Are social isolation, lack of social support or loneliness risk factors for cardiovascular disease in Australia and New Zealand? A systematic review and meta‐analysis

Abstract Background An international systematic review concluded that individuals with poor social health (social isolation, lack of social support or loneliness) are 30% more likely to develop coronary heart disease (CHD) and stroke. Notably, the two included Australian papers reported no association between social health and CHD or stroke. Objective We undertook a systematic review and meta‐analysis to investigate the association between social isolation, lack of social support and loneliness and cardiovascular disease (CVD) incidence among people living in Australia and New Zealand. Methods Four electronic databases were systematically searched for longitudinal studies published until June 2020. Two reviewers undertook title/abstract screen and one reviewer undertook full‐text screen and data extraction. Quality was assessed using the Newcastle – Ottawa Quality Assessment Scale. Results Of the 725 unique records retrieved, five papers met our inclusion criteria. These papers reported data from three Australian longitudinal datasets, with a total of 2137 CHD and 590 stroke events recorded over follow‐up periods ranging from 3 to 16 years. Reports of two CHD and two stroke outcomes were suitable for meta‐analysis. The included papers reported no association between social health and incidence of CVD in all fully adjusted models and most unadjusted models. Conclusions Our systematic review is inconclusive as it identified only a few studies, which relied heavily on self‐reported CVD. Further studies using medical diagnosis of CVD, and assessing the potential influence of residential remoteness, are needed to better understand the relationship between social health and CVD incidence in Australia and New Zealand.


| INTRODUC TI ON
In Australia, cardiovascular disease (CVD) dominates as the greatest cause of Australia's morbidity, mortality and healthcare expenditure. 1 Similarly, in New Zealand CVD is a leading cause of premature mortality 2 and monopolises nearly a quarter of health expenditure on noncommunicable diseases. 3 In 2003, an Expert Working Group of the National Heart Foundation of Australia 4 concluded from a systematic review that there is strong and consistent evidence of an independent causal association between social isolation and the causes and prognosis of coronary heart disease (CHD). Yet our understanding of the link between social health and CVD is limited, especially compared to other key CVD risk factors highlighted by the National Heart Foundation of Australia, including elevated cholesterol or blood pressure, diabetes, significant family history, smoking, poor nutrition, physical inactivity, adiposity and depression. 5 Social health refers to an individual's ability to form satisfying and meaningful relationships, their ability to adapt in social situations, and their interactions with and support from other people, institutions and services. 6 The concepts of social isolation, social support and loneliness are often discussed in relation to social health. 7 Social isolation is "an objective state in which a person has minimal contact with others and low involvement in local community life". 8 Social support is a subjective measure of how social connections are operationalised, and loneliness "is a subjective, unwelcome experience of lack of or loss of companionship". 8 Assessment of social health varies and the inconsistency is a common limitation of this research area. 9,10 However, there is a helpful framework to compare and contrast common tools. 10 Furthermore, a recent theoretical debate has emerged to separate the concepts of social isolation, social support and loneliness, as they likely have different implications for health and well-being. 7 Before the COVID-19 outbreak, 50% of Australians reported feeling lonely at least 1 day a week, 25% reported currently experiencing an episode of loneliness, and 10% reported that they lack social support. 11 Prevalence rates seemed to be slightly lower in New Zealand; 36% reported feeling lonely at least 1 day a week and 14% reported feeling lonely all, most or some of the time. 12 A number of conceptual frameworks illustrate the pathways between social health and health, [13][14][15][16][17][18][19][20] with several 13,18,20,21 being particularly relevant to our review as they describe the link between poor social health and CVD. The main, broad pathway that tends to be described is from poor social health, through molecular mechanisms, health behaviours, and chronic disease risk-factors, leading to chronic mental and physical ill-health and mortality; with each step being impacted by socio-demographics, the sociological environment, and personality. 18,20,22,23 For example, being socially isolated or feeling lonely can overstimulate the body's stress response through increased levels of the stress hormone cortisol, raise blood pressure and decrease blood flow to vital organs through higher tonic vascular resistance, impair the immune system's ability to fight infections through lower production of white blood cells, and reduce sleep quality leading to less restorative sleep and daytime fatigue. 24 People with poor social health also tend to have more unhealthy lifestyles, such as undertaking less physical activity or eating unhealthily, which increases their risk of CVD. 25 Additionally, a bi-directional pathway is also described "with health and social relationships interacting to influence each other, in virtuous circles or spirals of despair". 14 The bi-directional pathway also incorporates the health selection model, which explains how deterioration in health (such as a CVD event or decline in cognitive functioning) may limit or reduce social involvement, which leads to greater ill-health. Hence, poor social health likely impacts health and well-being over the life-course.
In 2016, an international systematic review of 23 studies concluded that individuals with poor social health were 30% more likely to develop CHD and stroke. 26 The systematic review included 181 006 participants, aged 18 years and over, mainly from Europe (38%) or North America (33%), followed from 1965 to 1996 for 3-21 years. Notably, the systematic review only included two Australian papers, which reported that a combined measure of social isolation and social support was not associated with CHD or stroke in fully adjusted models. 27,28 No eligible studies were identified from New Zealand. Given the unique geographical spread of the Australian and New Zealand populations, along with differences in political and cultural support systems (especially in terms of social systems and health care), research undertaken in other countries may not be generalisable. With the rise of CVD in Australia and New Zealand, along with the emerging knowledge of the role of social health in CVD, it is important to further the understanding of these issues in order to better address and mitigate them.
The aim of this systematic review is to investigate the association of social isolation, lack of social support and loneliness with CVD incidence among people living in Australia and New Zealand.

| ME THODS
This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses 29 (PRISMA) statement. Our protocol was followed unless otherwise defined (Prospero CRD42018093503).

| Criteria
We included longitudinal observational data. We included studies conducted in the general population (not clinical location or for specific health reasons), generalisable to people living in Australia and New Zealand of all ages. We included the social health measures of social isolation, social support and loneliness. The primary outcome was incidence of CVD, obtained through self-report or medical records.

| Search methods
We searched four databases from the earliest record to 21st June 2020 (Appendix A). There were no language or date restrictions.
One author (AZZP) scanned the references of included studies, and several relevant review articles identified in the initial search, for additional studies and found one additional article. The first authors of each included paper were asked if they knew of any additional studies which might be relevant. No additional studies were supplied. 30-32

| Data collection and analysis
Papers were provided an ID based on the first author's last name and year of publication. Two people (authors or contributors Noria Akbari, Aghnia Naim) undertook an initial screening of titles and abstracts independently, with discrepancies included in the fulltext screen. There were no relevant non-English articles or conference abstracts. One author (RF) screened all full-text papers and performed data extraction. A second author (AZZP) checked the extracted data. Two authors (RF & AZZP) independently assessed the quality of included studies using the Newcastle -Ottawa Quality Assessment Scale, 33 with discrepancies resolved with independent assessment by a third author (FB). A score of eight or nine was deemed as a low risk of bias. In our protocol we stated the use of the STROBE Statement, 34 however it was developed as a guideline for reporting observational studies rather than assessing quality.

| Data synthesis
Meta-analysis followed the Cochrane Collaboration 35 guidance, with at least two studies required. We converted odds ratios 36 and hazard ratios 37 to relative risks. We initially misunderstood the available methodology, 38 and were unable to convert the findings from a continuous exposure to a categorical exposure (as recorded in our protocol). Statistical heterogeneity was evaluated by using the I 2 statistic 35 and the metan STATA command for the 95% confidence intervals, and funnel plots and Egger's test were used to assess publication bias (added since protocol). We intended to report by CVD type, gender, and partner status, dichotomous baseline data collection year, dichotomous length of follow-up and by country (Australia vs. New Zealand).

| Description of studies
Five papers met our inclusion criteria ( Figure 1, Table 1), which encompassed the two Australian studies included in the international systematic review. 26 All samples were part of large Australian longitudinal cohort studies of non-institutionalised adults. Three papers (Strodl 2003, 27 Strodl 2008 Byles 2015 39 ) used wave 2 data from the Australian Longitudinal Study on Women's Health (ALSWH) cohort born between 1921 and 1926 (aged 70 years or more at baseline). ALSWH was established in 1996 as a nationally representative cohort, with recruitment through Medicare records, people living in rural/remote double sampled and women too ill to participate excluded. Strodl 2003 27 andStrodl 2008 27,28 further restricted the sample by excluding prevalent CHD or prevalent stroke at baseline (wave 2), while Byles 2015 39 stratified by prevalent or incidence of stroke. The ALSWH collected the Duke Social Support Index (DSSI) which incorporates both aspects of social isolation and social support.
One included paper (Simons 2013

| Risk of bias
All studies were rated as high risk of bias, with NOS scores ranging between four and seven stars (Appendix C). All studies lost a point in the "Comparability" category for not adjusting for our list of most important factors. Most studies lost a point in the "Outcome" category as assessment of CVD was self-report and the follow-up was not long enough (defined as 5-year or more).

| Findings
The five included papers reported six CVD outcomes; two studies assessed CHD (Strodl 2003 27 andSimons 2013 40 ), three studies assessed stroke (Strodl 2008, 28 Byles 2015 39 and Simons 2013 40 ), and one study assessed heart disease (Sahle 2020 45 ). Social health was not associated with incidence of CVD in fully adjusted models and most unadjusted models.

| Meta-analysis
Methodological considerations were required prior to meta-analysis.
Strodl 2008 28 and Byles 2015 39 used the same source data, followup period, social health exposure and CVD outcome. We chose Strodl 2008 28 (rather than Byles 2015 39 ) in the meta-analysis as their analysis considered the social health exposure as a potential predictor of the incidence of CVD outcome. The four remaining included papers chose to assess the exposure as either categorical (n = 3; Strodl 2003, 27 Strodl 2008, 28 Simons 2013 40 ) or continuous (n = 1; Sahle 2020 45 ). At protocol stage we misinterpreted the available biostatistical methodology and conversion from continuous to categorical is not possible.
Hence, the paper reporting the social health exposure as continuous (Sahle 2020 45 ) was excluded from meta-analysis. Three papers (Strodl 2003, 27 Strodl 2008 Simons 2013 40 ) remained, with four outcomes, which was sufficient for meta-analysis.
For the first meta-analysis method, we converted odds ratios or hazard ratios into relative risk (Figure 2A), demonstrating no association between social health and CVD. We speculate that including the non-statistically significant findings from Sahle 2020 45 (excluded from our meta-analysis based on exposure assessed as continuous), would not alter our finding. However, caution is required when interpreting Figure 2 due to a number of assumptions that were required to compile these results. Strodl  assessed DSSI as three categories, but to make it more comparable to Simons 2013, 40 we re-categorised the data into two categories ("very high-high" vs. "low-fair") and therefore unadjusted estimates are presented. Through this process we noticed that in Strodl 2008 28 's Table 1, the numbers for high DSSI did not align with the odds ratio and speculate that the reference category was incorrectly reverted. As Simons 2013 40 did not provide the prevalence of those not-exposed (ie 'good') we used the Strodl 2003

| Sub-analyses
No association was observed regardless of CVD subtype (CHD or stroke) or data source (ALSWH or Dubbo) ( Figure 2). Further subanalyses were not possible as they would likely reflect data source differences rather than differences by gender, baseline year, or years of follow-up. No study provided stratification by partner status and all studies recruited people living in Australia.

| DISCUSS ION
The five included papers reported six CVD outcomes and represented three Australian longitudinal cohort studies. The included papers reported no association between social health and incidence of CVD in fully adjusted models, and most unadjusted models. Notably, all included papers were assessed as high risk of bias, mainly due to lack of adjustment for important confounders, self-report of CVD and short follow-up.
Since our systematic search in June 2020, we have identified only one further publication that would have been included in this review.

| Comparison to international literature
The few papers we included provided findings that contradict the international systematic review reporting that poor social relationships are associated with a 29% and 32% increase in risk of CHD and stroke incidence respectively. 26

| Strengths and limitations
Our review mirrored the methodology of an international systematic review, 26 which allowed direct comparison of findings. Inclusion of longitudinal data provided directional assessment of social health as a risk factor for incidence of CVD, reducing the issue of reverse causation. Conversion of odds ratios and hazard ratios to relative risks allowed direct comparison between studies, however we did misinterpret at protocol stage that biostatistical methodology is not available for conversion from continuous to categorical ratios. We improved upon the protocol by using a specified quality assessment tool for the risk of bias and assessing publication bias.

| Conclusion
Our systematic review is inconclusive regarding the association between social isolation, lack of social support and loneliness and CVD incidence among people living in Australia and New Zealand. We identified five eligible papers from three longitudinal Australian cohort studies. No eligible studies were identified from New Zealand, highlighting a huge deficit in current research, as even research in Australia may not be generalisable due to different social, political and cultural systems. Included papers reported no association between social health and incidence of CVD in fully adjusted models, and most unadjusted models. The included papers relied heavily on self-reported CVD prevalence and incidence, hence further studies in Australia using medical diagnosis of CVD (through medical records, death certificates or national registers) are required.
Additionally, further research should explore whether Australia's and New Zealand's unique geographical population spread plays a role in the relationship between social health and CVD.

ACK N OWLED G EM ENTS
We thank Noria Akbari or Aghnia Naim for undertaking title/abstract screen of 575 publications each. Open access publishing facilitated by Monash University, as part of the Wiley -Monash University agreement via the Council of Australian University Librarians.

CO N FLI C T O F I NTE R E S T
None declared. The data collection, analysis and interpretation of data; the writing of the manuscript; and the decision to submit the manuscript for publication were solely at the discretion of the researchers, independent of funders.

AUTH O R CO NTR I B UTI O N S
RF, AZZP and JH contributed to data screening and data extraction.

Not social health
Tse 2017 Reduction in retained activity participation is associated with depressive symptoms 3 mo after mild stroke: an observational cohort study