Content uploaded by J. de Jonge
Author content
All content in this area was uploaded by J. de Jonge on Nov 10, 2015
Content may be subject to copyright.
Integration of prospective and
retrospective methods for risk analysis
in hospitals
M. KESSELS-HABRAKEN1, T. VAN DER SCHAAF2, J. DE JONGE1,C.RUTTE
3AND K. KERKVLIET4
1
Industrial Engineering and Innovation Sciences, Eindhoven University of Technology, The Netherlands,
2
Business Economics, Hasselt
University, Diepenbeek, Belgium,
3
Social Psychology, Tilburg University, Tilburg, The Netherlands, and
4
Hospital Pharmacy, Alysis
Hospitals, Arnhem, The Netherlands
Abstract
Objective. To explore how hospital management could gain a better picture of risks to support them in setting priorities for
patient safety.
Methods and Setting. This study deals with the combined application of prospective and retrospective methods for risk
analysis on two units of a Dutch general hospital. In the prospective analyses, employees identified and assessed possible
risks in selected processes. In the retrospective analyses, incidents were reported by employees and subsequently investigated.
The methods were integrated by using information from retrospective incident reports for prospective risk identification and
assessment, and by matching their categorization schemes. Two evaluation forms provided insight into the perceived useful-
ness of the methods and their integration.
Results and Conclusions. For both units, the prospective and retrospective analyses resulted in divergent overviews of risks
in terms of nature and magnitude, which suggests that one or both methods were subject to biases. Findings from the evalu-
ation forms showed that both methods were perceived as useful and that triangulation provided additional insight into risks.
Due to the convergent evidence, triangulation of prospective and retrospective methods can provide hospital management
and frontline staff with a more complete and less biased picture of risks. An integrative approach might be advantageous in
terms of efficiency of analysis, setting priorities for patient safety and improving the methods themselves.
Keywords: patient safety, prospective risk analysis, incident reporting, retrospective incident analysis
Introduction
Hospitals use retrospective methods to analyse errors and to
prevent their recurrence. However, the objective of minimal
patient harm [1] stresses the need to identify risks pro-
spectively and to foresee errors [2]. This is endorsed by the
requirement of the Joint Commission on Accreditation of
Healthcare Organizations (JCAHO) to conduct one prospec-
tive analysis every 18 months (The Joint Commission, 2009:
Standard LD.04.04.05). Several methods for prospective
analysis are available, such as (health-care) failure mode and
effect analysis ((H)FMEA), hazard analysis and critical
control points (HACCP) and probabilistic risk assessment
(PRA). Despite differences between these methods, such as
the consideration of combinatorial events in PRA and the
use of a decision tree in HFMEATM and HACCP, they all
aim to identify, assess and eliminate or reduce risks before
errors may occur [3 – 5].
Perfect prospective analyses would anticipate all errors and
therefore make retrospective analyses redundant [6].
However, both methods are subject to biases (see Table 1).
For instance, judgement variability could influence the
reliability of risk identification in prospective analyses [7], and
prospective risk assessments might be inaccurate due to a
lack of insight into error rates [4, 8, 9]. Retrospective incident
reporting and analysis is susceptible to problems such as
underreporting [10 – 18], incomplete data [11, 19], hindsight
and recall bias [20] and unreliable classifications [12, 14].
The question arises how to overcome those biases. Since a
‘golden standard’ is still lacking, triangulation could be the
answer for now. By using prospective and retrospective
methods, their strengths could be combined and their weak-
nesses minimized, which could yield a better picture of risks
[1, 6, 21, 22]. Recently, the National Quality Forum rec-
ommended such a combined approach to improve patient
safety [23]. But will the advantages outweigh the additional
Address reprint requests to: M. Kessels-Habraken. Tel: þ31 (0)40 848 58 68; Fax: þ31 (0)40 848 58 99;
E-mail: m.kessels@infoland.nl
International Journal for Quality in Health Care vol. 21 no. 6
#The Author 2009. Published by Oxford University Press in association with the International Society for Quality in Health Care;
all rights reserved 427
International Journal for Quality in Health Care 2009; Volume 21, Number 6: pp. 427 – 432 10.1093/intqhc/mzp043
Advance Access Publication: 14 October 2009
by guest on November 10, 2015Downloaded from
resources required to conduct two analyses instead of just
one? Probably yes, because the extra efforts could be limited
if the methods are integrated in terms of matching categoriz-
ations for risk identification and assessment. Then, efficiency
of analysis might be increased, for instance by making use of
retrospective data for the development of prospective failure
scenarios [10, 24]. Moreover, through integration the analysis
results will be directly comparable, thereby facilitating the
process of making sense of risks and determining interven-
tions [3].
In only a few prior studies have researchers concentrated
on integration of methods, for instance by using retrospective
error rates for prospective analyses [9, 25], or by comparing
prospectively and retrospectively identified causes of risks
[26]. However, those studies did not consider the perceived
usefulness of such integration. In the present study, we
examined how integration of prospective and retrospective
methods could be realized and whether it would be per-
ceived as useful. We integrated the methods by using infor-
mation from retrospective incident reports for prospective
risk identification and assessment, and by matching their cat-
egorization schemes.
Methods
Setting
The study was conducted at two units of a Dutch general
hospital. At the pharmacy a project called RISC (Risk analy-
sis by Incident reporting and Scenario analysis in the
Cytostatics dispensing process) concentrated on the process
from ordering up to and including delivering chemotherapy
drugs and archiving. A project at the nuclear medicine unit
called NUSAFE (NUclear medicine SAFE) included the
complete process from planning an examination or treatment
up to and including archiving.
Study design
The projects comprised both prospective risk analyses and
retrospective incident reporting and analysis. For both units,
the quality coordinators constructed flowcharts of the
selected processes by means of process mapping [27]; all
process steps were sequential. In the prospective analyses,
employees identified and assessed possible risks for each
process step; in the retrospective analyses, all process steps
that had contributed to the occurrence of reported incidents
were registered. During feedback sessions, the employees
were informed about preliminary results.
For a 4-month period, all 46 employees who were involved
in the selected processes were asked to report any deviation
from normal patient care. At the pharmacy, employees used
a hardcopy reporting form, while an electronic form was
used at the nuclear medicine unit. Moreover, clerical staff
from the latter unit scored each occurrence of a predefined
set of minor deviations in the subprocess of planning. For
both units, the first author together with one or more
............................................................................................................................................................................................................................................
Table 1 Possible biases of prospective risk analysis and retrospective incident reporting and analysis
Prospective risk analysis Retrospective incident reporting and analysis
Unreliable risk identification due to judgement variability during brainstorming [7] Limited number of reported incidents [10 –12], for instance due to a lack of error
recognition, a tendency to keep errors in-house, feelings of fear or shame, time
pressure and a lack of feedback [12, 14, 15, 17, 18]
Inaccurate risk assessment due to a lack of insight into error rates [4, 8, 9] Limited spectrum of reported incidents, partly due to the lack of incident reports
from doctors [12–17]
Failure to consider combinatorial events
a
[4, 8] Incomplete data for instance due to anonymity, confidentiality, shame and fear [11,
19]
Hindsight and recall bias [20]
Poor quality of classifications [12, 14]
a
PRA does explicitly consider combinatorial events.
Kessels-Habraken et al.
428
by guest on November 10, 2015Downloaded from
employees analysed the reported incidents. Information
about the incidents and the process steps involved was
registered in special databases.
Two months after the start of the incident reporting, 22 of
the 46 employees participated in the prospective analyses. For
each unit, two teams were composed, which were comparable
in terms of disciplines involved and participants’ work experi-
ence. Each team conducted a condensed version of an
HFMEATM analysis [28]. We decided to use HFMEATM
because the suggested components of a prospective analysis
as proposed by JCAHO are all part of HFMEATM (The Joint
Commission, 2009: Standard LD.04.04.05), because
HFMEATM has been applied in a diverse range of hospital
settings, and because a manual and DVD are available. The
analysis consisted of the identification of risks in the selected
processes and the assessment of their frequencies. The esti-
mated frequencies were corrected for the 4-month study
period to enable direct comparison with the incident analyses.
At each unit, one team was provided information from the
incidents database, such as the type and frequency of reported
incidents, while the other team had to rely completely on the
expertise and judgement of its team members.
We used two self-developed evaluation forms to examine
the perceived usefulness of the prospective and retrospective
methods and their integration. After the prospective analyses
had been finalized, the 22 participants received an evaluation
form (Form 1); 19 (86.4%) were completed and returned. At
the end of the project, all 46 employees received another
evaluation form (Form 2); 34 (73.9%) were completed and
returned.
Data analysis
To explore the benefit of the integration, we used chi-square
tests to compare the prospective and retrospective evalu-
ations of risks per process step. Since some expected cell
counts did not exceed the minimum level [29], Pareto ana-
lyses were used to identify those process steps that accounted
for the majority of the risks. The remaining process steps
were combined into a single category, called ‘other’. For
setting priorities and determining interventions, exact fre-
quencies might be not that important [7], as opposed to
rankings of risks. Therefore, for each analysis we ranked the
process steps in terms of the identified frequencies of risks.
Next, Spearman’s rank correlation coefficients (r
s
) were cal-
culated to explore differences between the analyses regarding
the rankings of the 10 highest risk process steps. For all stat-
istical analyses, an alpha level of 0.05 was used.
Results
We integrated prospective and retrospective methods by
using similar categorization schemes. This enabled us to
compare the analysis results directly. Tables 2 and 3 present
the results of the analyses in terms of the identified frequen-
cies of risks per process step and accompanying rankings.
For both units, the results clearly showed a lack of
congruence between prospective and retrospective analyses.
For instance, Table 2 shows that the prospective analysis
teams estimated that in a period of 4 months about 700
process deviations would occur in the process step ‘check
labels and dispensing protocol’, while in the 4-month study
period only 119 of such process deviations had been actually
identified by the retrospective analysis of reported incidents.
At the hospital pharmacy (RISC), 503 incident reports
were analysed, which revealed 1421 process deviations. When
corrected for the study period, the prospective analysis teams
predicted that risks would have resulted in 7062 and 12 654
process deviations, respectively. The frequencies of risks were
significantly different (P,0.001). At the nuclear medicine
unit (NUSAFE), 552 incident reports were analysed, which
showed 1169 process deviations. After correction for the
study period, the prospective analysis teams estimated that
risks would have caused 8677 and 4756 process deviations
to occur, respectively. Assessment of differences in those
overviews yielded a significant result (P,0.001). The signifi-
cant results for RISC and NUSAFE indicate that prospective
and retrospective analyses can result in divergent overviews
of the nature and magnitude of risks.
This finding might make it difficult for management to
determine interventions to improve patient safety. However,
for priority setting the relative magnitude of risks might be
more important than their exact frequencies [7]. Therefore,
we calculated the correlations between the rankings of the
10 process steps that were provided with the highest frequen-
cies of risks. For RISC, significant positive correlations were
found between the retrospective incident analyses and the
two prospective analyses (r
s
¼0.59, P¼0.04; r
s
¼0.79, P¼
0.001). No significant correlation was found between the two
prospective analyses (r
s
¼0.35, P¼0.24). For NUSAFE, no
significant correlations were found at all (r
s
¼0.16, P¼0.54;
r
s
¼0.18; P¼0.48; r
s
¼0.22, P¼0.39).
Although the prospective and retrospective analyses
showed a lack of congruence regarding the frequencies of
risks, the analysis of risk rankings yielded a different con-
clusion. For NUSAFE, management might still feel uncertain
about resource allocation, due to the lack of substantial con-
sensus on risk rankings. Conversely, for RISC, predictions
were supported by actual data (as reflected by the two signifi-
cant correlations). This might convince management to allo-
cate resources to the process step of entering data and
printing labels, which was identified as a high risk process
step by all analyses.
Evaluation forms
The evaluation form of the entire project (Form 2) revealed
that 33 respondents (97.1%) agreed that incident reporting
and analysis was useful for improving patient safety and opti-
mizing processes. Also, most respondents felt that prospec-
tive analysis was useful for improving patient safety (n¼26;
76.5%) and optimizing processes (n¼27; 79.4%).
Furthermore, prospective and retrospective analyses provided
insight into new risks according to 16 (47.1%) and 20
(58.8%) respondents, respectively.
Integrating risk analysis methods
429
by guest on November 10, 2015Downloaded from
Form 2 also showed that 16 respondents (47.1%) thought it
was the combination of the analyses that provided most insight
into risks. Others felt it was either the prospective (n¼3;
8.8%) or retrospective (n¼7; 20.6%) analysis that yielded most
insight. In those numbers, the participants in the prospective
analyses are included, but they also answered this question in
Form 1. Interestingly, in Form 1 a much higher percentage of
the respondents (n¼14; 73.7%) thought it was the combination
of the analyses that provided most insight into risks.
Regarding the integration of the methods, 10 participants
in the prospective analyses (52.6%) felt that information
about incidents and their frequencies was or would have
been useful; information about causes of incidents was or
would have been useful according to 11 participants (57.9%).
Form 1 also revealed that seven participants (50%, excluding
management) were more willing to report incidents after par-
ticipation in the prospective analysis. This could imply that
participation in a prospective analysis could enhance incident
reporting behaviour. For both units, follow-up chi-square
tests indicated that, after the start of the prospective analyses,
participants reported other incident types than non-
participants in terms of the subprocesses that contributed to
the occurrence of the reported incidents (P¼0.006 and P¼
0.04, respectively). This endorses the assumption that partici-
pation in a prospective analysis is positively associated with
incident reporting behaviour.
Discussion
In this study, we examined how prospective and retrospective
methods for risk analysis could be integrated and whether this
integration is perceived to be useful. Our findings show that
both methods are considered valuable in terms of improving
patient safety and optimizing processes. Our study supports
earlier findings that prospective and retrospective analyses are
partly complementary because both can yield divergent over-
views of risks in terms of nature and magnitude [6, 22]. Hence,
our study empirically endorses the theoretical contention that
due to the convergent evidence, triangulation of the methods
can provide hospital management and frontline staff with a
more complete and less biased picture of risks [1, 6, 21, 22].
Provided that risks are categorized similarly, integration of
prospective and retrospective methods enables direct com-
parison of the analysis results. Then, follow-up research
could reveal biases, whereby the methods could be further
improved [10]. Moreover, integration might limit the
additional resources that could be required due to the appli-
cation of two methods instead of just one. As we proposed,
information about incidents and their retrospectively
reported frequencies could be used as a reference point in
the prospective analyses, which might facilitate frontline staff
in the risk assessment. Conversely, prospectively developed
failure scenarios could be used as guideline for retrospective
..........................................................................................................
.......................... .......................... .............................
.............................................................................................................................................................................
Table 2 RISC: identified frequencies (freq.) of risks per process step and accompanying rankings (rank) by analysis
Analysis
Process step RIA PRA RISC 1 PRA RISC 2
Freq. Rank Freq. Rank Freq. Rank
Ordering
Fill in prescription form
a
207 2 600 1250 5
Pre-check prescription form 11 366 33
Sending
Fax prescription form to pharmacy 114 5 649 5 704
Processing
Fill in dispensing protocol 140 3 917 3 1758 4
Enter data and print labels 255 1 1109 1 2100 2
Check labels and dispensing protocol 119 4 675 4 700
Add prescription form 77 350 1834 3
Sort prescription form by date 83 284 2516 1
Dispensing
Put medication ready 74 944 2 433
Dispense chemotherapy drugs 53 375 272
Release chemotherapy drugs 56 8 333
Delivering
Transport chemotherapy drugs 66 176 167
Other 166 609 554
Total 1421 7062 12 654
Note: Frequencies (freq.) have been corrected for the study period of 4 months. Rankings (rank) are only presented for the five highest risk
process steps; all other cells are left empty. RIA, retrospective incident reporting and analysis. PRA RISC 1, prospective risk analysis
without information from the retrospective incidents database. PRA RISC 2, prospective risk analysis with information from the
retrospective incidents database.
a
Diagnosis errors have been excluded.
Kessels-Habraken et al.
430
by guest on November 10, 2015Downloaded from
incident analyses. Besides the probably consequential increase
in efficiency of analysis, integration of the methods could
also support hospital management in making sense of risks
and justifying their decisions regarding interventions [3].
Our study has several limitations. We did not test all possi-
bilities for integration. However, we purposely selected those
possibilities that could be easily applied by hospitals them-
selves to gain a better picture of risks. In future studies,
more possibilities could be tested and one could establish
whether integration actually increases efficiency of analysis.
The results of our retrospective analyses might have been
affected by hindsight bias; that is, the tendency for people to
overstate the extent to which they would have predicted
events beforehand [20]. We have tried to limit this by analys-
ing incidents as soon as possible after they had been
reported and by interviewing the people involved [30].
The perceived usefulness of the integration of prospective
and retrospective methods could be influenced by respon-
dents logically tending to evaluate the triangulation better
than the application of only one method; conversely, respon-
dents could tend to evaluate the triangulation negatively
because of the extra effort required. Since the former posi-
tively affects the perceived usefulness, while the latter nega-
tively affects it, future studies could examine whether the
perceived benefits of combining the methods actually out-
weigh the perceived drawbacks.
Similar studies should be carried out in other health-care
settings to assess the external validity of our results. However,
independent-samples t-tests and ANOVA did not reveal any
significant differences between the two units or the four pro-
spective analysis teams, which confirms our findings. Further,
our results could suggest that participation in a prospective
analysis positively influences health-care employees’ willing-
ness to report incidents. Therefore, future studies could focus
on the effects of participation in and taking notice of a pro-
spective analysis on incident reporting behaviour.
In conclusion, notwithstanding the fact that either pro-
spective or retrospective methods can be used to improve
patient safety, hospital management should seriously consider
their integration. Such an integrative approach might increase
efficiency of analysis and can yield a better picture of risks,
which could support hospital management in setting priori-
ties for patient safety and allocating resources to the most
important problems. Moreover, integration of the methods
could bring about advances in safety research by improving
the methods themselves. Together, such progress in theory
and practice could make health-care safer and reduce patient
harm accordingly.
............................................................................................................
........................... ........................... ...........................
.............................................................................................................................................................................
Table 3 NUSAFE: identified frequencies (freq.) of risks per process step and accompanying rankings (rank) by analysis
Analysis
Process step RIA PRA NUSAFE 1 PRA NUSAFE 2
Freq. Rank Freq. Rank Freq. Rank
Planning
Receive order 168 2 383 1202 1
Code order 69 291 217
Plan examination or treatment 247 1 584 5 333
Inform or instruct patient 61 180 184
Execution
Refer patient to waiting room 42 160 175
Prepare examination or treatment 90 5 95 100
Call patient and check patient data 69 2673 1 120
Select protocol and equipment 71 48 171
Carry out examination or treatment 121 3 861 4 348 3
Assess, edit and provide images 21 1814 2 50
Execution—other process steps 99 4 0 258
Reporting
Type report 10 1012 3 350 2
Archiving
Correct report 5 0 337 4.5
Send report and hardcopy 19 6 220
Archiving—other process steps 4 0 337 4.5
Other 73 570 354
Total 1169 8677 4756
Note. Frequencies (freq.) have been corrected for the study period of 4 months. Rankings (rank) are only presented for the five highest risk
process steps; all other cells are left empty. Ties have been assigned the average value of the associated ranks [29]. RIA, retrospective
incident reporting and analysis. PRA NUSAFE 1, prospective risk analysis without information from the retrospective incidents database.
PRA NUSAFE 2, prospective risk analysis with information from the retrospective incidents database.
Integrating risk analysis methods
431
by guest on November 10, 2015Downloaded from
Acknowledgements
We thank all employees from the two units for their
cooperation. In particular, we would like to thank Gonda
Nienhuis, Hanneke Stoffels and Adriaan van Sorge for their
help in the study design, data collection and data analysis.
Funding
This study was funded by Alysis hospitals and Infoland.
References
1. Battles JB, Lilford RJ. Organizing patient safety research to
identify risks and hazards. Qual Saf Health Care 2003;12:ii2– ii7.
2. Hollnagel E. Riskþbarriers¼safety? Saf Sci 2008;46:221– 9.
3. Battles JB, Dixon NM, Borotkanics RJ et al. Sensemaking of
patient safety risks and hazards. Health Serv Res
2006;41:1555– 75.
4. Marx DA, Slonim AD. Assessing patient safety risk before the
injury occurs: an introduction to sociotechnical probabilistic
risk modelling in health care. Qual Saf Health Care
2003;12:ii33– ii38.
5. McDonough JE, Solomon R, Petosa L. Quality improvement
and proactive hazard analysis models: deciphering a new tower
of Babel. In Aspden P, Corrigan JM, Wolcott J et al. (eds).
Patient Safety: Achieving a New Standard for Care. Washington, DC:
National Academies Press, 2004, 471– 508.
6. Senders JW. FMEA and RCA: the mantras of modern risk
management. Qual Saf Health Care 2004;13:249–50.
7. Bonnabry P, Cingria L, Ackermann M et al. Use of a prospec-
tive risk analysis method to improve the safety of the cancer
chemotherapy process. Int J Qual Health Care 2006;18:9 – 16.
8. Israelski EW, Muto WH. Human factors risk management in
medical products. In Carayon P (ed). Handbook of Human Factors
and Ergonomics in Health Care and Patient Safety. Mahwah, NJ:
Lawrence Erlbaum Associates, 2007, 615 – 47.
9. Trucco P, Cavallin M. A quantitative approach to clinical risk
assessment: the CREA method. Saf Sci 2006;44:491– 513.
10. Aspden P, Corrigan JM, Wolcott J et al. (eds) Patient Safety:
Achieving a New Standard for Care. Washington, DC: National
Academies Press, 2004.
11. Barach P, Small SD. Reporting and preventing medical mishaps:
lessons from non-medical near miss reporting systems. BMJ
2000;320:759– 63.
12. Evans SM, Berry JG, Smith BJ et al. Attitudes and barriers to
incident reporting: a collaborative hospital study. Qual Saf Health
Care 2006;15:39– 43.
13. Hogan H, Olsen S, Scobie S et al. What can we learn about
patient safety from information sources within an acute hospi-
tal: a step on the ladder of integrated risk management? Qual
Saf Health Care 2008;17:209– 15.
14. Johnson CW. How will we get the data and what will we do
with it then? Issues in the reporting of adverse healthcare
events. Qual Saf Health Care 2003;12:ii64 – 7.
15. Kingston MJ, Evans SM, Smith BJ et al. Attitudes of doctors
and nurses towards incident reporting: a qualitative analysis.
Med J Aust 2004;181:36 – 9.
16. Olsen S, Neale G, Schwab K et al. Hospital staff should use
more than one method to detect adverse events and potential
adverse events: incident reporting, pharmacist surveillance and
local real-time record review may all have a place. Qual Saf
Health Care 2007;16:40–4.
17. Shojania KG. The frustrating case of incident-reporting
systems. Qual Saf Health Care 2008;17:400– 2.
18. Waring JJ. Beyond blame: cultural barriers to medical incident
reporting. Soc Sci Med 2005;60:1927– 35.
19. Cannon MD, Edmondson AC. Failure to learn and learning to
fail (intelligently): how great organizations put failure to work to
innovate and improve. Long Range Plann 2005;38:299– 319.
20. Henriksen K, Kaplan H. Hindsight bias outcome knowledge
adaptive learning. Qual Safety Health Care 2003;12:ii46 – 50.
21. Herzer KR, Mark LJ, Michelson JD et al. Designing and imple-
menting a comprehensive quality and patient safety manage-
ment model: a paradigm for perioperative improvement.
J Patient Saf 2008;4:84– 92.
22. Runciman WB, Williamson JAH, Deakin A et al. An integrated
framework for safety, quality and risk management: an infor-
mation and incident management system based on a universal
patient safety classification. Qual Saf Health Care 2006;15:i82 – 90.
23. National Quality Forum. Safe Practices for Better Healthcare—2009
Update: A Consensus Report. http:// www.qualityforum.org
(30 July 2009, date last accessed).
24. Harms-Ringdahl L. Relationships between accident investigations,
risk analysis, and safety management. J Hazard Mater 2004;111:13–9.
25. Wetterneck TB, Skibinski KA, Roberts TL et al. Using failure
mode and effects analysis to plan implementation of smart I.V.
pump technology. Am J Health-Syst Pharm 2006;63:1528 – 38.
26. Van der Hoeff NWS. Theory and Practice of In-hospital
Patient Risk Management. Unpublished doctoral dissertation, Delft
University of Technology. Delft, The Netherlands, 2003.
27. Barach P, Johnson JK. Understanding the complexity of rede-
signing care around the clinical microsystem. Qual Saf Health
Care 2006;15:i10– 6.
28. Habraken MMP, Van der Schaaf TW, Leistikow IP et al.
Prospective risk analysis of health care processes: a systematic
evaluation of the use of HFMEATM in Dutch health care.
Ergonomics 2009;52:809 – 19.
29. Siegel S, Castellan NJ. Nonparametric Statistics for the Behavioral
Sciences, 2nd edn. New York: McGraw-Hill, 1988.
30. Carthey J, De Leval MR, Reason JT. The human factor in
cardiac surgery: errors and near misses in a high technology
domain. Ann Thoracic Surg 2001;72:300–5.
Accepted for publication 15 September 2009
Kessels-Habraken et al.
432
by guest on November 10, 2015Downloaded from