1. CIHI’s Population Grouping Methodology
Health System Use Summit
February 11, 2016; Track #2
Douglas Yeo, CIHI

2. Outline
Case Mix
An example

3. Better data. Better decisions. Healthier Canadians.
RAI assessment, completed on the client at regular intervals, is the foundation data for the main RAI outputs:
CAPs
Outcome Scales
Quality Indicators
Case Mix
These outputs can then be used across multiple levels for various purposes:
Care planning and monitoring of individuals
Quality improvement and planning at the organization level
Performance reporting and comparability at the system level
Two key messages here:
This work would not be possible without a standard like the RAI. Collecting the info in a standardized fashion allows the development and continuous improvement of outputs, and allows comparisons across all levels and between jurisdictions. Apples to apples.
Second: Strong, reliable data is needed at the point of collection. If it’s not good there, it’s not good all the way up. One of the best ways to ensure this is also one of the more challenging – the RAI assessment needs to be built into the culture of resident care.

4. The Case for Case Mix in the Health System
For over three decades, CIHI has been providing to health facilities:
Standards
Specifications for creating data
Decision support tools
Now we’re building a methodology to help jurisdictions understand their entire populations
Case Mix has been part of CIHI’s core business from the beginning.
Case Mix work at CIHI is world renowned
Primary use is for facility and system level planning
Secondary use in funding health care system

5. CIHI Population Grouping Methodology
Provides a clinical profile of each individual in the population
Each person eligible for publicly funded health care
Based on historical person-level clinical information from across the continuum of care
Over extensive time period (e.g. two to ten years)
Indicators
Cost weights - current and future burden of morbidity

6. Clinical Classification: Assigning Health Conditions to Persons (illustration)
A07 Paralytic Syndrome / Spinal Cord Injury
D44 Acute & Other Respiratory Diseases
F81 Signs, Symptoms Digestive & Hepatobiliary System
CIHI source data
Diagnoses
Hospital inpatient
Day surgery
Physician claims
Continuing care
ICD-10-CA
ICD-9
RAI-MDS 2.0
So what is this clinical classification? It’s a set of 225 Health Conditions. For each person, a flag is set to no or yes (0 or 1) for each of the 225 conditions. Each Health Condition flag indicate whether or not the person has the condition. Assigning these health condition flags can also be thought of as disease tagging. The 225 Health Condition flags paint a picture of the health of the person (i.e. their morbidity).
Note that although we use the terminology Health “Conditions” some of the 225 are in fact not conditions, but events (e.g. stroke) or states (e.g. palliative state). But most are conditions.
The methodology looks at information from multiple sectors of care to assign these 225 flag values for a person.
For the alpha release, the methodology looks at the inpatient, day surgery, and physician care. The ICD codes in that data are inform the setting of the flags (i.e. appropriate ICD codes are mapped to the health conditions; many ICD codes to one health condition category). Note that not all ICD codes are mapped to a health condition (e.g. No health condition category for normal newborn so those ICD codes are not utilized).
For conditions that one would expect that a person would see a doctor at least twice over a 2 year period (i.e. Chronic conditions that do not resolve), the methodology requires that the condition be found in the data at least twice in the 2 year period. The quality of the hospital data is considered to be much better than that of the physician claims data and so this requirement is not applied to hospital data.
The methodology can also look at long term care data and assign the values of the many of the Health Condition categories based on RAI assessment data.
This set of health condition flags describes the clinical profile of the person and more importantly describes clinical profile of the population.
As you can see here on this slide, this fictitious Joe (don’t worry we don’t really have access to or use real names) has 2 health conditions, identified by looking at hospital and physician data over a 2 year period.
Joe

7. Assigning Cost Weights to Persons (Illustration)
A07 Paralytic Syndrome / Spinal Cord Injury
D44 Acute & Other Respiratory Diseases
F81 Signs, Symptoms Digestive & Hepatobiliary System
Cost Weights
(hospital + physician costs)
Cost Indicator Effect
Concurrent
Prospective
A07
4.95
1.07
D44
1.74
0.76
F81
0.11
0.24
A07 x F81
-0.69
0.00
Total
6.12
2.07
As mentioned, the methodology also has predictive indicators. The predictive indicators for the alpha version consist of health resource (i.e. cost) indicators, or cost weights:
Healthcare cost for concurrent period
Healthcare cost for prospective period
These weights are relative weights and indicate the relative cost of one person compared to another person, and reflect the variation in treatment costs due to morbidity and treatment required. For a person with a concurrent cost weight of 1.24, their morbidity burden is estimated to be 24% higher than that of the average person.
The cost indicators are based on 2 years of cost data from Alberta (the development data). The average cost in that date for the concurrent period is $3,258; the average cost for the prospective period is $1,752
The costs include costs from hospital inpatient, day surgery, and emergency department episodes, and publically funded physician care
The value of the cost weight assigned to a person is based on the 225 health conditions that the person has. Each condition present adds to the cost weight (i.e. the expected cost) of the person.
As you can see here on this slide, Joe with his 2 health conditions, has a concurrent cost weight of 6.12
Other indicators will be added over time (e.g. Some possibilities are cost by sector, probability of admission to hospital, expected number of ED visits, mortality)
Joe’s concurrent cost weight is 6.12
Joe’s prospective cost weight is 2.07
Joe

8. Some Population Grouping Methodology Applications
Profiling and planning
Coordinated care and targeted care management
High users
Risk adjustment
Health system indicators
Funding
Physician
Regions
Although we apply the population grouping methodology to each person and it does provide a clinical profile and expected indicators for each person, the value of the methodology is in using it the population, or sub-population level
The clinical profile and predictive indicators of the population can be used to segment the population and look at variations across population segments with regard to the morbidity of population and the morbidity burden. This has various health policy applications such as
Population profiling and health system planning
Risk adjustment of indicators
Funding
Profiling and planning
Coordinated care and targeted care management
E.g. High Users
Identify high users health conditions can identify clinically complex sub-populations (e.g. 3+ health conditions)
Identify high users - Cost weights can identify sub-populations with high resource usage
Profile high users – identify high users of acute care using CMG+ and profile them using 214 health conditions (e.g. number of health conditions) or overall cost to the system (i.e. population grouping methodology cost weights)
Risk Adjustment
Adjust for population morbidity when comparing health burden across populations
The 214 health conditions provide a basis for risk adjustment
Examples:
Efficiency – use the existing cost weights from the population grouping methodology
Mortality – can develop a model specific to mortality
Useful to academics, administrators, and CIHI
Physician capitation
Equitable setting of capitation rates
Basis for fair compensation of physicians for enrolling patients with high needs
Regional funding
Account for differences in health status of the regional populations in population-based funding

9. Distribution of Health Conditions, by Age Group
This graph provides an illustration of how the population grouping methodology is useful in profiling.
Each vertical bar represents an age group. Each vertical bar shows distribution of # of health conditions within that group
Some observations
As population ages, the number of health conditions increases.
For 65+, over half of the people have 5+ conditions, and very few have 0 conditions
High percentage of non-users in older population is likely problems with frame; lag in reporting of deaths
This graph is based on 2 years of Ontario, Alberta and BC data used in the development of the beta version
* and Ontario, Alberta, and B.C. data used in methodology development

10. Profiling of Population (Concurrent Cost)
Decile
Volume
Average Actual Cost
Average Modeled Cost
Proportion of Costs
Avg. # of Health Conditions
Average Age (in Years) 1
2.3M 56
0.2%
0.2 39 2
200
0.7%
0.8 27 3
317
145
1.1%
1.0 34 4
488
260
1.7%
1.8 5
725
474
2.5%
2.6 6
1,046
830
3.7%
3.2 37 7
1,507
1,359
5.3%
3.9 41 8
2,356
2,380
8.3%
4.7 47 9
4,252
4,608
14.9%
5.6 49 10
17,612
18,470
61.7%
8.0
All
23M
2,856
2,860
100% 40
This table provides another illustration of how the population grouping methodology is useful in profiling, this time using the concurrent cost weight
We sorted the persons according to their concurrent cost weight and the split them into 10 groups of equal size. Each of the 10 groups is represented by rows 1 to 10 in the table. The first row is the persons estimated to be the least expensive (most of these are non-users in the concurrent period and have a cost weight of 0). Row 10 is the persons estimated to be the most expensive. The last row is the overall population.
So the overall population is ~ 23M. Each of rows 1 to 10 contain 10% or ~2.3M people.
Some observations
Top 10% consume 62% of costs
The median number of health conditions is just over 3, and that the number rises rapidly in the higher deciles
High users can be clearly identified
In top 10%, the average actual cost is over 6 times the overall average
* and Ontario, Alberta, and B.C. data used in methodology development

11. Estimated Cost for a Person (Illustration)
A07 Paralytic Syndrome / Spinal Cord Injury
D44 Acute & Other Respiratory Diseases
F81 Signs, Symptoms Digestive & Hepatobiliary System
Estimated Cost for a Person (Illustration)
Joe’s concurrent cost weight is 6.12
Joe’s prospective cost weight is 2.17
Population average concurrent cost is $2,861
Population average prospective cost is $1,483
Where did we get those modelled costs on the previous slide?
For a person, his or her cost weight, along with a population (could be region or province) average cost, provide an estimated cost for that person. Strictly speaking, the correct average cost to use is the risk adjusted average cost (RAAC) discussed on slide 11
Note that both concurrent and prospective costs are estimated—in this case, Joe’s set of conditions result in a high predicted cost in the future.
Joe’s expected concurrent cost:
$17,509 = $2,861 x 6.12
Joe’s expected prospective cost:
$3,218= $1,483 x 2.17
Joe
Mock data

12. Number of Cases (i.e. people)
Case Mix Index (CMI)
Weighted cases = sum of cost weights for a sub-population or group of people
In population grouping, the person is the case
CMI= Weighted cases Number of cases = Sum of cost weights Number of cases
Example:
Region
Number of Cases (i.e. people)
Weighted Cases
Case Mix Index (CMI) A
1.45M
1.3M
0.897 B
0.3M
0.344M
1.147 C
0.45M
0.556M
1.236
Total
2.2M
1.000
The population grouping methodology cost weights can also be used in the calculation of a summary measure of the health complexity of a population
This measure, called a Case Mix Index, is simply the average cost weight for that set of persons (i.e. the sum of the cost weights divided by the number of cases.)
In the example here the CMI has been calculated for three regions, based on all persons who are resident in those regions. One can see that clinical complexity varies substantially across the regions.
Similar analyses can be done for physicians rather than regions (or other sub-populations)
This CMI concept applies to our other case mix methodologies as well, not just the population grouping methodology
Mock data

13. Risk Adjusted Average Cost (RAAC)
RAAC= Average Cost (5) CMI (6)
Example:
(1) Region
(2)
Cases
(i.e. people)
(3)
Total Cost
(4)
Weighted Cases
(5)
Average Cost
(6)
CMI
(7)
RAAC A
1.45M
$2,455M
1.3M
$1,693
0.897
$1,888 B
0.3M
$556M
0.344M
$1,853
1.147
$1,616 C
0.45M
$889M
0.556M
$1,976
1.236
$1,599
Total
2.2M
$3.9B
$1,773
1.000
I mentioned risk adjustment earlier and the CMI can be used in risk adjustment.
As an example we have the 3 regions from the previous slide. If we want to compare the costs across the 3 regions then cost per case (i.e. cost per person or per capita cost) is a natural starting point. But depending on your question, these numbers may not meet the need. As an example, you may want to know how different the 3 regions are regarding efficiency. Some of the differences across the regions in straight average cost may be a result of differences in clinical complexity.
To adjust or control for this difference in clinical complexity when comparing costs, we can adjust the average cost of the region up or down using its CMI. This provides a Risk Adjusted Average Cost (RAAC)
Observations
Region C has the highest average cost
Region C has highest CMI; higher morbidity, more resource intensive population
Region C has the lowest RAAC or lowest adjusted cost
Lowest cost if all three regions were clinically similar
But be careful… variations in RAAC may be due to many reasons and may be justifiable (e.g. inefficiency perhaps, but could be a result of remoteness or other factors that affect healthcare costs and are outside the system’s control)
This RAAC concept (aka CPWC, cost per weighted case) applies to our other case mix methodologies as well, not just the popn grouping methodology. In the hospital world we have the Cost of a Standard Hospital Stay (CSHS)
Mock data

14. Population Based Funding
Funding for upcoming period
Proportion of weighted cases (i.e. case mix and volumes) is used to divide overall budget
(1) Region
(2)
Historical Funding
(3)
Weighted Cases
(4)
Proportion of Weighted Cases
(5)
Population Based Funding A
$2,455M
1.3M
59.1% = 1.3M 2.2M
$2,304M = .591 x $3.9B B
$556M
0.344M
15.6% = .344M 2.2M
$610M = .156 x $3.9B C
$889M
0.556M
25.3% = .556M 2.2M
$986M = .253 x $3.9B
Total
$3.9B
2.2M
100%
I mentioned funding applications earlier and here is an example of how the population grouping methodology can be used in funding allocations. Can look at the funding allocation as a splitting up of the pie. Each region receives a percentage of the healthcare budget. Their percentage is based on their weighted cases (persons), where the weights are the cost weights (could be concurrent or prospective depending on the design of the funding formula) from the population grouping methodology.
In this example of the same 3 regions from previous slides we can see the current allocations of funds across the 3 regions (column 2) and a proposed allocation based on the population risk adjustment grouping methodology (i.e. that takes the actual morbidity burden of the population in account and so is more equitable).
Note that this example is a very big simplification of how a case-mix based allocation would work and is meant to illustrate the basic concept of the use of case mix methodologies in funding. In reality the allocations would be more complex and would take other factors into account.
Mock data

15. The Future Continued silo-busting Ongoing evolution of Case Mix
Supporting government policy-making using high-level analysis of Case Mix
More population grouping, with more data 15