TABLE SA-2: MULTI-VARIABLE SENSITIVITY ANALYSES

Background

“Multi-variable sensitivity analyses” examine the effects of changes in two or more variables at once. The uni-variable sensitivity analysis table and graphs help the user understand how variations in the estimated value of one input at a time can affect the final cost-effectiveness results. But what if the true values of two or more inputs at a time depart significantly from their estimated values? In that case the final cost-effectiveness estimate could deviate much further from the true value than would be indicated in the sensitivity analysis of any one variable. This table explores the combined effect of simultaneous “worse” and simultaneous “better” estimates for five key variables. These variables are:

· HIV prevalence
· VCT cost per HIV+ mother
· HIV/AIDS treatment cost
· Non-HIV infant mortality rate at 12 months
· Infant formula cost per kg

As in the case of the uni-variable sensitivity analysis (Table SA-1), the range of each input indicated by the values in “Worse CE scenario” through “Base Case Scenario” to “Better CE Scenario” were determined by the user elsewhere in the CET. The difference is that the values in “Cost per DALY” and “Cost per case averted” reflect the cost-effectiveness results obtained if all five variables simultaneously assumed the values shown in “Worse CE scenario”, “Base Case Scenario”, or “Better CE Scenario”.

While very helpful in determining what the extreme outcomes of the analysis could be, this type of multi-variable sensitivity analysis has a severe limitation - it does nothing to inform the user of the likelihood that the “Worse CE” or “Better CE” scenarios will actually occur. In fact the probability of the simultaneous occurrence of extreme values in five variables could be very small. If an intervention’s “Worse CE scenario” is still below $50 per DALY (or other cost-effectiveness threshold) one can have a high degree of confidence that the intervention will be cost-effective. On the other hand, if an intervention’s “Worse CE scenario” is not cost-effective, one must assess the probability of this scenario actually occurring. One might wish to isolate the inputs that have the greatest influence on the results and re-assess whether the low or high end estimates are in fact plausible for these inputs. The point is not to force any particular result, but rather to explore contingencies that have a reasonable likelihood of occurring.

AL12 - AN16. Range of input values. This area shows the value of “Worse cost-effectiveness scenario”, “Base case cost-effectiveness scenario” and “Better cost-effectiveness scenario” for each of the five inputs.

AR12 - AT16. Cost per DALY. This area shows the “Worse cost-effectiveness scenario”, “Base case cost-effectiveness scenario” and “Better cost-effectiveness scenario” expressed in dollars per DALY that correspond to the “Worse”, “Base case” and “Better” scenarios defined in AL12 - AN16.

AU12 - AW16. Cost per case averted. This area is identical in structure to the previous section “Cost per DALY” except that the results are expressed as cost per case averted.