Cost Effectiveness Analysis- Chapter 5

• compare resources with consequences

• compare 2 or more (similar) alternatives
• Only similar tx are compared
• Identifies and compares costs and outcomes of competing
interventions
- outcomes must be “similar”
- Measured with the same denominator "Effectiveness"
• Costs: measured in $$$ Drug A Drug A
<less effective than< >more effective than>
• Consequences/outcomes: measured in natural units Drug B Drug B

e.g. symptom free days, cures, lives saved, years of life saved Drug A
Dominated It depends…
• Benefits are units not money (money is Cost-Benefit) >more expensive than>
Drug B (select B) (trade-off)
"Cost"
• Results are expressed as a ratio: Drug A
It depends… Dominant
- Costs / Outcome Unit <less expensive than<
Drug B (trade-off) (select A)
Ex: $$$ / cure or $ $$ / life saved
• Outcomes must always have some measurable denominator- Ex: mm Hg
• Often the term “cost-effectiveness” is misused
- Cost Utility is a “subset” of CEA
“Efficacy” vs “Effectiveness”
“Can it work?” vs “Will it work?”

These steps should not surprise you:

1. Define the problem or objective being tested
- select objectives/purpose (more specific the better)
- define perspective!

2. Identify Alternative Interventions

- consider relevant treatment options
- choose appropriate comparators

3. Describe Inputs and Outputs

• how resources affect the outcomes
• decision analysis models

4. Identify costs and outcomes

- can be final (cure / survival / death)
- or intermediate (mm Hg., FEV1, days)
• Have proof they’re related to one another

5. Interpret and present results

- C/E ratios; graphical representation of data

• This technique maximizes “bang for the $$”

- recognizes limited resources
- counter-intuitive to traditional medicine! When MDs bear no financial risk
- operating on a fixed budget often changes the preferred outcome
- This is the single most controversial issue in pharmacoeconomics
Cost of treatment:
Total money available to spend X 1pt Tx/$ tx X #cured/100 = total cures w/ $ available

Simple or Average: (cost per cure)

Amount of money to spend
Number pts Cured

Incremental/Marginal C/E ratio:

Total Cost option A - Total Cost option B
Total Outcomes A - Total Outcomes B

• Incremental costs measure variable costs…

• Tells us whether I treat with A or B there are fixed costs that I don’t care about and this ratio makes them
go away

• If I am going to use A for pts.. I am spending an xtra $$$ (positive) for every additional pt cured
compared to if I was going to use B
• Positive is that I am spending extra money. (bad)
When its days in the hospital more is bad so you want positive.
• Negative you are saving money (good)
• Use these to determine is it worth it??
• Generate a number that tells us if we use a more expensive drug how much more will it
cost us.

• Incremental C/E ratio are more useful for decision makers

• Using C/E ratios:
- decide in advance what’s “good enough”
- usefulness depends on data quality
- some assumptions are always needed; Demonstrate with sensitivity analyses
• Displaying CE results: League tables and Cost-effectiveness “plane”
 Cost Utility Analysis- Chapter 6

- can only measure things with a FINAL outcome

QALY
• integrates costs and outcomes within “utility analysis” framework
- “quality” x “quantity” (i.e. years)

• “Utilities” = Patient preferences

• Ratio = Costs
QALY (Quality Adjusted Life Year)- most common

Other demoninators: 1. QALY (Quality adjusted life year) *

2. HYE (Health Year Equivalents)
3. TWiST (“time without symptoms”)
4. “well-years gained”
5. “functional years”
6. Quality adjusted life expectancy

When should you use this method?

1. When “quality” of life is important- As the primary outcome
2. Treatments affect mortality and morbidity
3. When diverse treatments can achieve a common basis (i.e. outcome)
4. When studying projects where CUA used previously
**Basically when you are worried about quality of life**

NOT: 1. When only intermediate outcomes available

2. When the alternatives are equally effective
3. When “quality” isn’t measured separately
4. When its just too darn’ expensive...

Calculation QALY:
QALY =
# of years [time] x [proportion “Q”]
“Q” is the “quality” adjustment or utility

Ex: “Perfect” health: Q = 1.0 Death: Q = 0.0

Ex: 20 years of perfect health QALY = 20 yrs x 1.0 (healthy) = 20 QALYs
Ex: 20 years of a chronic illness QALY = 20 yrs x 0.5 (est. “Q”) = 10 QALYs

How to Calculate Q ?:
3 methods currently employed to get a value:
1. From the literature...
2. Health professionals judgements
3. Direct measurement from patients (surveys from pts)
a. Category (case) rating scales- mark X on the scale to rep health (~vis analog scale)
b. Standard Gamble *- how much are you willing to risk for perfect health? 2 choices live without
perfect health or gamble new tx with risk of death
c. Time trade off *- if have the tx you will die sooner but u would be in perfect health while
alive… so basically, how many years are you willing to give up to live in perfect health?
d. Magnitude estimation- given a reference case then compare it to related situations 2x or ½
desirable
e. Person trade off- measure different attributes of life.
Limitations and weakness:
• Utility values vary a lot
• Different methods = different QALYs
• Estimating “utilities” may not be linear
• may under-estimate the value of treatments
• Is 0.2 -> 0.3 the same as 0.7 -> 0.8 ?
• There are other measures besides QALYs but less popular and more complex

When reviewing a CUA study:

- Describe the “utilities” in layman’s terms
- Report the source of the weights used
- Explain the results in layman’s terms
- “Transparency”
- Sensitivity Analysis
 Decision Analysis- Chapter 8
• A quantitative approach to decision making
• Uses a diagram for choices and outcomes
• Quantifies uncertain events
• Imposes logical thinking
• Method used “inside” CEA, CUA, etc.
• Emphasis on ‘expected values’

Expected Values:
• In probability theory the expected value of a random variable is the sum of the probability of each
possible outcome of the experiment multiplied by its payoff ("value").
• It represents the average amount one "expects" as the outcome of the random trial when identical odds
are repeated many times.
• The expected value itself may be unlikely or even impossible (ex. Dice)
• These strategies are used in gambling, attempts at medical therapies, or other problem solving situations.

Box: A “choice” node: what follows is result of a decision, no probability required.

Circle: A “chance” node: what follows is uncertain & requires probabilities

Steps:
1: Identify a decision which needs to be made (eg. Medical decision)
2: Diagram the decision & all plausible results
- diagram consequences over time Decision Analysis (“Cosler method”
method”)
- include probabilities for each result 1 ?? % $ ???,??? $ ???
- calculate “expected values” for each
decision
- identify the preferred alternative
• Row for every branch and 3 columns 2 ?? % $ ???,??? $ ???

• % of pt that make it to that point

• Total cost of 1 pt that’s makes it to that point
• Cross multiply column 1 by column 2
• Add column 3 (separate options—decision 3 ?? % $ ???,??? $ ???

box)
- All probabilities after a decision box must
=100%!!

1. “Kickin’ it up a notch !”
• Expected values of non-$$$ outcomes
• Calculating incremental ratios
2. Applied examples: Monte Carlo simulations, Example output

- Could calculate Incremental-Cost Effectiveness Ratio (ICER) (Total $$$ A - Total $$$ B)
** usual grid wont apply if its length of stay bc less is better (Total Outcomes A - Total Outcomes B)
**cheaper and less days in hospital = no brainer but not according to the previous grid!

1. What is the expected cost per patient and expected 2. What’s the expected cost per healthy patient
LOS for pts with Ab Tx and without Ab Tx? with Ab Tx and without Ab Tx?
Expected Costs
Expected Costs Expected Length of Stay
Results (per Pt) (per PT) Results (per Pt) Healthy Patients

Ab Prophylaxis $ 7,940 3.32 Ab Prophylaxis $ 7,940 0.92

No Ab $ 9,000 4 No Ab $ 9,000 0.75

 2.Answer is negative so ur saving $ for every healthy pt. (use 0 and 1 for totals in chart)
Sensitivity Analysis: 76-77;105-107;162; 371-372;384
What is it?
- Process of varying assumptions and variables over “plausible” ranges to test the “robustness” of results
and conclusions
- Example: When you change the cure rate of an antibiotic, does it change your conclusions?

Why do it? Sources of uncertainty:

1. For effect of assumptions on conclusions 1. Variability in the data
2. Identifies your critical assumptions 2. Generalizability to real world practice
3. Trying to predict the future effects 3. Extrapolation of results
4. Now a key indicator of a quality study 4. Choice of Analytical method(s)

Conducting it:
Assume you have the following :
- Rx adverse event profile range: (1% to 5%)
- “realistic” discount rates ( 3% to 5%)
- Cost of one alternative (e.g. surgery) varies from $ 100,000 to $150,000
• How many different combinations do you have?

Types:
1. “Simple”- modify key assumptions across a reasonable range; if conclusion changes- the model is “sensitive”
2. Threshold- sort of a “breakeven point”
3. Analysis of extremes- “best case” & “worst case” scenarios
4. Probabilistic (Monte Carlo)- where values in ranges selected at random

Limitations:
• Are the ranges tested realistic?
• If results are “sensitive” - so what?-- You can make any variable “sensitive”
• Simple vs multi-way sensitivity
• SA is not being used enough in literature: not at all – or - not sufficiently

Where do we get realistic estimates?

• Clinical trials- Individual RCTs or meta-analyses
• Health care claims data- Hospital(s) / Managed Care Plans; Medicare / Medicaid
• Nationally public data sets- NCHS
• Epidemiology is fundamental

Health Related Quality of Life Psychometrics: (The Science of Surveys): chapter 7

• Health Status includes: functional status, morbidity (disability), well-being (mental/social/role

functioning)

• Quality of life is an opinion or assessment: based on Pt. perceptions and judgements, influenced with
their level of satisfaction

Patient outcomes are not clinical outcomes

“General” health survey(s)

- Creation of population “normal” values- comparisons of sub-populations
- These measure several health categories
• functional status (physical / mental)
 • social / role functioning
• overall assessment of health

Global or Generic Survey:

SF 36, SF12; Sickness Impact Profile (SIP), Nottingham health Profile (NHP), General Health
Questionnaire (GHQ28), Psychological General Well-Being Index (PGWBI), EQ-5D
(EuroQol)
SF-36 measures health in 8 dimensions:
Physical: Physical Functioning (RF), Role-Physical (RP), Bodily Pain (BP), General Health (GH)
Mental: Vitality (VT), Social Functioning (SF), Role-Emotional (RE), Mental Health (MH)

• Disadvantages or limitations
- lengthy & time consuming
- lower response rates
- complex scoring (need a computer or $$$)
- may not address concerns for specific Dxs.

Disease Specific:
• Early Dx specific forms weren’t surveys - measured functional limitations
Ex: - Karnofsky Performance Status Scale
• American Rheumatism Association Functional Classification
• NY Heart Association classification
- Originally designed by clinicians
• Advantages: • Disadvantages:
- more sensitive to disease specific effects - not comparable to larger groups
- fewer questions - may miss “unexpected” effects
- more relevant & more patient focused

Check Literature: 2 Areas:

1. Development and testing of survey itself- sociology / psych / educational literature
2. Applied research using the selected survey- health / medical sources
Components of a well-written report describing the use of a HRQOL instrument
1. Statement of the research questions and hypotheses
2. Description of the measuring instruments included in the study
3. Description of the subjects, sampling procedures, sampling frame
4. Description of the conditions under which the data were collected
5. Discussion of the statistical analysis
6. Presentation and discussion of the results, statement of conclusions, and
description of the limitations

Psychometrics:
Several Important properties of surveys:
1. Reliability 2. Validity 3. Responsiveness or “Precision” 4. Sensitivity / Specificity

1. Reliability: is this a ‘true’ score ?

- Statistics used to measure reliability:
- “Chronbach’s alpha”
- prefer values > 0.8 or > 0.9

2. Validity: Are you measuring what you think you’re measuring?

A. Content validity- does survey contain appropriate questions?
B. Criterion validity- do scores survey compare to other “gold standard” instrument?
 C. Construct validity- checks the theory behind the test; Relates to other measures in plausible ways
D. Factorial validity- do items within a dimension relate? (factor or cluster analysis)

Reliability: must be established first

- a valid instrument will be reliable, a reliable instrument may / may not be valid

3. Responsiveness/precision: if “the target” changes, will you be able to detect the changes?
4. Sensitivity: The ability to detect a change Specificity: The ability to detect “no
when it exists. change” when there really is no change
Reality (objective measurement) Reality (objective measurement)

"Present" "Absent" "Present" "Absent"

"Present" A B "Present" A B

Survey says: Survey says:

"Absent" C D
"Absent" C D

A/ (A+ C) =D/(B+D)