MedlinePlus Connect: Planning for Clinical Coding System Changes

LISTER HILL MAPPING EVALUATION

Olivier Bodenreider and Lee Peters from LHC ran the original algorithm to determine whether the algorithm would map ICD-10-CM and the Health Topics and if it could be improved. They ran the algorithm which used relationships within the Unified Medical Language System (UMLS) to map from ICD-10-CM to MedlinePlus Health Topics.

Four types of mapping methods are used in the algorithm:

Mapping Method	Percent of total output
I (synonymous)	0.8%
A (explicit mapping found in a given source)	2.4%
G/x (mapping identified from ancestors of source concept)	71.2%
O (other)	0.5%
None available	25.2%

The initial review of a sample from the initial mapping proved that this method does not provide a “magic” solution. While it does provide a mapping between ICD-10-CM and the Health Topics, the initial output resulted in a large number of mappings that are not appropriate or reasonable for MedlinePlus Connect. While mappings are of a higher quality when they have a close or synonymous relationship within the UMLS, less than 4% of initial mappings resulted from these close relationships.

The LHC researchers plan to continue refining the algorithm. This process is ongoing and will continue with the help of Rex Robison, who is currently on detail with MedlinePlus Connect from the NIH Library, and Anna Ripple from Lister Hill. Once these mappings are analyzed, a sample from Lister Hill can be compared with the GEMs findings to determine which provides the largest number of accurate mappings and the least amount of manual review.

Specific recommendations for LHC-UMLS analysis next steps are available in Recommendations.

ADDITIONAL OUTCOMES & LESSONS LEARNED

The process of asking these first two questions also resulted in several additional and unexpected project outcomes.

• 
  Several new Health Topics were identified during the analysis process and are being evaluated and potentially added to the Health Topics list. Head Injuries is one example.
  

• 
  The process of creating the Access database identified inconsistencies within the Health Topics file which the team has since corrected.
  
• 
  The team is fine-tuning the current rules for how Health Topics are assigned. It should complete this step before beginning the mapping process.
  
• 
  In addition to his help with ICD-9-CM and ICD-10-CM, Kin Wah provided an expanded SNOMED CT CORE Problem List subset file of the descendents of the 5000+ SNOMED CT CORE Problem List subset concepts already associated with a MedlinePlus Health Topic. Naomi Miller has reviewed these and is in the process of updating the SNOMED file used by MedlinePlus Connect. This more than doubled the number of mappings between SNOMED concepts and Health Topics from approximately 5,500 mappings to 11,500 mappings.
  

The team also learned several valuable lessons while conducting the project. First, random sampling of codes should be used in combination with other methods of analysis. The review of selected code groups, while smaller and not as statistically significant, provided greater insight into the GEMs structure since one could view the patterns within ICD-10-CM’s hierarchy. Second, there is currently no one perfect solution. While a fully automated system would be ideal, for the purposes of MedlinePlus Connect, review is still required in order to provide the best possible solution for health care providers and their patients. While automation is the ideal, semi-automation is the reality. Third, the project also reinforced the fact that projects that require in-depth analysis and large scale pattern recognition can benefit from multiple perspectives.

Several personal lessons learned also resulted from the project. The author chose the project because of an interest in integrating consumer health education within EHRs and a desire to gain knowledge about clinical vocabularies and the world of vocabulary mapping. The project provided valuable insight about clinical topics in general, mappings, and clinical vocabularies and their uses.

RECOMMENDATIONS

Both the initial review of the LHC-UMLS mapping and the review of the GEMs demonstrate that there is no perfect solution that will provide an easy, one-step, fully automated solution.

GEMs RECOMMENDATIONS

If the GEMs are used to support ICD-10-CM, the team should review the GEMs specific recommendations discussed below:

1. 
   Work should begin within the year and anytime after the next release of ICD-9-CM and ICD-10-CM this upcoming October. At this point, both codes will be partially “frozen” until October 2014, after the 2013 transition to ICD-10-CM. Waiting for this next release will ensure that the MedlinePlus Connect team is working with the most up-to-date files without worrying about excess additions. If the GEMs are used, the updated GEMs files, ICD-9-CM descriptions, ICD-10-CM descriptions, and MedlinePlus Health Topics will all need to be inserted into Access before analysis. Instructions for where to find each of the files is available in APPENDIX F: ACCESS & EXCEL TIPS & INSTRUCTIONS.
   

2. 
   Begin review with chapters identified as requiring the most review, as identified in the 13 unique chapters below. It is recommended to analyze them for patterns (similar to the patterns found in the S00-T89 chapter on Injuries and Poisonings) and to use these patterns throughout the review. The chapters are listed according to the method used to identify the chapter and therefore include some duplication (italicized chapters included in each list):
   

1. 
   Chapters with substantial growth:
   
   1. 
      M00-M99: Diseases of the musculoskeletal system and connective tissue,
      
   2. 
      O00-O9A: Pregnancy, childbirth, & puerperium
      
   3. 
      S00-T88: Injury, poisoning, and certain other consequences of external causes
      
   4. 
      V00-Y99: External causes of mortality
      

2. 
   Chapters with the majority of issues as identified from the random sample:
   
   1. 
      E00-E89: Endocrine, nutritional and metabolic diseases
      
   2. 
      K00-K94: Diseases of the digestive system
      
   3. 
      M00 – M99: Diseases of the musculoskeletal system and connective tissue
      
   4. 
      P00 - P96: Certain conditions originating in the perinatal period
      
   5. 
      S00 – T88: Injury, poisoning and certain other consequences of external causes
      
   6. 
      Z00 – Z99: Factors influencing health status and contact with health services
      

3. 
   Chapters that include an overwhelming number of code pairs from the unanalyzed backward mapping file and subjectively identified as needing greater review based on an initial scan:
   
   1. 
      E: Endocrine, nutritional, and metabolic diseases
      
   2. 
      F: Mental and behavioral disorders
      
   3. 
      H: Disease of the eye and adnexa; Diseases of the ear and mastoid process
      
   4. 
      I: Diseases of the circulatory system
      
   5. 
      L: Diseases of the skin and subcutaneous tissue
      
   6. 
      M: Diseases of the musculoskeletal system and connective tissue
      
   7. 
      O: Pregnancy, childbirth, and the puerperium
      
   8. 
      S & T: Injury, poisoning, and certain other consequences of external causes
      
   9. 
      V, W, X, Y: External causes of morbidity
      
   10. 
       Z: Factors influencing health status and contact with health services
       

3. 
   Focus on code pairs with combination flags during review. The combination flags often indicated a code pair with an issue in the S00-T89 chapter. While these combination flags are important to review, all codes must ultimately be reviewed for quality.
   

4. 
   In addition to reviewing the chapters with the majority of issues for patterns, it is recommended to also review each of the ICD-10-CM chapters for patterns similar to those in the injuries chapter. Manual review will be reduced if patterns are identified and Health Topics applied to sections of codes from the beginning, especially for those terms lower down the ICD10CM hierarchy. These patterns can be used to automate mappings of unmapped codes and to automatically apply Health Topics to sections that require minimal review. This will help speed up the process on chapters that already require minimal review.
   

5. 
   Use the master file that includes forward and backward mappings for the GEMs review.
   
   1. 
      View duplicates and ensure consistency throughout the file.
      
   2. 
      Focus on the approximate code pairs, especially those in the chapters identified for issues and growth. Identical pairs will require minimal review.