Loss to follow-up and selection bias

Of the original 1673 live births included in the cohort, 113 (6.8%) had died before the age of 18, and were therefore excluded from these analyses of CVD in adults. Data on hospital admissions in the HES database were identified for 931 of the remaining 1560 cohort members (see Figure 2). Of the 629 cohort members not found in the HES database, 172 were recorded in the Guernsey death registers as having died prior to the introduction of the HES database in 1997, leaving 457 cohort members unaccounted for and lost to follow-up. Some of these cohort members may have left Guernsey permanently before or after the 1940–45 occupation, while others may be residents who have yet to be admitted to the island's hospital. Of the 931 cohort members found in the HES database, 873 also had complete birth data and were included in the analyses that follow. This represents 52% follow-up of the original cohort and 63% follow-up of those who were alive when the HES database was first introduced. The matching process identified 225 individuals who were registered as being resident in Guernsey during the 1940–45 occupation, and these were therefore classified as exposed. Since the 'occupation register' used for this purpose provided a complete list of all islanders who were resident at the time of the 1940–45 occupation, none of the remaining cohort members (N = 648) would have been exposed to the occupation, and these were therefore classified as unexposed.

Sensitivity analyses and power calculations

The sensitivity analyses, which assumed that all those lost to follow-up had not been admitted to hospital for CVD, produced results that were no different for any of the three models in Tables 1, 2, 3. Meanwhile, the power calculations suggested that the number of cohort members with lower than mean birth weight in each of the analyses summarised in Table 1 (n = 425) was sufficient to detect just under a doubling of the hazard for CVD (or a HR of 1.92) at 90% power and a significance level of 0.05. If the Cox regression models in Table 1 had been calculated with 'lower than mean birth weight' as an exposure (rather than per kg increase in birth weight), the hazard ratio observed in the unadjusted model would have been 1.47. This is substantially smaller than that powered by the study, suggesting that the study was underpowered to examine a relationship of smaller magnitude between birth weight and CVD in later life and is therefore at risk of a type II error. For the models exploring the relationship between exposure to the occupation and CVD (Table 2), similar power calculations suggested that the number of cohort members who had been exposed to the occupation in the analyses summarised in Table 2 (n = 225) was sufficient to detect a doubling of the hazard for CVD (i.e. an HR of 1.99) at 90% power with a significance level of 0.05 and without adjustment for potential confounders. This is less than the actual effect size detected by either of the analyses presented in the second set of models (Table 2), and it therefore seems likely that the study was adequately powered for these models.

Potential limitations

This study had a number of potential limitations which should be taken into account when interpreting its results. First, the cohort comprises only home births delivered by a single community midwife, and as such is unlikely to be representative of all births in Guernsey at that time. Indeed, when these births were compared to a 10% representative sample of births during the same period drawn from the Guernsey birth registers, it was found that the births delivered by the community midwife were significantly more likely to have taken place in urban parishes [29]. Sociodemographic differences were found in separate analyses comparing those who had been resident in Guernsey during the occupation and those who had been evacuated and subsequently returned [30], as well as analyses comparing cohort members of above and below average birth weight. However, because all the models adjusted for age (as this was the temporal variable included in the Cox regression analyses), parish of residence at birth (for models B), and sex (for models A), any potential selection bias related to these three sociodemographic variables should have been ameliorated in these analyses.

Nonetheless, the study suffers from substantial loss to follow-up, since the main analyses only include those cohort members who could be linked to the HES database. Those lost to follow-up would include any cohort members who had died prior to the introduction of the HES database in 1997, any who were evacuated from Guernsey and did not return, and any still living in Guernsey who had yet to attend the island's hospital and thereby enter the HES database (whether for CVD or non-CVD complaints). However, a comparison of cohort members included in the HES database with those who were not found few sociodemographic or clinical differences, although those included in the HES database were less likely to have died in Guernsey and more likely to have been exposed to the 1940–45 occupation. While such differences will affect the generalisability of the results, they may well be the result of population movement rather than pre-existing sociodemographic differences between those included in the analyses and those lost to follow-up. In this regard, the results of the sensitivity analyses were encouraging since these were not substantially different from the main analyses when all those lost to follow-up were assumed not to have experienced a CVD event.

Finally, the power calculations indicated that the analyses in the first set of models (A: birth weight) are likely to have been underpowered. If so, it is possible that a significant relationship may have been found between birth weight and CVD had a larger sample been available for analysis. However, because the modest relationship between birth weight and CVD amongst men and women combined was positive rather than inverse, and was only inverse amongst men when the analysis was stratified by sex, it seems unlikely that a larger sample size would have generated the statistically significant inverse relationship between birth weight and CVD that has commonly been observed in previous studies [4].

Interpretation of findings

Notwithstanding these potential limitations, the lack of a significant inverse relationship between birth weight and CVD in this cohort runs contrary to the findings of numerous previous studies [6, 31–33]. There was also no apparent interaction between birth weight and exposure to the occupation, suggesting that this had not masked any underlying association between birth weight and CVD. However, in many of the previous studies examining the relationships between birth weight, CVD and associated risk factors, these relationships were only apparent after adjusting for one or more measures of current body size – an analytical approach which can invoke a statistical artefact known as the 'reversal paradox' [34, 35]. Although it was not possible to adjust for current body size in the present study, it would not have been appropriate to do so had data on current body size been available. Indeed, the absence of a significant negative relationship between birth weight and CVD in the present study adds to concerns that previous reports of such a relationship may be partly artefacts of the 'reversal paradox'.

Meanwhile, the results of both the first and second sets of models (see Tables 1 and 2) seem to provide clear evidence that exposure to the 1940–45 Channel Islands occupation was associated with an increased likelihood of hospital admission for CVD in later life. Differential exposure to undernutrition and related deprivation in early life between those who remained in Guernsey during the 1940–45 occupation and those who left or were evacuated before the occupation began [16, 27] is one plausible explanation for these findings. Likewise, it is possible that the stress associated with the occupation may have had a number of permanent psychosocial effects that led to a differential risk of CVD in later life. However, those cohort members who were evacuated from the islands before the occupation began often left their families behind and were sent in large groups to unfamiliar towns in the UK, many of which experienced sustained bombing during the war. It therefore seems likely that both resident and evacuated islanders would have experienced some form of stress, whilst only those who remained on the island would have experienced the undernutrition and related deprivation associated with the 1940–45 occupation.

It is nonetheless possible that these findings reflect residual confounding for pre-existing sociodemographic differences between islanders resident in Guernsey throughout the 1940–45 occupation and those who left the island or were evacuated before the occupation began [30]. Given that the analyses presented here had limited access to sociodemographic variables (such as parish of residence and paternal occupation at birth), it seems likely that these analyses will have suffered from residual confounding. However, it is reassuring that the increase in CVD amongst cohort members resident on the island during the 1940–45 occupation remained statistically significant when paternal occupation was recoded as 'agricultural' vs. 'non-agricultural' and in the sensitivity analyses which assumed that cohort members missing from the HES database had not experienced CVD.

We are therefore confident that the results of these analyses strengthen the evidence provided by two previous studies of cohorts exposed in childhood to nutritional deprivation during the Leningrad siege, which found that cardiovascular mortality and systolic blood pressure were significantly elevated among those exposed at the age of 9–15 years [15], while men exposed at the age of 6–8 years also had increased mortality from ischaemic heart disease [8]. While these are similar findings to those observed in the present study, it is nonetheless important to note that conditions during the 1941–44 Leningrad siege were much more severe than those in Guernsey during the 1940–45 occupation and siege, with widespread starvation of the civilian population [36]. Nonetheless, one explanation for the Leningrad studies' findings was that children over the age of 12 appeared to have received disproportionately fewer rations than younger children, and a similar scenario has been described in accounts of the 1940–44 German occupation of Belgium [37] and in accounts of rationing from the 1940–45 occupation of the Channel Islands, which suggest that the young and the elderly suffered most [26, 27, 38]. Indeed, as in Leningrad, some accounts from the Channel Islands describe a scenario in which those in their early teenage years bore the brunt of inadequate food supplies during the occupation because those over 12 years old were often expected to work yet did not receive proportionately higher rations [39].

Moreover, there is further evidence that children and adolescents might have experienced longer-term developmental delays as a result of the Channel Islands siege, in the form of delayed age at menarche [23] and slower growth rates [22]. In particular, a study of Jersey schoolchildren during the occupation found that children who remained on the island displayed significantly slower growth compared to children in the UK [22]. While the lower rates of weight gain were temporary, the lower rates of height gain appeared longer lasting, indicating that exposure to the occupation may have had some longer-lasting developmental effects. Similar patterns of impaired growth were observed following the 1940–44 German occupation of Belgium, where girls who reached puberty during this time grew up to be shorter and lighter adults [37]. Likewise, in a study of women exposed to the Dutch famine, those exposed at the ages of 0–9 and 12–16 were found to have shorter stature as adults [40]. It is therefore possible that these effects on the stature of children and adolescents may reflect the developmental mechanism(s) through which exposure to the 1940–45 occupation of the Channel Islands might have led to an increased risk of CVD in later life.

This interpretation is supported by the significant interaction between exposure to the occupation and parish of residence at birth (used as a marker of parish of residence during the occupation), which indicated that exposure to the occupation was more strongly associated with CVD amongst those born/living in urban parishes where there was less access to off-ration foodstuffs [26, 27]. Authors studying the immediate health effects of the Dutch famine have found similar differences between those living in rural and urban areas, and also attributed these to disparities in food availability therein [41]. As such, the statistically significant interaction between exposure to the occupation and parish of residence at birth observed in the present study supports the conclusion that the differences in hospital admission for CVD were due to undernutrition and related deprivation during the 1940–45 occupation. The interaction between exposure to the occupation and parish of residence at birth also raises the possibility of either a linear relationship or threshold effect involving the severity of food deprivation during the occupation and subsequent CVD.