Value of Information Analysis

Value of information (VOI) analysis is a decision analytic framework used to quantify the expected value of reducing uncertainty by collecting additional data before making a decision. In the context of health technology assessment and evidence based policy, VOI analysis estimates how much the expected net benefit of the optimal decision would increase if the current uncertainty about key parameters were resolved. The analysis produces monetary estimates that can be compared to the cost of proposed research, helping decision makers determine whether conducting additional studies is worthwhile and, if so, which parameters should be the focus of future research. The two main metrics are the expected value of perfect information (EVPI), which represents the maximum value of eliminating all parameter uncertainty captured in the model, and the expected value of partial perfect information (EVPPI), which quantifies the value of resolving uncertainty about specific individual parameters or groups of parameters.

Research resources are limited, and not all uncertainty is equally consequential for decisions. VOI analysis provides a rational framework for prioritizing research investments by identifying the parameters whose uncertainty has the greatest impact on the expected value of the decision. If the population EVPI is smaller than the cost of additional research, there is no economic justification for delaying the decision to collect more data, and the best available evidence should guide action immediately. If the EVPI is large, VOI analysis identifies which parameters (treatment effectiveness, adverse event rates, quality of life values, costs) contribute most to decision uncertainty through EVPPI calculations, directing research resources where they will have the greatest impact. This approach transforms research prioritization from an intuitive judgment into a quantitative exercise grounded in decision theory.

VOI analysis builds on a probabilistic decision analytic model, typically a cost effectiveness model, in which uncertain parameters are represented by probability distributions. The expected value of perfect information is calculated as the difference between the expected net benefit of the decision made with perfect information and the expected net benefit of the decision made with current (imperfect) information. Computationally, EVPI is the expected value of the maximum net benefit across all decision options (calculated by identifying the best option for each simulated parameter set and averaging across all simulations) minus the maximum of the expected net benefits (the net benefit of the option that is best on average). Population EVPI multiplies the per patient EVPI by the number of people affected by the decision over a relevant time horizon (future patients usually discounted, so later years count for less), producing a monetary ceiling on the value of future research. EVPPI extends this by holding specific parameters at their true (simulated) values while other parameters remain uncertain, isolating the value of resolving uncertainty about each parameter or parameter group. EVPPI values for different parameters are not shares of the overall EVPI and generally do not add up to it. The expected value of sample information (EVSI) further refines the analysis by estimating the value of a specific proposed study design, accounting for the fact that a finite sample will reduce but not eliminate uncertainty.

Use VOI analysis after completing a probabilistic cost effectiveness analysis when the decision is uncertain, meaning that the probability that the intervention is cost effective is not close to 0% or 100%. VOI analysis is most valuable when the decision has large resource implications, affects many patients, and when additional research is feasible but costly. It has been used to inform research recommendations by health technology assessment bodies such as NICE, though it is not a routine requirement of their appraisals. EVPI should be calculated first to determine whether any further research is potentially worthwhile. If population EVPI exceeds the likely cost of research, EVPPI analysis identifies the priority parameters, and EVSI analysis evaluates specific study designs and sample sizes.

A common mistake is calculating EVPI but not EVPPI, which tells decision makers that uncertainty is costly without indicating where to focus research. Another error is using population EVPI without carefully specifying the effective population size and the time horizon over which the decision applies, which can dramatically inflate or deflate the estimate. Researchers sometimes confuse the EVPI with the probability that the current decision is wrong; EVPI quantifies the expected cost of uncertainty, not the probability of making the wrong choice. Computationally, nested Monte Carlo simulation for EVPPI can be extremely slow, and efficient methods can be used instead, such as regression based approaches for EVPPI and moment matching for EVSI. Failing to update the VOI analysis as new evidence becomes available wastes the potential of this dynamic framework.

A health technology assessment evaluates a new oral anticoagulant versus warfarin for stroke prevention in atrial fibrillation. The probabilistic cost effectiveness model estimates an incremental cost effectiveness ratio of $32,000 per QALY with a probability of cost effectiveness of 62% at a willingness to pay threshold of $50,000 per QALY. EVPI per patient is $1,450, and with an effective population of 200,000 patients (the discounted total of patients treated over a 10 year time horizon), the population EVPI is $290 million. EVPPI analysis gives a population EVPPI of $112 million for the long term risk of major bleeding with the new anticoagulant, and $38 million for treatment effectiveness on stroke prevention; these partial values are not shares of the $290 million and need not add up to it. This analysis suggests that a long term registry study focused on bleeding outcomes (estimated cost: $5 million) is potentially worthwhile, as the most that resolving this uncertainty could be worth far exceeds its cost. An EVSI calculation for a proposed registry of 10,000 patients followed for five years estimates an expected value of $42 million for the population, or about $37 million after subtracting the study's $5 million cost, confirming that the study is a worthwhile investment.