Stark choices: exploring health sector costs of policy responses to COVID-19 in low-income and middle-income countries

Objectives COVID-19 has altered health sector capacity in low-income and middle-income countries (LMICs). Cost data to inform evidence-based priority setting are urgently needed. Consequently, in this paper, we calculate the full economic health sector costs of COVID-19 clinical management in 79 LMICs under different epidemiological scenarios. Methods We used country-specific epidemiological projections from a dynamic transmission model to determine number of cases, hospitalisations and deaths over 1 year under four mitigation scenarios. We defined the health sector response for three base LMICs through guidelines and expert opinion. We calculated costs through local resource use and price data and extrapolated costs across 79 LMICs. Lastly, we compared cost estimates against gross domestic product (GDP) and total annual health expenditure in 76 LMICs. Results COVID-19 clinical management costs vary greatly by country, ranging between <0.1%–12% of GDP and 0.4%–223% of total annual health expenditure (excluding out-of-pocket payments). Without mitigation policies, COVID-19 clinical management costs per capita range from US$43.39 to US$75.57; in 22 of 76 LMICs, these costs would surpass total annual health expenditure. In a scenario of stringent social distancing, costs per capita fall to US$1.10–US$1.32. Conclusions We present the first dataset of COVID-19 clinical management costs across LMICs. These costs can be used to inform decision-making on priority setting. Our results show that COVID-19 clinical management costs in LMICs are substantial, even in scenarios of moderate social distancing. Low-income countries are particularly vulnerable and some will struggle to cope with almost any epidemiological scenario. The choices facing LMICs are likely to remain stark and emergency financial support will be needed.


Scenarios
The epidemiological model uses data from low-and middle-income countries. For each country, the model produces estimates on the number of cases, hospitalisations, number of days in hospital for severe cases (general ward) and critical cases (intensive care unit), and deaths for 57 distinct epidemiological scenarios (2).
For this study, four epidemiological scenarios were chosen out of the set of 57 possible scenarios. Scenario 1 represents an unmitigated epidemic. Scenarios 2-4 scenarios were chosen because they represent a variety of plausible policy options. Descriptions of the scenarios are presented below in Table SM3. Number of cases, days in hospital and deaths per country per scenario can be found in Table SM4.

Summary
We summarise the main parameters used in the estimates of health resources and costing. Further details and references are then provided in the following sections.
In summary, there are five steps in our calculations: 1. Calculation of unit costs per activity for three base countries: Ethiopia (low-income country or 'LIC'), Pakistan (lower-middle income country or 'lower-MIC') and South Africa (

.1 General Approach
A full economic costing was carried out over a one-year time horizon. Costs were constructed using a bottom-up ingredients-based technique. The costing was carried out from a health systems perspective and included both direct (e.g. medicines) and indirect costs (e.g. facility overheads). No above-service delivery costs were included.
The 76 countries chosen met three inclusion criteria: 1) classify as low-income, lower-middle income or upper-middle income by the World Bank (17), 2) be included in the list of 92 countries for which epidemiological modelling data was available from Pearson et al (2020) (2), and 3) have recent available GDP per capita (adjusted for PPP) data in order to carry out cost extrapolation between countries (17).

Intervention costs
We used official WHO guidance to identify areas related to critical preparedness, readiness and response actions for COVID-19 to define a set of interventions involved in a national response to the pandemic (18). We identified 6 priority areas of work and is further subdivided into 13 activities.
• For the first five areas of work we considered only WHO guidance to define the resource use. For case management costs we assumed less resource-intensive activities thought to be more plausible in low-and middle-income settings ('real-world'). Assumptions on 'real world' resource use were based on the clinical expertise of members of the research team and are detailed below.
Following this guidance on areas of work, we generated a list of activities for which we needed to estimate unit costs (see Table SM5). These unit costs were brought together with the COVID epidemiological model to estimate resource needs.

Defining inputs, inputs quantities and input costs
In order to calculate a unit cost for each of the abovementioned activities, we used an ingredients-based costing to identify a series of input required. For each input we estimated quantities needed and a country-specific price per quantity (see Table SM6). The costs of each input were identified using a range of sources, according to availability of recent primary cost data and appropriateness of cost estimates to the COVID-19 pattern of care. More details can be found below.
To obtain yearly costs per country, the unit costs below were then multiplied by the number of country-specific units (see Table SM12 for more details). Example: In the case of Emergency Response Mechanisms: National level (1a) we aim to calculate a cost per day. We assumed that the three inputs required per day are: (i) 10 junior-level government officials, (ii) 10 senior-level government officials, as well as (iii) meeting space and equipment for those 20 people. The salary for one day of work for one junior-level government official in Ethiopia was estimated at US$12.27, for one senior-level government official at US$17.29 and the cost of one day's worth of space and equipment necessary for meetings was estimated at US$13.18 per person. We multiplied inputs by prices: (US$12.27 x 10) + (US$17.29 x 10) + (US$13.18 x 20), which equals US$559.26. This represents the cost per day of the emergency response mechanism at the national level.
In order to determine the annual costs per country, this number was then multiplied by the total number of working days, assumed to be 260 (see Table SM12).

Input quantities
Activities 1-6: Quantities of working days required for planning and management and communication were estimated from expert consultation as part of the Disease Control Priorities 3-Universal Health Coverage (DCP3-UHC) project (19). For case finding, surveillance and diagnostic activities, quantities were estimated based on requirements for similar activities for tuberculosis (TB) such as contact tracing from the VALUE TB study and previous studies in South Africa (more below) (20,21).

Activities 7:
The number of days per patient in general ward and in ICU was set at 8 and 10 respectively and was set to match the assumptions in the epidemiological model (2,14,22). Following expert clinician advise we assumed that one-third of critical patient bed days would be treated the general ward and two-thirds in the ICU.
The likelihood of additional COVID-related complications (per day) were estimated using evidence on the clinical course of COVID from patients in Wuhan, China (23), as were assumptions on the duration of symptoms (24,25). The number of diagnostic tests per hospitalisation was carried out in consultation with expert clinicians in essential critical care.

Estimation of non-bed-day costs (Pakistan)
An ingredients-based approach was used to calculate most of the service costs and prices for Pakistan. The data used was collected as part of the Disease Control Priorities 3-Universal Health Coverage (DCP3-UHC) project (19). For other countries primary data from the TB studies was used (see below).
For Pakistan, staff-related costs were constructed using federal-level pay scales. For most outputs, the number of minutes of staff required per activity were estimated via expert opinion obtained from clinicians working in the Health Planning, System Strengthening & Information Analysis Unit (HPSIU) in the Ministry of National Health Services Regulations and Coordination of Pakistan. For outputs where this was unavailable, health economists agreed a plausible assumed value.
Drug regimens were costed using resource use data obtained through expert opinion (HPSIU) and a number of price sources. An assessment of strengths and weaknesses of different price sources was conducted and hierarchy of sources was established. The primary source of price data was the Sindh Health Department Procurement Price list. If a price was unavailable, the Federal Wholesale Price List for Generic Medicines was used as a second option. As a last resort, private sector market prices were used.
Cost on supplies and equipment were similarly constructed. Resource use was determined through expert opinion (HPSIU) and price source hierarchy established. The primary source BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) For all countries, for additional diagnostic and radiology costs (beyond those available from the TB data) were estimated using available literature and market prices. We assessed strengths and weaknesses of different price sources. For example, we used the 'Costing and Pricing of Services in Private Hospitals of Lahore: Summary Report' as our primary source as it contained a methodological appendix that suggested that an ingredients-based approach consistent with ours was followed. If some prices were unavailable we used user fees from the Pakistan Institute of Medical Sciences, procurement prices from the Medical Emergency Resilience Fund procurement prices and user fees from the Aga Khan University Hospital.
Space costs were estimated using data from budget documents from the Federal government (Islamabad Capital Territory Health Infrastructure PC-1).
Oxygen therapy costs per bed-day were calculated by estimating the number of cylinders consumed in 24 hours at different flow rates, assumed to be 10L per minute in the general ward and 30L per minute in the ICU. Cylinder duration (hours) was estimated by dividing pressure by the number of litres per minute, assuming a standard cylinder size of 4.6kg, filled at 1,900 psi pressure (26). Cost per cylinder was obtained from the South African online catalogue of a manufacturer that is active in both South Africa and Pakistan (27).

Estimation of non-bed-day costs (Ethiopia and South Africa)
For Ethiopia and South Africa the main source of cost data was the VALUE TB study (20,21). Cost data were collected from a health provider perspective to estimate the economic costs of TB-related health services. Full costs of health services were estimated. Cost data collection was retrospective, over a one-year period to minimize the risk of bias due to seasonality. Resource use was measured in the VALUE TB study using both top-down and bottom-up methods wherever possible, to allow for comparison. The costs included in the current cost model reflected an average of top-down and bottom-up costs by site. For South Africa, we also used primary data from the XTEND trial (nurses and lay health workers) (28).
Some of the COVID-19 interventions were outside the scope of the VALUE TB and XTEND studies. Values for which a primary unit cost was partially or entirely unavailable from Value TB are listed below. For these interventions, resource use data from Pakistan was used with local Ethiopian or South African prices. Where Pakistan health care inputs were applied to other settings, we classified them as tradeable or non-tradeable. For tradable inputs, where country-specific price estimates were not available from primary data or from the published literature, the estimate from Pakistan was applied to other countries. For non-tradable inputs, the estimate from Pakistan was adjusted by an amount reflecting the difference in the two countries' GDP (adjusted for purchasing power parity, or PPP) (see Table SM7). The rationale behind this approach is that, while exchange rate may be influenced by government policy, PPP seeks to equalise the purchasing power of different currencies and, as such, may better reflect differentials in non-tradable prices across countries. More details on this method of price adjustment can be found in Section 2.3. Staff costs did not need to be extrapolated as we had country-specific salary information for the three countries.

Estimation of bed-day costs (all countries)
We took an ingredients-based approach to estimating the costs of general ward and ICU ward bed days, as these were major cost drivers in our cost model. We estimated the plausible number of nursing hours per bed day in an LMIC setting through consultation with members of the research team who have expertise in critical care in LMICs. In ICU the assumption of nurse-to-patient ratio would be 1:1; in the general ward the ratio would be 1:6 during the day time and 1:20 in the night. To understand the full range of inputs required we obtained the underlying costing data set provided by the authors of a recent costing of hospital-based care (29). The paper reports the results of a detailed activity-based costing in a hospital in Karachi, disaggregated by phase of care. We used the cost data for the ward stay phase, removing any supplies or equipment specific to the surgery, to estimate the average generic costs of a bed-day.
All bed-day costs were compared to and validated against available country-specific estimates from the published literature and from ongoing research and WHO CHOICE (see Table SM8). Rapid literature searches were conducted on the Medline, Embase and EconLit databases on 8-9 April 2020 to identify records reporting on the costs of ICU care in each of the study countries.
We estimated the additional costs of ICU beds compared to standard hospital beds using an ingredients-based approach to cost the equipment and supplies not present in standard hospital wards. We used the procurement price of equipment and assumed depreciation over ten (ventilators and suction pumps) or five years (all other equipment). Supply costs included central and arterial lines, ventilator tubing, and sedatives.

COVID-19 specific costs
Finally, we calculated costs of supplies and inputs specific to COVID-19. Personal protective equipment (PPE) per health worker per day (see Table SM8) was calculated and allocated a cost per PPE per minute to clinical staff. We also calculated costs of hygiene per bed day (see Table SM9). We estimated the costs of PPE and hygiene supplies using a list of necessary supplies from a COVID-related budget from the Ministry of Health of Pakistan, which included local prices sourced by the Aga Kahn University. This was complemented for other countries using the WHO's Essential Supplies Forecasting Tool (ESFT) (30). We divided supplies into single-use and disposable. We determined plausible quantities and useful life for supplies following clinical guidelines and expert opinion.
Oxygen supplementation therapy is the main form of treatment for COVID-19. There are different methods of oxygen delivery which utilise different types of supplies, equipment and require different average levels of oxygen flow. We calculated costs for 6 types of oxygen delivery techniques and assumed a distribution across severe and critical patients according to members of our research team with clinical expertise in critical care in LMICs. Table SM10 shows the assumptions used in our model and how they differ from normative standards. BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s)

Extrapolation of unit costs in base countries to calculate unit costs across LICs, Lower-MICs and Upper-MICs
We used the unit costs obtained in our three base countries to extrapolate unit costs to other LICs, Lower-MICs and Upper-MICs. We grouped countries according to income group. Costs for LICs were extrapolated using unit costs from Ethiopia, costs for LMICs were extrapolated from the unit costs from Pakistan, and those for UMICs from the unit costs from South Africa.
In order to carry out the extrapolation, each cost ingredient for each of the unit costs was classified as a tradeable good, non-tradeable good, or staff cost.
Tradeable goods are generally defined as those that can easily be traded in the international market and include goods such as medical or other supplies and medications. The unit costs for our three base countries were initially converted from each local currency into 2019 US$ using market exchange rates. To convert the tradeable good from the base country (e.g. Ethiopia) to a 'second' country (e.g. Afghanistan) we apportioned the percentage of the unit cost that was composed of tradeable goods in 2019 US$ from the base country to the second country.
Non-tradeable goods include buildings, heavy machinery, and other equipment. To convert these costs from a base country to a second country we used purchasing power parity (PPP) conversion rates. We multiplied the proportion of the unit cost that was defined as nontradeable (in 2019 US$) by the ratio of the GDP per capita (adjusted for PPP) of the second county, divided by the GDP per capita (adjusted for PPP) of the base country. Data on GDP per capita (adjusted for PPP) can be found in the World Bank database (17).
To convert staff costs from a base country to a second country we used conversion rates from Serje et al (2018) (31). Serje et al (2018) use regression analysis on a dataset containing wages from health workers of different skill levels for 193 countries in order to predict wages by country income level relative to GDP per capita. We used the multiples per GDP per capita presented in the paper in order to convert the staff wages from the base country to the second country. See Table SM11.

Calculation of country-specific number of units per activity
The unit cost in each of the 76 countries was used to calculate the total costs per activity per country. Table SM12 shows the quantities that those unit costs were multiplied by in order to calculate the total costs per country, as well as their justification and source. Note: Scenario 1 modelled an unmitigated epidemic. Therefore, only activities marked with ‡ were included in calculating the costs for Scenario 1. Scenarios 2-4 included costs in all the activities mentioned in Table SM13.  (39): 1) total health expenditure including out-of-pocket payments, 2) total health expenditure excluding out-of-pocket payments, and 3) government health spending per capita. Data on GDP per capita and health expenditure per capita per country can be found in Table SM14.