Skip to main content Accessibility help
×
Home

OP53 Comparing Approaches To Univariate Sensitivity Analysis

  • Conor Teljeur, Patricia Harrington and Máirín Ryan

Abstract

Introduction:

Fully probabilistic analyses are now standard for economic models, with all parameters varied according to probability distributions. Using univariate sensitivity analyses to explore the influence of different parameters on the model results are also standard. Although there are several approaches available, there has been little discussion of the merits of each or justification for the method used in any given analysis. The aim of this study was to compare three approaches to univariate sensitivity analysis using a case study.

Methods:

We considered three univariate sensitivity analysis approaches: (i) set one parameter at its upper and lower bounds while all others are set at their mean value; (ii) analysis of variance; and (iii) set one parameter at its mean and vary all others. We compared these approaches using an economic model of mechanical thrombectomy for the treatment of acute ischemic stroke, considering outcomes of incremental costs, incremental quality-adjusted life-years (QALYs), and net monetary benefit (NMB).

Results:

For incremental costs and QALYs the correlation between the approaches was moderate to high, with correlation coefficients between 0.46 and 0.94. For NMB the correlation between approaches was also high (range 0.89 to 0.98), but some of the most influential parameters were ranked differently. Setting one parameter at its upper and lower bounds was the only method that facilitated an analysis of direction of influence.

Conclusions:

The three approaches addressed different but relevant questions. Setting individual parameters at their bounds is effectively a systematic scenario analysis and may be misleading to decision makers. Analysis of variance may be more easily interpreted, but it has disadvantages. Setting a parameter at its mean, while varying other parameters, is similar to value of information analysis. As with any sensitivity analysis, it is imperative that the uncertainty associated with each parameter is adequately captured in the model.

Copyright

Corresponding author

Metrics

Full text views

Total number of HTML views: 0
Total number of PDF views: 0 *
Loading metrics...

Abstract views

Total abstract views: 0 *
Loading metrics...

* Views captured on Cambridge Core between <date>. This data will be updated every 24 hours.

Usage data cannot currently be displayed