A decision tree was developed based on Burden-eu’s protocol guideline and consensus model to estimate the DALY loss from the modelled COVID-19 cases of Australia’s national roadmap to re-opening [6, 10, 11].
Model structure
The model in this study was developed in Microsoft Excel 2016 (see Additional file 1 for the DALY COVID-19 model) [10, 11]. Figure 1 presents the COVID-19 outcome model adapted in this study [10, 11]. Patients infected with COVID-19 were simulated in the model and could progress to various health states: asymptomatic, symptomatic (acute), or death [10, 11]. “Asymptomatic” refers when a patient contracted the virus but do not show any symptoms [10, 11]. In the symptomatic health state, acute cases may experience mild/moderate symptoms but are not hospitalised, severe symptoms with hospitalisation, or a critical condition managed in intensive care unit (ICU). Asymptomatic and symptomatic patients may experience full recovery, Long COVID, or a permanent functional impairment [10, 11]. Long COVID patients experience longer-term symptoms after an acute COVID-19 infection, and may not recover fully for some time. Over time, those with Long COVID either fully recover or develop permanent functional impairment. Permanent functional impairments included diabetes, Parkinson’s disease, dementia, anxiety disorders, and ischemic stroke. Patients managed in ICU may recover fully or develop post-intensive care syndrome (PICS). PICS is defined as “new or worsening impairments in physical, cognitive, or mental health status arising after critical illness and persisting beyond acute care hospitalisation" [12]. It is assumed that patients in the state of permanent functional impairment could not transition back to post-acute consequence and recovery state.
Disability adjusted life years
Disability-adjusted life years (DALYs) were estimated by summing the life years lost due to premature mortality (YLL) and the years lived with disability (YLD) primarily using an incidence-based approach [10, 11].
DALY=YLL +YLD
YLL is calculated from the COVID-19 mortality statistics (M) and average life expectancy (LE), presented into a 10-age group band generated from public national data sources [6, 13].
YLL=M x LE
YLD for acute COVID-19, Long COVID and PICS (YLDinc) was calculated by multiplying the number of COVID-19 cases (N), average duration of health state until recovery or death (D), and disability weight (DW). The disability weight accounts for the extent of health loss associated with the specific health outcomes which range from 0 to 1 (0= no impact or having full health, 1= occurrence of death) [14].
YLDinc=N x D x DW
Some COVID-19 survivors have developed permanent illness or disability (distinct from Long COVID) [7-9]. However, the evidence on permanent impairment is less robust than that for acute, Long COVID and PICS. We therefore illustrate separately some of the potential for long-term disability due to COVID-19. Conservatively, we only include the incidence of certain permanent functional impairments post-COVID most commonly observed in large cohort studies (e.g., new onset of diabetes, Parkinson’s disease, dementia, anxiety disorders, and ischemic stroke) [8, 9]. To estimate the DALYs associated with these conditions, we sourced DALYs per person with these conditions from the 2019 Australian Burden of Disease study (See Additional file 2 Table 4) [15].
The results of this model are presented in three scenarios. The base case DALYs consist of the mortality and morbidity impact of acute COVID-19, Long COVID and PICS. “Total Burden One” presents the base case result plus the potential impact of all listed permanent functional impairments. “Total Burden Two” excludes diabetes as a permanent impairment.
Data Inputs
The data were obtained from the literature and Doherty’s modelling report[6], as update in September 2021. See Additional file 2 for more information. Ethics approval was not required as this study analysed publicly available data.
Doherty COVID-19 modelling
The Doherty COVID-19 model[6] was developed to inform Australia’s national COVID-19 reopening plan; it estimated the potential health and health system impacts of COVID-19 after eligible Australians achieve different coverage levels of full doses of COVID-19 vaccines (i.e. 50-80%). Transmission potential of COVID-19 delta variant, different bundles of public health and social distancing measures (PHSM), the efficacy of test-trace-isolate-quarantine activities (TTIQ), and the seeding infection rate (“initial number of daily cases present in the population at a given vaccination threshold”) were included in the Doherty analysis. Doherty model outputs over the first 180 days were reported in much detail [6], enabling us to calculate the potential DALY burden for each strategy and allowing us to illustrate the likely DALY burden arising from the post-acute consequences of COVID-19. The results and the analysis of our model are generated based on the most applicable hypothetical Doherty scenarios, given actual developments towards reopening to date across Australia. All other scenarios were also presented and calculated in the additional file 1 and additional file 2 table 1.
- Scenario 2C: Outbreaks seeded with 1,000 to 4,500 cases given partially effective TTIQ. The community has achieved COVID-19 vaccination coverage of 70% while maintaining low PHSM.
- Scenario 2D: Outbreaks seeded with 1,000 to 4,500 cases given partially effective TTIQ. The “medium PHSMs are overlaid between the 70 and 80% coverage thresholds with reversion to low PHSMs thereafter”.
- Scenario 3B: Outbreaks seeded with 300-1,000 cases given partially effective TTIQ. The community has achieved COVID-19 vaccination coverage of 80% with baseline PHSM.
- Scenario 3C: Outbreaks seeded with 1,000 to 4,500 cases given partially effective TTIQ. The community has achieved COVID-19 vaccination coverage of 80% with baseline PHSM.
The full definition of TTIQ and different levels of PHSM are reported elsewhere [6, 16]. Briefly, due to the high volume of cases, expected delays in TTIQ responses are noted for “partial” TTIQ [6]. No “stay-at-home” orders, but low-density requirements (2 sqm rule) are imposed for “baseline” PHSM. Social distancing rules are still mandated, but retail trade and travel restrictions are not imposed [6, 16]. Rules mandated for baseline are also similar for “low” PHSM, however there are some limitations in recreational, retail and workplace capacity under “low” PHSM [6, 16]. Under medium PHSM, “stay-at-home” orders are imposed unless for work, study and essential activities. However, work from home is recommended when possible. Schools, childcare and indoor recreational venues are closed. Intra and interstate travel are not allowed under medium PHSM [6, 16].
Asymptomatic Health State
According to a systematic review and meta-analysis, approximately 17% (95%CI:14%-20%) of total cases are asymptomatic [17]. However, asymptomatic cases are not presented in Doherty’s modelled result and for this reason we did not include asymptomatic cases in this calculation.
Symptomatic Health State
COVID-19 cases were obtained from the Doherty Modelling Interim Report to national Cabinet (17th September 2021) [6], and a period of 14-days was conservatively and collectively used for the recovery duration for acute-COVID-19 state [18].
Post-acute Consequences
Our model assumed that Long COVID symptoms start directly after the symptomatic phase. Given the United Kingdom (UK) evidence that some patients are still reporting Long COVID over 12 months after infection [19], our model assumed that Long COVID can potentially last up to two years. Due to the lack of longitudinal data regarding the length of Long COVID, we extrapolated available data on numbers of Long COVID cases over time from the UK Office of National Statistics (ONS)[19] and from a population-based cohort study in New South Wales(NSW)[20] until it reached 0% using a fitted decay function [21]. This estimate was only applied to COVID-19 survivors.
According to a large case-control study in the UK, fully vaccinated individuals appear less likely to experience Long COVID following “breakthrough” infection compared with their unvaccinated counterparts (odds ratio=0.51, 95%CI:0.32-0.82) [22]. This OR was converted to relative risk using the Cochrane formula[23] and was then applied in the vaccinated cohort.
Permanent functional impairment
To quantify the incidence of permanent disability post COVID-19, we used data from two studies that investigated these conditions [8, 9]. A large US cohort study indicated the incidence of 0.11% (95%CI:0.08-0.14) for Parkinson’s disease, 0.67% (95%CI:0.59-0.75) for dementia, 7.11% (95%CI:6.82-7.41) for anxiety disorders and 0.76% (95%CI:0.68-0.85) for ischemic stroke within six months post-COVID-19 [9]. For diabetes, data from a large cohort study in the UK found that 2.83% (95%CI:2.57-3.12) of hospitalised patients were diagnosed with diabetes (type-1 or-2) over a mean follow-up of 140 days [8]. However, it is arguable whether diabetes was an undetected pre-existing condition or whether COVID-19 induced type-1 or type-2 diabetes [24, 25]. Thus, we presented our DALY calculation with (”Total Burden One”) and without the impact of diabetes (“Total Burden Two”).
Post-Intensive care syndrome
We also included the debilitating effects often seen after ICU admission, commonly referred to as post-intensive care syndrome (PICS) [26]. The only cohort study that investigated PICS in the COVID-19 population found that 90.60% of ICU survivors had PICS [26]. Unfortunately, there are no existing disability weights for PICS. A QALY utility score of 0.75 (95%CI: 0.63–0.84) was available for ICU COVID-19 survivors 12-16 weeks post discharge [27], hence a comparable illness with similar utility score was assumed. We selected atrial fibrillation and flutter which has a utility score of 0.75[28] and a disability weight of 0.22 (95%CI: 0.15–0.31) [29]. A lifetime duration was assumed for PICS reflecting study findings that followed ICU survivors over 5-years and 10-years [30, 31].
Uncertainty Analysis and pilot model testing
Uncertainty analyses were undertaken to propagate parameter uncertainty (i.e. sampling error) from the input parameters to the final model outputs. Monte Carlo simulation with 2,000 iterations via the add-in tool Ersatz (Ersatz, Version 1.35) was used [32]. Estimates of DALYs were presented with 95% uncertainty intervals (95% UI).
The pilot testing of the model was conducted using actual observed Australian COVID-19 cases, hospitalisations and deaths data from January 2020 to January 2021. In this pilot, we have conducted different sensitivity analysis on rates of acute health states, post-acute rates, and a 28-day recovery period from the acute health state, to explore whether these input parameters have impacted on the robustness of the results (See additional file 3). Findings showed only a very small change in DALYs. Therefore, we have not included these sensitivity analyses in the current study. In this study, we have conducted a sensitivity analysis to investigate the impact of changes in the rates for post-acute consequences, using NSW Long COVID data points starting at 34% at week three and 0% at week 104 [20].