The effects of in-utero exposure to influenza on mental health and mortality risk throughout the life-course

Highlights

• Evidence on the effects of in-utero exposure to influenza at an individual level and on later-life mental health is limited

• We use unique mother-reported data on influenza during pregnancy coupled with lifecourse data on mental health and mortality

• Influenza exposure is associated with reductions in mental health in childhood for the full sample and in mid-life for males

• Influenza exposure is also associated with increased mortality, driven primarily by a 75% increase in the risk of stillbirth

Abstract

Studies examining the later-life health consequences of in-utero exposure to influenza have typically estimated effects on physical health conditions, with little evidence of effects on mental health outcomes or mortality. Previous studies have also relied primarily on reduced-form estimates of the effects of exposure to influenza pandemics, meaning they are unlikely to recover effects of influenza exposure at an individual-level. This paper uses inverse probability of treatment weighting and “doubly-robust” methods alongside rare mother-reported data on in-utero influenza exposure to estimate the individual-level effect of in-utero influenza exposure on mental health and mortality risk throughout childhood and adulthood. We find that in-utero exposure to influenza is associated with small reductions in mental health in mid-childhood, driven by increases in internalising symptoms, and increases in depressive symptoms in mid-life for males. There is also evidence that in-utero influenza exposure is associated with substantial increases in mortality, although these effects are primarily driven by a 75% increase in the probability of being stillborn, with limited evidence of additional survival disadvantages at later ages. The potential for mortality selection implies that estimated effects on mental health outcomes are likely to represent a lower bound.

Introduction

Mental health disorders account for 13% of the global disease burden (Collins et al., 2011) and are the second leading cause of years lived under disability worldwide (Ferrari et al., 2017). It is also estimated that mental health problems are associated with a 4% reduction in GDP across 28 European Union countries, driven by increases in direct spending on healthcare, greater spending on social security programmes, and lower rates of employment and in-work productivity (OECD, 2018). Approximately half of mental disorders in adults begin before the mid-teenage years (Kessler et al., 2007, World Health Organization, 2013). Given this, a well-documented upward trend in child and adolescent mental health problems is worrying (Collishaw, 2015, McManus et al., 2019, Sadler et al., 2018). Understanding the determinants of mental health problems at multiple stages of the life-course is therefore vital.

A large body of literature examines how health shocks occurring in the prenatal (in-utero) period may impact the development of disease in later life (Almond et al., 2018, Currie and Almond, 2011). However, this literature has focused almost solely on physical health problems, with limited, although growing, evidence on how in-utero shocks may impact mental health (Currie, 2020). Studies assessing effects on mental health have typically identified mental health problems using mental health-related hospital admissions (Brown et al., 2000) or drug prescriptions for mental disorders (Persson and Rossin-Slater, 2018), reducing their ability to detect effects on milder mental health symptoms and to capture problems faced by individuals not seeking treatment or not having access to healthcare.

Stemming from the seminal work of Almond (2006), exposure to influenza has emerged as one of the most widely-examined markers of the in-utero environment. Influenza is highly prevalent, infecting 5–20% of the population in the seasonal influenza season each year, and has significant effects on both mortality and morbidity (Cassini et al., 2018, Iuliano et al., 2018), particularly for pregnant women (Campion et al., 2014, Jamieson et al., 2006). A large number of studies have established effects of exposure to influenza in-utero on the prevalence of diabetes (Almond and Mazumder, 2005, Garthwaite, 2008, Lin and Liu, 2014), cardiovascular diseases (Garthwaite, 2008, Mazumder et al., 2010), metabolic and kidney diseases (Garthwaite, 2008, Lin and Liu, 2014), respiratory problems (Lin and Liu, 2014), and the incidence of strokes (Almond and Mazumder, 2005).¹ However, there is currently no evidence on how influenza exposure may impact the development of mental health problems. This is surprising given several hypothesised mechanisms linking the in-utero environment to these outcomes (Schlotz and Phillips, 2009). This includes hypothermia and inflammation which may directly damage the foetal brain during its early stages of development (Rasmussen et al., 2008) and indirect damage to the brain through under-nutrition (Colombo et al., 2004, Gale et al., 2008, Parsons et al., 2008), driven by influenza suppressing appetite, interfering with the absorption of nutrients such as fats and proteins, and through fever-induced increases in the rates of nutrient loss and energy consumption (Tomkins et al., 1994).

Previous studies are also limited by being unable to examine the impacts of in-utero exposure to influenza at an individual level. Long-run effects on offspring have primarily been examined by studying the impacts of influenza pandemics, making use of temporal and/or geographical variation in pandemic intensity under the assumption that this variation in influenza cases is randomly assigned. However, a drawback of these empirical strategies is that in-utero exposure to an influenza pandemic, or exposure to a high-virulence area, is not equivalent to in-utero exposure to influenza itself. Even in areas where the virulence of a pandemic is high, only a fraction of pregnant women will contract influenza. Also, given that influenza is still a prevalent disease even in non-pandemic periods, some pregnant mothers will contract influenza even in areas where the severity of the pandemic is low or even non-existent. As a result, reduced-form effects of pandemic exposure will not reflect consistent estimates of the average causal effect of in-utero exposure to influenza on later-life outcomes.

Schwandt (2017) uses individual-level data in a sibling fixed-effects design to study the effects of seasonal influenza on health at birth and labour market outcomes. However, influenza cases are identified using admissions to hospital for influenza-like illnesses, meaning cases which are not severe enough to warrant hospital attendance are not recorded, and so mothers with these milder symptoms will be incorrectly identified as not contracting influenza. Estimates may therefore not test the effects of in-utero exposure to influenza versus no exposure, but the effect of being exposed to particularly severe cases of influenza compared to not contracting influenza or contracting a form of influenza with milder symptoms.

This study addresses these limitations using data from the 1958 National Child Development Study (NCDS); a rare dataset which collects longitudinal information on mental health symptoms of a cohort of individuals in both childhood and adulthood, and also includes mother-reported information on a range of diseases contracted during pregnancy. As data on influenza exposure is self-reported and mental health outcomes are collected via validated scales rather than recorded diagnoses, milder cases of influenza and mental health problems can be detected. Importantly, the NCDS follows all children in Great Britain born from 2nd to 9th March 1958, meaning that mothers of the NCDS cohort were pregnant during the end of the 1957 Asian Flu epidemic, which reached its peak around June when 95% of the NCDS cohort were between 16 and 25 weeks into gestation.² Throughout the 1957 epidemic in the UK, there was evidence that a large proportion of influenza cases were concentrated amongst young adults, and therefore not necessarily concentrated among the old and less-healthy (Clarke et al., 1958, Ministry of Health, 1960, Vynnycky and Edmunds, 2008, Woodall et al., 1958). This has the potential to aid identification of the effects of in-utero influenza exposure estimated through a simple comparison of outcomes for exposed and non-exposed groups, as exposure is less likely to be correlated with (unobserved) markers of maternal health. We find evidence of this in our data, with mothers reporting having contracted and not contracting influenza during pregnancy being similar on markers of maternal health and socio-demographic characteristics.

However, to adjust for small remaining differences in characteristics of those exposed and not-exposed to influenza, we estimate the effects of in-utero exposure to influenza using inverse propensity of treatment weighting and “doubly-robust” inverse-probability-weighted regression adjustment methods to ensure mothers contracting influenza in pregnancy are identical in observed characteristics to those not contracting influenza.

In a further contribution, we use linked data on NCDS and administrative death records to examine whether in-utero exposure to influenza impacts mortality risk at different stages of the lifecourse. This allows an assessment of whether effects on mental health are likely to be affected by mortality selection, but also addresses the paucity of studies examining the later-life mortality effects of in-utero influenza. Fletcher (2019) finds that exposure to the 1918 Spanish Flu did not lead to increases in overall or cause-specific mortality, but uses only death information for individuals with data on education and labour market outcomes in adulthood, meaning impacts on mortality risk in early-life and how this risk changes as individuals age could not be examined.

Our results indicate that in-utero exposure to influenza is associated with small reductions in mental health in mid-childhood, driven by increases in internalising symptoms of mental health, and increases in depressive symptoms in mid-life for males. Findings are robust to the methods used to adjust for observed covariates, and tests indicate they are likely to be unaffected by attrition bias. We also demonstrate that, consistent with small differences in observed child and family background characteristics between exposed and non-exposed groups, estimates are stable to the removal of these characteristics from the covariate set, providing evidence that omitted variable bias is unlikely to be impacting results greatly. However, we find that in-utero influenza is associated with a significant reduction in survival rates, primarily driven by effects on the rate of stillbirths, suggesting effects on mental health may be impacted by survival bias. Estimating simple bounds for treatment effects under survival bias, we find that estimated mental health effects are likely to be conservative.

The remainder of this article is organised as follows. The data is described in Section 2. Section 3 outlines the empirical strategy. The main findings are presented in Section 4. Robustness checks are presented in Section 5. Section 6 is devoted to the discussion and concluding remarks.