A New, Evidence-based Estimate of Patient Harms Associated with Hospital Care

This article provides an updated estimate of patient harms associated with hospital care based on modern studies published from 2008 to 2011. The study found that a lower limit of 210,000 deaths per year were associated with preventable harm in hospitals, with the true number estimated to be more than 400,000 per year. The article emphasizes the need for increased patient engagement, transparency, and accountability to address the epidemic of patient harm in hospitals.

A New, Evidence-based Estimate of Patient Harms
Associated with HospitalCare
John T.James,PhD
Objectives:Based on 1984 data developed from reviews of medical
records of patients treated in New York hospitals, the Institute of Med-
icine estimated that up to 98,000 Americans die each year from medical
errors.The basis of this estimate is nearly 3 decades old;herein,an
updated estimate is developed from modern studies published from
2008 to 2011.
Methods:A literature review identified 4 limited studies thatused
primarily the Global Trigger Tool to flag specific evidence in medical
records, such as medication stop orders or abnormal laboratory results,
which point to an adverse event that may have harmed a patient.Ulti-
mately, a physician must concur on the findings of an adverse event and
then classify the severity of patient harm.
Results:Using a weighted average of the 4 studies,a lower limitof
210,000 deaths per year was associated with preventable harm in hos-
pitals.Given limitations in the search capability of the Global Trigger
Tool and the incompleteness of medical records on which the Tool de-
pends, the true number of premature deaths associated with preventable
harm to patients was estimated at more than 400,000 per year. Serious
harm seems to be 10- to 20-fold more common than lethal harm.
Conclusions:The epidemic of patient harm in hospitals must be taken
more seriously if it is to be curtailed. Fully engaging patients and their
advocatesduring hospitalcare,systematically seeking the patients’
voice in identifying harms,transparentaccountability forharm,and
intentionalcorrection of rootcauses of harm willbe necessary to ac-
complish this goal.
Key Words: patient harm,preventable adverse events,transparency,
patient-centered care,Global Trigger Tool,medical errors
(J Patient Saf 2013;9: 122Y128)
‘‘Allmen make mistakes, but a good man
yields when he knows his course is wrong,
and repairs the evil. The only crime is
pride.’’V Sophocles, Antigone’’
M edical care in the United States is technically complex at
the individualprovider level,atthe system level,and at
the nationallevel.The amountof new knowledge generated
each year by clinical research that applies directly to patien
can easily overwhelm the individualphysician trying to opti-
mize the care of his patients.1 Furthermore, the lack of a well-
integrated and comprehensive continuing education system
the health professions is a major contributing factor to know
edge and performance deficiencies at the individual and sys
level.2 Guidelines forphysicians to optimize patientcare are
quickly outof date and can be biased by those who write the
guidelines.3Y5At the system level,hospitals struggle with staff-
ing issues, making suitable technology available for patient
and executing effective handoffs between shifts and also be
inpatient and outpatient care.6 Increased production demands in
cost-driven institutions may increase the risk of preventable
verse events (PAEs). The United States trails behind other d
oped nations in implementing electronic medical records fo
citizens.7 Hence,the information a physician needs to optimize
care of a patient is often unavailable.
At the nationallevel,our country is distinguished for its
patchwork of medical care subsystems that can require pat
to bounce around in a complex maze of providers as they se
effective and affordable care.Because of increased production
demands, providers may be expected to give care in subop
working conditions,with decreased staff,and a shortage of
physicians, which leads to fatigue and burnout. It should be
surprise that PAEs that harm patients are frighteningly com
in this highly technical, rapidly changing, and poorly integra
industry.The picture is further complicated by a lack of trans-
parency and limited accountability for errors that harm pati8,9
There are at least 3 time-based categories of PAEs reco
nized in patients that are or have been hospitalized. The bro
definition encompasses allunexpected and harmfulexperience
thata patientencounters as a resultof being in the care of a
medicalprofessionalor system because high quality,evidence-
based medical care was not delivered during hospitalization
harmful outcomes may be realized immediately, delayed fo
or months, or even delayed many years. An example of imm
harm is excess bleeding because of an overdose of an antic
lant drug such as that which occurred to the twins born to D
Quaid and his wife.10 An example of harm that is not apparent
for weeks or months is infection with Hepatitis C virus as a r
of contaminated chemotherapy equipment.11 Harm thatoccurs
years later is exemplified by a nearly lethal pneumococcal i
tion in a patient that had had a splenectomy many years ag
was never vaccinated against this infection risk as guideline
prompts require.12
METHODS
The approach to the problem ofidentifying and enumer-
ating PAEs was 4-fold: (1) distinguish types of PAEs that ma
occur in hospitals, (2) characterize preventability in the con
of the GlobalTriggerTool (GTT), (3) search contemporary
medical literature for the prevalence and severity of PAEs th
have been enumerated by credible investigators based on m
REVIEW ARTICLE
122 www.journalpatientsafety.com J Patient Saf&Volume 9, Number 3, September 2013
From the Patient Safety America, Houston, Texas.
Correspondence: John T. James, PhD, Patient Safety America,
14503 Windy Ridge Lane, Suite 200, Houston, TX 77062
(email: [e-mail]).
The author discloses no conflict of interest.
Sources of support: none.
Copyright * 2013 by Lippincott Williams & Wilkins
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

records assessed by the GTT, and (4) compare the studies found
by the literature search.
Types of PAEs
The cause of PAEs in hospitals may be separated into these
categories:
& Errors of commission,
& Errors of omission,
& Errors of communication,
& Errors of context, and
& Diagnostic errors
Thesedistinctionsare importantbecauseinvestigators
searching for preventable harm must be aware of what they can
find and whatthey cannotfind.The easiesterror to detectin
medical records is an error of commission. This occurs when a
mistaken action harms a patient either because it was the wrong
action or it was the right action but performed improperly. For
example,the patientmay need his gallbladderremoved,but
during the surgery,the intestine is nicked,and the patientde-
velops a serious infection,such as was alleged to be the cause
leading to the death of Representative John Murtha.Errors of
omission can be detected in medical records when an obvious
action was necessary to healthe patient,yetit was notper-
formed at all. For example, a patient may need a A-blocker, but
because itwas notprescribed,the patientdied prematurely.13
Errors of omission because of failure to follow evidence-based
guidelines are much more difficultto detect,partly because
there are many complex guidelines and also because adverse
consequences of failure to follow guidelines may be delayed
until after discharge.14,15
Errors ofcommunication can occurbetween 2 ormore
providers or between providers and patient.One example of a
lethalerrorof communication between providerand patient
occurred when cardiologists failed to warn their19-year-old
patient not to run.The patient had experienced syncope while
running,and 5 days ofinpatient,diagnostic testing were in-
conclusive;however,his cardiologists knew he was notready
to return to running butfailed to warn him againstthis risk.
Having not been warned against running,he resumed running
and died 3 weeks later while running.15
Contextual errors occur when a physician fails to take into
account unique constraints in a patient’s life that could bear on
successful,postdischarge treatment.For example,the patient
may lack the cognitive ability to comply with a medicaltreat-
mentplan ormay nothave reasonable accessto follow-up
care.16 Diagnostic errorsresulting in delayed treatment,the
wrong treatment,or no effective treatmentmay also be con-
sidered separately,although a smallsubsetof these mightbe
included as errors of commission or omission.For example,a
diagnostic error may lead to harm from errors of commission by
overtreatment or mistreatment of the patient until the mistake is
discovered.The apparent eagerness of the U.S.health-care in-
dustry to over diagnose patients often leads to harmful conse-
quences for patients.17
Preventability and the Global Trigger Tool
The prevailing view is that ‘‘preventability’’ of an adverse
eventlinks to the commission ofan identifiable errorthat
caused an adverse event. Adverse events that cannot be traced to
a likely error should not be called ‘‘preventable.’’ The portion of
adverse events thatare deemed preventable tends to be about
50% to 60%;however,recently,experts have postulated that
virtually alladverse events they identified with the ‘‘GTT are
preventable.’’18 The GTT dependson systematic review of
medicalrecords by persons trained to find specific clues or
triggers suggesting thatan adverse eventhas taken place.For
example, triggers might include orders to stop a medication
abnormallab result,or prescription of an antidote medication
such as naloxone. As a final step, the examination of the rec
mustbe validated by 1 or more physicians.As will be shown
shortly,the methods used to find adverse events in hospital
medicalrecords targetprimarily errors of commission and are
much less likely to find harm from errors of omission,com-
munication,context,or missed diagnosis.19 There are some
overlaps in these categories and cascades of harmful events
ensue from a single rootcause.A ‘‘perfectstorm’’ of unrec-
ognized butcorrectable medicalerrors can resultin serious
harm or death.15,20
Literature Search
Our literature search included the following three terms:
medicalerror,globaltriggertool,and hospital.We searched
Pub Med and ‘‘reports and publications’’ from the governme
Web site http://oig.hhs.gov. Those searches turned up 20 ar
published between 2006 and 2012, of which, 4 were found t
suitable for the present analysis. The unsuitable studies incl
studies of populations outside the United States,studies con-
fined to narrow hospital populations (e.g., intensive care uni
studies of ambulatory patients,studies involving only method-
ologicalcomparisons,adverse-eventissue papers,failures of
incident reporting systems, and studies that did not classify
severity of the harm associated with adverse events.
Characterization of the Core Studies
The 4 key studies were reviewed for similarity and differ
ence in methods used to find adverse events. It was found t
each one employed similar methods to flag,confirm, and then
classify adverse events according to level of harm.All studies
used a 2-tier approach thatconsisted of screening of medical
records by nonphysicians, usually nurses or pharmacists, to
suspectevents.In the second tier,physiciansexamined the
suspect events to determine if a genuine adverse event had
curred and,if so, the levelof seriousness of the event.In all
studies, the GTT from the Institute for Healthcare Improvem
was the primary screening tool;21however, there were variations
in the supplementary tools used to detect potential adverse
A 2008 pilotstudy by the Office ofInspectorGeneral
(OIG) of the Department of Health and Human Services used
5 methods in its search for adverse eventsVnurse reviews us
the GTT, conditions that were not present on admission (POA
beneficiary interviews,hospitalincidence reports,and patient
safety indicators.22 The pilot study revealed that the GTT cap-
tured the highestpercentage (78%)of the events ultimately
deemed to be adverse events in the second tier review by p
sicians.The use ofPOA indicatorcodes was second bestat
61%.Together,these methods were found to identify 94% of
the flags thatled physicians to declare thatan adverse event
had taken place.A more comprehensive OIG study in 2010
employed these 2 screening methodsand a third based on
whetherthe patienthad been readmitted to the hospitalwith
30 days of discharge from the last discharge during the Octo
2008 index period.23
A study by Classen and colleagues also employed the GT
along with Agency for Healthcare Research and Quality Patie
Safety Indicators (PSIs) and hospital reports of adverse even
Of the 167 flagged events thatultimately were deemed true
adverse events by physician review,the GTT detected 90% in
the severity levels F through I(Table 1).18 The longitudinal
J Patient Saf&Volume 9, Number 3, September 2013 Patient Harms Associated with HospitalCare
* 2013 Lippincott Williams & Wilkins www.journalpatientsafety.com123
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

study by Landrigan and colleagues relied on the GTT and POA
indicators to flag possible adverse events. Like the other studies,
the ultimate determination of a genuine adverse event and the
severity ofthe eventwere judged by physicians during the
second-tieranalysis.24 Although there are slightvariations in
the approach used to discover flags in the records examined by
the 4 studies, the GTT was the core method placed in the hands
of trained and experienced nurses. All studies used a second tier
requiring physicians to determine whethera flag signaled a
genuine adverse event and, if so, then assign a severity level to
that event.All studies used the National Coordinating Council
for Medication Reporting and Prevention scale (Table 1).
RESULTS
Recentdata from the 4 key studies provide a more com-
prehensive, evidence-based estimate of the number of lethal and
serious medical errors than the one provided by the Institute of
Medicine (IOM).25These data are compiled in Table 2, and the
studies are described below.
A pilot study by the OIG was published in 2008 in an effort
to explore the effectivenessof search methodsfor adverse
events.21As noted in the methods section, this study relied on 5
search methods for flagging potential adverse events in medical
records but did not specify whether such events were prevent-
able. The 278 medical records reviewed by screeners and phy-
sicianswere notrandomly selected to be representative of
Medicare hospitalizations; instead, they originated from hospi-
tals in 2 unspecified counties. Of the 51 serious adverse events
identified,only 3 were on the National Quality Forum’s list of
serious reportable events and only 11 were on Medicare’s Hospital
Acquired Condition (HAC) list. In 2010, the OIG estimated ad-
verse events in hospitalized Medicare patients.23
Investigators looked atthe medicalrecords of780 ran-
domly selected patients chosen to represent the 1 million Medi-
care patients‘‘discharged’’from hospitals in the month of
October 2008.The total number of hospital stays for the 780
patients during this period was 838 because some of the ben-
eficiaries were hospitalized and discharged more than once
during the 1-month index period.Using primarily the GTT
developed by the Institute for Healthcare Improvement to find
adverse events, investigators found 128 serious adverse events
(level of harm F, G, H, or I) that caused harm to patients, and an
adverse event contributed to the deaths of 12 of those patients.
Seven of these deaths were medication related,2 were from
blood stream infections,2 were from aspiration,and the 12th
one was linked to ventilator-associated pneumonia. Only 2 of
these events were on the National Quality Forum list, and none
were on the Medicare HAC list.The authors ofthis report
estimated that ‘‘events’’ contributed to the deaths of 1.5 % (12/
780) of the 1 million Medicare patients hospitalized in October
2008.Thatamounts to 15,000 per month or 180,000 per year.
TABLE 1. Adverse Events Classified as Serious
Level of Harm Description
F Required prolonged hospital stay
G Permanent harm
H Life sustaining intervention required
I Contributing to death of patient
Adapted from the National Coordinating Council for Medication
Errors Reporting and Prevention.
TABLE 2. Recent Studies of Preventable Adverse Events
Reference
Source of Medical
Record Data
Time Covered
by Records
No. records
Reviewed
Search Tool
or Method
Serious Adverse
Events (Class F to I)
Found (%)
% Deemed
Preventable
Lethal Adverse
Events (%)
Major Causes of
Lethal Events
OIG (2008) Medicare beneficiaries
in 2 counties
1 wk in August
2008
278 Global trigger tool 43 (15%) n/s 3 (1.1%) n/s
OIG (2010) Representative
Medicare patients
October 2008 838 Global trigger tool 128 (15%) 44% 12 (1.4%) 7-medication, 2-sepsis,
2-aspiration, 1-other*
Classen et al.
(2011)
3 tertiary-care hospitalsOctober 2004 795 Global trigger tool 167 (21%) ~100% 9 (1.1%) 4-procedure, 2-pulmonary,
1-infection, 2-not
specified
Landrigan, et al.
(2010)
10 hospitals in North
Carolina
Jan 2002 through
Dec 2007
2341 Global trigger tool 332 (14%) 63% 14 (0.6%) 7- HAI, 3-Renal/endocr.
4-other systems†
* Ventilator-associated pneumonia.
† Cardiac arrest, pulmonary embolism, hematologic event, neurological event.
James J Patient Saf&Volume 9, Number 3, September 2013
124 www.journalpatientsafety.com * 2013 Lippincott Williams & Wilkins
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Note that the percentage of deaths per hospitalization was slightly
lower at 1.4% (12/838).The authors did not explicitly state the
percentage of the lethal adverse events that were preventable, but
given theirdescription of the events,it seems thatmostwere
preventable. Overall, physician reviewers estimated that 44% of
serious medical events were preventable.
In a somewhat similar study published in March 2011 in
the journal Health Affairs,investigators examined the medical
records of 795 patients treated in 1 of3 tertiary hospitals in
the month of October 2004.18 These hospitals had been recog-
nized for theireffortsto improvepatientsafety.The in-
vestigators also used the GTT to discover adverse events. They
found 167 adverse events in the categories F through I, and 9 of
the adverse events contributed to the deaths of patients (cate-
gory I).Thus,an adverse eventcontributed to death in 1.1%
of these patients.The causes were as follows:procedure re-
lated (not infection)V4,nosocomialinfectionV1,pulmonary/
venous thromboembolismV2,and unspecified otherV2.In-
terestingly,none of the deaths were explicitly associated with
medication errors,which were the primary causes of death in
the Medicare patients studied by the OIG.23 Medication-related
errors caused 35% of the category-F harms in the Health Affairs
study.18 The average age ofthe patients whose records were
examined was 59 years. The 10 authors of the original study did
notformally assess the preventability oferrors,declaring in-
stead that it is their belief that all adverse events are preventable.
In a fourth recent study targeting changes in patient safety in
10 hospitals in North Carolina,there was a lowerincidence of
deaths associated with adverse events.24Hospitals in North Carolina
were chosen because hospitals in that state had shown a ‘‘high level
of engagement in efforts to improve patient safety.’’ In that state,
96% ofthe hospitals had enrolled in a nationalcampaign to
improve patient safety, whereas the average in other states was
only 78%.A priori,a lower rate of preventable adverse events
than the national average could be expected.The investigators
studied the change in incidence of adverse events using the GTT
on 10 randomly selected medical records per quarter from the
firstquarter of 2002 to the lastquarter of 2007. The tool was
applied by internal and external reviewers; however, the internal
reviewers had better kappa scores (a measure of agreement) when
compared with experienced external reviewers, so the results of
internal reviews, which were the only ones given in detail in the
original paper, will be used here. Based on 2341 admissions and
the finding of 14 cases where adverse events contributed to death,
the percentage of lethal adverse events was 0.60%. The primary
causes of death were hospital-acquired infections (HAIs) (7) and
acute renal failure (2). Other causes are shown in Table 2. This
study involved many more medicalrecords than the OIG or
Health Affairs study, but the hospitals and patients were not se-
lected to be representative of hospitals around the country.The
hospitals were selected because the investigators felt that North
Carolina had made a concerted effortto improve patient safety
over the study period. It is not surprising that the percentage of
serious or lethal adverse events was lower than in the other studies
summarized in Table 2.
All 4 studies (Table 2) have similar, 2-tier search methods
to identify serious adverse events.The GTT,supplemented by
other less comprehensive methods,was applied to medical re-
cords by experienced nonphysicians to identify possible adverse
events,and then,physician reviewers determined which flags
were associated with an adverse event.However,the study
populations were quite different.One would expectthe OIG
studies ofMedicare patients,who tend to have more comor-
bidity than the average hospitalized patient, to show the highest
incidence of lethalPAEs.One would expectthe incidence of
lethaladverse events in tertiary hospitals to be above the na
tionalaverage forall hospitalizations because more complex
illnesses are treated there with longer hospital stays. One w
expect, as the original authors did, that the incidence data f
North Carolina would be below the national average for leth
adverse events because of concerted efforts in that state to
prove patient safety in hospitals compared with the average
other states in the United States.
It is our opinion that none of the 4 studies alone can pro
vide a defensible estimate for hospitals across the United St
however, by combining the studies, an evidence-based estim
of the number of lethal PAEs across the country can be deve
oped. The most favorable way to combine the 4 studies to fi
the lowest reasonable estimate is to weigh the studies accor
to how many medical records from a single hospital stay we
reviewed by each team ofinvestigators.This means thatthe
study ofpatients hospitalized in North Carolina was heavily
weighted compared with the other studies.Thus,there were a
total of 4252 records reviewed (compiled from Table 2). Amo
the records reviewed, there were 38 total deaths associated
adverse events.The ratio projects to a death rate from adverse
eventsof 0.89%.This is well below the percentagesfrom
Medicare and tertiary-care studies (1.1%Y1.4%) and well ab
the data from the North Carolina study (0.60%). There were
estimated 34.4 million hospitaldischarges in 2007,26 and the
average percentage ofpreventable adverse events among all
adverse events in the 3 studies where this was reported or p
tulated was 69% (averaged from Table 2).Thus,the best esti-
mate from combining these 4 studies is 34,400,000 0.69
0.0089 = 210,000 preventable adverse events per year that
tribute to the death of hospitalized patientsVbased primarily
evidence in hospital medical records found by the GTT meth
DISCUSSION
There has been no lack of contention about the prevalen
of PAEs,which herein willbe considered synonymous with
medical errors that cause harm to patients; this does not inc
near misses that do not harm patients.27,28
The first estimate of
medicalerrors thatreceived widespread attention was declared
by the IOM in its now- famous book called ‘‘To Err is Human.25
The IOM provided 2 estimates of the number of deaths from
medical errors, but careful inspection of the origin of these e
timates show thatthey were based on data thatare now quite
old. The earliest estimate originated from the Harvard Medic
Practice Study in which 30,000 randomly selected discharge
records from 1984 in 51 New York hospitals were examined.29
The investigators found that serious adverse events occurre
3.7% of the hospitalizations.Of the adverse events, 58% were
attributable to error (i.e.,they were preventable).Of this frac-
tion,13.6% resulted in death.Extrapolated to 33.6 million
hospitalizations nationwide in 1997,simple arithmetic yielded
the following:33,600,000 0.037 0.136 0.58 = 98,000
deaths per year. Another study of 15,000 medical records fr
Colorado and Utah in 1992 found lower rates of adverse eve
and death,from which the IOM estimated 44,000 deaths na-
tionwide per year.25 Although physician reviews reveal adverse
events due to ‘‘negligence,’’ which was about28% to 29% in
both studies,a later publication from the IOM suggested that
the 44,000 to 98,000 deaths did notinclude errors ofomis-
sion.30 Because the New York study included a larger sample,
the deaths-per-year figure of 98,000 attributed to the IOM is
estimate most often quoted.In fact,the IOM declared that the
‘‘number of deaths [per year] due to medicalerror may be as
high as 98,000.’’
J Patient Saf&Volume 9, Number 3, September 2013 Patient Harms Associated with HospitalCare
* 2013 Lippincott Williams & Wilkins www.journalpatientsafety.com125
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Why is the present estimate of the number of lethal PAEs
so much higherthan the highestestimate (98,000)from the
IOM? It is likely that the bar for identification of a PAE in the
New York/IOM study was much higher than in the 4 modern
studies and that the GTT is better able to identify adverse events
than general reviews by physicians, which was the method used
in the older studies cited by the IOM.19 It is also possible that
the frequency of preventable and lethalpatientharms has in-
creased from 1984 to 2002Y2008 because ofthe increased
complexity of medicalpractice and technology,the increased
incidence of antibiotic-resistant bacteria, overuse/misuse of med-
ications,an aging population, and the movement of the medical
industry toward higherproductivity and expensive technology,
which encourages rapid patient flow and overuse of risky, inva-
sive, revenue-generating procedures.31Y33
Several observations aboutthe 4 varied studies described
in the ‘‘Results’’ section are in order. Although they used varied
selection criteria for the patientpopulations and hospitals,the
results in terms of the portion of adverse events found and the
portion ofdeath-associated events are notremarkably varied.
The percentage of serious adverse events (class F to I) ranged
from 14% to 21%,and the percentage of death-associated ad-
verse events (class I) varied from 0.60% to 1.4%.The result
found in records from North Carolina hospitals (0.60%) is likely
to be below the national average because patient safety efforts in
thatstate have been more intense when compared with other
states. The results from the other studies would be expected to
be above the national average because of the age of the patients
and seriousness of the illnesses. This dispersion of percentages
makes sense and gives one confidence that the estimate of the
average number of preventable, lethal adverse events based on
hospital medical records screened by the GTT approach is rep-
resentative of the nation as a whole. The portion of serious ad-
verse eventsthatwere notlethal(classF, G, and H) were
roughly 10- to 20-fold larger than the portion of lethalPAEs.
This leads to a rough estimate of 2 to 4 million serious,PAEs
per year that would be discoverable in medical records using the
GTT approach.
There are important limitations to the 4 modern studies that
mustbe considered.Premature deaths as a resultof medical
errors may occur many years after the hospital stay because the
patient’s care was notoptimalor did notfollow guidelines.12
Furthermore,lethal PAEs can been missed by the GTT and by
physician reviews. The GTT does not detect diagnostic errors or
errors of omission,especially those involving failure to follow
guidelines.19 Lethaldiagnostic errors have been estimated to
affect 40,000 to 80,000 people per year including outpatients.34
Physicians have been indefensibly slow to adopt guidelines that
would potentially preventpremature deathsor harm.35 One
egregious example is the estimated 100,000 heartfailure pa-
tients that died prematurely each year in the late 1990s because
they did notreceive beta-blockers.13 The efficacy ofbeta-
blockers was established by a study published in the JAMA
in 1982.36
The 4 modern studies also rely heavily on information in
medicalrecords.One study ofmedicalrecords showed that
quality scores of607 randomly selected medicalrecords on
cardiac patients treated in 219 hospitals from January 2004 to
June 2005 averaged 12.5/20 points, which suggests rather poor
medical record keeping.37 The quality scores were determined
based on the medical records including cardiac history, perfor-
mance and cognition levels, current medications and medication
allergies,differentialdiagnosis,and planned use of evidence-
based medicine.Hospitalswith low-scoring records(0Y10
points) had a 40% higher in-hospital death rate than those that
scored high (15Y20 points). Furthermore, the larger OIG stu
noted that‘‘To the extentthatthe study did notidentify an
event, it was likely because the three screening methods fa
to flag the case for physicians review or because document
in the medical records was incomplete.’’23
A few years afterthe seminalpublication by the IOM,
another IOM panel recognized the limitations of using medic
records provided by medicalinstitutions as the basis for iden-
tifying medical errors. When an adverse event is alleged an
evaluation is undertaken,the ‘‘sentineleffectcan significantly
alter the data that are recorded.’’30There are anecdotal accounts
of data altering or omission of critical data when mistakes a
alleged;however,to our knowledge,scientific studies of this
phenomenon have been lacking until recently.
In a study thatbroke pastthe wallof silence aboutdis-
covery of medicalerrors thatwere missing from medicalre-
cords, Weissman and colleagues found that 6 to 12 months
their discharge, patients could recall 3 times as many seriou
preventable adverse events as were reflected in their medic
records.14 This study involved review of 998 medical records
of patients hospitalized in Massachusetts for medicalor sur-
gical treatment from April to October 2003. Record reviews
specially trained nursesand doctorsidentified 11 serious
PAEs from the records.The method was one adapted from
the Harvard Medical Practice Study, which is the method us
by the core result in the report from the IOM asserting up to
98,000 deaths per year occur from medical errors.25 However,
interviews with patients identified 21 additional serious PAE
thatwere notdocumented in the medicalrecords.Of the
21 undiscovered, serious PAEs,12 occurred predischarge and
9 occurred postdischarge.The predischarge serious PAEs in-
cluded the following: adverse drug events (3), nerve or vess
injury orwrong operation (4),deep venous thrombosis (2),
hospitalacquired infection (2),and postoperative respiratory
distress (1). The serious PAEs postdischarge included the fo
lowing: wound infection (6), deep venous thrombosis (1), op
erative wound dehiscence (1),and operative organ injury (1).
Even in this study,the investigators found only those errors
that patients were aware had happened. There certainly ma
more serious errors thatwentundocumented and were un-
known to patients. Weismann’s finding that evidence of ma
serious adverse events is notapparentin medicalrecords is
reinforced by some olderstudies.For example,it has been
pointed out that some medical errors are not known by clini
cians and only come to lightduring autopsies,which have
found misdiagnoses in 20% to 40% of cases.38 ‘‘Aggressive’’
searches foradverse drug events and prompted self-reports
from clinicians have shown a much higherrate ofadverse
drug events than are evident in the medical records.39 A com-
parison of direct observation for medication errors with revi
of documentation in medicalrecordsin 36 hospitalsand
skilled-nursing facilities found that far more errors were fou
by direct observation than by inspection of medical records40
A recent national survey showed that physicians often r
fuse to report a serious adverse event to anyone in authorit41
In the case of cardiologists,the highest nonreporting group of
the specialtiesstudied,nearly two-thirdsof the respondents
admitted thatthey had recently refused to reportat leastone
serious medical error, of which they had first-hand knowled
to anyone in authority.It is reasonable to suspectthatclear
evidence of such unreported medical errors often did not fin
their way into the medicalrecords of the patients who were
harmed.
The bottom line on total, lethal PAEs as a result of care
hospitals cannotbe estimated in a statistically rigorous way.
James J Patient Saf&Volume 9, Number 3, September 2013
126 www.journalpatientsafety.com * 2013 Lippincott Williams & Wilkins
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Based on our extrapolation from the 4 modern studies, there are
at least 210,000 lethal PAEs detectable by the GTT approach to
record reviews. To deal with other factors that should be applied
to this estimate,the ‘‘weightof evidence’’approach mustbe
engaged. In addition to the core estimate of 210,000, one must
consider evidence of the following:
& life-shortening errors ofomission due to failure to follow
medical guidelines that the GTT approach misses,19
& a factorfor evidence oferrors ofcommission thatare not
documented in medical records,37,39
& failure to make life-saving diagnoses.38
In lightof the evidence above,and especially thatof the
Weisman study,14and although it is probably an underestimate,
a minimum estimateof a 2-fold increasein the medical
recordYbased estimateis reasonable to compensatefor the
known absence ofevidence in medicalrecords oferrors of
commission and the inability ofthe GTT to detecterrors of
omission even when the evidence thatguidelineswere not
followed may be presentin the medicalrecord.Note thatthe
Weisman study suggests a factor of 3 (32/11) for undocumented
evidence ofserious PAEs caused during hospitalization,but
here,we settle for a factor of 2.14 To this,one should add the
undetected diagnostic errors.If we begin with the minimum
estimate of 40,000 and assume that only half of these occur in
hospitals, then the math looks like this: (210,000 2) + 20,000
~ 440,000 PAEs thatcontribute to the death of patients each
yearfrom care in hospitals.This is roughly one-sixth ofall
deaths that occur in the United States each year. The problem of
PAEs must emerge from behind the ‘‘Wall of Silence’’ and be
addressed for the sake of prolonging the lives of Americans.
Needed changes involve not only doctors and hospitals but
increased participation by patients in their health-care decisions.
Perhaps it is time for a national patient bill of rights for hospi-
talized patients that would empower them to be thoroughly in-
tegrated into their care so that they can take the lead in reducing
their risk of serious harm and death.15All evidence points to the
need for much more patient involvement in identifying harmful
events and participating in rigorous follow-up investigations to
identify rootcauses.42 Even for those harms identified in the
medical records of Medicare patients, only 14% become part of
the hospital’s incident reporting system.9 Physician observers of
our hospitalshave made Congresspainfully aware thatthe
hospital peer-review system has widespread failures that permit
negligent care by physicians.43 Hospitals are simply not going
to heal withoutattentive,systematic listening to those harmed
patients or their survivors.
CONCLUSIONS
There was much debate afterthe IOM reportaboutthe
accuracy of its estimates. In a sense, it does not matter whether
the deaths of 100,000, 200,000 or 400,000 Americans each year
are associated with PAEs in hospitals.Any of the estimates
demands assertive action on the partof providers,legislators,
and people who willone day become patients.Yet,the action
and progress on patientsafety is frustratingly slow;however,
one musthope thatthe present,evidence-based estimate of
400,000+ deaths peryearwill fosteran outcry foroverdue
changes and increased vigilance in medical care to address the
problem of harm to patients who come to a hospitalseeking
only to be healed.
REFERENCES
1. Zilberberg MD. The clinical research enterpriseVtime to change
course? JAMA. 2011;305:604Y605.
2. IOM (Institute of Medicine). Redesigning Continuing Education
in the Health Professions. Washington, DC: The National Academies
Press; 2010.
3. Sniderman AD, Furberg CD. Why guideline-making requires reform.
JAMA. 2009;301:429Y431.
4. Ferket BS, Colkesen EB, Visser JJ, et al. Systematic review of guidelin
on cardiovascular risk assessment. Arch Intern Med. 2010;170:27Y40
5. Mendelson TB, Meltzer M, Campbell EG, et al. Conflicts of interest in
cardiovascular clinical practice guidelines. Arch Intern Med.
2011;171:577Y585.
6. Gittell JH. High Performance HealthcareVUsing the Power of
Relationships to Achieve Quality, Efficiency and Resilience. New York
NY: McGraw Hill; 2009
7. American College of Physicians. Achieving a high performance health
care system with universal access: What the United States can learn
from other countries. Ann InternMed. 2008;148:55Y75.
8. Reid, RO, Friedberg MW, Adams JL, et al. Associations between
physician characteristics and quality of care. Arch Intern Med.
2010;170:1442Y1449.
9. Levinson DR. Hospital Incident Reporting Systems Do Not Capture
Most Patient Harm. DHHS, OIG. 2012, OEI-06-09-00091.
10. California Injury Lawyers Blog. Available at: http://www.
californiainjurylawyersblog.com/2009/06/california-medical-malpracti
dennis-quaids-twins-to-receive-500000-for-heparin-overdose.html.
Accessed July 12, 2012.
11. McKnight EV, Bennington TT. A Never EventVExposing the Largest
Outbreak of Hepatitis C in American Healthcare History. Fremont, NE
History Examined, LLC; 2010
12. Ghandi TK, Zuccotti G, Lee TH. Incomplete careVOn the trail of flaws
in the system. N Engl J Med. 2011:365:486Y488.
13. Gheorghaide M, Gattis WA, O’Conner CM. Treatment gaps in the
pharmacologic management of heart failure. Rev Cadiovasc Med.
2002;3:S11YS19.
14. Weismann JS, Schneider EC, Weingart SN, et al. Comparing
patient-reported hospital adverse events with medical records review
Do patients know something that hospitals do not? Ann Intern Med.
2008;149:100Y108.
15. James JT. A Sea of Broken HeartsVPatient Rights in a Dangerous,
Profit-Driven Health Care System. Bloomington, IN:
AuthorHouse; 2007.
16. Weiner SJ, Schwartz A, Weaver F, et al. Contextual errors and failures
individualizing patient care. Ann Intern Med. 2010;153:69Y75.
17. Welch HG, Schwartz LM, Woloshin S. Over-diagnosedVMaking
People Sick in the Pursuit of Health. Boston, MA: Beacon Press; 2011.
18. Classen, DC, Resar R, Griffin F, et al. ‘‘Global trigger tool’’ shows that
adverse events in hospitals may be ten times greater than previously
measured. Health Aff. 2011;30:581Y589.
19. Parry G, Cline A, Goldmann D. Deciphering harm measurement.
JAMA. 2012;307:2155Y2156.
20. Walter D. Collateral DamageVA Patient, a New Procedure, and the
Learning Curve. Charleston, SC: CreateSpace; 2010.
21. Institute for Healthcare Improvement. IHI Global Trigger Tool Guide.
Cambridge MA, 2008. Available at: http://www.ihi.org/knowledge/
Pages/Tools/IHIGlobalTriggerToolforMeasuringAEs.aspx. Accessed
July 12, 2012.
22. Department of Health and Human Services, Office of the Inspector
General. Adverse Events in Hospitals: Case Study of Incidence among
Medicare Beneficiaries in Two Selected Counties. Washington, DC;
2008, Available at: http://oig.hhs.gov/oei/reports/OEI-06-08-00220.pd
23. Department of Health and Human Services, Office of the Inspector
General. Adverse Events in Hospitals: National Incidence among
J Patient Saf&Volume 9, Number 3, September 2013 Patient Harms Associated with HospitalCare
* 2013 Lippincott Williams & Wilkins www.journalpatientsafety.com127
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.

Medicare Beneficiaries. Washington, DC; 2010, Available at:
http://oig.hhs.gov/oei/reports/OEI-06-09-00090.pdf.
24. Landrigan, CP, Parry GJ, Bones CB, et al. Temporal trends in rates of
patient harm resulting from medical care. N Engl J Med.
2010;363:2124Y2134.
25. IOM (Institute of Medicine). To Err is HumanVBuilding a Safer Health
System. Washington, DC: The National Academies Press; 2000.
26. Hall MJ, DeFrances CJ, Williams SN, et al. National Health
Statistics Report. CDC Report Number 29; 2010. Available at:
http://www.cdc.gov/nchs/data/nhsr/nhsr029.pdf.
27. Hayward RA, Hofer TP. Estimating hospital deaths due to medical
errors. JAMA. 2001;286:415Y420.
28. Goodman JC, Villarreal P, Jones B. The social cost of adverse medical
events, and what we can do about it. Health Aff. 2011;30:590Y595.
29. Brennan TA, Leape LL, Laird NM, et al. Incidence of adverse events
and negligence in hospital patients: results of the Harvard Medical
Practice Study. N Engl J Med. 1991;324:370Y376.
30. IOM (Institute of Medicine). Patient SafetyYAchieving a New Standard
of Care. Washington, DC: The National Academies Press; 2004.
31. Moody J, Cosgrove SE, Olmsted R, et al. Antimicrobial stewardship: a
collaborative partnership between infection preventionists and
healthcare epidemiologists. Infect Control Hosp Epidemiol.
2012;33:328Y330.
32. Schneider EL, Campese VM. Adverse drug responses, an increasing
threat to the well- being of older patients. Arch Intern Med.
2010;170:1148Y1149.
33. Stergiopoulos K, Brown DL. Initial coronary stent implantation with
medical therapy vs medical therapy alone for stable coronary artery
disease. Arch Intern Med. 2012;172:312Y319.
34. Newman-Toker DE, Pronovost PJ. Diagnostic errorsVthe next frontier
for patient safety. JAMA. 2009;301:1060Y1062.
35. Kotchen TA. Why the slow diffusion of treatment guidelines into
clinical practice? Arch Intern Med. 2007;167:2394Y2395.
36. A randomized trial of propranolol in patients with acute myocardial
infarction. I. Mortality results. JAMA. 1982;247:1707Y14. Available at
http://www.ncbi.nlm.nih.gov/pubmed/7038157.
37. Dunlay SM, Alexander KP, Melloni C, et al. Medical records and qualit
of care in acute coronary syndromes. Arch Intern Med. 2008;
168:1692Y1698.
38. Leape L. Institute of Medicine medical error figures are not
exaggerated. JAMA. 2000;284:95Y97.
39. Weingart SN, Wilson RM, Gibberd RW, et al. Epidemiology of medica
error. BMJ. 2000:320:774Y777.
40. Flynn EA, Barker KN, Pepper GA, et al. Comparison of methods for
detecting medication errors in 36 hospitals and skilled-nursing
facilities. Am J Health-System Pharm. 2002;59:436Y446.
41. Campbell EG, Regan S, Gruen RL, et al. Professionalism in medicine:
results of a national survey of physicians. Ann Intern Med. 2007;
147:795Y802.
42. Junya Z, Struver S, Epstein A, et al. Can we rely on patients’ reports
adverse events? Med Care. 2011;49:948Y955.
43. Rogan GN, Sebat F, Grady I. How Peer Review Failed at Redding
Medical Center, Why It Is Failing Across the Country and What Can
be Done About It. Congressional Report, June 1, 2008. Available at:
http://www.allianceforpatientsafety.org/redding-failure.pdf. Accessed
July 12, 2012.
James J Patient Saf&Volume 9, Number 3, September 2013
128 www.journalpatientsafety.com * 2013 Lippincott Williams & Wilkins
Copyright © 2013 Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited.