Birth characteristics and adult cancer incidence: Swedish cohort of over 11,000 men and women

Authors


Abstract

Associations between larger size at birth and increased rates of adult cancer have been proposed but few empirical studies have examined this hypothesis. We investigated overall and site-specific cancer incidence in relation to birth characteristics in a Swedish population-based cohort of 11,166 singletons born in 1915–1929 for whom we have detailed obstetric data and who were alive in 1960. A total of 2,685 first primary cancers were registered during follow-up from 1960 to 2001. A standard deviation (SD) increase in birth weight for gestational age (GA) was associated with (sex-adjusted) increases of 13% (95% CI = 0.03–0.23) in the rates of digestive cancers and of 17% (95% CI = 0.01–0.35) in the rates of lymphatic cancers. Women who had higher birth weights also had increased rates of breast cancer under age 50 years (by 39% per SD increase; 95% CI = 0.09–0.79), but reduced rates (by 24%; 95% CI = 0.07–0.38) of endometrial (corpus uteri) cancer at all ages. There was no evidence of associations with other cancer sites. For overall cancer incidence, men had an 8% increased risk at all ages per SD increase in birth weight for GA while women only had an increased risk under age 50 years (mainly driven by the association with breast cancer). These findings provide evidence of a modest association of birth size and adult cancer risk, resulting from positive associations with a few cancer sites and a possible inverse association with endometrial cancer. © 2005 Wiley-Liss, Inc.

Over the past 15 years, much research has been conducted to investigate in utero exposures in relation to the risk of certain noncommunicable diseases in adult life.1 Smaller birth size (as a marker of the fetal environment) has been found to be associated with increased risks of ischemic heart disease2, 3 and type 2 diabetes.4

Prenatal influences on cancer risk have also been postulated,5, 6, 7, 8 whereby greater birth size is associated with higher risk. However, only 2 studies to our knowledge have investigated associations between birth weight and overall adult cancer risk; one was in women only9 and the other was in young adults (<40 years).10 Most previous studies have focused on specific cancer sites. In a (Medline) literature search, we found 17, 9 and 5 studies on female breast, testicular and prostate cancer, respectively. Knowledge of whether birth weight-cancer associations are site-specific or are similar across sites would provide evidence with which hypotheses relating to underlying biologic mechanisms can be evaluated. Additionally, prenatal influences on cancer risk need to be compared to those on ischemic heart disease and type 2 diabetes in an attempt to examine the net effect on adult morbidity.

In this article, we investigate the relationship of birth characteristics with site-specific and overall cancer incidence in a cohort of over 14,000 men and women in Sweden who have been followed up for 40 years during adult life. This cohort has previously been used to investigate breast cancer incidence11 as well as circulatory morbidity12 and mortality2 in relation to birth characteristics.

Material and methods

The Uppsala Birth Cohort Study is a prospective longitudinal study of all deliveries at the Uppsala Academic Hospital in Sweden during 1915–1929.2 Detailed obstetric notes were kept, providing data on maternal age, birth order, birth weight and placental weight (recorded to the nearest 10 g) and birth length and head circumference (recorded to the nearest 0.5 cm). Ponderal index, a measure of weight relative to height in infants, was calculated as birth weight/length3 (kg/m3). Gestational age was calculated as the number of days between delivery and the mother's recall of last menses.

Follow-up of the cohort was initially made through parish records where deaths and changes of residence were recorded. Personal identity numbers introduced in the 1950s were subsequently used to link the cohort to Swedish Registers of Deaths and Migration, decennial National Censuses and, from 1958, the Swedish Cancer Registry. WHO International Classification of Diseases, revision 7, was used to (back)code cancer sites (ICD7: 140–207). Linkage to these registers enabled calculation of cancer incidence rates. Data recorded at birth (mother's marital status and maternal social class) were considered as potential confounders. Data on adult socioeconomic confounders were obtained from the 1960 census (marital status, household car ownership, educational level and occupation).

Statistical methods

As markers of the rate of fetal growth, birth size measures (birth weight, length and head circumference) for gestational age (GA) were calculated as z-score=(observed value −mean)/standard deviation, where the mean and standard deviation (SD) of each birth size measure are gestational week-specific. One z-score corresponds to 450 g for birth weight, 2.0 cm for length and 1.4 cm for head circumference at 40 weeks of gestation. Z-scores and gestational age were included as continuous variables in all models; departures from linearity were checked using a Wald test of the quadratic term and categories of the explanatory variable were utilized when this test provided evidence, at the 5% significance level, of nonlinearity. Associations with categories of absolute birth weight (<3,000, 3,000–3,499, 3,500–3,999, ≥4,000 g) were also examined for all sites and are presented for selected sites.

Proportional hazards models were used to estimate hazard ratios for cancer (overall and site-specific) by birth characteristics. Individuals' follow-up times were calculated on an age time scale from 1 November 1960 (the census date) to the earliest of date of emigration, date of first primary cancer diagnosis (regardless of site), date of death, or 31 December 2001 (when complete registration data were available). For site-specific analyses, a subject was considered to have an event if his or her first primary cancer was of that site. We stratified by year of birth (1915–1919, 1920–1924, 1925–1929) to account for cohort effects. Robust standard errors were implemented to account for possible correlations between siblings. If there was a suggestion that the proportional hazards assumption was violated (from Nelson-Aalen plots), then follow-up time was split at an appropriate age and a test of interaction performed. In analyses of nonsex-specific cancers, we adjusted for sex if sex differences in cancer rates were constant across all ages and if birth size-cancer associations were not found to differ between men and women (tested using a Wald test of the interaction term).

Results

Of the 14,610 subjects born in the Uppsala Academic Hospital during 1915–1929, 13,749 were live singleton births. Of these, 1,660 died and 93 emigrated before the start of follow-up on 1 November 1960. Personal identity numbers were traced for 11,630 members (97%) of the remaining cohort, 11,529 of whom were linked to the 1960 census. We further excluded people previously diagnosed with cancer (n = 11), without birth weight (n = 12) or gestational age (n = 314) data, or with an unreliable gestational age of under 30 weeks (n = 26), thus leaving 11,166 people (5,820 men and 5,346 women) eligible for inclusion in analyses.

Birth characteristics and basic follow-up information of the eligible study population are given in Table I and the number of incident first primary cancers by site (with ICD7 codes) in Table II. The median age at the start of follow-up was 37 years (range, 30–45 years). A total of 2,685 (24%) individuals were diagnosed with cancer during the potential maximum follow-up of 41 years (Tables I and II). Reproductive-related cancers (specified in Table II) accounted for almost half of cancer incidence in women and 30% in men.

Table I. Birth Characteristics and Follow-Up Data of Study Sample
Birth characteristics% missingCategory/unitsMen (n = 5,820)Women (n = 5,346)
MeanSDMeanSD
  • Singleton men and women alive and linked to the 1960 census who had no prior cancer diagnosis and for whom data on birth weight and gestational age (of at least 30 weeks) were available.

  • 1

    Without prior cancer diagnosis.

Birth weight0g35275023397498
Birth length0.3cm51.22.250.52.2
Head circumference2.7cm35.11.534.41.4
Placental weight0.6g667.1138.5658.0143.1
Ponderal index0.3kg/m326.12.626.42.7
Gestational age0weeks39.92.040.02.0
   n%n%
Year of birth01915–19191,49725.71,39526.1
  1920–19242,00634.51,80633.8
  1925–19292,31739.82,14540.1
Birth order0.112,32840.02,14140.1
  21,33723.01,27423.9
  374112.871313.4
  ≥41,40824.21,21322.7
% unmarried mothers0.2%1,13719.61,09920.6
Maternal age (years)0.04<251,86432.01,77633.2
  25–291,67128.71,45227.2
  30–841,14819.71,11120.8
  ≥351,13519.51,00418.8
Follow-up data  n (%)Median age (years)n (%)Median age (years)
Start of follow-up (1 November 1960)  5,820 (100)37.35,346 (100)37.2
Main event/censoring reason      
 Cancer diagnosis  1,419 (24.4)68.41,266 (23.7)65.7
 Died1  1,827 (31.4)68.8963 (18.0)72.2
 Emigrated1  67 (1.2)50.837 (0.7)48.8
 Censored on 31 December 2001  2,507 (43.1)76.73,080 (57.6)77.3
Table II. Number of Incident First Primary Cancers by Site and Sex among 11,166 People During 385,000 Person Years
Malignant neoplasmICD7 codesMenWomenAll
  • 1

    61% colon, 39% rectum.

  • 2

    90% lung cancer.

  • 3

    36% kidney, 64% bladder cancer.

  • 4

    42% non-Hodgkin lymphoma, 20% lymphatic leukemia, 18% myeloma plasmocytoma.

  • 5

    28% thyroid, 72% cancer of endocrine glands.

Nonreproductive-related cancers140–207, excl 170–99966371,633
 Digestive organs and peritoneum of which150–159328214542
  Stomach151562783
  Colorectal1153–154155104259
  Liver and biliary passage155–156373269
  Pancreas157543791
 Respiratory system2160–16519889287
 Urinary system3180, 18115460214
 Lymphatic and hematopoietic tissues4200–20911080190
 Brain and nervous system193414182
 Skin melanoma190443377
 Endocrine glands5194, 195195574
 Buccal cavity and pharynx140–148321143
Reproductive-related cancers    
 Women    
  Breast170 367 
  Corpus uteri (endometrium)172 112 
  Ovary175 89 
  Cervix171 45 
 Men    
  Prostate177405  
  Testis1788  
All cancers140–2071,4191,2662,685

Nonreproductive-related cancers

Birth weight for GA was positively associated with 2 of the 7 nonreproductive cancer groups analyzed (Table III) and there was no evidence of an association with the other 5 groups. One SD increase in birth weight for GA was associated with a 13% (95% CI = 3–23) increase in rates of digestive cancers and a 17% (95% CI = 1–35) increase in rates of lymphatic/hematopoietic cancers. The increased risk for digestive cancers was mainly due to increases in the risk of stomach (19%; 95% CI = −1–43), colorectal (15%; 95% CI = 1–31) and pancreatic cancers (18%; 95% CI = −3–44). Birth length and particularly head circumference had results in similar directions to those for birth weight, but no associations with ponderal index were statistically significant at the 5% level (Table V). Larger birth size was also associated with an increase, although of smaller magnitude, in rates of all nonreproductive cancers combined (Tables III, IV and V). Comparing the extremes of the birth weight distribution, rates of all nonreproductive cancers combined and of cancers of the digestive organs and lymphatic/hematopoietic tissues were 38% (95% CI = 13–68), 43% (95% CI = 1–102) and 145% (95% CI = 24–383) higher, respectively, in adults who had a birth weight of at least 4,000 g compared to those whose birth weight was less than 3,000 g (Table IV).

Table III. Hazard Ratios and 95% Confidence Intervals for Overall and Site-Specific Cancer Incidence Associated with One Standard Deviation Increase in Birth Weight for Gestational Age
Malignant neoplasm1Adjusted for sexAdjusted for sex, birth order and socioeconomic factors at birth2 and during adulthood3
HR95% CIpHR95% CIp
  • 1

    ICD codes given in Table II. All adult ages, and both sexes unless specified otherwise.

  • 2

    Mother's marital status (married, divorced, single, widowed) and maternal social class.

  • 3

    Socioeconomic indicators in 1960; marital status (married and living together, married and living apart, single, widowed, divorced), household car ownership (none, at least 1 car), educational attainment (nonacademic, A-levels, tertiary) and occupational status (nonmanual, manual, self-employed, unemployed).

Nonreproductive-related cancers1.081.03–1.140.0021.091.03–1.150.001
 Digestive organs and peritoneum1.131.03–1.230.0071.131.03–1.230.007
  Stomach1.190.99–1.430.071.130.93–1.380.23
  Colorectal1.151.01–1.310.041.161.02–1.330.02
  Liver and biliary passage0.920.72–1.180.510.960.74–1.250.79
  Pancreas1.180.97–1.440.091.180.97–1.430.10
 Respiratory system1.060.93–1.200.371.070.93–1.220.35
 Urinary system1.020.88–1.180.801.050.90–1.220.51
 Lymphatic and hematopoietic tissues1.171.01–1.350.031.191.02–1.390.02
 Brain and nervous system1.020.81–1.270.881.050.84–1.330.66
 Skin melanoma1.030.85–1.260.751.010.82–1.260.89
 Endocrine glands0.850.63–1.130.250.880.64–1.200.43
Reproductive-related cancers      
 Women      
  Breast      
   <50 years1.391.09, 1.790.0091.401.08, 1.810.012
   ≥50 years1.000.89, 1.120.981.000.88, 1.130.99
  Corpus uteri (endometrium)0.760.62, 0.930.0080.790.64, 0.970.02
  Ovary1.060.83, 1.340.651.020.79, 1.310.89
 Men      
  Prostate1.010.91, 1.110.900.980.89, 1.090.72
All cancers      
 Men, all ages1.081.02, 1.140.0061.081.02, 1.140.005
 Women <50 years1.241.08, 1.430.0031.231.06, 1.440.007
 Women ≥50 years0.970.91, 1.040.360.980.91, 1.050.52
Table IV. Hazard Ratios and 95% Confidence Intervals for Overall and Selected Site-Specific Cancer Incidence Associated with Categories of Birth Weight, Adjusted for Sex, Gestational Age and Socioeconomic Factors at Birth and in Adulthood
Birth weight (g)Malignant neoplasm1
HR95% CIHR95% CIHR95% CI
  • 1

    ICD codes given in Table II. Birth and adult socioeconomic factors are listed in footnotes 2 and 3 of Table III.

 Nonreproductive-related cancers, men and womenDigestive organs, men and womenLymphatic/hematopoietic tissues, men and women
<3,0001 1 1 
3,000–3,4991.191.01–1.411.200.90–1.612.061.17–3.62
3,500–3,9991.311.10–1.551.431.07–1.922.261.26–4.07
≥4,0001.381.13–1.681.431.01–2.022.451.24–4.83
 Female breast, <50 yearsFemale breast, ≥50 yearsEndomentrium (corpus uterus)
<3,0001 1 1 
3,000–3,4991.810.77–4.260.860.62–1.190.730.45–1.19
3,500–3,9992.661.09–6.461.060.75–1.490.610.36–1.05
≥4,0004.001.49–10.720.910.57–1.460.550.36–1.17
 All cancers, menAll cancers, women <50 yearsAll cancers, women ≥50 years
<3,0001 1 1 
3,000–3,4991.341.10–1.631.901.17–3.080.920.77–1.09
3,500–3,9991.411.16–1.732.001.20–3.340.960.80–1.15
≥4,0001.481.18–1.852.731.49–4.990.960.75–1.22
Table V. Hazard Ratios and their 95% Confidence Intervals for Overall and Site-Specific Cancer Incidence Associated with Birth Length and Head Circumference for Gestational Age and Ponderal Index
Malignant neoplasm1Effect of standard deviation increase in
Birth length for gestational ageHead circumference for gestational agePonderal index
HR95% CIHR95% CIHR95% CI
  • 1

    ICD codes given in Table II. All adult ages and both sexes unless specified otherwise.

  • 2

    p <0.05.

  • 3

    p <0.01.

Nonreproductive-related cancers1.061.01–1.1221.081.02–1.1431.040.99–1.09
 Digestive organs and peritoneum1.080.99–1.191.121.01–1.2421.060.98–1.15
  Stomach1.250.96–1.611.281.04–1.5821.010.83–1.23
  Colorectal1.040.91–1.191.140.97–1.341.131.00–1.27
  Liver and biliary passage1.010.76–1.341.140.91–1.440.870.69–1.09
  Pancreas1.090.91–1.311.080.87–1.341.120.95–1.32
 Respiratory system1.060.94–1.200.910.81–1.031.010.90–1.15
 Urinary system0.980.85–1.131.090.95–1.251.060.94–1.19
 Lymphatic and hematopoietic tissues1.110.96–1.281.140.99–1.321.100.96–1.26
 Brain and nervous system1.070.83–1.381.050.81–1.361.030.72–1.45
 Skin melanoma1.000.80–1.261.130.95–1.341.020.82–1.28
 Endocrine glands1.000.76–1.310.970.74–1.260.820.65–1.03
Reproductive-related cancers      
 Women      
  Breast      
   <50 years1.411.07–1.8621.381.20–1.5931.060.79–1.42
   ≥50 years1.050.93–1.200.990.88–1.110.930.84–1.04
  Corpus uteri (endometrium)0.840.69–1.020.900.75–1.080.840.70–1.00
  Ovary1.140.89–1.450.870.71–1.060.890.73–1.07
 Men      
  Prostate0.950.86–1.041.030.94–1.141.090.99–1.19
All cancers      
 Men, all ages1.030.98–1.091.091.03–1.1531.071.02–1.132
 Women <50 years1.261.08–1.4831.241.10–1.4031.040.88–1.22
 Women ≥50 years1.000.93–1.070.980.92–1.040.970.91–1.03

There was no evidence of an association of maternal age with any of the cancer sites (not shown), with the exception of thyroid and other endocrine cancers, where a 5-year increase in maternal age was associated with a 17% reduction (95% CI = 1–30; p = 0.05) in rates. Adjustment for maternal age did not, however, alter the findings for birth weight for GA. Birth order was not associated with cancer incidence for any of the sites (results not shown), with the exception of stomach cancer. For stomach cancer, a birth order of 4 or more was associated with increased incidence (hazard ratio of 2.03 compared to being first born; 95% CI = 1.20–3.44) and adjustment for birth order also attenuated the association with birth weight (Table III). For all other cancer sites, adjusting for birth order and socioeconomic measures at birth and in adulthood did not alter the findings (Table III).

Longer gestation, independent of birth weight for GA, was associated with a small increase (3% per week; p = 0.03) in rates of all nonreproductive cancers combined (sex-adjusted) and in cancers of the urinary system (Table VI). Nonlinear associations with GA were observed for lymphatic cancers, whereby both premature and late births were associated with reduced risks.

Table VI. Hazard Ratios and their 95% Confidence Intervals for Overall and Site-Specific Cancer Incidence Associated with 1-Week Increase in Gestation, Adjusted for Sex and Birth-Weight for Gestational Age
Malignant neoplasm1HR95% CIp
  • 1

    ICD codes given in Table II.

  • 2

    Evidence of nonlinear trend. HRs (95% CIs) compared to 40 weeks of gestation are: <37 weeks: 0.68 (0.36–1.28); 37–38 weeks: 0.77 (0.50–1.19); 39 weeks: 0.99 (0.37–1.45); ≥41 weeks: 0.68 (0.45–1.01).

  • 3

    Evidence of non-linear trend. HRs (95% CIs) compared to 40 weeks of gestation are: <37 weeks: 0.88 (0.70–1.10); 37–38 weeks: 0.87 (0.74–1.01); 39 weeks: 0.90 (0.78–1.04); ≥41 weeks: 0.84 (0.72–0.97).

Nonreproductive-related cancers1.031.00–1.050.03
 Digestive organs and peritoneum0.990.95–1.030.58
 Respiratory system1.051.00–1.110.07
 Urinary system1.091.02–1.170.01
 Lymphatic and hematopoietic tissues0.9920.93–1.050.64
 Brain and nervous system1.070.95–1.210.26
 Skin melanoma0.980.88–1.080.63
 Endocrine glands1.050.94–1.180.35
Reproductive-related cancers   
 Women   
  Breast   
   <50 years0.940.83–1.070.37
   ≥50 years0.990.93–1.040.59
  Corpus uteri (endometrium)0.940.85–1.040.23
  Ovary0.950.85–1.060.35
 Men   
  Prostate1.020.98–1.070.30
All cancers   
 Men1.0130.98–1.030.61
 Women1.020.99–1.050.11

Male reproductive-related cancers

Only prostate cancer could be analyzed separately as there were too few cases of other male cancers. There was no evidence of an association between birth weight, length, or head circumference and prostate cancer rates, but weak evidence of increased rates with higher ponderal index (p = 0.08; Tables III and V). The linear trend with gestational age was not significant due to weak evidence of nonlinearity (p = 0.10). Compared to men born at term (40 weeks), those who were born prematurely (<39 weeks of gestation) or postterm (≥41 weeks) had lower rates of this cancer (hazard ratios of 0.74, 95% CI = 0.56–0.97, and 0.70, 95% CI = 0.53–0.91, respectively).

Female reproductive-related cancers

We found positive associations between birth weight, birth length and head circumference for GA with breast cancer under age 50 years, where there were 63 breast cancers (39% increase per SD increase in birth weight for GA; 95% CI = 9–79), but no association over age 50 or with ponderal index at any age (Tables III and V). After adjusting for the fetal growth rate as measured by birth weight for GA, there was no association with GA (Table VI). The positive association with breast cancer under age 50 represented a 4-fold (95% CI = 1.49–10.72) difference in cancer rates between the lowest (<3,000 g) and highest (≥4,000 g) categories of birth weight (Table IV).

Exceptionally, women with greater birth size had lower rates of endometrial (corpus uteri) cancer. Rates of this cancer in women who weighed at least 4,000 g were almost half that of women who weighed under 3,000 g (HR = 0.55; 95% CI = 0.36–1.17; Table IV), with similar results for each measure of birth size. There was some suggestion that this result may not hold under age 50, but there was insufficient power (11 cancers) to analyze separately. Cancer of the ovary and cervix (not shown) were not associated with birth characteristics.

Overall cancer risk

The effect of birth size on overall cancer incidence varies between men and women due to the influence of strong associations with some female reproductive cancers. In men, one SD increase in birth weight was associated with an 8% increase in overall cancer incidence rates at all adult ages, with associations of very similar magnitude for head circumference and ponderal index. Expressed in terms of absolute birth weight, the GA-adjusted hazard ratio in men associated with 1,000 g increase in birth weight was 1.17 (95% CI = 1.05–1.31). There was also evidence of reduced cancer rates in men of short and long gestation (nonlinear association).

In women, birth size-cancer associations differed by age. Under age 50 years, with a total of 179 cancers in women, there was an increase of 24% (95% CI = 8–48) in cancer rates with one SD increase in birth weight for GA, driven by the positive association with breast cancer. Over age 50, there was no overall effect in women as positive associations with nonreproductive cancers were counterbalanced by inverse associations with endometrial cancer.

Adjustment for socioeconomic indicators at birth and in adulthood did not affect these findings. We also carried out separate analyses of smoking- (buccal cavity, pharynx, pancreas, respiratory and urinary cancers) and nonsmoking-related cancers in order to address, indirectly, the possibility of confounding by smoking status. Among men, where the association with birth weight did not differ by age, we found similar associations for smoking- (hazard ratio 1.07; 95% CI = 0.98–1.18) and nonsmoking-related cancers (1.08; 95% CI = 1.01–1.15).

Discussion

Comparisons with other studies

Several studies have reported higher breast cancer risk in young women who had greater birth size,13, 14 but in one a J-shaped association was observed.15 The lack of association with prostate cancer was also found in 3 other studies,16, 17, 18 but not in a smaller study (21 cases) where 5-fold differences in rates were found.19 For prostate cancer, 2 different studies have found lower risk with longer17 and shorter16 gestation, but neither specifically reported a U-shaped association, of which there was weak evidence in our study. The (linear) raised risk of digestive cancers and of its constituent sites with larger birth size found here was consistent with findings from a retrospective cohort of women,9 but J-shaped associations have been reported for colorectal cancer in the EPIC Norfolk study.20 Our findings for lymphatic cancers are similar to those found in children, e.g., for acute myeloid leukemia.21

Women who were small at birth were at an increased risk of endometrial cancer in this study. Without any directly comparable studies with independent consistent results, there remains the possibility that this may be a chance finding. Alternatively, the association may be explained by prenatal associations with risk factors for this cancer. Low birth weight is associated with increased risks of obesity and insulin resistance syndrome,22 which are in turn associated with polycystic ovarian syndrome, all of which are well-established risk factors for endometrial cancer.23

In the absence of many comparable studies of birth size, comparisons can instead be made with studies examining adult height, as birth weight and length for GA are both good predictors of final adult height.24 Our findings would be in agreement with other studies that have reported increased risks in taller adults of pancreatic,25 colorectal,26, 27 breast28 and overall29, 30 cancer risk. For example, the association of overall cancer with birth weight in men found here (hazard ratio per SD increase in birth weight for GA = 1.08; 95% CI = 1.02–1.14) is very similar to that found for adult height in a male cohort in Wales29 (hazard ratio per SD increase in adult height = 1.09; 95% CI = 0.97–1.21) and in the U.S. Physicians' Health Study.30

Strengths and weaknesses

This study is unique in assessing possible fetal origins of adult cancer due to the availability of prospective data on various measures of birth size and to the possibility of analyzing, with sufficient power, cancer sites separately. However, this has led to multiple tests of statistical significance being performed, so the overall false positive rate is larger than 5%. Future studies will reveal whether some of our results are chance findings. We were not able to investigate associations with cancers that occur in young adults (e.g., testicular cancer). Our findings are unlikely to be influenced by selection bias as this is an inclusive population-based cohort with very few losses to follow-up. Cancer registration is also likely to be very complete.31

We attempted to adjust for potential confounding by adjusting for indicators of socioeconomic position at birth and in adulthood, but there remains the potential for residual confounding. Our results for breast and colorectal cancers may be positively confounded as the risks of these cancers increase with higher socioeconomic position32 (e.g., due to correlations with reproductive-related factors for breast cancer). Such confounding would not, however, explain the increased risk of stomach or liver cancer as incidence of these cancers is lower in higher socioeconomic groups,32 nor would it explain that for lymphatic or endometrial cancers, which do not have strong socioeconomic gradients in risk. We were not able to adjust for smoking, but the similarity of results for smoking- and nonsmoking-related cancers suggests that confounding by smoking status may not be a large problem.

Biologic plausibility

Birth size for GA is unlikely to be a risk factor for adult disease in itself, but may be a marker for some aspects of the fetal environment that are related to risk. Fetal growth retardation may be a consequence of impaired fetal nutrition. It has been suggested that larger birth size may indicate a greater number of cells at risk of carcinogenesis, e.g., for breast cancer or kidney cancer. Programming of the insulin-like growth factor (IGF) system in utero may lead to increased postnatal cell proliferation rates,33 and IGF-1 has been linked to an increased risk of breast34 and other epithelial cancers.35 We were not able to investigate other features of the growth trajectory, such as postnatal catch-up growth. Evidence from the 1946 British birth cohort, however, seems to indicate that at least the relationship of birth weight with premenopausal breast cancer is not mediated through childhood growth.36 Other prospective longitudinal studies with repeated postnatal growth measurements will be valuable in investigating cancer in relation to other growth features. It has been hypothesized that maternal age may be related to cancer risk due to poorer oocyte quality in older mothers as a result of greater risk of genetic changes with age.37 However, with the exception of endocrine cancers, we found no evidence of an association between maternal age and cancer incidence.

We have found some evidence supporting the hypothesis that larger birth size is associated with increased risk of certain adult cancers. However, our findings suggest that positive associations were not uniform across all cancer sites, but were particular to just a few sites (digestive and lymphatic at all ages and in women, breast cancer under age 50). Furthermore, our findings generate the hypothesis, which merits further study, that rates of endometrial cancer are lower in women who had higher birth weights. For all cancer sites combined, men who had larger birth size had a modest increase in rates at all ages and women only had an increase under age 50 because at older ages positive associations with some cancer sites were counterbalanced by the strong inverse association with endometrial cancer.

The association of larger birth size with increased cancer incidence in men found here is in the opposite direction to that found with circulatory diseases in this cohort2 and in other studies.3 Furthermore, the magnitude of the overall increase in cancer incidence in men observed here is approximately half the magnitude of the decrease in ischemic heart disease in the same cohort2 (hazard ratio of 1.17 for cancer vs. 0.77 for IHD for 1,000 g increase in birth weight). These findings suggest that both circulatory disease and cancer need to be considered in determining the net effect of prenatal influences on adult morbidity and mortality.

Acknowledgements

The authors thank Rawya Mohsen for data management and linkage. The Uppsala Birth Cohort Study was established and later supported by grants provided by the U.K. Medical Research Council (grant 9322050), the Swedish Council for Working Life and Social Research (grants 94/0157, 2003-0101 and 2004-1439) and the Swedish Research Council (grants 5446 and 345-2003-2440). I.K. is currently funded by the Swedish Council for Working Life and Social Research.

Ancillary