This action might not be possible to undo. Are you sure you want to continue?
Optimal Timing of Transfer Out of the Intensive Care Unit
Allan Garland, MD, MA, Alfred F. Connors, Jr, MD Am J Crit Care. 2013;22(5):390-397.
Abstract and Introduction
Background Little other than subjective judgment is available to help clinicians determine when a patient should be transferred out of the intensive care unit. Objective To assess whether remaining in the intensive care unit longer than judged to be medically necessary is associated with increased 30-day mortality. Methods This prospective, observational cohort study was performed in a 13-bed, closed-model, adult medical intensive care unit of a county-owned, university-affiliated hospital that often has difficulty transferring patients to general care areas because of a lack of available beds. Analysis included all 2401 survivors of intensive care from the study period. Delay in discharge from the intensive care unit was defined as time elapsed between the request for transfer and the actual transfer. Logistic regression was used to assess the association of discharge delay with 30-day mortality, adjusting for demographics, comorbid conditions, type and severity of acute illness, care limitations in the unit, and other potential confounding variables. Nonlinear relationships with continuous variables were modeled with restricted cubic splines. Results Overall, 30-day mortality was 10.1%. Mean discharge delay was 9.6 (SD, 11.7) hours; 9.9% had a discharge delay exceeding 24 hours. The relationship of 30-day mortality to discharge delay was statistically significant and Ushaped, with the nadir at 20 hours. Conclusions These data indicate an optimal time window for patients to leave the intensive care unit, with increased mortality not only if they leave earlier but also if they leave later than this optimal timing.
Little other than subjective clinical judgment is available to help clinicians determine when a patient should be transferred out of the intensive care unit (ICU).[1,2] The criterion of no longer requiring care that can be supplied only in the ICU seems straightforward to apply for mechanical ventilation and vasopressors, but less so for many interventions that are used in ICUs but are often performed outside of them as well. The decision about the need for ICU nursing care is largely subjective. As in other domains of clinical judgment in the ICU, large individual variation between physicians is likely.[3,4] In addition, bed availability can influence when patients leave the ICU. Admissions when the ICU is fully occupied can result in patients being transferred out before the physician would like, whereas limited availability of beds in step-down areas can lead to transfer occurring later than desired. Some evidence indicates excess mortality among patients transferred out of the ICU prematurely.[5,6] We are unaware of any data assessing outcomes associated with remaining in the ICU longer than appears medically required; such delays may increase mortality rates by prolonging the risk of ICU-acquired complications. Our goal was to assess how the timing of transfer out of ICU influences mortality. We were able to study the consequence of remaining in the ICU longer than the treating physicians deemed to be medically indicated because, owing to frequently inadequate availability of beds in the general care areas, patients in our ICU commonly experience such transfer delays. We hypothesized that delays in transferring patients out of the ICU due to lack of available beds in step-down units would increase mortality.
The data for this study were collected prospectively in the 13-bed medical ICU of MetroHealth Medical Center, a 520-bed urban, county-owned, university-affiliated hospital located in Cleveland, Ohio. The data span August 2002 to March 2007, excluding March to May 2004 when data collection was suspended because of personnel limitations. Care in this
To avoid erroneous mortality rates due to patients with multiple ICU admissions. Decisions to transfer patients out of the ICU were made by the intensivist and fellow.com/viewarticle/811246_print 2/11 . which differ between hospitals and can change over time. no set policies regarding these decisions were in place. wherein milder chronic conditions are less likely to be coded for more severely ill patients. We used the 2002 to 2006 Ohio death registry to assess posthospital mortality. (3) the request to transfer the patient out of the ICU. Rotations were 4 weeks long for the ICU fellow and house officers and 2 weeks long for the intensivists. sex. a comparison of dates of death for 200 randomly chosen patients showed more than 95% agreement for the 3 sources. Care limitations were represented by 2 mutually exclusive variables representing the presence. Patients were included if they survived their stay in the ICU. We also excluded patients discharged from our hospital directly to another acute care hospital and patients with ICU discharge delays exceeding 96 hours. variables found to have such relationships were included in the model as splines. The nurse to patient ratio averaged 1:2. or terminally withdraw any form of life-supporting therapies that had been initiated.mized into white versus nonwhite. We tested for nonlinear relationships between the dependent variable and continuous covariates by using restricted cubic splines. The 2 comorbid illness variables included in our models were the number of conditions present within each of these groups. we included only each patient's first ICU admission during the study period. and whether the transfer request was initiated at night or on a weekend. at any time in the ICU. Group 2 comprises comorbid conditions that have previously been associated with lower mortality.10] Similar models were constructed for hospital mortality and mortality to 60 days after ICU admission.[8. The date and time were recorded for (1) hospital admission. or (2) withhold life-supporting therapies within the context of otherwise providing all indicated care. of orders to (1) provide only comfort care. 2006. We used 30-day mortality instead of the more commonly reported in-hospital mortality because the latter is highly sensitive to interhospital and posthospital transfer patterns. The source of ICU admission was coded as the emergency department. The time from ICU admission until the transfer request is denoted as ICULOSdesired. Our primary analysis used multivariable logistic regression to evaluate the impact of ICU discharge delay on mortality to 30 days after ICU admission. Demographics were age. ICULOSdesired. To explore the causes of ICU transfer delay. Models included adjustments for patients' demographics. and (5) hospital discharge. These models included slightly different numbers of individuals because the Ohio death data available for this project ended December 31. and the need for invasive mechanical ventilation. as recorded in the hospital's administrative and billing database. gastrointestinal. PreLOS. with the expectation of imminent death.com/viewarticle/811246_print closed-model ICU was provided by a rotating team comprising a board-certified intensivist. Time intervals were measured in fractional hours. Formal rounds were made twice daily by this team. or surgical conditions/trauma). We compared the 2 data sources against each other and against the US Social Security Death Index. We determined in-hospital mortality from the hospital's computerized information system. whether ICU admission occurred at night (8 PM-8 AM) or on a weekend (Saturday or Sunday). The pre-ICU length of stay (PreLOS) was the time from hospital admission to ICU admission. other care areas in the hospital. cardiovascular. an outside hospital.medscape. Existing information indicates that in-hospital mortality and 30-day mortality are influenced by similar factors. ICU discharge delay was defined as the time elapsed from when a request for transfer to a bed in a general care area was received in the hospital's admitting office to the time that the patient left the ICU. (4) transfer out of the ICU. an ICU fellow. other ICUs. (2) ICU admission. this association most likely represents the effect of coding bias. Acute diagnostic groupings were coded as the organ system responsible for ICU admission (respiratory. Group 1 comprises conditions that are predictive of poorer outcomes.medscape. or other sources.[9. miscellaneous medical conditions. type and severity of acute illness. comorbid conditions. The best-fitting logistic regression models were determined as those with the lowest value of Akaike's information criterion.9] a fixed time point is much less influenced by such artifacts. We subdivided the 31 conditions into 2 groups. The intensivist took calls from home at night. existence of care limitation orders in ICU. source of ICU admission. we similarly constructed a quantile regression model for the 90th percentile www. Severity of acute illness was measured by using the worst value in the initial ICU day for the acute physiology score (APS) from the Acute Physiology and Chronic Health Evaluation II. it was combined with the most closely related category. and a group of 5 house officers who took overnight call on rotation.09/10/13 www. When preliminary analysis indicated that all members of a category had the same outcome. neurologic. and race dichoto . Comorbid illness was quantified as the presence of 31 specific preexisting conditions.
5) 1. for which a nonsignificant P value can be taken as indicating adequate calibration.0) 1384 (52. 12. Statistical analysis was done with Stata 10.6 (18.9 (1. and 72. and 0. (%) of patients At night (8 PM-8 AM) On weekend (Saturday or Sunday) ICU admission source. 11.0) 1269 (52. Most patients (85%) were transferred from ICU to general care areas.6) Patients with vital status data to hospital discharge 2624 55. Six percent died before hospital discharge. No. respectively.2 (0. 30-day and 60-day mortality information was unavailable for 223 and 285 patients. P values less than .7) hours (range.7%.9 (1. hospital survival status. 48.7 hours represents the unconditional 90th percentile of this variable.2 (0.6 (7.medscape.6 (7. (%) of patients Emergency department General care area Other ICU Outside hospital Others/miscellaneous Acute diagnostic grouping.7) 975 (37.8%. whereas for quantile regression.01 to 24. Continuous data are presented as mean (SD). 23. (%) of patients Respiratory Cardiovascular Neurologic Gastrointestinal Miscellaneous Surgical or trauma APACHE II acute physiology score.bmean (SD) Group 1 Group 2 Admitted to ICU.8) 1992 (76) 508 (19) 45 (2) 21 (1) 58 (2) 708 (27) 475 (18) 477 (18) 356 (14) 584 (22) 24 (1) 13. we used bootstrapped standard errors with 1000 repetitions.com/viewarticle/811246_print . 12% left the hospital directly from the ICU. The fractions having ICU discharge delay of 0 to 12. Model calibration was addressed via the Hosmer-Lemeshow H statistic.6 (SD.01 to 96 hours were 78.9) 1089 (45.4) 608 (25.5%.2) 1.2%.6) 3/11 www. No.05 are considered significant.01 to 48. mean (SD) Patients with 30-day vital status data 2401 55.0 software. and the number of samesex patients contemporaneously awaiting transfer out of this ICU.09/10/13 www.9) 1209 (46.9%. No. This model additionally included post-ICU location.com/viewarticle/811246_print of ICU discharge delay. of comorbid illnesses. 7. Characteristics and outcomes of ICU survivorsa Characteristic Number Age. (%) of patients No. 24.1%. Our primary analysis included the 2401 unique ICU survivors with known 30-day mortality (). mortality at 30 days after ICU admission was 10. Categorical data are presented as proportions.6 (18.9) 887 (36. 1. which waived informed consent. 0–93 hours).5) 1. This study was approved by the hospital's institutional review board. Their mean delay in being transferred out of ICU was 9. Table 1. respectively. mean (SD). Standard errors in logistic regression were calculated with the Huber-White robust sandwich estimator.1) 676 (25. No. (%) of patients Nonwhite race.01 to 72. 11. y Male sex. Results Of 2624 eligible patients.medscape.9) 1. Discrimination of logistic regression models was evaluated with the c statistic. Goodness of fit for quantile regression was assessed as the pseudo R2. No.3) 1816 (75) 467 (20) 42 (2) 21 (1) 55 (2) 658 (27) 433 (18) 433 (18) 344 (14) 519 (22) 24 (1) 13. also shows the characteristics of the entire cohort of 2624 unique ICU survivors.
No.9) 72.1) 70 (2.5) 656 (25.9 (1.11 Table 1.2) 15. No.0) 1269 (52.4 (66.9) 1209 (46.0) 180. No.7) 6.com/viewarticle/811246_print Patients with 30-day vital status data 2401 55.0) 275 (11. mean (SD). (%) of patients General care areas Other ICU in our hospital Left hospital directly from ICU ICU discharge delay.medscape.5) 12.1) 93.3 (173.2 (0.3) 246 (10.3 (1.7) 975 (37. h Mean (SD) Median (interquartile range) ICU length of stay.3) 794 (33.2 (66.0) 1384 (52. 10.0) 297 (11. actual. No.7) 1471 (61. (%) of patients At night (8 PM-8 AM) On weekend (Saturday or Sunday) www.0) 243 (10. (%) of patients Comfort care or withdrawal of life support Withholding of life support Request for transfer out of ICU. b Group 1 comprises conditions that are predictive of poorer outcomes.1) 676 (25.9) 601 (25.0) 106 (4.4) 93.com/viewarticle/811246_print APACHE II score (total).1 (88.2 (1.5) 12.6 (11.2) 1. of comorbid illnesses. No.0 (110.0) 9.0) 1.5) 1.9) 1089 (45. ICU. No. (%) of patients Died Alive. of patients simultaneously awaiting transfer from ICU.9) 73 (2. (%) of patients Care limitation orders in ICU.4.8) 4/11 . mean (SD) Intubated.0) 1. h Before ICU After ICU Hospital discharge status.9) 655 (25. intensive care unit.9 (1.5) 1. (%) of patients No. mean (SD) ICU discharge location.5 (85.3) 9.4) 608 (25. No. Acute Physiology and Chronic Health Evaluation.6 (18. No.2 (3.6 (18.7) 15.9) 288 (12.1) 2229 (84.6 (11.0) 96 (4.2 (3.5) 17.5) 863 (32.9) Abbreviations: APACHE. Characteristics and outcomes of ICU survivorsa Characteristic Number Age. h Hospital length of stay.7) 6.5 (8. 10. (%) of patients (out of 2401) 17. discharged to other than home 30-day mortality.2 (3.4.9) 1.4) 276 (10.8 (111. mean (SD).7 (172. y Male sex.6) 1614 (61.4) 136 (5. discharged home Alive.3) 600 (25. a The patients with vital status data to 30 days after ICU admission comprised the cohort for the primary analysis.2 (0. mean (SD).09/10/13 www. group 2 comprises those comorbid conditions previously associated with better outcomes.4) 2043 (85.9) 887 (36.4) 147 (5.3) 180. No.8) 322 (12.medscape.2 (3. (%) of patients Nonwhite race.3) Patients with vital status data to hospital discharge 2624 55. (%) of patients At night (8 PM-8 AM) On weekend (Saturday or Sunday) No.5 (8.9) 71. mean (SD) Glasgow Coma Scale score.bmean (SD) Group 1 Group 2 Admitted to ICU.
6 (11.0) 180.6 (7.9) 71. of patients simultaneously awaiting transfer from ICU. (%) of patients Comfort care or withdrawal of life support Withholding of life support Request for transfer out of ICU.2 (3. 10. mean (SD) APACHE II score (total). taken as the 90th conditional percentile of ICU discharge delay. No.2) 15.4) 136 (5.com/viewarticle/811246_print 5/11 .0) 106 (4. b Group 1 comprises conditions that are predictive of poorer outcomes. No. No.7) 6.0 (110.3) 180. h Before ICU After ICU Hospital discharge status. (%) of patients At night (8 PM-8 AM) On weekend (Saturday or Sunday) No. No. (%) of patients (out of 2401) 1816 (75) 467 (20) 42 (2) 21 (1) 55 (2) 658 (27) 433 (18) 433 (18) 344 (14) 519 (22) 24 (1) 13.4) 147 (5.5) 863 (32.9) 73 (2.1 (88.1) 70 (2.0) 9.3 (173.0) 297 (11.3) 9.7) 1471 (61.0) 275 (11.6) 17.9) 288 (12.medscape.09/10/13 www. No. we used all 2624 ICU survivors to construct a quantile regression model for long delays.2 (3. mean (SD) Glasgow Coma Scale score.5 (8.0) 243 (10. (%) of patients Emergency department General care area Other ICU Outside hospital Others/miscellaneous Acute diagnostic grouping.1) 93.6 (7. (%) of patients Died Alive. (%) of patients Care limitation orders in ICU.11 Noting that 90% of transfer delays were less than 24 hours. group 2 comprises those comorbid conditions previously associated with better outcomes.0) 96 (4.2 (3. a The patients with vital status data to 30 days after ICU admission comprised the cohort for the primary analysis.5) 12. discharged to other than home 30-day mortality.2 (3.5) 656 (25.6) 1614 (61.com/viewarticle/811246_print ICU admission source. mean (SD) ICU discharge location. No.5 (85. 10. (%) of patients Respiratory Cardiovascular Neurologic Gastrointestinal Miscellaneous Surgical or trauma APACHE II acute physiology score.3 (1.7 (172.5) 12.4) 93.9) 655 (25.8 (111. h Mean (SD) Median (interquartile range) ICU length of stay. Acute Physiology and Chronic Health Evaluation. No. (%) of patients General care areas Other ICU in our hospital Left hospital directly from ICU ICU discharge delay.4) 2043 (85.9) 601 (25.0) 1.8) 322 (12.4.0) 1. mean (SD).1) 2229 (84.3) 600 (25.4. mean (SD).7) 6. The only variables www.2 (1. intensive care unit. No. ICU.medscape.3) 794 (33.5 (8.7) 15.3) 246 (10. actual.6 (11. discharged home Alive.4 (66.9) Abbreviations: APACHE.4) 276 (10. mean (SD) Intubated.6) 17.2 (66.5) 1992 (76) 508 (19) 45 (2) 21 (1) 58 (2) 708 (27) 475 (18) 477 (18) 356 (14) 584 (22) 24 (1) 13. h Hospital length of stay.9) 72.
60. to 47.002). then progressively increases to values of 0. Table 3.99. orders limiting use of life-supporting therapies.09/10/13 www. by far. Form in the model Coefficient. 1. 1. 1.003 <. sex.97) 1. mechanical ventilation status.3 NA -8.com/viewarticle/811246_print Form in the model Odds ratio (95% CI) Cubic splines Cubic splines Binary Binary Linear Linear 4 indicator variables 4 indicator variables Linear Binary Linear Linear Binary Multiple terms Multiple terms 1. Using an ICU transfer delay of 0 hours as reference (odds ratio. APS.002 <.45 .89. Logistic regression model of mortality to 30 days after ICU admission among 2401 ICU survivorsa Independent variable ICU discharge delay. not applicable.001 NA <. 0.4 47.53 (1.90 (0. y Nonwhite race Male sex Number of comorbid illnesses d In group 1 In group 2 Source of ICU admission Acute diagnostic category APACHE II APS. the strongest association with ICU discharge delay. Significant covariates were age.09. NA.05.08 after this variable was removed from the model. respectively.73 (0.08 (1.01 <.36 including all variables to just 0. No measures of illness type or severity were significant in this model. Table 2. Wald test of nonlinear spline terms). source of ICU admission.001 <.00 (0.58 Direction of the effectb U-shapedc Increases Increases Increases Decreases Varies Varies Increases Decreases 6/11 .91.99 (0.00) 0.001 .medscape.001 <. 2. and 1.11) 0. the pseudo-R  of the model decreased from 0.7 16. APS. of same-sex patients awaiting transfer from ICU 0 1 2 3 ≥4 ICU discharge location General care areas Other ICUs Left hospital directly from ICU Abbreviations: ICU. 72.04 . The relationship between the odds of 30-day mortality and ICU discharge delay was significant (P = . 0. This variable had.01 <. 1. comorbid conditions. 0. nonlinear (P = . respectively.com/viewarticle/811246_print significant in this model were the number of same-sex patients contemporaneously awaiting transfer from the ICU and the ICU discharge location ().26 (1.56. and 93 hours.34. intensive care unit.57 (0. with nadirs at 19 and 21 hours.0).2 -7.medscape.12. 1. acute diagnosis. h (Reference group) Binary Binary Binary Binary (Reference group) Binary Binary NA 2. The corresponding models of hospital and 60-day mortality evidenced similar relationships.2 22.99.64. ranged from 2.3 extra hours if a patient was vying for beds outside the ICU with 4 or more other patients. aOnly significant covariates are shown.28) P .885 and a Hosmer-Lemeshow P value of .39 for transfer delays of 48.001. 1.87) 1.7 extra hours of delay with 1 other ICU patient in competition.001 .18 .003 . a measure of competition for transferring out of the ICU. and U-shaped (see Figure).29 (0.9 P NA .e h Admitted to ICU At night www. points Required mechanical ventilation Pre-ICU length of stay. and ICU discharge delay (). h Desired ICU length of stay.15 . h Age.14) 1. including age. the odds ratio decreases to a nadir of 0.35 with 20 hours of delay. Quantile regression model of 90th percentile of ICU discharge delay among 2624 ICU survivorsa Independent variable No.001 <.72.001 .006 . or in-hospital death. The first of these.42) 0.001 The model of 30-day mortality had a c statistic of 0.89) Multiple terms Multiple terms 1.00) 0. 1.
04 (0.44. not applicable.com/viewarticle/811246_print On weekend Request for transfer out of ICU At night On weekend End-of-life orders Withhold life support Withdrawal of life support Binary Binary Binary Binary Binary 1.medscape.09 <. 1. aDirection of effect indicated only for statistically significant variables. 1.001 3. Acute Physiology Score.75) 0. intensive care unit. dGroup 1 comprises conditions which predict poorer outcomes.56) 1. 1.71.08.11 eTime in ICU before request for transfer.medscape.62.6) Increases Increases Abbreviations: APACHE.68 (0.05) . ICU.001 <.71) 27. 53.09/10/13 www. group 2 comprises those comorbid conditions previously associated with better outcomes. 4.13 (2.com/viewarticle/811246_print 7/11 . Acute Physiology and Chronic Health Evaluation.78 .8 (14. APS. c See Figure. ICU discharge delay is the interval from www.4. NA.86 . Estimate and 95% confidence intervals are shown. Figure. bDirection in which predicted 30-day mortality changes with an increase in the value of the indicated variable. Effect of delay in discharge from intensive care unit (ICU) on the odds ratio for death at 30 days after ICU admission among 2401 ICU survivors.05 (0.
com/viewarticle/811246_print the transfer request until actual transfer out of the ICU.medscape. One would expect delays due to worsening illness to be associated with higher mortality.18] The reason for delayed ICU discharge is central to the interpretation our findings. Three additional comments are in order about our methods and results.medscape. which easily break down for non-Gaussian. U-shaped relationships often indicate the confluence of 2 opposing influences. with worse outcomes if they are discharged earlier or later than the optimal timing. quantile regression does not share the problematic assumptions of linear regression. That mortality rates progressively decreased with increasing ICU discharge delay up to 20 hours suggests that. The observed increase in mortality with longer delays in leaving the ICU is a phenomenon that has not been previously reported. Although logarithmic transformation sometimes ameliorates the problematic assumptions of linear regression. We used restricted cubic splines to assess for nonlinearity because they are flexible. we restricted our analysis to patients with ICU discharge delays less than 96 hours. special methods are then needed to interpret the results. our findings indicate that clinical judgment is inadequate for determining when it is safest for patients to leave the ICU. A possible explanation for the observation that our patients had higher survival rates if they spent approximately 1 extra day in the ICU is suggested by investigators who found that higher severity of illness on the day of ICU transfer was associated with higher post-ICU mortality. whereas ordinary least-squares regression assesses only the mean value. It is rarely justified to assume that continuous variables are linearly related to the dependent variable in health care research. powerful. patients may alternatively be delayed in leaving the ICU because their medical condition worsened before the requested transfer occurred. Although further clarification could come from assessing severity scores on the final ICU day. similar to our experience. that approach fails to make optimal use of the available data. Discussion Our analysis indicates complex effects of delays in leaving the ICU.09/10/13 www. Although our data set did not include the reason for patients' delayed ICU discharge. these patients would have benefited from an extra day in the ICU.[17. most delays from an adult ICU in Australia were due to unavailability of beds in general care areas. First.16] Although the alternative strategy of performing logistic regression after dichotomizing ICU discharge delay to generate a variable representing long delays has been used to analyze ICU transfer delays. Because our patients were believed by their physicians to be ready for transfer. our model for ICU discharge delay indicates that nonmedical reasons for ICU transfer delays were predominant. The prior literature on delayed discharge from the ICU is sparse. our model of ICU discharge delay used quantile regression instead of ordinary least squares (linear) regression. However. and it can render incorrect results.com/viewarticle/811246_print 8/11 . No measures of severity of illness. Second. our data set does not contain that information. Our results did not differ substantially if we instead limited analysis to patients with ICU discharge delays less than 72 hours. the predictive power of that model was almost entirely related to the number of other ICU patients who were competing for beds outside the ICU. except to observe that they most likely increase hospital costs. We made this choice because our goal was to identify variables associated with long transfer delays.[12. highly skewed variables such as time intervals. including eventual death in the hospital. First. Indeed. and simple to interpret. expecting worsening illness to cause longer transfer delays than the delays due to unavailability of beds in general care areas. We are not aware of prior studies in which the consequences of such delays were assessed. 2 lines of reasoning make it unlikely that many patients included in our analysis experienced such delays because of worsening illness. Delaying ICU discharge beyond 20 hours increasingly elevated mortality. An example of the latter would be complications from vascular catheters that are no longer needed but are not removed until just before the patient leaves the ICU. we achieved the goal. Discharge delays from a neonatal ICU in New York were associated with greater severity of illness. were significant predictors of ICU discharge delay. www. Possible mechanisms for this phenomenon include prolongation of exposure to ICU-acquired pathogens or other ICU-acquired complications.42). this evidence validates the belief that ICUs are dangerous places for patients. our regression results for 30-day mortality illustrate the importance of allowing for nonlinearity. These data suggest that there is an optimal window of time for patients to leave the ICU. However. Second. By modeling the 90th conditional percentile of ICU discharge delay. on average. if we had modeled ICU discharge delay linearly. Specifically. we would have incorrectly concluded that no significant relationship with mortality existed (P = . Also. Although our purpose was to evaluate the impact of remaining in the ICU longer than deemed medically necessary (because of the unavailability of beds in generalcare areas). it requires justifying the cut point chosen.
18 .05 (0.4. Additional exploration of the data demonstrated that mechanical ventilation was. points Required mechanical ventilation Pre-ICU length of stay. Acute Physiology and Chronic Health Evaluation.86 .08 (1.60. Table 3.001 . 1. we chose not to include GCS scores and these interaction terms in our models. Acute Physiology Score.57 (0.002 <.87) 1.00 (0. group 2 comprises those comorbid conditions previously associated with better outcomes. bDirection in which predicted 30-day mortality changes with an increase in the value of the indicated variable.68 (0. but our results indicated the reverse ().006 . however.29 (0. 1. Proving causality would require an interventional study. 1.58 . the apparent paradoxical protective effect of mechanical ventilation reflects high mortality among those patients with low GCS scores for whom mechanical ventilation was not applied. medical ICU. the association we observed in this observational study between ICU discharge delays and mortality does not prove a causal relationship. as expected.64.09/10/13 www.001 Direction of the effectb U-shapedc Increases Increases Increases Decreases Varies Varies Increases Decreases Increases Increases Abbreviations: APACHE.09 <. neurologic function.99.com/viewarticle/811246_print Third.15 .14) 1.001 <. h Desired ICU length of stay.97) 1.45 . 1.09.6) P . 2.com/viewarticle/811246_print 9/11 .04 . For those with poor neurologic function.00) 0. The key additional observation is that most patients with low GCS scores who did not receive mechanical ventilation had orders to withhold life support and died soon after leaving the ICU.11) 0.53 (1.e h Admitted to ICU At night On weekend Request for transfer out of ICU At night On weekend End-of-life orders Withhold life support Withdrawal of life support Form in the model Odds ratio (95% CI) Cubic splines Cubic splines Binary Binary Linear Linear 4 indicator variables 4 indicator variables Linear Binary Linear Linear Binary Binary Binary Binary Binary Binary Multiple terms Multiple terms 1.56) 1. where patients are assigned to various degrees of predetermined www.01 <.78 .medscape.89) Multiple terms Multiple terms 1. y Nonwhite race Male sex Number of comorbid illnesses d In group 1 In group 2 Source of ICU admission Acute diagnostic category APACHE II APS. Because they did not materially influence the observed relationship between mortality and ICU discharge delay.05) 3.medscape. h Age.00) 0.08.42) 0. NA. In addition. aDirection of effect indicated only for statistically significant variables. Thus. as identified by a high score on the Glasgow Coma Scale (GCS) at ICU admission. intensive care unit. 1.003 .26 (1. 1. dGroup 1 comprises conditions which predict poorer outcomes.89.90 (0. 1.8 (14. Additional analyses demonstrating the generalizability of our findings in other sites and in other types of ICUs are needed.04 (0. ICU. 0.05.001 . c See Figure.73 (0.62. The major limitation of our results is that they derive from a single. associated with higher 30-day mortality among patients with preserved neurologic function. we expected that needing mechanical ventilation reflects a higher severity of illness and would be associated with higher 30-day mortality. and end-of-life wishes in this cohort of ICU survivors.13 (2.11 eTime in ICU before request for transfer.44. mechanical ventilation was associated with lower mortality. 1.71.12.01 <. APS. not applicable.75) 0.34.28) 1. 0. 1.99.001 . 4. 53. Logistic regression model of mortality to 30 days after ICU admission among 2401 ICU survivorsa Independent variable ICU discharge delay. The likely explanation lies in the interplay between mechanical ventilation.99 (0.71) 27.
2011. New York.ancestry. Garland A. Steiner C. 2001.10:1298– 1305. 2002. 2009. Indeed. 9. Our findings further imply that clinical judgment is not reliable for determining that optimal time window. Johansen M. Garland A. Accessed June 21. Thomas CL.322:1274–1276. Med Care. Comparison of Charlson comorbidity index with SAPS and APACHE scores for prediction of mortality following intensive care. Rationing in the intensive care unit. 12. J Palliat Med. www. et al. 2006. 5. 13. http://search. Goldfrad C. 2000. Shaman Z. 2009.3:203– 211. Comorbidity measures for use with administrative data. Einstadter D. Chang RWS. J Clin Epidemiol. New York. such a study would be difficult to perform. create or contribute to an online discussion on this topic. References 1.355:1138–1142.27:633–638. 10. Crit Care Med. and triage. Vasilevskis EE. Baron J. Med Care. they indicate that the importance of determining when patients should be transferred from the ICU is not limited to the economic consideration of improving ICU bed utilization.org and click "Responses" in the second column of either the full-text or PDF view of the article. 2006. Am J Respir Crit Care Med.62:510–516.medscape. Christensen S. Cook DJ. 4. Garland A.[1. Harris DR. practical methods of determining when patients should be transferred out of the ICU. Connors AF Jr. Elixhauser A. Harrell FE Jr. Society of Critical Care Medicine. Reduction in mortality after inappropriate early discharge from intensive care unit: logistic regression triage model. Physician-attributable differences in intensive care unit costs: a single-center study. Consequences of discharges from intensive care at night. Connors AF Jr. Clin Epidemiol.medscape. Rowan K. 14. discharge. Applied Logistic Regression. 1998.34:958–963. Kuzniewicz MW. 1989. 3. Social Security Death Index (SSDI).36:8–27.47:803–812. Mortality trends during a program that publicly reported hospital performance. Cebul RD. Lancet. Med Care. 7. BMJ. Visit www. Guidelines for intensive care unit admission. Marrie RA. 2. Crit CareMed.com/search/db. 1999. with an increasing risk of subsequent death if patients leave the ICU either too early or too late. Jensen R. Regression Modeling Strategies.09/10/13 www. Brock DW. Daly K. Lemeshow S. Husak SS. Christiansen C. Beale R. Task Force of the American College of Critical Care Medicine. 8. Baker DW.40:879–890. Quantile regression and restricted cubic splines are useful for exploring relationships between continuous variables. 2 professional task forces have recognized that intensivists have little except subjective clinical judgment to guide them in determining when patients should be discharged from the ICU. et al. 6. 2007. Dawson NV.2] Although further work is needed to confirm and clarify our findings. 2013. NY: Springer. Hosmer DW.174:1206–1210. NY: Wiley. Conclusion We observed that there is an optimal timing for patients to leave the ICU.com/viewarticle/811246_print discharge delay. 2001.com/viewarticle/811246_print 10/11 . evidence-based. Lemeshow S. 11. Sidebar eLetters Now that you've read the article. Physicians' influence over decisions to forgo life support. Relationship between discharge practices and intensive care unit in-hospital mortality performance evidence of a discharge bias.ajcconline. Truog RD. Gordon NH. Coffey RM. Future research is needed to discover objective.aspx?dbid=3693. Dean ML.
1999.28:87– 96. reprints@aacn. Australian Health Rev.com/viewarticle/811246_print 11/11 . A gentle introduction to quantile regression for ecologists. Duan N.1:412–420. PsycholMethods. 2002. 2004.medscape.org. 18. (949) 362-2049. Zhang S. 19.78:605– 610. A program to reduce discharge delays in a neonatal intensive care unit. Machado JAF. To purchase electronic or print reprints. © 2013 American Association of Critical-Care Nurses www. (800) 899-1712 or (949) 362-2050 (ext 532). J Am Statistical Assoc. contact American Association of Critical-Care Nurses.7:19–40. MacCallum R. fax. Krauss A. Am J Manag Care. CA 92656. Preacher K. Cade BS. Koenker R. Smearing estimate: a nonparametric retransformation method. Auld P. Perlmutter D. Noon BR. 101 Columbia.22(5):390-397.medscape. e-mail.4:548–552. Frontiers Ecology Environment. On the practice of dichotomization of quantitative variables. Suico C. 1983. Delayed discharges from an adult intensive care unit. Leslie G. 17. 1998. Phone. 16.94:1296–1310. Rucker D.09/10/13 www. Am J Crit Care. Williams T. Goodness of fit and related inference processes for quantile regression. J Am Statistical Assoc. 20. 2003. Aliso Viejo. 2013.com/viewarticle/811246_print 15.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.