ANNEX A: Modelling methodology

INTRODUCTION

Our analysis relies on original modelling that incorporates the best- available current data on AT need and the impacts of AT on users’ lives. Because disability and AT have long been low priorities on the international research agenda, there is limited empirical evidence on the potential health and economic benefits of AT, as well as the return on investment in AT provision. Given the data-poor environment, this model necessarily relies on simplifying and generalising assumptions.

This work aims to complement ongoing efforts in the AT space and to spur increased data collection and additional analysis. This analysis aims to expand the literature and provide new perspective to ministries of health, finance, and social protection, as well as global donors, on the need to broaden their efforts to provide AT. These results should be viewed as a strategic guide to decision makers.

The fundamental objective of the model is to estimate and bring increased specificity to the costs and associated benefits of delivering high-priority AT products in LMICs. The model assessed both the ‘critical path’ investments needed to strengthen systems for AT delivery and costs associated with AT usage, including end-to-end product delivery costs and the ongoing lifetime cost of AT for users. Though increased access to AT leads to social, economic, and health benefits—as described throughout the report—the model specifically calculates the benefits of economic and health improvements for users in quantitative terms.

APPROACH

Three parameters defined the scope of this work:

• Products: Eyeglasses, hearing aids, prostheses, and wheelchairs
• Geographical coverage: All lower- and middle- income countries (LMICs)
• Time: Remaining lifetime of all individuals in need of the four priority products alive today (55 years) 70

The products included in the analysis are a subset of WHO’s ‘Priority Assistive Devices List’ and represent four of ATscale’s five selected

priority products for increased utilisation. We selected these four for two reasons: a majority of people in need of AT require at least one of these four products, and the research into their potential benefits is relatively extensive. 71

A set of underlying global assumptions support the model. These include:

1. The lifetime costs of AT delivery (including initial procurement and ongoing servicing and upkeep) are based on today’s prices and do not change over time72
2. The products delivered are suitable for users in the local context— this implies negligible abandonment of devices
3. Ideal implementation of supportive systems and policies prior to product delivery—these programme costs are accounted for, but uptake estimates do not include lag time in systems strengthening, demand generation, or similar efforts
4. Global averages of demand and impact will provide reasonable estimates of cost-effectiveness and ROI in specific country-level environments
5. Distribution across age brackets is flat in all LMICs

METHODOLOGY

COHORT SIZING

The model is based on product-specific estimates of unmet need in LMICs. These figures are based on existing literature and account for total global need, the respective share of need in LMICs, and the relative rates of product delivery to date.

The model makes a simple adjustment to address potential double-counting across the four products. It roughly estimates the percentage of the adult population that would have received one product as a child and then a second later in life—for example, due to ageing. The economic and health gains portion of the model treat each AT received as its own case and therefore may overestimate the income gains for ‘multiple AT’ users. To find the number of adults needing more than one AT today, we first used the childhood need rate to estimate the share of the adult population that first needed AT as

a child. We then applied the rate of ‘novel’ adult AT need to the same population to get a rough estimate of the overlap in populations for all product combinations. This gave an estimate of approximately 2.5% of the total cohort requiring two products. While this same approach could be applied again to find those needing >2 products, we assumed that this population would be negligible relative to the total cohort.

For any individual using more than one AT, we assumed that the model would overstate their marginal income gains from AT by approximately 10%. Applying this 10% to 2.5% of users requiring two ATs suggests that without an adjustment for double counting, the model would overestimate the total benefits by approximately 0.25%. In order to make a conservative adjustment and avoid overstating the potential benefits of AT, we rounded this up and applied a flat 1% reduction to all economic and health benefit estimates. We also incorporated the 1% adjustment into the estimation of case-finding costs (assuming fewer individuals to identify).

The cohort sizes used are shown in Table 3.

Table 3: Estimated cohort sizes, by product and age group 73

	CHILDREN	ADULTS
HEARING AIDS	4 million	50 million
PROSTHESES	5 million	30 million
EYEGLASSES (prescription)	20 million	110 million
EYEGLASSES (readers)	-	720 million
WHEELCHAIRS	10 million	50 million

DISCOUNT FOR TIME TO STRENGTHEN AT DELIVERY SYSTEMS

One of ATscale’s primary areas of work focuses on improving the enabling ecosystem to support sustainable and effective delivery of AT products and services to those in need. While this work, along with more targeted market shaping efforts, is ongoing, we assumed for the purposes of modelling the economic and health benefits that it would take some time for full implementation. Therefore, we first modelled the maximum possible benefits that each AT user (across both children and adults) could accrue in a given year. Then, we assumed that they would accrue 35% of this maximum benefit in the Year 1 (2020), and that the share of the maximum accrued in each subsequent year would increase linearly up to 100% in Year 15. The benefits then continue to accrue at 100% of the maximum in each following year, assuming full development of the AT delivery system.

BENEFITS

ECONOMIC IMPACT

The economic impact of AT is threefold: i) increased rates of employment and productivity (affecting adult users as well as children once they reach working age); ii) improved educational outcomes (affecting child users); and iii) unpaid family support providers taking up more paid work. We modelled each of these components separately and then aggregated them across the three groups. In all cases, we use GDP per capita as a benchmark for average LMIC earnings and adjust it to account for economic growth and inflation.

These factors are all influenced by ‘disability severity’; the model uses QALY weights as proxies. The severity of their disability in large part dictates the extent to which AT users are better able to access jobs, attend and succeed at school, or reduce their reliance on family support providers. The model uses the product-specific QALY weightings (see ‘Health impact’ below for more on QALYs) as quantitative proxies of disability severity throughout the economic estimates.

We based the estimate of increase in employment and productivity on previous work by the ILO. The approach accounts for changes in both willingness / capacity to work (workforce participation) and ability to obtain a job (employment and unemployment rates) based on disability severity and the impact of AT. The model estimates the total earnings gains based on the following formula: 74,75

where N = # AT users, V = GDP per capita, y = income adjustment factor, ß = disability severity, e = employment rate, u = unemployment rate, d = inactivity rate, i = AT product, * = post-intervention

Note: This assumes i) average employment statistics can be applied to estimate LMIC-wide shifts, and ii) working life spans from the ages of 18 to 64.

The educational component is based on the impact of increased schooling on lifetime earning potential. Research has shown that each additional year of schooling is linked to a 10% increase in personal earnings. 11 The model scaled each year of school for which a child had AT by the relative increase in ability to attend and perform to estimate the effective number of increased school-years gained. Due to the limited available data regarding the impact of AT on education attendance and performance, the model again used QALY weight values as proxies for increased ability to attend and learn. We then multiplied the result in order to estimate lifetime earnings gains. The following formula describes this calculation: 76

where N = # children using AT, V = GDP / capita, r = percent earn- ings increase per year of schooling, T = total years of schooling while using AT, Q = QALY weight, i = AT product, * = post-intervention

Finally, the model estimates the economic gains of family supporters who take on additional paid work outside the home. We assume that the increase in quality of life and independence (assessed by proxy according to QALY weightings) leads to a proportional reduction in need for dedicated support from family members. In turn, those previously providing support may then be able to pursue part- time or full-time employment. Because data on support providers in

LMICs are limited, largely due to the difficulty in estimating the informal family supporter population, the model uses U.S. benchmarks scaled up to the level of AT need in LMICs.

Supporters’ income increases derive from their pre- and post- intervention employment statuses. The model differentiates between those working part-time (‘high’ = 25 hours per week; ‘low’ = 15 hours per week) or not working (zero hours per week) pre- intervention and then accounts for changes between groups post- intervention. Only three of these transitions produce employment gains: i) no work to low part-time work, ii) no work to high part- time work, and iii) low part-time work to high part-time work. We determined the allocation to each category based on average employment statistics for the AT users and the severity of the user’s disability (again using QALY weights as a proxy). 77

The following formula captures the estimated annual income gains from this increase in paid work:

where CG = caregiver, N = # support providers, V = GDP / capita / hour, p = share of support providers in category, ∆h = change in hours worked per week, i = AT product, j = support provider employment group

The combined economic gains due to adult, children, and family supporters having the opportunity to perform additional paid work come to nearly USD 10.5 trillion. Table 4 below breaks down economic benefits by product and population group.

Table 4: Breakdown of modelled economic benefits

	HEARING AIDS		PROSTHESES		EYEGLASSES		WHEELCHAIRS
	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS
Cohort size	4 million	50 million	5 million	30 million	20 million	830 million	10 million	50 million
Avg. lifetime gains per user	~USD 59,500	~USD 2,800	~USD 246,300	~USD 8,400	~USD 76,800	~USD 4,200	~USD 106,200	~USD 8,100
Total lifetime user gains	~USD 200 billion	~USD 100 billion	~USD 1,200 billion	~USD 300 billion	~USD 1,700 billion	~USD 3,600 billion	~USD 1,000 billion	~USD 400 billion
Total lifetime user gains across products	~USD 8,500 billion
Total family supporter gains	~USD 70 billion	~USD 110 billion	~USD 180 billion	~USD 150 billion	~USD 160 billion	~USD 740 billion	~USD 330 billion	~USD 140 billion
Sum of family supporter gains across products78	~USD 1,900 billion
Total economic gains15	~USD 10 trillion

HEALTH IMPACT

The health impact assessment is based on the quality-adjusted life year (QALY). This is a standard metric used to capture changes in AT users’ reported quality of life, despite a lack of direct change to their underlying physical condition as a result of receiving AT. 79 Existing literature provides data on users’ reported quality of life before and after receiving either of the four assistive products. These data tracked changes in quality of life over the residual life expectancy, beginning with the average age of receiving each AT product (addressing adults and children separately), in order to estimate the total gain in QALY.

The QALY weightings are based on pre- and post-intervention EQ-5D values (a standardised instrument used to measure health status) available in the existing literature. 80, 81, 82, 83 Given the data-poor environment, some of these estimates came from small-scale or localised studies. The QALY weighting values used in the model are shown in Table 5 below.

Table 5: QALY weightings by product (EQ-5D)

	Pre-Intervention	Post-Intervention	Difference
HEARING AIDS	0.830	0.853	0.023
PROSTHESES	0.398	0.724	0.326
EYEGLASSES (prescription)	0.895	0.961	0.066
EYEGLASSES (readers)	0.915	0.961	0.046
WHEELCHAIRS	0.537	0.638	0.101

The health model does not account for changes to mortality or health system expenditure. Desk research and expert interviews indicated that there is insufficient empirical evidence directly linking AT uptake to reductions in mortality. Similarly, evidence examining the impact of increased access to AT on health care expenditures was inconclusive. This may be due to difficulty in assessing the effects of two opposing and uncertain forces: i) increased access to AT helps users overcome some barriers to accessing health services, thereby increasing utilisation of services, and ii) increased use of preventive services reduces incidence of serious complications, which are associated with more costly reactive emergency treatments.

We estimate the gain in QALYs using the following equation: 84

For the case of AT, in which there is no change in life expectancy post- intervention, the equation reduces to:

where N = # users, Q = QALY weight, l = residual life expectancy, r = discount rate, i = AT product , * = post-intervention

This approach yields a final estimated gain of 1.3 billion QALYs. Table 6 below breaks down the change in QALYs by age group and product.

Table 6: Breakdown of modelled health benefits

	HEARING AIDS		PROSTHESES		EYEGLASSES		WHEELCHAIRS
	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS
Total QALYs	~3 million	~20 million	~40 million	~170 million	~40 million	~950 million	~30 million	~90 million
Total QALYs across products	~1.3 billion
Cohort size	4 million	50 million	5 million	30 million	20 million	830 million	10 million	50 million
Avg. QALYs / user	~0.6	~0.4	~8.9	~5.2	~1.8	~1.1	~2.7	~1.8
Avg. QALYs / user across products	~1.3

COSTS

As described above, the costing analysis consists of two components: initial investments critical to ensuring that systems are fully supportive and structured to effectively deliver appropriate AT and the user-incurred costs of accessing and receiving assistance.

Estimating the fixed investment costs to strengthen AT delivery systems is important in order to capture the full and realistic requirements for sustainable AT delivery. However, cost estimates for these activities (which may include policy change, advocacy, public awareness raising, and stigma reduction) are subject to significant uncertainty as needs may be highly country-dependent and / or vary with regard to cost and efficacy (e.g. stigma reduction, demand

generation, etc.). Meanwhile, these costs are likely very small relative to those for ongoing service delivery. Therefore, the model takes a single line-item estimate for the total cost of activities to strengthen systems for AT delivery. We fixed this value at USD 10 billion—a conservatively high estimate of the total required costs for most core activities.

The user-incurred costs begin with one-off case-finding activities. These costs are based on benchmark estimates from comparable health interventions, scaled to the prevalence of unmet need for each AT product. 85, 86 These benchmarks considered health worker wages, transportation, field training, and screening and diagnostic tests.

Users then experience additional recurring costs over the rest of their lifetimes. Across the support provision pathway, individuals typically require appointments for initial referral to a specialist, detailed assessment, AT fitting and training, and subsequent regular follow-up and servicing. 87 Meanwhile, equipment needs typically include the device itself and ongoing replacement parts (depending upon equipment type and usage patterns). To estimate procurement costs, the model uses estimated LMIC market prices for each product. 88 To approximate the cost of delivery, fitting, and training we used WHO estimates of outpatient costs for primary-level hospitals in selected LMICs. 89 Given the lack of data on AT-specific delivery channels, this estimation aimed to reflect the cost of health worker time and any tests necessary to accurately diagnose and prescribe appropriate AT to individuals who need it. These costs, as well as servicing and maintenance, recur over the lifetime of the equipment, starting at an estimated ‘midpoint age’ of receiving AT. 90

The model addresses the recurrent costs for adult and child users separately. The model weights total unmet need between adult and child populations, and accounts for the difference in lifetime needs between the two groups. It then sums across the two cases to find the total cost of meeting today’s unmet need.

Combining these elements, the model follows this formula:

The model does not account for other potential secondary costs of accessing AT. These may include potential time missed from work for health appointments, travel to and from the health centre, and more. However, these costs are expected to be small relative to the others described above, and data estimating these costs are limited. Therefore, secondary costs have not been included in the analysis.

The formula yields an estimated total required investment of approximately USD 730 billion over the cohort’s lifetime. Table 7 below breaks down these costs.

Table 7: Breakdown of modelled cost 91

	HEARING AIDS		PROSTHESES		EYEGLASSES		WHEELCHAIRS
	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS	CHILDREN	ADULTS
Lifetime cost per person (USD)	~7,200	~2,400	~14,300	~5,200	~1,200	~100	~5,300	~2,500
Cohort size	4 million	50 million	5 million	30 million	20 million	830 million	10 million	50 million
Total user costs (USD)	~30 billion	~130 billion	~70 billion	~170 billion	~30 billion	~130 billion	~50 billion	~120 billion
System strengthening costs	~USD 10 billion
Total costs across products and system strengthening	~USD 730 billion

ROI

We estimated the final roi using the following equation:

This includes a summation of the economic benefits across all three drivers. It excludes the health and social benefits also described above, meaning the true ROI (including both financial and nonfinancial benefits) could be even higher than the value estimated here implies. The dollar values of both the benefit and cost components are also discounted according to their net present value92 with a discount rate of 5% over the AT users’ remaining lifetimes (55 years from start). Given total discounted costs of USD 400 billion, yielding total discounted benefits of USD 4.1 trillion, the model gives a final ROI of approximately 9:1.

SENSITIVITY ANALYSIS

Following completion of the modelling as described above, we conducted a sensitivity analysis to understand the impact of variation in key input parameters on the final ROI output value (using ROI as a composite measure of multiple other intermediate outputs in the model).

We conducted a basic one-at-a-time sensitivity analysis on a set of 18 key parameters, covering components of cohort sizing, costing, benefit accrual, and others. We independently varied each parameter to upper and lower bounds and recorded the impact on the output ROI value. To normalise the findings across parameters, we measured results as the ratio of the percentage-change in ROI over the percentage-change in the parameter value. The formula for this calculation was as follows:

where R = baseline ROI value, dR = change in ROI, P = baseline pa- rameter value, dP = change in parameter value

The analysis showed that four variables have significant impact on the ROI outcome: retirement age of AT users, the pre- and post- intervention QALY weighting values, and life expectancy at birth for users that receive AT during childhood. For example, a one-percent change in retirement age leads to a nearly-five-percent shift in the overall ROI. This affirms that the model outputs are most strongly determined by the total time over which users accrue benefits (with each additional year of work and life contributing heavily to total economic gains, and outweighing the additional cost of maintaining the AT over that time) and the degree of benefit derived from receiving AT (with greater QALY differentials leading to greater impact in school and in the workplace).

Figure 6 below shows the percentage change in the economic output figure per one-percent change in the input parameter.

Figure 6: Sensitivity analysis findings

Sensitivity of ROI output based on input parameter variation

Ratio of percent-change in ROI to percent-change in input parameter value