• To determine what belonged in the numerator and denominator of a cost effectiveness (C/E) ratio.
  
  
• How to measure resource use for the numerator, how to value health in the denominator.
  
  
• How to estimate the effectiveness of interventions.
  
  
• How to incorporate discounting and patient preferences.
  
  
• How to handle uncertainty.
  
  

Although welfare economic theory holds that health benefits should be determined by willingness to pay, the panel decided to use QALY's. Using this measure, health benefits accruing to people of different ages, or health states particularly affecting those with a lower socio-economic status, would not be affected by those underlying issues of status that affect payment (and hence not conflict with the Americans with Disabilities Act and other societal moral concerns).

What goes in the Numerator and Denominator?

First, anything that impacts human health or use of resources must go somewhere in the C/E ratio. The denominator of the ratio is reserved for improvement in health. Although willingness to pay for certain health benefits have in the past been used, and still are valid, that approach is not consistent with the standard set forth here.

The numerator should be a change in resource use - costs of health care services, costs of patient time, costs of care-giving (paid and unpaid), costs such as necessary travel or childcare expenses, economic costs borne by employers and other employees, and non-health care costs such as might be borne by the education system or other societal endeavors (e.g. welfare) as a result of the intervention.

Patient time spent accessing health care in pursuit of an intervention is valued in monetary terms. For the sake of consistency, time spent sick should be considered in the denominator as part of the health effect of the treatment under consideration.

Measuring Terms in the Numerator:

The costs here should reflect the difference between two therapies being compared. (One therapy could be doing nothing.) Fixed costs unaffected by the level of implementation should be excluded. However many costs described as "fixed" by the hospital are in fact variable in the long run (e.g. administrative costs) and should be included. Common approximations of cost include converting charges by using a hospital's cost-to-charge ratio (all hospitals have this), use of management accounting systems to estimate true costs (at Duke University, Transition 1 is an example of one such system), and use of third party payments for provider costs.

Costs Should Be Measured in Constant Dollars:

When the data covers more than one year, values from the earlier year must be inflated. Goods and services in the economy should be inflated by the CPI. Goods and services of medical care should be inflated by the medical component of the CPI.

For reasons of political correctness, wage information (routinely reported by age and gender) should not be so segregated as the lower pay of some ethnic and gender groups will decrease the value of an intervention for that group. Furthermore, average wage information discounts the value of time spent in leisure (retirees). Time lost from these activities should be valued as if the person were working.

One usually counts costs related to an intervention that occur during the years of life added - e.g. rx of strokes during years of life added from antihypertensive treatment. What about unrelated costs of rx during added years of life - e.g. preventing an MI but then having to treat cancer? These are usually small and can be excluded. However, where they might be large it is suggested to use a sensitivity analysis to figure out how much of an effect there is. Future non health care costs as a result of added longevity (consumption minus productivity) are not included.

Valuing Health Consequences - QALY's:

QALY's measure both longevity and the quality of that added longevity. The scale must be 0-1. Examples of such scales are given in the articles. Community preferences should be used as described in the first article. In particular, the SF-36, a common instrument used in medical outcome research is NOT to be used since it is not based on preferences. Patient preferences should be used only when a community rating is unavailable for that disease state.

Standard gamble and time trade off provide a way to scale preferences that are consistent and deemed appropriate ways to make the scale meaningful. However, any of a number of approaches may be used, and this area of contention is noted to be an area that might make comparisons among studies inexact.

QALY calculation should take into account average quality of life that is gender and age specific. It should not assume that a successful therapy leads to a perfect quality of life. This means that two equal therapies - one to a 20 year old population and one to an 80 year old population - will favor the 20 year old as the average quality of life is higher for that age group. In these instances, one should conduct a sensitivity analysis to estimate the effect of using the age specific quality of life rating, as one then gets into the ethical minefield of age discrimination by valuing therapies higher for a younger population.

Estimating Effectiveness of Interventions:

Essentially, all valid available resources should be used to estimate the anticipated effect of a health intervention. Randomized clinical studies may produce the most consistently reproducible results, but extension of an intervention to the uncontrolled population may obliterate the effect seen under controlled circumstances. Hence other types of studies should also be considered, especially meta analyses, population based modeling, etc.

Time Preference and Discounting:

Discounting to Net Present Value is commonly used. In simple terms, this is the concept of a pension plan. You invest now, and it grows into larger sums later. Therefore, a benefit that occurs in the future with a certain monetary value can be "bought" now with fewer funds (with the assumption those funds are invested and earning interest). The interest rate (after inflation) should be 3% by convention. Most studies have previously used 5% (and not always net of inflation). For the next ten years, values should be given with both 3% and 5% rates to allow comparisons to previous studies.

What about valuing health effects that occur in the future (e.g. added life span) versus health effects that occur now (e.g. improved quality of life immediately following cataract surgery)? It is worth more, presumably, to have immediate benefit. Health effects occurring in the future should be discounted as by an equivalent 3% to keep parity between health and monetary values.

Handling Uncertainty:

Estimates of effectiveness obviously can vary depending on the studies/sources consulted. The actual cure rates, the influence on quality of life, etc. can differ. Sensitivity analysis addresses this by ranging all estimates across a reasonable range for numbers used in the computation of CE. The recommended way is a 1-way invariate analysis. By this, they mean that the value for each particular variable in the CEA (life years added, quality of life after intervention, cost of the intervention, rate of complications associated with the intervention, etc.) is varied and the effect on the CE determined. If a value varied 10 fold (e.g. the rate of a rare complication of a therapy) but only minimally affected the CE value, then further research to pinpoint that value is not warranted. If a variable (e.g. such as a 25% increase in the cost of a particular intervention) significantly alters the CE ratio, a more exact cost accounting may be indicated to help define whether that intervention should be done or not.

This one way sensitivity analysis underestimates the severity of uncertainty as multiple factors may all work in the same direction. Therefore a multivariate analysis should also be done ranging values across reasonable ranges and generating a confidence interval for the CE ratio.

The panel recognized that it is not possible to get data for some of their suggested approaches. They noted that in the event shortcuts, expert opinion, or estimates were used, it was the responsibility of the analyst to defend any deviation from the ideal reference case and justify a sub-optimal approach.

Read Part 1 of this series.
Read Part 3 of this series.