By the same authors

From the same journal

From the same journal

Cost-Effectiveness Uncertainty Analysis Methods: A Comparison of One-Way Sensitivity, Analysis of Covariance, and Expected Value of Partial Perfect Information

Research output: Contribution to journalArticle

Published copy (DOI)

Author(s)

  • Jonathan Campbell
  • R. Brett McQueen
  • Anne M. Libby
  • Eldon Spackman
  • Joshua J. Carlson
  • Andrew Briggs

Department/unit(s)

Publication details

JournalMedical Decision Making
DateE-pub ahead of print - 27 Oct 2014
DatePublished (current) - Jul 2015
Issue number5
Volume35
Number of pages12
Pages (from-to)596-607
Early online date27/10/14
Original languageEnglish

Abstract

Objectives: To compare model input influence on incremental net monetary benefit (INMB) across three uncertainty methods: (i) one-way sensitivity analysis; (ii) probabilistic analysis of covariance (ANCOVA); and (iii) expected value of partial perfect information (EVPPI). Methods: Preliminary Model: We used a published cost-effectiveness model and assumed £20,000/QALY willingness-to-pay (Case 1: lower decision uncertainty) and £8,000/QALY willingness-to-pay (Case 2: higher decision uncertainty). One-way sensitivity analysis identified ten influential inputs. From these ten inputs, we estimated ANCOVA results (10,000 Monte Carlo draws) and EVPPI for each input (1,000 inner and 1,000 outer draws). We ranked inputs based on their influence on variation of INMB and compared input ranks across methods within case using Spearman’s rank correlation. We completed similar methods within three follow-up models with variation on the linearity of the models selected. Results: Preliminary Model: Case 1: The two most influential inputs were the same across all uncertainty methods, contributed 78% of variation in outcome (ANCOVA), and were the only inputs with non-zero EVPPI values. Case 2: The two most influential inputs accounted for 49% of variation in outcome (ANCOVA); all inputs had non-zero EVPPI values. The influential input rank order correlations across uncertainty methods within each case ranged from 0.70 to 0.99 (p-values < 0.05 for all correlations). Follow-Up Models suggested similar trends for more linear models with uncorrelated parameters, but less linear models with or without correlated parameters had lower input rank order correlations across uncertainty methods. Conclusions: Evidence across models suggest influential input rank agreement between one-way and more advanced uncertainty analyses for relatively linear models with uncorrelated parameters, but less agreement for less linear models with correlated parameters. Although each method provides unique information, the additional resources needed to generate and communicate advanced analyses should be weighed, especially when the outcome decision uncertainty is low and the model is relatively linear with uncorrelated parameters.

Discover related content

Find related publications, people, projects, datasets and more using interactive charts.

View graph of relations