MedlinePlus Connect: Planning for Clinical Coding System Changes

PHASE II: ANALYSIS

ANALYSIS METHOD I: GENERAL EQUIVALENCE MAPPINGS (GEMS)

GEMS RANDOM SAMPLE

The Associate Fellow created an Access database with the GEMs text files and the existing MedlinePlus Connect files (APPENDIX F: ACCESS & EXCEL INSTRUCTIONS). A query retrieved 9cm to 10cm equivalence mappings, the code definitions, and all mapped Health Topics. Results showed the ICD-9-CM codes, the descriptions of the ICD-9-CM codes, the associated ICD-10 codes and their descriptions, MedlinePlus topics associated with each ICD-9-CM code, and the GEMs flags. While checking the data for errors, the Associate Fellow discovered a number of duplicates within the MedlinePlus Health Topics file and sent these topics to the MedlinePlus Connect team for review and clean-up. The Associate Fellow then uploaded the updated file into Access.

The forward mapping file (9-CM to 10-CM) contained 23,484 associations. Out of these 23,484 associations, 17,156 were associated with Health Topics. These 17,156 associations were pulled into an Excel workbook. A sample size of 376 was determined such that it would allow analysis of results with a 95 percent confidence level and a 5 percent margin of error⁹.

Prior to review, the author filtered the sample according to the flags (identical, etc.) to confirm that the percentages of the sample and population were consistent (results in Table 6). The author noted if the ICD-10-CM code mapped to the ICD-9-CM code matched the associated Health Topic.

Table : ICD-9-CM to ICD-10-CM GEMs

	Population (17156)		Sample (376)
Flag	#	%	#	%
Identical	2789	16.26%	52	13.83%
Approximate	14367	83.74%	324	86.17%
No Map	14	0.08%	0	0.00%
Combination	1568	9.14%	44	11.70%
Scenario	1568 (1-6)	9.14%	44 (1-4)	11.70%
Choice	1568 (1-2)	9.14%	44 (1-2)	11.70%

Having determined that the sample and population percentages of each type of data were approximately consistent based on the flags, the Associate Fellow reviewed all identical mappings and approximate mappings within the sample and indicated if the ICD-10-CM code and the Health Topic were a good association or if the ICD-10-CM code and the health topic were not a good association.

The Associate Fellow categorized the mappings within several categories:

• 
  No Problem: The mapping was appropriate and provided a good association to a MedlinePlus Health Topic.
  
• 
  ‘Delete Health Topic’ or ‘Delete Health Topic. Add Health Topic’: The mapping requires a Health Topic to be deleted and in some cases another Health Topic added in order to provide an appropriate mapping.
  
• 
  Need Additional Health Topic: The initial Health Topic assigned using the GEMs is correct. Adding an additional Health Topic will further enhance the mapping.
  
• 
  Delete Health Topic: The Health Topic should be deleted because it cannot be correctly associated with the code.
  
• 
  M+ Problem: The GEMs mapping is correct but the original MedlinePlus Health Topic assigned to the ICD-9-CM term is incorrect and therefore this mistake still exists.
  
• 
  Delete association. The association between the ICD-10-CM code and the MedlinePlus Heath Topic is not appropriate and no MedlinePlus Health Topic is appropriate for the ICD-10-CM code.
  

Discussion of this analysis is found in the Results section.

GEMS SELECTIVE SAMPLE

The Associate Fellow analyzed an additional sample of 1,070 mappings from S00 – T89: Injury, Poisoning and Certain Other Consequences of External Causes chapter of the GEMs forward mapping file to better determine patterns from one chapter of ICD-10-CM. The Associate Fellow chose this chapter because it included the largest number of issues from the random sampling, the largest number of new overall codes, and different organization than previously found in ICD-9-CM where injuries were grouped by body part instead of injury. The author reviewed sections of codes from each distinct subset within S00 and T89. The sample was not a true random sample and, while suggestions can be made from the chapter, it is not guaranteed to provide an accurate representation of the entire population of the chapter or of ICD-10-CM and the GEMs mappings. Discussion of this analysis is found in the RESULTS section.

ANALYSIS METHOD 2: LISTER HILL CENTER (LHC) – UMLS ALGORITHM

Olivier Bodenreider and Lee Peters from LHC used the algorithm originally created to map ICD-9-CM codes to MedlinePlus Health Topics to create a mapping from ICD-10-CM to MedlinePlus Health Topics, using relationships found within the Unified Medical Language System (UMLS). Their mapping did not include ICD-10-PCS codes.

The algorithm uses three main methods to provide the mappings:

1. 
   (I) Synonymy: Use the UMLS to determine synonymous terms. Synonymous terms are codes with the same UMLS concept, or CUI value (Concept Unique Identifier). Approximately 1% of the mappings are synonymous.
   
2. 
   (A) Explicit mapping found in a given source: If no direct equivalence or synonymy is found, the algorithm finds mappings asserted by another source in the UMLS. Mappings are used when one source points to a concept that is synonymous to a Health Topic.
   
3. 
   (G/x) Mapping identified from the ancestors of the source concept: For any given ICD-10-CM concept, take all UMLS ancestors of ICD-10-CM and determine if these ancestors are synonymous with any of the Health Topics or mappings found in method 2. Filter concepts that are ancestors of these synonyms and keep the ancestors closest to the child. The majority of mappings are from this method. Three categories came from this method: G/Parents (G/P), G/Children (G/C), and G/Siblings (G/S)
   
4. 
   (O) Other
   

The Associate Fellow reviewed code sections from the first two versions of the algorithm’s output and identified areas with correct and incorrect mappings. The Associate Fellow also realized that Health Topic “see” references from the MedlinePlus Connect XML files were included in the mappings. The Associate Fellow worked with Olivier Bodenreider and Lee Peters on this issue. The LHC researchers resolved all “see” references to the Health Topics so the final mapping only includes ICD-10-CM codes and Health Topics. At the end of the Associate Fellowship Spring project, the team conducted a handoff to Rex Robison and Anna Ripple who will continue the review and analysis with the MedlinePlus Connect and LHC teams.