Nursing workload, patient safety incidents and mortality: An observational study from Finland

Abstract

Objective
To investigate whether the daily workload per nurse (Oulu Patient Classification (OPCq)/nurse) as measured by the RAFAELA system correlates with different types of patient safety incidents and with patient mortality, and to compare the results with regressions based on the standard patients/nurse measure.

Setting
We obtained data from 36 units from four Finnish hospitals. One was a tertiary acute care hospital, and the three others were secondary acute care hospitals.

Participants
Patients’ nursing intensity (249 123 classifications), nursing resources, patient safety incidents and patient mortality were collected on a daily basis during 1 year, corresponding to 12 475 data points. Associations between OPC/nurse and patient safety incidents or mortality were estimated using unadjusted logistic regression models, and models that adjusted for ward-specific effects, and effects of day of the week, holiday and season.

Primary and secondary outcome measures
Main outcome measures were patient safety incidents and death of a patient.

Results
When OPC/nurse was above the assumed optimal level, the adjusted odds for a patient safety incident were 1.24 (95% CI 1.08 to 1.42) that of the assumed optimal level, and 0.79 (95% CI 0.67 to 0.93) if it was below the assumed optimal level. Corresponding estimates for patient mortality were 1.43 (95% CI 1.18 to 1.73) and 0.78 (95% CI 0.60 to 1.00), respectively. As compared with the patients/nurse classification, models estimated on basis of the RAFAELA classification system generally provided larger effect sizes, greater statistical power and better model fit, although the difference was not very large. Net benefits as calculated on the basis of decision analysis did not provide any clear evidence on which measure to prefer.

Conclusions
We have demonstrated an association between daily workload per nurse and patient safety incidents and mortality. Current findings need to be replicated by future studies.

Full text

1
FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
Nursing workload, patient safety
incidents and mortality: an
observational study from Finland
Lisbeth Fagerström,1,2 Marina Kinnunen,3 Jan Saarela1
To cite: FagerströmL,
KinnunenM, SaarelaJ.
Nursing workload, patient
safety incidents and mortality:
an observational study
from Finland. BMJ Open
2018;8:e016367. doi:10.1136/
bmjopen-2017-016367
►Prepublication history and
additional material for this
paper are available online. To
view these les, please visit
the journal online (http:// dx. doi.
org/ 10. 1136/ bmjopen- 2017-
016367).
Received 12 March 2017
Revised 31 January 2018
Accepted 26 March 2018
1Faculty of Education and
Welfare Studies, Åbo Akademi
University, Vaasa, Finland
2Faculty of Health and Social
Sciences, University College of
Southeast Norway, Drammen,
Norway
3Vaasa Central Hospital, Vaasa,
Finland
Correspondence to
Professor Lisbeth Fagerström;
lisbeth. fagerstrom@ abo. 
Research
ABSTRACT
Objective To investigate whether the daily workload
per nurse (Oulu Patient Classication (OPCq)/nurse) as
measured by the RAFAELA system correlates with different
types of patient safety incidents and with patient mortality,
and to compare the results with regressions based on the
standard patients/nurse measure.
Setting We obtained data from 36 units from four Finnish
hospitals. One was a tertiary acute care hospital, and the
three others were secondary acute care hospitals.
Participants Patients’ nursing intensity (249 123
classications), nursing resources, patient safety incidents
and patient mortality were collected on a daily basis during
1 year, corresponding to 12 475 data points. Associations
between OPC/nurse and patient safety incidents or
mortality were estimated using unadjusted logistic
regression models, and models that adjusted for ward-
specic effects, and effects of day of the week, holiday
and season.
Primary and secondary outcome measures Main
outcome measures were patient safety incidents and
death of a patient.
Results When OPC/nurse was above the assumed optimal
level, the adjusted odds for a patient safety incident were
1.24 (95% CI 1.08 to 1.42) that of the assumed optimal
level, and 0.79 (95% CI 0.67 to 0.93) if it was below
the assumed optimal level. Corresponding estimates for
patient mortality were 1.43 (95% CI 1.18 to 1.73) and 0.78
(95% CI 0.60 to 1.00), respectively. As compared with the
patients/nurse classication, models estimated on basis
of the RAFAELA classication system generally provided
larger effect sizes, greater statistical power and better
model t, although the difference was not very large. Net
benets as calculated on the basis of decision analysis did
not provide any clear evidence on which measure to prefer.
Conclusions We have demonstrated an association
between daily workload per nurse and patient safety
incidents and mortality. Current ndings need to be
replicated by future studies.
INTRODUCTION
Many studies have shown that insufficient
nurse staffing in hospital-based care negatively
affects outcomes such as mortality, infections
and failure to rescue.1–6 However, the results
are inconsistent and indicate a complex and
non-linear relationship between nursing
workload (NWL), mortality and other patient
outcomes.7–12 The strength of the evidence
underpinning the association between nurse
staffing and outcomes in previous studies
can be challenged. Poor research designs,
measurement problems and/or imprecise
data that do not take into account daily vari-
ations in patients’ care needs may contribute
to the mixed findings.8 Higher nurse staffing
and richer skill mix are associated with
improved patient outcomes.4 8 10 Therefore,
higher ratios have been recommended for
improving patient safety and outcomes.1 9
However, it is difficult to set fixed, standard
patient-to-nurse ratios for units in acute care
hospitals, as evidenced in systematic reviews
and other studies.7 10 13–15 Staffing levels
must instead match patients’ nursing care
needs.8 16 17
In an attempt to accommodate some of
these issues, the RAFAELA patient classifi-
cation system was developed in the 1990s in
Finland.16 18 19 As compared with most other
patient classification systems that use fixed
patient-to-nurse ratios, the RAFAELA system
use daily data on patients’ care needs and the
workload per nurse. The main purpose of the
RAFAELA system is to ensure an appropriate
allocation of nurse staff resources and, thus,
a preferable NWL, which has been labelled
as an optimal NWL. The latter term refers to
Strengths and limitations of this study
►The study is the rst to assess the relationship
between nursing workload and patient outcomes
based on data obtained on a daily basis.
►The instrument used here takes patient characteris-
tics, such as age, sex and diagnoses, into account.
►The study provides some evidence to suggest that
the traditional nurse stafng measure, the patients-
to-nurse ratio, may partly fail to control for patient
severity and casemix.
►The study does not address the potential inuence
of skillmix, competence level, work experience or
the professionals’ patient-related direct time.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

2FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
a situation when patients’ care needs are assumed to be
in balance with the nursing resources, and that working
conditions can be assumed as being favourable, most
desirable or satisfactory for the realisation of good nursing
care.16 18–22 While certain realities such as economic
restraints cannot be disregarded, the intention with the
RAFAELA system is to provide a NWL measure dedicated
to the reduction or elimination of adverse events.
In the RAFAELA system, NWL is based on daily assess-
ments of patients’ care needs and the registration of the
nursing staff resources.16 The PAONCIL method (‘Profes-
sional Assessment of Optimal Nursing Care Intensity
Level’) is used to establish an assumed optimal NWL
for a specific ward. Daily measurements of NWL (Oulu
Patient Classification (OPC)/nurse) are subsequently
compared with this level, and resources are considered to
be appropriately allocated when the actual NWL is at this
level.19 21 This would mean that a satisfactory number of
nurses, neither too many nor too few, are being allocated
to provide care for the actual patient group.
We have found only two studies8 18 on the relation-
ship between NWL based on assessed requirements for
care (as opposed to nurse patient ratios or equivalent
measures) and patient outcomes. Needleman et al8 found
a significant association between patient mortality and
increased exposure to unit shifts when nurse staffing was
below the target level. In a recent study by Junttila et al18
based on monthly means, the incidence rate of death
when average daily NWL was above the assumed optimal
level was 13-fold that when the average daily NWL was
below this level. However, to our knowledge, no studies
exist on this relationship using daily-level data.
The aim of this observational study was therefore to
investigate whether the daily workload per nurse (OPC/
nurse), as a measure based on the RAFAELA system,
correlates with patient safety incidents and patient
mortality, using data collected on a daily basis. In addi-
tion, we wanted to compare the estimates with those
based on the standard patients-to-nurse ratio (patients/
nurse).
METHODS
Study setting
We obtained data from 36 units from four Finnish hospi-
tals. One (A, 9 units) is a tertiary acute care hospital,
whereas the three others (B, 14 units; C, 4 units; D, 9 units)
are secondary acute care hospitals. The following special-
ties were included in the data material: internal medi-
cine (eight units), surgical (eight units), paediatrics (five
units), gynaecology (four units), maternity (two units),
neurology (two units), orthopaedics (two units), oncology
(one unit), rehabilitation (one unit), lung (one unit) and
otology (one unit). Inclusion criteria were daily use of the
RAFAELA system according to standards, reliable nursing
intensity data as expressed in terms of a yearly reliability
test done by parallel classifications (requirement is that
unanimity is over 70%), and applicable nursing intensity
level measured with the PAONCIL method.16 19–21 Units
that had undergone major organisational changes over
the previous year were excluded. The A and B data
represent the period 1 January to 31 December 2012,
and the C and D data represent the period 1 January to
31 December 2013.
We did not include any sensitive health-related data of
patients in the study, or any information regarding char-
acteristics of the nurses. The RAFAELA is owned by the
Association of Finnish Local and Regional Association
Authorities and governed by non-commercial Finnish
Consulting Group.
Measurement of NWL in the RAFAELA nursing intensity and
stafng system
The RAFAELA is a standardised, person-centred, evidence-
based system for nurse staffing that was developed in the
1990s.16 19 The feasibility, validity and reliability of the
RAFAELA have been tested with good results.16 18 21 22
It is now used in about 90% of the hospitals in Finland,
and has lately been implemented in Iceland, the Nether-
lands, Sweden and Norway.22 A requirement for users of
the RAFAELA system is that the inter-rater reliability for
nursing intensity measurements should be tested yearly.
The daily nursing intensity of each unit is assessed by
all the responsible registered nurses on each day. One
registered nurse usually classify one to six patients per
day. The assessment is done every day by classifying each
patient’s care needs by the OPC instrument. This instru-
ment consists of six subareas of nursing care. The nursing
intensity level varies from 6 to 24 points for an individual
patient per calendar day.16 19 The nurses’ workload is
calculated by dividing the total amount of nursing inten-
sity points on the unit, for example, 350, with the number
of nurses who take care of patients, for example, 12,
during the same 24 hours. In this example, the patient-re-
lated NWL will then be 29.2 OPC points per nurse (here-
after referred to as OPC/nurse).
The underlying assumption of the RAFAELA system is
that the nature and characteristics of nursing care differ
between wards. The recommended NWL of each ward
therefore has to be determined by the PAONCIL method.
The development, testing and description of this method
has been reported in several publications.16 19–22 Thus,
the method is used to assess each ward’s recommended
optimal NWL including various contextual and organi-
sational factors.21 The recommendation is that this level
has to be reassessed by conducting the PAONCIL study
every second year. The ones used in this study were not
older than 2 or 3 years. The basic idea of the RAFAELA
system is that the observed NWL (eg, 29.2 points/nurse)
is compared with the established preferred for the same
unit (eg, 22–30 points/nurse). If the observed NWL
lies within the established limits, the nursing intensity is
considered to be at the assumed optimal level.
The data we use in this paper consist of daily measure-
ments based on the RAFAELA system.19 They correspond
to every admitted patient’s nursing intensity during 1 year
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

3
FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
and were based on 249 123 classifications of patients’
nursing intensity (OPC classifications). Each day, the
patient-related nurse resources were also recorded, using
a standardised model where non-patient-related time was
excluded. Apart from each day’s staff data (OPC/patient,
OPC/nurse, etc), there was daily information also on
patient incidents and patient mortality.
All data were collected during a period of 1 year,
meaning that there were 12 475 data points (not approx-
imately 13 140, since some wards were closed for shorter
periods, foremost because of holidays). Table 1 provides
the central variables of the data in terms of each unit’s
PAONCIL level, daily mean number of classified patients,
daily mean number of OPC classifications, total OPC
points, nursing staff resources, number of patients per
nurse, OPC points per nurse, incidents and deaths.
Outcomes
Data on incidents were collected daily from The Reporting
System for Safety Incidents in Health Care Organizations
(HaiPro), which is a comprehensive and standardised
patient safety system in Finland.23 24 As defined by HaiPro,
an incident is a safety hazard that may harm or harm the
patient. Incidents are classified into 14 categories,24 but
there are two main categories: near miss, which may
have caused harm to the patient, but was prevented
by chance or by timely preventive actions and adverse
events, which are negative events that caused harm to
the patient. To roughly capture the severity of an event,
we categorised incidents in four ways1: whether at least
one incident, of any type, occurred (incident),2 whether
a patient was affected to any degree (patient affected),3
whether the incident caused harm to the patient (harm
to patient), and4 whether there was more than one inci-
dent, of any type, on the same day (>1 incident), within
the available follow-up of 365 days. In addition, we used
patient mortality (death) as a fifth type of adverse event.
Some wards had no deaths during the study period, but
excluding them from the analyses would not affect the
results to any noteworthy degree. The mortality data
were retrieved from the local mortality register of each
hospital.
Statistical analyses
Using logistic regression analyses, associations were esti-
mated on the daily level between nursing intensity per
nurse (OPC/nurse) in relation to the assumed optimal
level and each type of outcome, that is, each of the four
types of incidents and mortality. For each of type of
outcome, the event was coded as 0 or 1, meaning that
either there was no event during that specific day, or
there was an event. The use of logistic regression models
accommodate any issues related to non-normal distri-
butions. We estimated associations both in unadjusted
models and in models that adjusted for ward-specific
effects and effects of day of the weak, holiday and season,
using dummy variables. Thus, we allowed for heteroge-
neity in the intercept term, which was motivated by the
fact that across-ward variability was fairly modest. The
categories of the variables are described in the footnotes
of table 2. Parallel analyses were performed with the stan-
dard measure of NWL, patients/nurse. Supplementary
electronic files provide full details of the models esti-
mated (see online supplementary file 1).
We report results in which evaluations based on the
RAFAELA system (OPC/nurse) were assessed using the
assumed optimal level with a ±15% deviation around this
point,16 19 21 and in which the patients/nurse measure was
assessed using a categorisation with three equally large
groups. The results reported in table 2 were consequently
based on 20 different regressions. Model fit indices (−2
log likelihood, Akaike information criterion and Nagelk-
erke’s R2) are provided to facilitate comparisons between
regressions based on the OPC/nurse measure and the
patients/nurse measure. The analyses were performed
using SPSS V.21. All estimates are expressed in terms of
ORs with 95% CIs.
Apart from comparing the predictive accuracy of
the models that use the OPC/nurse measure and the
patients/nurse measure, respectively, we have also used
decision-analytic methods.25 These ascertain the value
of prediction models by incorporating information on
consequences and they require explicit valuation of
outcomes. The technique may thus help in deciding on
which measure to prefer, that is, the one with a higher
net benefit.
RESULTS
When OPC/nurse was above the assumed optimal level,
the unadjusted odds for a patient safety incident were
1.28 (95% CI 1.13 to 1.45) that of the assumed optimum
level (see table 2). Corresponding ORs for the other types
of incidents, patient affected, harm to patient and >1 inci-
dent were 1.13 (95% CI 0.96 to 1.32), 1.16 (95% CI 0.93
to 1.45) and 1.25 (95% CI 0.95 to 1.66), respectively. ORs
for patient mortality were even higher or 1.42 (95% CI
1.19 to 1.69). If OPC/nurse was below the recommended
optimal level, the ORs for incidents and patient mortality
were conversely lower or around 0.67 for the different
types of incidents, and 0.55 for patient mortality.
When ward-specific effects and effects of day of the week,
holiday and season were adjusted for, the ORs diminished
somewhat (see table 2). NWL above the assumed optimal
level was associated with 8%–34% higher odds of an inci-
dent, depending on the type of incident, and 43% higher
odds of patient mortality, as compared with if it was at
the assumed optimal level. If OPC/nurse was below this
level, the OR for an incident and for patient mortality
was approximately 25% lower. Adding the ward-specific
effects improved model fit considerably. Also the variables
for weekday, holiday and season improved the model
fit, except for the outcomes >1 incident and death. The
odds for incidents were in general least likely to occur on
Saturdays and on holidays, whereas there were no obvious
seasonal effects (not shown here). Complete descriptions
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

4FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
Table 1 Assumed optimal workload, and daily mean values of workload, staff resources and adverse events per ward in the data
Optimal
load,
lower
bound
Optimal
load,
upper
bound
OPC
per
nurse
Patients
per
nurse OPC
OPC
per
patient
Number of
nurses Incident
Patient
affected
Harm
to patient
>1
incident Death n
Number
of
OPC
classifications
Ward, id
D Infectious* 18.90 25.56 21.06 1.48 251.22 14.20 12.03 0.09 0.06 0.02 0.01 0.04 365 17.70
D Internal† 21.85 29.55 24.45 1.60 308.81 15.28 12.74 0.13 0.07 0.04 0.02 0.07 362 20.18
D Cardiology 21.88 29.60 25.98 1.89 355.99 13.75 13.98 0.26 0.11 0.08 0.03 0.08 365 25.90
D Surgical 16.85 22.79 23.88 1.41 425.36 16.87 17.80 0.05 0.02 0.01 0.00 0.08 365 25.19
D Orthopaedics 15.99 21.63 21.92 1.41 350.84 15.52 16.15 0.12 0.08 0.04 0.02 0.05 365 22.62
D Surgical 14.54 19.66 17.58 1.24 243.46 14.18 13.75 0.07 0.05 0.04 0.01 0.02 330 17.15
D Maternal 25.47 34.47 33.83 2.48 343.10 13.62 10.12 0.04 0.02 0.01 0.00 0.00 364 25.17
D Oncology 14.36 20.12 22.84 1.38 249.46 16.50 10.88 0.26 0.11 0.05 0.03 0.18 362 15.10
D Neurology 20.05 27.13 25.12 1.34 392.29 18.69 15.76 0.13 0.05 0.03 0.02 0.08 365 20.96
B Internal† 16.14 21.83 20.40 1.49 182.62 13.75 8.95 0.12 0.09 0.07 0.02 0.06 341 13.29
B Internal† 20.37 27.56 25.13 2.00 290.05 12.61 11.59 0.23 0.16 0.10 0.06 0.08 365 23.04
B Internal† 17.16 23.21 19.94 1.47 171.98 13.54 8.59 0.20 0.11 0.07 0.06 0.05 365 12.69
B Internal† 19.65 26.59 24.16 1.82 292.34 13.24 12.11 0.11 0.07 0.05 0.01 0.13 365 22.09
B Surgical 21.47 29.04 29.16 2.00 544.95 14.59 18.73 0.20 0.16 0.10 0.01 0.16 365 37.34
B Surgical 18.50 25.10 23.29 1.77 355.67 13.18 15.37 0.08 0.05 0.03 0.02 0.08 365 26.95
B Surgical 23.27 31.48 28.04 1.88 474.66 14.90 17.01 0.12 0.08 0.01 0.02 0.05 365 31.85
B Gynaecology 18.87 25.54 23.17 1.68 131.87 14.04 6.21 0.13 0.10 0.02 0.02 0.02 177 9.45
B Maternity 19.05 25.78 20.69 1.37 338.39 15.04 16.40 0.03 0.02 0.02 0.00 0.00 365 22.48
B Paediatrics 11.51 15.58 13.08 0.75 134.87 17.39 10.23 0.04 0.03 0.01 0.00 0.00 362 7.72
B Paediatrics 12.16 16.45 8.65 0.49 108.10 17.52 12.38 0.02 0.01 0.00 0.00 0.00 365 6.17
B Otology 17.45 23.60 22.45 1.81 224.67 12.36 9.87 0.11 0.07 0.04 0.01 0.01 298 18.12
B Neurology 16.25 21.98 19.37 1.37 245.29 14.24 12.68 0.08 0.06 0.01 0.01 0.07 365 17.30
B Lung 20.02 27.09 19.43 1.50 273.30 12.97 14.05 0.10 0.08 0.01 0.02 0.11 365 21.05
C Surgical 22.60 30.60 24.33 1.41 321.42 17.29 13.22 0.17 0.08 0.03 0.07 0.06 365 18.58
C Internal† 18.70 25.30 25.15 1.65 397.05 15.20 15.93 0.17 0.09 0.01 0.05 0.15 366 26.12
C Internal† 20.40 27.60 21.77 1.29 208.62 16.87 9.52 0.10 0.09 0.01 0.02 0.03 291 12.33
C Surgical 19.40 26.30 23.70 1.51 228.64 15.64 9.49 0.04 0.03 0.00 0.01 0.00 291 14.58
A Gynaecology 19.50 26.40 24.92 1.83 290.61 13.60 11.39 0.03 0.02 0.01 0.00 0.00 366 21.14
A Gynaecology 25.60 42.10 39.84 2.81 510.25 14.21 12.83 0.04 0.03 0.02 0.00 0.00 225 35.94
A Gynaecology 33.50 45.30 40.54 2.85 607.06 14.22 15.06 0.04 0.02 0.02 0.00 0.00 366 42.64
A Orthopaedics 16.80 22.70 20.29 1.33 430.29 15.18 21.19 0.10 0.05 0.02 0.02 0.04 366 28.30
A Paediatrics 12.30 14.90 13.33 0.97 165.18 13.85 12.34 0.15 0.10 0.08 0.04 0.00 366 11.95
Continued
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

5
FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
of all estimates and the models estimated, with predictive
indices, can be found in the supplementary electronic file
(see online supplementary file 1).
The two lower panels in table 2 provide results of
parallel analyses when NWL was measured according to
the standard patients-to-nurse ratio (patients/nurse). As
compared with results based on the RAFAELA system,
there are three main issues to be pointed out. First,
effects sizes in terms of ORs were consistently smaller with
the patients/nurse approach than with the OPC/nurse
approach, irrespective if unadjusted or adjusted models
are compared. For instance, in the fully adjusted model,
the OR of an incident was 1.13 (95% CI 0.96 to 1.33) if
workload was in the highest one-third, and 0.89 (95% CI
0.75 to 1.05) if it was in the lowest one-third, as compared
with if it was in the middle one-third. These effects were
notably smaller than the estimated relative effect sizes
for being above and below the recommended optimum
according to the RAFAELA system, which were 1.24
(95% CI 1.08 to 1.42) and 0.79 (95% CI 0.67 to 0.93),
respectively. Second, in almost all instances, the estimates
of the patients/nurse approach had smaller statistical
power in terms of wider CIs (ie, larger SEs). However,
far from all estimates for the OPC/nurse measure, or for
the patients/nurse measure, were statistically significant
at the 5% level. Third, when comparing results for the
patients/nurse measure to the OPC/nurse measure for
otherwise similar models and outcomes, the model fit
of the former was consistently poorer (values of the log
likelihood and AIC were higher and R square lower). It
nevertheless needs to be stressed that the difference was
not very large.
We experimented also with other ways to catego-
rise NWL. For OPC/nurse, we used an alternative with
a halved deviation from the recommended optimal
point, that is, ±7.5% instead of ±15%, and with a doubled
deviation, that is, ±30% from the optimal point. The
patient-to-nurse measure was also assessed using alterna-
tive categorisations, such as five and seven equally large
groups, respectively. Results of these additional regres-
sions supported the overall conclusions as reported above.
In models using the patients/nurse measure, associations
were mostly weaker, came with lower statistical power, and
they were less systematic, as compared with models based
on the OPC/nurse measure (see online supplementary
file 1).
Hence, our analyses suggest that, in terms of predictive
accuracy, models estimated on basis of NWL according to
the RAFAELA system are slightly to be preferred above
otherwise similar models that use the standard patients/
nurse measure. It is not evident, however, which measure
is to be preferred when it comes to decision-making.
Figures 1–5 summarise net benefit values calculated
based on the models estimated for each type of patient
safety incident and patient mortality, respectively; see
Vickers and Elkin25 for technical details. The values have
been calculated over a reasonable range for the proba-
bility of an event (type of incident or mortality). Models
Optimal
load,
lower
bound
Optimal
load,
upper
bound
OPC
per
nurse
Patients
per
nurse OPC
OPC
per
patient
Number of
nurses Incident
Patient
affected
Harm
to patient
>1
incident Death n
Number
of
OPC
classifications
A Paediatrics 11.00 14.20 11.16 0.70 117.12 16.05 10.57 0.09 0.06 0.02 0.01 0.01 360 7.31
A Paediatrics 8.90 12.00 9.29 0.54 91.18 17.36 9.67 0.09 0.06 0.02 0.01 0.00 363 5.24
A Surgical 21.10 28.50 24.45 1.48 273.20 16.50 11.24 0.02 0.02 0.01 0.00 0.06 364 16.54
ARehabilitation 15.10 20.50 20.57 1.49 198.98 13.89 9.51 0.14 0.12 0.03 0.04 0.00 315 14.52
Total 18.52 25.22 22.41 1.53 294.59 14.97 12.93 0.11 0.07 0.03 0.02 0.05 12 475 19.97
Assumed optimal load refers to the established interval of optimal workload according to the PAONCILmeasurement (OPC per nurse).
The variables are described in more detail in the main text.
*Infectious diseases.
†Internal medicine.
OPC, Oulu Patient Classication; PAONCIL, Professional Assessment of Optimal Nursing Care Intensity Level.
Table 1 Continued
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

6FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
based on the OPC/nurse measure and the patients/nurse
measure are to be compared by looking at the net benefit
values (see figures 1–5). The one with higher net benefit
values is to be preferred above the other. As shown by the
figures, there is no clear discrepancy. For some threshold
probabilities, the OPC/nurse measure lies above the
patients/nurse measure, while for others, the situation is
the opposite. In addition, for each event (type of incident
and mortality), the two curves are essentially overlapping,
and in most instances the difference in net benefit values
is rather modest. In terms of the magnitude of the benefit
for patients, it is consequently not evident which measure
of NWL is to be preferred.
DISCUSSION
We find that the odds for a patient safety incident were
10%–30% higher, and for patient mortality about 40%
Table 2 OR for an adverse event (with 95% CI) for four types of patient safety incidents and for patient mortality, according
to NWL measurement by the RAFAELA system (OPC/nurse) and the standard NWL measurement system (patients/nurse),
unadjusted and adjusted estimates
Incident Patient affected Harm to patient >1 incident Death
OPC/nurse, unadjusted model
Below optimum 0.67 (0.58–0.78) 0.68 (0.56–0.82) 0.66 (0.50–0.88) 0.67 (0.47–0.95) 0.55 (0.43–0.70)
At optimum 1 1 1 1 1
Above optimum 1.28 (1.13–1.45) 1.13 (0.96–1.32) 1.16 (0.93–1.45) 1.25 (0.95–1.66) 1.42 (1.19–1.69)
−2 log likelihood 8577.5 6169.3 3523.0 2406.4 4958.6
Akaikeinformation criterion 8561.5 6173.3 3527.0 2410.4 4962.6
Nagelkerke’sR20.0106 0.0056 0.0052 0.0056 0.0160
OPC/nurse, adjusted model
Below optimum 0.79 (0.67–0.93) 0.78 (0.64–0.96) 0.85 (0.63–1.14) 0.73 (0.50–1.07) 0.78 (0.60–1.00)
At optimum 1 1 1 1 1
Above optimum 1.24 (1.08–1.42) 1.08 (0.91–1.28) 1.11 (0.88–1.41) 1.32 (0.98–1.79) 1.43 (1.18–1.73)
−2 log likelihood 8010.8 5856.3 3211.1 2187.9 4286.5
Akaikeinformation criterion 8106.8 5952.3 3307.1 2283.9 4382.5
Nagelkerke’s R20.0960 0.0688 0.1050 0.1041 0.1733
Patients/nurse, unadjusted model
First group 0.74 (0.64–0.86) 0.85 (0.71–1.02) 0.79 (0.61–1.04) 0.80 (0.58–1.10) 0.47 (0.38–0.58)
Second group 1 1 1 1 1
Third group 1.09 (0.95–1.25) 1.18 (0.99–1.41) 1.24 (0.96–1.58) 0.95 (0.70–1.30) 0.97 (0.81–1.17)
−2 log likelihood 8589.1 6180.9 3525.1 2416.5 4958.8
Akaikeinformation criterion 8593.1 6184.9 3529.1 2420.5 4962.8
Nagelkerke’s R20.0055 0.0033 0.0045 0.0010 0.0159
Patients/nurse, adjusted model
Firstgroup 0.89 (0.75–1.05) 0.98 (0.80–1.21) 0.90 (0.66–1.23) 1.01 (0.71–1.44) 0.86 (0.68–1.08)
Second group 1 1 1 1 1
Third group 1.13 (0.96–1.33) 1.15 (0.94–1.41) 1.03 (0.77–1.39) 1.15 (0.81–1.64) 1.20 (0.97–1.49)
−2 log likelihood 8029.8 5863.1 3213.4 2196.1 4301.8
Akaikeinformation criterion 8.125.8 5959.1 3309.4 2292.1 4397.8
Nagelkerke’s R20.0931 0.0674 0.1043 0.1004 0.1698
Number of events 1367 848 400 246 636
The table summarises results from 20 different models estimated on 12 475 calendar days, representing 36 wards at 4 hospital units.
Adjusted model refers to models adjusted for ward-specic effects and effects of the of the week, holiday and season.
Estimates for ward-specic effects and effects of day of the week, holiday and season are found in the onlinesupplementary electronic les.
At optimum refers to the assumed optimal nursing intensity point with±15% deviation, as dened by the RAFAELA system.
Patients/nurse refer to a categorisation into three equally large groups.
Categories used for day of the week are Monday, Tuesday, Wednesday, Thursday, Friday, Saturday and Sunday.
Categories used for holiday are No or Yes, where Yes refers to Easter, Midsummer, Christmas and New Year.
Categories used for season are January–March, April–May, June–August, September–October and November–December.
NWL, nursing workload; OPC, Oulu Patient Classication.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

7
FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
higher, if the NWL as measured by the RAFAELA system
(OPC/nurse) was above the assumed optimal level, as
compared with if it was at this level. If OPC/nurse was
below the level, the odds for a patient safety incident and
for mortality were approximately 25% lower. The latter
situation would mean that nurses have more time for
caring and observing each patient, which may reduce
the risk for adverse events and accordingly prevent the
patient’s health condition from deteriorating.
Previous research9 did not find significant changes
in patient safety associated with decreased NWL and
could not confirm compliance with ratios per shift.
Other studies used hospital-level administrative data that
imprecisely allocated staffing to patients’ care needs.8 11
We think that such associations between nurse staffing,
patient outcomes and mortality may be challenged.12 18
Needleman et al8 found similar results between mortality
and day-to-day, shift-to-shift variation in staffing, and
Junttila et al18 between mortality and days with NWL
over optimal level on a monthly level. The OPC/nurse
measure is more detailed than the traditional patients/
nurse measure. While comparable to the ‘hours per
patient day model’,26 its accuracy of nursing resources
is higher. For example, if a nurse becomes sick during a
Figure 1 Decision curves for incident according to the OPC/nurse measure and patients/nurse measure,
respectively.OPC,Oulu Patient Classication.
Figure 2 Decision curves for patient affected according to the OPC/nurse measure and patients/nurse measure,
respectively.OPC, Oulu Patient Classication.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

8FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
shift and leaves the unit, the nurse in charge will deduct
these hours from the unit’s resources.
Several factors affect the reporting of incidents, for
example, staff’s lack of motivation or knowledge, nurse
staff shortage, stressful situations or burn-out. A reason-
able argument is therefore that a very high NWL indi-
cates a working situation where the nurse staff resources
are too low. Still, too few resources can result in the depri-
oritisation of the registration of adverse events and thus
the under-reporting of incidents connected to high NWL,
which may affect the results of our study and the conclu-
sions that we draw.
Our study provided results based on daily measures of
all-in-hospital patients’ actual nursing intensity, including
detailed registration of used staff resources and the associa-
tion with incidents and mortality on daily levels. The HaiPro
database, on which our analyses were based, meets WHO
criteria for a good reporting system.23 24 However, we know
that despite a good reporting system, incidents reports are
missing due to several reasons, such as lack of time, person-
nel’s involvement etc. The Global Trigger Tool is another
method to analyse patient safety, which has been recom-
mended.27 However, it collects triggers and patient safety inci-
dents from treatment periods, not on a daily basis, whereas
Figure 3 Decision curves for harm to patient according to the OPC/nurse measure and patients/nurse measure,
respectively.OPC, Oulu Patient Classication.
Figure 4 Decision curves for>1 incident according to the OPC/nurse measure and patients/nurse measure, respectively.OPC,
Oulu Patient Classication.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

9
FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
data on incidents collected from HaiPro can be targeted to
certain days.24 Units that underwent major organisational
changes over the previous year were excluded from our
study, because they may negatively influence the data quality
including incident reports. The accuracy of the data used,
in terms of NWL, incidents and mortality, is highly reliable
and probably better than in previous studies on NWL and
adverse outcomes. The staffing measurement determined
by the RAFAELA system implicitly considers specific charac-
teristics of each ward, such as organisational factors in terms
of unit size, leadership and physical environment.16 19 21
We found evidence that a staffing measure based on
daily measurement of individual patient care needs and
the recommended NWL (OPC/nurse) is slightly better
in predicting incidents and mortality rates as compared
the standard patient-to-nurse measure. Yet it needs to
be stressed that, based on decision curve analysis, it was
not clear which measure of NWL will produce higher
net benefit in terms of avoiding patient safety incidents
and patient mortality. Current findings therefore ought
to be further investigated and the findings replicated in
larger, longitudinal multicentre studies.
A strength of this study is that the analyses were
conducted based on nurses’ independent classifica-
tions of patients’ nursing intensity. The data used was
based on a scientifically tested NWL system, which
enabled comparisons,16 since the patient-casemix and
patient severity groups require different staff resources
to maximise positive patient outcomes.4 8 18 28 29 NWL
consequently ought to be monitored daily using reliable
instruments to ensure good patient outcomes. Such
optimal resource allocation is needed for successful
leadership and clinical governance, and it is crucial for
favourable outcomes, to preventing adverse events and
to reducing patient mortality.
Our study nevertheless has certain limitations. The
reliability of incident reports can always be ques-
tioned, despite that the HaiPro system has been in
systematic use for almost 10 years. Although we could
control for ward-specific effects and effects of day of
the week, holiday and season, there might be other
confounding factors. Hospital settings are character-
ised by complexity regarding factors that may affect
total NWL.1 2 13 28–31 While a list of central organisa-
tional and contextual factors were included in the
PAONCIL instrument, we did not address the effects
of skill mix, competence level or work experience on
patient outcomes. Physicians’ patient-related direct
time and healthcare support should also probably be
included in further studies.32 Further analyses of other
patient characteristics, such as age, sex or diagnoses,
were not conducted because the OPC instrument takes
these variables into account. Earlier studies have shown
that the OPC instrument identifies patients’ individual
characteristics such as functional ability, symptoms
of diseases and the effect on nursing intensity of the
most central patient characteristics.16 22 Hence, the
measurement by the OPC covers the actual patient
casemix for each day. However, the contribution of
these aspects, especially age and sex, may be analysed
in more details in further studies. Another limitation
was that a death or an incident caused by low staffing
on a ward on 1 day may not always occur on that same
day or at that same ward. This could be explored by
analysing patient records around the critical days and
at multiple wards. Although this study was the first
about the relationship between the assumed optimal
NWL and daily outcomes, a multicentre study with
several hospitals is needed to further test the general-
isability of the results.
Figure 5 Decision curves for death according to the OPC/nurse measure and patients/nurse measure, respectively.OPC,
Oulu Patient Classication.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from

10 FagerströmL, etal. BMJ Open 2018;8:e016367. doi:10.1136/bmjopen-2017-016367
Open Access
CONCLUSIONS
This study has showed that a work situation above the
assumed optimal level increases the risk for adverse events
and patient mortality. However, the resources for nursing
staff are limited in all organisations. Nurse managers, there-
fore, have to use available resources in the most optimal way.
This study provided some new evidence to suggest that the
traditional nurse staffing method, the patient-to-nurse ratio,
is not necessarily preferable when it comes to controlling for
patients’ severity and casemix. The staffing measure based
on the assumed optimal NWL may therefore be consid-
ered a novel attempt to fill a gap in the existing knowledge
on leadership and clinical governance. Efficient resource
allocation is needed for successful leadership and clinical
governance and it is crucial for favourable outcomes, for
preventing adverse events and for reducing the mortality
risk. Future research is needed to ascertain whether good
patient outcomes are ensured by daily monitoring of nurses’
workload with instruments like the one studied here.
Contributors LF did the literature search. LF and JS designed the study. LF
collected the data. JS prepared the data and performed the analyses. LF, MK and JS
contributed to data interpretation, writing and revision of the report.
Funding The authors disclose receipt of the following nancial support for the
research and authorship for this article: The State Research Funding of Vaasa
Hospital District based on the Health Care Act (1326/2010).
Competing interests None declared.
Patient consent Not required.
Ethics approval This study received approval from the chief administrative
physicians of all four hospitals involved. No further ethical approval was therefore
necessary, which is in accordance with the regulatory regime for conducting health
research in Finland.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Full descriptions of all models estimated and their
estimates are found in the supplementary electronic les.
Open Access This is an Open Access article distributed in accordance with the
Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which
permits others to distribute, remix, adapt, build upon this work non-commercially,
and license their derivative works on different terms, provided the original work is
properly cited and the use is non-commercial. See: http:// creativecommons. org/
licenses/ by- nc/ 4. 0/
© Article author(s) (or their employer(s) unless otherwise stated in the text of the
article) 2018. All rights reserved. No commercial use is permitted unless otherwise
expressly granted.
REFERENCES
1. Aiken LH, Clarke SP, Sloane DM, et al. Hospital nurse stafng and
patient mortality, nurse burnout, and job dissatisfaction. JAMA
2002;288:1987–93.
2. Aiken LH, Cimiotti JP, Sloane DM, et al. Effects of nurse stafng and
nurse education on patient deaths in hospitals with different nurse
work environments. Med Care 2011;49:1047–53.
3. Aiken LH, Sloane DM, Bruyneel L, et al. Nurse stafng and education
and hospital mortality in nine European countries: a retrospective
observational study. Lancet 2014;383:1824–30.
4. Seago JA, Williamson A, Atwood C. Longitudinal analyses of nurse
stafng and patient outcomes: more about failure to rescue. J Nurs
Adm 2006;36:13–21.
5. Rafferty AM, Clarke SP, Coles J, et al. Outcomes of variation in hospital
nurse stafng in English hospitals: cross-sectional analysis of survey
data and discharge records. Int J Nurs Stud 2007;44:175–82.
6. Zingg W, Holmes A, Dettenkofer M, et al. Hospital organisation,
management, and structure for prevention of health-care-associated
infection: a systematic review and expert consensus. Lancet Infect
Dis 2015;15:212–24.
7. Lang TA, Hodge M, Olson V, et al. Nurse-patient ratios: a systematic
review on the effects of nurse stafng on patient, nurse employee,
and hospital outcomes. J Nurs Adm 2004;34:326–37.
8. Needleman J, Buerhaus P, Pankratz VS, et al. Nurse Stafng
and Inpatient Hospital Mortality. N Engl J Med Overseas Ed
2011;364:1037–45.
9. Donaldson N, Bolton LB, Aydin C, et al. Impact of California's
licensed nurse-patient ratios on unit-level nurse stafng and patient
outcomes. Policy Polit Nurs Pract 2005;6:198–210.
10. Kane RL, Shamliyan TA, Mueller C, et al. The association of
registered nurse stafng levels and patient outcomes: systematic
review and meta-analysis. Med Care 2007;45:1195–04.
11. Shuldham C, Parkin C, Firouzi A, et al. The relationship between
nurse stafng and patient outcomes: a case study. Int J Nurs Stud
2009;46:986–92.
12. Van den Heede K, Sermeus W, Diya L, et al. Nurse stafng and
patient outcomes in Belgian acute hospitals: cross-sectional analysis
of administrative data. Int J Nurs Stud 2009;46:928–39.
13. Tarnow-Mordi WO, Hau C, Warden A, et al. Hospital mortality in
relation to staff workload: a 4-year study in an adult intensive-care
unit. Lancet 2000;356:185–9.
14. Lang TA, Hodge M, Olson V, et al. Nurse-patient ratios: a systematic
review on the effects of nurse stafng on patient, nurse employee,
and hospital outcomes. J Nurs Adm 2004;34:163–74.
15. National Institute for Health and Care Excellence. Safe Stafng for
nursing in adult inpatient wards in acute hospitals. Safe stafng
guideline 1. 2014 http//: www. nice. org. uk/ sg1
16. Rauhala A. The validity and feasibility of measurement tools for
human resources management in nursing case of the RAFAELA
system: Department of Health Policy and Management, University of
Kuopio, 2008. Doctoral dissertation.
17. Adomat R, Hewison A. Assessing patient category/dependence
systems for determining the nurse/patient ratio in ICU and HDU: a
review of approaches. J Nurs Manag 2004;12:299–308.
18. Junttila JK, Koivu A, Fagerström L, et al. Hospital mortality and
optimality of nursing workload: A study on the predictive validity of
the RAFAELA Nursing Intensity and Stafng system. Int J Nurs Stud
2016;60:46–53.
19. Rauhala A, Fagerström L. Determining optimal nursing intensity: the
RAFAELA method. J Adv Nurs 2004;45:351–9.
20. Rauhala A, Kivimäki M, Fagerström L, et al. What degree of work
overload is likely to cause increased sickness absenteeism among
nurses? Evidence from the RAFAELA patient classication system. J
Adv Nurs 2007;57:286–95.
21. Fagerström L, Rainio AK, Rauhala A, et al. Professional assessment
of optimal nursing care intensity level. A new method for resource
allocation as an alternative to classical time studies. Scand J Caring
Sci 2000;14:97–104.
22. Andersen MH, Lønning K, Bjørnelv GM, et al. Nursing intensity and
costs of nurse stafng demonstrated by the RAFAELA system: liver
vs. kidney transplant recipients. J Nurs Manag 2016;24:798–805.
23. Kinnunen M. Learning from errors in organization – Factors
preventing or making it possible: University of Vaasa,Acta
Wasaensia:230. Doctoral thesis.
24. Holmström A-R. Learning from Medication errors in healthcare –
how to make medication error reporting systems work? Doctoral
dissertation: Faculty of Pharmacy, University of Helsinki, 2017.
25. Vickers AJ, Elkin EB. Decision curve analysis: a novel method for
evaluating prediction models. Med Decis Making 2006;26:565–74.
26. Spetz J, Donaldson N, Aydin C, et al. How many nurses per patient?
Measurements of nurse stafng in health services research. Health
Serv Res 2008;43:1674–92.
27. Classen DC, Resar R, Grifn F, et al. 'Global trigger tool' shows that
adverse events in hospitals may be ten times greater than previously
measured. Health Aff 2011;30:581–9.
28. Needleman J, Buerhaus P, Mattke S, et al. Nurse-stafng levels and the
quality of care in hospitals. N Engl J Med 2002;346:1715–22.
29. Grifths P, Ball J, Drennan J, et al. Nurse stafng and patient
outcomes: Strengths and limitations of the evidence to inform
policy and practice. A review and discussion paper based on
evidence reviewed for the National Institute for Health and Care
Excellence Safe Stafng guideline development. Int J Nurs Stud
2016;63:213–25.
30. Fasoli DR, Fincke BG, Haddock KS. Going beyond patient
classication systems to create an evidence-based stafng
methodology. J Nurs Adm 2011;41:434–9.
31. Cho SH, Hwang JH, Kim J. Nurse stafng and patient mortality in
intensive care units. Nurs Res 2008;57:322–30.
32. Kerlin MP, Small DS, Cooney E, et al. A randomized trial of
nighttime physician stafng in an intensive care unit. N Engl J Med
2013;368:2201–9.
on 24 April 2018 by guest. Protected by copyright.http://bmjopen.bmj.com/BMJ Open: first published as 10.1136/bmjopen-2017-016367 on 24 April 2018. Downloaded from