This paper describes the findings of an exploratory study on reverse mapping of ICD-10-CA, the Canadian Adaptation, to SNOMED CT. For this study a set of 5,000 most frequent ICD-10-CA codes from the health ministry of a Canadian province was used. The methods included applying six mapping algorithms to each ICD-10-CA description to find the matching SNOMED CT concepts, and comparing the output against the UK SCT-ICD10 cross map for accuracy. Overall, we found successful SNOMED CT matches for ~63% of the ICD-10-CA codes. Issues requiring further attention include ways to increase successful matches and independent validation of mapping output. This study provides a glimpse of the methods that could lead to a SNOMED CT to ICD10-CA cross map. It should be of interest to those responsible for secondary use of discharge abstracts in epidemiological and statistical reporting.
The Systematized Nomenclature of Medicine Clinical
Terms (SNOMED CT) is a terminology system used
to capture information relating to a patient’s
condition and care in a consistent manner. Currently,
there are ~376000 concepts in SNOMED CT,
organized into 19 hierarchies such as clinical finding,
observations, body structure and social context.
There are another ~1 million commonly used terms
to describe these concepts, and ~1.4 million semantic
relationships to define the logical connections
between concepts [
While SNOMED CT is the terminology of choice for
capturing details of a clinical encounter, it is
considered too fine grained for non-clinical purposes
such as the reporting of resource use and billing.
Many have advocated the need to link SNOMED CT
to established classification systems, such as the
International Statistical Classification of Diseases
and Related Health Problems Version 10 (ICD-10),
that are already used extensively in statistical
reporting [
In this paper, we describe the initial findings of an exploratory study to create a reverse map from ICD10-CA to SNOMED CT. It originated as part of a Master of Science project by the lead author. We contend that reverse mapping could be one way to produce the SNOMED CT to ICD-10-CA cross map. This paper describes the mapping algorithms and process used, the key results on matches found, and the lessons and implications from the study.
The ICD-10-CA is an enhanced version of the
ICD10 published by the World Health Organization
(WHO). The ICD-10-CA has 23 chapters and is used
for classifying morbidity, diseases, injuries and
causes of death in Canada. It also covers non-disease
situations and conditions that pose a risk to health
including occupational and environmental factors,
lifestyle and psycho-social circumstances. The
ICD10-CA has an alphanumeric coding format of 3-6
characters. The major difference between ICD-10
and ICD-10-CA is that the latter has two additional
chapters: XXII on morphology of neoplasms and
XXIII on provisional codes for research and
temporary assignment. There are also minor changes
in some chapters in the form of addition, subdivision,
deletion and revision of selected ICD codes [
For this study, we obtained a set of 5,000 most frequently reported ICD-10-CA codes and their long descriptions for the fiscal year of 2005/06 from the health ministry of a Canadian province. These source mapping terms were from inpatient separations in acute care settings including designated sub-acute care facilities for patients that require more care and time before returning home. The profile of the discharge abstracts for the 5,000 ICD-10-CA codes selected for the study is in Table 1.
Description Total separations 2005/06 in province Total diagnosis codes reported Average no. of codes reported per separation Total discrete diagnosis codes (all) Frequency of top 5,000 diagnosis codes % of total diagnosis in top 5000 codes % of total discrete diagnosis in top 5000 codes Total discrete most responsible diagnosis codes Table 1. Profile of the Discharge Abstracts Count 364,977 1,481,285
4.1 10,529 1,460,730 98.6% 47.5% 6,651
After conducting a detailed review of the literature on cross mapping of terminology systems, we adopted five related mapping algorithms and created Web-based versions of these algorithms in to find matching SNOMED concepts for each of the ICD10-CA descriptions in the data set [5]. Four of the algorithms are lexical techniques for exact-match, match-all-words-only, match-all-words and partialmatch. The fifth is semantic matching that involves retrieving the current concepts based on entries in the SNOMED historical relationship table if the initial concepts found are inactive. These mapping algorithms are summarized in Table 2.
Algorithm Explanation 1. Exact match Exact string match where all words are same and in same sequence for both source and target terms, including punctuation 2. Match all only String match where all words are same but not necessary in same order; additional words not allowed in target term 3. Match all String match where all words are same but not necessary in same order; additional words allowed in target term 4. Partial match String match where one or more words in source term is found in target term 5. Semantic match For inactive concepts found use historical relationships of Was-A Same-As, May-Be
A, Replaced-By to find current concepts 6. Unmappable Assigned when no match is found
Table 2. Mapping algorithms used in this study
In addition to using the original SNOMED CT terms and the ICD-10-CA long descriptions in mapping, we normalized all of these original terms to remove “noise” such as genitives and spelling errors using the Unified Medical Language System (UMLS) normalization steps, as shown in Table 3a [6]. To improve successful mapping, we expanded step-2 to remove both “stop words” and “exclude words,” as well as SNOMED prefixes, shown in Table 3b. For step-5 we included both the lookup and stemming methods to uninflect the phrase. The lookup method uses the UMLS SPECIALIST Lexicon’s inflection table with ~1 million entries, whereas the stemming method uses the computational technique first published by Porter Stemming that reduces word variants to a single canonical form [7,8].
Steps 1 to 6 Example Remove genitive Hodgkin’s disease, NOS o Hodgkin diseases, NOS Remove stop words Hodgkin diseases, NOS o Hodgkin diseases, Convert to lowercase Hodgkin diseases, o hodgkin diseases, Strip punctuation hodgkin diseases, o hodgkin diseases Uninflect phrase hodgkin diseases o hodgkin disease Sort words hodgkin disease o disease hodgkin Table 3a. UMLS six normalization steps[7, slide 20] SNOMED Prefixes Step-2 Stop words Exclude words
Explanation Frequent short words that do not affect the phrase: and, by, for, in, of, on, the, to, with, no, and (nos) Words that may change meaning of the word but if ignored help to locate a term otherwise missed: about, alongside, an, anything, around, as, at, because, before, being, both, cannot, chronically, consists, covered, does, during, every, find, from, instead, into, more, must, no, not, only, or, properly, side, sided, some, something, specific, than, that, things, this, throughout, up, using, usually, when, while, without [X] – concepts with ICD-10 codes not in ICD-9 [D] – concepts in ICD-9 XVI and ICD-10 SVII [M] – morphology of neoplasm concepts in ICD-O [SO] – concepts in OPCS-4 chapter Z in CTV3 [Q] – temporary qualifying terms from CTV3 [V] – concepts in ICD-9 and ICD-10 on factors influencing health status and contact with health services (V-codes and Z-codes)
Table 3b. Expanded UMLS normalization step-2
The reverse mapping of ICD-10-CA terms to SNOMED CT concepts involved cycling through the mapping algorithms one at a time to find the best candidate SNOMED CT concepts as the target terms. For each algorithm we always started with the original terms, then the UMLS normalized terms, followed by the stemmed terms. In each cycle, we would review the candidate concepts found to see if it was a match, and if so, what type of match it was based on the algorithm applied. When no matching concepts were found, we would label the term as unmappable. Our experience with the matching techniques was that, the sooner we could find a match in the cycle, i.e. first-match, the greater confidence we would have that the candidate concept is appropriate. The preferred order of matched terms was always exact-match first, match-all-only, then match-all, with partial-match last. Whenever inactive concepts were found a semantic-match was done to find the current concepts through their historical relationships. During mapping we tallied frequency statistics on the different types of matches with summary/detailed outputs. Only the first-matches were counted to determine the effectiveness of each mapping algorithm.
To determine the accuracy of the mapping results
from this study, we compared our output with the UK
SNOMED CT to ICD-10 (SCT-ICD10) cross map.
To do so, the 5,000 ICD-10-CA codes were matched
with the TargetCodes of the SCT_CrossMapTargets
table from the July 2007 version of the IHTSDO
distribution set [
Of the 5,000 ICD-10-CA descriptions used in this study, we were able to match 1,619 source ICD terms (32.38%) to 2,625 target SNOMED concepts by the exact-match technique. Next, we matched 63 ICD terms (1.26%) to 87 SNOMED concepts by matchall-only; another 1,478 ICD terms to 4,829 concepts by match-all; and 1,839 ICD terms to ~25 million concepts by partial-match. One ICD term C8800 Waldenstr was umappable. A summary of the mapping output by match-type is shown in Table 4. Match Type Exact match Match all only Match all Partial match Unmappable Total
Each ICD term was cycled through all the matching techniques to determine the number of candidate target SNOMED concepts found for each match type. The first-match reported for each match type excluded the target concepts already identified in previous iterations to avoid duplicate counting. We tracked not only the total matches but also which technique found the first match. The output produced suggested exact-match, match-all-only and match-all could be considered as successful matches, since they returned one or more identical or similar SNOMED concepts based on the ICD term provided. The number of first-matches found for these match types by ICD Chapter are shown in the Appendix. One can see that the percentages of matches were very low for Chapters IV Endocrine, nutritional and metabolic diseases at 36%; XIII Diseases of the musculoskeletal system and connective tissue at ~36%; and XV Pregnancy, childbirth and the puerperium at ~4%. Of the overall 3,160 ICD terms or ~63% that were mapped to one or more SNOMED concepts, most were found by exact-match and match-all during the first-match. The profiles of first-matches found by each match type are briefly described below. Exact Match – Table 5 shows 1,237 original ICD terms had exact-matches with 2,064 candidate concepts. Another 364 ICD terms had exact-matches with 527 concepts using the UMLS normalized version, and 18 ICD with 34 concepts using the stemmed version. In all, 2,625 candidate SNOMED concepts were found, which means that there were multiple exact matches for some of the ICD terms.
Match All Words First Match Original Term 1,343 UMLS Version 114 Stemmed Version 21
Total 1,478
Table 7. Match all words output Exact Match First Match Original Term 1,237 UMLS Version 364 Stemmed Version 18
Total 1,619
Table 5. Exact match output Match All Only – Table 6 shows 33 original ICD terms had match-all-only with 48 candidate concepts; 29 UMLS normalized terms had 37 concepts, and 1 stemmed term had 2 only. In all, 87 candidate SNOMED concepts were found, which means that there were multiple match-all-only for some terms. Match All Words Only First Match Original Term 33 UMLS Version 29 Stemmed Version 1
Total 63
Table 6. Match all only output Match All Words – Table 7 shows 1,343 original ICD terms had match-all with 4,558 candidate concepts; 114 UMLS normalized terms had 217 concepts, and 21 stemmed terms had 54. In all, 4,829 SNOMED concepts were found, which means that there were multiple match-all for some terms.
Partial Match – Table 8 shows 1,839 ICD terms had partial-matches with 25 million SNOMED concepts. We found the results of partial matches to be more unpredictable than the previous match types. If a source term was long and contains common words such as disorder or procedure, the results returned could be numerous as only one word from the source term needed to be present in the target term.
Six comparisons were made between our mapping output and the UK map to see if: (a) both contained the same results; (b) both contained similar results; (c) both contained dissimilar results; (d) only UK map contained the results; (e) only our mapping output contained the results; (f) both had unmappable results. The overall results are shown in Table 9. Only (b), (c) and (f) are illustrated in this paper. Similar Results - Where both maps contained similar results, the UK map usually had more mapped terms than our output, as shown in Table 10. An example is with the ICD term Q61.2 Polycystic kidney, autosomal dominant where the UK map had six SNOMED concepts but only four in ours. Description UK map had more results than mapping outputs Mapping outputs had more results than UK map UK and mapping outputs had same no. of results
Total ConceptId Fully Specified Name UK 66091009 Congenital disease (disorder) ¥ 204955006 Polycystic kidney disease ¥ 204962002 Multicystic kidney (disorder) ¥ 28728008 Polycystic kidney disease, adult ¥ ¥ type (disorder) 253878003 Adult type polycystic kidney ¥ ¥ disease type I (disorder) 253879006 Adult type polycystic kidney ¥ ¥ disease type II (disorder) 274567009 [EDTA] Polycystic kidneys, adult type (dominant) associated with renal failure (disorder) Table 10. Comparing both with similar results Total 2,125 224 63 2,401 CA ¥
Dissimilar Results – Where both had dissimilar results, our output were more specific as each concept must contain all the words in the source term. For 100 (82%) of these terms the UK map had more candidate concepts; for 9 terms (7.4%) both had same number of concepts; whereas for 13 (10.7%) our mapping output had more concepts. An example is the ICD term S597 Multiple injuries of forearm, shown in Table 11, where both maps had four concepts but none are similar.
ConceptId 122549002
Fully Specified Name Injury (disorder) UK ¥
CA 210860005 211290004 212308001 212464002 Unmappable Results – These were in almost every ICD chapter but most notable in XVII: Congenital malformations, deformations and chromosomal abnormalities; XIX: Injury, poisoning and certain other consequences of external causes; and XIII: Diseases of the musculoskeletal system and connective issue (Table 12). It is possible these ICD terms have further refinement making it difficult to find concept and lexical matches. An example is the ICD-10-CA term O2450 Pre-existing Type 1 diabetes mellitus arising in pregnancy, which could be refined as: delivered with or without antepartum condition (1), delivered with postpartum complication (2), or antepartum condition or complication (3).
This study was our initial effort to apply a set of mapping algorithms on a set of ICD-10-CA terms to find the matching target SNOMED concepts. Our output showed most of the matches were found using the exact-match and match-all algorithms. The match-all-words-only algorithm did not add a great deal to the number of matches found, and the partialmatch was considered too unpredictable with respect to the candidate target concepts returned. Due to space limitation, we did not report on additional matches found after normalization with UMLS and stemming techniques were applied to the original ICD terms, or those found by semantic matching. A major issue is how one should define “successful match.” In our output we had just over 60% of the matches found by exact-match and match-all, which we reviewed and deemed correct. However, more formal validation preferably by an independent source is needed. While our results showed successful matches in only ~63% of the 5,000 ICD10-CA codes, we were surprised to find the UK cross map had similar successful matches of ~68% against the same 5,000 ICD-10-CA codes (see Table 9). Equally intriguing were the different matches found between the two maps. Almost 50% of the concepts found were similar but not identical, whereas ~20% were dissimilar or found only in the UK map. One possible explanation is the minor differences that exist between ICD-10 and ICD-10-CA with respect to the addition, subdivision, deletion and revision made in some ICD-10-CA chapters. Another is that a concept-based method was used to create the UK cross map, which seemed to outperform the lexical techniques in this study. One possible solution to improve mapping precision is to combine methods, such as the use of semantic and lexical mapping between SNOMED CT and ICD-9-CM by Fung.9 Another issue is the extent that our semi-automated matching algorithms can aide in the cross-mapping process by health records staff when encoding the inpatient discharge abstracts. The current abstracting process is mostly an intellectual and manual exercise. As such, explicit cross-mapping guidelines need to be established, including the use of any computerbased mapping tools, to improve this abstracting process. With our mapping algorithms, a consensusbased process is needed for the health record staff to verify the accuracy of the ~63% successful matches. Guidelines are also needed to reconcile the remaining ~37% partially-matched terms.2,10
Still, we contend there is merit in exploring the use of reverse mapping with lexical algorithms to identify candidate SNOMED concepts for a given set of ICD10-CA terms. Our next steps are to enhance the mapping algorithms to include contexts, incorporate these algorithms into the abstracting process, and conduct further field evaluation. Last, the idea of applying reverse mapping to identify candidate SNOMED CT concepts for a set of mapping terms can be a helpful approach when creating a cross map from SNOMED CT to another terminology system.
This study provides a glimpse of the feasible mapping methods that could eventually lead to a SNOMED CT to ICD-10-CA cross map for Canada. We believe the intent, methods and results of this current study should be of interest to those responsible for secondary use of patient discharge abstracts in epidemiological and statistical reporting. The notion of reverse mapping is also highly generalizable to include the encoding of local terms that already exist in legacy systems within many health organizations to a reference terminology such as SNOMED CT.
We wish to thank the Provincial Ministry that provided the 5,000 ICD-10-CA codes for the study.
We also thank Ms. Robyn Kuropatwa in facilitating the process to obtain the ICD codes from the ministry. Funding support for this work was provided by the Canadian Institutes for Health Research through its Strategic Training Initiative. Note that the views presented in this paper are those of the authors only and do not represent the official position of any Canadian government agencies.
Appendix. Mapping Output for top 5,000 ICD-10-CA codes by ICD Chapter Chapter
Title Source Exact
Only III IV V VI IX X XI XII XIII XIV XV XVI XVII XVIII XIX XX XXI XXII
Lee DHK. Reverse Mapping ICD-10-CA to SNOMED CT. UVic Master of Science research project report, Oct 2007. Unpublished.
National Library of Medicine. The SPECIALIST Lexicon. http://lexsr3.nlm.nih.gov/LexSysGroup/Projects/ Summary/lexicon.html Kleinsorge R, Willis J, et al. UMLS Overview – Tutorial T12. AMIA Annual Symposium 2006. http://165.112.6.70/research/umls/pdf/AMIA_T1 2_2006_UMLS.pdf. Jan15/2006.
Goldsmith JA, Higgins D, Soglasnova S.
Automatic Language-specific Stemming in Information Retrieval. Springer-Verlag Berlin Heidelberg 2001.
Certain infections and parasitic disease Neoplasms Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism Endocrine, nutritional and metabolic diseases Mental and behavioural disorders Diseases of the nervous system Diseases of the eye and adnexa Diseases of the ear and mastoid process Diseases of the circulatory system Diseases of the respiratory system Diseases of the digestive system Diseases of the skin and subcutaneous tissue Diseases of the musculoskeletal system and connective tissue Diseases of the genitourinary system Pregnancy, childbirth and the puerperium Certain conditions originating in the perinatal period Congenital malformations, deformations, chromosomal abnormalities Symptoms, signs and abnormal clinical and laboratory findings not elsewhere classified Injury, poisoning and certain other consequences of external causes External causes of morbidity and mortality Factors influencing health status and contact with health services Morphology of neoplasms A00-B99 D50-D89 M00-M99 8000/09989/1 136 343 80 225 218 196 89 42 279 165 276 105 383 226 313 169 205 181 691 297 333 28 20 47 174 35 56 66 75 56 24 136 67 136 120 42 78 5 57 105 99 175 9 29 28 2 1 1 3 1 3 1 4 9 1 3 1 2 2 8 4 17 141 57 58 20 24 56 18 11 74 41 56 20 61
Provisional codes for research and temporary assignment
Total