Risk-Adjustment Methodology:

Algorithmic application of risk levels to the U.S. population for COVID-19 infection

Risk adjustment tools are used to account for the underlying differences in expected healthcare outcomes between patients. These tools estimate comorbidity indexes based on actuarial data where each patient is assigned a score based on information on both demographic and health status. Without the use of risk adjustment models, payers and providers may game the system and only insure or choose to treat healthier than average individuals. Comorbidity indexes adjust for the presence of chronic conditions and confounding comorbidities and establish baseline estimates of health status among populations and cohorts.

There are several risk adjustment models available for consideration:

• The Charlson Comorbidity Index (CCI) [1]. This method uses data on 19 specific secondary diagnoses (comorbidities) to estimate a weighted index that predicts a patient’s risk of death within one year of hospitalization. Limitations to this index include the lack of inclusion of comorbidities, its reliance on having complete medical records, and not differentiating between comorbidities that are present on admission or are complications.

•  The Charlson Comorbidity Index (CCI) [1]. This method uses data on 19 specific secondary diagnoses (comorbidities) to estimate a weighted index that predicts a patient’s risk of death within one year of hospitalization. Limitations to this index include the lack of inclusion of comorbidities, its reliance on having complete medical records, and not differentiating between comorbidities that are present on admission or are complications.

• Elixhauser Comorbidity Index (ECI) [2]. This method was also developed using pre-determined clinical categories. It uses 30 categories (comorbidities) while relying on the ICD-9 coding system. ECI does not summarize risk in an index and has been shown to outperform CCI in predicting one-year mortality [3].

• Hierarchical Condition Categories (HCC) [4]. This model was developed by the Centers for Medicare and Medicaid Services (CMS). This model adjusts capitation payments by estimating a risk score of future medical cost for each beneficiary based on demographic information and patient health status. CMS HCC ranks diagnoses into unique categories that represent comorbidities with similar healthcare costs. Higher categories represent higher predicted healthcare costs. CMS HCC has been shown to outperform both CCI and ECI in predicting in-hospital and six-month mortality rate [5].

• Hierarchical Condition Categories (HCC) [4]. This model was developed by the Centers for Medicare and Medicaid Services (CMS). This model adjusts capitation payments by estimating a risk score of future medical cost for each beneficiary based on demographic information and patient health status. CMS HCC ranks diagnoses into unique categories that represent comorbidities with similar healthcare costs. Higher categories represent higher predicted healthcare costs. CMS HCC has been shown to outperform both CCI and ECI in predicting in-hospital and six-month mortality rate [5].

PurpleLab uses the CMS HCC model in our risk adjustment of the population for poor outcomes related to the COVID-19 pandemic.

We chose the CMS HCC as Risk Adjustment for the following reasons:

We chose the CMS HCC model for our map as in comparison to either the CCI or the ECI, CMS HCC:

It is designed for an older population

It is nearly exhaustive of relevant chronic conditions and comorbidities

It includes complications

The current version of CMS HCC (v24) includes 86 categories. Four categories have been either replaced or phased out since the model was first proposed. We include the four inactive categories in our risk adjustment algorithm. We maintain the mappings as concepts groups within HealthNexus™ (PurpleLab’s Medical Terminology Master Data Management Platform). These concepts groups were derived from:

It is designed for an older population.

It is nearly exhaustive of relevant chronic conditions and comorbidities.

It includes complications

In addition to being skewed towards an older population, it includes chronic conditions and comorbidities related to immuno-compromise as well as cardio-respiratory compromise.

The current version of CMS HCC (v24) includes 86 categories. Four categories have been either replaced or phased out since the model was first proposed. We include the four inactive categories in our risk adjustment algorithm. We maintain the mappings as concepts groups within HealthNexus™ (PurpleLab’s Medical Terminology Master Data Management Platform). These concepts groups were derived from:

Anonymized patients in PurpleLab’s Claims Repository were evaluated for whether or not they have an ICD9 or ICD 10 CM code that maps to a CMS HCC, to establish whether or a patient has one or more chonic conditions. Overlapping chronic conditions represent comorbidities. Using the CMS HCC model implies a 90 by 90 intersection which is too complex to visualize. For this reason, we have developed the following coding scheme:

Anonymized patients in PurpleLab’s Claims Repository were evaluated for whether or not they have an ICD9 or ICD 10 CM code that maps to a CMS HCC. Therefore,

1. If a patient appears in the Claims Repository but does not have ICD9 or ICD10 CM codes that map to a CMS HCC AND their age is less than 60 years old, then the patient is classified as having Low Risk.

2a. If a patient appears in the Claims Repository but does not have ICD9 or ICD10 CM codes that map to a CMS HCC AND their age is greater than 60 years old, then the patient is classified as having Moderate Risk.

2b. If a patient appears in the Claims Repository and has an ICD9 or ICD10 CM code that maps to one CMS HCC that is NOT indicative of either (i) immuno-compromise; and/or (ii) cardio-respiratory compromise. Then the patient is classified as having Moderate Risk.

3a. If a patient appears in the Claims Repository and has an ICD9 or ICD10 CM code that maps to more than one CMS HCC that is NOT indicative of either (i) immuno-compromise; and/or (ii) cardio-respiratory compromise. Then the patient is classified as having High Risk.

3b. If a patient appears in the Claims Repository and has an ICD9 or ICD10 CM code that maps to a single CMS HCC indicative of either: (i) immuno-compromise; and/or (ii) cardio-respiratory compromise. Then the patient is classified as High Risk.

4. If a patient appears in the Claims Repository and has an ICD9 or ICD10 CM code that maps to a more than one CMS HCC indicative of either: (i) immuno-compromise; and/or (ii) cardio-respiratory compromise. Then, the patient is classified as having Severe Risk.

We used ICD9 CM and ICD10 CM patient claim data spanning several years. This defines a prevalence-based approach (PBA) as opposed to an incidence-based approach (IBA). A PBA provides a cross-sectional view of the risk regardless of when the chronic condition or comorbidity first occurred. IBA estimates, on the other hand, are more suitable for calculating the value of prevention.

References:

1. Charlson ME, Pompei P, Ales KL, MacKenzie CR: A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. Journal of Chronic Diseases 1987, 40(5):373-383.

2. Elixhauser A, Steiner C, Harris DR, Coffey R: Comorbidity Measures for Use with Administrative Data. Medical care 1998, 36(1):8-27

3. Chu, Y., Ng, Y. & Wu, S. Comparison of different comorbidity measures for use with administrative data in predicting short- and long-term mortality. BMC Health Serv Res 10, 140 (2010).

4. Pope G, Kautter J, Ellis R, Ash AS, Ayanian JZ, Iezzoni LI, Ingber MJ, Levy JM, Robst J: Risk Adjustment of Medicare Capitation Payments Using the CMS-HCC Model. Health Care Financing Review 2004, 25(4):119-141.

5. Li, P., Kim, M.M. & Doshi, J.A. Comparison of the performance of the CMS Hierarchical Condition Category (CMS-HCC) risk adjuster with the charlson and elixhauser comorbidity measures in predicting mortality. BMC Health Serv Res 10, 245 (2010).