Genetic and environmental influences on quality of life: The COVID‐19 pandemic as a natural experiment

Abstract By treating the coronavirus disease 2019 (COVID‐19) pandemic as a natural experiment, we examine the influence of substantial environmental change (i.e., lockdown measures) on individual differences in quality of life (QoL) in the Netherlands. We compare QoL scores before the pandemic (N = 25,772) to QoL scores during the pandemic (N = 17,222) in a sample of twins and their family members. On a 10‐point scale, we find a significant decrease in mean QoL from 7.73 (SD = 1.06) before the pandemic to 7.02 (SD = 1.36) during the pandemic (Cohen's d = 0.49). Additionally, variance decomposition shows an increase in unique environmental variance during the pandemic (0.30–1.08), and a decrease in the heritability estimate from 30.9% to 15.5%. We hypothesize that the increased environmental variance is the result of lockdown measures not impacting everybody equally. Whether these effects persist over longer periods and how they impact health inequalities remain topics for future investigation.


| INTRODUCTION
Natural experiments pose a particularly interesting set of circumstances where an intervention is implemented that is not under the control of researchers. 1 With respect to research in the domain of public health and human behaviour, a great advantage of research on natural experiments is that it corresponds to 'real world' conditions, in contrast to many controlled experiments. Additionally, natural experiment studies are essential for evaluating population-scale (health) interventions and changes where experimental manipulation or random allocation is not feasible. As a result, natural experiments can provide unique ecologically valid insights into health processes as they are naturally occurring. A well-known example of a population-level natural experiment is the compulsory schooling age reform in the United Kingdom, where the minimum age at which students were allowed to leave school increased from 15 to 16 for everyone born on or after 1 September 1957. An interesting finding in the context of this reform is that the additional year of education reduced the gap in unhealthy body size between those in the top and bottom terciles of genetic risk for obesity from 20 to 6 percentage points, thus benefitting those with a higher genetic risk for obesity. 2 Another interesting set of natural experiments is the introduction of national tobacco control policies in different countries. For example, a workplace smoke-free legislation was introduced in Ireland in March 2004. One of the results of this legislation was an immediate significant reduction in small-for-gestational birth rates, which was sustained over the postban period. 3 In the Netherlands, smoking prevalence decreased from 40%-51% to 22%-23% between 1993-1995 and 2009-2010, but no effect was seen on the heritability estimates of smoking. 4 These examples involve national-level policy changes aimed at improving population health. Another set of natural experiments is (natural) disasters with population-level consequences. For example, on March 11, 2011, Japan was struck by an earthquake and consequent tsunami, leading to the loss of ±18,500 lives and ±345,000 people suffering damages to (or loss of) their house 5 and many people suffered from posttraumatic stress disorder (PTSD) after this disaster.
Hikichi and colleagues studied these events from a natural experiment perspective in order to gain knowledge on the association between social cohesion and the risk for PTSD. 5 They found that individualand community-level social cohesion before the disaster were associated with a lower risk of showing PTSD symptoms following the disaster. Another disastrous event with population-level consequences was World War 2 (WW2). During the horrific events of WW2, many children were separated from their parents. Pesonen and colleagues 6  March 2020, the World Health Organization (WHO) officially declared a pandemic, as the virus spread quickly across many countries in the world. As a result, many countries enforced a lockdown with varying levels of regulations. In the Netherlands, a so-called 'intelligent lockdown' was installed, meaning that public spaces, schools, restaurants, and so forth were closed and that people were encouraged to work from home, but could still leave their house for walks and other outdoor activities. As a result, many people's lives changed profoundly from an economic, social, and physical perspective.
What these different aspects (economic, social, physical) have in common is that they are all related to mental health and well-being. In a meta-analysis by Prati and Mancini, 7 the psychological impact of the COVID-19 pandemic lockdowns across 25 studies was evaluated in terms of both positive and negative psychological functioning. They found that lockdowns had a small but detrimental effect on mental health, as expressed in negative psychological functioning (i.e., anxiety, depression, substance use, sleep disturbances, suicide risk, negative affect, and general distress), but surprisingly the effects on positive psychological functioning were not significant. In a Dutch sample, specifically people without severe or chronic mental health disorders showed a slight increase in depression, anxiety, worry, and loneliness symptoms, whereas people with depressive, anxiety, or obsessive-compulsive disorders did not seem to have increased symptom severity during the pandemic compared with before. 8 Besides the effects of this large natural experiment on mean population levels of mental health, such an impactful natural experiment enables a unique study into causes of individual differences in mental health.
From a behaviour genetic perspective, the focus goes beyond mean level changes to explain the causes of individual differences. It is well established that individual differences in well-being are influenced by both environmental factors and genetic factors: research indicates that about 40% of individual differences in wellbeing is explained by genetic factors (the heritability), with the other 60% being explained by non-shared/unique environmental factors. 9 Research combining behaviour genetics and experiments is relatively scarce, and typically focuses on short-term interventions. For example, one might use the 'method of co-twin control', where only one member of an identical twin pair receives an intervention. 10 This is an interesting way of studying the possible effect of the intervention while controlling for genetic confounding. Alternatively, we can study individual differences in the effect of an intervention by applying an intervention in a classical twin design (CTD). This design also provides information on stability and change of the sources of individual differences pre-and post-intervention. For example, Haworth and colleagues examined the influence of a 10-week positive psychology intervention on well-being in a sample of 750 twins, and found that the relative influence of genetic and environmental influences remained stable, but that (partly) different non-shared environmental factors influenced well-being post-intervention. 11 In a more recent study, a brief online mindset intervention increased the relative influence of additive genetic factors to individual differences in mindset. 12 The COVID-19 pandemic can serve as a natural experiment for the investigation of absolute and relative changes in the genetic and environmental causes of variation in well-being since we can compare the variance decomposition during the pandemic to before the pandemic.
For two well-being related constructs, optimism and meaning in life, it was already found that the heritability during the pandemic was slightly lower compared with before the pandemic. 13 In addition to estimates of quantitative change such as lower heritability estimates, a study focusing on the qualitative aspects of the psychological responses to the COVID-19 crisis in young adults found a genetic correlation of 1 between pre-pandemic and pandemic purpose in life, indicating that the same genes affect this trait before and during the pandemic. 14 Optimism and meaning in life can be viewed as facets of well-being, 15 but whether these effects are similar for other wellbeing measures, such as quality of life (QoL), remains unexplored.
In the present study, we explore the impact of the COVID-19 pandemic on individual differences in well-being, quantified as QoL, in the Netherlands. We use a unique dataset that is comprised of data from twin families (e.g., twins, siblings, parents, aunts, uncles, nephews, nieces: pedigree data) both before and during the pandemic to provide a useful account of how genetic and environmental influences may be impacted by substantial environmental change.

| Participants
Participants were voluntary registrants of the Netherlands Twin Register (NTR). 16 NTR participants are recruited through birth felicitation services, city councils, and online platforms. Every couple of years, biological and non-biological family members are invited to partake in survey research on development, health, behaviour, and lifestyle.
Relations among participants, that is, pedigree structure information, is stored in the 'Person Administration of the Netherlands Twin Register' (PANTER) database. 17 Within this database, family roles and relations (e.g., mother-offspring, sibling-sibling) among participants are stored, with unlimited one-to-one relation possibilities for each individual. Participants can have multiple roles and relations in the database. For example, a person can be a mother and a twin.
For the current project, we selected a sample with pre-pandemic QoL data, and a (partly overlapping) sample with pandemic QoL data.
All participants were 16 years or older. For the pre-pandemic sample, QoL data were available for multiple waves of data collection. If multiple observations were available for an individual, we selected the most recent pre-pandemic observation (assessment data between January 2014 and February 2020). Within each family, if data for multiple siblings were available, we only selected data collected in the same data collection wave, in order to reduce potential timedependent confounders. Additionally, if data from both parent or spouses were available, we selected their data such that the data from both parents/spouses were included from the same wave of data collection.
During the pandemic, we made use of a single wave of data collection, which took place in April and May 2020, during the first lockdown in the Netherlands. Because we were interested in the effects of the lockdown on genetic and environmental influences on QoL, and not the effect of being infected itself, we excluded individuals with an (expected) COVID-19 infection (see below for details). We included twins and higher-order multiples (e.g., triplets), parents, siblings, and spouses (of multiples). Nuclear family information and age per type of family member is presented in Table 1. In total, prepandemic QoL data were available for 25,772 individuals, and pandemic QoL data were available for 17,222 individuals, of whom 11,232 had data available at both time points. Across the whole sample, age ranged between 16 and 102. Figures S1 and S2 visualize the pre-pandemic and pandemic age distributions, respectively.

| Quality of life
Well-being was assessed as QoL using a Dutch version of Cantril's Self-Anchoring Striving Scale. 18 Participants were asked the question: 'Where on the scale would you place your life in general? A score of 10 means the best life you can imagine, 1 means the worst life you can imagine'. In one of the pre-pandemic questionnaires, the question was scored on a scale from 0 to 10 instead of 1 to 10. As almost no participant scored a 0 (N = 6) or 1 (N = 3) on this question, these two answers were pooled together as one so that the question was scored similarly from 1 to 10 across the different questionnaires.  19,20 to predict whether a person likely had COVID at the time of assessment. Detailed information on the development and application of this variable can be found in the original study paper. 20

| Pre-pandemic to pandemic comparison
Means and standard deviations for pre-pandemic and pandemic QoL for individuals with different roles within families were calculated using R. 21 We selected a subsample of genetically unrelated individuals (n = 8529) and performed a paired-samples t test to examine if QoL significantly changed from before to during the pandemic. Effect sizes were calculated using Cohen's d for paired samples. Additionally, we calculated within-individual difference scores that reflect individual change from before to during the pandemic.

| Kinship correlations
Kinship correlations were obtained to acquire a first indication of familial resemblance for QoL during and before the pandemic. We cal- were cross-time correlations. These last two correlations between pre-pandemic and pandemic QoL were pooled with fixed effect metaanalysis in the meta package in R so that one cross-phenotype correlation is computed to be used for further interpretation. As the tool does not provide standard errors or confidence intervals (CIs), these were calculated manually (s r ¼ ffiffiffiffiffiffiffi ffi 1Àr 2 nÀ2 q ).

| Genetic analyses
We used the Mendel 16.0 software package 'Variance Components' analysis option 22 to decompose (co)variation in QoL into additive genetic (A), dominant genetic (D), common/household environmental (C), and unique environmental (E, also includes measurement error) sources of (co)variation. Effects of age and sex were regressed out prior to the Mendel analyses, and subsequent analyses were conducted on the residual QoL scores. 23 Shared environmental influences were defined as influences that are shared by members of the same household. As we are examining adults only, most (adult-aged) children within a nuclear family will not live in the same household.
Therefore, a household effect was specified for spouses.
To perform variance decomposition in Mendel, three input files are required: 1. an input pedigree file, 2. a control file, and 3. a definition file.
1. The input pedigree file contains all the familial and phenotype data,  Table S1. 'sufficient' QoL before the pandemic (indicated by a 6 or higher), to 'insufficient' QoL during the pandemic (indicated by a 5 or lower).

Figures 2 and 3 depict the number of individuals and percentage
of individuals, respectively, that increased, decreased, and remained stable per QoL pre-pandemic score. As can be seen in Figure 2, prepandemic QoL scores are relatively skewed with most people indicating good pre-pandemic QoL. In general, the most common change was a decrease in QoL. Examining the group of respondents with decreased QoL during the pandemic in more detail (Figure 3), we see that individuals with high pre-pandemic QoL scores more often decreased during the pandemic compared with individuals with lower pre-pandemic QoL scores. With respect to the group of respondents that indicated increased QoL during the pandemic, it was especially individuals with lower pre-pandemic QoL scores that indicated higher scores during the pandemic. We also plotted the percentage of individuals that decreased, increased, or remained stable for QoL for different age groups separately in Figure S3. Visual inspection of the plot does not show large differences between the age groups, with only a very slight trend of younger individuals being more negatively impacted in terms of QoL. Pandemic QoL correlations were similar to or lower than prepandemic QoL correlations. As seen in Table 2

| Genetic analyses
The total phenotypic variance in   returned to more normal levels later on. It should be mentioned that some of the studies included in the reviews above also examined effects during the first lockdown. Therefore, it is likely that there are different effects across different countries, even during similar lockdown periods.
Important in the context of these results is that our sample, and the Netherlands in general, scores relatively high on QoL and other well-being measures compared with other countries. The Netherlands scores among the top happiest countries according to the 2019 World Happiness Report, 30 which was unchanged in the 2021 World Happiness Report that reported on the data collected in 2020 (during the pandemic). 31 Importantly, we found a pandemic average QoL of 7.02 which is significantly lower than the pre-pandemic average, but still a good score indicating that people were still quite satisfied with their QoL. We found that especially those with higher QoL scores were prone to decreases in QoL during the pandemic. Given the skewed distribution of QoL in our sample, this was the majority of our sample.
Increases in QoL, however, were found mostly for individuals with lower QoL scores. While this is a relatively small part of our sample, it was surprising that it was especially individuals with lower baseline QoL that showed improvements in QoL during the pandemic. A potential explanation is that we only examined individuals that did not have a COVID-19 infection around the time of assessment. Individuals with low levels of baseline QoL might have evaluated their QoL differently during the pandemic as they started comparing themselves with others that did become ill. In this way, they might have altered their perception, causing them to provide a different judgment during the pandemic. 27 Individuals with higher levels of baseline QoL, on the other hand, might not have focused on these kinds of comparative mechanisms because they did not think their QoL was worse than average to begin with. Importantly, another possibility is that these findings might (partly) result from regression to the mean (RTM), the phenomenon whereby the second assessment of a trait results in values closer to the mean than at initial assessment purely by chance.
However, pre-pandemic QoL was measured on multiple occasions for some individuals, in which case we chose the latest available timepoint. By selecting participants from different measurement occasions, we attempt to acquire a better estimate of the participants' true baseline mean, which in turn decreases RTM. 32 In a way, our prepandemic QoL measure is formed by taking a random sample around an individual's baseline mean levels. This makes it less likely that high/ low scorers will inevitably go down/up at the next measurement occasion (i.e., during the pandemic).Therefore, while we cannot rule out regression to the mean completely, this does make it less likely that our results are (fully) attributable to this phenomenon.
We used Mendel, instead of the CTD, to decompose the variance into genetic and environmental sources of variation. In the CTD, the variance components are estimated based only on the MZ and DZ twin covariances. As a result, only three parameters can be estimated simultaneously, so that an a priori choice needs to be made between an ACE or ADE model. The advantage of the Mendel software is that it allows for efficient analysis of whole pedigree data, allowing us to examine a large sample and estimating A, C, D, and E simultaneously.
There are extensions of twin designs where other family members can be included, such as the Cascade model 33 that are more flexible in terms of model specification (e.g., constraining paths and sex-specific heritability). However, the advantage of Mendel is that it easily allows for the inclusion of complex family relations and irregular pedigrees, as are present in large twin-family registers like the Netherlands Twin Register. Yet, we did not find any evidence for dominant genetic effects (D), that is, alleles acting in a multiplicative fashion (dominance or epistasis). Based on the correlations between the different types of family members ( It is important to interpret these results within the context of our sample and the time-frame in which we collected the data. Since it was the first lockdown, when the WHO had just announced a pandemic, individuals were likely still psychologically adjusting to the new situation. Whether the effects found in this study would be similar in later stages of the pandemic is a question that remains to be answered. Additionally, different countries employed different strategies to contain the virus, with the Netherlands installing the intelligent lockdown where people were encouraged to stay at home, but were still allowed to freely move around outside at all times of day. In this light, the finding of the large increase in environmental variance is even more remarkable, as the regulations in the Netherlands were less strict than those in many other countries. As such, the environmental effects of more stringent lockdown measures may be even larger. In any case, it is reasonable to expect that countries with different regulations will find different results than presented here, as these regulations impact the extent to which people had to alter their lives. Finally, a limitation of our sample was that we had more female respondents than male respondents in both the pre-pandemic sample (65% female, 35% male), and the pandemic sample (71% female, 29% male). The representativeness was further limited by there being roughly twice as many highly educated individuals in the sample than expected based on the Dutch population.

| CONCLUSIONS
In conclusion, in this study we used data from before and during the

ACKNOWLEDGMENTS
We would like to thank all the twins and their family members for participation in the Netherlands Twin Register.

CONFLICT OF INTEREST
The authors declare no relevant financial or non-financial interests to disclose.

ETHICS APPROVAL
All procedures performed in studies involving human participants

CONSENT TO PARTICIPATE
Written informed consent was obtained from all individual participants included in the study.