HIV Modelling Consortium

1

Daniel KeeblerDST/NRF Centre of Excellence in Epidemiological Modelling

and Analysis (SACEMA)Stellenbosch University, South Africa

HIV Modelling Consortium

How Should HIV Programmes Monitor Adults on ART? A Combined

Analysis of Three Mathematical Models

2

AcknowledgementsPaul Revill2, Scott Braithwaite3, Andrew Phillips4, Nello Blaser5, Annick Borquez6*, Valentina Cambiano4*, Andrea Ciaranello7*, Janne Estill5*, Richard Gray8*, Andrew Hill9*, Olivia Keiser5*, Jason Kessler3*, Nicolas A. Menzies10*, Kimberly A. Nucifora3*, Luisa Salazar Vizcaya5*, Simon Walker2*, Alex Welte1*, Philippa Easterbrook11, Meg Doherty11, Gottfried Hirnschall11 & Timothy Hallett6 for the HIV Modelling Consortium

*Listed in alphabetical order  2 Centre for Health Economics, University of York, York, UK3 School of Medicine, New York University, New York, US 4 University College London, London, UK5Division of International and Environmental Health, Institute of Social and Preventive Medicine (ISPM) University of Bern, Bern, Switzerland 6 Imperial College London, London, UK7 Massachussetts General Hospital, Boston, MA, US8 Kirby Institute, University of New South Wales, Sydney, Australia9 University of Liverpool, Liverpool, UK10 Centre for Health Decision Science, Harvard University11 HIV Programme, World Health Organization, Geneva, Switzerland

The question addressed

"How can available resources for HIV treatment and monitoring be allocated among alternative adult patient monitoring strategies to maximize health benefits in a population?"

3

The question addressed… considering:

the criteria used (clinical, immunological, virological)

the frequency of monitoring (6, 12, 36 months) the decision rules applied (when to switch to 2nd-

line)

Examining the: Benefits (Life-years saved, infections averted, DALYS averted) Costs (USD $ over short and long-terms)

4

Previous modelling work has been done on this question.

Compare alternative potential monitoring strategies “head-to-head” in each model, with standardized cost inputs.

Focus on costs in three settings: South Africa, Zambia and Malawi

Approach: model comparison

5

Models

1. Phillips AN, Pillay D, Miners AH, Bennett DE, Gilks CF, Lundgren JD. Outcomes from monitoring of patients on antiretroviral therapy in resource-limited settings with viral load, CD4 cell count, or clinical observation alone: a computer simulation model. Lancet. 2008; 371(9622): 1443-51. Epub 2008/04/29.

2. Braithwaite RS, Nucifora KA, Yiannoutsos CT, Musick B, Kimaiyo S, Diero L, et al. Alternative antiretroviral monitoring strategies for HIV-infected patients in east Africa: opportunities to save more lives? Journal of the International AIDS Society. 2011; 14:38. Epub 2011/08/02.

3. Estill J, Aubriere C, Egger M, Johnson L, Wood R, Garone D et al. Viral load monitoring of antiretroviral therapy, cohort viral load and HIV transmission in Southern Africa: a mathematical modelling analysis. AIDS. 2012; 26(11):1403-1413. Epub 2012/03/17.

6

Model Group Structure Type Time horizon

Patient benefit Transmission Adherence Resistance

HIV Synthesis1 UCL Individual

based Stochastic 15 years YES YES YES YES

Braithwaite2 New York University

Cohort model, Monte Carlo progression

Stochastic Lifetime YES NO YES YES

Estill3 University of Bern Cohort model Stochastic 5 years YES NO NO NO

Strategies investigatedAbbreviation ScenarioNM No switching or monitoring CM, S4 Switch on new stage 4 event CM, S3/4 Switch on new stage 3 or 4 eventCD4<100/S4 Switch if CD4 <100 or new stage 4CD4-CA Currently widely-used CD4 monitoring algorithmCD4/TGVL Switch if CD4 a) falls >50% from peak on treatment, b) falls below

baseline, with confirmation of viral load >1,000 copies.CD4/TGVL+ Switch if CD4 a) falls >50% from peak on treatment, b) falls below

baseline, with confirmatory VL >1,000, or new Stage 4, PLUS routine 12 monthly VL at 1,000 (CD4/VL mixed scenario)

VL36 36 monthly VL; switch if VL>1000VL12 12 monthly VL; switch if VL>1000 VL6 6 monthly VL; switch if VL>1000 VL6/VL>1K 6 monthly VL; switch if VL>10000 VL6/VL>5K 6 monthly VL; switch if VL>5000 VL6/VL>500 6 monthly VL; switch if VL>500

7

Results

8

9

HIV Synthesis: ZambiaRange of additional benefits modest

Six monthly VL greatest

impact. Cost per DALY

averted highClear pathway of

prioritisation

10

Cost-Effectiveness Frontier Plots for Zambia (Cost per DALY Averted, 2012 USD)

(a) Estill model; (b) Braithwaite et al. model; (c) HIV Synthesis model. Unfavoured (i.e. dominated/extendedly dominated; see Methods) strategies are shown in light grey while most efficient strategies are shown in black and their code is highlighted in bold. The frontier line that represents a most efficient pathway of spending as resources increase is shown in red together with the ICERs, i.e. the incremental cost per DALY averted of moving from one strategy to the next along the frontier.

Opportunity Costs

11

If ART coverage targets not met, resources allocated to increasing ART coverage would generate much greater benefits than if allocated to better patient monitoring.

Costs and benefits (DALYS averted) of alternative uses of resources (Braithwaite model). Results given are per 1 million HIV-infected persons with both benefits and costs discounted at 3%.

12

Opportunity Costs & Initiation Thresholds

13.6

13.8

14.0

14.2

14.4

14.6

14.8

15.0

15.2

CD4 monitoring, ART initiation at CD4 <350

CD4 monitoring, ART initiation at CD4 <500

Routine VL, ART initiation at CD4 < 350

Routine VL, ART initiation at CD4 < 500

DALYs Averted per Person

Monitoring Strategy at ART Initiation Threshold (ordered by mean cost of strategy per patient, 2012 US$)

Cost:$4,716

Cost:$5,973

Highest cost

Cost:$7,836

Cost:$9,495

Lowest cost

Costs per patient lifetime and DALYs averted from alternative uses of resources (Braithwaite model).

13

Reduced 2nd-Line, Test Costs

14

15• Includes net effect of prevention of unneeded switch and earlier switching in failing patients

16NB: Estill et al. not included owing to model structure

17

InterpretationNot all models could run all scenarios, complicating comparisons (e.g. VL6/500)

Bulk of model scenarios more sensitive to test costs than to 2nd-line costs in sensitivity analyses

But proportion of total costs taken up by tests & 2nd-line, and life-years on 2nd-line, varied between models

Virological failure may also impact on results: if overlap is higher, then virological monitoring may convey less useful information in targeted VL scenarios TGVL performed well in Braithwaite; not as well in Phillips Life-years as proxy for predictive value of CD4 VL failure

Conclusions: Modelling Results

18

Conclusions: Modelling Results (1)Regular viral load (every 6-12 months) provides the most

benefits both in reducing patient morbidity and mortality and in reducing HIV transmission in the population…

… but is the most expensive strategy for monitoring patients.

Infrequent viral load monitoring (every 36 months) or conditional viral load measurement (for confirmation of immunologic failure suggesting a need for second line drugs) would have an intermediate cost but may maximise health benefits for that budget. However, if infrequent/conditional viral load monitoring would

lead to machines being used at less the full capacity then these alternatives are less likely to be cost-effective.

19

20

Conclusions: Modelling Results (2)The question is not whether viral load provides greater

benefit to individual patients than CD4 or clinical monitoring.

The question is whether, given available resources, the opportunity costs in morbidity and mortality of forgoing the use of these resources for other efforts is acceptable.

In particular, it is expected that resources would produce greater health benefits if they were committed to increasing the coverage of ART to those in need.

Equity should also be a concern: lower-cost but less-effective monitoring strategies may reach a larger number of people than high-cost, more-effective strategies.

Conclusions: Linking Modelling and Operational

Research

21

22

Linking Modelling and Operational Research

While we expect the qualitative distribution of results to remain similar, when placed into real-world contexts, precise results given here will change

If I devote resources to patient monitoring, what benefits will it yield for program performance in my particular context?

The answer depends on…

23

Linking Modelling and Operational Research

Delays in getting results to lab, back from lab, and to the patient

Staff time & capacity for test processing & interpretation (can they do VL or CD4 better?)

Capacity for test coverage: centralization vs. de-centralization, lower-cost tests reaching more people

Role of monitoring in adherence interventions Program-specific ART coverage and feasible targets for

scale-up Willingness-to-pay per DALY averted (level of

resources) And more…

24

Integration of operational research and modelling is key to assessing program-specific impacts, within a framework of opportunity costs For interpreting and for generating models

Development of flexible modelling tools/user-friendly platforms that can incorporate all of the above, and be deployed to programs for their own use in planning, should be investigated

Linking Modelling and Operational Research

25

Linking Modelling and Operational Research

Modelling allows for the examination of strategies’ costs and health impacts over a longer timeframe than trials may allow, and without needing to withhold health interventions for the purpose of comparison

Modelling can provide a clear picture of which costs within a given intervention carry the greatest impact

Useful for advocates as well as program staff/operational researchers Models can inform programming and operational research, but

models must themselves be informed by this: What factors impact how technologies work in the real world? How

can these be incorporated into models? Interpreting and building models requires strong linkage between

modellers, operational researchers, implementers and advocates