This site is 100% ad supported. Please add an exception to adblock for this site.

Group Ins.- Health Risk Adjustment


undefined, object
copy deck
Risk Adjustment
Risk Adjustment is the process of adjusting payments to insurance companies or providers in order to reflect the health status or riskiness of the insureds.
Examples of Risk Adjustment:ï‚Ÿ Provider Reimbursement
An HMO may pay doctors with sick or old patients more than doctors with healthy or young patients, to compensate them for the extra risk.
ï‚Ÿ Provider Profiling
Many doctors’ bonuses depend on their utilization level. Risk adjustment can determine whether high utilization was due to especially risky patients, or due to the doctor’s inefficiency.
ï‚Ÿ Transferring money between insurance companies
 Insurance companies and MCO’s have an incentive to market only to young, healthy applicants, especially if there are rating restrictions which prevent the company from charging unhealthy applicants more than healthy ones.
 Under a Risk Adjustment program, the ins cpys with safe enrollees must pay money to ins cpys with risky enrollees. This enables a carrier to sell insurance to risky ph’s without undue punishment, and therefore benefits society.
 Specifically, the insurers (or MCOs) with the healthiest policyholders must contribute money into a risk pool. The insurers with the riskiest policyholders receive distributions from the pool, either prospectively (in advance) or retrospectively (after the year’s experience is over).
ï‚Ÿ Medicare+Choice HMO contracts
 The federal government uses risk adjustment to determine the appropriate premiums to pay HMO’s to cover its Medicare patients.
ï‚Ÿ Underwriting
 A risk adjuster (or risk assessment formula) is a formula which takes as input information about the enrollees in a group (demographics, health status, past claims, prescription drug information, etc.), and outputs a single number representing the riskiness (that is, expected medical claim cost) of the group.

 If a risk adjuster is sophisticated enough, it can be used as a kind of “underwriting machine”.
When a group comes up for renewal, its claims data is entered into the MCO’s risk adjustment system, which assesses the level of deterioration or improvement in the group’s health status.
ï‚Ÿ To protect ins cpys from Guaranteed Issue and Community Rating regulations
Under these conditions, the ins cpys cannot match their premiums to the risks that they take on. Not only are they at risk of going insolvent, but they also have a strong incentive to market only to healthy people, which damages society.
 To protect ins cpys’ solvency
 To encourage companies to compete on the basis of efficiency, rather than by screening and avoiding risks.
 To encourage competition and marketing in the small group and individual markets
Small groups and Individuals have volatile claim levels, so they are risky policyholders to insure. But Risk Adjustment will allow a small-group or individual carrier to share its risks with other carriers.

ï‚Ÿ To facilitate comparison shopping by consumers
 Just because one ins cpy charges less than another doesn’t mean it is a more efficient company; it might just have healthier policyholders.
ï‚Ÿ To protect ins cpys from antiselection (see example on next page)
The two steps in Risk Adjustment are:
1. Risk assessment
(using a formula to determine the riskiness of policyholders)
which consists of:
1A. Risk Classification (dividing the insureds into classes)
1B. Risk Measurement (quantifying the claim cost level for each class)

2. Payment adjustment
(adjusting payments based on the differences in risk, as measured by the risk assessment step)
1A. Risk Classification
Insureds can be classified according to:
ï‚Ÿ Demographics (age, sex, location)
ï‚Ÿ Claim / Utilization levels while insured
ï‚Ÿ Medical history
ï‚Ÿ Diagnosis codes
ï‚Ÿ prescription drug use
ï‚Ÿ Perceived health status (i.e. self-reported data from patient surveys)
ï‚Ÿ Lifestyle / Behavior (smoking, weight, substance abuse)
1B. Risk Measurement
Carriers A and B (from the above example) will each create three classes, “low-risk”, “avg risk”, and “high-risk”.
Relative Risk Factors by Class
 Each class can be given a relative risk factor, equal to the average expected claim cost (CC) for the class divided by the average expected claim cost for the carrier’s whole portfolio.

Avg CC for the class
relative risk factor = ———————————
Avg CC for the carrier

 Recalling that the claim costs by class are $100, $200, and $600, and that Carrier A’s overall average cost PMPM is $175, Carrier A’s relative risk factors for each class are 0.571, 1.143, and 3.429. For example, 3.429 = $600 / $175. Carrier B’s relative risk factors are 0.444, 0.888, and 2.667 for those same three classes.
This type of risk measurement can be used for ratesetting purposes, but it cannot be used for risk adjustment because the carrier’s portfolio is not being compared to any other carrier’s. There was no need for the textbook to include this example. It is not relevant to risk adjustment.
Relative Risk Factors by Carrier
ï‚Ÿ This method can be used for risk adjustment (the transfer of money between carriers)
ï‚Ÿ Each carrier is given a relative risk factor, which determines whether its risks are favorable or unfavorable compared to the market.

Avg CC for the carrier
relative risk factor for a carrier = ————————————
Avg CC for the market

ï‚Ÿ The total market, companies A and B together, has:
 80 low risks
 100 avg risks
 20 high risks. (Again, always work with whole numbers, to make the math easier)
 The market average claim cost is thus $200.

 Carrier A’s risk factor is $175/$200 = 0.875.
 Carrier B’s risk factor is $225/$200 = 1.125
ï‚Ÿ A carrier whose risks are average for the market would have a risk factor of 1.0.
In the above examples, each insured fell into one of three discrete classes; plus, the classification was known to be correct. In real life, not only are insureds difficult to classify, but each insurer will be using a different classification system.
The state government, which adminsters a Risk Adjustment program, must have an objective, identical way of assessing the riskiness of each ins cpy’s enrollees so that it can determine who must pay whom.
The easiest type of risk assessment formula would be one based on each insurer’s historical and expected claim costs per member. However, such a formula has three major problems:
ï‚Ÿ High claim costs may just be due to poor claims administration or poor provider contract negotiation skills. Insurers should not be compensated for these types of inefficiencies.
 Some insurers don’t have enough data to accurately predict expected claims.
 Insurance companies can falsify their “expected claims” prediction in order to get more money from the pool.

Instead, risk adjustment formulas should only be based on specific, unmistakable characteristics of the insureds that will nevertheless be strongly correlated with claim costs. Characteristics commonly used include Age, Sex, Location, and specific medical conditions like AIDS or cancer. This way, the risk adjustment formula is fair and equitable to every participating company.
A risk adjuster (risk-assessment method) must [be]:
ï‚Ÿ Accurate in predicting claim costs
ï‚Ÿ Simple and Cheap to administer
ï‚Ÿ Use data that is routinely available
A risk adjuster that’s 100% accurate but that uses all kinds of combined medical diagnoses is useless: Ins Cpys won’t be able to collect the needed data.
ï‚Ÿ Equitable to the participating ins cpys, and
ï‚Ÿ Have industry support
else there will be lawsuits, which will jeopardize the effectiveness of the system.
ï‚Ÿ Mandatory to participate in, not voluntary.
Else, ins cpys with favorable risks would choose not to participate (since they would be subsidizing other insurers’ bad risks), thereby defeating the purpose of pooling.
 Protect the confidentiality of patients’ diagnoses
 Compensate insurers for claim costs due to their policyholders’ risks. Not for claim costs due to the insurer’s administrative inefficiencies or failure to properly control utilization.
ï‚Ÿ Not subjective or discretionary.
For example, the indicator diseases must be ones that don’t involve subjectivity in diagnosis.
ï‚Ÿ Cannot be gamed
i.e. insurers must not be able to falsify information in order to get higher payouts.

(continued on next page)
Assessment Method Advantages Disadvantages

Demographics alone (age, sex, location) ï‚Ÿ Objective
ï‚Ÿ Data easily available
ï‚Ÿ Easy to understand ï‚Ÿ A poor predictor of medical claim costs.
Insurer’s claim costs  Good predictor of future claim costs
 useful for large groups, because doesn’t require collecting data.  Could be because of inefficiency, not risk
ï‚Ÿ Subject to gaming
ï‚Ÿ New ins cpys have no data
Pharmacy data ï‚Ÿ Good predictor
ï‚Ÿ Drug records complete and available ï‚Ÿ drug use is influenced by physician practice patterns
Specified Medical Diagnoses; medical info
(e.g. blood pressure) ï‚Ÿ A useful addition to other systems in use ï‚Ÿ difficult to collect this data
ï‚Ÿ not all ins cpys have it
ï‚Ÿ lack of patient confidentiality
Perceived health status or behavioral / lifestyle factors (smoking) This cell was
left blank ï‚Ÿ subjective
ï‚Ÿ can be gamed (by form of the survey)
ï‚Ÿ expensive to collect data (only good for small groups)
A Combination ï‚Ÿ More accurate than a single method ï‚Ÿ Complicated / Expensive.
Measures of Predictive Accuracy
Standard R-squared:
ï‚Ÿ a single number
ï‚Ÿ standardized to between 0 to 1, so can compare different studies

ï‚Ÿ Overly sensitive to effect of large claims (since it squares every error)
Mean Absolute Prediction Error
= AVERAGE{ |A – E| }
over each person in the study, where A is actual claim cost and E is expected claim cost, as predicted by the risk adjuster.
ï‚Ÿ a single number
ï‚Ÿ not overly sensitive to large claims

ï‚Ÿ not scaled to between 0 and 1.
Cumming’s Prediction Measure
Mean Absolute Prediction Error
Cumming’s Measure = 1 – ———————————————————
Mean Absolute Deviation from Average

Mean Absolute Deviation from Average = AVERAGE{ |E – | }
( is the average actual claim cost)

NOTE: These formulas are only verbally described in the textbook; my interpretation might not be correct.

Advantages of Cumming’s Measure:
ï‚Ÿ All three.
ï‚Ÿ None.
Group Level Measures
These measures don’t examine the A vs. E for each person individually. They just compare the A for a group of people vs. the sum of the E’s for the individuals in that group.

ï‚Ÿ Predictive Ratio = Egroup / Agroup.
 close to 1 means the risk adjuster works well.

ï‚Ÿ The group studied can be:
 randomly chosen individuals
 a non-random group like “people with a certain condition”
 a particular employer group.

Deck Info