Group discussion on selected HIV/AIDS costing and cost-effectiveness articles

1. Group discussion on selected HIV/AIDS costing and cost-effectiveness articles Michelle Remme (michelle.remme@lshtm.ac.uk) HIV Modelling and Economics, Social and Mathematical Epidemiology Group London School of Hygiene and Tropical Medicine Improving health worldwide www.lshtm.ac.uk

2. Why are these Vfm / efficiency measures useful? • UNAIDS Investment Framework calls for more efficient HIV investments – this will require evidence • Inputs into decisions on: • resource allocation (allocative efficiency – what interventions are most efficient where?) • programme optimisation (technical efficiency – how can these interventions be implemented more efficiently)

3. Prioritising most efficient interventions Source: http://www.givingwhatwecan.org/where-to-give/charity-evaluation/health/hiv-aids

4. Some useful resources • Recent systematic reviews: • Siapka et al (under review): unit costs and efficiency of 6 basic HIV/AIDS programmes • Guinness et al (under review): cost-effectiveness of harm reduction programmes • Beck et al (2013): cost and cost-effectiveness of HIV community services • Sweeney et al (2012): economies of scope/efficiency gains from integration • Galarraga et al (2011): unit costs of ART and PMTCT • Santa-Ana-Tellez (2011): costs of interventions for OVC • Galarraga et al (2010): cost-effectiveness of HIV prevention  comparable and quality data is limited • Databases/ compilation of cost/CEA data: • LSHTM/UNAIDS inventory • Futures unit cost database • Disease Control Priorities (2nd edition) • WHO CHOICE

5. Growing evidence base in LMICs Source: LSHTM/UNAIDS • How to interpret the growing evidence base for priority-setting and programming?

6. Next hour Learning objectives • Know what to consider when interpreting a cost or cost-effectiveness analysis • Understand how scale, scope, quality and costing methods can influence study results and conclusions Overview

7. Selected papers • Guinness et al – HIV prevention programmes for FSWs in India • Obure et al – Cost of HCT in Kenya and Swaziland • Marseille et al – Determinants of costs &cost-effectiveness of ART in Zambia • Rosen et al – Cost and outcomes of different ART models in South Africa • Menzies et al – Cost of PEPFAR ART programmes in 5 countries • Meyer-Rath & Over – Cost functions in modelling with TasP example • Shrestha et al – Comparison of costing methods for HCT in the USA*

8. Discussion points • What are the key findings? Pick one figure/table that summarises these best • What do these findings mean / why are they relevant for HIV policy and programming? • As an HIV policymaker or programme manager, (how) would you use these results? • What are some key methodological limitations? Drummond checklist • How do these limitations influence the interpretation of the study results?

10. (Guinness et al, 2005, WHO Bulletin)

11. Key issues • U-shaped relationship between unit cost and scale • Suggesting that efficiency gains can be realised by increasing scale until a cost-minimising point, which was found to be relatively low (1,000 – 1,700 FSWs reached) • Despite high share of variable costs in cost profile • Resource requirement estimates that assume constant unit cost are over- or underestimating the real cost of delivering services • Scale can be measured in different ways and may not always be modifiable by the programme, i.e. defined by duration of programme, reachability/size of target population • Limitations • Small sample size and variation in service package provided – is the output/service provided comparable? Is it of equal quality (effectiveness) across 17 NGOs? • Retrospective costing based on routine monitoring systems could have errors

13. (Obure et al, 2012, Sex Transm Infect)

14. Key issues • Variation in average costs per client counselled and tested in both Kenya and Swaziland suggest potential for efficiency gains • Kenya: PITC was delivered at a lower average cost and reached HIV-positive clients at lower cost than VCT (see Bautista-Arredondo et al, 2008, AIDS for targeting efficiency) • VCT in provincial and district hospitals more efficient at identifying HIV-positive clients • Swaziland: public facilities providing both PITC and VCT are more efficient at identifying HIV-positive clients than private (possibly more ‘voluntary’) • Variation in average costs driven by human resource costs, and in particular staff workload (little variation in salaries) – this can be explained by different models, i.e. Individual or group counselling, use of staff for multiple services • Limitations: • HIV-positive tests may not have been first tests  may be overestimating cost-effectiveness • Caution: VCT and PITC are not perfect substitutes

16. (Marseille et al, 2012, PLoS One)

17. Key issues • Determinants of costs • Large variation in average cost per ART patient-year, suggesting potential inefficiencies • On-site costs represented 2/3 of total average costs • Sites with higher adherence had higher unit costs (quality costs more) • Average costs decrease with scale, through effect on off-site costs • ART appears more efficient at hospitals  could this be partly scale or scope effect? Equity considerations? • On-site cost savings through rural/private sites can be offset by higher centralised support costs to these sites • Determinants of cost-effectiveness • Counterfactual in CEA is important: no intervention or existing intervention (CTX) • Complexity of patient case load increases cost, but not an inefficiency • Adherence may explain cost-effectiveness more than cost per se

19. (Rosen et al, 2008, Trop Med Int Health)

20. Key issues • 4 different models of ART delivery had different average costs per patient initiated (US$ 756 – 1,126) • likely due to scale, factor prices (access to public sector prices), scope (integration within primary care clinic or hospital), intensity of care provided (number of monitoring visits or lab tests) • For the quality-adjusted output, i.e. patient in care and responding, average costs were higher due to higher resource use (more drugs, lab tests, clinic visits) [US$ 903 – 1,210] • Average cost to produce a patient in care and responding (incorporating costs of patients no longer in care or not responding) is even higher (US$ 1,128 – 1,723) • If the quality of the treatment programme increases, total costs will increase, but the average cost of producing quality-adjusted outputs will fall  or cost-effectiveness/value for money will increase • Cost variation increases when patient outcomes are factored in • see interpretation of efficiency in Site 4 and Site 2, before and after adjusting for quality

22. Fig. 2. Change in median per-patient financial costs in successive 6-month periods, from start of HIV treatment scale-up in eachsite through 2006–2007 (2009 US$). (Menzies et al., 2011, AIDS)

23. Key issues • Annual per patient costs varied widely, esp. when ARV costs excluded: • Price differentials and different stages of programme development • Service package varied per site - supportive care could include OI prophylaxis & treatment, nutritional support, adherence and community-based interventions • Newly initiated ART patients (first 6 months) cost 15-20% more than established ART patients (50% more, excl. ARVs) –more frequent monitoring • Average costs higher for paediatric patients • Per patients financial costs dropped rapidly over 1st year (46.8% between 1st and 2nd 6-month periods) and additional 29.5% the following year • Costs decline as sites mature • Cost reductions most important for investment costs (scale effect) • Recurrent costs also drop, presumably due to learning by doing

25. (Meyer-Rath & Over, 2012, PLoS Medicine)

26. Key issues • Determinants of costs include: patient health status, treatment regimen, factor prices, programme/facility scale, facility experience, facility type, quality of care, technical efficiency • Most modelling studies (e.g. for ART scale-up) only consider patient health status and treatment regimen • Flexible cost functions incorporating various cost determinants would be more appropriate than current cost accounting identities • Replication of Granich et al (2012) model of TasP with a cost function that incorporates scale/level of delivery finds inefficiencies of small scale could add up to 42% to total future programme cost

28. (Shrestha et al, 2012, J Public Health Management Practice)

29. Key issues • Different costing methods lead to different estimates that can be up to 78% lower than microcosting (direct measurement) or 61% higher • Programme budget costing (where programme costs = total funding)  highest estimate • Gross costing (Medicaid payment for HIV testing used as unit cost)  lowest estimate • Policy implications: depending which methods is used, conclusions on a programme’s cost-effectiveness and eligibility for resource allocation could differ • Different methods can respond to different questions, but when comparing the relative value for money of interventions, it is important to understand if the costing methods underpinning the economic evaluation are comparable • Microcosting generally considered to be preferred, but underestimates waste as it assumes that unused resources are used by other programmes/ for other services

30. Concluding remarks • Determinants of costs and cost-effectiveness (from economic perspective) • Complexity of patient case mix • Factor input prices • Scale • Type/level of facility • Geographic setting (urban/rural) • Programme/site maturity • Targeting • Quality / adherence • Above service level costs • Discount rates • Costing methods • Gaps: • Economies of scope/integration • Above service level/health system costs • Programme management characteristics • Dealing with multiple (non-HIV) outcomes in cost-effectiveness analysis

31. Thank you http://blogs.lshtm.ac.uk/samemodellingandeconomics/ https://strive.lshtm.ac.uk