Pharmacoeconomics - Making Adjustments To Costs And Outcomes

Front 1

The quality of your economic evaluation is dependent on the quality of the cost and outcome data used in the study.

Back 1

Dr. Pizzi stressed knowing the table titled "What adjustments are made and to which components of the ratio?"

This lesson covers the "tinkering" they do with the ratio after outcome data is obtained to minimize uncertainty.

Front 2

Sensitivity analysis

Back 2

Modifying the costs in your numerator AS WELL AS the outcomes in the denominator to account for variation or uncertainty in the real world.

Front 3

Discounting

Back 3

is applied to account for the time preference

Discounting is applied to costs if the time horizon of your analysis is greater than 1 year.
Discounting is sometimes applied to outcomes if the time horizon is greater than 1 year and outcome is a utility measure

Front 4

Inflation

Back 4

is used if cost data are more than 1 year old and is applied to costs and not outcomes
Old cost data is inflated to present day value

Front 6

What is risk adjusting?

Back 6

Another way to tinker to make sure our calculation is as good as we can get it
Trying to fix/correct observational data that we used in our denominator for the outcome

Front 7

Continuation

Back 7

Example: Looking at improvement in viral load and you got your data from observational studies of people with HIV. Patients were not randomized and so these people may have certain characteristics that could make them more or less prone to your outcome measure.
You do risk adjustment to level the playing field when you can't get randomized data. You only have observational data

Front 8

Sensitivity Analysis - Vary inputs of your input ratio over an acceptable range of values

Back 8

Doing a PE analysis and you only have medicare reimbursements so you know what medicare pays
You also want to understand a private payer population
You know that private payers pay 15-20% more so in your sensitivity analysis you would change your class inputs to be 20% more and see what happens to the ratio.
If your result changes, your ratio is sensitive to that variable.

Front 9

Three major sensitivity analysis

Back 9

One way - You have each of your inputs in the numerator and denominator and your going to vary them one by one to see how it impacts your ratio
Two way - changing two variables at the same time to see how it impacts your ratio

Front 10

Continuation

Back 10

Probabilistic - Simulating random iterations of your ratio - using a computer to simulate real world variation
This is called a Monte Carlo simulation and this is preferred

Front 11

One way sensitivity analysis is very popular.
Major limitation?

Back 11

Its so simple that it can be misleading
Works well when there is one input to your ratio that has a big influence
Does not work well when there are several variables that have influence.

Front 12

What is a tornado diagram?

Back 12

It shows several different one way sensitivity analyses.

Front 13

When looking at a tornado diagram, how do you know which input variable is most sensitive?

Back 13

There are horizontal bars lined on top of each other. The longer of two bars will be more sensitive, as there is a greater variability in expected value for that one input variable.

A skinny bar means there your ratio does not change nor is not impacted by variation.

Front 14

Two Way Sensitivity Analysis

Back 14

Modifying two of your inputs to the ratio at the same time.
Instead of looking at disease incidence, now you may be looking at incidence and efficacy

Front 15

continuation

Back 15

In the two way diagram, the white area shows when it would be good to vaccine and the black area shows when it would be good to just give support.
The line shows the dividing line between. If your on the line, either would work.

Front 16

Two way sensitivity analysis is not as popular as probabilistic or one way analysis. Why?

Back 16

By showing a two way graph, you could be misleading the reader to this they are the only two variables that matter. Besides incidence and efficacy of the vaccine, cost could be very important. You have to be careful about which two variables are chosen.

Front 17

Probabilistic Sensitivity Analysis - Monte Carlo

Back 17

Your taking all of the inputs to the ratio where you believe there is gonna be variation and simulating what happens by having random combinations of those variables within the ranges you set on the computer. These results are graphically shown on a cost effectiveness plane with 4 quadrants.

Front 18

Continuation

Back 18

The origin is the gold standard you are comparing your treatment to. Each dot on the plane represents a simulated patient.
Quadrant 1 treatment is more effective but also more costly
Quadrant 2 treatment is less effective but costs more
Quadrant 3 treatment is less effective and costs less
Quadrant 4 treatment is more effective and costs less

Front 19

Discounting

Back 19

Adjusting for the time preference for money and health where current value is not the same as the future value. Health and money are worth more in the future.
Future dollars are discounted when your analysis goes beyond one year. If your ICER looks at total cost and outcomes achieved in a diabetes medication compared to the gold standard. If your time horizon is greater than one year you need to discount your cost.

Front 20

Continuation

Back 20

If the analysis is one year or less you do not need to discount.
Discount benefits? You can. C/U ratio does not change because both cost and benefits are being discounted. If you discounted costs and not benefits, your C/U would decrease.

Front 21

Continuation

Back 21

Typically discounted at a rate of 3-5%. You can't discount things that are in natural clinical units. There is a preference for having health now versus later.

Front 22

Inflation

Back 22

Applies to PE analysis because we are putting costs in our numerator.
Adjusting older cost data up to its present value.
You inflate the cost, but not the outcome. Outcomes are a health or utility measure. Therefore the numerator is inflated

Front 23

Continuation

Back 23

How is this done?
Inflation rates for good and services -->
Consumer Price Index - obtain the rate of healthcare inflation for the years of your analysis.
For labor costs -->
Bureau of Labor Statistics

Front 24

Continuation

Back 24

If you are looking at data from 2008, you inflate it up to 2011 using the consumer price index.
If your numerator involves either reimbursements or microcosts (itemized costs such as supplies, doctor time, nurse time etc.)

Front 25

Continuation

Back 25

If you are microcosting and your numerator involves labor costs such as a nurses salary, you can obtain labor costs from the Bureau of Labor Statistics. You only need to inflate labor costs if you are doing a microcost approach or if the data is from previous years. Most analyses involve reimbursements and can be obtained from the consumer price index. Microcosting is not very popular bc it is difficult to get all the itemized costs using a time and motion study.

Front 26

Continuation

Back 26

Inflate or Discount?
When your doing a PE analysis you need to discount because future health benefits are worth more today and they are tomorrow. You are discounting the cost today and discounting it tomorrow for each year beyond 1 year.

Front 27

Inflation-adjusted discount rate

Back 27

With historical costs you inflate them up to present day value.
Discounting is applied from the current day to the future.Inflation-adjusted discount rate --> Future costs need to be both inflated and discounted!
If you are using a 5% discount rate, and the inflation rate for healthcare goods and services is 2%, then your inflation-adjusted discount rate is 5-2 = 3%.

Front 28

Continuation

Back 28

This means that you need to increase your costs by a net of 3%.
If your analysis is less than 1 year, you do not need to discount. You do not need to inflate!

Front 29

Continuation

Back 29

If you are looking at data from 2008, you inflate it up to 2011 using the consumer price index.
If your numerator involves either reimbursements or microcosts (itemized costs such as supplies, doctor time, nurse time etc.)

Front 30

Continuation

Back 30

If you are microcosting and your numerator involves labor costs such as a nurses salary, you can obtain labor costs from the Bureau of Labor Statistics. You only need to inflate labor costs if you are doing a microcost approach or if the data is from previous years. Most analyses involve reimbursements and can be obtained from the consumer price index. Microcosting is not very popular bc it is difficult to get all the itemized costs using a time and motion study.

Front 31

Continuation

Back 31

Inflate or Discount?
When your doing a PE analysis you need to discount because future health benefits are worth more today and they are tomorrow. You are discounting the cost today and discounting it tomorrow for each year beyond 1 year.

Front 32

Inflation-adjusted discount rate

Back 32

With historical costs you inflate them up to present day value.
Discounting is applied from the current day to the future.Inflation-adjusted discount rate --> Future costs need to be both inflated and discounted!
If you are using a 5% discount rate, and the inflation rate for healthcare goods and services is 2%, then your inflation-adjusted discount rate is 5-2 = 3%.

Front 33

Continuation

Back 33

This means that you need to increase your costs by a net of 3%.
If your analysis is less than 1 year, you do not need to discount. You do not need to inflate!

Front 34

You are completing a cost effectiveness analysis using cost data from 2007. The analysis projects cost effectiveness from 2011-2014. You need to...

Back 34

Inflate historical costs to bring them to 2011 value
Inflate future costs for 2012-2014
Discount future costs for 2012-2014

You don't count the first year for both inflation and discounting. They start at 2012.

Front 35

If benefits were also presented as quality adjusted life years, would they also need to be discounted?

Back 35

Yes

If you didn't discount benefits, your Cost/Utitliy ratio would go down over time. You should not see this. It is recommended to discount both costs and benefits.

Front 36

Now we are going to shift gears away from discounting and inflation and step into the idea of fixing or improving the clinical outcomes we are using in our denominator. We are trying to build a ratio of cost per outcome and we have discussed way to adjust the ratio.

Back 36

What could be clinically wrong with the data we are putting in the denominator. Does the outcome measure effectively represent the population? What do you do with clinical variables where you think the source may be biased or problematic. How do you tweak the denominator.

Front 37

Risk Adjustment: methods that account for differences in characteristics that might lead to variation in outcomes. This is often applied before making inferences about treatment outcomes.

Back 37

If your denominator uses observational data from an observational study, you should check to see if it is risk adjusted. You are tweaking the denominator to make sure the data you got is not biased bc it came from an observational vs a randomized study.

Front 38

Why do we care if our data is from an observational study?

Back 38

When you randomize, each subject has an equal chance of getting a treatment. You are washing out the bias that can occur when selecting people to receive a treatment.
Observational data has no randomization. This can lead to bias. The people in the study may have certain characteristics that could bias the data. This tries to fix the observational data to make it better.

Front 39

Graphical depiction

Back 39

Two treatments, a and b
Individuals who get treatment a have a certain probability of getting treatment a based on their symptoms, age, previous treatments and the same for treatment b.
Each a and b will have an outcome. That outcome could be due to the treatment. But the outcome could also be due to certain confounders (characteristics the person has) or they may fail because they are too old or too sick (confounders to your outcome)

Front 40

Continuation

Back 40

In a randomized trial, there is an equal chance of getting treatment a or b is equal. The effect of confounders are washed out.
Observational studies patients are selected to get a or b. The probability of a is not the same as b.