It is well established that prenatal biologic processes are important for the development of some childhood cancers, whereas less is known regarding their influence on adult cancer risk. High birth weight has been associated with risk of breast cancer, whereas studies of other specific cancers and all cancers together have been less conclusive.
The authors established a cohort of more than 200,000 men and women who were born between 1936 and 1975. Birth weights were obtained from school health records and information concerning cancer from the Danish Cancer Registry. Follow‒up was performed between April 1, 1968 and December 31, 2003. During 6,975,553 person-years of follow‒up, a total of 12,540 primary invasive cancers were diagnosed.
Analyses of site‒specific cancers revealed that the majority of cancers had a positive linear association with birth weight. Departures from a positive linear association were found to be statistically significant for cancers of the pancreas and bladder, which demonstrated a V‒shaped association, and testicular cancer, which demonstrated an inverse association with birth weight. Excluding these 3 exceptions, the trends for the individual cancer sites were not heterogeneous, and the overall trend was a relative risk of 1.07 (95% confidence interval, 1.03–1.11) per 1000‒g increase in birth weight. This trend was the same in men and women and in all age groups.
Several studies have established that the prenatal period is significant for the later development of breast cancer.1–25 Although to our knowledge the biologic mechanisms are still poorly understood, much attention has been given to the role of estrogens and insulin growth factors. Maternal pregnancy levels of both hormones have been correlated with birth weight,26–29 which therefore has been used as a proxy variable in several epidemiologic studies.
A positive association between birth weight and the risk of breast cancer has been established in several population‒based studies, including our own.1–25 In addition, in a recent study of women from the current study population, we found that the risk of breast cancer associated with birth weight was independent of the effect of subsequent growth patterns and the timing of puberty.25
It has been hypothesized that prenatal biologic processes are important for several other cancers as well,30–32 but to our knowledge it has to date only been convincingly demonstrated for the development of childhood leukemia.33 Testing this hypothesis for adult cancers with sufficient power requires very large cohorts with long follow‒up periods. To our knowledge, only 2 studies published to date have addressed the significance of birth weight in the development of cancer in general. In their cohort study of Swedish women, Andersson et al. found a linear association with the overall incidence of cancer,7 although the study was limited in strength due to its size (262 cases with known birth weights). In a study of 11,166 men and women, McCormack et al. found birth weight was associated with cancer in men at all ages and women aged <50 years.18
The majority of studies focusing on specific cancer sites have reported a positive association between prostate cancer and birth weight,34–39 whereas studies of testicular cancer have found that low birth weight increases the risk.40–46 In a case‒control study by Bergstrom et al.,47 an increased risk of renal cell cancer was observed among men with a birth weight of >3500 g, whereas no association was found for women. Another case‒control study reported a nonlinear association between birth weight and colorectal cancer,48 with children having both low and high birth weights found to be at an increased risk, a finding that was not confirmed by Nielsen et al.49
In the current study, we explored the association between birth weight and the risk of cancer between the ages of 6 to 73 years in a large population‒based cohort of women and men.
MATERIALS AND METHODS
The study cohort consisted of 161,063 females and 164,155 males born between 1930 and 1975 who attended school in the Copenhagen municipality and for whom a school health record was kept. These records contain information concerning birth weight and annual measurements of weight and height that have been computerized. To a varying degree, the records also contain additional information regarding childhood infections, vaccinations, and general health in the preschool years (with annual updates throughout school years) that at the time of last follow‒up had not been systematically computerized. Birth weight was reported by the parent(s), typically the mother, attending the child's first visit to the school health services.
The Danish Civil Registration System (CRS) was established April 1, 1968, and all residents and newborn infants in Denmark are given a unique personal identification number (CRS number). The CRS number is stored along with information regarding the names of all Danish residents and is updated daily with respect to vital statistics and immigration status. All other national registries in Denmark, which record individual information, are based on the CRS number, thereby serving as a unique key for linkage studies. Name and birth date from the school health records was computerized and linked to the CRS by the Institute of Preventive Medicine to obtain individual CRS numbers, and was successful for 141,393 females (88%) and 145,140 males (88%). The missing identification of CRS numbers is most likely due to death or emigration before 1968.
Ascertainment of Cases
Information regarding incident cases of invasive cancer was obtained from the Danish Cancer Registry, which is considered close to complete with respect to cases of malignant diseases diagnosed in Denmark since 1943.50
The association between birth weight and the incidence of cancer was estimated in a cohort design using log-linear Poisson regression with the PROC GENMOD procedure in the SAS statistical software package (release 8.02; SAS Institute, Inc, Cary, NC).51 Follow‒up for a specific cancer began on April 1, 1968 or at age 6 years, whichever came last, and continued until a diagnosis of the specific cancer, death, emigration, or December 31, 2003, whichever came first. Adjustment was made for age (quadratic splines with knots: 35 years, 40 years, 45 years, 50 years, 55 years, and 60 years) and calendar period in 5‒year intervals.52
The main focus of the analyses was to investigate whether the risk of cancer according to birth weight could be described by a linear trend throughout the entire birth weight spectrum. The association between birth weight and cancer risk was analyzed in a 2‒step approach. First, a linear spline model with a birth weight of 3500 g as the knot was estimated to investigate the linearity of the association in birth weights being less than or ≥3500 g. Second, if these 2 trend estimates could be considered equal (based on a likelihood ratio test), an overall trend was estimated. The trends were estimated by treating birth weight categorized in intervals (501–2499 g, 2500–2999 g, 3000–3499 g, 3500–3999 g, 4000–4499 g, and 4500–5999 g) as a continuous variable. Persons with recorded birth weights that were ≥6000 g or ≤500 g were excluded from the analyses due to a high risk of misclassification in these extreme groups (N = 663). The numeric value assigned to a given category was chosen as the median of the distribution of birth weight within the category.
Testing for differences in the site‒specific trends using a competing risks model53 is not feasible due to the large dataset involved. Therefore, we used a meta‒analytic approach, in which we tested whether the site‒specific trends were significantly different using inverse variance‒weighted regression. More specifically, the test was performed in a linear regression model by means of the PROC GENMOD procedure in the SAS software package with the logarithm of the site‒specific relative risk (RR) for trend as the outcome, with tumor site as the exposure variable and the inverse of the squared site‒specific standard error for the RR as weight. The trends for different sites were further compared using inverse variance‒weighted regression with a common fixed effect and a random site effect with the PROC MIXED procedure in the SAS software package. There was no indication of a random site effect.
In the cohort of 106,504 women and 110,825 men with recorded birth weights, a total of 12,540 cases of primary invasive cancer were diagnosed during 6,975,553 person‒years of follow‒up. Table 1 shows number of cancer cases and person‒years of follow‒up by age, calendar period, and birth cohort.
Table 1. Total Number of Cancer Cases and Person-Years of Follow–up by Age, Calendar Period, and Birth Cohort
Cases (%) (Total = 12,540)
Person-years/1000 (%) (Total = 6976)
Follow-up began on April 1, 1968 and ended on December 31, 2003.
We focused on whether the risk of a specific cancer according to birth weight could be described with a linear trend. To ensure a linear fit, we used a 2‒step approach. First, trends were calculated for children with birth weights <3500 g and children with birth weights ≥3500 g (data not shown). The next step was to calculate a trend for the entire birth weight spectrum for the cancers in which the trends for the 2 halves of the birth weight spectrum were not significantly different. This was the case for the majority of cancers except for tumors of the pancreas and bladder (Table 2), and therefore when all cancers combined were analyzed these 2 cancers were not included. Furthermore, testicular cancer was not included in analyses of a common trend because previous studies have reported an inverse trend with birth weight.
Table 2 presents the association between birth weight and the incidence of site‒specific cancer (adjusted for age and calendar period). Point estimates of the RR for the different birth weight categories (501–2499 g, 2500–2999 g, 3000–3499 g, 3500–3999 g, 4000–4499 g, and 4500–5999 g) are presented without a 95% confidence interval (95% CI) to simplify the table. The different cancer sites are sorted according to the magnitude of the trends.
The majority of the cancers demonstrated a positive linear association with birth weight, with Hodgkin lymphoma and cancers of the uterus and pharynx being notable exceptions. The trends were only statistically significant for kidney and lung cancers as well as for malignant melanoma. Because the majority of cancers had a positive linear association with birth weight, we analyzed the difference in trends for all cancers (excluding cancers of the pancreas, bladder, and testis) and found that the site‒specific trends were not significantly different (P = .56) from each other. Analyses of all cancers combined demonstrated a significantly positive association between birth weight and cancer equivalent to a RR of 1.07 (95% CI, 1.03–1.11%) per 1000‒g increase in birth weight (Table 2) (Fig. 1). Inclusion of cancers that demonstrated a nonlinear association with birth weight (prostate and bladder cancers) did not appear to influence the overall trend estimate (RR of 1.07; 95% CI, 1.03–1.11%) per 1000‒g increase in birth weight.
We made separate analyses for those cancers for which smoking (according to the American Cancer Society) is a major cause (cancers of the lung, larynx, oral cavity, pharynx, esophagus, and bladder) and cancers for which smoking is not considered to be a major cause. The analyses revealed no significant difference in trends because we found a RR for smoking of 1.06 (95% CI, 0.98–1.14) and a RR for not smoking of 1.06 (95% CI, 1.02–1.10).
Further analyses were performed to ensure consistency and to rule out that the common trend in women were solely dependent on the known association with breast cancer. Analyses according to attained age demonstrated a similar birth weight trend (P = .85) for age <50 years (RR of 1.05; 95% CI, 1.01–1.11) as for age ≥50 years (RR of 1.06; 95% CI, 1.01–1.11). Analyses according to sex also demonstrated a similar birth weight trend (P = .69) in men (RR of 1.05; 95% CI, 1.00–1.11) and women (RR of 1.06; 95% CI, 1.02–1.11). In the overall trend for women, no statistically significant difference was noted when excluding breast cancer (RR of 1.07; 95% CI, 1.01–1.13). Finally, when using only primary cancers (defined as a first registered cancer) in the analyses, trend estimates were found to be similar to those reported in Table 2.
The prenatal period has been hypothesized to be a critical time window in relation to exposures associated not only with chronic nonmalignant diseases but also with cancer. The original hypothesis by Trichopoulos31 that breast cancer originates in utero has recently been refined into an integrated model for breast cancer etiology.54 The hypothesis focuses on the prenatal period and the creation of susceptible stem cells; however, to our knowledge the biologic mechanisms are still not fully understood. For the most part, the focus has been on maternal estrogens and insulin growth factor levels because the pregnancy levels of these 2 hormones appear to be correlated with birth weight, which is widely used as a proxy variable in epidemiologic studies. It is plausible that the processes responsible for affecting breast cancer risk, especially the creation of susceptible stem cells, also may affect the risk of cancer in other organs.
However, the testing of this hypothesis with sufficient power in relation to cancer at all sites implies several challenges. In addition to studies regarding breast cancer,1–25 the majority of the evidence has come from studies of a number of hormone‒related cancers (ie, prostate and testicular cancer).34–47 We used birth weight as a proxy variable to explore the possible association between intrauterine exposure and cancer in a Danish cohort of >200,000 men and women who were born between 1930 and 1975. The magnitude of the current study made it possible to perform separate analyses of several cancers not previously investigated.
We first analyzed the association between birth weight and different site‒specific cancers separately. To ensure that a deviation from a linear fit was not overlooked, we analyzed the association between birth weight and cancer in 2 steps. The first step was to examine whether the same linear trend was present in both the upper and lower end of the distribution of birth weights. Analyses of site‒specific cancers revealed that the majority of cancers have a positive linear association with birth weight (Table 2). However, nonlinear associations were found for some cancers, a few of which were significantly departing from linearity (ie, bladder and pancreatic cancers), which revealed significant V‒shaped associations with birth weight. Therefore, the next step was to calculate a common trend for each site‒specific cancer (Table 2).
Having observed that most cancers have a positive linear association with birth weight, we went on to examine whether there was an overall trend between birth weight and the risk of cancer. We fully acknowledge that cancers at different sites have different postnatal etiologies; however, this does not exclude the possibility of a common prenatal risk factor that could be noted in common associations between birth weight and the risk of cancer in general.
Cancers having a significant nonlinear fit (ie, cancers of the pancreas and bladder) were excluded in calculations of a common trend along with testicular cancer because previous studies strongly supported an inverse trend with birth weight. The association between all remaining cancer sites and birth weight was found to be similar and to fit a linear trend of a 7% increase in cancer risk per 1000‒g increase in birth weight (95% CI, 3‒11%). In additional analyses, we found the overall trend to be the same in all age groups and in both sexes. A separate analyses of all cancers in women with and without breast cancer revealed that the common trend is not only determined by the known association with breast cancer but is apparent for other cancers as well.
Birth weight was not found to be linearly associated with the risk of bladder and pancreatic cancers, but being the only 2 cancers deviating from the general trend, this finding requires confirmation. Apart from studies concerning cancers of the breast, prostate, and testis, to our knowledge few studies have been conducted on the association between cancer and birth weight. Similar to several previous reports, we found a linear association between breast cancer and birth weight.1–17 We also found an inverse association between birth weight and testicular cancer, which thereby strengthens the previous findings.40–46
To our knowledge, only 2 smaller studies published to date have previously addressed the association between the overall risk of cancer and birth weight. In a Swedish cohort study of women, a linear association was found with the overall incidence of cancer.7 However, when analyzing cancers in subgroups, inconsistent results were obtained due to very small sample sizes. In the study by McCormack et al, the association between cancer incidence and 1 standard deviation increase in birth weight was found to be different between men and women.18 Although birth weight was found to be linearly associated with cancer in men at all ages, only a linear association with cancer was noted in women age <50 years.18 However, to our knowledge none of these studies attempted to test whether the different trends could in fact be considered as a common trend.
The experience from previous studies as well as the current study therefore emphasize the necessity of having a very large study sample with which to address the relatively small effect of birth weight on cancer risk. Even in our cohort of nearly 7 million person‒years of follow‒up and 12,540 cancer outcomes, some cancers were still relatively rare due to the age distribution in the cohort. Clearly, previous inconsistencies should be explained to a large degree by small sample sizes.
The possibility that in‒utero growth has an equal effect on the risk of cancer at nearly all if not all sites is intriguing. The question remains how birth size is implicated in cancer risk. It could be explained by the existence of a common pathway with ≥1 opposing factors influencing male hormone‒related cancers. The common pathway could establish a “base risk” on which other later risk factors would have independent influence. Several exposures have been shown to be associated with size at birth, and the focus has especially been on levels of estrogen, insulin growth factor‒1,28, 29 and insulin in the mother during pregnancy, but other, still unknown factors also may be important.55, 56
It has been suggested that breast cancer risk is correlated with the number of stem cells.31, 32, 54 We believe that this model could be broadened to include most if not all types of cancer, and that, accordingly, large babies could have an increased risk of cancer due to persistently increased numbers of susceptible cells. Therefore, the association between birth weight and cancer risk could either reflect a simple correlation between birth weight and the number of stem cells or reflect that those factors that govern birth weight are also associated with an increased risk of cancer (eg, by initiating a multistep carcinogenesis).
The current study had several strengths due to its design and sample size. The Danish national health registries contain continuously updated mandatory recordings of vital status, emigration status, and cancer diagnoses, which enabled us to follow our cohort members for nearly 70 years. The social structure of the Danish healthcare system further diminished the risk for bias because equal access to healthcare is provided for all citizens. Birth weights, reported by the parents and recorded at an early age, have been found to be very accurate.57 Birth weights were furthermore recorded decades prior to and independently of a possible cancer diagnosis, making differential misclassification unlikely.
Historically, there has been little change in the distribution of birth weights in Denmark; therefore, the median birth weight remained stable between 1936 and 1983.58 This finding can be explained in part by the fact that Denmark has not been as affected by external factors as other countries; for example, during World War II, Denmark remained more or less untouched. Although the country was occupied by the Germans, the provisions were good, and the consumption of calories was reduced by only approximately 10%.59
In the current study, we were not able to adjust birth weight for gestational age or other birth size indicators. Furthermore, we were unable to adjust for the effect of smoking. Smoking exhibits a parent‒offspring association, maternal smoking reduces birth weight, and smoking is a strong risk factor for certain cancers60; therefore the inability to control for smoking limits the interpretation for smoking‒associated cancers. However, it is interesting to note that the association between birth weight and smoking‒associated cancers was found to be similar to that found for cancers not associated with smoking, implying that controlling for smoking in the mother and the offspring would expectedly strengthen the association.
The results of the current study provide evidence to suggest that prenatal biologic processes are important in the pathogenesis of all cancers. For the majority of cancers, the risk increased in a linear manner with increasing birth weight, with the only clear exception being testicular cancer, which demonstrated an inverse association with birth weight. We hypothesize that the biologic explanation behind the association noted between birth weight and cancer at different sites should be sought in a common pathway, eg, an increased cancer risk with an increased number of susceptible stem cells, with some opposing factor(s) influencing testicular cancer.
Supported by grants from the U.S. Department of Defense Congressionally Directed Medical Research Programs, the Danish Medical Research Council, the Danish National Research Foundation, the Danish Cancer Society, Augustinus Fonden, Dagmar Marshalls Fond, Fabrikant Einar Willumsens Mindelegat, Aase og Ejner Danielsens Fond, Else og Mogens Wedell-Wedellsborgs Fond, and Andersen-Isted Fonden.