- Top of page
Influenza can be especially serious for pregnant women.[1, 2] During the 2009 H1N1 pandemic, pregnant women experienced disproportionate hospitalization and mortality,[3-5] similar to reports from the 1918 and 1957 influenza pandemics.[6-8] Studies also have demonstrated an association between influenza virus infection and spontaneous abortion, preterm birth, stillbirth, and low birth weight.[5, 9-14] Furthermore, influenza can be serious for neonates and infants.[15, 16] While the acute effects of influenza on pregnant women and their children are well established, a growing literature is examining possible long-term consequences for prenatally exposed offspring.
The impact of prenatal exposure to maternal influenza virus infection has been evaluated using a variety of approaches. Numerous studies have examined the relationship between in utero exposure to the 1957 influenza pandemic and development of schizophrenia, with conflicting results.[17-22] Several researchers have examined birth cohort effects associated with the 1918 pandemic. A birth cohort study encompassing 24 countries found no evidence of increased late-life mortality among those exposed prenatally or shortly after birth to the 1918 pandemic. A study examining birth cohort effects among 60- to 82-year-olds born in the USA between 1915 and 1923 found that persons born in 1919, who would have been in utero during the 1918 pandemic, were more likely to report negative health outcomes (e.g., fair-to-poor health; trouble hearing) compared with those born in 1915–1918 and 1920–1923. A time-series analysis of prenatal exposure to the 1918 pandemic showed an increased rate of self-reported prevalent cardiovascular disease among American adults born in 1919.
This study was designed to examine whether having been in utero during an influenza pandemic, a proxy for prenatal influenza exposure, is associated with an increased risk of physician-diagnosed acute myocardial infarction (MI) or stroke. We used Danish surveillance data to identify months when influenza activity was highest during three previous pandemics (1918, 1957, and 1968) and defined persons as exposed or unexposed based on birth month and year.
- Top of page
We identified 1 530 189 persons who lived in Denmark as of April 2, 1968, when the Civil Registration System was established. Table 1 shows the number of persons and amount of person-time contributed in the exposed and unexposed groups for each pandemic period. The average number of person-years contributed per person was similar for exposed and unexposed persons within each period.
Table 1. Characteristics of the study population
|1918||Total (n) before restrictions||Exposed||Unexposed|
|Born 1915–1922: 470 390a||Male||Female||Total||Average person-years of follow-up||Male||Female||Total||Average person-years of follow-up|
|Acute MI||158 885||54 315||59 101||113 416||16·49||21 505||23 934||45 439||16·53|
|Stroke||160 216||55 161||59 184||114 345||16·27||21 930||23 941||45 871||16·28|
|1957||Total (n) before restrictions||Exposed||Unexposed|
|Born 1954–1960: 517 719a||Male||Female||Total||Average person-years of follow-up||Male||Female||Total||Average person-years of follow-up|
|Acute MI||222 718||33 880||32 173||66 053||25·25||79 897||76 768||156 665||25·22|
|Stroke||222 671||33 876||32 164||66 040||25·20||79 882||76 749||156 631||25·17|
|1968||Total (n) before restrictions||Exposed||Unexposed|
|Born 1966–1972: 542 080a||Male||Female||Total||Average person-years of follow-up||Male||Female||Total||Average person-years of follow-up|
|Acute MI||283 749||66 407||62 767||129 174||25·48||79 238||75 337||154 575||25·45|
|Stroke||283 727||66 401||62 764||129 165||25·46||79 233||75 329||154 562||25·42|
Figure 1 shows the data used to identify exposure groups. The date ranges used to define exposed and unexposed groups for each pandemic are provided in Table 2.
Table 2. Birth date ranges used to define exposure, by influenza pandemic
|1918||October 1918–November 1920||March 1921–December 1921|
|1957||October 1957–August 1958||November 1958–December 1960|
|1968||January 1969–October 1970||December 1970–December 1972|
Among persons born after the peaks of the pandemics, rates of MI and stroke were plotted by birth month and by birth-month quarter to assess presence of seasonality for the outcomes under study. Seasonality was not visually evident in any of the cohorts born after the three pandemics. For birth dates from January 1959 through December 1960, the peak-to-low ratios for acute MI and stroke were 1·10 (95% CI 1·00, 1·26) and 1·11 (95% CI 1·01, 1·23), respectively. For birth dates from January 1971 through December 1972, the peak-to-low ratios for acute MI and stroke were 1·04 (95% CI 1·00, 1·52) and 1·10 (95% CI 1·00, 1·31), respectively. We did not include adjustment for seasonality in our models. For all three pandemics, the crude, sex-stratified, and sex-adjusted IRRs and 95% CIs were generally very similar for acute MI and stroke (Table 3). There was no apparent difference in the rate of acute MI for those exposed prenatally to influenza compared with those unexposed during the 1918 and 1957 pandemics. The sex-adjusted rate ratios for 1918 and 1957 were 1·02 (95% CI: 0·99, 1·05) and 0·96 (95% CI: 0·87, 1·05), respectively. For 1968, the adjusted IRR was 1·18 (95% CI: 0·96, 1·45).
Table 3. Crude and adjusted incidence rate ratios for (A) prenatal influenza exposure and acute myocardial infarction and (B) prenatal influenza exposure and stroke
| ||Crude IRR (95% CI)||Sex-adjusted IRR (95% CI)||Cases (n)|
|1918||1·04 (1·00, 1·08)||0·98 (0·93, 1·02)||1·02 (0·99, 1·05)||1·02 (0·99, 1·05)||15 940||6288|
|1957||0·94 (0·89, 1·05)||1·02 (0·85, 1·22)||0·96 (0·87, 1·05)||0·96 (0·87, 1·05)||597||1475|
|1968||1·24 (0·97, 1·59)||1·05 (0·72, 1·54)||1·18 (0·96, 1·45)||1·18 (0·96, 1·45)||180||182|
|1918||0·99 (0·96, 1·03)||0·99 (0·96, 1·03)||0·99 (0·97, 1·02)||0·99 (0·97, 1·02)||22 759||9211|
|1957||1·05 (0·96, 1·15)||0·92 (0·83, 1·01)||0·99 (0·92, 1·05)||0·99 (0·92, 1·05)||1223||2940|
|1968||0·86 (0·75, 0·99)||0·84 (0·73, 0·96)||0·85 (0·77, 0·94)||0·85 (0·77, 0·94)||709||997|
There was no indication of increased cerebrovascular stroke risk for persons prenatally exposed to the 1918 and 1957 influenza pandemics compared with the unexposed; sex-adjusted IRRs were 0·99 (95% CI: 0·97, 1·02) and 0·99 (95% CI: 0·92, 1·05), respectively. For 1968, the adjusted IRR was 0·85 (95% CI: 0·77, 0·94).
The results of the analyses to evaluate the potential bias due to misclassification of exposure to ILI in utero generally yielded ‘corrected’ odds ratios that were very similar to the observed rate ratios for all pairwise combinations of sensitivity and specificity, for both MI and stroke outcomes, and in all three cohorts (Table 4). This result indicates that, assuming an accurate bias model, non-differential and independent exposure misclassification is unlikely to explain the near null associations that we have reported. At the low end of the range for specificity (0·6) for the 1968 cohort, corrected odds ratios were further from the null. This is likely explained by the sparse number of outcomes, which can unduly influence this type of bias analysis.
Table 4. Bias analysis with corrected odds ratios for non-differential, independent exposure misclassification for the (a) 1918 pandemic and acute myocardial infarction, (b) 1918 pandemic and stroke, (c) 1957 pandemic and acute myocardial infarction, (d) 1957 pandemic and stroke, (e) 1968 pandemic and acute myocardial infarction, and (f) 1968 pandemic and stroke
- Top of page
In this study, we found generally no evidence of an association between prenatal influenza exposure and acute MI or stroke in later life. This was true across the pandemics, although the results for the 1968 pandemic suggested a slightly protective effect for stroke.
A major advantage of our study was the availability of person-time for every Danish resident surviving to approximately 1977, starting from the birth years of interest. We based exposure definitions on surveillance data from a national source and the periods of high activity identified in this dataset correspond well with other published reports on pandemics in Denmark and other parts of Europe.[9, 31, 36]
Being in utero during a peak influenza period is a proxy for actual prenatal exposure. Thus, there is certain to be exposure misclassification for both the exposed and unexposed groups. However, we improved the exposure classification by excluding persons who would have been in utero during ‘buffer’ months, when ILI was at varying levels in the months immediately before and after the peaks. The exposure misclassification is likely non-differential and independent, and the results of our bias analysis were generally consistent with our observed data.
Use of the DNRP to identify the outcomes of interest has both strengths and limitations. While there is clear benefit to using physician-diagnosed events, misclassification (albeit likely non-differential) is a concern, particularly for stroke diagnoses. Two research groups have demonstrated that the quality of the stroke and TIA data in the DNRP is low for some diagnostic subgroups and the data should be used with caution.[37, 38]
The potential for healthy survivor bias, particularly in the 1918 pandemic analysis, must be considered. In Denmark, there was a decrease in the birth rate in 1919, followed by an increase in 1920. This trend has been noted in other European countries that, like Denmark, were not heavily involved in World War I. In 1919, Harris reported an increased risk of miscarriage or premature birth in pregnant women who had influenza complicated by pneumonia during the 1918 pandemic. The results of a recent study provide compelling evidence that the birth rate decline in Denmark in 1919 was caused by miscarriages among women with influenza during their first trimester. If many of the fetuses exposed to the 1918 pandemic were miscarried, and there is indeed an association between prenatal influenza exposure and the outcomes of interest, this effect would not be apparent in our analyses. In addition, any perinatal mortality occurring as a direct result of prenatal or perinatal infection, or as an indirect result of prenatal infection via premature birth, could also contribute to a healthy survivor effect. Unfortunately, perinatal mortality data are not readily available. Birth rate declines were not apparent around the time of the 1957 and 1968 pandemics, so we do not believe survivor bias is present in analyses for those pandemics.
Another important factor related to the 1918 pandemic analysis is left truncation of the outcomes. The DNRP was established in 1977, so we lacked information on MI or stroke events occurring before that year. The 1918 pandemic-related cohorts were in the prime age for cardiovascular disease during the post-World War II period when the developed world experienced an epidemic of cardiovascular disease. Because this occurred before 1977, these outcomes could not be included in the present study. Danish cause-of-death data are available in hard-copy form from 1943 on and could provide the basis for an important follow-up to the analysis reported here. At the same time, persons included in the 1957 and 1968 pandemic analyses are still relatively young for the outcomes of interest in this study. However, while it is true that the risk for some of these outcomes is low in the ages attained by the exposed cohorts, it should also be noted that low-risk populations have been endorsed for studies of weak associations.
Data are unavailable on variables such as socioeconomic status and other variables of interest such as birth weight (the Danish Medical Birth Registry began in 1973). While such factors may be important for the question at hand, and may be effect modifiers, we are not very concerned with residual confounding in this study because there do not appear to be factors that would be associated with both birth year and the outcomes of interest. Finally, any changes to the health system or to health habits over time would have affected the exposed and unexposed nearly uniformly. The unexposed cohorts were born within a couple years of the exposed cohorts, so these social and healthcare changes would have impacted both nearly identically.
In the 1980s and 1990s, Barker and colleagues articulated what has come to be known as the fetal origins hypothesis. The hypothesis states that adverse prenatal conditions can ‘program’ the fetus and cause physical and metabolic changes during key developmental stages that result in increased risk of certain conditions such as hypertension and diabetes. There is accumulating evidence regarding environmental exposures and the development of cardiovascular disease, including prenatal exposures. Congenital cardiovascular abnormalities have been associated with maternal HIV infection, maternal diabetes, maternal febrile illness, certain drugs and chemicals, and deficiencies in particular nutrients during gestation. In addition, low birth weight for gestational age has been associated with adult health outcomes including acute myocardial infarction. The impact of prenatal exposure to maternal influenza infection on offspring has been evaluated by others. There is evidence from rodent studies, suggesting that prenatal exposure to cytokines, specifically IL-6 which is produced in response to influenza and other inflammatory processes, is associated with hypertension in adult offspring.[44, 45] As discussed earlier, the relationship between in utero exposure to the 1957 influenza pandemic and development of schizophrenia has been studied,[17-22] and several groups have examined birth cohort effects associated with the 1918 pandemic.[23-25]
Mazumder et al. reported an increase in cardiovascular disease among Americans born in 1919 compared with those born in the years immediately before (1915–1918) and after (1920–1923). While that study examined self-reported prevalent cardiovascular disease among persons aged 60–82 years, our study examined registry-recorded events that occurred during ages 62–87 for the 1918 pandemic-related cohorts. These differences in study design could partly explain the difference in results.
In conclusion, we did not observe an association between prenatal influenza exposure and either acute MI or stroke in the present study. The results are generally consistent across the three pandemics. The pandemics allowed us to consider the association of interest with varying age ranges for incident events as well as with different influenza strains (H1N1 in 1918, H2N2 in 1957, and H3N2 in 1968), adding strength to our finding of a uniformly null association. We have performed a fairly comprehensive evaluation of the associations of interest over different age ranges, although survivor bias and left truncation of the outcomes for the 1918 pandemic may have impacted our findings. The current relatively young ages of the persons included in the 1957 and 1968 analyses may warrant re-evaluation at a later time.