1 INTRODUCTION

Preterm births account for approximately 11% of all births globally, and are considered the leading cause of childhood deaths.1 Previous work has indicated that preterm births may be associated with an increased risk of certain childhood cancers, such as hepatoblastomas, compared to being born at term.2 Preterm births may also influence cancer risk beyond childhood. There is conflicting evidence regarding whether those born preterm have increased risks of breast and testicular cancer.3-5 Increased risks of testicular cancer could be driven by altered sex hormone levels in the pre- and postnatal periods of individuals that are born prematurely.3 Exposure to oxidative stress due to immature antioxidant defense systems may also play a role.2 It may also be that preterm birth and cancer share genetic risk factors.

Cancer accounts for approximately 15% of deaths globally,6 and is a major cause of death in childhood.7 It is now also the leading cause of death for middle-aged adults in high-income countries.8 Although many etiologic factors for cancer are well recognized, the contribution of preterm birth to cancer risk is uncertain. Previous work has shown that gestational age at birth is inversely correlated with cancer-related mortality earlier in life (<30 years).9 A modest (albeit nonsignificant) increase in cancer mortality was observed for shorter gestational age categories up to the age of 50.10 Further work is required to expand upon these findings. Earlier studies of cancer risks have been hampered by small sample sizes, the categorization of gestational ages, lack of adjustment for potential confounding (including unmeasured familial confounding) and, importantly, have not considered the risk across the full range of gestational ages (including postterm births).

Our study aimed to examine the association between gestational age and risk of any primary cancer and to determine whether the risk patterns differed by sex, birth weight for gestational age, family history of cancer, age of onset, cancer site and accounting for unmeasured familial confounding.

2 METHODS 2.1 Study design

The association between gestational age and cancer risk was determined using a population-based cohort. The work is descriptive in nature and reported using the RECORD guidelines for the reporting of studies conducted using observational routinely collected health data.11

2.2 Study population

The population consisted of individuals live-born in Sweden between January 1, 1974 and December 31, 2013. To ensure greater coverage of confounding factors and to reduce the potential effects of changing immigration patterns throughout the observation period, only people with Swedish-born parents (mother and father) were included. People with missing information on gestational age, sex or with implausible death/emigration dates were excluded.

2.3 Exposure

Gestational age in weeks was extracted from the Medical Birth Register12 (MBR) where it is determined from routine ultrasound examinations (routine practice since the mid-1990s) or the date of the last menstrual period. Gestational age was primarily analyzed as weeks. In addition, it was categorized as extremely preterm (≤27 6/7 weeks gestation), very preterm (28 0/7-31 6/7 weeks gestation), moderate to late preterm (32 0/7-36 6/7 weeks gestation), term (37 0/7-41 6/7 weeks gestation) and postterm (≥42 0/7 weeks gestation).13, 14

2.4 Outcome

The primary outcome was any primary cancer reported to the Swedish Cancer Register.15 Over 96% of newly detected cancer cases in Sweden are captured within the Cancer Register and the reporting has been mandatory since 1958.16 Cancers at any site were considered and identified using ICD-7 codes17 140-209, except 156, 163 and 199, as these indicate possible secondary malignancies. Only cancers with malignant behavior were included. We also performed analysis by cancer sites classified according to ICD-7 codes (Table S1).

2.5 Covariates

Covariates potentially informative for confounding or risk modification were extracted from the MBR,12 including sex, birth weight and birth year. The MBR contains data on all births in Sweden since 1973. Birth weight for gestational age categories were calculated according to sex-specific birth weight distributions per 5-year birth brackets and categorized as small for gestational age (≤10th percentile of population), appropriate for gestational age (11-89th percentile population) and large for gestational age (≥90th percentile of population).18 Maternal smoking during pregnancy, which has been recorded since 1983 and is available for almost all births from 1990 onwards, was classified as smoker or nonsmoker.

The Multi-Generation Register19 was used to obtain information about the fathers of the study population. Through linkage using patient identification numbers, data on the parents' cancer history and highest level of education at the date of delivery were extracted. Family history of any primary, malignant cancer in either parent, occurring before or on the date of birth of the index child, was determined and classified as present or absent. Educational level was determined from the Longitudinal Integrated Database for Health Insurance (LISA),20 which contains data on the educational level of all individuals aged ≥16 years from 1990 onwards and aged ≥15 years from 2010 onwards. Highest parental education measured as years of formal education, available only for the study population born 1990 or later, was categorized as low (≤9 years), medium (10-14 years) and high (≥15 years).

2.6 Observation period

The population was followed from date of birth to the earliest occurrence of one of the following events: (a) first cancer, (b) death, (c) emigration or (d) December 31, 2016. Dates of death and emigration were gathered from the Total Population Register.19 Because the main outcome considered was any cancer, including those developing in utero and presenting at birth, cancers recorded as early as day 0 (ie, date of birth) were included. For individuals with cancer at birth to be included in the analyses, 1 day was added to the event date. Similarly, where death or emigration occurred on the date of birth, it was assumed that the neonates were alive at birth (hence registered as live-born), and 1 day was added to the censoring date.

2.7 Statistical analysis

We first calculated crude cancer rates per 100 000 person-years for any primary malignant cancer. Follow-up was calculated from birth to censoring date. Next, we examined the association between gestational age and any primary malignant cancer. Using Cox proportional hazards models with attained age as the underlying time scale,21 we calculated hazard ratios (HRs) and 95% Wald type confidence intervals (CIs) for the cancer outcome. Gestational age was analyzed as gestational age in weeks (continuous) and by categories. In addition, we determined the change in risk (slope) within each gestational age category. To avoid making any assumptions about the form of the association, natural cubic splines with four equally placed knots were applied for gestational age in weeks and continuous covariates. In a first “crude” model, we adjusted only for birth year using splines. In the second model, considered the adjusted HR (aHR), we additionally adjusted for potential confounding by maternal and paternal age at birth using splines. Unless otherwise specified, all data presented are aHR.

To better understand the relationship between cancer and gestational age, we examined the risk separately for males and females, by birth weight for gestational age category, by family history of cancer, by childhood or adolescent and young adult (AYA) cancers and by cancer site. In addition, we analyzed for a potential interaction between sex and gestational age through the inclusion of a sex-by-gestational age interaction term to the model.

When examining childhood cancers, patients were followed up from birth through 14 years of age or censoring. For AYA cancers, the observation period commenced on the 15th birthday. On examination of individual cancer sites, a person was not considered censored by the occurrence of a cancer at a different site, and each person could therefore contribute a cancer event across multiple cancer sites. The study population was restricted only to females for the examination of breast, cervical, uterine and ovarian cancer. Conversely, only males formed the population for the examination of testicular cancer.

Sensitivity analyses were conducted to examine potential confounding. For this purpose, gestational age was analyzed as categories. The population was restricted to full siblings discordant for gestational age category and familial confounding was examined by stratifying analyses by siblings. This approach allows unmeasured confounders shared by full siblings, such as childhood environmental exposures, pregnancy-related factors and lifestyle factors, to be accounted for. Further sensitivity analyses were conducted to determine the potential confounding effect of smoking and parental education using change-in-estimate methods for live births from 1990 onwards.

Additional sensitivity analyses were conducted by excluding cancers reported in the first 28 days or 6 months after birth. For these analyses, the observation time was started at 28 days or 6 months. Finally, the impact of congenital malformations was examined through sensitivity analyses (a) restricted to people with a congenital malformation, (b) restricted to people without a congenital malformation and (c) adjusting for the presence of congenital malformations in the full population.

All tests of statistical hypotheses were performed on the two-sided 5% level of confidence corresponding to two-sided 95% CIs for the HR not overlapping unity. No adjustment for multiplicity of statistical tests was performed. The proportional hazards assumption was examined using Schoenfeld residuals.22, 23 The SAS software 9.4 (proc phreg) was used for all analyses.

3 RESULTS 3.1 Study population

Exactly, 4 063 803 individuals were live-born in Sweden between January 1, 1974 and December 31, 2013 (Figure S1). Of these, 3 144 742 were born to Swedish-born parents. Following the exclusion of 7051 people with missing gestational age or sex, or with implausible death/emigration dates (eg, death prior to birth date), the final cohort consisted of 3 137 691 live births.

3.2 Population characteristics

The characteristics of the 3 137 691 males and females live-born between 1974 and 2013 are outlined in Table 1. They were followed up to a median attained age of 23 years, reached a maximum attained age of 43 years and totaled 71 691 112 person-years of follow-up. Of the births, 2 702 538 (86.1%) occurred at term, 180 363 (5.8%) were born preterm and 254 790 (8.1%) were born postterm. Overall, 1 613 390 (51.4%) of the population were male and 1 524 301 (48.6%) were female.

TABLE 1. Cohort characteristics by gestational age category (in completed weeks) in 3 137 691 live births in Sweden included in our study Characteristic Gestational age category Extremely preterm (<28 weeks) Very preterm (28-31 weeks) Moderate to late preterm (32–36 weeks) All preterm (<37 weeks) Term (37-41 weeks) Postterm (42-45 weeks) All N (%) N (%) N (%) N (%) N (%) N (%) N (%) No. individuals (% of total) 5846 (0.2) 17 014 (0.5) 157 503 (5.0) 180 363 (5.8) 2 702 538 (86.1) 254 790 (8.1) 3 137 691 No. males 3133 (53.6) 9253 (54.4) 85 227 (54.1) 97 613 (54.1) 1 377 161 (51.0) 138 616 (54.4) 1 613 390 (51.4) Birth year 1974-1979 423 (7.2) 1979 (11.6) 20 734 (13.2) 23 136 (12.8) 404 676 (15.0) 62 278 (24.4) 490 090 (15.6) 1980-1984 476 (8.1) 1760 (10.3) 19 222 (12.2) 21 458 (11.9) 323 127 (12.0) 33 351 (13.1) 377 936 (12.1) 1985-1989 637 (10.9) 2276 (13.4) 23 039 (14.6) 25 952 (14.4) 374 526 (13.9) 29 529 (11.6) 430 007 (13.7) 1990-1994 915 (15.7) 2779 (16.3) 23 743 (15.1) 27 437 (15.2) 406 439 (15.0) 32 781 (12.9) 466 657 (14.9) 1995-1999 770 (13.2) 2081 (12.2) 18 099 (11.5) 20 950 (11.6) 299 853 (11.1) 24 747 (9.7) 345 550 (11.0) 2000-2004 874 (15.0) 2194 (12.9) 18 790 (11.9) 21 858 (12.1) 303 746 (11.2) 26 811 (10.5) 352 415 (11.2) 2005-2009 972 (16.6) 2232 (13.1) 19 004 (12.1) 22 208 (12.3) 326 337 (12.1) 25 559 (10.0) 374 104 (11.9) 2010-2013 779 (13.3) 1713 (10.1) 14 872 (9.4) 17 364 (9.6) 263 834 (9.8) 19 734 (7.8) 300 932 (9.6) Birth weight for gestational age Small 617 (10.6) 1719 (10.1) 15 788 (10.0) 18 124 (10.1) 273 401 (10.1) 25 766 (10.1) 317 291 (10.1) Appropriate 4420 (75.6) 13 239 (77.8) 124 837 (79.3) 142 496 (79.0) 2 150 524 (79.6) 202 661 (79.5) 2 495 681 (79.5) Large 619 (10.6) 1719 (10.1) 15 834 (10.1) 18 172 (10.1) 272 839 (10.1) 25 805 (10.1) 316 816 (10.1) Missing 190 (3.3) 337 (2.0) 1044 (0.7) 1571 (0.9) 5774 (0.2) 558 (0.2) 7903 (0.3) Maternal age <20 years 150 (2.6) 434 (2.6) 3773 (2.4) 4357 (2.4) 46 961 (1.7) 5996 (2.4) 57 314 (1.8) 20-24 years 865 (14.8) 2795 (16.4) 27 414 (17.4) 31 074 (17.2) 448 735 (16.6) 48 234 (18.9) 528 043 (16.8) 25-29 years 1719 (29.4) 5126 (30.1) 51 160 (32.5) 58 005 (32.2) 929 382 (34.4) 88 994 (34.9) 1 076 381 (34.3) 30-34 years 1829 (31.3) 5219 (30.7) 46 381 (29.5) 53 429 (29.6) 840 326 (31.1) 75 181 (29.5) 968 936 (30.9) 35–39 years 1005 (17.2) 2785 (16.4) 23 326 (14.8) 27 116 (15.0) 362 985 (13.4) 30 892 (12.1) 420 993 (13.4) ≥40 years 278 (4.8) 655 (3.9) 5449 (3.5) 6382 (3.5) 74 149 (2.7) 5493 (2.2) 86 024 (2.7) Paternal age <20 years 44 (0.8) 130 (0.8) 948 (0.6) 1122 (0.6) 10 703 (0.4) 1257 (0.5) 13 082 (0.4) 20–24 years 485 (8.3) 1556 (9.2) 14 747 (9.4) 16 788 (9.3) 221 316 (8.2) 24 338 (9.6) 262 442 (8.4) 25–29 years 1473 (25.2) 4424 (26.0) 43 336 (27.5) 49 233 (27.3) 754 083 (27.9) 76 046 (29.9) 879 362 (28.0) 30–34 years 1854 (31.7) 5419 (31.9) 50 793 (32.3) 58 066 (32.2) 925 773 (34.3) 85 058 (33.4) 1 068 897 (34.1) 35-39 years 1230 (21.0) 3486 (20.5) 30 743 (19.5) 35 459 (19.7) 534 228 (19.8) 45 966 (18.0) 615 653 (19.6) ≥40 years 760 (13.0) 1999 (11.8) 16 936 (10.8) 19 695 (10.9) 256 435 (9.5) 22 125 (8.7) 298 255 (9.5) Maternal smoking status Smoker 849 (14.5) 2762 (16.2) 22 769 (14.5) 26 380 (14.6) 312 592 (11.6) 23 221 (9.1) 362 193 (11.5) Missing 1638 (28.0) 4437 (26.1) 38 650 (24.5) 44 725 (24.8) 665 227 (24.6) 90 587 (35.6) 800 539 (25.5) Maximal attained parental education at delivery Low (0–9 years) 196 (3.4) 499 (2.9) 3870 (2.5) 4565 (2.5) 53 546 (2.0) 3930 (1.5) 62 041 (2.0) Medium (10–14 years) 2836 (48.5) 7316 (43.0) 62 749 (39.8) 72 901 (40.4) 1 032 189 (38.2) 82 766 (32.5) 1 187 856 (37.9) High (≥15 years) 1277 (21.8) 3175 (18.7) 27 846 (17.7) 32 298 (17.9) 514 140 (19.0) 42 914 (16.8) 589 352 (18.8) Missing 1537 (26.3) 6024 (35.4) 63 038 (40.0) 70 599 (39.1) 1 102 663 (40.8) 125 180 (49.1) 1 298 442 (41.4) 3.3 Any primary malignant cancer

A total of 22 604 primary malignant cancers were recorded during the follow-up period, equating to an overall cancer rate of 31.5 (95% CI 31.1-31.9) per 100 000 person-years. Crude rates were highest in the postterm group and lowest in the preterm group (Table 2). After adjusting for birth year, maternal age and paternal age a higher risk of any primary malignant cancer was observed in those born moderate to late preterm (32-36 weeks) compared to those born at term (aHR 1.07, 95% CI 1.01-1.14). Relative to week 40, this increased risk was observed specifically for those born at weeks 30 to 35 (Figure 1). Although not supported by our statistical inference, elevated aHR point estimates, but not lower confidence limits, were also observed for weeks 29 and 36. The risk was highest for births at 31 weeks (aHR 1.18, 95% CI 1.05-1.32) (Figure 2, Table 2). No difference in risk was observed for postterm compared to term births.

TABLE 2. Cancer rate per 100 000 person-years and hazard ratio (95% CI) for any primary malignant cancer by gestational age in weeks and categories, using 40 weeks or term category as the reference Gestational age Week Cancer events Person-years Rate (95% CI) Model 1a HR (95% CI) Change in risk per week (95% CI) Model 2b HR (95% CI) Change in risk per week All births 23-45 22 604 71 691 112 31.5 (31.1-31.9) Extremely preterm 23 20 78 292.4 25.6 (16.5-39.6) 1.00 (0.64-1.54) 0.80 (0.43-1.48) 0.88 (0.61-1.26) 1.00 (0.64-1.54) 0.80 (0.43-1.48) 0.87 (0.61-1.24) 24 0.85 (0.50-1.44) 0.85 (0.50-1.44) 25 0.90 (0.58-1.40) 0.90 (0.57-1.40) 26 0.95 (0.66-1.37) 0.95 (0.66-1.37) 27 1.01 (0.76-1.34) 1.01 (0.76-1.34) Very preterm 28 105 335 483.1 31.3 (25.9-37.9) 1.08 (0.89-1.31) 1.06 (0.86-1.31) 1.04 (0.87-1.24) 1.08 (0.89-1.31) 1.06 (0.86-1.31) 1.03 (0.86-1.24) 29 1.12 (0.95-1.30) 1.12 (0.95-1.30) 30 1.16 (1.02-1.31) 1.16 (1.02-1.31) 31 1.17 (1.05-1.32) 1.18 (1.05-1.32) Moderate to late preterm 32 1143 3 495 387.8 32.7 (30.9-34.7) 1.07 (1.01-1.14) 1.17 (1.05-1.30) 0.96 (0.92-1.01) 1.07 (1.01-1.14) 1.17 (1.05-1.30) 0.96 (0.92-1.01) 33 1.14 (1.04-1.25) 1.14 (1.04-1.25) 34 1.10 (1.03-1.18) 1.10 (1.03-1.18) 35 1.07 (1.02-1.12) 1.07 (1.02-1.12) 36 1.03 (1.00-1.07) 1.03 (1.00-1.07) Term 37 19 142 61 423 085.2 31.2 (30.7-31.6) Ref 1.02 (0.99-1.04) 1.00 (0.99-1.01) Ref 1.01 (0.99-1.04) 1.00 (0.99-1.01) 38 1.00 (0.99-1.02) 1.00 (0.99-1.02) 39 1.00 (0.99-1.01) 1.00 (0.99-1.01) 40 Ref Ref 41 1.00 (0.99-1.01) 1.00 (0.99-1.01) Postterm 42 2194 6 358 864.1 34.5 (33.1-36.0) 1.00 (0.96-1.04) 1.00 (0.98-1.02) 1.04 (0.95-1.12) 1.00 (0.96-1.04) 1.00 (0.98-1.02) 1.03 (0.95-1.12) 43 1.00 (0.97-1.03) 1.00 (0.97-1.04) 44 1.00 (0.96-1.05) 1.00 (0.96-1.05) 45 1.00 (0.95-1.06) 1.00 (0.95-1.06) a Model 1 is adjusted for birth year. b Model 2 is adjusted for birth year, maternal age and paternal age. Adjusted hazard ratio (HR) for any primary malignant cancer according to gestational age, adjusted for birth year, maternal age and paternal age [Color figure can be viewed at wileyonlinelibrary.com]