anobel / icdtohcc

Generating Medicare HCC codes from ICD codes

ICD to HCC

Generate Medicare Hierchical Condition Categories (HCC) based on ICD codes

Introduction

• many to many relationship
• ICD9/10 are mapped to Condition Categories (CC), and a hierarchy based on disease severity is applied to the CCs to generate HCCs.

Data

The ICD/HCC mappings were obtained from CMS. CMS provides SAS macros for assigning HCCs and HCC scores based on ICDs, adjusted annually. We have implemented assigning HCCs (but not scores) in R. The original data is available from CMS, or in the github repository at /crosswalks/originalCMS_xw.

Between 2007-2012, there were 70 HCCs (Version 12). In 2013, this was expanded to 87 (Version 20). We use the labels for the more inclusive mappings (post-2013) for all years for consistency. This means that in pre-2013 data 17 HCCs will be structurally zero.

Source

icdtohcc.R

• Import condition categories, hierarchy rules, and labels, by year from CMS (for both ICD9/10)
• Apply CCs and implement hierarchies to generate HCCs, starts with wide patient data, ends with long data Source with markdown

Future Plans

• split into separate functions
• ?incorporate into icd package

icdtohcc's Issues

Applying Hierarchies

HCCs are defined when a severity hierarchy has been applied to the condition categories (CCs). This hierarchy sometimes differs from year to year. For many purposes, the most inclusive hierarchy is appropriate, but we would like to be able to apply year-specific hierarchies.

In each year, there are situations with overlapping hierarchies that must be applied in successive order. While it would make sense to loop through them, it would be nice to apply this in a vectorized fashion. Efficiency will also be important, as administrative data sets can get quite large. The current data set I'm working with as 12.5million patients, with up to 25 diagnoses per patient.

Also, much of the administrative data with patient diagnoses is in wide format, with many diagnoses per patient (up to 25, on per column). It may be faster to convert this to long format, apply hierarchies, and then covert back to wide format...I'm still experimenting with this aspect. thoughts, @mpdakkak?

create as functions

rmarkdown -> script

eventually convert the markdown document to an rscipt that can be run. will unchunk and edit comments once major development issues resolved

integrate data and functionality into icd package