Sensitivity in Cost Effectiveness Analyses Sensitivity of a

Sensitivity in Cost Effectiveness Analyses Sensitivity of a CEA on prevention strategies against Neural Tube Defects using DPL software. Radboud Duintjer Tebbens Sensitivity in CEA's

Contents • • Objectives Problem Description The Model Implementation and Deterministic Results Sensitivity Analysis Probabilistic Model Conclusions 9/16/2020 Sensitivity in CEA's 2

Objectives • Exploring the possibilities of DPL for sensitivity analyses • Getting familiar with Cost Effectiveness Analyses • How can sensitivity analyses on CEA’s be improved? 9/16/2020 Sensitivity in CEA's 3

Problem Description • What is a Cost Effectiveness Analysis? Analysis that compares health interventions using C/E ratio = Incremental Cost/Incremental Effectiveness = Incremental Cost / Incremental QALY’s QALY: Quality-Adjusted Life Year 1 QALY = 1 year in full health 0 QALY = 1 year dead 9/16/2020 Sensitivity in CEA's 4

Problem Description (2) • The CEA on Neural Tube Defects (NTD’s) considers 5 health programs for prevention of NTD’s on newborns: 1. No Program (baseline) 2. Supplementation: education program to stimulate women aged 15 -45 to take Vitamin Supplements 3, 4&5. Fortification with folic acid: a rule that forces food producers to fortify cereals with folic acid, 0. 14 , 0. 35 or 0. 70 mg per 100 g of cereals • The CEA on neural tube defects is described in the appendix of the book: Costeffectiveness in Health and Medicine, M. R. Gold et. al, Oxford University Press 1996 9/16/2020 Sensitivity in CEA's 5

Problem Description (3) • Adverse Effect in Fortification Programs: intakes of more than 1. 0 mg of folic acid can cause delayed diagnosis of vitamin B 12 deficiency which can lead to neurological complications, mostly by elderly persons 9/16/2020 Sensitivity in CEA's 6

The Model Simple deterministic model that: • Calculates the number of babies born with NTD with each program • Calculates the number of Adverse Effects caused by any of the programs • Adds costs of the various programs and their health outcomes • Uses expert judgment outcomes to evaluate the number of QALY’s gained compared to the baseline case ‘no program’. 9/16/2020 Sensitivity in CEA's 7

The Model (2) • Important Uncertain Parameters: parameter QALY’s per NTD Costs per NTD QALY’s per Adverse Effect Costs per Adverse Effect p. NTD 0. 4+ p. NTD 0. 4 p. OAdv. Eff p. Beg. Supp 9/16/2020 value 12. 591 180972. 8 0. 31 3897. 38 0. 000035 0. 000085 0. 000023 0. 15 Sensitivity in CEA's sensitivity range 5 -20 160 000 - 200 0. 20 - 0. 40 3500 - 4300 0. 000021 -0. 000049 -0. 000007 -0. 000047 0. 0 -1. 0 8

Model (3) • p. NTD 0. 4+ is the probability that a women aged 15 -45 who consumes at least 0. 4 mg of folic acid daily will have an NTD affected pregnancy. • p. NTD 0. 4+ is the probability that a women aged 15 -45 who consumes less than 0. 4 mg of folic acid daily will have an NTD affected pregnancy. • The difference between these variable accounts for the number of prevented NTD’s by the health intervention. • p. OAdveff, the probability that someone who is not a woman aged 1545 and who consumes more than 1. 0 mg of folic acid daily, eventually quantifies the occurrence of neurological complications 9/16/2020 Sensitivity in CEA's 9

Implementation and Deterministic Results Straightforward Influence Diagram with • 1 Decision Node: Program • Many Value Nodes containing all model parameters • Objective function to be minimized: Incremental C/E = Costs/QALY’s Gained NB: I used Costs instead of Incremental Costs 9/16/2020 Sensitivity in CEA's 10

Implementation and Deterministic Results (2) • Deterministic Results: Program No Program Costs(M$) 710 QALY’s lost Incremental C/E 49 398 - Suppl. Fort. 0. 14 Fort. 0. 35 Fort. 0. 70 724 706 636 503 45 642 48 373 43 119 32 144 9/16/2020 Sensitivity in CEA's 195 089 688 821 101 333 31 106 11

Sensitivity Analysis • Univariate: Rainbow Diagrams or Tornado Diagrams • Many Univariate Sensitivity Analyses at once: Tornado Diagrams • Multivariate? • (See accompanying DPL file: NTDdeterm. da) 9/16/2020 Sensitivity in CEA's 12

Probabilistic Model • Not conceptually very clear • Use of Risk Profiles and probability distributions, but very basic • No improvement for sensitivity visualization • (See accompanying DPL file: NTDprob. da) 9/16/2020 Sensitivity in CEA's 13

Conclusions • DPL is good for a first look at sensitivity of a model on its parameters. • First look at CEA on NTD prevention shows that the decision based on the C/E ratio is very little sensitive to most uncertain parameters, even though the cost and health outcomes are quite sensitive to them. 9/16/2020 Sensitivity in CEA's 14

Conclusions (2) • Multivariate sensitivity analyses for CEA’s? • It would be interesting to look at uncertainty analysis for CEA’s, taking into consideration the (joint) probability distribution of the uncertain parameters (but is it necessary? ) 9/16/2020 Sensitivity in CEA's 15