University of South Carolina, Arnold School of Public Health, HSPM J712, updated Sept. 2012

Copyright © 1999-2000 Samuel L. Baker

New! Added comment about cost-saving and cost-effectiveness.

Cost-Effectiveness Articles Sound applet

1. Eddy, "Screening for Cervical Cancer"
2. Fries, "Reducing Health Care Costs by Reducing the Need and Demand for Medical Services"
3. Russell, "Prevention"
4. Huntington,"For Every Dollar Spent -- The Cost-Savings Argument for Prenatal Care"
5. Eddy, "Oregon's Methods: Did Cost-effectiveness Analysis Fail?"

Eddy, "Screening for Cervical Cancer"

Neuhauser (stool guaiac study discussed in preceding lecture) method applied to real example. Marginal cost per year of life saved by PAP test rises quickly as testing gets more frequent.

Eddy -- data sources and false positives

Eddy uses IARC's best false positive rate of 0.5%, which is achieved in countries with
a centralized government-run lab system. Big controversy broke in 1990 over US lab accuracy, with false positive rates as high as 5%.

False positives lead to expensive further tests as well as anxiety for the woman. Even at an 0.5% false positive rate, a woman who gets 20 tests in her lifetime is 4 times as likely to get a false positive as to actually have cervical cancer.

Eddy -- cancers found and years of life saved

Eddy reports years of life saved rather than lives saved. This makes saving a younger woman more valuable than saving an older woman.

He calculates that women who get screened every four years live an average of 93.8 days longer than women who don't get tested at all.

Increasing the frequency of screening increases life expectancy, but the gains are smaller. E.g., going from testing every 4 years to testing every year increases life expectancy by about 5 days.

Eddy -- years of life saved at different frequencies of testing

Eddy reports how many days of life are saved on average at various frequencies of testing. The number in each second row cell is how many days of life are saved by testing at that column's frequency, compared with not having any tests at all.

Eddy discounts future years of life savedSound applet

In the third row above, Eddy discounts the future years of life saved at a 5% discount rate.

Eddy bases his numbers on a hypothetical 20-year-old woman looking forward to the rest of her life, trying to decide how often to have PAP tests.

Sound applet
Suppose a 20-year-old woman decides to get regular PAP tests. One possibility is that, when she is 70 years old, a cancer will be found that would have caused her death at age 75. The early detection allows the woman to live to be 80. Those 5 years of life saved are 55 to 59 years in the future, so Eddy says they are only worth this much today to the 20-year old woman:

This totals about 0.31. Saving 5 years of life 55 years from now is valued at 0.31 years today, by this methodology.
(If the formula above is mysterious to you, please take time out to look at the interactive lecture on discounting future income.)

Sound applet
Suppose instead that her first PAP test finds a cancer that would have caused her death at age 21. Now Eddy would calculate the value this way:

These 59 terms add up to 18.9 years. 59 years of life saved are valued at only 18.9 years, when the future years are discounted to the present.

Sound applet
To sum up: Discounting the future years of life saved, using this methodology, has two effects:

1. It makes the savings total smaller, so the cost per year of life saved looks bigger.
2. It puts even greater relative value on saving a young person's life, compared with saving an older person's life.

A general point about economics: Even what appear to be technical analyses have moral and ethical issues in their assumptions and methods.

Eddy calculates the net costs of testing regimensSound applet

• (The $470 figure is different from what's in the printed article, but I believe the article's number is in error.)

This is an important point. This test does not save money. Giving all women PAP tests does not reduce total health care spending.

The myth that "health maintenance" pays for itself with dollar savings.

Eddy -- average costs per year of life saved

For testing every year, the most expensive (and most recommended) regimen, the average cost per year of life saved doesn't look bad.

Even with discounting future benefits, the cost, at $39,644 per year of life saved, is less than a kidney machine (renal dialysis).  Testing every year looks pretty good!

Eddy -- marginal cost per year of life saved

In the table below, the marginal cost in any one column is the difference in cost between that column's cost and the cost of the column to the left. For the first column, the marginal cost is the difference between every four years and no tests at all.
 

Test every this many years	4	3	2	1
Net cost over lifetime, discounted at 5%	$264	$355	$470	$1,093
Marginal lifetime cost, discounted at 5% (The difference between net costs)	$264	$91	$115	$623
Days of life gained over no testing	93.8	95.4	96.9	98.6
Marginal gain in days of life  (The difference between days of life gained)	93.8	1.6	1.5	1.7
Days of life gained, discounted at 5%	9.54	9.72	9.88	10.07
Marginal gain in days discounted at 5% (The difference between numbers in the row above)	9.54	0.18	0.16	0.19
Marginal cost per year of life saved (365 times marginal cost divided by  marginal gain in days of life saved)	$1,028	$20,774	$28,003	$133,853
Marginal cost per discounted year of life saved (365 times marginal cost divided by  marginal gain in discounted days of life saved)	$10,108	$184,654	$262,523	$1,197,636

If you accept the idea of discounting future years of life saved, annual (as opposed to bi-annual) PAP tests cost $1.2 million per year of life saved.

Eddy -- rapidly diminishing returns

This is because most cervical cancers develop slowly.

Nevertheless, most doctors recommend annual testing.

If you were running an HMO, what frequency of testing would you pay for?

Further readingSound applet

This gives cost-effectiveness numbers -- dollars per year of life saved -- for a number of procedures.  Most are inexpensive per year of life saved, but none of them reduce health care costs. All of them have a net cost, even after subtracting savings from the future health care that's made unnecessary.

For further reading on methodology:

Fries, J.F., Koop, E., et al, "Reducing Health Care Costs by Reducing the Need and Demand for Medical Services," N Engl J Med, July 29, 1993, 329(5), pp. 321-325.
Sound applet

Sound applet
Example of the common argument that prevention saves money.

Foil for Russell's article, which follows.

"Risky Behavior Costs Money," is one subhead. Under it, Fries et al say there was a $1360 difference in annual medical claims cost between high- and low-risk persons in one group of employees.

But how much of that risk difference is amenable to intervention, and at what cost?

Sound applet Fries et al say: "Cure for Terminal Illness Has Become Extraordinarily Expensive and Inhumane."

But, James D. Lubitz and G. F. Riley, "Trends in Medicare Payments in the Last Year of Life," NEJM, April 15, 1993, 328(15), pp. 1092-1096, report that the 29% of Medicare spending is for people in their last year of life, but this showed no upward trend from 1976 to 1988.

So the expense of treatment for terminal illness isn't growing any faster than the rest of expense of prevention and care.

Neuhauser and Eddy show that "cheap" preventive procedures can be quite expensive, if judged on dollars spent per year of life saved.

Sound applet
Fries et al cite studies showing cost savings for health promotion at the workplace.

A possible project for a paper in lieu of an exam in this course would be to look up Fries' and Russell's references and evaluate them, plus look on Medline for more recent studies. Are Fries et al right that these programs save money, or is Russell right when she says that they don't? To what extent are these authors talking about different programs, whose potential for savings differ?

Russell, L.B., "The Role of Prevention in Health Care Reform,"

Sound applet
Russell asks:  Should we claim, or require, that preventive measures pay for themselves?  Her answer is No.

She criticizes justifying prevention by cost savings to the health care system because:

• It's too strict. We spend money (representing social resources) on health care to feel better and live longer, not necessarily to save money later.
• Hypertension control example.
• Eddy's findings on PAP test are another example.
• There's the paradox that total costs may rise simply because prevention helps people live longer.
• Smoking cessation may add to future Social Security costs.

Russell cites studies showing that prevention programs, for risk factors like hypertension and cholesterol, generally don't pay for themselves in health care cost savings.

Russell talks about a national health program, but this question has relevance to prepaid health plans -- so-called "health maintenance organizations," or HMOs. That name was coined to convey the idea that the health plan would have an incentive to promote preventive medicine to maintain the health of the members, thereby avoiding expensive treatment later. But, if Russell is right, what does that imply about how HMOs will behave in practice? Will they really promote prevention?

• For marketing purposes, yes.
• But not to save cost.

S.C. Mental Health Waiver example

* Community Mental Health Center (CMHC) costs are based on what would have been billed to Medicaid at fee-for-service rates. CMHC's were actually paid a capitated rate.

It is possible that the added costs, consequent to case management, reduced medical spending needs years later, after the study was over.  If this were a private HMO, however, it might not realize those savings.  People change HMOs frequently.  That's necessary for a well-functioning market, but it further reduces the incentive for an HMO to prevent illness, if some other HMO is going to get the benefit.

Huntington, J., Connell, F.A., "For Every Dollar Spent -- The Cost-Savings Argument for Prenatal Care,"

But if you examine the studies that support this claim, you see a lot of

1. shoddy methodology, such as failure to control for confounding effects, and
2. wishful thinking, basically assuming that the assertion is true.

Should we claim prenatal care is cost-saving even though it isn't?

Cost-benefit claims helped get Medicaid eligibility expansions through Congress and state legislatures.

Studies vary considerably in how much effect is found for prenatal care.
Self-selection makes the effect of prenatal care appear larger than it actually is.  One study didn't correct for the fact that early delivery itself means fewer prenatal visits.
Most studies assumed that an average or routine level of prenatal care was effective at reducing premie births.  Since the 1990's the trend has been for intensive prenatal care of high-risk women.  This intensive prenatal care does appear to be effective, but it costs more than routine care, so the studies are, in effect, underestimating the cost of prenatal care in their calculations.
Getting more pregnant women to prenatal care earlier requires overcoming such barriers as lack of transportation, the attitudes of the women, the attitudes of providers, and enrollment paperwork.  The cost of giving more prenatal care would be greater than the cost of the care itself.
On the other hand, narrow cost-benefit analysis can overlook benefits such as happier pregnancies, better relationships between moms and health care providers, leading to better parenting, greater likelihood that the child will be immunized, better diet, hygene, or other health-related habits. 

And there's the Russell point -- prenatal care can be a cost-effective way to improve health, even if it doesn't save money.

For your interest, one kind of prevention that does seem to be cost-saving -- reducing total health care system costs -- is contraception. See http://www.agi-usa.org/pubs/journals/2924897.html . Thanks to Cheryl Brannon for finding this one.

Eddy, D.M., "Oregon's Methods: Did Cost-effectiveness Analysis Fail?"

Oregon -- Cost-effectiveness in practice?Sound applet

• Based on this, one can decide how frequently to repeat, to best balance the marginal cost of additional repetitions against the marginal reduction in risk.

Oregon's situationSound applet

The story begins in 1987, when Oregon's legislature decided to try to control Medicaid spending by stopping funding for most organ transplants.

Medicaid spending -- like a box with three dimensions

Sound applet In 1987, Oregon's Medicaid spending box was big, and growing fast. The Oregon legislature had to choose which dimension of the box to limit.

• It didn't want to cut payments to providers -- it was already having trouble getting providers for Medicaid patients, and there was the Boren Amendment, requiring payment rates "… reasonable and adequate to meet the costs which must be incurred ..."
• It didn't want to reduce eligibility. There were already many poor not eligible.
• So it went to work on the third dimension, the services per enrolled person, using the conventional wisdom that expensive procedures are expensive.  In particular, it decided that Medicaid would not pay for most organ transplants.

This sparked public outcry and debate.

Legislators and officials in Oregon needed a defensible way to make decisions about what services to cover.

The legislature developed a proposal to expand Medicaid and health insurance to all Oregonians, while reducing services per person, in order to control total spending.

Under the proposal, cost-effectiveness analysis would be used to objectively determine which services to fund and which not to fund. That way, when controversy erupted, they could argue that funds are limited and that, in return for forgoing some low-benefit-per-dollar services, all Oregonians were getting access to basic, high-value-per-dollar health care services.

Sound applet If you are interested, Fox, D.M., Leichter, H.M., "Rationing Care in Oregon: The New Accountability," Health Affairs, Summer 1991, 10(2), pp. 7-27, gives details on the legislative history.

In 1990, the Oregon passed the legislation, which

• expanded health insurance to cover nearly all Oregonians.
• mandated employer-provided private insurance for job holders
• established a state-subsidized private insurance for high-cost-risk people ("risk pool")

• 130,000 Medicaid eligible before expansion
• 210,000 with expansion.

Oregon -- ranking services by cost-effectivenessSound applet

1. Calculate the cost-effectiveness of all services (procedure-condition combinations).
2. List all the services in order by cost-effectiveness, with the most cost-effective at the top of the list.
3. Calculate expected spending on each procedure.
4. Go down the list, totaling expected spending. Draw a line where the total equals the amount the legislature is willing to fund.
5. Procedures above the line are paid for. Procedures below the line are not.

Oregon -- The cost-effectiveness calculationSound applet

• Eddy, p. 2137, calls this the "Cost-Benefit Ratio," a term I would not use, because the benefit here is in years of life, not in dollars.

Oregon -- quality-adjusted life yearsSound applet

First, though, we need to discuss the "quality-adjusted" life year.

So far, the studies I've shown you have used just two possible outcomes, life and death.

To include procedures that relieve suffering but don't necessarily save lives, Oregon used the Quality-Adjusted Life-Year.

• For a full description, by the person who developed this for Oregon, see Kaplan, R.M., "Value Judgement in the Oregon Medicaid Experiment," Medical Care, October 1994, 32(10)pp. 975-988.

• Dead = 0. Alive and healthy = 1.
• Suffering and disability states are valued between 0 and 1.
• For example, a year of chronic "pain, stiffness, weakness, numbness, or other discomfort" was valued at 0.747.
• Theoretically, there could have been a suffering state so miserable that its value was less than 0, but, in Kaplan's results, that did not happen.
• Kaplan calculated these values based on surveys that asked people what they thought of life in various conditions.

Oregon -- expected quality-adjusted of life-years saved for each procedure-problem combination

1. Estimate the probability of each possible outcome with and without the procedure
2. Multiply each outcome's probability by its quality-adjusted life-year value.
3. Add up all the products for all outcomes with the procedure and, separately, for all outcomes without the procedure. This gives you two expected values:
1. The expected value of life state if the procedure were done. Eddy calls this QWB_Rx. I'll call it EQWB_Rx (E is for expected.)
2. The expected value of life state if the procedure were not done. Eddy calls this QWB_No. I'll call it EQWB_No
4. The difference between the expected values is the expected improvement in quality of life from the procedure. Multiply that by the duration of the effects.

Oregon -- cost-effectiveness calculation continuedSound applet

Oregon -- cost-effectiveness calculation example

• Eddy, p. 2137, top right shows how the EQWB numbers are calculated by multiplying each outcome's value times its probability.

Oregon -- cost-effectiveness calculation simplifications

• As an aside, notice that, in contrast with what Eddy did in his other article, Oregon did not discount future costs or benefits, nor did Oregon calculate future health care costs saved or incurred as a consequence of the treatment today.

Oregon's list was done, but then ...Sound applet

Some items seemed out of place. Some rankings seemed counter to common sense.

Eddy says, in Oregon's defense:

1. That's the whole idea of using cost-effectiveness analysis. If you already know what's important, you don't need cost-effectiveness analysis to tell you.
2. Some critics didn't recognize the cost side of argument.
• For example, dental caps for pulp exposure ranked nearly as high as surgery for ectopic pregnancy, but the tradeoff is one surgery for ectopic pregnancy against capping 105 broken teeth.

• Certain categories of procedures were given higher or lower priority.
• Within categories, the cost-effectiveness numbers were used for ranking but with hand-adjustment.

Oregon -- Were the list's problems technical or conceptual?

• categories too broad
• errors in evaluating outcomes and durations

• Does cost-effectiveness violate the Rule of Rescue?
• Eddy (my interpretation) says: Life-saving services have external value
• External means, for example, that it makes me happy to see you get medical services you need.
• Eddy calls this "vicarious utility."

That should take care of Rule of Rescue concerns.

But, the questions that Eddy would ask the public seem even harder to answer than the questions that Kaplan asked.

Did Eddy's PAP test article do this?

Oregon went ahead and implemented the priority list, after more changes.Sound applet

The methodology was further revised to remove the quality-of-life adjustment. Kaplan is angry about this, because it discounts treatments that improve a disabled person's condition but don't save his life or fully cure the disability.

More hand-adjustments were made, so that the influence of the cost-effectiveness ranking further receded.

Some categories of care, such as for extremely immature babies, were exempted from the list, so that funding for them was guaranteed.

In 1993, the Clinton Administration granted the waiver.

Cost-effectiveness studiesSound applet

• Idea: Face squarely the Economic Problem
• Decide how best to spend society's limited sources
• Actually doing it raises methodological questions
• Cost-effectiveness analysis seems to have been judged OK for evaluating repetitions of tests.
• For decisions among procedures, as tried in Oregon, the cost-effectiveness results were heavily modified before put into practice, and no other state has copied Oregon's method.

Take-away points

These ideas came up in class:

1. Cost-benefit and cost-effectiveness studies are not definitive. There's no one right way to do them. All cost-benefit studies involve choices about what costs and benefits to include and about about how to value the costs and benefits that are included. Cost-effectiveness studies all involve choices about how to specify the goal, as well as about what costs to include and how to value them.
2. Cost-saving and cost-effectiveness are not the same thing. Nor are cost-saving and cost-benefit the same thing.  The mental health patients in S.C. who had standard case management did cost less than the patients with enhanced case management. That does not mean that neglect is cost-effective.  Nor does it mean that the costs exceed the benefits.  It only means that neglect is cost-saving to the health care system. The same goes for giving poor people primary care through the hospital emergency department. If you get more effect for more cost, then what you gain on the effect side can be worth the extra cost, making spending more cost-effectivene. Cost-effectiveness analysis is supposed to be about comparing alternative ways of meeting the same goal. If one alternative meets more goals, or meets the goals better, you can't ignore that and just compare the costs and call that a cost-effectiveness analysis.

I can imagine three possibilities for a preventive measure:

1. It pays for itself. The future health care cost savings are more than the cost of the intervention today. Axnick argues that the measles vaccine does this. I call this "cost-saving." Cost-saving preventive actions have no opportunity cost. You don't give up anything, in the long run, when you do them.
2. It does not pay for itself. The cost of the intervention is more than the expected future health care cost savings. The intervention has an opportunity cost.
   1. The intervention is less costly per person than other things we do to save a life or improve someone's health. This makes the intervention "cost-effective." Cost-effective interventions are "economically feasible." PAP tests are an example of this.
   2. The intervention is more costly than other things we do to save a life. This makes the intervention not cost-effective. This is what you might mean by "not economically feasible." An example of this is that we, in the U.S., generally don't stop vehicular traffic at intersections in all directions at the same time to allow pedestrians to cross safely. Instead, we allow cars to turn and drive across cross walks while pedestrians are crossing the street. Presumably, we do this because stopping cars in all directions is not cost-effective at saving lives or preventing accidents.