Professional Documents
Culture Documents
Swift 4
Swift 4
Measurements and Main Results: Baseline (n = 1,906) and Keywords: care transitions; readmissions; risk stratification;
implementation (n = 1,938) cohorts differed with respect to quality
(Received in original form December 9, 2013; accepted in final form March 12, 2014 )
Author Contributions: V.H., A.A., O.G., B.W.P., and C.J.F. contributed to the study’s conception, design, implementation and data gathering. R.K. and S.C.
were responsible for data analysis and interpretation. U.R.O. and S.C. were responsible for drafting the manuscript. V.H., O.G., B.W.P., and C.J.F. critically
revised the article. All eight authors assisted in the subsequent revisions and have read and approved of the final manuscript.
Correspondence and requests for reprints should be addressed to Uchenna R. Ofoma, M.D., Division of Critical Care Medicine, Geisinger Medical Center, 100
North Academy Avenue, Danville, PA 17822. E-mail: uofoma@geisinger.edu
This article has an online supplement, which is accessible from this issue’s table of contents online at www.atsjournals.org
Ann Am Thorac Soc Vol 11, No 5, pp 737–743, Jun 2014
Copyright © 2014 by the American Thoracic Society
DOI: 10.1513/AnnalsATS.201312-436OC
Internet address: www.atsjournals.org
Unplanned readmissions to the intensive costs (1, 2). There is growing concern admission. Broad guidelines have been
care unit (ICU) are associated with that early readmissions to the ICU may published regarding appropriate ICU
increased length of stay, mortality, and indicate premature discharge from index discharge (3). However, decisions about
Ofoma, Chandra, Kashyap, et al.: Readmission Prediction Tool in ICU Discharge Workflow 737
ORIGINAL RESEARCH
which patients are ready to be discharged and nurses received education from Mayo (17). METRIC Data Mart contains a near-
from the ICU are often based on the quality coaches regarding the prognostic real-time copy of pertinent ICU patient
subjective intuition of the clinician (4, 5). performance of the score and about its physiologic monitoring data, medication
Physicians’ approach toward decision- incorporation into the rounding workflow. orders, laboratory and radiologic
making at the time of ICU discharge can The SWIFT score for each patient was investigations, physician and nursing
be further influenced by a variety of factors calculated with an automatic web-based notes, and respiratory therapy data,
not related to patient physiology and acuity tool at 6:45 A.M. every day, based on the which were used for automatic SWIFT
of illness. These include the availability most recently available values of SWIFT score calculation. Access to the database
of beds, staffing, and workload pressure components (15). If a laboratory value is accomplished through open database
(6–8). was not obtained for a given patient, connectivity.
It has also been suggested that severity then it was assumed to be normal or
of illness and prognostic scoring systems, noncontributory. The charge nurse was Subjects
as adjuncts to clinical judgment, may instructed to provide the bedside nurse and A consecutive cohort of adult (aged 18 yr
better identify patients ready for discharge the consulting physician with a copy of or older) patients discharged from the
from the ICU (3, 9). To this effect, the SWIFT score for all patients in the medical ICU during the study period
multiple scoring systems have been ICU every morning on rounds. However, was enrolled. Patients who were
proposed (10–12). However, there are no for purposes of implementation, SWIFT discharged to a comfort/palliative care
prospective studies of their implementation discussion was limited to patients facility or to another hospital or those
in clinical practice, and it remains concerning whom a discharge decision transferred to another ICU were excluded
unknown whether such implementation had been made on a specific rounding day. from the study analysis. Only index
can improve processes and unexpected Criteria to conclude that a physician had discharges and readmissions were
outcomes of care for patients after ICU decided to discharge a patient included studied.
discharge. the discussion of possible discharge,
In the present study, we incorporated followed by an agreement and explicit
the Stability and Workload Index for instruction by the consultant to the house Sample Size
Transfer (SWIFT) score into the usual staff to discharge. After a decision to With a baseline composite 7-day
discharge rounds workflow of the same discharge the patient from the unit readmission rate of the study ICU of
ICU where it was initially derived and (without having seen the score) was made, 8.5% (11), and with 1,000 patients before
validated and observed how the SWIFT the patient’s nurse revealed the SWIFT and 1,000 patients after the intervention,
score influenced physician discharge score to the physician team. The patient’s the study was calculated to have a power
decision-making. We also measured nurse was instructed to encourage the of 0.84 to detect a 3.5% decrease in
readmission rates after SWIFT introduction bedside providers to comment on the composite readmission rate, using a one-
and compared them with baseline pre- discharge plan after revelation of the sided chi-square comparison and assuming
implementation rates. Some of the results of SWIFT score. The nurse then recorded 50% compliance with the intervention.
this study have been previously reported in the outcome of the discussion, using the In choosing a one-sided test, we
the form of an abstract (13). study reporting tool. When the SWIFT hypothesized that introduction of the
score was not calculated because of SWIFT score would lead to more
limitations of the automatic tool, (e.g., conservative discharge strategies and
Methods for patients who were not physically located thereby lower readmission rates.
in the ICU at 6:45 A.M.), no “SWIFT
Study Design and Settings discussion” took place and these patients The SWIFT Score
The study was conducted in an adult were analyzed as such. Plan-Do-Study-Act The SWIFT score (11) is a previously
medical ICU at the Mayo Clinic in (PDSA) cycles were performed on validated tool that predicts the likelihood
Rochester, Minnesota, from December 2007 a quarterly basis during the implementation of ICU readmission within 24 hours of
to December 2009. The characteristics of period to continually assess the impact discharge for individual patients, based
the ICU have been previously published of our change. on a number of easily collected parameters
(14). The institutional review board available at the point of ICU discharge.
approved the study protocol and waived Electronic Resources SWIFT score parameters include source
the need for informed consent. The current Mayo Clinic Electronic of ICU admission, ICU length of stay, last
A baseline (December 2007–November Medical Record (EMR) fits the definition measured PaO2/FIO2 (fraction of inspired
2008) period of usual discharge activity of a comprehensive EMR (16). A oxygen) ratio, Glasgow Coma Scale score
was followed by an interventional comprehensive EMR is defined as having at time of ICU discharge, and last arterial
(December 2008–November 2009) period. 24 key functions in all clinical units. The blood gas PaCO2 (Table 1). The minimum
As part of a quality initiative to reduce ICU Critical Care Independent Multidisciplinary and maximum possible scores are 0 and
readmission rates, an automatically Program of the hospital has an established 64, respectively. A SWIFT score value
calculated SWIFT score was provided as near-real-time relational database, called greater than 15 has a sensitivity and
part of the nursing and attending rounding METRIC Data Mart. Details of METRIC specificity of 50% and 85%, respectively,
flow sheets, as well as to the charge nurse Data Mart and the Mayo Clinic EMR for predicting readmission within 24 hours
for the duration of the study. Physicians have been previously published elsewhere of ICU discharge.
Ofoma, Chandra, Kashyap, et al.: Readmission Prediction Tool in ICU Discharge Workflow 739
ORIGINAL RESEARCH
Resource Utilization
The observed measures of resource
Figure 1. Flow diagram. *Transferred to one of nine specialized care units at the Mayo Clinic utilization before and after implementation
Rochester, based on need for specialized care. ICU = intensive care unit. of the SWIFT score are outlined in Table 2.
Compared with the baseline period, the
cohort from the SWIFT implementation
period had decreased ICU length of stay
implementation rate was 25%. Rates of with 15 or less) was significantly associated (2.05 vs. 2.19 d; P = 0.04), decreased rate of
discussion of SWIFT score varied over the with physician behavior to change noninvasive ventilation use (14.6 vs. 17.6%;
study duration (P , 0.001), with a peak discharge plan (discharge to a monitored P = 0.01), fewer days of noninvasive
in the second quarter (Table 3). Discussion setting or postpone discharge; P , 0.001). ventilation (0.15 vs. 0.20; P = 0.007), and
of the SWIFT score at the time of discharge marginally decreased number of days of
led to changes in discharge plan (change Readmission Rates invasive ventilation (0.53 vs. 0.63; P = 0.06).
from intended general care unit discharge In the implementation cohort, 26.2% of There was no difference in the proportion
destination to a monitored unit or decision subjects had SWIFT scores greater than of patients using invasive ventilation (19.8
to keep in ICU for longer) for 12.6% of 15 and thus were predicted to have a higher vs. 21.1%; P = 0.29). These differences were
patients and to discharge with enhanced risk of unplanned readmissions. The largely attenuated in subgroup APACHE
verbal communication with the receiving number and rate of observed 24-hour III–matched secondary analysis using the
services for 17.7% discharges (Figure 2). readmissions were 36 (1.9%) during the 356 implementation cohort subjects
A SWIFT score of more than 15 (compared baseline period and 47 (2.4%) during the concerning whom the SWIFT score was
discussed and comparing them with
subjects from the baseline cohort. There
Table 2. Characteristics, resource utilization measures, and readmission rates in were also no differences in 24-hour and
baseline and SWIFT implementation cohorts 7-day readmission rates in this matched
comparison (Table 4).
Characteristic Baseline SWIFT P Value
(n = 1,906) (n = 1,938)
Table 3. Compliance with SWIFT implementation and observed changes in provider Given the observed compliance rates,
discharge decisions the effective study sample likely was
underpowered to detect any true difference
Variable First Second Third Fourth in readmission rates.
Quarter Quarter Quarter Quarter Apart from low adoption rates, other
factors could also potentially explain why
Total discharges 504 451 468 515 our intervention did not impact on
SWIFT score . 15* 140 (30.0) 126 (31.6) 122 (29.8) 120 (25.8) readmission rates. It is possible that the
SWIFT score discussed† 86 (17.1) 119 (26.4) 77 (16.5) 74 (14.4) SWIFT score may not adequately capture
Change in discharge plan‡ 13 (15.1) 17 (14.3) 8 (10.4) 7 (9.4)
Monitored setting 5 (5.8) 13 (10.9) 7 (9.1) 4 (5.4)
all factors that are critical to preventing
Discharge postponed 8 (9.3) 5 (4.2) 1 (1.3) 2 (2.7) readmissions. The derivative components of
Enhanced communication‡ 20 (23.3) 22 (18.5) 10 (13.0) 11 (14.9) the SWIFT score are composed mostly of
Readmissions within 24 h† 17 (3.4) 6 (1.3) 16 (3.4) 8 (1.6) patient-centered factors, and data from
Table E1 in the online supplement, showing
Definition of abbreviation: SWIFT = Stability and Work Load Index for Transfer.
*SWIFT score was calculated for 1,738 discharges. Quarterly breakdown not shown. association between scores greater than 15
†
Numbers expressed as percentage of total discharges. and increased resource utilization, suggest
‡
Numbers expressed as percentage of total cases in which SWIFT was discussed. that the SWIFT score can be considered
a marker for severity of illness. Apart from
We observed a SWIFT implementation SWIFT data. As illustrated in Table 3 severity of illness and other patient-related
rate of 18% of all discharges and 25% of (footnotes), the electronic SWIFT score, factors (11, 19–21), studies also suggest that
discharges with SWIFT score above 15. Such which was computed at 6:45 A.M. daily, there may be a relationship between ICU
low compliance rates are reflective of was available for only 1,738 of 1,938 readmissions and physician- or hospital-
a failure in change management, given that discharges. Second, the SWIFT score lacked related factors such as patient inflow
this implementation was performed in the precision for individual patients and volumes (22), ICU occupancy (23), and
same ICU where the SWIFT score was despite efforts at provider education, it decision-making practices (24). These
initially derived and validated. Coiera (18) often did not make sense to rotating ICU factors are not part of the SWIFT score
outlined four major causes for the failure providers and its implementation did not calculation.
of decision support systems to be used fit naturally into the established routine Another potential limitation of this
clinically. Several of these factors process of care. These factors were study is that the design makes it difficult to
may partly explain why the SWIFT identified in our PDSA assessments as directly link SWIFT introduction with
implementation rate at our study ICU barriers to full-fledged adoption and outcomes. For example, we cannot attribute
was low. First, we depended totally on an probably led to a waning of enthusiasm the significantly lower rates of resource
electronic system to calculate and supply with the implementation over time. utilization during the implementation
Figure 2. SWIFT implementation and provider discharge decision flow chart. *Subgroup utilized for post hoc APACHE-matched comparisons. †Includes
200 subjects for whom the SWIFT score was not available at discharge. APACHE = Acute Physiology and Chronic Health Evaluation; ICU = intensive care
unit; SWIFT = Stability and Work Load Index for Transfer.
Ofoma, Chandra, Kashyap, et al.: Readmission Prediction Tool in ICU Discharge Workflow 741
ORIGINAL RESEARCH
Table 4. Readmission rates and resource utilization in APACHE-matched subgroups power to provide clinicians with confidence
to use the results to guide decision-making,
Variable Before* After† Difference and the need to be able to modify user
(n = 93) (n = 93) (P Value) behavior while improving patient outcomes
(27). Further studies addressing our
Readmission within 24 h 3 (3.2) 0 (0) 0.25 observed and perceived barriers to
Readmission within 7 d 8 (8.6) 6 (6.5) 0.8 adoption and other study limitations are
Invasive ventilation use, n (%) 25 (26.9) 25 (26.9) 1 needed to further elucidate any potential
Invasive ventilation, d‡ 0.85 (2.4) 0.67 (1.6) 0.9 impact of readmission predictive tools
Noninvasive ventilation use, n (%) 21 (22.6) 16 (17.2) 0.5
Noninvasive ventilation, d‡ 0.16 (0.4) 0.11 (0.4) 0.3 on readmission rates and other ICU
Hospital length of stay, d 6.3 6.1 0.68 outcomes.
ICU length of stay, d 1 2 0.12
Definition of abbreviations: APACHE = Acute Physiology and Chronic Health Evaluation; ICU =
intensive care unit. Conclusions
*APACHE-matched subgroup of baseline cohort (exact APACHE III scores).
†
Subgroup with SWIFT score discussed during implementation (n = 356).
‡
Median values of 0, thus mean and SD reported. Readmission prediction tools can help
physicians optimize the timing of ICU
discharge. This study suggests that physician
period to the SWIFT introduction. Also, our The Society of Critical Care Medicine decision-making at the time of patient
study collected data on a limited number has recommended using the unplanned discharge is potentially modifiable by one of
of baseline characteristics and adjusted ICU readmission rate within 48 hours such tools. However, introducing such tools
only for severity of illness. The study could as a clinical performance measure (25). into the discharge workflow may present
have been strengthened by the addition However, in order for readmission rates change management challenges that limit
of more characteristics of the study cohorts, to be a viable quality measure, physicians the evaluation of the impact of such tools on
which could have allowed for better must be able to alter the rate (26). readmission rates and other relevant ICU
adjustment for potential differences in Theoretically, one of the proposed ways outcomes. n
patient demographics using logistic that they can do this is by the use of risk
regression analysis. This becomes more stratification and clinical prediction tools Author disclosures are available with the text of
important, given that matching based on such as the SWIFT score as an adjunct this article at www.atsjournals.org.
exact APACHE III scores markedly reduced to clinical judgment, thereby potentially
the sample available for adjustment analysis. making safer ICU discharge decisions. Acknowledgment: The authors thank Andrew
Also, the overall quality improvement Standards for the evaluation of C. Hanson (Mayo Clinic Department of
atmosphere during the implementation a clinical prediction tool such as the SWIFT Biomedical Statistics and Informatics) for his work
with SWIFT tool programming, as well as the staff
may have contaminated the discharge score include the need to be validated to and nurse quality coaches of the 6 Mary Brigh
decision-making process, even when the provide evidence of reproducible accuracy, Medical Intensive Care Unit at the Mayo Clinic
SWIFT score was not discussed. the need to have sufficient predictive Rochester.
References 9 Zimmerman JE, Wagner DP, Draper EA, Knaus WA. Improving
intensive care unit discharge decisions: supplementing physician
1 Rosenberg AL, Hofer TP, Hayward RA, Strachan C, Watts CM. Who judgment with predictions of next day risk for life support. Crit Care
bounces back? Physiologic and other predictors of intensive care Med 1994;22:1373–1384.
unit readmission. Crit Care Med 2001;29:511–518. 10 Frost SA, Tam V, Alexandrou E, Hunt L, Salamonson Y, Davidson PM,
2 Rosenberg AL, Watts C. Patients readmitted to ICUs: a systematic Parr MJ, Hillman KM. Readmission to intensive care: development
review of risk factors and outcomes. Chest 2000;118:492– of a nomogram for individualising risk. Crit Care Resusc 2010;12:
502. 83–89.
3 Task Force of the American College of Critical Care Medicine, Society 11 Gajic O, Malinchoc M, Comfere TB, Harris MR, Achouiti A, Yilmaz M,
Schultz MJ, Hubmayr RD, Afessa B, Farmer JC. The Stability and
of Critical Care Medicine. Guidelines for intensive care unit
Workload Index for Transfer score predicts unplanned intensive care
admission, discharge, and triage. Crit Care Med 1999;27:633–638.
unit patient readmission: initial development and validation. Crit Care
4 Shear LM, Steingrub JS, Teres D, Shear L. ICU patient triage ranking:
Med 2008;36:676–682.
a flawed practice. Crit Care Med 1988;16:409.
12 Ouanes I, Schwebel C, Français A, Bruel C, Philippart F, Roucaud N,
5 Heidegger CP, Treggiari MM, Romand JA. A nationwide survey of
Soufir L, Adrie C, Garrouste-Orgeas M, Timsit J-F, et al. Prediction of
intensive care unit discharge practices. Intensive Care Med 2005;31: early readmission after a stay in the ICU: validation of the SWIFT
1676–1682. score and construction of the FAIR score in a multicenter cohort.
6 Strosberg MA. Intensive care units in the triage mode: an organizational Intensive Care Med 2009;35:296.
perspective. Hosp Health Serv Adm 1991;36:95–109. 13 Ofoma U, Kashyap R, Daniels C, Gajic O, Pickering B, Farmer C. Does
7 Sax FL, Charlson ME. Utilization of critical care units: a prospective implementation of a previously validated prediction tool reduce
study of physician triage and patient outcome. Arch Intern Med readmission rates into a medical intensive care unit? [abstract].
1987;147:929–934. Chest 2012;142:278A.
8 Goldfrad C, Rowan K. Consequences of discharges from intensive care 14 Afessa B, Keegan MT, Hubmayr RD, Naessens JM, Gajic O, Long
at night. Lancet 2000;355:1138–1142. KH, Peters SG. Evaluating the performance of an institution
using an intensive care unit benchmark. Mayo Clin Proc 2005;80: 21 Ho KM, Dobb GJ, Lee KY, Finn J, Knuiman M, Webb SA. The effect of
174–180. comorbidities on risk of intensive care readmission during the same
15 Chandra S, Agarwal D, Hanson A, Farmer JC, Pickering BW, Gajic O, hospitalization: a linked data cohort study. J Crit Care 2009;24:101–107.
Herasevich V. The use of an electronic medical record based 22 Baker DR, Pronovost PJ, Morlock LL, Geocadin RG, Holzmueller CG.
automatic calculation tool to quantify risk of unplanned readmission Patient flow variability and unplanned readmissions to an intensive
to the intensive care unit: a validation study. J Crit Care 2011;26:634. care unit. Crit Care Med 2009;37:2882–2887.
e9–634.e15. 23 Chrusch CA, Olafson KP, McMillan PM, Roberts DE, Gray PR. High
16 Jha AK, DesRoches CM, Campbell EG, Donelan K, Rao SR, Ferris TG, occupancy increases the risk of early death or readmission after
Shield A, Rosenbaum S, Blumenthal D. Use of electronic health transfer from intensive care. Crit Care Med 2009;37:2753–2758.
records in U.S. hospitals. N Engl J Med 2009;360:1628–1638. 24 Yoon KB, Koh SO, Han DW, Kang OC. Discharge decision-making by
17 Herasevich V, Pickering BW, Dong Y, Peters SG, Gajic O. Informatics intensivists on readmission to the intensive care unit. Yonsei Med J
infrastructure for syndrome surveillance, decision support, reporting, 2004;45:193–198.
and modeling of critical illness. Mayo Clin Proc 2010;85:247–254. 25 Society of Critical Care Medicine Quality Indicators Committee.
18 Coiera E. Guide to health informatics, 2nd ed. London: Hodder Arnold; Candidate critical care quality indicators. Anaheim, CA: Society of
2003. pp. xxvii, 440. Critical Care Medicine; 1995.
19 Frost SA, Alexandrou E, Bogdanovski T, Salamonson Y, Davidson PM, 26 Kramer AA, Higgins TL, Zimmerman JE. Intensive care unit
Parr MJ, Hillman KM. Severity of illness and risk of readmission to readmissions in US hospitals: patient characteristics, risk factors,
intensive care: a meta-analysis. Resuscitation 2009;80:505–510. and outcomes. Crit Care Med 2012;40:3–10.
20 Rosenberg AL, Hofer TP, Hayward RA, Strachan C, Watts CM. Who 27 Laupacis A, Sekar N, Stiell IG. Clinical prediction rules: a review and
bounces back? Physiologic and other predictors of intensive care suggested modifications of methodological standards. JAMA 1997;
unit readmission. Crit Care Med 2001;29:511–518. 277:488–494.
Ofoma, Chandra, Kashyap, et al.: Readmission Prediction Tool in ICU Discharge Workflow 743
Reproduced with permission of the copyright owner. Further reproduction prohibited without
permission.