VP131 Comparison Of Variation Methods For One-Way Sensitivity Analyses

Abstract

INTRODUCTION:

A common approach to one-way sensitivity analysis is to vary inputs by a constant percentage. An alternative is to derive ranges using evidence-based probability distributions from published sources. Our objective was to compare one-way sensitivity analysis results when using these two approaches for a reference case model, along with two additional case studies.

METHODS:

For the reference case, we replicated a published Human Immunodeficiency Virus/Acquired Immunodeficiency Syndrome (HIV/AIDS) cost-effectiveness Markov model (zidovudine versus zidovudine plus lamivudine in the UK) using TreeAge®. Health states included three HIV/AIDS states and death. We generated one-way sensitivity analyses by varying inputs in two ways: (i) using ±15 percent for all inputs, and (ii) using the 2.5 and 97.5 percentile values of the evidence-based probability distributions for all inputs. Our outcome was the mean difference between lower and upper incremental cost-effectiveness ratios (ICERs) for each variation method for the ten most influential inputs. We assessed the number of inputs with a mean difference between lower and upper ICERs of >10 percent of the deterministic ICER.

RESULTS:

The deterministic ICER was GBP7,654/QALY (quality adjusted life year) for combination therapy versus monotherapy. The mean difference in ICER uncertainty for the evidence-based vs. ±15 percent variation method was GBP3,251/QALY (p = .0096). Six inputs had a mean difference in ICER uncertainty of >10 percent of GBP7,654/QALY (that is, mean difference in ICER uncertainty > GBP765) for the evidence-based variation method, compared to only two inputs for the constant percentage variation method.

CONCLUSIONS:

For the reference case, the magnitude of uncertainty in the outcome was larger for the evidence-based variation method compared to the constant percentage variation method. Evidence-based uncertainty in inputs should be used in all sensitivity analyses to reflect realistic uncertainty in an outcome and aid decision-making about future research strategies. Additional case studies will be presented using validated models in diabetes and asthma.