Validity of computerized diagnoses, procedures, and drugs for inflammatory bowel disease in a...

pharmacoepidemiology and drug safety 2009; 18: 1086–1093.interscience.wiley.com) DOI: 10.1002/pds.1824
Published online 11 August 2009 in Wiley InterScience (www
ORIGINAL REPORT

Validity of computerized diagnoses, procedures, and drugs forinflammatory bowel disease in a northern California managed careorganizationy,z

Liyan Liu MD, MSc1, James E. Allison MD1,2 and Lisa J. Herrinton PhD1*

1Division of Research, Kaiser Permanente Northern California, Oakland, CA, USA2Division of Gastroenterology, Department of Internal Medicine, University of California, San Francisco, CA, USA

SUMMARY

Purpose Resources for studying inflammatory bowel disease (IBD) are needed in evaluations of drug safety including traditional drugs andnew biologics agents. We developed an IBD registry, with ascertainment from computerized visit information.Objective We sought to characterize the positive predictive value (PPV) of IBD case-finding using computerized data compared with chartreview.Methods We identified 2906 persons aged 89 years or younger with one or more IBD diagnoses in computerized visit data during the periodof 1996–2002. The diagnosis of IBD was confirmed through chart review. Adopting chart review as the gold standard, the validity ofcomputerized encounter data to determine IBD was estimated.Results Among the 2906 study subjects with one or more ICD-9 diagnosis codes of 555 or 556 in computerized data, 81% were confirmedas having IBD by chart review. Defining cases as those who underwent two or more visits without regard to diagnostic procedures or drugutilization maximized the correct classification of cases (PPV, 95%).Conclusions The quality of IBD diagnoses in computerized data is adequate to meet the aims of a wide range of research studies. Copyright# 2009 John Wiley & Sons, Ltd.

key words—inflammatory bowel disease; Crohn’s disease; ulcerative colitis; epidemiology; health maintenance organizations;computerized medical information

abbreviations—IBD, inflammatory bowel disease; CD, Crohn’s disease; UC, ulcerative colitis; PPV, positive predictive value; ICD-9,international classification of diseases, 9th revision

Received 27 January 2009; Revised 8 June 2009; Accepted 9 July 2009

INTRODUCTION

Few registries or other population-based resourcesexist to accurately identify cases of inflammatorybowel disease (IBD) in community settings. This hashindered drug safety research in the United States,where population-based case identification is needed.Lack of registries is a key constraint in the developmentof knowledge about the safety of biologic therapy in

*Correspondence to: Dr L. J. Herrinton, Division of Research, KaiserPermanente Northern California, 2000 Broadway Ave., Oakland, CA94612, USA. E-mail: [email protected] conflict of interest was declared.zThis work was originated in the Division of Research, Kaiser PermanenteNorthern California, Oakland, CA, USA.

Copyright # 2009 John Wiley & Sons, Ltd.

IBD, with infliximab, adalimumab, certolizumab, andnatalizumab currently approved for the disease. KaiserPermanente is a pre-paid, comprehensive, integratedcare organization that maintains computerized data-bases of all visits, procedures, pharmacy dispensings,and other medical goods and services provided to itsmembers. These databases provide the opportunity tobuild disease registries for efficient study of chronicdiseases that otherwise cannot be easily identified in alarge population. We recently used the computerizeddata to construct an IBD registry. In this report, weprovide information on the validity of the registry bothto inform those who seek to use computerized medicaldata for research in IBD and for those who use researchfindings from such studies. Measures of validity used in

validity of computerized ibd diagnoses 1087

this report include the positive predictive value (PPV),in this instance comparing the computerized data todiagnoses based on review of the comprehensivemedical record. In addition, we estimated theproportion of cases with one or more diagnosisascertained by the algorithm. This work is part of aseries of investigations into the epidemiology, naturalhistory, and mortality of IBD, as well as disparities inits presentation and course.

METHODS

The methods have been described in detail elsewhere.1

The study was conducted with the approval of theKaiser Foundation Research Institute InstitutionalReview Board.

Study population

We constructed a community-based IBD registry fromthe 3.2 million-person membership of Kaiser Perma-nente. The registry comprised persons with incident orprevalent IBD evaluated by a health care providerbetween 1996 and 2002.

Data sources and collection

Five certified medical record analysts reviewed themedical records to confirm case status. They receivedongoing training and monitoring by the study gastro-enterologist (JA) during regular bi-weekly meetings. Inaddition, the study gastroenterologist reviewed 775chart review forms, mostly during the initial months ofthe data collection effort. Whenever the gastroenter-ologist and the reviewer disagreed on categorizations,the casewas discussed during the studymeeting both toresolve the disagreement and to provide furthertraining. In addition, we performed a quality assurancesub-study in which 43 randomly selected records werereviewed by a second gastroenterologist. We examinedagreement between the two gastroenterologists withregard to actual diagnosis (Crohn’s disease (CD),ulcerative colitis (UC), indeterminate, uncertain, or notIBD) and extent of disease for those with CD and UC.In this sub-study, the two reviewers disagreed on five ofthe 43 cases, with the disagreement being as follows:(1) UC versus CD; (2) unclear versus CD; (3) unclearversus UC; (4) CD with versus without small bowelinvolvement; and (5) CD with versus without perianalinvolvement. For all but the fourth case, discussionresulted in agreement between the two gastroenterol-ogists. In the fourth case, the study gastroenterologist’sopinion was taken as final.To confirm each case, we reviewed all available

hospital discharge summaries and clinic notes, endo-


scopy reports, pathology and radiology reports, andcomputerized laboratory data to confirm or rule-out thediagnosis of IBD and to differentiate CD, UC, andindeterminate IBD. We also supplemented the chartreview with computerized data on enrollment, demo-graphic information, date of IBD-related visits andhospitalizations, IBD-related surgical procedures, andIBD-related medications.Each department has its own outpatient data

collection form, pre-coded with the most commondiagnoses and procedures for that department. IBDcodes were listed only on outpatient encounter formsfor internal medicine, which contained two IBDdiagnoses (CD and UC), and for gastroenterology,which contained four IBD diagnoses: two for CD(ileum and colon) and two for UC (colon and rectum).From coded, computerized pharmacy information, weobtained details on prescriptions and dispensings of thefollowing IBD-related drugs: mesalamine, sulfasala-zine, 6-mercaptopurine, azathioprine, methotrexate,glucocorticoids, and infliximab.

Case definitions

A possible case of IBDwas defined as a person with thefollowing: (1) an international classification ofdiseases, ninth edition (ICD-9) code for CD (ICD-9code 555) or UC (ICD-9 code 556) in the computerizedoutpatient or hospitalization database, and (2)� 12months of enrollment in Kaiser Permanente between1996 and 2002. A total of 12 059 persons wereascertained as possible cases, of which we randomlysampled 24% (n¼ 2906) for comprehensive review ofpaper and computerized information to determinewhether the case was confirmed and the type of colitis.Among the 2906 possible cases, 97.2% were enrolledin Kaiser for 2 years or longer. A confirmed case ofIBD was defined as a possible case of IBD with anendoscopy, pathology, or radiology report indicative ofIBD. Among the 2906 possible cases selected for chartreview, 2325 had at least one inpatient or twooutpatient diagnoses of IBD and 581 had only oneoutpatient diagnosis of IBD. For the 2325 possible IBDcases with at least one inpatient or two outpatientdiagnoses, we reviewed all available endoscopyreports, both on computer and on paper. For the 581possible IBD patients with a single outpatient diagnosisof IBD, we reviewed only computerized data sources,including computerized endoscopy reports. Amongthese 581 possible cases, 288 had an endoscopy duringthe study period. For the 293 without an endoscopyreport, we accepted� 12 dispensings of an IBD-relateddrug (5-aminosalicylate, 6-mercaptopurine, azathiopr-

Pharmacoepidemiology and Drug Safety, 2009; 18: 1086–1093DOI: 10.1002/pds

1088 l. liu ET AL.

ine, or steroid) and no report of a pathogen(Salmonella, Shigella, Campylobacter, and Yersinia)isolated from stool culture as confirmatory.1 CDlocation (colon, ileocolon, terminal ileum, and uppergastrointestinal tract) was categorized using the Viennaclassification.2 UC and indeterminate colitis werecategorized using the most extensive degree ofmacroscopic or microscopic inflammation recordedin the chart as pancolitis, right-sided, left-sided, distal,or proctitis.

Development and validation of the case-findingalgorithm

For a previous validation study,3 we evaluated whetherCD and UC were misclassified as other noninfectiousgastroenteritis and colitis (ICD-9: 558.9), irritablecolon (564.1), functional diarrhea (564.5), fistula of theintestine (569.81), ulceration of the intestine (569.82),or cholangitis (576.1). Using chart review to confirmthe diagnosis, we found very few patients in our healthplan with these codes who actually had IBD.The variables that were examined for inclusion in the

case-finding algorithm included: (1) outpatient visitscoded IBD (ICD-9 code 555 and 556), (2) inpatientvisits coded IBD, (3) visits for IBD specifically to agastroenterologist, (4) use of IBD-related drugs, and(5) endoscopy services. We evaluated 11 possible case-finding algorithms categorized into seven broad groupsbased on these utilization patterns. The algorithm ‘1þinpatient or outpatient visits to primary care orgastroenterology coded 555 or 556’ was used as thebasis for comparison with all other algorithms.For each of the 11 possible case-finding algorithms,

we determined the PPV and the proportion of cases

Table 1. Counts and proportions of cases ascertained through computerized data

Diagnosis in computerized data Diagnosis confirmed du

IBD

CD UC I

At least one diagnosisCD 785 (74%)� 27 (3%)UC 62 (4%) 1228 (74%)Both CD and UC 101 (54%) 62 (33%)

Total 948 (33%) 1317 (45%)At least two diagnosesCD 701 (88%) 21 (3%)UC 51 (4%) 1007 (87%)

Both CD and UC 101 (54%) 62 (33%)Total 853 (40%) 1090 (51%)

Kaiser Permanente IBD Registry, Northern California, 1996–2002. CD¼Crohn’�Row %.


with one or more diagnosis ascertained by thealgorithm. For most analyses, the PPV was definedas the proportion of possible IBD cases captured by thealgorithm that were confirmed with IBD during chartreview. However, for one set of analyses, we restrictedthe definition to CD or UC. The 95% confidenceinterval (CI) was determined by approximating thebinomial distribution with a normal distribution.4

Persons presenting with CD may be initiallymisdiagnosed with UC, and UC may evolve intoCD. In addition, there may be cross-classification ofCD with UC because of coding errors. To betterunderstand the meaning of combinations of ICD-9codes 555 and 556 assigned to the same patient, weconducted an analysis restricted to this subgroup. Forthis analysis, we compared the most common code(between 555 and 556) and the most recent codeagainst the diagnosis determined during chart review.

RESULTS

Overall PPV

Among the 2906 study subjects with at least oneinpatient or outpatient visit given ICD-9 diagnosiscodes of 555 or 556 in computerized data, 81% wereconfirmed with IBD, regardless of type, during chartreview (Table 1). The PPV rose from 81 to 95% whencase-finding was focused more narrowly on the 2142study subjects with at least two inpatient or outpatientvisits given ICD-9 codes 555 or 556 in computerizeddata (Table 1). Using this algorithm, 14% (n¼ 337) ofcases with one or more diagnosis were not ascertained.In addition, 88% of computerized CD being confirmed

with at least one diagnosis that were confirmed with IBD during chart review

ring chart review Total

Not IBD

ndeterminate Any IBD

21 (2%) 833 (79%) 222 (21%) 1,055 (100%)61 (4%) 1351 (81%) 314 (19%) 1,665 (100%)19 (10%) 182 (98%) 4 (2%) 186 (100%)101 (4%) 2366 (81%) 540 (19%) 2906 (100%)

16 (2%) 738 (93%) 57 (7%) 795 (100%)51 (4%) 1109 (96%) 52 (4%) 1161 (100%)19 (10%) 182 (98%) 4 (2%) 186 (100%)86 (4%) 2029 (95%) 113 (5%) 2142 (100%)

s disease, UC¼ ulcerative colitis, IBD¼ inflammatory bowel disease.



as CD and 87% of UC being confirmed as UC throughchart review.

PPV in relation to utilization patterns

The 11 possible case-finding algorithms were categor-ized into seven groups as shown in Table 2. The firstgroup, ‘inpatient or outpatient visits to primary care orgastroenterology coded 555 or 556’ included threepossible case-finding algorithms (1þ, 2þ, and 3þvisits), with the first serving as the base for comparison.The PPV and proportion of cases with one or more

diagnosis ascertained by the case-finding algorithmvaried somewhat with utilization pattern. In Group 1,the PPV increased as the number of inpatient oroutpatient visit was increased from 1þ to 2þ to 3þ (81to 95 to 97%); the proportion of cases with one or morediagnosis who were ascertained decreased correspond-ingly (from 100%, by definition, to 86 and 73%,respectively). Requiring two or more visits gave thealgorithm with the fewest incorrectly classified cases(proportion of cases with one or more diagnosis, 86%,PPV 95%, sum of false positives and false negatives450). Group 2 evaluated inpatient visits withoutoutpatient visits: proportion of cases with one or morediagnosis 53%, PPV 88%. Group 3, outpatient visitswithout inpatient visits, was inferior to using both visittypes in the algorithm. Group 4, requiring at least one

Table 2. Case ascertainment and PPVof various computerized algorithms to identdata, of whom 2366 were confirmed with IBD during chart review and 540 wer

Group Algorithm No. ofvisits

No. offalse

negatives

No.fals

posit

Base Inpatient or outpatient visits toprimary care or gastroenterologycoded 555 or 556

1þ 0 54

1 Inpatient or outpatient visits toprimary care or gastroenterologycoded 555 or 556

2þ3þ

337649

115

2 Inpatient visits coded 555 or 556 1þ 1106 173 Outpatient visits coded 555 or 556 1þ 133 40

2þ 482 74 Outpatient visits to gastroenterology

coded 555 or 5561þ 607 22

5 Inpatient or outpatient visits coded555 or 556 plus IBD-related drugs

1þ2þ

393607

218

6 Inpatient or outpatient visits coded555 or 556 plus endoscopy

1þ 533 36

7 2þ visits coded 555 or 556, or1 such visit with either IBD-relateddrug or endoscopy

— 60 42

Kaiser Permanente IBD Registry, Northern California, 1996–2002. IBD¼ inflaminterval.�By definition.


outpatient visit to the gastroenterology department,yielded 74% of the cases with one or more diagnosiswith a relatively low PPV (89%). Groups 5 and 6,requiring an IBD-related drug dispensing or anendoscopy procedure, respectively, markedly reducedascertainment of the proportion of cases with one ormore diagnosis. Group 7, allowing 2þ inpatient oroutpatient visits or a single visit with either an IBD-related drug dispensing or an endoscopy procedure,overall was similar to 2þ visits alone, but offeredimproved case ascertainment with a reduced PPV.We evaluated the PPV separately for CD and UC and

observed the result to be similar to those for overallIBD (data available upon request).

PPV in relation to demographic characteristics

We further evaluated whether the PPV of the bestperforming algorithm (2þ inpatient or outpatient visitscoded 555 or 556) changed in relation to patientcharacteristics (Table 3). The PPV for overall IBD waslower in women than men (94 vs. 96%); lower inpersons with co-morbid conditions (Charlson-Deyo 1þvs. 0: 92 vs. 95%); and lower in patients aged 70–89years (89% compared with 93–96% in youngerpersons). The PPVs were similar across ethnic groupsand lengths of enrollment.

ify IBD in 2906 patients with one or more diagnosis in inpatient or outpatiente ruled-out

ofeives

Sum offalse negatives

and falsepositives

Proportionof confirmedcases, %

95%CI PPV, % 95%CI

0 540 100� — 81 80–83

33

450702

8673

84–8770–75

9597

94–9696–98

9 1285 53 51–56 88 86–893 536 94 93–95 85 83–863 555 80 78–81 96 95–976 833 74 72–76 89 87–90

91

612688

8374

82–8572–76

9096

89–9195–97

8 907 77 76–79 83 82–85

3 483 97 97–98 84 83–86

matory bowel disease, PPV¼ positive predictive value, CI¼ confidence


Table 3. Positive predictive value of 2þ computerized IBD diagnoses (inpatient or outpatient, any department) in relation to demographic and clinicalcharacteristics among 2142 cases

Characteristic fromcomputerized data

Confirmed withIBD (n¼ 2029), %

Not confirmed withIBD (n¼ 113), %

PPV, % 95%CI

GenderMale 47 37 96 95–97Female 53 63 94 92–95

Age, years0–18 10 12 93 90–9719–29 11 8 96 94–9930–39 20 17 96 94–9740–49 22 18 96 94–9850–59 17 15 95 93–9760–69 12 12 95 92–9870–89 8 19 89 84–93

EthnicityCaucasian 69 69 95 94–96African-American 15 14 95 93–97Asian 9 11 94 90–97Hispanic 5 4 96 93–100Unknown 2 3 93 86–100

Enrollment during study period (years)1–3 13 19 93 90–964–5 14 19 93 90–966–7 73 63 95 94–96

Charlson comorbidity index0 79 69 95 94–961þ 21 31 92 90–95

The gold standard for confirmation of IBD was chart review. Kaiser Permanente IBD Registry, Northern California, 1996–2002. IBD¼ inflammatory boweldisease, PPV¼ positive predictive value, CI¼ confidence interval.

1090 l. liu ET AL.

We further evaluated the PPVof the best performingalgorithm (2þ inpatient or outpatient visits coded 555or 556) separately for CD and UC separately, observingsimilar results (data available upon request).

PPV among 186 persons with both CD and UCcodes

Among 186 patients with both CD and UC codes oncomputerized data, we determined the PPV of (1) themost common diagnosis recorded in the GI depart-ment; and (2) the most recent diagnosis recorded inoutpatient or inpatient visits (Table 4). Using mostcommon diagnosis recorded in outpatient or inpatientdata, the PPV for CD (57%) was somewhat better thanfor UC (36%), although both were poor. Using the mostrecent diagnosis recorded in the GI department, thePPV was better for CD (83%) than for UC (60%). Allbut four of these 186 cases were confirmed with IBD.

PPV of disease location among patients with2þ visits

We evaluated the validity of computerized informationon disease location recorded during the most recentvisit (Table 5). The PPVs were 64% for Crohn’s


enteritis, 50% for Crohn’s colitis, 75% for UC, and39% for ulcerative proctitis. We further sought toassess whether use of rectal medications indicatedulcerative proctitis confined to the rectum (Table 6).Rectal medications include both suppositories andenemas. Among 150 patients with ulcerative proctitisrecorded in the gastroenterology department, only 58(PPV¼ 39%) were confirmed by chart review. Among92 patients (150� 58¼ 92) not confirmed as ulcerativeproctitis, 50 patients used rectal mesalamine during thestudy period and four were not confirmed with IBDduring chart review.

DISCUSSION

Biologic agents offer new modalities for treatment ofIBD, but identification of case groups is needed toassess drug safety in the United States. Serious adverseevents, especially infections and malignancies, includ-ing T-cell lymphomas in children and young adults, arenot adequately understood, particularly given thebaseline risk of these conditions among those withsevere inflammation. We sought to determine thevalidity of IBD diagnoses on computerized inpatientand outpatient data using medical record review as the


Table 4. Among 186 persons with both CD and UC codes, cross-tabulation of computerized codes compared with chart review

Final diagnosis recorded in the chart

Most recent diagnosis recorded incomputerized data

CD UC Indeterminate IBD Not IBD Total

Most recent diagnosis recorded in theGI departmentCD 67 (83%) 7 (9%) 6 (7%) 1 (1%) 81 (100%)UC 18 (26%) 42 (60%) 10 (14%) 0 (0) 70 (100%)No gastroenterology visit 16 (46%) 13 (37%) 3 (9%) 3 (9%) 35 (100%)

Most common diagnosis recorded in anyinpatient or outpatient visitCD 66 (57%) 37 (32%) 10 (9%) 3 (3%) 116 (100%)UC 35 (50%) 25 (36%) 9 (13%) 1 (1%) 70 (100%)

Total 101 62 19 4 186

Kaiser Permanente IBD Registry, Northern California, 1996–2002. CD¼Crohn’s disease, UC¼ ulcerative colitis, IBD¼ inflammatory bowel disease.

Table 5. PPV of IBD disease location as recorded in computerized data from the gastroenterology department among 2142 patients with 2þ diagnoses

Disease location recorded in mostrecent computerized data

Diagnosis confirmed during chart review

Ileocolitis Crohn’s colitis Proximalulcerativecolitis

Proctitis Unknown/not IBD

Total

Ileocolitis (ICD9 555.0, 555.2, 555.9) 333 (64%) 111 (21%) 11 (2%) 1 (0.2%) 67 (13%) 523 (100%)Crohn’s colitis (ICD9 555.1) 54 (29%) 93 (50%) 10 (5%) 0 (0%) 29 (16%) 186 (100%)Proximal ulcerative colitis(ICD9 556.0, 556.1, 556.3-556.9)

21 (3%) 32 (4%) 599 (75%) 49 (6%) 101 (13%) 802 (100%)

Proctitis (ICD9 556.2) 0 (0%) 3 (2%) 66 (44%) 58 (39%) 23 (15%) 150 (100%)No gastroenterology visit 87 (18%) 65 (14%) 181 (38%) 29 (6%) 119 (25%) 481 (100%)Total 495 (23%) 304 (14%) 867 (41%) 137 (6%) 339 (16%) 2142 (100%)

The gold standard is chart review. Kaiser Permanente IBD Registry, Northern California, 1996–2002. IBD¼ inflammatory bowel disease.


gold standard. Compared with previous validationstudies,5–7 this study was much larger, including 2906patients in contrast to 157,5 448,6 and 830 7 in earlierreports. In addition, this community-based study used arelatively longer follow up period of 7 years, providingtime for patients with quiescent disease to recur.In the Kaiser Permanente population, the PPV for

identification of IBD using computerized data was 81%overall among persons with at least one diagnosis ofCD or UC. A marked improvement in accuracy was

Table 6. Use of rectal medications among patients with ulcerative proctitis

Visit c

With rectal medication

Yes No

Ulcerative proctitis confirmed 36 (42%) 49 (9%Ulcerative proctitis ruled out 50 520Total 86 569


achieved by expanding the number of diagnosis to 2 ormore (PPV, 95%; proportion of cases with one or morediagnosis ascertained, 86%).We examined key utilization variables simul-

taneously to understand their marginal effect on thePPV. The PPV increased rapidly with increasingnumbers of visits, with inpatient and outpatient visitsbeing similarly accurate. It also increased withincreasing drug utilization. Taken by itself, a visit tothe gastroenterology department appeared to be a

oded ulcerative proctitis

Total

Without rectal medication

Yes No

) 22 (34%) 58 (3%) 16542 2129 276164 2187


KEY POINTS

� Resources for studying IBD are needed inevaluations of drug safety including traditionaldrugs and new biologics agents. We developed anIBD registry, with ascertainment from computer-ized visit information.

� We characterized the PPV of IBD case-findingusing computerized data compared with chartreview.

� We identified 2906 persons aged 89 years oryounger with one or more IBD diagnoses incomputerized visit data during the period of1996–2002.

� We found that defining cases as those whounderwent two or more visits without regard todiagnostic procedures or drug utilization maxi-mized the correct classification of true cases(PPV, 95%) while minimizing the incorrectclassification of false cases (14%).

� We conclude that the quality of IBD diagnoses incomputerized data is adequate to meet the aims ofa wide range of research studies.

1092 l. liu ET AL.

useful marker for a correct diagnosis; however, thisproved to be strongly related to the total number ofvisits and to drug utilization. Similarly, endoscopyutilization was not useful in detecting true IBD casesfrom computerized data, possibly due to the lowcolonoscopy rate in Kaiser. This is in contrast withMoun’s study 7 who found colonoscopy at follow upincreased the diagnostic accuracy.Two case-definition algorithms yielded high PPVs

while minimizing loss of confirmed cases: (1) two ormore visits during inpatient or outpatient visits; and (2)one visit with any use of an IBD-related drug.Considering the counter-balance of false negativeand false positive ascertainment other algorithmsproduced either a relatively large number of falsepositive or false negative cases.The PPV varied little (from 93 to 96%) among

patients from different ethnic groups or with length ofenrollment. Computerized diagnoses among womenwere significantly worse than among men, although thedifference in PPV was quite small (2%). Computerizeddiagnoses were also worse among patients with co-morbid conditions, and possibly among those aged 70years and older. Upon closer examination of each of the19 components of the Charlson comobidity index, onlydiabetes with chronic complications (PPV¼ 76%,p¼ 0.004) was significantly associated with the PPV.This could be due to misclassification of othergastrointestinal symptoms and diseases as IBD. Thismisclassification should be taken into considerationwhen interpreting incidence and prevalence rates orother IBD-related study results.Two percent of the patients with an IBD diagnosis

on computerized data were determined not to have IBDon chart review. Although we did not track the reasonfor misclassification, we noticed that many personsmisclassified with IBD had other gastrointestinaldiseases such as diverticulitis, irritable bowel syn-drome, ischemic colitis, radiation enteritis, andinfectious gastroenteritis. Depending on the aims ofa research study, the inclusion of persons with a singlecomputerized diagnosis may not be worthwhile.Our results are comparable with most published data.

Lewis et al.5 have validated IBD diagnoses in theGeneral Practice Research Database from the UnitedKingdom. They mailed surveys to general practitionerscaring for a stratified random sample of 170 IBDpatients. They found a PPVof 92% (95%CI: 86–96%)among 155 usable surveys. In Norway, Moun et al.7

also examined misclassification of IBD diagnosesamong 791 IBD patients ascertained and followed forone to two years after initial case ascertainment. Theyobserved a PPV of 88% for CD and 91% for UC.


Bernstein et al.6 in Canada also measured the accuracyof administrative data diagnoses compared with chartreview. They included cases with more than three IBD-associated visits. Although they did not report the PPV,they reported a sensitivity (proportion of confirmedcases) of 89.2% for CD and 74.4% for UC, comparedwith our levels of 86% among cases with 2þ IBD-associated visits. In collaboration with a fee-for-service health plan, we have previously reviewed 400charts (200 from each site) to validate IBD diagnoses.3

We found a relatively poor PPV (61%, 95%CI: 57–64%) due to the shorter follow up period (30months) ofthat study and possibly to differences between claimsdata and data collected for provision of care(PPV¼ 84%, 95%CI: 79–89%).In conclusion, our study demonstrates the value of

computerized data for identifying IBD patients, with aPPVof 81% among persons with one or more diagnosisand a PPV of 95% among persons with two or morediagnoses in our capitated health plan. Requiring agastroenterology visit, endoscopy utilization, or IBD-related drugs uses results in underascertainment ofcases and generally would not be justified foridentifying prevalent cases. The validity of computer-ized data for identifying subtypes of IBD (CD 88% vs.UC 87% among 2þ diagnoses) and location of disease(64, 50, 75, and 39% for Crohn’s enteritis, Crohn’s



colitis, UC, and ulcerative proctitis, respectively) wasmore modest. The methods used for case-finding in anyparticular study will depend on the research question,the research setting, the relative costs of overascertain-ment and underascertainment, and the resourcesavailable for confirmation of cases. A case-findingstrategy that tolerates underascertainment will likelyyield more severe cases receiving higher levels ofmedical utilization. Characterization of the validity ofcase ascertainment in the IBD registry improves theinterpretability of results generated from this resource.

ACKNOWLEDGEMENTS

This research was supported by grants from the Crohn’s andColitis Foundation of America (CCFA) and the KaiserFoundation Research Institute. We appreciate Dr JamesLewis’ involvement in the study design and data collection.


REFERENCES

1. Herrinton LJ, Liu L, Lewis JD, et al. Incidence and prevalence ofinflammatory bowel disease in a Northern California managed careorganization. Am J Gastroenterol 2008; 103(8): 1998–2006.

2. Gasche C, Scholmerich J, Brynskov J, et al. A simple classification ofCrohn’s disease: report of the working party of the world congresses ofgastroenterology, Vienna 1998. Inflamm Bowel Dis 2000; 6: 8–15.

3. Herrinton LJ, Liu L, Lafata JE, et al. Estimation of the period prevalenceof inflammatory bowel disease among nine health plans using compu-terized diagnoses and outpatient pharmacy dispensings. Inflamm BowelDis 2007; 13(4): 451–461.

4. Brown LD, Cai TT, DasGupta A. Interval estimation for a binomialproportion. Statistical Science 2001; 16(2): 101–117.

5. Lewis JD, Brensinger C, Bilker WB, et al. Validity and completeness ofthe general practice research database for studies of inflammatory boweldisease. Pharmacoepidemiol Drug Saf 2002; 11: 211–218.

6. Bernstein CN, Blanchard JF, Rawsthorne P, et al. Epidemiology ofCrohn’s disease and ulcerative colitis in a central Canadian province:a population-based study. Am J Epidemiol 1999; 149: 916–924.

7. Moum B, Ekbom A, Vatn MH, et al. Inflammatory bowel disease: re-evaluation of the diagnosis in a prospective population based study insouth eastern Norway. Gut 1997; 40(3): 328–332.


Validity of computerized diagnoses, procedures, and drugs for inflammatory bowel disease in a...

Documents

Transcript of Validity of computerized diagnoses, procedures, and drugs for inflammatory bowel disease in a...