The assessment of adverse events
in hospitals in Brazil
WALTER MENDES, MOˆ NICA MARTINS, SUELY ROZENFELD AND CLAUDIA TRAVASSOS
National School of Public Health, Oswaldo Cruz Foundation, Rio de Janeiro, Brazil
Abstract
Objective. To evaluate the incidence of adverse events in Brazilian hospitals.
Design. Retrospective cohort study based on patient record review.
Setting. Three teaching hospitals in the State of Rio de Janeiro, Brazil.
Participants. Random sample (1103) of 27 350 adult patients admitted in 2003. Patients under 18 years old, psychiatric
patients and patients whose length of stay was less than 24 hr were excluded, and obstetric cases were included.
Main Outcome Measure(s). Incidence of patients with adverse events; proportion of preventable adverse events; number of
adverse events per 100 patients and incidence density of adverse events per 100 patient-days.
Results. The incidence of patients with adverse events was 7.6% (84 of 1103 patients). The overall proportion of preventable
adverse events was 66.7% (56 of 84 patients). The incidence density was 0.8 adverse events per 100 patient-days (103 of 13
563 patient-days). The patient’s ward was the most frequent location of adverse events (48.5%). In regard to classification,
surgical adverse events were the most frequent ones (35.2%).
Conclusions. The incidence of patients with adverse events at the three hospitals was similar to that in international studies.
However, the proportion of preventable adverse events was much higher in the Brazilian hospitals.
Keywords: adverse events, retrospective patient record review, patient safety, quality in health care
Introduction
The Harvard Medical Practice Study, conducted in
New York State in 1984, drew attention to the occurrence of
adverse events in hospitals, as a severe and little-known
problem [1]. Several other studies were published sub-
sequently in developed countries, using a similar method-
ology based on retrospective patient record reviews: in the
USA [2], Canada [3], Denmark [4], France [5], Australia [6],
New Zealand [7], UK [8] and more recently Spain [9].
The retrospective patient record review method, con-
sidered the ‘gold standard’ for the identification of adverse
events, has been criticized due to its cost and limitations for
monitoring the occurrence of adverse events in health ser-
vices [10]. However, the measurement of frequency of
adverse events and the identification of the most common
causes have been important stages adopted by several
countries to draw attention to the seriousness of the problem
and to guide the development of policies for patient safety.
A recent review indicates that prospective or retrospective
studies do not explain the differences observed in adverse
outcome rates between studies. However, study methodology
is an important factor, which can influence the mortality
differences [11].
In Brazil, as in other developing countries, research on
adverse events has focused on the frequency of adverse
events associated with specific causes, e.g. medication use
[12] and emergency care [13]. Still, it is important to know
the overall incidence of patients with adverse events during
hospital stay in order to understand the problem’s magnitude
in Brazilian hospitals and to encourage and orient the devel-
opment of policies to improve quality. This study aims to
evaluate the incidence of adverse events in Brazilian hospitals
based on a retrospective patient record review.
Methods
Adverse event was defined as an unintended injury or harm
resulting in death, temporary or permanent disability or dys-
function, or prolonged hospital stay that arises from health
care. The study design was a retrospective patient record
review, based on the assessment tools developed by the
Canadian Adverse Event Study [3]. At our request, the
Address reprint requests to: Claudia Travassos, Avenida Brasil 4365, Manguinhos, Rio de Janeiro/RJ. 21045-900, Brazil.
Tel: +55 (21) 38653244; Fax: +55 (21) 22901696; E-mail: [email protected]
International Journal for Quality in Health Care
© The Author 2009. Published by Oxford University Press in association with the International Society for Quality in Health Care;
all rights reserved
Page 1 of 6
International Journal for Quality in Health Care 2009; pp. 1–6
10.1093/intqhc/mzp022
International Journal for Quality in Health Care Advance Access published June 23, 2009
Canadian Study team authorized us to use their assessment
instruments. The study was performed in three public teach-
ing hospitals in the State of Rio de Janeiro that provide acute
care and emergency care. Obstetric care is provided by two
of them. Teaching hospitals were selected because of their
willingness to participate and the comparatively higher quality
of their patient records.
The study population consisted of 27 350 patients admitted
in the year 2003. A random sample of patients was selected.
The sample frame excluded patients under 18 years old,
patients who stayed in the hospital less than 24 hr and cases
with psychiatric diagnoses. Unlike the Canadian study, obste-
tric cases were included in the sample due to the persistently
high maternal mortality rate in Brazil (73 maternal deaths per
100 000 live births in 2003). It must be noted that the great
majority of deliveries in Brazil occur in hospitals (97%). The
parameters used to define the sample size were based on
the Canadian study: a 10% expected incidence of patients
with adverse events (maximum absolute error 3%) and a
50% proportion of potential adverse events and, with a
significance level of 5%. The loss rate was estimated at 10%.
The rate of ineligible patients was estimated at 20%. The
final sample size was 1628 patients, with 1103 eligible for
the study.
Assessment of adverse events involved two phases. Phase
1 was an explicit patient record review by nurses to screen
for patients potentially with adverse events based on screen-
ing criteria (trigger tools). At least one screening criterion
determined that the record should be included in the second
phase review. Phase 2 was an implicit structured review by
physicians.
Reviewers were trained using standard patient records
specially selected for this purpose. Reviewers were only auth-
orized to perform fieldwork after reaching at least 80%
agreement with these patient records. All the physicians and
nurses had more than 20 years of professional experience.
The phases 1 and 2 review forms, developed by the
Canadian study, were translated and adapted to the context of
Brazilian hospitals. The forms were translated from English to
Portuguese by two different translators, followed by a compari-
son of the two versions. An expert panel decided by consensus
on the best translation of key terminology. The expert panel
consisted of nine physicians with clinical, surgical, intensive-
care, epidemiological and pharmaceutical backgrounds.
The expert panel also decided on the list of screening cri-
teria to be used, based on the Canadian study list of screen-
ing criteria. The expert panel suggested the exclusion of two,
the addition of one and the modification of five screening
criteria. Excluded criteria were: Criterion 1, ‘Unplanned hos-
pitalization (including readmission) as a result of any health
care provided during the 12 months prior to the index
admission’, and Criterion 2, ‘Unplanned admission to any
hospital during the 12 months following discharge from the
index admission’. These were judged inappropriate in the
Brazilian context, because of lack of systematic documen-
tation in the patient records about details of previous admis-
sions to the same hospital or to other hospitals. The added
criterion refers to increases in creatinine level during hospital
stay. This criterion was included to identify patients who
developed acute renal failure while in hospital. Other modifi-
cations were based on the writing for accuracy of under-
standing. The adapted forms were pre-tested and
back-translated [14]. Software was developed for data
collection.
The exclusion of the two screening criteria suggested by
the expert panel had an impact on both the incidence of
patients with adverse events and the proportion of preventa-
ble adverse events, but these differences were not statistically
significant (P . 0.05). The new criterion of ‘starting from a
normal creatinine in admission, the value increased to twice
the admission value during hospital stay’ was identified
17 times but did not change either the incidence or the pro-
portion of preventable adverse events. For purposes of com-
parison, the results presented in this article adhered strictly
to the Canadian list of screening criteria.
In the phase 2 review, the physicians first identified the
occurrence of unintentional injuries or harm. Then, they ana-
lyzed injuries or harm for any association with temporary or
permanent disability and/or prolonged hospital stay or death.
Finally, using the six-point scale from the Canadian study,
they determined whether the injury or harm was caused by
the care provided to the patient. This scale ranges from (1)
‘virtually no evidence’ to (6) ‘virtually certain evidence’. An
injury or harm is classified as an adverse event when it is
rated as 4 or more. The preventability of the adverse events
is also judged according to a six-point scale, with an adverse
event classified as preventable when rated as 4 or more.
The target patient safety measures included: incidence of
patients with adverse events (number of patients with at least
one adverse event/total number of patients); proportion of
preventable adverse events (number of patients with preven-
table adverse events/total number of patients with adverse
events), number of adverse events per 100 patients and inci-
dence density of adverse events per 100 patient-days
(number of adverse events/sum of hospital days across all
study patients).
Descriptive analysis included patient characteristics: sex
and age (age group: 18–20; 21–30; 31–40; 41–50; 51–60;
61–70 years; and 70 years and older); adverse event
classification—diagnostic error, surgical procedure, orthope-
dic care (fractures), medication, anesthesia, obstetrics, clinical
procedure, system (organization); type of error—omission
and commission; location of the adverse event—inside the
hospital (ward, surgery room, intensive care unit, emergency
room, delivery room, procedure room, service area and out-
patient service) and outside the hospital (home and other
places)—and time of occurrence and detection of the
adverse event.
Reliability in screening criteria between nurse reviewers
was measured at a significance level of 5%, using simple
agreement statistics given the small number of comparisons.
After each 10 cases reviewed, the following case was also
reviewed by a second reviewer, previously assigned as the
first reviewer’s pair for purposes of comparison. Cases for
testing inter-rater agreement were automatically selected by
the software. Simple agreement between nurse reviewers in
Mendes et al.
Page 2 of 6
assessing potential adverse events was judged as good,
although it was better between nurses A and B (80.6%; 95%
CI: 74.0–92.0) than between nurses C and D (77.8%; 95%
CI: 64.7–93.2). Inter-rater reliability was not tested in the
phase 2 review, because the assessment was performed by a
single physician.
The software generated a Microsoft Software Access data-
base, which was exported to Microsoft Software Excel
format, and the data were analyzed using Stata 10.0.
Results
A total of 1103 patients were considered eligible cases. Of
these, 888 (80.5%) were non-obstetric patients. Of the total
cases, 676 (61.3%) were female. Even excluding obstetric
cases, the majority of cases were women: 462 (52%). The
mean age of patients was 46.9 years (standard deviation 19.1)
with a median age of 46 years. The most frequent age group
was 18–30 years old: 292 (26.5%). When excluded the
obstetric cases, the predominant age group turned to be 51–
60 years old: 160 (18.0%). Inpatient mortality rate was 8.5%
(94 of 1103) (95% CI: 6.9–10.2), and there were no cases of
maternal death.
Of all the patients, 452 (41.0%; 95% CI: 38.1–43.9) had
at least one potential adverse event, i.e. met at least one posi-
tive screening criterion. When obstetric patients were
excluded, 395 patients had at least one potential adverse
event (44.5%; 95% CI: 41.2–47.8).
The total number of positive screening criteria was 834
(Table 1). The two predominant positive screening criteria
were: Criterion 18, ‘Any unwanted events not mentioned
above’ and Criterion 1, ‘Unplanned hospitalization (including
readmission) as a result of any health care provided during the
12 months prior to the index admission’. Screening Criterion
17, ‘Documentation or correspondence indicating litigation,
whether merely intent to sue or actual lawsuit’ was not
detected in any case (Table 1).
Of all the patients, 84 had at least one adverse event.
The total number of positive screening criteria among those
84 patients was 349. The three most frequent positive screen-
ing criteria in adverse events were: Criterion 3, ‘Occurrence
of injury or harm to patients during hospitalization’,
Criterion 15, ‘Hospital infections/septicemia’ and Criterion
9, ‘Other unexpected complications during index admission
which are NOT a normal development of patient’s disease or
expected result of treatment’ (Table 1).
The incidence of patients with adverse event was 7.6% (84
of 1103 patients) (95% CI: 6.0–9.2). The proportion of pre-
ventable adverse event was 66.7% (56 of 84 patients) (95%
CI: 56.4–77.0). In the 84 patients, 103 adverse events were
identified. Number of adverse events per 100 patients was
9.3 (103 of 1103) (95% CI: 7.6–11.1) and the incidence
density of adverse events was 0.80 adverse events per 100
patient-days (103of 13 563 patient-days) (95% CI: 0.60–
0.90). Among patients with adverse events, 17.9% have had
more than one adverse event. On average, among patients
with adverse event, it was observed 1.2 events per patient.
Incidence of non-obstetric patients with adverse events
was 8.6% (76 of 888 patients) (95% CI: 6.7–10.4), and the
proportion of preventable adverse events was 65.8% (50 of
76 patients) (95% CI: 54.9–76.7). Among obstetric patients,
eight experienced one or more adverse events or a 3.7%
(8 of 215 patients) incidence of obstetric patients with
adverse events (95% CI: 1.2–6.3). The proportion of pre-
ventable adverse events in these patients was 75.0% (6 of 8)
(95% CI: 36.3–113.7).
Among the 94 deaths in the study population, 32 (34.0%)
had an adverse event, and 25 (78.1%) of these were classified
as having experienced a preventable adverse event. The most
frequent class of adverse events was surgical procedures
(35.2% of all cases). Second in frequency were medical
procedures (30.6%) (Table 2). The proportion of errors by
omission was 35.0% (36 of 103), and that of errors due to
the care provided (commission) was 65.0% (67 of 103). The
highest frequency of adverse events was observed in the
wards (48.5%), followed by the operating room (34.7 %) and
intensive care unit (11.9 %) (Table 3). The most frequent
times of occurrence (85.4%; 88 of 103) and detection
(91.3%; 94 of 103) of adverse events were during admission.
Discussion
The incidence of patients with adverse events was 7.6%, and
66.7% of the adverse events were preventable. Excluding
obstetric cases, the incidence of patients with adverse events
was 8.6%. Studies focusing on health care quality improve-
ment have shown incidence similar to this study: New
Zealand (11.3%), Australia (16.6%), UK (10.8%), Denmark
(9.0%), France (14.5%), Spain (9.3%) and Canada (7.5%)
[3–9].
However, the proportion of preventable adverse events
(66.7%) observed in this Brazilian study was higher than that
observed in other studies: New Zealand (61.6%), Australia
(50%), UK (52%), Denmark (40.4%), France (27.6%) and
Spain (42.6%) [4–9]. The Canadian adverse event study [3]
showed a proportion of preventable adverse events of 36.9%
(95% CI%: 32.0–41.8). The contrast between the proportion
of preventable adverse events in Brazil and these other
countries suggests that patient safety problems may be more
frequent in Brazil than in developed countries. Moreover, the
proportion of preventable adverse events was higher among
patients who died in hospital compared with those who
survived.
Assessment of the preventable nature of unwanted out-
comes poses a major challenge, due mainly to the inherent
complexity of health care processes. There is no valid instru-
ment to unequivocally judge whether an event could have
been prevented. In general, other factors can have synergistic
or antagonistic effects. Outcome assessment is usually based
on subjective (or implicit) criteria and standards, backed by
the expertise, practical experience and clinical decision-
making of specialists [15].
As observed in other studies [3, 8, 9], 17.9% of patients
with adverse events had more than one event. However, the
Adverse events in Brazil
Page 3 of 6
incidence density of adverse events per 100 patient-days in
this Brazilian study was lower (0.80 adverse events 95% CI:
0.60–0.90) than the one observed for all hospital in the
Spanish study (1.2 adverse events), but near to the one
observed for medium-sized hospitals (0.99 adverse events
95% CI: 0.85–1.12). Possibly these differences are related to
variations in hospitals characteristics, as the length of stay
that was 12.3 days for all patients in this Brazilian study.
Extended length of stay has previously been shown to be
associated with increased adverse events [6].
The modifications suggested by the expert panel in the list
of screening criteria did not produce statistically significant
changes in the outcome measures. This finding corroborates
that it is more important to focus on the comprehensiveness
of the group of screening criteria (i.e., criteria are not
mutually exclusive) than on a single criterion’s inclusion/
exclusion. Still, non-specific criterion, such as Criterion 18,
‘Any unwanted events not mentioned above’, showed high
frequency in the study, despite representing fewer cases with
adverse events. Possible causes for the high percentage of
cases selected on the bases of unspecified criterion might be
the recurrent problem of lack of medication in hospitals in
the study and other problems related to the hospital’s
structure/system.
The fact that six out of eight cases (75%) with a principal
diagnosis related to obstetrics were preventable adverse
events highlights the relevance of including obstetric admis-
sions in studies on adverse events in Brazilian hospitals.
.............................................................................................................................................................................
Table 1 Distribution of screen-positive criteria for potential adverse events and adverse events
Screening criteriaa
Potential adverse
events, N (%)
Adverse
events, N (%)
1. Unplanned admission (including readmission) as a result of any health care
provided during the 12 months prior to the index admission
124 (14.9)
20 (5.7)
2. Unplanned admission to any hospital during the 12 months following discharge
from the index admission
334 (4.0)
4 (1.1)
3. Occurrence of injury or harm to patient during hospitalization (including any harm,
injury, or trauma occurring during index admission)
87 (10.4)
53 (15.2)
4. Adverse drug reaction
24 (2.9)
11 (3.2)
5. Unplanned transfer to intensive or semi-intensive care unita
24 (2.9)
17 (4.0)
6. Unplanned transfer from or to another acute care hospital (excluding transfers for
specialized exams, procedures, or care not available in the original hospital)a
18 (2.1)
3 (0.9)
7. Unplanned return to surgery room
20 (2.4)
17 (4.9)
8. Unplanned removal, injury, or repair of an organ or structure during surgery,
invasive procedure, or vaginal delivery
6 (0.7)
3 (0.9)
9. Other unexpected complications during index admission which are NOT a normal
development of the patient’s disease or an expected result of the treatment
109 (13.1)
50 (14.3)
10. Development of a neurological alteration absent at admission, but present at the
time of discharge from the index admission (includes neurological alterations related to
procedures, treatments, or investigations)a
16 (1.9)
14 (4.0)
11. Deatha
94 (11.3)
46 (13.2)
12. Inappropriate hospital discharge/inadequate discharge plan from index admission
(excludes unauthorized discharge)
15 (1.8)
4 (1.1)
13. Reversed cardio-respiratory arresta
19 (2.3)
10 (2.9)
14. Injury related to abortion or labor and delivery
6 (0.7)
2 (0.6)
15. Hospital infection/septicemia (excludes infections/septicemia occurring fewer than
72 hr after admission)
76 (9.1)
51 (14.6)
16. Dissatisfaction with care received as documented on patient record, or evidence of
complaint lodged (includes documents, documented complaint, conflicts between
patient/family and health care professionals, and unauthorized discharge)
22 (2.6)
4 (1.1)
17. Documentation or correspondence indicating litigation, whether merely intent to
sue or actual lawsuit
0 (0.0)
0 (0.0)
18. Any unwanted events not mentioned above
141 (16.9)
40 (11.5)
Total
834 (100.0)
349 (100.0)
aModified criteria: Criterion 5, the word ‘semi-intensive’ was included; Criterion 6, the word ‘from’ was included (this matches the definition
of the criteria in the Canadian study); Criterion 9, the word ‘unexpected’ was included (this matches the definition of the criteria in the
Canadian study); Criterion 10, neurological deficit was changed to neurological alteration; and Criterion 11, the word ‘unexpected’ was
excluded (this match the criteria in the Harvard Medical Practice study).
Mendes et al.
Page 4 of 6
In Brazilian hospitals, more adverse events were due to
commission (65%) than omission (35%) of care. This differs
from the proportion observed in the Canadian study, adverse
events due to omission was 57.1% for medicine and 50.8%
for surgery services [3].
Several methodological limitations should be mentioned.
The quality of data in the patients’ records has still not been
systematically assessed. Quality problems can cause underes-
timation of the incidence of adverse events. The preliminary
results of phase 1 predictive validity comparing trigger
assessment performed by physicians (gold standard) to nurse
assessment showed a sensitivity of 69%, suggesting that the
incidence of patients with adverse events in this study was
underestimated. Importantly, the validity of the implicit
evaluation (phase 2) depends on the physician reviewer’s
experience and knowledge.
Conclusion
The study analyzed only three hospitals, representing a
limited sample of all the hospitals in Rio de Janeiro State,
Brazil. In addition, the criteria used led to selection of the
best hospitals. Teaching hospitals can be expected to have
better patient records and better quality-of-care than the
average non-teaching hospital. However, the hospitals
included here could have a higher incidence of patients with
adverse events, because they are teaching hospitals with
higher patient acuity. The reported incidence of patients with
adverse events in teaching hospitals in other studies was
higher than in non-teaching hospitals [3]. Finally, the high
volume of obstetric cases decreased the relative proportion
of non-obstetric cases, a point to be considered in future
studies. However, this study’s findings clearly support the
inclusion of obstetric cases in countries with low-quality
maternal care.
The study showed a high proportion of preventable
adverse events in Brazilian teaching hospitals, thus emphasiz-
ing the magnitude and characteristics of patient safety pro-
blems in Brazilian hospitals in general. The adverse events
assessment tools adopted in this study can provide the basis
for the development of monitoring strategies and can be
applied in association with methods to evaluate adverse
events of specific origins. These are approaches for creating
learning opportunities and the potential for change in the
development of safer health care.
Acknowledgements
The authors wish to thank the researchers from the Canadian
Adverse Event Study, G. Ross Baker, Peter G. Norton and
Virginia Flintoft, for their help in the development of this
research in Brazil. They also wish to thank Ana Luiza Braz
Pava˜o for her assistance in the data processing.
Funding
The study was funded by the Brazilian National Research
Council, CNPq (grant 403647/2004-5 and 400162/06-7) and
the Brazilian National Health Surveillance Agency (ANVISA).
References
1. Kohn LT, Corrigan JM, Donaldson MS et al. To Err Is Human.
Washington, DC: National Academic Press, 2000.
2. Thomas EJ, Studdert DM, Burstin HR et al. Incidence and
types of adverse events and negligent care in Utah and
Colorado. Med Care 2000;38:261–71.
3. Baker GR, Norton PG, Flintoft V et al. The Canadian Adverse
Events Study: the incidence of adverse events among hospital
patients in Canada. Canadian Med Assoc J 2004;170:1678–86.
....................................................................................
Table 2 Classification of adverse events
Classification
Adverse event, N (%)
Surgical
38 (35.2)
Medical procedures
33 (30.6)
Diagnosis
11 (10.2)
Obstetric
9 (8.3)
Medication
6 (5.6)
Fractures
2 (1.9)
Anesthetic
1 (0.9)
System events
7 (6.5)
Others
1 (0.9)
Total
108 (100.0)a
aIn five adverse events, two were classified in two classes.
....................................................................................
Table 3 Percentage distribution of adverse events by location
Location
Adverse event, N (%)
Inside the hospital
Room or ward
49 (48.5)
Surgery room
35 (34.7)
ICU
12 (11.9)
Emergency room
1 (1.0)
Delivery room
1 (1.0)
Procedures room
1 (1.0)
Service area
1 (1.0)
Ambulatory
1 (1.0)
Sub total
101 (100.0)
Out of hospital
Home
2 (50.0)
Other place
2 (50.0)
Sub-total
2 (100.0)
Total
105 (100.0)b
ICU, intensive care unit.
aIn two adverse events, two locations were identified.
Adverse events in Brazil
Page 5 of 6
4. Schioler T, Lipczak H, Pedersen BL et al. Danish Adverse
Event Study, incidence of adverse events in hospitals. A retro-
spective study of medical records. Ugeskr Laeger 2002;164:
4377–9.
5. Michel P, Quenon JL, Sarasqueta AM et al. Comparison of
three methods for estimating rates of adverse events and rates
of preventable adverse events in acute care hospitals. Br Med J
2004;328:199–202.
6. Wilson RM, Runciman WB, Gibbert RW et al. The quality in
Australian health care study. Med J Aust 1995;163:458–71.
7. Davis P, Lay-Yee R, Schug S et al. Adverse events regional
feasibility study: indicative findings. N Z Med J 2001;114:
203–5.
8. Vincent C, Neale G, Woloshynowych M. Adverse events in
British hospitals: preliminary retrospective record review. Br
Med J 2001;322:517–9.
9. Aranaz-Andrés JM, Aibar-Remón C, Vitaller-Murillo J et al.
Incidence of adverse events related to health care in Spain:
results of the Spanish National Study of Adverse Events.
J Epidemiol Community Health 2008;62:1022–9.
10. Murff HJ, Patel VL, Hripcsak G et al. Detecting adverse events
for patient safety research: a review of current methodologies.
J Am Med Assoc 2003;36:131–43.
11. Marang-van de Mheen PJ, Hollander EJ, Kievit J. Effects of
study methodology on adverse outcome occurrence and mor-
tality. Int J Qual Health Care 2007;19:399–406.
12. Rozenfeld S. Agravos provocados por medicamentos em hospi-
tais do Estado do Rio de Janeiro, Brasil. Rev Saúde Pública
2007;41:108–15.
13. Daud-Gallotti R, Dutilh Novaes HM, Lorenzi MC et al.
Adverse events and death in stroke patients admitted to the
emergency department of a tertiary university hospital. Eur J
Emerg Med 2005;12:63–71.
14. Mendes W, Travassos C, Martins M et al. Adaptaça˜o dos instru-
mentos de avaliaça˜o de eventos adversos para uso em hospitais
brasileiros. Braz J Epidemiol 2008;11:55–66.
15. Donabedian A. An Introduction to Quality Assurance in Health
Care. Oxford: Oxford University Press, 2003.
Accepted for publication 28 May 2009
Mendes et al.
Page 6 of 6