Coding Algorithms for Defining Comorbidities in ICD-9-CM ...

1 ORIGINALARTICLEC oding Algorithms for Defining Comorbidities inICD-9-CM and ICD-10 Administrative DataHude Quan, MD, PhD,* Vijaya Sundararajan, MD, MPH, FACP, Patricia Halfon, MD, Andrew Fong, BCOMM,* Bernard Burnand, MD, MPH, Jean-Christophe Luthi, MD, PhD, L. Duncan Saunders, MBBCh, PhD, Cynthia A. Beck, MD, MASc,* Thomas E. Feasby, MD,**and William A. Ghali, MD, MPH,* , Objectives:Implementation of the International Statistical Classi-fication of Disease and Related Health Problems, 10th Revision(ICD-10) Coding system presents challenges for using administrativedata. Recognizing this, we conducted a multistep process to developICD-10 Coding Algorithms to define Charlson and Elixhauser co-morbidities in administrative data and assess the performance of theresulting :ICD-10 Coding Algorithms were developed by transla-tion of the ICD-9-CM codes constituting Deyo s (for Charlsoncomorbidities) and Elixhauser s Coding Algorithms and by physi-cians assessment of the face-validity of selected ICD-10 codes.

2 Theprocess of carefully developing ICD-10 Algorithms also producedmodified and enhanced ICD-9-CM Coding Algorithms for the Charl-son and Elixhauser Comorbidities . We then used data on in-patientsaged 18 years and older in ICD-9-CM and ICD-10 administrativehospital discharge data from a Canadian health region to assess thecomorbidity frequencies and mortality prediction achieved by theoriginal ICD-9-CM Algorithms , the enhanced ICD-9-CM algo-rithms, and the new ICD-10 Coding :Among 56,585 patients in the ICD-9-CM data and 58,805patients in the ICD-10 data, frequencies of the 17 Charlson comor-bidities and the 30 Elixhauser Comorbidities remained generallysimilar across Algorithms . The new ICD-10 and enhanced ICD-9-CM Coding Algorithms either matched or outperformed the origi-nal Deyo and Elixhauser ICD-9-CM Coding Algorithms in predictingin-hospital mortality. The C-statistic was for Deyo s ICD-9-CM Coding algorithm , for the ICD-10 Coding algorithm , for the enhanced ICD-9-CM Coding algorithm , for theoriginal Elixhauser ICD-9-CM Coding algorithm , for theICD-10 Coding algorithm and for the enhanced ICD-9-CMcoding :These newly developed ICD-10 and ICD-9-CM co-morbidity Coding Algorithms produce similar estimates of comor-bidity prevalence in administrative data, and may outperform exist-ing ICD-9-CM Coding Words:ICD-9, ICD-10, comorbidity, risk adjustmentoutcome, administrative data(Med Care2005;43: 1130 1139)Patient clinical characteristics usually are measured andcontrolled in clinical outcomes research.

3 Similarly, whenadministrative data are used for such research, comorbiditycoding Algorithms are essential for Defining comorbidity measurement tools developed by Charlsonet al1and Elixhauser et al2are used widely to measure burdenof disease or case-mix with administrative data. Charlson etal1defined 17 Comorbidities using clinical conditions re-corded in charts. Deyo et al,3 Romano et al,4and D Hoore etal5,6independently developed International Classification ofDisease, 9th Revision, Clinical Modification ( ICD-9-CM ) Coding Algorithms for the Charlson Comorbidities . Deyo scoding algorithm1and the Dartmouth-Manitoba Coding algo-rithm developed by Romano et al3are similar in generatingCharlson index scores and in their ability to predict 9D Hoore et al5,6only used the first 3 characters ofICD-9-CM codes and did not distinguish subgroups of certainclinical conditions (such as diabetes with or without compli-cations).

4 Elixhauser et al2defined 30 Comorbidities usingdistinctive ICD-9-CM codes as a starting point. The originalElixhauser ICD-9-CM Coding algorithm has been revisedFrom the *Department of Community Health Sciences, University of Cal-gary, Calgary, Alberta, Canada; Centre for Health and Policy Studies,University of Calgary, Calgary, Alberta, Canada; Victorian Departmentof Human Services, Australia; Health Care Evaluation Unit, InstitutUniversitaire de Me decine Sociale et Pre ventive, University of Lausanne,Switzerland; Department of Public Health Sciences, University ofAlberta, Edmonton, Alberta, Canada; Department of Psychiatry, Uni-versity of Calgary, Calgary, Alberta, Canada; **Department of Medicine,University of Alberta, Edmonton, Alberta, Canada; and Department ofMedicine, University of Calgary, Calgary, Alberta, by an operating grant from the Canadian Institutes of HealthResearch, Canada.

5 Dr. Quan is supported by a Population Health Inves-tigator Award from the Alberta Heritage Foundation for Medical Re-search, Edmonton, Alberta, Canada, and by a New Investigator Awardfrom the Canadian Institutes of Health Research. Dr. Ghali is supportedby a Health Scholar Award from the Alberta Heritage Foundation forMedical Research, Edmonton, Alberta, Canada, and by a Government ofCanada Chair in Health Services Research. Dr. Beck is supported by aClinical Fellowship from the Alberta Heritage Foundation for MedicalResearch, Edmonton, Alberta, : Dr. Hude Quan, Department of Community Health Sciences,University of Calgary, 3330 Hospital, Dr. NW, Calgary, Alberta, Canada,T2N 4N1. E-mail: 2005 by Lippincott Williams & WilkinsISSN: 0025-7079/05/4311-1130 Medical Care Volume 43, Number 11, November 20051130twice, and the most recent revision (version ) was postedon the website of the Agency for Healthcare Research andQuality (we refer to this as the Elixhauser AHRQ-WebICD-9-CM Coding algorithm ).

6 10 This revised algorithm con-tains more ICD-9-CM codes than the original ElixhauserICD-9-CM Coding algorithm , but excludes cardiac arrhyth-mias from the list of 1992, the 10th Revision of ICD (ICD-10)11wasintroduced by the World Health Organization as a potentialenhancement to ICD-9-CM . An obvious merit of ICD-10 is thatit contains more codes than ICD-9-CM , allowing for the richercoding of clinical Coding uses a newalphanumeric system and many codes are not directly convert-ible to corresponding ICD-9-CM codes. Therefore, ICD-10coding Algorithms to define Comorbidities must be , Halfon et al13in Switzerland and Sundarara-jan et al14in Australia independently developed ICD-10coding Algorithms to define Charlson Comorbidities . Al-though many of the ICD-10 codes are similar, there arediscrepancies between the 2 Coding Algorithms , which mayrelate to the different approaches used to developing thealgorithms.

7 Halfon et al13used clinical judgment surroundingindividual ICD-9-CM codes following Charlson s clinicaldefinitions for Comorbidities . In contrast, Sundararajan et al14employed a computerized ICD-9-CM and ICD-10 mappingfile to translate the ICD-9-CM codes to ICD-10 codes. AnICD-10 Coding algorithm has not yet been reported forElixhauser objectives of this study were: (1) to developICD-10 Coding Algorithms for Charlson and Elixhauser co-morbidities by a consensual approach among 3 internationalresearch groups in Switzerland, Australia and Canada, (2) toback-translate the newly developed ICD-10 Coding algo-rithms into ICD-9-CM codes to improve the original Deyo(for Charlson Comorbidities ) and Elixhauser ICD-9-CM cod-ing Algorithms , (3) to use administrative data from a Cana-dian health region to demonstrate the degree of consistencybetween ICD-9-CM and ICD-10 Coding Algorithms in defin-ing these Comorbidities , and (4) to assess the performance ofthe Coding Algorithms for predicting in-hospital of ICD-10 Coding Algorithms forCharlson ComorbiditiesStep 1 Three lists of ICD-10 codes were generated for theICD-10 Charlson Comorbidities .

8 List 1 contained all codes inthe ICD-10 Coding Algorithms for Defining Charlson comor-bidities developed by Halfon et al13and Sundararajan et List 2, 2 coders with clinical and Coding experience inICD-9-CM and ICD-10 independently coded the 17 Charlsoncomorbidities using the ICD-10 Canadian version (Interna-tional Statistical Classification of Disease and Related HealthProblems, Tenth Revision, Canada ICD-10-CA ) computer-ized code provided the coders with 1 clinicalterm per comorbidity (such as myocardial infarction). Theywere instructed to find all codes relevant to each clinical 2 coders met to compare their codes to generate a consensuslist of ICD-10 codes. If no consensus could be reached, aphysician was involved in the discussion until consensus wasachieved. For List 3, the 2 coders above recoded the 17 comor-bidities. At this time, clinical terms taken from the ICD-9-CMmanual for all ICD-9-CM codesthat were developed by Deyoet al3for each Charlson comorbidity were provided to thecoders.

9 The 2 coders followed a methodology similar to thatused in generating List 2 described 2 The 3 lists were combined for a comprehensive list,with their specific ICD-10 codes and their clinical descrip-tions as listed in the ICD-10 manual. Four physicians re-viewed the comprehensive Coding list independently and thenmet to discuss whether each condition met the clinical defi-nition and whether there were missing conditions based onCharlson s clinical definition of Comorbidities . For the fewcodes without consensus, 1 additional physician was con-sulted to finalize this Coding of the ICD-10 Coding Algorithmfor Elixhauser ComorbiditiesBecause 6 Comorbidities are common to the Charlsonand Exliahsuer lists (ie, congestive heart failure, peripheralvascular disease, chronic pulmonary disease, hemiplegia/paraplegia, AIDS/HIV, and metastatic solid tumors), allcodes allocated to these 6 Comorbidities in the aforemen-tioned process were retained as is.

10 For the remaining 24 Elixhauser Comorbidities , we proceeded as 1 Three lists of ICD-10 codes were generated for theICD-10 Elixhauser Comorbidities . For List 1, Elixhauser sclinical terms were directly translated into ICD-10. This wasperformed initially by the 2 coders independent of each otherusing the ICD-10-CA computerized code ,they met to create a consensus list. If any disagreementbetween coders occurred, a physician discussed the discrep-ancies with them. For List 2, the same 2 coders recoded the30 Comorbidities . At this time, they coded each clinical termtaken from the ICD-9-CM manual for all ICD-9-CM codes inElixhauser s original Coding algorithm . The same methoddescribed in List 1 was followed to generate List 2. For List3, Elixhauser s original ICD-9-CM codes were converted toICD-10 using a cross-table mapping algorithm from theNational Centre for Classification in Health at the Universityof Sydney in ,17 Step 2 The 3 lists were combined to yield a comprehensivelist, with their specific ICD-10 codes and their clinical de-scriptions.