Effect of ICU Strain on Timing of Limitations in Life-Sustaining Therapy and Death
May Hua, MD, MSc, Scott D. Halpern, MD, PhD, Nicole B. Gabler, PhD, MPH, MHA, Hannah Wunsch, MD, MSc
Electronic Supplementary Material
Details of linear regression modeling
When assessing interaction terms between strain and staffing models, we used cluster-robust standard errors to adjust for clustering to avoid collinearity, and models were adjusted for ICU characteristics including the presence of nighttime intensivist staffing and staffing model (open/closed, where a closed unit was defined as a model of care where patients were managed by a critical care physician and team).1
Linear relationships between continuous variables were assessed graphically prior to modeling. Because time to DNR, time to death and time to ICU discharge did not appear to satisfy assumptions of normality, these outcomes were log-transformed prior to use in the model. The MPM0III score also required log-transformation to satisfy a linear relationship to the outcome. Normality, linearity and heteroscedasticity were assessed using residual plots; robust standard errors were used if heteroscedasticity was present. The presence of multi-collinearity was assessed using variance inflation factor and tolerance.
Details of mediation analysis
For time to DNR to mediate the relationship between ICU strain and time to death, we would expect four relationships to exist: 1) ICU strain is associated with time to DNR; 2) time to DNR is associated with time to death; 3) ICU strain is associated with time to death; and 4) the magnitude of the effect of strain on time to death is attenuated when time to DNR is included in the model.2 To declare mediation, we required that all four of these criteria be met, but did not require that each was statistically significant itself, as such a requirement would reduce power to detect mediation.3
Each strain variable was tested separately, with the other two strain metrics included as covariates. We used mixed-effects linear regression to account for clustering of outcomes by ICU-year. We used bootstrapping with resampling to obtain bias-corrected confidence intervals, as bootstrapping does not rely on assumptions of normality.3 Both time to death and time to DNR were log-transformed; all models were fully adjusted for the demographic, clinical and admission characteristics listed above.
References
E1. Williams RL. A note on robust variance estimation for cluster-correlated data. Biometrics. Jun 2000;56(2):645-646.
E2. Krull JL, MacKinnon DP. Multilevel modeling of individual and group level mediated effects. Multivar Behav Res. 2001;36(2):249-277.
E3. MacKinnon DP, Fairchild AJ, Fritz MS. Mediation analysis. Annu. Rev. Psychol. 2007;58:593-614.
Table 1. Characteristics of ICUs in Participating in Project IMPACT, 2001-2008
N (%) / Patients
N (%)
Type of ICU
Medical / 31 (19.8) / 1,540 (15.8)
Surgical / 15 (9.6) / 511 (5.2)
Mixed medical-surgical / 84 (53.5) / 6,527 (66.8)
Trauma/Burn / 25 (15.9) / 1,167 (11.9)
Neurological / 2 (1.3) / 26 (0.3)
Type of Hospital
Academic / 39 (24.2) / 3,003 (30.4)
Government / 7 (4.4) / 366 (3.7)
Community-non-profit / 104 (64.6) / 6,171 (62.4)
Community-for-profit / 11 (6.8) / 351 (3.6)
Hospital location
Urban / 97 (60.6) / 5,968 (60.5)
Suburban / 46 (28.8) / 2,211 (22.4)
Rural / 17 (10.6) / 1,686 (17.1)
ICU Model
Open / 149 (92.6) / 8,348 (84.4)
Closed / 12 (7.5) / 1,543 (15.6)
Nighttime Intensivist Staffing
No / 122 (75.8) / 6,547 (66.2)
Yes / 39 (24.2) / 3,344 (33.8)
ICU; intensive care unit
Table 2. Linear Regression Results for the Relationship Between Metrics of ICU Strain and Time to Do-Not-Resuscitate Orders and Death for Decedents with any Limitation in Life-Sustaining Therapya
(days) / Time to Death
(hours)
β
(95% CI) / P-value / β
(95% CI) / P-value
Unit Census / 0.01
(-0.01 to 0.03) / 0.52 / 0.00
(-0.02 to 0.02) / 0.83
Admissions per day / -0.24
(-0.52 to 0.04) / 0.09 / -0.27
(-0.50 to -0.03) / 0.02
Unit Acuity / -0.20
(-0.37 to -0.04) / 0.02 / -0.14
(-0.28 to 0.01) / 0.06
Log MPM0III / 0.32
(0.27 to 0.37) / <0.001 / 0.26
(0.22 to 0.30) / <0.001
Male / 0.05
(0.02 to 0.08) / <0.001 / 0.04
(0.01 to 0.06) / 0.005
Race
Caucasian / ref / - / ref / -
Black / 0.05
(-0.01 to 0.10) / 0.11 / 0.03
(-0.02 to 0.08) / 0.23
Other / -0.02
(-0.10 to 0.05) / 0.53 / -0.03
(-0.09 to 0.04) / 0.44
Functional Status on Admission
Independent / ref / - / -
Partially Dependent / 0.08
(-0.12 to -0.04) / <0.001 / -0.06
(-0.09 to -0.02) / 0.001
Fully Dependent / -0.09
(-0.15 to -0.02) / 0.01 / -0.04
(-0.09 to 0.01) / 0.12
Insurance
Private / ref / - / ref / -
Medicare / -0.04
(-0.08 to -0.01) / 0.11 / -0.01
(-0.05 to 0.03) / 0.65
Medicaid / 0.02
(-0.06 to 0.10) / 0.58 / 0.03
(-0.04 to 0.10) / 0.38
Self-pay / -0.15
(-0.22 to -0.07) / <0.001 / -0.11
(-0.17 to -0.05) / <0.001
Government or Other / -0.02
(-0.13 to 0.10) / 0.76 / -0.05
(-0.14 to 0.05) / 0.34
Type of admission
Medical / ref / - / ref / -
Surgical / 0.10
(0.03 to 0.18) / 0.007 / 0.08
(0.02 to 0.14) / 0.001
Mechanical ventilation / 0.45
(0.39 to 0.51) / <0.001 / 0.37
(0.33 to 0.41) / <0.001
Vasopressor use / 0.19
(0.15 to 0.23) / <0.001 / 0.17
(0.14 to 0.20) / <0.001
Weekend admission / -0.03
(-0.07 to 0.00) / 0.06 / -0.03
(-0.06 to -0.01) / 0.02
Source of Admission
Emergency room / ref / - / ref / -
General care / 0.05
(0.00 to 0.10) / 0.04 / 0.05
(0.01 to 0.09) / 0.01
Step-down unit / 0.07
(0.01 to 0.14) / 0.02 / 0.07
(0.02 to 0.12) / 0.005
Procedure / 0.01
(-0.06 to 0.09) / 0.74 / 0.01
(-0.05 to 0.07) / 0.71
Another ICU / 0.15
(0.06 to 0.24) / 0.001 / 0.17
(0.09 to 0.25) / <0.001
Other / 0.16
(0.08 to 0.23) / <0.001 / 0.14
(0.08 to 0.20) / <0.001
DNR, do not resuscitate; CI, confidence interval.
aAll models were adjusted for ICU-year as a fixed effect. All outcomes were log-transformed. There was no evidence of multi-collinearity; robust standard errors were used.
Table 3. Interactions Between Selected Metrics of ICU Strain and Staffing Models
(days) / Time to Death for Decedents with Limitations in Life-Sustaining Therapy (hours)
Interaction P-value / β
(95% CI) / Stratified p-value / Interaction P-value / β
(95% CI) / Stratified p-value
Census and acuity / 0.41 / 0.40
Acuity at 25th percentilea / 0.00
(-0.02 to 0.03) / 0.78 / -0.00
(-0.02 to 0.02) / 0.90
Acuity at 50th percentile / 0.01
(-0.01 to 0.03) / 0.50 / 0.00
(-0.02 to 0.02) / 0.82
Acuity at 75th percentile / 0.01
(-0.01 to 0.03) / 0.34 / 0.01
(-0.01 to 0.03) / 0.56
Admissions and staffing model / 0.24 / 0.35
Open / -0.74
(-1.04 to -0.44) / <0.001 / -0.71
(-0.95 to -0.47) / <0.001
Closed / -1.24
(-2.00 to -0.48) / <0.001 / -1.04
(-1.67 to -0.41) / <0.001
Census and staffing model / 0.04 / 0.003
Open / 0.02
(-0.00 to 0.04) / 0.07 / 0.02
(-0.00 to 0.03) / 0.12
Closed / -0.03
(-0.07 to 0.01) / 0.17 / -0.05
(-0.08 to -0.01) / 0.01
CI, confidence interval.
aEffect of census for a patient with value of acuity at a given percentile.
Table 4. Relationship Between Metrics of ICU Strain and Time to Death or DNR Orders for Decedents with any Limitation in Life-Sustaining Therapy, Adjusting for Significant Interactionsa
β
(95% CI) / P-value / β
(95% CI) / P-value
Census
Open Staffingb / 0.02
(-0.00 to 0.04) / 0.07 / 0.02
(-0.00 to 0.03) / 0.12
Closed Staffingb / -0.03
(-0.07 to 0.01) / 0.17 / -0.05
(-0.08 to -0.01) / 0.006
Admissions / -0.80
(-1.08 to -0.52) / <0.001 / -0.76
(-0.98 to -0.53) / <0.001
Acuity / -0.18
(-0.33 to -0.03) / 0.02 / -0.07
(-0.21 to 0.06) / 0.29
ICU, intensive care unit; DNR, do not resuscitate; CI, confidence interval.
aAdjusted for age, gender, race, functional status on admission, type of insurance, type of patient (surgical, non-surgical), log MPM0-III score, use of mechanical ventilation, use of vasopressors, weekend admission, presence of nighttime intensivist staffing, location prior to admission, closed staffing model. An interaction term for census and staffing model was included. All outcomes were log-transformed. There was no evidence of multi-collinearity; cluster-robust standard errors were used.
bClosed staffing model was operationally defined as a model of care where patients had a mandated transfer to a unit where they were managed by a critical care physician and team. An open staffing model was defined as any other type of staffing model, including critical care consultation.
Table 5. Linear Regression Results for the Relationship Between Metrics of ICU Strain and Time to ICU Discharge for Patients with Limitations for Life-Sustaining Therapy who Died in Hospital
(n=5,011)
β
(95% CI) / P-value
Census / -0.04
(-0.07 to -0.01) / 0.007
Admissions*
Open / -1.09
(-1.45 to -0.73) / <0.001
Closed / -2.52
(-3.77 to -1.27) / <0.001
Acuity / 0.02
(-0.21 to 0.26) / 0.88
Log MPM0III / 0.36
(0.29 to 0.43) / <0.001
Male / 0.02
(-0.02 to 0.07) / 0.36
Race
Caucasian / ref / -
Black / 0.15
(0.07 to 0.22) / <0.001
Other / 0.01
(-0.09 to 0.11) / 0.84
Functional Status on Admission
Independent / ref / -
Partially Dependent / -0.06
(-0.12 to -0.00) / 0.04
Fully Dependent / -0.11
(-0.17 to -0.04) / 0.002
Insurance
Private / ref / -
Medicare / -0.01
(-0.07 to 0.05) / 0.69
Medicaid / 0.08
(-0.04 to 0.20) / 0.19
Self-pay / 0.02
(-0.12 to 0.17) / 0.73
Government or Other / 0.01
(-0.14 to 0.17) / 0.85
Type of admission
Medical / ref / -
Surgical / 0.15
(0.05 to 0.25) / 0.003
Mechanical ventilation / 0.85
(0.79 to 0.90) / <0.001
Vasopressor use / 0.33
(0.28 to 0.38) / <0.001
Weekend admission / 0.01
(-0.03 to 0.06) / 0.59
Source of Admission
Emergency room / ref / -
General care / 0.04
(-0.03 to 0.11) / 0.23
Step-down unit / 0.09
(0.02 to 0.17) / 0.02
Procedure / 0.03
(-0.08 to 0.14) / 0.60
Another ICU / 0.29
(0.14 to 0.45) / <0.001
Other / 0.18
(0.07 to 0.28) / 0.001
Nighttime intensivist staffing / -0.01
(-0.08 to 0.06) / 0.80
Closed staffing model / 0.18
(-0.06 to 0.43) / 0.14
ICU; intensive care unit
*Interaction between admissions and staffing model, p=0.03.
Table 6. Linear Regression Results for the Relationship Between Metrics of ICU Strain and Time to Death for Decedents without Limitations in Life-Sustaining Therapy and Decedents with Lower Severity of Illness
Decedents without Limitations in Life-Sustaining Therapy
(n=825) / Decedents with Lower Severity of Illness
(n=2,486)
β
(95% CI) / P-value / β
(95% CI) / P-value
Census / 0.01
(-0.08 to 0.10) / 0.79 / -0.01
(-0.05 to 0.03) / 0.60
Admissions / 0.52
(-0.61 to 1.65) / 0.37 / 0.07
(-0.45 to 0.58) / 0.80
Acuity / -0.08
(-0.99 to 0.83) / 0.91 / -0.02
(-0.36 to 0.32) / 0.91
Log MPM0III / 0.13
(0.00 to 0.34) / 0.25 / 0.79
(-0.38 to 1.96) / 0.19
Male / 0.00
(-0.13 to 0.14) / 0.92 / 0.00
(-0.06 to 0.07) / 0.94
Race
Caucasian / ref / - / ref / -
Black / -0.01
(-0.17 to 0.15) / 0.96 / -0.04
(-0.13 to 0.06) / 0.48
Other / 0.12
(-0.10 to 0.35) / 0.76 / 0.03
(-0.10 to 0.15) / 0.66
Functional Status on Admission
Independent / ref / - / ref / -
Partially Dependent / -0.00
(-0.19 to 0.18) / 0.96 / -0.07
(-0.15 to 0.01) / 0.10
Fully Dependent / 0.04
(-0.23 to 0.32) / 0.76 / -0.04
(-0.17 to 0.09) / 0.56
Insurance
Private / ref / - / ref / -
Medicare / -0.15
(-0.31 to 0.00) / 0.04 / -0.04
(-0.12 to 0.03) / 0.31
Medicaid / -0.12
(-0.37 to 0.13) / 0.28 / -0.01
(-0.12 to 0.10) / 0.78
Self-pay / -0.13
(-0.43 to 0.16) / 0.31 / -0.07
(-0.18 to -0.04) / 0.24
Government or Other / -0.18
(-0.54 to 0.17) / 0.30 / 0.02
(-0.18 to 0.21) / 0.84
Type of admission
Medical / ref / - / ref / -
Surgical / 0.09
(-0.08 to 0.26) / 0.67 / 0.10
(-0.03 to 0.24) / 0.14
Mechanical ventilation / 0.38
(0.22 to 0.53) / 0.11 / 0.33
(0.25 to 0.41) / <0.001
Vasopressor use / 0.15
(0.03 to 0.27) / 0.01 / 0.11
(0.05 to 0.18) / 0.001
Weekend admission / -0.02
(-0.05 to -0.00) / 0.48 / -0.02
(-0.09 to 0.05) / 0.52
Source of Admission
Emergency room / ref / - / ref / -
General care / -0.04
(-0.25 to 0.16) / 0.64 / 0.02
(-0.06 to 0.11) / 0.64
Step-down unit / 0.14
(-0.11 to 0.39) / 0.28 / 0.09
(-0.03 to 0.21) / 0.13
Procedure / 0.13
(-0.17 to 0.42) / 0.42 / -0.04
(-0.19 to 0.11) / 0.60
Another ICU / -0.21
(-0.40 to -0.02) / 0.03 / 0.03
(-0.12 to 0.19) / 0.67
Other / 0.05
(-0.22 to 0.32) / 0.85 / 0.09
(-0.03 to 0.22) / 0.15
ICU; intensive care unit
Table 7. Interactions Between Selected Metrics of ICU Strain and Staffing Models for Decedents without Limitations in Life-Sustaining Therapy and Decedents with Lower Severity of Illness
Interaction P-value / β
(95% CI) / Stratified p-value / Interaction P-value / β
(95% CI) / Stratified p-value
Census and acuity / 0.72 / 0.10
Acuity at 25th percentilea / 0.00
(-0.09 to 0.10) / 0.93 / -0.01
(-0.05 to 0.03) / 0.64
Acuity at 50th percentile / 0.01
(-0.08 to 0.11) / 0.77 / 0.01
(-0.04 to 0.05) / 0.80
Acuity at 75th percentile / 0.02
(-0.09 to 0.14) / 0.70 / 0.02
(-0.03 to 0.07) / 0.43
Admissions and staffing model / 0.34 / 0.43
Open / 0.10
(-0.05 to 0.07) / 0.77 / -0.61
(-1.05 to -0.17) / 0.007
Closed / 1.27
(-1.08 to 3.63) / 0.29 / -0.88
(-2.08 to 0.32) / 0.15
Census and staffing model / 0.23 / 0.48
Open / -0.01
(-0.08 to 0.06) / 0.83 / 0.01
(-0.03 to 0.05) / 0.59
Closed / 0.11
(-0.07 to 0.29) / 0.24 / -0.02
(-0.11 to 0.06) / 0.60
CI, confidence interval.