Index: Heath 2015 - Review of Expected Value of Information Methods

Heath 2015 - Review of Expected Value of Information Methods

Routing Summary

Fully captures Heath, Manolopoulou and Baio (2015), A Review of Methods for the Analysis of the Expected Value of Information (arXiv:1507.02513): the EVPI/EVPPI definitions, nested Monte Carlo, the single-parameter and regression estimators, and the 28-parameter influenza-vaccine comparison with Table 1 transcribed. 7 notes.

• Need the paper on one page or the section-to-note map? → Heath 2015 - Review of Expected Value of Information Methods - Overview
• Need net benefit, EVPI or the EVPPI formula (Eq. 1)? → Heath 2015 - Net Benefit, EVPI and EVPPI Definitions
• Need nested Monte Carlo, unit loss integral, or the linear-nuisance shortcut? → Heath 2015 - Nested Monte Carlo and Restrictive EVPPI Approximations
• Need the Strong-Oakley or Sadatsafavi single-parameter algorithms? → Heath 2015 - Single-Parameter EVPPI Strong-Oakley and Sadatsafavi
• Need EVPPI by GAM or Gaussian-process regression, or runnable code? → Heath 2015 - Regression-Based EVPPI (GAM and Gaussian Process)
• Need Table 1, Figures 1-2, run times, or the vaccine model? → Heath 2015 - Influenza Vaccine Case Study and Table 1
• Need which method to pick, limitations, or key references? → Heath 2015 - Discussion Limitations and Practical Guidance

Concept Map

Concept	Note	Type	Depends On	Key Result
Net benefit; known-distribution net benefit	Heath 2015 - Net Benefit, EVPI and EVPPI Definitions	definition	Decision Analysis	$n{b}_t=k{e}_t-c_t$
EVPPI (Eq. 1); EVPI	Heath 2015 - Net Benefit, EVPI and EVPPI Definitions	definition	net benefit	${\mathbb{E}}_{\varphi }[{\max }_t{\mathbb{E}}_{\psi \mid \varphi }N{B}_t]-{\max }_t\mathbb{E}N{B}_t$
Two-step Monte Carlo	Heath 2015 - Nested Monte Carlo and Restrictive EVPPI Approximations	definition	EVPPI	benchmark; about 55 min per parameter
Linear-nuisance approximation	Heath 2015 - Nested Monte Carlo and Restrictive EVPPI Approximations	definition	EVPPI	needs linearity and independence
Strong-Oakley blocking	Heath 2015 - Single-Parameter EVPPI Strong-Oakley and Sadatsafavi	definition	sorted PSA draws	EVPPI $\approx 0$ for small $M$, $\to$ EVPI for large $M$
Sadatsafavi segmentation	Heath 2015 - Single-Parameter EVPPI Strong-Oakley and Sadatsafavi	definition	sorted PSA draws	asymptotically unbiased at the true number of decision changes
Regression EVPPI	Heath 2015 - Regression-Based EVPPI (GAM and Gaussian Process)	theorem	EVPPI	$\frac{1}{S}\sum {\max }_t{\hat{g}}_t-{\max }_t\frac{1}{S}\sum {\hat{g}}_t$
Vaccine case study	Heath 2015 - Influenza Vaccine Case Study and Table 1	example	all estimators	all methods agree; ${\beta }_1,{\omega }_1$ top
Practical guidance	Heath 2015 - Discussion Limitations and Practical Guidance	concept	case study	GAM for 1 parameter, GP for $\ge 6$

Notes

• Heath 2015 - Review of Expected Value of Information Methods - Overview — CONTAINS: research question, contributions, key results, motivating PyMC examples (E87, E89, E90), Source Map.
• Heath 2015 - Net Benefit, EVPI and EVPPI Definitions — CONTAINS: Def. net benefit, known-distribution net benefit, $\varphi /\psi$ split, EVPPI Eq. (1), EVPI; ordering theorem with proof sketch; worked two-option example.
• Heath 2015 - Nested Monte Carlo and Restrictive EVPPI Approximations — CONTAINS: two-step MC, numerical integration, unit loss, linear-nuisance and re-parametrisation.
• Heath 2015 - Single-Parameter EVPPI Strong-Oakley and Sadatsafavi — CONTAINS: both algorithms step by step, bias-controlled $M$, cumulative-sum visual tool, code sketch.
• Heath 2015 - Regression-Based EVPPI (GAM and Gaussian Process) — CONTAINS: regression model, five-step estimator, SAVI/BCEA, verified Python example, PyMC sketch.
• Heath 2015 - Influenza Vaccine Case Study and Table 1 — CONTAINS: model, settings, full Table 1 (28 parameters x 5 methods), Figures 1-2, run times.
• Heath 2015 - Discussion Limitations and Practical Guidance — CONTAINS: decision guide, limitations, key references, checklist.

Sources

• Heath Manolopoulou Baio 2015 - A Review of Methods for the Analysis of the Expected Value of Information — arXiv:1507.02513v1 (9 Jul 2015), 13 PDF pages.

See Also

• Value of Information Concept Note - EVPI, EVPPI and EVSI — EVSI and the price-of-information view (concept note).
• Decision Analysis, From Inference to Decision, Expected Information Gain.