European Journal of Epidemiology

, Volume 28, Issue 3, pp 257–265

Childhood infectious disease and premature death from cancer: a prospective cohort study

Authors

  • Peter W. G. Tennant
    • Institute of Health & SocietyNewcastle University
  • Louise Parker
    • Departments of Medicine and PediatricsPopulation Cancer Research Program
  • Julian E. Thomas
    • Newcastle upon Tyne NHS Foundation Trust, Sir James Spence InstituteRoyal Victoria Infirmary
  • Sir Alan W. Craft
    • Northern Institute of Cancer ResearchNewcastle University
    • Sir James Spence InstituteRoyal Victoria Infirmary
    • Institute of Health & SocietyNewcastle University
    • Sir James Spence InstituteRoyal Victoria Infirmary
CANCER

DOI: 10.1007/s10654-013-9775-1

Cite this article as:
Tennant, P.W.G., Parker, L., Thomas, J.E. et al. Eur J Epidemiol (2013) 28: 257. doi:10.1007/s10654-013-9775-1

Abstract

Studies of the association between early life infections and cancer have produced inconsistent findings, possibly due to limited adjustment for confounding and retrospective designs. This study utilised data from the Newcastle Thousand Families Study, a prospective cohort of 1,142 individuals born in Newcastle-upon-Tyne in 1947, to assess the impact of various childhood infectious diseases on cancer mortality during ages 15–60 years. Detailed information was collected prospectively on a number of early life factors. Deaths from cancer during ages 15–60 years were analysed in relation to childhood infections, adjusting for potential early-life confounders, using Cox proportional-hazards regression. In a subsample who returned questionnaires at aged 49–51 years, additional adjustment was made for adult factors to predict death from cancer during ages 50–60 years. Childhood history of measles and influenza, were both independently associated with lower cancer mortality during ages 15–60 years (adjusted hazard ratios = 0.39, 95 % CI 0.17–0.88 and 0.49, 95 % CI 0.24–0.98 respectively). In contrast, childhood pertussis was associated with higher cancer mortality during ages 15–60 years (adjusted hazard ratio = 4.88, 95 % CI 2.29–10.38). In the subsample with additional adjustment for adult variables, measles and pertussis remained significantly associated with cancer mortality during ages 50–60 years. In this pre-vaccination cohort, childhood infection with measles and influenza were associated with a reduced risk of death from cancer in adulthood, while pertussis was associated with an increased risk. While these results suggest some disease-specific associations between early-life infections and cancer, further studies are required to confirm the specific associations identified.

Keywords

EpidemiologyCommunicable diseasesMalignancyPertussisMeasles

Introduction

Childhood infectious diseases are intimately related to early-life disadvantage [1], but their influence on adult health is unclear [2]. Several studies have demonstrated associations between childhood respiratory infections and impaired lung function in adulthood [3], while some viral infections have known oncogenic function [4]. Nevertheless, it is frequently proposed that acute childhood infectious diseases may have long-term health benefits, including reducing the risks of autoimmune diseases and cancer [5, 6].

The apparent inverse association between febrile infection and the risk of cancer has long been anecdotally recognised [7]. A number of epidemiological studies have since explored this phenomenon, with inconsistent findings [816]. Several early studies reported large protective effects for increasing numbers of infectious diseases [810], but with serious methodological limitations [11]. More recently, Abel et al. [11] and Albonico et al. [12] found more modest associations, while Hoffman et al. identified increased risks of cancer among those with a history of mumps or pertussis [13]. Regardless, early-life infections are frequently cited to explain associations with adult cancer, such as for achieved height [17].

Previous studies of the association between childhood infectious disease and adult cancer are impaired by their retrospective designs and lack of account for potential early-life confounding influences such as socioeconomic circumstances, gestational age, and birth weight [1113]. Initially established in 1947 to investigate the burden of childhood infectious disease in Newcastle-upon-Tyne, a city in northern England, the Newcastle Thousand Families Study is a prospective birth cohort study that collected comprehensive information on childhood environment, socioeconomic circumstances, and history of infectious disease [18]. Details of all deaths in adulthood are routinely notified to the study team. Although less informative than total cases of cancer, this nevertheless presents a unique opportunity to study the association between childhood infectious diseases and all fatal cases of cancer in adulthood.

Methods

Study participants

Full details of the Newcastle Thousand Families Study are available elsewhere [18]. Briefly, all 1,142 babies born to mothers resident in Newcastle-upon-Tyne (UK) during May–June 1947 were recruited into a prospective cohort study. Information various factors, including birth weight, birth order, gestational age, duration breast fed, social class, and housing conditions were recorded prospectively. Birth weight, as recorded by the midwife, was standardised for gestational age and sex [19]. Birth order was dichotomised into ‘first’ and ‘subsequent’. Social class was classified into three summary categories (I–II, III, and IV–V) according to the UK Registrar General’s Standard Occupation Classification (where I is the most advantaged and V the least) using paternal occupation at birth. Housing conditions were assessed by the city’s Public Health Department, and scored for the presence of overcrowding, lack of hot water, toilets shared between households, and dampness or poor repair, with scores of two or more being combined.

Over the following 15 years, episodes of illness were reported to the study team by health visitors, who visited the families regularly throughout childhood, and general practitioners if a child presented to them. The study team was also informed whenever the children were referred to, or attended, hospital. Histories of influenza, measles, mumps, pertussis, rubella, tuberculosis, and varicella were coded as dichotomous variables (‘had’, ‘did not have’), while number of episodes of lower respiratory tract infections (LRTIs), upper respiratory tract infections (URTIs), and gastrointestinal infections (GIs) were recorded as counts.

All deaths and emigrations are ‘flagged’ by the UK National Health Service Central Register (NHSCR), enabling deaths from cancer to be identified from death certificates (as supplied by the NHSCR). Individuals who emigrated during adulthood, and have been lost-to-follow-up, were censored at their time of emigration. Emigrants whose date of departure was prior to age 15 years or where the exact date is unknown were excluded, as were those who died before age 15 years.

Information from adulthood

At age 49–51 years, 574 study members self-completed a comprehensive questionnaire on health and lifestyle [18]. Adult social class was estimated from details of the household’s main wage earner at age 49–51 years, and again classified into three categories (I–II, III, and IV–V) according to the UK Registrar General’s Standard Occupation Classification. Current alcohol consumption was divided into four categories, based on average units consumed per week. Light drinking was defined as less than five units per week for women (ten for men) and moderate drinking as up to 21 units for women (28 for men). In the UK, 1 unit is 10 ml (equivalent to 7.9 g) of pure alcohol. Moderate or vigorous physical activity was assessed over four domains (at work, cycling/walking, sporting/recreational, gardening/DIY) using questions from the Medical Research Council’s national survey of health and development [20], and summarised into low (no more than occasional in any domain), medium (regular in one domain, excluding sporting/recreational), high (regular sporting/recreational, or in two or more domains). The number of pack-years of cigarettes smoked (one pack-year equals one pack of 20 cigarettes smoked per day for 1 year) was estimated from self-reported smoking habits at ages 15, 25, 35 and 50 years. Use of hormone replacement therapy medication was estimated from self-reported medicinal intake. Ethical approval for the study was obtained from the appropriate local research ethics committees and the study has been performed in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki and its later amendments.

Statistical analysis

All variables with a hypothesised effect on the risk of developing cancer or of death from incident cancer were examined. The overall influence of each early-life factor on death from cancer during ages 15–60 years was assessed using Cox proportional hazards regression. The independent impact of each variable was estimated by constructing an adjusted model using a backwards stepwise approach. To assess the potential influence of factors acting in adulthood, a sensitivity analysis was performed among those participants who returned the questionnaire at aged 49–51 years. Cox proportional hazards regression was used to predict death from cancer during ages 50–60 years, assessing the independent impact of each variable again using a backwards stepwise approach. Variables or categories with complete estimability were excluded from Cox proportional hazards regression models; but their effects were approximated by exact Poisson regression. The proportional hazards assumption was examined using the Grambsch and Therneau test. Interactions between significant variables in the adjusted models, and with sex, were examined using cross-product terms. Participants with missing data were excluded from individual analyses by listwise deletion. Statistical analyses were performed using Stata 11.1 (StataCorp, College Station, USA). p < 0.05 was considered statistically significant.

Results

Figure 1 shows the flow of study members through this investigation. Of the original 1,142, 54 died and 23 emigrated before aged 15 years while five emigrated and/or were lost to follow-up at an unknown age. The remaining 1,060, known to be alive at age 15 years, constitute the participants in this investigation. Tables 1 and 2 detail the clinical and socio-demographic characteristics of the sample. Briefly, around half were male (n = 539, 50.9 %), three quarters were firstborn (n = 786, 74.2 %), and the majority lived in houses with at least one indicator of adverse environment (n = 540, 56.5 %). Over half (n = 573, 57.1 %) had fathers who were employed in skilled manual or non-manual jobs (III), and nearly a third (n = 314, 31.3 %) had fathers who were employed in partly skilled or unskilled manual jobs (IV–V). The median standardised birth weight was −0.2 standard deviations (IQR: −0.8–0.5) at a median gestational age of 40 weeks (IQR: 40–40). The most common childhood infectious diseases were measles (n = 768, 72.5 %), influenza (n = 707, 66.7 %), and pertussis (n = 534, 50.4 %). Participants had a median of 6 (IQR: 3–11) URTIs, 1 (0–2) GI, and 0 (IQR: 0–1) LRTIs recorded during their first 15 years.
https://static-content.springer.com/image/art%3A10.1007%2Fs10654-013-9775-1/MediaObjects/10654_2013_9775_Fig1_HTML.gif
Fig. 1

Showing the flow of participants through the study. The complete sample was used to investigate the association between childhood infectious diseases and risk of death from cancer during ages 15–60 years, adjusting for potential confounding factors acting in childhood. The sub-sample was used to investigate the association between childhood infectious diseases and risk of death from cancer during ages 50–60 years, adjusting for potential confounding factors acting in childhood and adulthood

Table 1

Descriptive statistics for continuous demographic variables and generic childhood infectious diseases

Variable (unit)

N

Range

Median (IQR)

Fetal and childhood

Duration breast fed (weeks)

919

0–63.3

8.7 (3.3–31.0)

Gestational age (weeks)

947

28–44

40 (40–40)

Standardised birth weighta (z-score)

959

−3.8–4.7

−0.2 (−0.8–0.5)

GIs during ages 0–15 years

1,060

0–11

1 (0–2)

LRTIs during ages 0–15 years

1,060

0–20

0 (0–1)

URTIs during ages 0–15 years

1,060

0–33

6 (3–11)

Adult

Cigarette smoking historyb (pack-years)

359

0.1–100.5

21.0 (8.0–31.0)

GIs gastrointestinal infections, IQR inter-quartile range, LRTIs lower respiratory tract infections, URTIs upper respiratory tract infections

aAdjusted for sex and gestational age

bExcluding 215 who had never smoked

Table 2

Descriptive statistics for categorical demographic variables and specified childhood infectious diseases

Variable

N (%)

Fetal and childhood

 Birth order

n = 1,060

  1st

786 (74.2)

  ≥2nd

274 (25.9)

 Housing conditions at birtha

n = 956

  0

416 (43.5)

  1

255 (26.7)

  ≥2

285 (29.8)

 Sex

n = 1,060

  Male

539 (50.9)

  Female

521 (49.2)

 Social class at birthb

n = 1,003

  I–II

116 (11.6)

  III

573 (57.1)

  IV–V

314 (31.3)

 Influenza during ages 0–15 years

n = 1,060

  No

353 (33.3)

  Yes

707 (66.7)

 Measles during ages 0–15 years

n = 1,060

  No

292 (27.6)

  Yes

768 (72.5)

 Mumps during ages 0–15 years

n = 1,060

  No

761 (71.8)

  Yes

299 (28.2)

 Pertussis during ages 0–15 years

n = 1,060

  No

526 (49.6)

  Yes

534 (50.4)

 Rubella during ages 0–15 years

n = 1,060

  No

858 (80.9)

  Yes

202 (19.1)

 Tuberculosis during ages 0–15 years

n = 1,060

  No

955 (90.1)

  Yes

105 (9.9)

 Varicella during ages 0–15 years

n = 1,060

  No

532 (50.2)

  Yes

528 (49.8)

Adult

 Alcohol consumption at age 49–51 years

n = 563

  None

65 (11.5)

  Low

237 (41.8)

  Moderate

217 (38.3)

  High

48 (8.5)

 Physical activity at age 49–51 years

n = 565

  Low

211 (37.4)

  Medium

191 (33.8)

  High

163 (28.9)

 Social class at age 49–51 yearsb

n = 535

  I–II

274 (51.2)

  III

180 (33.6)

  IV–V

81 (15.1)

 Use of hormone replacement therapy

n = 574

  No

490 (85.4)

  Yes

84 (14.6)

aFrom zero to two or more of: overcrowding, lack of hot water, shared toiled, and dampness or poor repair

bAs defined by the UK Registrar General’s Standard Occupational Classification, where I is the most advantaged, and V the least

Predictors of death from cancer during ages 15–60 years

Between ages 15–60 years, 89 participants (55 men and 33 women) died and 29 emigrated and/or were lost to follow-up. Cardiovascular disease was the leading cause of death among men (n = 21, 38 %), followed by cancer (n = 18, 33 %). Cancer was the leading cause of death among women (n = 19, 58 %). Of the deaths from cancer, most were lung cancer (n = 10, 26 %) followed by breast (n = 5, 13 %), haematological (n = 5, 13 %), ovarian (n = 4, 11 %), oesophageal (n = 3, 8 %), pancreatic (n = 3, 8 %), and skin (n = 3, 8 %). No other site had more than one case. There was no significant difference in the number of deaths from cancer between men and women (p = 0.80). The median age at death was 50.6 years for men and 53.2 years for women.

Table 3 shows the crude and adjusted hazard ratios for all available early-life factors in relation to death from cancer during ages 15–60 years. Before adjustment, only childhood history of pertussis was predictive of death from cancer, with an increased risk among those with a history of infection (unadjusted hazard ratio, HR = 3.17, 95 % confidence interval, CI 1.50–6.72, p = 0.003). This effect was increased (adjusted HR = 4.88, 95 % CI 2.29–10.39, p < 0.0001) when adjustment was made for childhood history of measles and influenza, both of which were associated with lower risks of death from cancer in the adjusted model (influenza: adjusted HR = 0.39 (0.17–0.89), p = 0.03; measles: adjusted HR = 0.49, 95 % CI 0.24–0.98, p = 0.04). Supplementary Table 1 shows the most common site-specific cancers, by childhood histories of influenza, measles, and pertussis.
Table 3

Crude and adjusted hazard ratios for death from cancer during ages 15–60 years

Variable

Unadjusted

Adjusteda

Hazard ratio (95 % CI)

p value

Hazard ratio (95 % CI)

p value

Birth order

 1st

Reference

0.27

  

 ≥2nd

1.59 (0.70–3.61)

   

Housing conditions at birthb

 0

Reference

0.56

  

 1

0.82 (0.37–1.81)

   

 ≥2

0.64 (0.28–1.46)

   

Sex

 Male

Reference

0.88

  

 Female

1.05 (0.55–2.00)

   

Social class at birthc

 I–II

0.95 (0.28–3.28)

   

 III

Reference

0.35e

  

 IV–V

1.63 (0.81–3.25)

   

Influenza during ages 0–15 years

 No

Reference

0.05

Reference

0.03

 Yes

0.45 (0.20–1.01)

 

0.39 (0.17–0.89)

 

Measles during ages 0–15 years

 No

Reference

0.32

Reference

0.04

 Yes

0.71 (0.36–1.39)

 

0.49 (0.24–0.98)

 

Mumps during ages 0–15 years

 No

Reference

0.92

  

 Yes

0.96 (0.47–1.99)

   

Pertussis during ages 0–15 years

 No

Reference

0.003

Reference

<0.0001

 Yes

3.17 (1.50–6.72)

 

4.88 (2.29–10.39)

 

Rubella during ages 0–15 years

 No

Reference

0.64

  

 Yes

1.21 (0.55–2.63)

   

Tuberculosis during ages 0–15 years

 No

Reference

0.81

  

 Yes

0.87 (0.27–2.82)

   

Varicella during ages 0–15 years

 No

Reference

0.26e

  

 Yes

0.68 (0.36–1.32)

   

Duration breast fed (weeks)

1.00 (0.98–1.02)

0.98

  

Gestational age (weeks)

1.02 (0.86–1.21)

0.80

  

Standardised birth weightd (z-score)

1.25 (0.98–1.60)

0.07

  

GIs during ages 0–15 years

1.04 (0.86–1.27)

0.68

  

LRTIs during ages 0–15 years

1.03 (0.90–1.19)

0.66

  

URTIs during ages 0–15 years

1.00 (0.95–1.05)

0.96

  

GIs gastrointestinal infections, LRTIs lower respiratory tract infections, URTIs upper respiratory tract infections

aAdjusted model was constructed using a backwards stepwise approach. All variables were entered into the model, and then non-significant variables were iteratively removed (according to decreasing p value) until only those with p < 0.1 remain, details of which are shown

bFrom zero to two or more of: overcrowding, lack of hot water, shared toiled, and dampness or poor repair

cAs defined by the UK Registrar General’s Standard Occupational Classification, where I is the most advantaged, and V the least

dAdjusted for sex and gestational age

eLog rank test result, due to violation of proportional hazards assumption

The only other variable to approach nominal significance was standardised birth weight. Before adjustment, increasing birth weight was associated with increasing risk of death from cancer (unadjusted HR, per z-score unit = 1.25, 95 % CI 0.98–1.60, p = 0.07), albeit not statistically significantly. Regardless, the association was attenuated when adjustment was made for histories of influenza, measles, and pertussis (adjusted HR = 1.20, 95 % CI 0.93–1.55, p = 0.17).

Sensitivity analysis using adult lifestyle data

The potential influence of factors acting in adulthood on the relationship between early-life factors and death from cancer in adulthood was assessed by examining predictors of death during ages 50–60 years in those participants (n = 574) who returned a health and lifestyle questionnaire at aged 49–51 years. 23 of the subsample died during aged 50–60 years. Cancer was the leading cause of death (n = 14) accounting for six male deaths and eight female deaths.

Hormone replacement medication usage and greater number of pack years of cigarettes smoked were the only adult variables (either before or after adjustment) that were significantly associated with death from cancer during ages 50–60 years (hormone replacement medication: adjusted HR = 4.43, 95 % CI 1.61–12.19, p = 0.004; smoking history, per pack-year: adjusted HR = 1.03, 95 % CI 1.01–1.05, p = 0.005). Of the early life factors, positive history of pertussis and negative history of measles were also associated with death from cancer during ages 50–60 years, independent of each other and of hormone replacement medication and cigarette smoking history (pertussis: adjusted HR = 6.39, 95 % CI 1.57–26.03, p = 0.01; measles: adjusted HR = 0.17, 95 % CI 0.06–0.49, p = 0.001). Unlike in the larger sample, no association was observed between history of influenza in childhood and death from cancer during 50–60 years (unadjusted HR = 0.85, 95 % CI 0.29–2.55, p = 0.78).

Two statistically significant interactions were observed, both in the model of death from cancer during ages 50–60 years. The effect of pertussis on risk of death from cancer during ages 50–60 years was greater in women (adjusted HR indeterminate, adjusted risk ratio = 9.50, 95 % CI 1.40–∞, p = 0.02) than in men (adjusted HR = 2.37, 95 % CI 0.53–10.68, p = 0.26) (p < 0.0001). The same interaction was present in the larger sample of death from cancer during ages 15–60 years (pertussis adjusted HR, in women = 14.09, 95 % CI 3.47–57.26, p = 0.0002; in men = 2.32, 95 % CI 0.88–6.10, p = 0.09), but was on the margin of statistical significance (p = 0.06). The second interaction in the model of deaths during ages 50–60 years was between history of pertussis and hormone replacement medication usage, and was explained by the fact that all hormone replacement users were women.

Discussion

Principal findings

Using prospectively-collected data from a pre-vaccination cohort, and adjusting for various early-life influences, childhood history of infection with either measles or influenza were both independently associated with a lower risk of death from cancer during ages 15–60 years. Conversely, history of pertussis was associated with a significantly increased risk of premature death from cancer, the effect being larger in women than in men.

In a sensitivity analysis, performed on a sub-sample of adults who returned health and lifestyle questionnaires at aged 49–51 years, both measles and pertussis remained significantly associated with death from cancer during ages 50–60 years. Neither of the associations was attenuated by adjusting for potential confounding variables in adulthood, including cigarette smoking history and hormone replacement medication, which both independently predicted of death from cancer during ages 50–60 years.

Strengths and weaknesses

This study utilised detailed, prospectively-collected data on a range of factors operating around birth and throughout childhood. Episodes of illness were reported prospectively by multiple sources. Follow-up for the primary outcome was relatively complete, with only 2.5 % of participants being lost to follow-up before aged 15-years. We were able to perform a sensitivity analysis in a subsample of over half the original cohort, a participation rate that compares favourably to other long-term follow-up studies.

This study has several limitations. Most prominently, by examining only deaths from cancer, rather than all cases of cancer (which are not known), it is not possible to say whether the identified associations are due to an increased risk of developing cancer, an increased risk of dying from cancer, or both. The reduction in the number of available cases will have also reduced the power to detect more modest associations. Most deaths from cancer in this cohort are likely to occur after age 60 years, but it is not yet possible to study these cases, since the cohort are not old enough. Those cases that are included may be untypical of incident cases in terms of severity and/or circumstances, thus the results may not be generalisable to all cancers. Information on death certificates is likely to be of variable quality, and some deaths from cancer may have been missed, although it seems unlikely this would be related to history of infectious disease in childhood. Cancer is a heterogeneous group of conditions with varied aetiologies, pathologies, and survival rates [21]. It seems unlikely that any infectious exposure would affect all cancer subtypes consistently. Potential associations with specific subtypes may therefore have been masked by differences in effects, such as associations acting in opposite directions. At the least, the overall effects are likely to have been driven by those types with the highest incidence and mortality.

Although we were able to adjust for various childhood factors, and the sensitivity analysis found no mediating influences in adulthood, the observed associations may still be influenced by residual confounding. No information was available on childhood diet, which may modify the risk of cancer [22], although food rationing (in place until study members were aged six years) will have reduced some of the early-life variation in diet. Exposure to parental smoking was also not collected, which may simultaneously increase the risks of early-life respiratory infections [23], and cancer [24], but cannot explain the apparently protective effects of measles and influenza.

Most childhood diagnoses were based on clinical presentation alone. Cases of influenza, in particular, may therefore include other febrile illnesses. Some outbreaks may not have been reported or recorded, although any resulting bias cannot explain all of the measles, influenza and pertussis results, given that they act in opposite directions.

Overall, the study had relatively modest power, because of the small number of deaths from cancer. This prevented us from formally analysing the associations with individual cancer subtypes. Although site-specific associations are presented in Supplementary Table 1, there were too few cases to perform significance tests, so these results should be considered largely exploratory. Low study power will also have hindered investigations of rarer childhood diseases, such as tuberculosis. To maximise power, no adjustment was made for multiple testing, despite a large number of infections being examined. A stepwise approach was used to produce simplified adjusted models, but this approach can result in overly sample-specific models with inflated precision. The possibility of both false negative and false positive results should hence not be discounted. The sensitivity analysis was performed on an incomplete subsample of participants who returned the health and lifestyle questionnaire at aged 49–51. Those who returned the questionnaire were not representative of the cohort at birth, with notably higher recorded incidences of influenza (39 vs. 27 %, p < 0.001), measles (82 vs. 61 %, p < 0.001), and pertussis (56 vs. 43 %, p < 0.001). This may suggest selection biases within the sensitivity analysis sample. However, since the associations with death from cancer were very similar in the subsample to the results observed in the complete cohort, this suggests such biases are unlikely to have had a meaningful impact.

Comparison with other studies

Comparisons with the existing literature are problematic because previous analyses typically focus on infections in the recent past or examine composite exposures. Most also utilise a case–control design, relying on self-reported information on childhood infectious illness and many have low statistical power, due to widespread vaccination against the exposures of interest.

Historical case–control studies reported markedly lower self-reported rates of childhood infectious diseases among cases with adult cancer [810]. However, the absolute rates of infectious disease were extremely low, raising serious concerns about ascertainment [11]. More recently, Abel et al. [11] reported a lower risk of stomach, colon, rectal, breast, and ovarian cancer associated with self-reported histories of pertussis and varicella, contrasting with the current study. These effects, however, were only present when comparing with population controls (pertussis: odds ratio, OR = 0.63, 95 % CI 0.41–0.96, varicella: OR = 0.66, 95 % CI 0.45–0.97), not hospital controls (pertussis: OR = 1.03, 95 % CI 0.61–1.74, varicella: OR = 0.83, 95 % CI 0.52–1.32), suggesting residual confounding [11]. Looking specifically at malignant melanoma, Kolmel et al. [14] found no evidence that either measles, mumps, rubella, varicella, or pertussis were associated with the cancer in adulthood. Albonico et al. [12] identified a trend for decreasing risk of malignant solid epithelial tumours for increasing number of febrile diseases in childhood, but again no specific illness was attributed, with notably flat odds ratios for both measles and pertussis (measles: OR = 0.87, 95 % CI 0.57–1.34, pertussis: OR = 0.92, 95 % CI 0.67–1.25). In a large study, McDuffie et al. [15] reported significantly lower risks of non-Hodgkin’s lymphoma among men with histories of measles (OR = 0.64, 95 % CI 0.51–0.79) and mumps (OR = 0.75, 95 % CI 0.60–0.93). Hoffman et al. [13] found no evidence that any childhood illnesses conferred a protective effect on the risk of cancer, however they did identify significantly increased risks associated with mumps (OR = 2.61, 95 % CI 1.18–5.80) and pertussis (OR = 2.71, 95 % CI 1.30–5.64), the latter result being notably similar to the current study. Wrotek et al. [16] reported that a lower proportion of cancer patients had mumps (OR = 0.50, 95 % CI 0.35–0.70), rubella (OR = 0.40, 95 % CI 0.27–0.60), and varicella (OR = 0.58, 95 % CI 0.41–0.82), compared with healthy area-matched controls. There was also suggestion of an increased risk associated with measles, although the result was not statistically significant (OR = 1.33, 95 % CI 0.94–1.88) [16]. Finally, three separate studies have shown a protective association between mumps and risk of ovarian cancer [2527]. West [26] and Newhouse et al. [25] also found evidence that measles conferred a similar protective effect of ovarian cancer, although for Newhouse et al. [25], this was only when comparing to population, not hospital, controls.

Potential mechanisms and implications

Two competing hypotheses seek to explain why childhood infectious diseases, such as measles and influenza, may confer a protective effect on the risk of cancer in adulthood. Krone et al. [28] propose that the recognition of cancer cells by anti-cancer elements of the immune system (the process of immuno-surveillance) might be enhanced by prior contact with pathogenic proteins that are structurally homologous to those found on cancerous cells (a phenomenon known as unexpected antigenic cross-reactivity). Cramer and Finn [29] alternatively propose that cell surface antigens may undergo similar changes in structure and expression when responding to infection as to when undergoing malignant transformation, thereby training an antigen-specific immune response.

Reasons for the observed association between pertussis and a higher risk of premature death from cancer are unclear. Both pertussis infection and cancer may share common risk factors not examined in the current study, such as parental smoking [30] or lower fruit and vegetable consumption during childhood [22]. Alternatively, the results could be explained by a pre-existing susceptibility to both conditions, such as poor immune function. Speculatively, the a more directly causal mechanism could be involved if exposure to the pertussis toxin (which arrests cell cycle progression by indirectly inhibiting adenylate cyclase [31]) provoked a relative increase in cell proliferation once the toxin was cleared.

Given the small sample size and risks of multiple testing, the current study should be viewed primarily as hypothesis-generating. Younger cohorts living in more economically developed nations are now routinely vaccinated against both measles and pertussis, and it is unclear how the results of the current study, obtained from an historical, pre-vaccinated cohort, would transfer to these populations. Nevertheless, measles and pertussis remain endemic in many parts of the world [32, 33], and there is also currently an epidemic of pertussis within the UK [34]. Further studies in larger and more recent cohorts (preferably with the power to look at the incidence of individual cancer subtypes) are needed to confirm the currently identified associations before any public health implications can be drawn.

Conclusions

In a cohort of men and women born in Newcastle-upon-Tyne during 1947, childhood infection with measles and influenza were associated with reduced risks of death from cancer in adulthood, while pertussis was associated with an increased risk. These results suggest that there may be some disease-specific effects of childhood infectious diseases on the risk of cancer in adulthood. However, further studies are required to confirm the specific associations identified, particularly given the current lack consensus within the literature.

Acknowledgments

We thank all the Thousand Family Study members for taking part and the study teams past and present.

Conflict of interest

All authors declare that they have no competing interests.

Supplementary material

10654_2013_9775_MOESM1_ESM.docx (17 kb)
Supplementary material 1 (DOCX 17 kb)

Copyright information

© Springer Science+Business Media Dordrecht 2013