Health-related quality of life of Australians during the 2020 COVID-19 pandemic: a comparison with pre-pandemic data and factors associated with poor outcomes

Australians aged 18–99 years who were members of an online panel managed by the survey and consumer insights company Toluna (https://​au.​toluna.​com) were invited to participate. Sampling was managed by the survey host company, Survey Engine [12]. Quota sampling was conducted at two levels: by region/pandemic intensity and then by age and sex. At the first level, participants were quota sampled by four regions that we defined according to the number of active cases regionally and the relative restrictiveness of prevailing public health orders at the time of the survey, noting that Melbourne and greater Victoria were experiencing a second wave of the pandemic and a long and strict lockdown period in the lead-up to, and during, data collection for this study (see Online Appendix 1). The following four regions of pandemic intensity were defined: (1) Melbourne, the capital city of Victoria: the greatest number of cases and the most restrictive public health orders, including city-wide lockdown; (2) Regional Victoria: the second highest number of cases and the second most restrictive public health orders; (3) NSW and Queensland: some localised hot-spots and border closures; (4) Western Australia, South Australia, Tasmania, the Australian Capital Territory, and the Northern Territory: very few cases, least restrictive public health orders. Note that Australian states are governed by state governments and the two territories are under federal governance; therefore, the nature and timing of specific public health orders and restrictions differed between states and territories. A summary of these is presented in Online Appendix 1.

The second level of quota sampling (by age and sex) occurred within each region, which ensured proportions corresponded to the Australian general population within each region and therefore for the overall sample [13]. Responses to each item were mandatory. Participants who successfully completed the survey were awarded ‘panel points’ by Toluna, which are used to redeem vouchers or consumer goods, with an approximate value of less than 1AUD.

Participants completed an online survey about the impact of the COVID-19 pandemic on their daily life, work and study, and about their current health status, HRQL, and mental health. Participants also completed questions about their demographic characteristics. Relevant to this analysis, the surveys completed were:

*measures currently undergoing validation. See details and references in Online Appendix 2.

Between March 2015 and February 2016, we collected QLQ-C30 reference value data from the Australian general population (N = 1979), representative by age and sex, to facilitate interpretation of QLQ-C30 data from Australians with cancer [18]. We also collected Kessler distress scale and GHQ data in the same online survey.

We imposed seven quality checks, including unreasonable completion time (completion in less than 30% of the sample’s median completion time, here 7.79 min), and six checks for inconsistent responses, to ensure our final dataset included high-quality and genuine responses, as recommended for social science surveys [19]. Participants who failed two or more quality checks or completed in less than 7.79 min were excluded from analysis. Online Appendix 3 describes the quality checks used and includes a flowchart of participant inclusions and exclusions.

We required a minimum sample size of n = 1456 for the primary study endpoint, allowing us to detect clinically “small” differences in any of the QLQ-C30 scales [20] with 80% power and a Bonferroni adjusted alpha of 0.003 (Bonferroni adjustment is conservatively recommended for sample size calculation when using the Hochberg method) [21]. Our recruitment target was increased to n = 2000 (n = 500 per sampling region) to allow for drop out and planned secondary analyses.

To address Aim 1, we calculated the mean and standard deviation of each scale within the QLQ-C30, and GHQ, for each age group (18–29, 30–39, 40–49, 50–59, 60–69, and 70 + years) and sex (male, female) according to their scoring manuals [17, 22]. We also calculated the QLQ-C30 27-item summary score [23].

We then compared the mean scores per scale to our 2015/16 reference sample using independent t-tests for each age group and sex. We hypothesised that health outcomes would generally be worse for the 2020 sample for the following functional and symptom QLQ-C30 domains, which were most applicable to a general population sample in the context of the pandemic [7]: role functioning, emotional functioning, social functioning, cognitive functioning, fatigue, sleep disturbances, financial impact, and global HRQL, as well as for the GHQ score. We interpreted the size of the differences in QLQ-C30 scales according to Cocks’ guidelines [20].

To address Aim 2, we conducted a series of simultaneous multiple regressions for each of the 15 QLQ-C30 domains and the GHQ (outcome variables), using the same 22 explanatory variables (see Online Appendix 4). A set of explanatory variables was chosen a priori based on the existing literature and hypotheses that key characteristics of the individual and environment would predict worse HRQL and health outcomes, in line with Wilson and Cleary’s model. These include increased commitments (including workload and carer responsibilities), higher COVID-related worries, more changes to one’s typical routine, recency of lockdown/severity of the COVID-19 situation within region, lower social support and poorer attitude towards restrictions and regulations generally. The explanatory variables were consistent within each regression model (per domain).

The overall model can be summarised by the following equation:where Y_i represents the domain of study for a person i, e.g. QLQ-C30 overall health/HRQL, a represents the intercept; β_1-22 represents a regression coefficient for each relevant explanatory variable (22 all together), and e represents error.

More details are presented in Online Appendix 5. All assumptions were checked, and with several caveats described below, satisfied. The descriptive statistics including skewness metrics are outlined in Table A5.1 in online Appendix 5. We applied a Bonferroni adjusted alpha of 0.05/16 scales = 0.003 to determine the significance of overall models; this adjustment is conservative given that domains of the QLQ-C30 are known to be correlated [17, 24]. We then applied the Hochberg adjustment method within each model to determine the significance level of individual explanatory variables based on ranked alphas [21].

All analyses were completed in SPSS v27 and all available item-level data from included participants were used.

Table 2 shows QLQ-C30 and GHQ scores for the 2020 sample compared to a pre-COVID-19 reference sample [18] from 2015 to 2016 by age group and sex. The largest differences seen were in the 18–29 year group, whereby the 2020 sample reported worse cognitive functioning, nausea, diarrhoea, and financial difficulties, with these differences deemed of a medium size according to interpretation guidelines [20]. Most HRQL domains were statistically significantly worse for the 2020 sample, by age group and sex, as compared to the reference sample. Notable exceptions include fatigue, pain and the QLQ-C30 summary score, which were statistically similar in both samples across all age groups. Emotional functioning was statistically significantly worse for the 2020 participants aged 18–59, however there was no difference for older subgroups aged 60 and over.

The 2020 sample’s emotional functioning scores were substantially lower than the reference sample, with the difference ranging from 7 to 15 points, with larger differences amongst the younger age groups.

GHQ scores were only statistically different between pre-pandemic and 2020 samples for the 18–29 age group, with the 2020 sample reporting worse general health. The global health/HRQL scale of the QLQ-C30 was statistically significantly worse for all age groups except 30–39 years, during the pandemic. For those aged 40–70 +, differences were considered “small”. For the 18–29 and 30–39 year old, the 2020 sample reported trivially worse global health/HRQL on average, as compared to the reference sample.

Interestingly, amongst men, pain and fatigue were statistically significantly better in the 2020 sample, although the sizes of these differences were considered trivial [20]. Amongst women, all scales, apart from the QLQ-C30 summary score, were statistically significantly worse in the 2020 sample. All QLQ-C30 differences were considered small, apart from fatigue, pain, and constipation, which were trivial according to Cock’s interpretation guidelines [20].

Descriptive statistics and correlations for all variables in the regression models are provided in Appendix 5. All models met the assumption of linearity. GHQ, global health/HRQL, and emotional functioning scales met the assumption of normality of the residuals. All other domains did not satisfy the assumptions of normality of the residuals and homoscedasticity. This may have been caused by these variables having skewed distributions (negatively for functioning domains and positively for symptoms, as observed via histograms and skewness statistics reported in Appendix 5). However, these distributions are expected of a non-clinical population. Deviations from normality of the residuals are less serious in large samples; therefore, no transformations were applied [25]. There were no indications of major multicollinearity issues, with Tolerance statistics ranging between .42 and .94 and variance inflation factor (VIF) estimates between 1.02 and 2.47 [26, 27]. Table 3 shows a summary of 16 regression models (see Eq. 1). It includes estimates of variance predicted for the overall model (R²) with the relevant F- and p-values and standardised regression coefficients (betas). More details are provided in Appendix 5.

All models were statistically significant at p < .001. The models that explained most of the variance in the distribution were emotional functioning (35.6%)), QLQ-C30 global health/HRQL (22.5%), Nausea/vomiting (18.8%), GHQ (18.7%), and financial difficulties (17.9%). Observation of zero-order correlations between explanatory variables and dependent variables (see Tables A5.2–A5.4 in online Appendix 5) suggest that 10 of the 95 significant beta coefficients (or 2.8% of the 352 coefficients reported in Table 3 in total) might have been due to a suppression effect. This occurs when an independent variable (IV; the suppressor) either correlates positively with another IV and negatively with the dependent variable (DV), or vice versa, or serves to control for (suppress) variance in the IV which is irrelevant to the DV [27]. The suppressor variable will increase the regression weight of the IV it is correlated with [27]. Thus, if a beta coefficient is high and the correlation is low, this signals a suppression effect [27].

Better emotional functioning was most strongly associated with lower personal/family concerns, lower financial concerns, better job security, finances, mental health, and social life due to pandemic-related health orders, better physical health/activity, sleep, and nutrition, having less loneliness, fewer perceived negative impacts of the COVID-19 pandemic, more positive compliance attitudes, being male, and living in NSW or Queensland. Living in Melbourne was also significant though possibly due to a suppression effect, so this finding should be interpreted with caution.

Better global health/HRQL was most strongly associated with less worry about self or family being infected with COVID-19, positive impacts on work/finances, mental health and relationships, positive impacts on physical health and activity, less family responsibilities, more positive impacts of lockdown/regulations and less negative impacts, positive attitudes towards compliance, and living with partner. Less loneliness and time, and younger age were also significant, albeit likely due to a suppressor effect.

More nausea/vomiting was most strongly associated with worries about self or family, worries about infrastructure and supplies, negative attitudes towards compliance, younger age, and not living in Melbourne.

The factors most strongly associated with a poorer (higher) GHQ score included a more negative impact of the COVID-19 pandemic on physical health/activity, job security/finances/mental health, personal/family concerns, fewer number of positive and other impacts, and older age.

Greater financial difficulties were most strongly associated with all aspects of COVID-19-related worry, worse impact of the COVID-19 pandemic on physical health, higher number of negative impacts of the COVID-19 pandemic, negative attitudes towards compliance with public health orders and not living in Melbourne.

Further details on the results for all scales, including reliability estimates, are provided in online Appendix 5.

Strengths of our study include the ability to compare our 2020 data to reference data from 2015 to 2016 across all domains to demonstrate change since the pandemic. Like our pandemic sample, our reference sample was recruited from an online panel and data collected via online survey. Other strengths include the large sample size, and the use of validated self-report measures.

Our survey was conducted towards the end of the 2020 lockdown period, where most Australian states and territories had eased restrictions (see online Appendix 1) and case numbers were lower. Australia benefited from strict lockdown orders and high rates of compliance, through significantly fewer COVID-19 infections in 2020 compared to other countries [5]. A study on the impact of policies on social distancing behaviours in the USA showed that state-wide stay-at-home orders and limits on hospitality were the strongest measures to reduce mobility in the community [35], and overall infection rates. It is possible that Australians recognised the success of 2020 lockdown compared to international infection rates, which may have impacted their survey responses. As a consequence, our estimates of the impact of the pandemic are likely smaller than we would have observed if we had collected data at the peak time of infections.

Another possible limitation is the use of a cancer-specific HRQL measure, the QLQ-C30, to assess health outcomes amongst the general population. We do not feel this has impacted our results, as the questions do not refer specifically to cancer, the symptoms and functional concerns are general in nature and known to everyday people. In fact, the use of the QLQ-C30 questionnaire could be seen as a strength, as it helped us to identify an unexpected impact of the pandemic on nausea and vomiting that we may not otherwise have seen. It was used similarly in a Spanish study [30]. We also used guidelines for interpreting the clinical significance of differences to interpret the differences between mean scores of our 2020 data compared to our 2015/16 data—both obtained from non-clinical samples. Cock’s guidelines do not provide effect size estimates for the emotional functioning domain [20].

We acknowledge that despite our use of quota sampling to ensure adequate representation of age and sex, which are key variables for HRQL in the general population given our chosen HRQL measure [36], our sample was not representative on other variables as shown in Table 1; therefore, readers must interpret results accordingly.

Finally, a small number of significant effects were likely the result of a suppressor effect [27]. Whilst it is difficult to hypothesise the presence of the suppressor effect a priori [27], these results may inform future studies examining similar issues. Importantly, despite utilising 16 simultaneous regression models with 22 identical explanatory variables, only a very small percentage of effects (less than 3%) were likely to be the result of a suppressor effect. An additional strength of this research is that no multicollinearity issues were detected within each model and the significance levels were adjusted based on Hochberg’s procedure [20].

It will be interesting to conduct further analyses to determine whether specific subgroups of the Australian population were differentially affected for certain domains—such as essential workers, front-line workers, parents of school-aged children, people without work, people with family abroad, residents of aged care facilities, and other vulnerable groups. Examining the impact of the 2021 lockdowns due to COVID-19 Delta variant outbreaks in Eastern Australia on self-reported outcomes as compared to our 2020 data would also be of interest. The 2021 outbreak occurred in a climate where many Australians were fatigued by repeated, long lockdown periods [37]. In early 2021, Australia gained access to COVID-19 vaccines. Uptake was initially slow, due to supply shortages and some community reluctance, however by mid-October 2021, 55% of the Australian population was fully vaccinated, compared to 36% internationally, and by December 2021, 76.9% was fully vaccinated [38]. Therefore, there are key differences and challenges that may have impacted the Australian population since our 2020 survey. Finally, international comparisons, and historical comparisons (to past pandemics) [1] would be of interest to a global audience.

In conclusion, our survey results suggest that the COVID-19 pandemic and associated restrictions and health orders likely impacted most heath domains—with larger differences observed amongst younger adults and women. Domains most strongly associated with pandemic-related variables were emotional functioning, general health, financial difficulties, and nausea and vomiting—although all health domains were statistically significantly associated with pandemic-related factors. Our data could be used to identify Australian sub-groups most at risk of poor health outcomes in the event of future pandemics. This information could be used to develop and target interventions to mitigate these risks or improve outcomes.