Health and economic burden of influenza‐associated illness in South Africa, 2013‐2015

Abstract Background Economic burden estimates are essential to guide policy‐making for influenza vaccination, especially in resource‐limited settings. Methods We estimated the cost, absenteeism, and years of life lost (YLL) of medically and non‐medically attended influenza‐associated mild and severe respiratory, circulatory and non‐respiratory/non‐circulatory illness in South Africa during 2013‐2015 using a modified version of the World Health Organization (WHO) worksheet based tool for estimating the economic burden of seasonal influenza. Additionally, we restricted the analysis to influenza‐associated severe acute respiratory illness (SARI) and influenza‐like illness (ILI; subsets of all‐respiratory illnesses) as suggested in the WHO manual. Results The estimated mean annual cost of influenza‐associated illness was $270.5 million, of which $111.3 million (41%) were government‐incurred costs, 40.7 million (15%) were out‐of‐pocket expenses, and $118.4 million (44%) were indirect costs. The cost of influenza‐associated medically attended mild illness ($107.9 million) was 2.3 times higher than that of severe illness ($47.1 million). Influenza‐associated respiratory illness costs ($251.4 million) accounted for 93% of the total cost. Estimated absenteeism and YLL were 13.2 million days and 304 867 years, respectively. Among patients with influenza‐associated WHO‐defined ILI or SARI, the costs ($95.3 million), absenteeism (4.5 million days), and YLL (65 697) were 35%, 34%, and 21% of the total economic and health burden of influenza. Conclusion The economic burden of influenza‐associated illness was substantial from both a government and a societal perspective. Models that limit estimates to those obtained from patients with WHO‐defined ILI or SARI substantially underestimated the total economic and health burden of influenza‐associated illness.


| INTRODUC TI ON
Influenza is a vaccine-preventable infectious disease that causes substantial morbidity and mortality globally every year, in particular among young and elderly individuals and persons with underlying medical conditions. [1][2][3][4] Global studies have suggested a higher burden of influenza-associated severe illness, including death, in Africa compared to other regions. 1,2 Nonetheless, influenza vaccine use remains low on the continent. 5 Disease and economic burden estimates are essential to guide policy-making on influenza vaccination, especially in resource-limited settings; however, the disease and economic burden of influenza-associated illness remain poorly understood in Africa.
In South Africa, a middle-income country of 54.8 million people in 2015, over 10 million mild, 128 000 severe-non-fatal and 11 000 fatal influenza-associated illness episodes are estimated to occur annually with a heavy burden among young and old individuals and persons with chronic medical conditions, including HIV and tuberculosis infection. 3,4,[6][7][8][9][10][11] Approximately one million doses of influenza vaccine are available in the private sector annually, and approximately the same number of doses is available annually in the public sector since 2010. Vaccination guidelines are revised annually, and influenza vaccination is recommended for groups at increased risk of influenza-associated severe illness with the highest priority given, in recent years, to pregnant women and HIV-infected individuals. 12 In addition, an influenza policy that also provides the implementation framework for influenza immunization was adopted in 2017. 13 Nonetheless, the number of doses of influenza vaccine remains insufficient to cover the recommended risk groups, estimated to be over 20 million individuals. 13 In South Africa, like in many other countries, primary drivers of vaccine policy are the cost of the vaccine, its delivery, and the economic impact of the disease. South Africa has a policy of introducing new vaccines or increasing vaccination coverage only after careful consideration of the evidence of disease burden and the potential costs and benefits from such vaccines. However, economic burden estimates of influenza-associated illness in South Africa are lacking.
In this study, we aimed to estimate the mean annual direct and indirect costs, absenteeism, and years of life lost (YLL) from influenzaassociated illness in South Africa from 2013 through 2015.

| Definitions
We estimated the economic burden from a societal perspective.
Costs in South African Rands were converted to 2015 US dollar using average monthly South African Rand to US dollar exchange rate in 2015. All costs were expressed in 2015 prices using the South Africa all-items Consumer Price Index. 14 We aggregated influenza-associated illness within four syn- illness. The main analysis focused on syndromes (b-d), and the total economic burden was obtained by summing the cost among these three syndromes, whereas syndrome (a) was used as sensitivity analysis (ie, minimum estimate). We considered healthcare attendance of influenza-associated illness in two categories: (a) medically attended: attended by a registered medical care provider/institution excluding pharmacies and (b) non-medically attended: not attended by a registered medical care provider/institution including pharmacies and traditional healers. We defined the severity of the abovementioned (15%) were out-of-pocket expenses, and $118.4 million (44%) were indirect costs.
The cost of influenza-associated medically attended mild illness ($107.9 million) was 2.3 times higher than that of severe illness ($47.1 million). Influenza-associated respiratory illness costs ($251.4 million) accounted for 93% of the total cost. Estimated absenteeism and YLL were 13.2 million days and 304 867 years, respectively. Among patients with influenza-associated WHO-defined ILI or SARI, the costs ($95.3 million), absenteeism (4.5 million days), and YLL (65 697) were 35%, 34%, and 21% of the total economic and health burden of influenza.

Conclusion:
The economic burden of influenza-associated illness was substantial from both a government and a societal perspective. Models that limit estimates to those obtained from patients with WHO-defined ILI or SARI substantially underestimated the total economic and health burden of influenza-associated illness.

K E Y W O R D S
economic burden, influenza, South Africa, years of life lost influenza-associated illnesses in three levels as follows: (a) mild: medically and non-medically attended illness not warranting hospitalization; (b) severe: medically and non-medically attended illness warranting hospitalization, including deaths; and (c) in-and out-ofhospital deaths.
We considered the cost of influenza-associated illness in three categories as follows: (a) direct medical costs: costs related to treatment incurred both within and outside health facilities (ie, costs of ambulatory care, hospitalization, pharmaceuticals, and consultation with traditional healers); (b) direct non-medical costs: illness-related expenditures that do not relate directly to medical treatment (eg, transportation costs to hospital/clinics, additional food costs, and extra expenses for accommodation for both patient and caregiver); and (c) indirect costs: the value of lost productivity because of reduced working time (for both patient and caregiver) during the illness episode or while receiving care. Medical and medication costs not covered by healthcare facilities, consultation with traditional healers, and direct non-medical costs were considered out-ofpocket expenditures.

| Data sources
For estimation of the economic burden and YLL from influenza-associated illness, we used a combination of published or publically available data 6,16-31 as well as data collected through laboratoryconfirmed influenza surveillance and costing studies. The list of the data sources and their use in the estimation approach are provided in Table 1, whereas the detailed description of each data source is • Population denominators. 14-16

DS3
Laboratory-confirmed influenza surveillance data among inpatients with SARI and outpatients with ILI at selected surveillance sites during 2013-2015.
• Length of hospitalization and proportion of in-hospital procedures (ie, admission to ICU, chest X-rays, oxygen therapy, medications, and laboratory tests) among influenza-positive patients with SARI. • Proportion of medications among influenzapositive patients with ILI.
Primary data collection DS4 Hospitalization data from a large private hospital network (NetCare) active in 7/9 provinces.
• Length of hospitalization among patients hospitalized with circulatory or non-respiratory/ non-circulatory syndromes.

DS5
Cost of hospitalization and outpatient consultation.

DS6
Costing study of laboratory-confirmed influenza inpatients with SARI and outpatients with ILI at selected surveillance sites during 2014.
• Direct medical and non-medical costs incurred by the patient/caregiver and absenteeism for medically attended illness (out-of-pocket expenditures).
Primary data collection DS7 Community costing survey implemented in 4/9 provinces of South Africa.
• Direct medical and non-medical costs incurred by the patient/caregiver and absenteeism for non-medically attended illness (out-of-pocket expenditures).

DS8
Healthcare utilization survey among individuals with SARI or ILI in three South African communities.
• Proportion of non-medically attended illness that sought care with pharmacies or traditional healers. 25 provided as Appendix S1. All estimates of influenza disease burden, either for medically or non-medically attended illness, were based on pre-existing mean annual national estimates for 2013-2015 (DS2 in Table 1 and Appendix S1). 6

| Estimation of the economic burden of influenza-associated illness
To estimate the national economic burden of influenza-associated illness, we used the WHO manual and toolkit, a worksheet based tool, for estimating the economic burden of seasonal influenza 32 with the following modifications: (a) we included direct and indirect cost estimates for patients and caregivers for non-medically attended severe illness, as a high proportion (56.1%) of influenzaassociated severe illness is not medically attended in South Africa, 6 and (b) we included indirect costs for patients and caregivers for non-medically attended mild illness, as a high proportion (74.4%) of influenza-associated mild illness is not medically attended in South Africa. 6 Furthermore, of these non-medically attended illness episodes, a high proportion (57.3%) results in absenteeism from school or work for the patient and/or the caregiver. 26 For each data input in the WHO toolkit, we inputted the main estimate as well as the 95% CI as required. 32 We reported total costs summed across the three main endpoints as well as cost estimates disaggregated by the three main syndromic endpoints considered in this study (ie, all-respiratory, all-circulatory, and non-respiratory/non-circulatory). Estimates were parametrized and generated within six age categories (ie, <1, 1-4, 5-19, 20-44, 45-64, and ≥65 years of age) and subsequently aggregated.
In addition, we implemented a sensitivity analysis using the WHO toolkit without the abovementioned modifications and restricted to influenza-associated ILI and SARI (a subset of all-respiratory illness) as suggested in the WHO manual. 32 We also reported the economic burden of influenza-associated illness per capita and as a proportion of the gross domestic product (GDP) for comparison with other studies globally. South Africa's mean annual GDP during 2013-2015 was $344.9 billion. 33

| Direct cost estimation
In the main analysis for medically attended severe illness, data on quantities of resources used during hospitalization were obtained from influenza-positive patients hospitalized with SARI (ie, number of admissions to ICU, chest X-rays, oxygen therapy, medications, and laboratory tests-DS3 in Table 1 and Appendix S1) through routine influenza surveillance conducted in seven public hospitals across the country. Unit costs of hospitalization/procedures, medications, and laboratory testing were obtained from the uniform patient fee schedule of the National Department of Health and the state price list of the National Health Laboratory Service (DS5 in Table 1 and Appendix S1). Unit cost was applied to average quantity of resources used per influenza-associated SARI episode to estimate ancillary costs (ie, costs of procedures, medications, and laboratory tests). Routine service costs per day (eg, facility and consultation fees) were multiplied by the mean length of hospitalization, which was estimated within syndromes using DS3 and DS4 in Table 1 and Appendix S1. Routine service costs and ancillary costs were summed to obtain the mean total cost of influenza-associated SARI hospitalization per episode. This figure was multiplied by national estimates of the number of influenza-associated severe illness episodes (obtained from DS1 and DS2 in Table 1 and Appendix S1) across the syndromes evaluated in this study (ie, acute-respiratory, all-respiratory, all-circulatory, and non-respiratory/non-circulatory). This was done because hospitalization costs among influenza-associated non-SARI illness were not available. In addition, we obtained the mean out-ofpocket expenditure (for patient and caregiver) per influenza-associated SARI episode using data from a costing study (DS6 in Table 1 and Appendix S1) conducted at the same hospitals where SARI surveillance was implemented. A similar approach was used to estimate the mean consultation cost and out-of-pocket expenses per episode of influenza-associated ILI and multiplied by national estimates of the number of influenza-associated mild illness episodes (using DS1 and DS2 in Table 1 and Appendix S1) across the syndromes evaluated in this study.
For non-medically attended mild and severe illness, the quantity of resources used was obtained from community costing surveys (DS7 in Table 1 and Appendix S1) and healthcare utilization surveys (DS8 in Table 1 and Appendix S1). The estimated costs per episode were multiplied by national estimates of the number of non-medically attended influenza-associated mild and severe illness (DS2 in Table 1 and Appendix S1.) across the syndromes evaluated in this study.
In addition, we postulated that the hospitalization and outpatient consultation costs may be different (probably higher) among patients with influenza-associated circulatory and non-respiratory/ non-circulatory illness compared to those with influenza-associated SARI or ILI. Hence, we implemented a sensitivity analysis by replacing the estimated cost of hospitalization and outpatient consultation among influenza-positive patients with SARI or ILI by the cost per patient day equivalent (PDE; any hospitalization cost) and any outpatient consultation cost to obtain an estimate that would account for differential treatment costs across different clinical presentations (DS5 in Table 1 and Appendix S1).

| Indirect cost estimation
We estimated costs associated with absenteeism due to mild and severe episodes of influenza-associated illness. For absenteeism due to hospitalization, number of days absent from work or school was estimated as the length of hospitalization plus 1.5 days as reported by Molinari et al. 34 Absenteeism was estimated for affected patients and their caregivers, parametrized within syndromes using length of hospitalization obtained from DS3 and DS4 for patients (Table 1 and Appendix S1) and DS6 for caregivers (Table 1 and Appendix S1). For mild illness, number of productive days lost was obtained from DS6 and DS7 (Table 1 and Appendix S1).
School absenteeism due to mild or severe illness was calculated for individuals aged 5-19 years, and work absenteeism was calculated for individuals aged 20-64 years including patients and their caregivers. School and work absenteeism for hospitalized patients were adjusted based on five working/school days per week with work absenteeism for patients and caregivers further adjusted to account for the unemployment rate in South Africa (DS9 in Table 1 and Appendix S1). The mean absenteeism per illness episodes was obtained by dividing the days of absenteeism as estimated above (ie, excluding individuals of non-school or working age, and adjusting for five working/school days per week and the unemployment rate) by the total number of illness episodes, including individuals of non-school/working age. This resulted in the mean absenteeism for school or work per illness episode to be lower than the mean number of hospitalization days per illness episode.
We estimated the indirect cost of medically and non-medically attended illness by multiplying minimum daily wages by the number of days absent from work as estimated above for patient and caregiver. We only considered value of lost time from paid activities but did not consider value of time for unpaid activities such as housekeeping or leisure.

| Estimation of years of life lost due to influenza-associated deaths
We estimated the YLL due to influenza-associated deaths using the national estimates of influenza-associated deaths (DS2 in Table 1 and Appendix S1) and the life expectancy for South Africa (DS10 in Table 1 and Appendix S1). YLL was estimated for all influenza-associated deaths as well as for influenza-associated SARI deaths.

| Ethical approval
The SARI protocol (DS2) was approved by the University of the

| Direct medical and non-medical costs incurred by the patient/caregiver and absenteeism due to medically attended illness (DS5 and DS6)
In  Abbreviations: CI, confidence intervals; ILI, influenza-like illness (outpatients); SARI, severe acute respiratory illness (inpatients). a Non-medical costs include cost for transportation and additional food.
percentiles: $191-$1156). Average direct out-of-pocket costs and number of days absent from school or work are provided in Table 2, while the itemized unit costs for patients with influenza-associated SARI or ILI are provided in Table 3.

| Economic burden of influenzaassociated illness
The mean direct and indirect cost per illness episode was $25 overall; $770 for medically attended severe illness; $40 for medically attended mild illness; and $15 for non-medically attended mild or severe illness. Mean annual cost of influenza-associated illness was estimated at $270.5 million (Table 4), 0.08% ($5.1 per capita) of the total mean annual GDP. Of this, the direct cost for medically and nonmedically attended illness was $152.1 million (56% of the total cost) and the direct and indirect cost of medically attended influenza-associated illness was $155.1 million (57% of the total cost) (Table 4 and   (Table 5 and Figure 1B). The cost of influenza-associated WHO-defined SARI and ILI (a subset of allrespiratory illness) was $95.3 million accounting for 35% of the total cost (Table S3). This cost using the unmodified WHO toolkit (ie, not accounting for direct and indirect cost of non-medically attended influenza-associated SARI and the indirect cost of nonmedically attended influenza-associated ILI) was estimated at $87.6 million (Table S3).
The cost of influenza-associated illness using the PDE was estimated at $292.4 million (Table S1). The cost by syndromes using the PDE is provided in Tables S2 and S4.

| Absenteeism for influenza-associated illness
The mean annual absenteeism associated with influenza-associated illness was estimated at 13.2 million days (1.2 days per illness episode; Table 4); 6.7 million days of absenteeism from school (51%; 0.6 days per illness episode); and 6.4 million days of absenteeism from work (49%; 0.6 days per illness episode; Table 4 and Figure 2C). Absenteeism due to non-medically attended influenzaassociated illness accounted for 70% (9.2 million days) of total absenteeism (Table 4 and Figure 2A).
Absenteeism for influenza-associated SARI and ILI (a subset of allrespiratory illness) was 4.5 million days accounting for 34% of the total absenteeism (Table S3). The absenteeism for influenza-associated illness using the unmodified WHO toolkit was 3.2 million days (Table S3).

| D ISCUSS I ON
We estimated the economic burden, absenteeism, and YLL from influenza-associated illness in a middle-income country in Africa. The In our setting, the indirect cost of influenza-associated illness accounted for 44% of the total cost. This was similar to studies conducted in the United States and Asia (range: 13%-56%), whereas in Europe, the indirect cost accounted for a higher proportion of the total cost (range: 88%-92%). 35 Some of these differences may be related to different unit costs in diverse settings as well as dissimilar estimation approaches. Whereas the cost per illness episode of non-medically attended illness ($15) was lower than that for medically attended illness, the cost of non-medically attended illness accounted for 43% of the total cost. This is because a large proportion of influenza-associated mild (~74%) and severe (~56%) illness is not medically attended in South Africa. 6 This was also reflected in the high proportional contribution of absenteeism from non-medically attended illness (70%) to the total.
Irrespective of syndromes, the cost of influenza-associated medically attended mild illness accounted for 70% of the cost of any medically attended illness (severe and mild). Whereas the cost of and absenteeism from influenza-associated severe illness per episode are higher than those of mild illness, the number of severe illness episodes is substantially lower (1.2% of total illness) than the number of mild illness episodes. 6 The cost of and absenteeism from influenza-associated respiratory illness accounted for 93% and 98% of the total, respectively. Influenza-associated non-respiratory illness accounts for a large proportion of influenza-associated severe illness (45%), but a small proportion of influenza-associated mild illness (<1%) 6 resulting in an overall minimal contribution of influenza-associated non-respiratory illness to the total cost (7%) and absenteeism (2%).
We found a substantial underestimation of the total cost, absenteeism, and YLL when restricting the analysis to individuals with influenza-associated SARI and ILI, a subset of all-respiratory severe and mild illness, respectively. This is because, in our setting, influenza-associated ILI, influenza-associated non-fatal SARI, and influenza-associated SARI deaths represent only 34%, 40%, and 20% of the total number of influenza-associated mild, severe-non-fatal, and fatal illness, respectively. 6 In addition, when we implemented the analysis among individuals with influenza-associated SARI and ILI using the unmodified WHO tool (ie, not accounting for direct and indirect costs of nonmedically attended influenza-associated SARI and the indirect cost of non-medically attended influenza-associated ILI), we further underestimated cost by $7.7 million and absenteeism by 1.3 million days.
Our study has some limitations that warrant discussion. First, we did not include the indirect cost associated with deaths including lost productivity because such estimates are not available for South Africa. Second, because we did not have specific estimates of costs for influenza-associated all-circulatory and non-respiratory/non-circulatory illness, we assumed the same costs as for influenza-associated respiratory illness. However, given the proportionally small number of influenza-associated non-respiratory illness (<1% of the total illness episodes), this category minimally influenced the total cost estimates. In addition, the economic burden estimates using the PDE costs (sensitivity analysis; $292.4 million) and the influenza-associated illness specific cost (main analysis; $270.5 million) were similar. Last, because of the high disparity of wages in South Africa, we used the minimum wages to estimate indirect cost; hence, this represents a minimum estimate.
In conclusion, we found a substantial economic burden, absenteeism, and YLL associated with influenza disease in South Africa, with influenza-associated mild and non-medically attended illness accounting for a large proportion of total cost and absenteeism in our setting.
The economic burden was high from both a government and a societal perspective. Estimates obtained from individuals meeting the SARI and ILI case definition as suggested in the WHO manual for estimating the economic burden of seasonal influenza 32 would underestimate the total burden in excess of 60% for each of the three endpoints evaluated in this study. This study provides the foundation for future cost-benefit analysis on potential interventions such as influenza immunization. However, despite the substantial disease 6 and economic burden of influenza-associated illness in South Africa, given the competing health priorities in the country, it is unlikely that the South African government will introduce universal influenza immunization in the near future. Cost-burden and cost-benefit studies on the potential impact of influenza immunization among groups at increased risk of influenza-associated severe illness, including different age groups, F I G U R E 2 Estimated mean annual proportional number of absenteeism (in days) from influenza-associated illness in South Africa during 2013-2015 by: A, healthcare attendance; B, syndrome; and C, absenteeism type are warranted to further guide the current influenza vaccination guidelines 12 and refine the existing influenza policy. 13 An alternative strategy to risk group-based vaccination is to target the community transmitters of infection, such as school-aged children, thereby potentially reducing overall community disease burden and associated cost. 36,37 Cost-benefit studies of these approaches are also warranted.

E THI C S
The SARI protocol (DS2 in The costing protocol (DS6 in Table 1) was approved by the HREC protocol number M121195. All other data sources were publically available.