CDPS risk scores are calculated at the individual level, but are applied at the group level. An individual risk score includes the sum of the intercept (or baseline), demographic (i.e. age and gender) weights, and weights for all indicated CDPS (or MRX or CDPS+Rx) categories. As described above, CDPS weights are hierarchical within major categories but are additive across major categories. In practice, this means, for example, that an individual will receive higher scores for multiple chronic conditions across different major categories, such as Cardiovascular and Diabetes, but will only receive the highest score for multiple conditions within a major category (e.g. for heart failure but not also for hypertension).

The CDPS model was developed in 2000 using data from seven Fee-for-Service (FFS) Medicaid programs.  The model received major updates in 2009 (using national FFS Medicaid data from 2002-2005) and in 2014 (using additional national FFS Medicaid data from 2011). CDPS has also received regular annual updates to include the most recent ICD and NDC codes.

We have completed updating CDPS using data from three national Medicaid managed care plans from 2017-2019. We expect two main improvements in the CDPS model:

• To the extent that new treatments and technology have changed how patients are treated, the relative weights estimated with 2011 data may not accurately reflect relative cost in 2019.
  
  
• To the extent that patients are treated differently in FFS vs. managed care, the regression weights estimated with FFS data may not to reflect patterns of care in managed care organizations.

Participating Medicaid managed care plans are large national organizations with Medicaid contracts in a broad cross-section of states.

Resource:

• Summary of CDPS revisions (PDF)

Updating The Chronic Illness and Disability Payment System

CDPS Analysis Methods
The 2024 revision focused on clinical review of the 2022 model. Clinical experts were engaged for each CDPS major category. Each was provided with a list of relevant diagnoses and their assignment in the CDPS model. The experts were asked to identify any diagnoses that were not in the model that they felt should be considered for the model, any diagnoses that were included that they thought should be excluded, and any diagnoses that they felt were assigned to an incorrect CDPS category (i.e. either too high or too low). Based on this feedback, regression analyses were conducted for each major category with (sometimes new) stage 1 groups. Results were considered, and changes were model to the model as appropriate.

Results
The 2024 CDPS model has 56 CDPS categories within 19 major categories. This is an increase of four categories from the previous model. New categories were created for CNSVH, CERH, METL, and HEMH. The 2024 model includes 22,646 ICD 10 diagnoses: 1,125 diagnoses were dropped from the previous model and 1,534 were added. The majority of dropped diagnoses were from EYEVL (321), METVL (315) and CNSL (134). The majority of added diagnoses were assigned to CAREL (317), GENEL (228), and PRGINC (207). Among the 21,203 diagnoses in both models, 19,369 (92%) were assigned to the same CDPS category. The majority of codes were reassigned in the major categories for CAN (503), CER (240), EYE (225), and CNS (207). R-squares increased slightly in this revision in the concurrent model, from .24 to .25 for the disabled, from .11 to .12 for children, and from .35 to .37 for adults. R-squares for the prospective model remained the same for the disabled at .13 and children at .06, but increased from .13 to .14 for adults.

• Improving Health-Based Payment for Medicaid Beneficiaries: CDPS (PDF)
• The Medicaid Rx Model (PDF)
• Updating the Chronic Illness and Disability Payment System (PDF)

Additional Resources:

• CDPS 7.0 FAQs (PDF)
• Summary of Risk Adjustment Efforts Across States (PDF)
• Herbert Wertheim School of Public Health and Human Longevity Science
• Division of Health Policy
• Guidelines for Using CDPS Models for Risk Score Development (PDF)
• Rare Diseases FAQ (PDF)