MedlinePlus Connect: Planning for Clinical Coding System Changes

APPENDIX D: ORIGINAL MAPPING GUIDELINES

ORIGINAL GUIDELINES FOR MAPPING ICD-9-CM CODES TO MEDLINEPLUS HEALTH TOPICS

Background

In order for electronic health record/personal health record systems to link their records to MedlinePlus, we need to provide some kind of mapping file from codes used in charts to MedlinePlus health topics.

There are over 15,000 ICD codes for diagnoses (corresponds to volumes 1 and 2.) We are using ICD9-CM (because that’s what Medicare uses.)

Initial Guidelines:

1. 
   We will map where there are clear, solid, unambiguous matches. We will not stretch meanings to include maps to MedlinePlus topics. Example: ICD code for tracheostomy status—do not map to M+ topic Critical Care. This is the only place where M+ has information on tracheostomy, and there is a M+ see reference Tracheostomy see Critical Care. However, a person may have a tracheostomy as an outpatient. We would consider this to have no match.
   
2. 
   We will not “explode” ICD9 codes at the present time. Reason: not always valid. Examples:
   
   1. 
      099 is code for Other Venereal Diseases, which could be mapped to Sexually Transmitted Diseases. However, under it, 099.3 is Reiter’s Disease, which we feel should map to Arthritis and not Sexually Transmitted Diseases, as per MeSH.
      
   2. 
      595 is Cystitis. 595.1 is Chronic Interstitial Cystitis. 595.9 is cystitis, unspecified. We would map 595 to Bladder Diseases and 595.1 to Interstitial Cystitis. But 595.9 is Bladder Diseases. Exploding would remove our interstitial cystitis term.
      
   3. 
      We do not think it would be possible for us to recommend exploding or not exploding on a case-by-case basis at this time.
      
3. 
   We will use the broad M+ topic for a disease or condition, not topics that include
   
   1. 
      Age groups: e.g., use Asthma, not Asthma in Children
      
   2. 
      Gender: e.g., use Heart Diseases, not Heart Diseases in Women
      
   3. 
      Pregnancy: If there is no existing pre-coordinated code a pregnancy-related topic, e.g., AIDS in pregnancy, we will use the separate topics AIDS and Pregnancy. However, if there is a pre-coordinated term, e.g., high blood pressure and pregnancy has 55 separate ICD-9 codes, we will map those to the specific topic High Blood Pressure in Pregnancy.
      
4. 
   If an ICD-9 code can be represented by more than one M+ health topic, we will provide maps to multiple topics. (We didn’t decide up to how many, but 4 seems reasonable.)
   
5. 
   If there is no good match for an ICD9 code, we will code as:
   
   1. 
      Never: there will never be a M+ topic match (e.g., History of Non-Compliance; Refusal of Flu Shot)
      
   2. 
      Search: the ICD description could provide results if searched in M+ but there is no good topic match.
      
   3. 
      No match: there is no good match at this time, but we could develop a topic that would match in the future.
      

We used the web site at http://www.icd9data.com/ to double check on codes. We only coded from volume 1 and 2, diagnoses. We did not use the drugs, procedures, or the HCPCS codes.

Things to know about ICD codes

There are several layers of codes. So under disorders of the esophagus, you have, for example (this is not the complete list)

• 
  530, diseases of esophagus
  
• 
  530.0, achalasia
  
• 
  530.1, esophagitis
  
• 
  530.10, esophagitis unspecified
  
• 
  530.11, reflux esophagitis
  
• 
  530.19, other esophagitis
  
• 
  530.8, other specified disorders of the esophagus
  
• 
  530.81, esophageal reflux
  
• 
  530.89, other diseases of the esophagus
  
• 
  530.9, unspecified disorder of the esophagus
  

APPENDIX E: NOTES FROM SELECT MEETINGS

Kin Wah Fung Meeting Notes

Date: 4/27/11 at 11am

Attendees: Kristen Burgess, Naomi Miller, Stephanie Dennis, Kin Wah Fung

1. 
   I began by describing the current setup of MedlinePlus Connect: two Excel files – one Excel file with 12,160 ICD-9-CM codes associated with up to 3 MedlinePlus Health Topics (semi-automated process) and one Excel file with 5478 SNOMED CT CORE codes associated with up to three MedlinePlus Health Topics.
   

2. 
   I explained the current challenge and purpose of this Spring Associate Fellowship Program (AFP) project: how to support ICD-10-CM codes (approximately 69,000 codes) beginning in 2013. Our goal is to have a file with associations between ICD-10-CM codes and up to three Health Topics. We need to support current Health Topics associated with ICD-9-CM at a minimum but also want to expand the list with the transition to ICD-10-CM.
   

3. 
   I asked Kin Wah for his thoughts on the project.
   
   1. 
      Kin Wah confirmed that the initial automated mappings used the UMLS and potentially MeSH. Naomi explained that the initial list of ICD-9-CM codes came from the Institute of Family Health and were chosen because they were the most commonly used diagnosis codes. She noted that the initial mappings from Lister Hill were either great or way off.
      
   2. 
      In addition, it was noted that updates occur to the ICD files annually. SNOMED-CT CORE Problem List updates and mappings are also done manually. Kin Wah noted that he could run the SNOMED codes through an algorithm to expand that list.
      
   3. 
      Kin Wah discussed the GEMs and suggested that we use them. If we use them, he suggested beginning with forward mapping (ICD-9-CM – ICD-10-CM) and then backward mapping (ICD-10-CM – ICD-9-CM) to fill in any areas that were not covered initially. He noted that he has read that complete mapping between the two requires two passes and it is necessary to go both forward and backward.
      
   4. 
      Kin Wah noted that the GEMs are considered to be of high quality because CMS developed them.
      

4. 
   We asked who at the NLM was working on or with the GEMS or who would know about them. He did not think anyone was using them at the NLM. He mentioned that Vivian Auld might have an interest in the GEMs maps.
   

5. 
   We asked about the UMLS and whether or not we could use it to map between the two classifications (ICD-9-CM and ICD-10-CM). Kin Wah said that the UMLS could be used because it contains all of the maps and data necessary. The GEMs are also all in the UMLS. The 2010 version of ICD-10-CM is available in the UMLS but not the 2011 version at the moment. The UMLS is a source that can be used to access ICD-10-CM content but Kin Wah noted that the UMLS is difficult for non-programmers to use and the GEMs might be easier when used outside of the UMLS.
   

6. 
   At this point, Kin Wah suggested that we try two methods and compare them to determine the best way to do the mapping from ICD-10-CM  Health Topics:
   

1. 
   Method 1: Use the UMLS (without using the GEMs) to map between ICD-10-CM and the M+ Health Topics. He will use the original algorithm to match ICD-10-CM codes to M+ Health Topics. We can determine the percentage of these matches that are useful for M+ Connect and the percentage that will require manual editing and changes.
   
   1. 
      Note: This process requires updated M+ Health Topics within the UMLS. Naomi confirmed with Chris Hui on 4/27/11 that these were updated in the last update of the UMLS Metathesaurus. He wrote that “the January 8, 2011 xml files were used for 2011AA”. 3 M+ Health Topics have been released since January 8 but these could be filled in manually.
      
   2. 
      The next UMLS update is May 2011. People who could help if updating is necessary include: Chris Hui, Betsy Humphreys, and Stuart Nelson
      
2. 
   Method 2: Use the GEMS for forward/backward mapping to determine associations between ICD-10-CM and M+ Health Topics.
   
   1. 
      Naomi, Stephanie, and Kristen discussed the potential of pulling the GEMs files into Access, or a similar program, to facilitate this process.
      
   2. 
      I asked about connecting the ICD-10-CM descriptions, the forward mapping file with ICD-9-CM ICD-10-CM flags, and the M+ Health Topics. Kin Wah mentioned that ICD has no machine readable file but that it does have an XML file with the descriptions. These can be extracted from the NCHS site (they have PDF and XML files). We can populate excel with these.
      

7. 
   Stephanie asked about whether one of the online translator tools could be used for the M+ Connect project. Kin Wah said that the GEMs are the standard and as far as he knows there is no other major source. The main issue is that the GEMS will only be maintained for 3 years after 2013. A lot of commercial work is going into mappings and implementation of ICD-10-CM.
   

8. 
   A comment was made about the importance of the project. The project has the potential to link the library’s focus on patient education with its focus on standards and the support of standards.
   

9. 
   The discussion then turned to SNOMED. We explained that the focus has been on ICD-9-CM for pragmatic reasons but that M+ Connect hopes to ramp up the work on SNOMED.
   
   1. 
      We asked whether we could use all of SNOMED or just the CORE subset. Kin Wah believed that we were licensed to use all of SNOMED and did not seem to see it as an issue because the user would not be able to see the entire mapping file. If a provider has the permission to send the code and they send a code to M+Connect, they are only provided with the M+ page of Health Topics and not the whole mapping.
      
   2. 
      Kin Wah noted that we would probably be interested in 3 out of the 19 top-level hierarchies (each contains sub-hierarchies): Clinical Finding, Event, Situation
      
   3. 
      Stephanie questioned whether we should write a memo to IHTSDO to determine what additional permissions we might need.
      
   4. 
      Kin Wah discussed the reasons behind the decision to map from SNOMED-CT CORE Problem List subset to ICD-10-CM. As changes occur within health IT and EHRs, the hope is that systems will use SNOMED as the main coding vocabulary and that by providing mappings to ICD-10-CM, it will also provide an easy mechanism for reimbursement through ICDs. The mapping from SNOMED to ICD10 is tentatively scheduled to be ready by the end of 2011. The NLM is creating an official, vetted map. This will help to address arguments against SNOMED since it will create an authoritative method of getting reimbursed through the ICD codes.
      
   5. 
      Kin Wah asked whether M+ system is scalable and can handle an increase in SNOMED codes. Stephanie and Naomi said yes and described the current input system. They noted that they hope to associate M+ Health Topics with ICD-9-CM, ICD-10-CM, and SNOMED codes. The hope is to have a system that will show suggestions or ideas for each health topic (potentially pulling from UMLS?). [Updates were not discussed]
      
   6. 
      Kin Wah suggested that we use our current health topic associations to SNOMED and include SNOMED descendents. Basically, we will use the tree structure to increase the number of SNOMED associations available to users. Kin Wah agreed to help determine the descendents of the SNOMED terms already associated with M+ Health Topics and to send a file with the suggested descendents and their descriptions.
      

10. 
    To Do Items Discussed
    
    1. 
       Kin Wah: Run the algorithm previously used for ICD9 to associate ICD10 codes with M+ topics
       
    2. 
       US: Make sure UMLS has the most recent version of M+ Health Topics.
       
       1. 
          [we know it was updated as of January. Should we push to have it update again in May or is it too late?]
          
    3. 
       Explore the expansion of current SNOMED mappings to show the descendents and increase the number of associations. Analyze these to determine appropriateness for M+Connect.
       

Vivian Auld Meeting Notes

Date: 6/9/11

1. 
   I began the meeting by describing the current setup of MedlinePlus Connect and the purpose of this project. I explained that the current priority is to prepare for the 2013 change to ICD-10-CM but that we are also looking into how to expand the current SNOMED CT mappings. I also described Kin Wah Fung’s provision of SNOMED-CT CORE Problem List Subset (or at least the CORE concepts mapped to M+ Health Topics) descendents.
   

2. 
   Vivian ensured that I use “ICD-10-CM” or “ICD-9-CM” instead of just “ICD-10” or “ICD-9” when discussing the classifications. While much of the US literature about the CMS change to ICD-10-CM simply refers to the classifications as “ICD-10” or “ICD-9”, the correct terminology includes the reference to the clinical modification (CM). It is important to distinguish the US version and the international version.
   

3. 
   I asked if M+ Connect could use SNOMED CT concepts outside of the SNOMED CT CORE Problem List subset or if there were any legal or licensing restrictions we should know about. Vivian said that no license or additional permissions are needed – with the assumption that anyone requesting a health topic using a SNOMED CT concept code would be going through an EHR. The issue is not as clear if the service is available outside of an EHR and if a user can possibly extract chunks of SNOMED. Vivian said that there were ways to create solutions if this was the case (ex: access provided but users must have a UMLS login, etc.)
   

4. 
   In terms of suggestions for additional subsets or sections of SNOMED CT recommended for future use within MedlinePlus Connect, Vivian suggested Kaiser Permanente’s (KP) Convergent Medical Terminology (CMT) Problem List Subset (http://www.nlm.nih.gov/research/umls/Snomed/cmt.html). The IHTSDO is currently deciding which of the codes they will use for the international release. Once this is determined, additional codes will be chosen for a US extension. Codes not used will be returned to KP. This process is currently ongoing. Vivian mentioned that Kin Wah could tell us which pieces of the CMT were included and were not included in the CORE Problem List and that M+ Connect could begin by looking at the CMT problem list concepts that are currently mapped to concepts within his CORE Problem List.
   

The Introduction section about CMT is copied below:

Convergent Medical Terminology (CMT) is a set of clinician- and patient friendly terminology, linked to US and international interoperability standards, and related vocabulary development tools and utilities.  Developed by Kaiser Permanente over many years for use within its health-IT systems, CMT now includes more than 75,000 concepts.

In September 2010 Kaiser Permanente, the International Health Terminology Standards Development Organization (IHTSDO) and the US Department of Health and Human Services jointly announced Kaiser Permante’s donation of their CMT content and related tooling to the IHTSDO. The donation consists of terminology content already developed, a set of tools to help create and manage terminology and processes to control the quality of terminology that is developed. CMT also includes mappings to classifications and standard vocabularies including SNOMED CT.

Additional information about CMT is available as an FAQ.

Vivian also mentioned that Phase I of Meaningful Use (M/U) requires the use of a problem list which can be accomplished using an existing system (ICD-9-CM or SNOMED CT) or one developed internally. Vivian noted that the CORE Problem List Subset of SNOMED CT is the section of SNOMED that can be used for Phase I. Vivian also mentioned that much of the nursing subset is already found in the SNOMED CORE subset. She said that Kin Wah would be a good person to contact regarding CMT and SNOMED CT.

My Notes: We should look up what is required for Phase II of M/U from SNOMED CT. If the Problem List is currently the only required section, I think that further expanding this section is the best next step. If additional sections that are relevant to M+ Connect are used in later phases of M/U, then it is worth expanding by mapping to those sections.

5. 
   Additionally, Vivian mentioned that she heard a lot about GEMs during her time at HIMMS. Of particular interest to me was her comment that CMS appeared to be saying that the GEMs are provided, people can use them, but there is no guarantee from CMS that they are correct. I asked if anyone else was working on a similar project, and she said that there were companies who were using the GEMs and their own algorithms but nothing widely / freely distributed. She discussed one group called IMO (Intelligent Medical Objects) that does a lot of work on vocabulary and interoperability within EHRS (http://www.e-imo.com/). IMO currently provides terminology tables and cross-maps that are reviewed by AHIMA.
   

Vivian commented on the enormous increase in ICD-10-CM terms. She noted that she expects additional analysis of the codes to be conducted once ICD-10-CM is implemented to determine which codes are used. Because of limited resources and time, she recommended that people focus their energy on problem lists.

6. 
   Lastly, Vivian and I discussed the implementation dates of ICD-10-CM as well as meaningful use. The last release of ICD-9-CM will occur this October. In addition, a release of ICD-10-CM will occur (normal maintenance with additions/updates/etc) in October. ICD-10-CM will then be frozen until after its 2013 implementation (this is the expectation). The next release will occur in October 2014. Vivian expects that the upcoming release of ICD-10-CM will be quite large since additional changes will not be made except for in the case of emergency for 3 years. This is great information to know since M+ Connect will be able to use the data released this October to plan for 2013 and additional updates and changes will not occur until after everyone has made the 2013 change.
   

Olivier Bodenreider and Lee Peters Second Meeting Notes

July 21, 2011

1. 
   For the first part of the meeting, Olivier and Lee further explained their algorithm to me since I had a number of questions about how exactly each of the methods worked and what they did in the latest file.
   

2. 
   For latest file, they took out relations from some “mapped from” sources. What this means is that for mapping method (A), the algorithm has better precision. When a source asserts close mapping relationships to ICD-10-CM, they are directional and for a purpose. There are two directions – mapped_to and mapped_from. The Restrict to MeSH algorithm does not consider the direction. This creates better recall but worse precision. By removing some “mapped_from” sources, the algorithm has better precision but worse recall. Other semantic relations can be added to the algorithm in addition to mapped_to and mapped_from if there are others that we think would be helpful.
   

3. 
   The G/x method includes all ancestors (Broader Terms). They build a graph of the ancestors and restrict the concepts to those with the same semantics as the source (ex: diseases get diseases). They only pick ancestors that would have passed the methods used in (I) or (A). They also look at the distance in order to only get terms with the closest relationship. Ancestors, Siblings, and Children are included and identified within the mapping (G/a = ancestors, G/s = siblings, G/c=children).
   
4. 
   They only use method (O), if they found nothing from the ancestors. They look at concepts outside of concepts that only share the same semantics. For example, this method would look at “lung cancer” and consider concepts that do not share the same semantic type as the source, such as “finding site_lung”, etc.
   

5. 
   We also spent quite a bit of time discussing the issue with the “see also” terms used in the Health Topics file. Olivier discussed several options for us to choose:
   
   1. 
      [already done] When there is a preferred term (PT), assert it as mapping to the Health Topic.
      
   2. 
      [already done, but not ideal] When there is no PT, list the “see also” terms as potential mappings that need review.
      
   3. 
      [new option] When there is 1 PT available (even if “see also” terms are available) map to just the PT and ignore the “see also” terms.
      
   4. 
      [new option]When there is a PT available, list it and all PTs used for a “see also” term.
      
   5. 
      [new option]When no PT is available, list the “see also” terms as potentials but indicate that they are “see also” terms.
      
   6. 
      [new option]When no PT is available, list the PT’s that the “see also” terms point to in the XML files.
      
      1. 
         Example: for the “see also” term Accident, the Health Topics for Fall, Injury, and First Aid would all be results because there is only 1 target code for accident, not a target code for each “type” of accident.
         

Note: The resulting files, except for the option to only review PTs and not all “see also” terms will create a large file with quite a bit of review. One good thing is that the output will distinguish between whether the mappings are PTs or “see also” terms.

We need to decide if we want separate files of these options or one file with all of them together.

My initial reaction was to receive a file with the following:

• 
  The next file should only include PTs. If a ‘see also’ term and not a PT is available, indicate that it was initially found as a ‘see term’. If both are found, indicate this if possible.
  
• 
  This will provide everything resolved to its PT but indicate if it was initially found as a PT or ‘see also term’.
  

Last, we discussed how much they thought the algorithm could actually be improved. Olivier said some bad mappings went away when they restricted the “mapped_from” direction. He mentioned that separating PT from “see also” terms will help in the review as well. While some sources can be removed from the mapping and algorithm, the other way to improve the bad mappings is by creating additional rules for the output. These are not a part of the algorithm, but are applied to the results. For example, we can create a rule that says we never want the health topic “rare diseases” applied to any ICD-10-CM code unless it comes from a particular chapter of ICD-10-CM.

Lee Peters created a graphical user interface that can be used to understand how particular mappings occurred. This can help the person creating rules for output. In addition, we can create a blacklist of topics if there are ICD-10-CM topics we never want to use or Health Topics we never want to use.