Approaches to projecting future healthcare demand

TY - UNPB

T1 - Approaches to projecting future healthcare demand

AU - Matias, Maria Ana

AU - Santos, Rita

AU - Kasteridis, Panagiotis

AU - Grasic, Katja

AU - Mason, Anne Rosemary

AU - Rice, Nigel

PY - 2022/4

Y1 - 2022/4

N2 - Background: Existing projections of healthcare expenditure in the UK describe a wide range of possible spending futures. In part, these reflect uncertainties about growth in demand, but they also reflect differences in modelling approaches and in their underlying assumptions.The rise in healthcare demand, and its consequent impact on expenditures, has stimulated interest among policymakers in better projecting future healthcare needs to aid the management and organisation of healthcare resources. More accurate projections are expected to allow the healthcare system to adapt and prepare for future challenges. However, with a plethora of different and emerging methodologies and approaches to project future outcomes and events, it is increasingly challenging to select appropriate techniques for a given research objective such as the demand for health care, within a specific context such as the UK National Health Service (NHS).Objectives: This work provides a review and critique of four approaches to projection modelling: macro-level modelling, macrosimulation, microsimulation, and machine learning. Our critique assesses these different techniques in terms of appropriateness depending on the projection objective(e.g. the impact of policy changes, drivers of demand, expenditure projection), their development and implementation costs (data requirements, maintenance, development, and running times), predictive accuracy and model fit, ease of use (implementation and interpretation), transparency, and capacityfor future updates when required.Discussion: Each of the four modelling techniques has both strengths and limitations. For any given scenario, the choice among the techniques depends on the relative importance and weight placed on the particular objective, data requirements, and on the time horizon for the projection. For example, if the research objective is the long-term forecast of healthcare demand and expenditure, machine learning and macro-level models are likely to provide the most accurate models. However, if theobjective is to focus on the impact of policy changes and policy scenarios, macrosimulation or microsimulation models are more suitable. The choice of time horizon of the projection, even for longterm projections, is particularly crucial, since the forecast error in the factors explaining the growth of healthcare demand and expenditure will grow as the time horizon increases.

AB - Background: Existing projections of healthcare expenditure in the UK describe a wide range of possible spending futures. In part, these reflect uncertainties about growth in demand, but they also reflect differences in modelling approaches and in their underlying assumptions.The rise in healthcare demand, and its consequent impact on expenditures, has stimulated interest among policymakers in better projecting future healthcare needs to aid the management and organisation of healthcare resources. More accurate projections are expected to allow the healthcare system to adapt and prepare for future challenges. However, with a plethora of different and emerging methodologies and approaches to project future outcomes and events, it is increasingly challenging to select appropriate techniques for a given research objective such as the demand for health care, within a specific context such as the UK National Health Service (NHS).Objectives: This work provides a review and critique of four approaches to projection modelling: macro-level modelling, macrosimulation, microsimulation, and machine learning. Our critique assesses these different techniques in terms of appropriateness depending on the projection objective(e.g. the impact of policy changes, drivers of demand, expenditure projection), their development and implementation costs (data requirements, maintenance, development, and running times), predictive accuracy and model fit, ease of use (implementation and interpretation), transparency, and capacityfor future updates when required.Discussion: Each of the four modelling techniques has both strengths and limitations. For any given scenario, the choice among the techniques depends on the relative importance and weight placed on the particular objective, data requirements, and on the time horizon for the projection. For example, if the research objective is the long-term forecast of healthcare demand and expenditure, machine learning and macro-level models are likely to provide the most accurate models. However, if theobjective is to focus on the impact of policy changes and policy scenarios, macrosimulation or microsimulation models are more suitable. The choice of time horizon of the projection, even for longterm projections, is particularly crucial, since the forecast error in the factors explaining the growth of healthcare demand and expenditure will grow as the time horizon increases.

M3 - Discussion paper

T3 - CHE Research Paper

BT - Approaches to projecting future healthcare demand

PB - Centre for Health Economics, University of York

CY - UK

ER -