ticipated that problems of technical competence, hand-off,
wouldfigurepromi-nently, so we explored these areas in depth.
Fivemalpracticeinsurancecompaniesin4regions(northeast-ern,mid-Atlantic,southwestern,andwesternUnitedStates)par-ticipated in the study. The companies covered approximately33000 physicians, 61 acute care hospitals (35 academic and26 nonacademic),and428 outpatientfacilities.Thestudywasapproved by ethics review boards at the investigators’ institu-tions and by the review sites.
Datawereextractedfromrandomsamplesofclosedclaimfilesateachinsurerinon-sitereviewsconductedbetween2002and2004. The claim file is the repository of information accumu-lated by the insurer during the life of a claim. While the claimis open, it includes medical records pertaining to the episodeof care at issue. For all sampled claims, we reacquired the rel-evant medical records from insured institutions. We defined a claim as a written demand for compensationformedicalinjury.
Wefocusedon4clinicalcategories—(1)obstetric,(2)surgical,(3)missedanddelayeddiagnoses,and(4)medication—andappliedauniformdefinitionofeachacrosssites.These categories cover approximately 80% of all medical mal-practice claims filed in the United States.
Insurers contrib-uted to the study sample in proportion to their annual claimsvolume.Thenumberofclaimsbysitevariedfrom84to662(me-dian, 294).
STUDY INSTRUMENTS AND CLAIM FILE REVIEW
Reviewers were board-certified attending physicians, fellows,or final-year residents in surgery (surgical claims), obstetrics(obstetricclaims),andinternalmedicine(diagnosisandmedi-cation claims). Physician investigators from the relevant spe-cialties trained the reviewers in the content of claims files, useofthestudyinstruments,andconfidentialityproceduresin1-daysessions at each site. Reviews took a mean of 1.6 hours per fileand were conducted by 1 reviewer.A sequence of 4 instruments guided the review. The reviewprocess has been described in detail in previous publica-tions.
Insummary,reviewersmadeaninitialjudgmentaboutwhether an adverse outcome had occurred, and if it did, theyscored the outcome on a severity scale ranging from emotionalinjury to death.
For claims with identifiable adverse out-comes,reviewers proceededto considerthepotentialcontribu-tory role of 17 possible contributory factors in causing the ad-verseoutcome.Next,theyjudgedwhethertheadverseoutcomewas due to medical error, defined as “the failure of a plannedactiontobecompletedasintended(i.e.,errorofexecution)ortheuse of a wrong plan to achieve an aim (ie, error of planning).”
Reviewers’ confidence in the error judgments was recorded ona6-pointscalerangingfrom1(littleornoevidencethatadverseoutcome resulted from error/errors) to 6 (virtually certain evi-dencethatadverseoutcomeresultedfromerror/errors).
Claimsthat scored 4 (more likely than not that an adverse outcome re-sulted from error/errors; more than 50-50 but a close call) orhigherwereclassifiedasinvolvinganerror.Finally,forthesub-setofclaimsjudgedtoinvolveanadverseoutcomeduetoerror,reviewersgathereddetailsoftheclinicalcircumstances,includ-ingthespecialtyofinvolvedcliniciansandtheircontributoryroleasratedona5-pointscale(1,somewhatimportant,to5,highlyimportant).To test the reliability of the claims file review, 10% of thefiles were rereviewed by a second physician from the relevantspecialty who was unaware of the first review. On the basis of 148 pairs of reviews,
scores were 0.78 (95% confidence in-terval,0.65-0.90)forthedeterminationofinjuryand0.63(95%confidenceinterval,0.12-0.74)forthejudgmentthaterroroc-curred. More detailed results of the reliability testing are re-ported elsewhere.
TheunitofanalysisinMIMEPSwastheepisodeofcareinclaims judged to involve errors that led to an adverse outcome. Forease of exposition, we henceforth refer to such episodes as“cases.” For this study, we drew a subsample of cases from thefull sample of cases identified in MIMEPS (n=889). The sub-sample consisted of cases in which the reviewer had rated thecontributory role of a medical student, intern, resident, or fel-low at 4 or 5 or rated it 3 and no other involved clinician hada higher rating. Reliability testing for the determination that 1or more trainees were involved, which was based on 47 pairsof original reviews, showed good agreement (89% agreement;
=0.64 [95% confidence interval, 0.34-0.94]).
AnalyseswereconductedusingtheStata/SE8.0(StataCorp,Col-lege Station, Texas) software package. We generated descrip-tive statistics to examine the characteristics of the trainees, pa-tients,andadverseoutcomesinthestudysample.Wecomparedthe frequency with which contributing factors were associatedwithtraineeerrorsvsnontraineeerrorsusingFisherexacttests. Weconductedanin-depthanalysisofcasesinvolvingprob-lemsofteamworkandtechnicalcompetence,respectively.Withrespect to teamwork problems, we investigated the personnelrelationshipsinwhichtheyoccurred.Caseswithteamworkprob-lems were defined as those in which the original reviewer had judgedthat1ormoreofthefollowingcontributoryfactorsplayeda role in the error: communication breakdowns, supervisionproblems,handoffproblems,failurestoestablishclearlinesof responsibility,andconflictamongclinicalstaff.Definingteam-work according to such behaviors is consistent with previousconceptualizations of teamwork in medicine
and the WorldHealth Organization’s definition of the term.
For problems of technical competence or knowledge, weidentified the task being performed by the physician when theerroroccurred.Thetaskoptionsweredrawnfromalistofcat-egories of general practitioner tasks provided by the Occupa-tionalInformationNetwork,
whicharegeneralenoughtoad-equatelycapturetheworkofcliniciansinarangeofspecialties. We made several modifications to the Occupational Informa-tionNetworklist.Specifically,wesplit1ofthecategories(“spe-cialized medical care to treat or prevent illness, disease, or in- jury”)intononproceduralandproceduralwork,andthenfurtherdivided the latter into procedures that were related to obstet-rical deliveries and those that were not.
Of 889 cases (claims with both error and injury) iden-tified in the parent study, 240 (27%) involved traineeswhose role in the error was judged to be at least moder-ately important. The claims were closed between 1984
(REPRINTED) ARCH INTERN MED/VOL 167 (NO. 19), OCT 22, 2007 WWW.ARCHINTERNMED.COM
©2007 American Medical Association. All rights reserved.