Risk Adjustments and Bundled Payments

The CMS guidelines for the Medicare bundled payment initiative suggest that CMS will be receptive to the use of risk adjustments in establishing the episode payment budget for DRGs to be paid under this program.  This prompted the Singletrack Analytics research team to fire up its HCC grouper to explore how risk adjustment might be integrated into a bundled payment process.

The Hierarchal Condition Categories (HCC) grouper develops risk scores for individual patients through an algorithm that applies a risk adjustment factor to various diagnoses, or combinations of diagnoses, and also to the patient’s age and gender.  The presence of a particular diagnosis generates a risk adjuster for the diagnosis, and multiple diagnoses can generate multiple risk adjusters.  The final risk adjuster for a patient is the aggregate of all of the individual risk adjusters.  An example of a risk adjustment calculation for an individual patient is shown below.

HCC Group

Risk Score

Age/Gender

1.09

Congestive Heart Failure

0.82

Hip Fracture/Dislocation (3)

0.86

Intestinal Obstruction/Perforation

0.62

Rheumatoid Arthritis and Inflammatory Connective Tissue Disease

0.69

Septicemia/Shock

1.52

Specified Heart Arrhythmias

0.59

Vascular Disease with Complications

1.22

Total:

7.40

Correlation between Risk Score and Cost

The objective of risk adjustment is to identify a set of factors based on a patient’s condition that can explain variations in cost.  While the adjusters are not expected to predict costs of individual patients, aggregations of the risk factors are expected to explain differences in costs among groups of patients, which those costs are related to the patient’s condition.  Therefore, in the examples below a high degree of correlation between risk score and individual patient cost is not expected; however moderate correlation would indicate some level of relationship between risk scores and cost.

In some DRGs, the HCC score was found to be generally correlated with the cost per episode.  In other DRGs less correlation was noted.  The charts below show the relationship between episode cost and HCC score, with each dot representing a bundled payment episode.  In these charts below, the HCC score shows a relatively high correlation (R2 = .48) with the episode cost for DRG 313 (Chest pain).  However, the relationship for DRG 470 (Major joint replacement) is significantly lower (R2 = .17).  This may indicate that medical DRG costs are more significantly affected by the presence of complicating factors than are surgical DRGs.

  

Episode Cost by HCC Category

For purposes of analysis, each patient was assigned a category based on their aggregate HCC score.  These categories were as follows:

HCC Score Range

HCC Cost Category

Percent of Patients

0 to 2

Low

60%

>2 to 5

Med

25%

> 5 to 10

High

10%

> 10

Extreme

5%

Using these groupings, the total cost to CMS (i.e. the payments to the providers) for each episode is shown in the graphs below.  The bars are colored to indicate the HCC cost categories as shown in the graph legend.

Several examples are shown below for a variety of DRGs or combinations of similar DRGs. Of interest in most of these examples is that the higher cost cases are generally correlated with higher category risk scores.  This is indicated by the fact that most of the lower-cost patients have blue bars, while the higher-cost patients are more likely to have green or purple bars. 

DRGs 233-236 - CABG

DRG 312 - Chest Pain

 DRGs 190-192 - COPD

DRGs 226-251-Percutaneous CV Procedure

 DRGs 713-714 - TURP

Risk Adjuster Effect on Post-Discharge Cost

An additional factor explored by the research team was the possibility that risk factors have a greater effect on post-discharge cost than on inpatient costs.  This appears to be the case for many DRGs, as shown below.  These charts show the percentage of total cost during each phase of the episode separated by the HCC categories.  The totals of each color bar add up to 100%, so the graphs indicate the percentage of the total occurring in each of the episode periods.

In most DRGs, the greatest percentage of cost for the “Low” HCC category was for inpatient costs, with smaller percentages for the 30 day and 90 day episode.  However, in three of the DRGs the post-discharge costs exceeded the inpatient cost for the “High” and “Extreme” HCC’s categories.

DRG 470 - Joint Replacement

DRG 312 - Chest Pain

DRGs 190-192 - COPD

Conclusions

Although this is not a scientific study, the anecdotal results appear to indicate a relationship between cost per episode and HCC score.  While it may be useful to include this type of risk adjustment in the bundled payment proposal, CMS cautioned that it will critically review any type of risk adjustment from the possibility of “diagnosis creep” that can occur when more accurate diagnosis coding occurs over time, which would raise the average HCC scores.  In the final regulations for accountable care organizations, CMS did not allow the same type of HCC-based risk adjustments for ACOs that are used for Medicare Advantage plans and PACE programs, presumably because of its concern over this effect.  Although these objections could be overcome by adjustments to the methodology, the development of such adjustments may not be worthwhile.

However, the use of HCC scoring may be of benefit in designing care plans, particularly in planning for post-discharge care.  Given the patients with higher HCC scores and therefore a greater number of medical complications have significantly higher post-discharge costs, it may be useful for the clinical care team to carefully review all of the diagnoses for all patients to identify those patients having medical conditions that may create significant post-discharge costs.  These patients may be tagged for more intense follow-up, with focus on care for their particular medical conditions.  This process can be done anecdotally without the use of the HCC grouping software by paying attention to key diagnoses that have high risk scores.  This process would require the availability of diagnosis information upon patient discharge, and would rely on careful and complete coding of all medical conditions by admitting and attending physicians throughout the patient stay.  Utilizing this information could allow the care team to focus on patients having the greatest opportunity for post-discharge costs, while potentially allowing lower-risk patients less intense follow-up.

In addition, HCC risk scores may have a place in the payment reconciliation process by providing explanatory information for cases whose costs exceeded the bundled payment targets.  By measuring the risk factors in the baseline population use to develop the bundled payment budget and comparing them to the risk scores of the patients in the bundled payment participation period, differences in patient mix can be identified and qualified.  This may assist in evaluating the success of the bundled payment system, and potentially provide information for improvements.

Data Used

This project utilized the 2010 “Medicare 5% sample” data, which contains data for 5 percent of the Medicare beneficiaries.  Episodes were built starting from admission date, with the timeframe categories including the inpatient stay period, the 30-day post-discharge period, and the 31-90 day post-discharge period.  A small amount of cost is attributed to the “91+” period, which occurs when a readmission crosses the end of the 90-day episode boundary.  In this case costs are included in the episode until discharge, even though those costs will occur after the end of the 90-day episode.