Figure. The proportion of a session used by a patient discussion declines as its relative position in the session increases (N = 262 patient discussions within 23 sessions). The trend line estimated with linear regression is y = 0.108 + 0.037 x (P < .005) (eTable 2).
Cohen MD, Ilan R, Garrett L, LeBaron C, Christianson MK. The earlier the longer: disproportionate time is spent on patients discussed early in attending physician handoffs. Arch Intern Med.. Published online November 12, 2012. doi:10.1001/2013.jamainternmed.65.
eTable 1. Characteristics of participating ICU attending physicians
eTable 2. Analysis of variance for trend line of Figure 1
eTable 3. Analysis of variance for linear approximation of response surface
eTable 4. Summary statistics for linear mixed model analysis
eFigure 1. Histogram of 100,000 mean tau values, randomly generated using exact structure of the 23 session data set
eFigure 2. Histogram of 100,000 median tau values, randomly generated using exact structure of the 23 session data set
eFigure 3. Antitonic regression for 1 session of size 6
eFigure 4. Antitonic regression for 3 sessions of size 8
eFigure 5. Antitonic regression for 3 sessions of size 9
eFigure 6. Antitonic regression for 4 sessions of size 10
eFigure 7. Antitonic regression for 5 sessions of size 11
eFigure 8. Antitonic regression for 1 session of size 12
eFigure 9. Antitonic regression for 2 sessions of size 13
eFigure 10. Antitonic regression for 1 session of size 15
eFigure 11. Antitonic regression for 1 session of size 16
eFigure 12. Antitonic regression for 1 session of size 18
eFigure 13. Antitonic regression for 1 session of size 23
eFigure 14. Share of session estimated as linear function of session size, ordinal position, and their interaction
Cohen MD, Ilan R, Garrett L, LeBaron C, Christianson MK. The Earlier the Longer: Disproportionate Time Allocated to Patients Discussed Early in Attending Physician Handoff Sessions. Arch Intern Med. 2012;172(22):1762-1764. doi:10.1001/2013.jamainternmed.65
Author Affiliations: School of Information, University of Michigan, Ann Arbor (Dr Cohen); Department of Medicine and Critical Care Program, Kingston General Hospital, Kingston, Ontario, Canada (Dr Ilan); Queens University School of Medicine, Kingston (Dr Ilan); Marriott School of Management, Brigham Young University, Provo, Utah (Mr Garrett and Dr LeBaron); and Rotman School of Management, University of Toronto, Toronto, Ontario (Dr Christianson).
Handoffs in hospitals have been widely recognized by both regulators and researchers as a locus of potential communication failure, with substantial risks to patient safety and quality of care.1,2 By conservative estimate, there are over half a billion patient handoff discussions annually in US hospitals. Most empirical studies have been performed in shift-change settings, where most handoffs occur, and where it is typical that responsibility for multiple patients is transferred during a single handoff session. However, theoretical analysis in the literature is entirely focused on how best to hand off a single patient.3- 5 As a result, research has overlooked what has been labeled the portfolio problem: how best to allocate across multiple patients the scarce time available for a handoff session.6
In the first study of this issue, to our knowledge, we used video recordings of 262 patient discussions in 23 handoff sessions among experienced attending physicians in the intensive care unit (ICU) of a tertiary medical center. We found that first-discussed patients received about 50% more time than those discussed last in a session. This occurred despite the order of cases being effectively random and therefore unrelated to severity or complexity of illness.
We recorded 23 end-of-week handoff sessions that occurred just prior to the transfer of responsibility for the 21-bed ICU. The unit was staffed by 2 teams, each led by an outgoing attending physician who handed off to an incoming one. Our study was approved by the Queen's University ethics board (Kingston, Ontario, Canada) and included 10 highly experienced physicians with a median of 9 years as an attending physician.
The procedure followed in this ICU was to discuss patients in bed-list order, not according to severity. We confirmed this in interviews and determined that ICU bed assignment itself did not relate to acuity or complexity. With unpredictable patient arrivals and all rooms equally equipped, the discussion order of the cases was effectively randomized, making severity of illness or other patient characteristics unrelated to discussion order.
Our main measures of interest were constructed from the videos: the number of patient discussions in each session, the ordinal position of each discussion in its session, and its duration. To determine whether our hypothesized negative relationship of order and length of discussion was statistically significant, we computed the Kendall τ rank order coefficient within each session. To determine the magnitude of the portfolio effect, we used 3 distinct methods: comparing means for early and late discussions, monotonic regression, the method most appropriate for ordinal predictors, and linear regression, a common technique for approximating such relationships. Details on participants, methods, and analyses (eTables and eFigures) are available in the online supplement.
We observed 262 patient discussions. The mean (SD) session duration was 142.73 (98.20) seconds. A median session had 11 discussions (range, 6-23 discussions).
Kendall τ correlations were negative for 19 sessions. Their overall mean (SD) was −0.186 (0.302) (median, −0.282). With no standard significance test for this complex situation, we performed a Monte Carlo test,7 a 100 000-replication random replacement simulation using our data set's exact structure of sessions and of patients per session. It determined that the observed pattern of correlations would occur by chance with P < .001.
To determine the magnitude of the portfolio effect, we used 3 approaches that produced highly similar estimates. To facilitate comparison, we converted absolute duration to the proportion of the session at which the patient discussion occurred.
We compared the mean (SD) fraction of a session used by discussions occurring first (0.119 [0.065]) and last (0.077 [0.064]) (Wilcoxon signed rank test; P < .04). Monotonic regressions8,9 performed separately for each session size yielded weighted-average estimates (details are in the online supplement) for first and last shares of total time of 0.126 and 0.075. Approximation by linear regression for our median session-size estimated shares of 0.118 and 0.071. The 3 approaches closely agree, as does a fourth analysis, reported in our online supplement, using a linear mixed-effects model to control for intrasession correlation. The average time allocated declined steadily with increasing ordinal position. For example, in a median-size session the first discussion used at least 50% more time than the last.
The portfolio effect is summarized in the Figure, showing the relationship of proportion of session to relative position in session for the pooled 262 cases together with a least-squares fitted line.
To our knowledge, this study of discussion order and duration is the first of its kind. It has the limitations of involving 1 site and 1 particular type of handoff. However, it is easily replicated, and, if confirmed, it suggests that shift-change handoffs and handoff training programs should include methods for explicitly controlling the allocation of scarce time across the portfolio of patients discussed. Examples of such methods include discussing the “sickest” patients first,10 or the newest first, or deliberately concluding with a block of time reserved for returning to cases requiring further discussion.
Correspondence: Dr Cohen, School of Information, University of Michigan, 3442 North Quad, 105 S State St, Ann Arbor, MI 48109-1285 (email@example.com).
Published Online: November 12, 2012. doi:10.1001/2013.jamainternmed.65
Author Contributions:Study concept and design: Cohen, LeBaron, and Christianson. Acquisition of data: Cohen, Ilan, Garrett, and LeBaron. Analysis and interpretation of data: Cohen, Ilan, Garrett, and LeBaron. Drafting of the manuscript: Cohen. Critical revision of the manuscript for important intellectual content: Cohen, Ilan, Garrett, LeBaron, and Christianson. Statistical analysis: Cohen and Garrett. Obtained funding: LeBaron. Administrative, technical, and material support: Cohen and LeBaron.
Conflict of Interest Disclosures: None reported.
Funding/Support: Dr Cohen is supported by a Robert Wood Johnson Foundation Investigator Award; Dr Ilan is supported by the Physicians' Services Inc Foundation.
Role of the Sponsors: The funding sources had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
Additional Information: Dr Cohen and Mr Garrett performed the data coding and analysis; Drs Cohen, Ilan, and LeBaron did the data collection; and all authors contributed to the writing of the manuscript.
Additional Contributions: Daren K. Heyland, MD, MSc, assisted with data collection; Mariom V. Ferrer, with transcription; Chandler Krynen, BA, with coding; and Andrew G. Day, MSc, with statistical methods.