Forecasting for Decision Making in International Public Health
Vineet Prabhu, PhDHIV Market Intelligence, CHAI
Presented at:Improving the Response of Global Public Health in a Fast-changing WorldJoint UNICEF, UNFPA and WHO meeting with manufacturers and suppliers of in vitro diagnostic products, vaccines & immunization devices,finished pharmaceutical products, active pharmaceutical ingredients, contraceptive devices and vector control productsUN City, Copenhagen, Denmark
December 3, 2019
• Highlight challenges both demand and supply side face in serving public health needs
• Encourage both sides to work together – flexibility and transparency are key
• Urge continuous improvement including learning from other disease areas
Objective of presentation
(With apologies for the heavy reliance on HIV examples as that is the world I live in day to day)2
Fact:
All forecasts are wrong!!
(forecasters are simply aiming to minimize how wrong they are)
3
At the very outset, important to consider what decision the forecasting exercise is informing
Timescale of the forecasting exercise becomes paramount – next 12 mo. or 3-5 yrs from now?
DevelopmentCapacity
scale planning
SourcingActual
production
• What is the investment case for a manufacturer?
• What is the public health impact for potential funders?
Stage
Relevant Questions
• How quickly can production be scaled if demand truly takes off?
• What factory or machinery (and potentially regulatory) limitations exist?
• What lead times, shelf life, and inventory cost considerations exist for upstream raw materials?
• What lead times, shelf life, and inventory cost considerations exist for finished product?
Forecast Type
Market need sizing(e.g. PLHIV)
Addressable market(e.g. PLHIV on ART)
Demand forecast(e.g. likely market share)
4
For ensuring sufficient supply (and thus confidence on the demand side), rough order of magnitude of market size (need) is most relevant pre-commercialization
5
5K 50Kvs. vs.
Albeit actual production will be proportional to actual demand (i.e. orders) in real time, production capacitycannot be rapidly changed on a sliding scale of in-between figures; a major learning from LPV/r pellet roll-out
500K
Different types of forecasting methodologies – think about context for appropriateness
Different contexts call for different approaches to be taken
Consumption based Morbidity based
E.g. malaria drugs (ACTs)
• Often inappropriately taken when fever develops – easily bought over-the-counter
• Consumption may not have any relation to actual malaria prevalence
• Past trends/seasonality reasonable inference of “demand”
Self-diagnosis
Self-medication
Self-diagnosis
Self-medication
E.g. ARVs for HIV
• Will reflect programmatic targets for ART scale-up, retention, and treatment optimization
• Based on national prevalence, with ART clinics as gatekeepers
• Past consumption levels may only reflect a “floor” for demand
6
Multiple steps to introduce new product introduction, each with varying probabilities of success and timing across countries and even within a country in different years
7
High level of uncertainty created – any global forecast must be grounded in quality country intelligence following the 80/20 principle
Beware false precision!
Uncertainty comes from underlying data quality – unnecessarily complicated models won’t fix that.Understand the level of precision that is reasonably possible and be transparent about it.
1,424
Pediatric population (ages 0-14) living with HIV in 26 high-burden LMICs
Differences in CLHIV estimates would mask whatever assumptions one might make on ART coverage, 1st- vs. 2nd-line numbers, and individual API or formulation market shares
(<20K annual new infections by 2020)
(<20K annual new infections by 2022)
-
200
400
600
800
1,000
1,200
1,400
1,600
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027P
ed
iatr
ic P
op
ula
tio
n (
age
s 0
-14
) liv
ing
wit
h H
IV
"Business as usual"Scenario
Moderate Scenario
Super Fast-track Scenario
8
Source: Prabhu VR, McGovern S, Domanico P. IAS, July 2017, Paris. Oral abstract WEAD0203
• In many LMICs, record keeping is poor or requires (error-ridden) manual compilation for central level visibility
• Particularly acute for pediatrics where formulation and dosing is age and weight dependent
• Thus, it can be very difficult to have an accurate baseline view of the situation in country(s)
• Forecast output cannot be more precise than the baseline input
Success/failure in one programmatic area can affect market dynamics for downstream commodities
Don’t do things in a vacuum – underlying numbers for one commodity need to inform forecast(s) for related commodities
Increasing # on ART # of viral load tests needed
to monitor ART patients
(if CD4 testing is available)
# patients identified with Advanced HIV Disease
(AHD)
ARV procurement
(if done with more testing)
(if done with high yield testing only)
# of HIV RDTs procured
# of TB LAM dipsticks procured
# of CrAg LFA tests procured
TB treatment and prophylaxis procurement
CM treatment and prophylaxis procurement
9
• Funding – how much of the need can be supported? At what price point(s)?– e.g. underlying assumption on ARV forecasts is that drug treatment funding is the last thing to be cut if HIV funding
decreases vs. for oral pre-exposure prophylaxis (PrEP) for prevention in otherwise healthy clients?
– At its most basic level, (available funding) ÷ (unit cost) = number of units that can be bought
• Finding the patients/clients – do they naturally interact with the health system? – E.g. Zn/ORS for childhood diarrhea was scaled up through consumer marketing and retail shop sales
• How much demand generation is required?– Is a market being created from scratch (e.g. oral PrEP), or is it more a question of changing market shares to a more optimal
product (e.g. TLD rollout)?
• What infrastructure is required to support market growth?– e.g. number of molecular diagnostic tests (HIV/HCV/HPV/TB) that can be run is limited by number of labs/devices in country
and their accessibility
What are the constraints and enablers?
Market context is a critical qualitative input into any forecast, including informing where one lies on the market sizing vs. demand forecasting continuum
10
Consider the bigger picture of market dynamics that can affect your sub-market of interest
11
2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
New 3L PI Pts(post-DTG)
New 2L PI Pts(post-DTG)
New 2L PI Pts(post-NNRTI)
Existing 2L PI Pts
How durable will DTG be in 1L? How many years before we start to see failures?
How quickly will existing 2L PI patients be proactively switched to DTG?
Nu
mb
er o
f Pa
tien
ts o
n P
rote
ase
Inh
ibit
ors
Question of when not if PI market shrinks in short termQuestion of when not if PI market grows in long term
Protease Inhibitor (PI) Market Over Time with Dolutegravir IntroductionBefore dolutegravir
1st-line NNRTI
Question: what will be the market shares between different PIs given a constant flow of patients from 1st-line, which in turn in growing?
2nd-line PI
NNRTI: non-nucleoside reverse transcriptase inhibitor; PI: protease inhibitor; DTG: dolutegravir
Treatment failure
After dolutegravir (more durable than NNRTIs)
1st-line NNRTI 1st-line DTGProactive Switch Trea
tmen
t fa
ilure
1st-line DTG
2nd-line PI
Treatment failure
2nd-line PI
2nd-line NNRTI 2nd-line DTGTreatm
ent
failu
re
3rd-line PI
Question: what will the overall PI market look like?
Proactive Switch
Source: Prabhu VR, McGovern S, Panos Z. HIV Glasgow, Oct 2018, Poster 281
It is important to adapt and change how one forecasts for a given market rather than accept the status quo approach
• Clarity from demand side on what type of forecast is being presented and underlying uncertainty
• Transparency from suppliers on realistic production capacity, lead times etc.
Being flexible and understanding the constraints of the other side to create a win-win
DemandSupply
Supply Demand
Seeking actual order commitments to sustain and de-risk business investments
Seeking supply security and affordable prices before committing
Chicken and egg
cycle
12
Example of supply and demand side working together – ARV Procurement Working Group (APWG)
13
Quarterly Order Cycles Monthly Business Calls
Biannual Newsletters Quarterly Demand Forecasts Annual KPI reviews
Ad-Hoc Market Support Product Availability Dashboards
Consolidating orders and coordinating timing Market intelligence sharing Monthly calls with suppliers for challenging products
Broad dissemination of market info Best available picture of demand from countries Continuous improvement
Creating a common platform for understanding, communication, and transparency while maintaining confidentiality as appropriate
• Be clear on what decision is being informed by a forecasting exercise – this affects what level of precision is needed
• Understand interdependencies of commodities – you are talking about the same patient/client!
• Understand limitations of your data sources and beware false precision
• Don’t accept the status quo – push for continuous improvement in forecasting
• Learn new approaches from other disease areas
• While uncertainty will never go away, supply- and demand-side transparency can help greatly reduce it
Parting thoughts…
14
Acknowledgements
15
CHAI’s market shaping work on HIV commodities is made possible through the generous support of Unitaid, with complementary support from the
UK Department for International Development (DFID) and the Bill & Melinda Gates Foundation