n = 1), Israel (n = 1), Norway (n = 1) and Switzerland (n = 1).

Table 1 Main characteristics of the included studies included in the meta-analyses
Full size table

The majority of studies (n = 17) identified maternal diabetes according to the medical birth registries or medical records [18, 20, 22,23,24,25,26,27,28, 30,31,32, 34,35,36, 38, 39]. However, five studies were based on self-report methods (questionnaire or telephone interview) [19, 21, 29, 33, 37]. All the included studies ascertained cancers based on cancer registries or medical records. Twenty studies reported risk estimates calculated using multivariate statistical models to adjust for significant confounders [18,19,20,21,22,23,24,25,26,27,28,29,30,31, 33,34,35,36, 38, 39], while two studies [32, 37] only provided data that can be used to calculate ORs. Fourteen studies [19, 20, 22,23,24, 28, 30, 31, 33,34,35,36, 38, 39] were adjusted for maternal age, five [18, 22, 23, 28, 34] were adjusted for birth weight, six [19, 28, 30, 34, 35, 39] were adjusted for birth order, and four [19, 20, 29, 39] were adjusted for maternal smoking status. As shown in Supplementary Table 2, seventeen studies [19, 21, 22, 33, 37] were considered as high quality, and the other 17 were of moderate quality [18, 20, 23,24,25,26,27,28,29,30,31,32, 34,35,36, 38, 39] according to the Newcastle–Ottawa quality tool.

Maternal diabetes and childhood cancer risk

Seven studies [20, 22, 28, 31, 34, 38, 39] explored the childhood cancer risk associated with maternal diabetes when all cancer subtypes were combined, with a total of 7,545,946 participants and 34,351 cases with childhood cancer. The ascertainment of exposure and outcome was based on birth registries, cancer registry and medical records. Three studies [22, 28, 39] reported a significant positive association, two [34, 38] found a positive, though non-significant association. The other two studies [20, 31] reported the ORs for GDM and PDM, respectively. Interestingly, they both reported that PDM conferred elevated risk of cancer, but GDM did not. As shown in Fig. 2, the summary RR was 1.30 (95% CI, 1.10–1.53; P = 0.002), with substantial heterogeneity in the included studies (P < 0.001; I2 = 73.3%). Egger’s and Begg’s tests suggested no evidence of publication bias (P value for Egger: 0.525; P value for Begg: 0.917).

Fig. 2
figure 2

Pooled effect of maternal diabetes on the risk of childhood cancers in the offspring. Data are presented as OR for each study (boxes), 95% CIs (horizontal lines) and summary as RR with 95% CI (diamond). OR odds ratio, CI confidence interval, GDM gestational diabetes mellitus, PDM preexisting diabetes mellitus

Full size image

Subgroup analyses were conducted to explore the sources of the high heterogeneity in the included studies (Table 2). First, we investigated the impact of diabetes types on the estimates of relative risk. We found a significantly positive association between maternal PDM and childhood cancer risk (The pooled OR, 1.41; 95% CI, 1.17–1.70; P < 0.001; I2 = 50.8%). However, this positive association was not significant in women with GDM (The summary OR, 1.10; 95% CI, 0.94–1.28; P = 0.223; I2 = 33.2%). The summary OR was 1.26 (95% CI, 1.05–1.51; P = 0.013; I2 = 56.8%) when all diabetes subtypes were combined. Univariate meta-regression analysis showed that maternal diabetes type was one of the important sources of variability between studies (Adj R2 = 18.91%). In addition, although some variations were discovered in the subgroup analyses of study type, study location, number of cancer cases and adjustment for birth weight and birth order, all the pooled ORs were still greater than 1.

Table 2 Subgroup analysis for the association of maternal diabetes and childhood malignancy
Full size table

Childhood leukemia

Of six studies [18, 19, 31, 34, 39] reported childhood leukemia risk, two [31, 34] documented a significant positive association, while the other four studies [18, 19, 24, 39] found a positive, but non-significant association. As shown in Fig. 3a, the summary OR revealed that children of diabetic mothers have a higher risk of childhood leukemia compared to the offspring of the non-diabetic mothers (The summary OR, 1.30; 95% CI, 1.15–1.48; P < 0.001; I2 = 0.0%). The funnel plot could not rule out potential publication bias (Fig. 4). However, Egger’s tests (P = 0.072) and Begg’s tests (P = 0.386) showed no evidence of publication bias.

Fig. 3
figure 3

Pooled effect of maternal diabetes on the risk of childhood leukemia (a) and ALL (b) risk in the offspring. ALL acute lymphoblastic leukemia

Full size image
Fig. 4
figure 4

Funnel plots of studies exploring the association between maternal diabetes and the risk of childhood cancer (a), leukemia (b) and ALL (c) in the offspring. ALL acute lymphoblastic leukemia

Full size image

Of seven studies [23, 26, 31, 34,35,36, 39] reported childhood acute lymphoblastic leukemia (ALL) risk, three [31, 34, 35] reported a significant positive association, and three studies [23, 26, 39] found a positive but non-significant association. A study documented that the increased incidence of childhood ALL in the offspring was linked to maternal PDM, but not GDM. The summary OR was 1.44 (95% CI, 1.27–1.64; P < 0.001; I2 = 0.0%), indicating that maternal diabetes was correlated with a significantly increased risk of ALL in the offspring (Fig. 3b). What’s more, both PDM (The summary OR, 1.69; 95% CI, 1.31–2.17; P < 0.001; I2 = 27.0%) and GDM (the summary OR, 1.32; 95% CI, 1.06–1.63; P = 0.013; I2 = 0.0%) were linked to a statistically significant increased risk of childhood ALL in the offspring. The results of sensitivity analysis were presented in Supplementary Table 3, indicating a robust association. There was no evidence of publication bias (P value for Egger’s test: 0.285; P value for Bgge’s test: 0.348) (Fig. 4c).

Other types of cancer

We also explored the risk of other common cancers in children of diabetic mothers. Three case–control studies explored the risk of neuroblastoma associated with maternal diabetes and did not report significant associations (Supplementary Fig. 3). However, the summary OR for women with GDM was 1.53 (95% CI, 1.00–2.35; P = 0.051) with no evidence of heterogeneity (P = 0.439; I2 = 0.0%). As to the brain cancers, 2 ORs were synthesized for maternal diabetes (any types) and yielded a summary OR of 0.59 (95% CI, 0.39–0.91), suggesting an inverse association (Supplementary Fig. 2). However, this inverse association was not evident in women diagnosed with PDM (OR, 0.69; 95% CI, 0.37–1.29) and GDM (OR, 0.73; 95% CI, 0.37–1.46). In addition, we found that maternal diabetes was not associated with the risk of lymphomas (OR, 1.29; 95% CI, 0.94–1.78) (Supplementary Fig. 1) and retinoblastoma (OR, 1.32; 95% CI, 0.71–2.47) (Supplementary Fig. 4) in the offspring. Funnel plots showed no evidence of publication bias (Supplementary Fig. 5). In addition, the risk estimates for some other cancer types that explored in insufficient studies are presented in Table 3.

Table 3 Cancer risk associated with maternal diabetes for cancer types with limited numbers of studies
Full size table

Discussion

The aim of this study was to determine whether maternal diabetes increases the risk of childhood cancer in the offspring. Based on the 18 observational studies, we found that children of mothers with PDM had a higher risk of developing childhood cancer compared to the offspring of the non-diabetic mothers. Surprisingly, the increased risk of childhood cancer was not significant in children born to GDM women, although the pooled OR was greater than 1. For site-specific cancer, both PDM and GDM conferred an elevated risk of leukemia in the offspring, especially for ALL. However, maternal diabetes was not associated with an increased risk of lymphoma and retinoblastoma in offspring.

We found that maternal diabetes was correlated with a 30% increased risk of childhood cancer in the offspring when all cancer subtypes were combined. Intriguingly, in the subgroup analyses of maternal diabetes type, the results suggested that mothers with PDM but not GDM conferred an elevated incidence childhood cancer in the offspring. This difference may be attributed to the fact that, in patients with GDM, the intrauterine environment of hyperglycemia only occurs in the later stages of pregnancy, whereas in patients with PDM, it is already present early in pregnancy. Thus, GDM showed a weaker effect on the development of childhood cancer in the offspring. Besides, metformin use in pregnancy is increasing worldwide, which may reduce the risk of pediatric cancer in the offspring to some extent. A case–control study conducted by Seppälä et al. has assessed the role of diabetes medication and found a possible risk-reducing impact on the risk of childhood cancer [39]. Furthermore, among the different hypoglycemic drugs, metformin has the most significant effect. Additionally, the summary RR for cohort studies was not statistically significant. However, the result might be due to the insufficient power. There were only three cohort studies with 173 cases of childhood malignancy on the association between maternal diabetes and childhood cancer in the offspring, which may reduce their statistical power.

Our study also reported that children born to women diagnosed with diabetes had a 29% increased risk of childhood leukemia and a 45% increased risk of ALL as compared with children born to non-diabetic mothers. Leukemia is the most common pediatric cancer worldwide, accounting for about 30% of all cases [1, 40], and ALL represents up to 80% of all pediatric leukemia [41]. Although approximately 90% are ultimately cured of their disease in many developed countries [42], ALL remains a significant cause of childhood morbidity and mortality. The positive association between birth weight and the risk for childhood leukemia, especially for ALL, has been declared in multiple epidemiological studies [43,44,45]. Besides, numerous studies have supported a link of high birth weight with maternal diabetes. Birth weight may thus mediate the association between maternal diabetes and childhood ALL in the offspring. Additionally, hyperglycemia and hyperinsulinism may also play an influential role in the development of childhood leukemia. First, maternal hyperglycemia could increase fetal growth, affect some epigenetic modifications in the offspring [46, 47], which may play an important role in offspring’s development of ALL. Second, hyperinsulinism could strongly activate the insulin receptor (IR) signaling pathways and increase the production of insulin-like growth factor (IGF)-1, which may cause proliferation of progenitor or preleukemic cells and lead to childhood leukemia [48, 49].

We also reported that women with GDM conferred an elevated risk of neuroblastoma in the offspring. Of note, only two studies on the association between GDM and the risk of neuroblastoma in the offspring were included. The inverse association between maternal diabetes and brain cancers has the same limitations. Therefore, due to the limited data, these findings need further studies to identify. In addition, our study failed to find any significant associations between maternal diabetes and the risk of lymphomas, retinoblastoma in the offspring. However, considering that the pooled OR for lymphoma was significantly greater than 1, further studies of the association between childhood lymphoma and maternal diabetes appear to be warranted. Retinoblastoma is related to the loss or mutation of both alleles of the RB1 tumor suppressor gene [50] and may be unaffected by diabetes.

Our study has two major strengths. First, this is the first meta-analysis to assess the association between maternal diabetes and the risk of childhood cancer in the offspring. Second, we distinguished between PDM and GDM and explored their effects on the risk of childhood malignancy in offspring separately. However, several limitations in our study should be recognized. First, the included studies have different adjusted models, and some did not take into account potential confounders such as hypertensive disorders, birth weight, other related environmental exposures and lifestyle factors associated with childhood cancer. There is an obvious need to study the role of these confounding factors. Second, whether the use of diabetes medication modifies the association between maternal diabetes and childhood cancer risk in the offspring is not clear and need more research to identify. Third, several studies included a limited number of childhood cancer cases born to women with diabetes, thus may reduce the precision of risk estimates. Fourth, most of the included studies were case–control studies, so recall bias was unavoidable. Finally, for certain types of childhood cancer, there were limited data on their risk associated with maternal diabetes.

Conclusion

Our meta-analysis provides evidence that maternal diabetes is associated with an increased risk of childhood cancer in offspring, particularly for ALL. Future studies should explore the role of potential confounders in the association between maternal diabetes and childhood cancer risk in the offspring, such as hypertensive disorders, diabetes medication, birth weight, other related environmental exposures and lifestyle factors.