RESEARCH—HUMAN—CLINICAL STUDIES

Testing the Impact of Protocolized Care of Patients

With Severe Traumatic Brain Injury Without
Randall M. Chesnut, MD ‡* Intracranial Pressure Monitoring: The
Nancy Temkin, PhD‡*
Walter Videtta, MD§* Imaging and Clinical Examination Protocol
Silvia Lujan, MD||*
Gustavo Petroni, MD, MCR|| BACKGROUND: Most patients with severe traumatic brain injury (sTBI) in low- or-middle-
Jim Pridgeon, MHA‡ income countries and surprisingly many in high-income countries are managed without
Sureyya Dikmen, PhD‡ intracranial pressure (ICP) monitoring. The impact of the first published protocol (Imaging
Kelley Chaddock, BA‡ and Clinical Examination [ICE] protocol) is untested against nonprotocol management.
Terence Hendrix, BA¶ OBJECTIVE: To determine whether patients treated in intensive care units (ICUs) using
Jason Barber, MS‡* the ICE protocol have lower mortality and better neurobehavioral functioning than those
Joan Machamer, MA‡ treated in ICUs using no protocol.
Nahuel Guadagnoli, MD# METHODS: This study involved nineteen mostly public South American hospitals. This is a
Peter Hendrickson, PhD‡ prospective cohort study, enrolling patients older than 13 years with sTBI presenting within 24 h
Victor Alanis, MD** of injury (January 2014-July 2015) with 6-mo postinjury follow-up. Five hospitals treated all sTBI
Gustavo La Fuente, MD†† cases using the ICE protocol; 14 used no protocol. Primary outcome was prespecified composite
Arturo Lavadenz, MD‡‡ of mortality, orientation, functional outcome, and neuropsychological measures.
Roberto Merida, MD§§ RESULTS: A total of 414 patients (89% male, mean age 34.8 years) enrolled; 81% had 6 months
Freddy Sandi Lora, MD|||| of follow-up. All participants included in composite outcome analysis: average percentile (SD) =
Ricardo Romero, MD¶¶ 46.8 (24.0) nonprotocol, 56.9 (24.5) protocol. Generalized estimating equation (GEE) used to
Oscar Pinillos, MD## account for center effects (confounder-adjusted difference [95% CI] = 12.2 [4.6, 19.8], P = .002).
Zulma Urbina, MD*** Kaplan–Meier 6-month mortality (95% CI) = 36% (30%, 43%) nonprotocol, 25% (19%, 31%)
Jairo Figueroa, MD††† protocol (GEE and confounder-adjusted hazard ratio [95% CI] = .69 [.43, 1.10], P = .118). Six-
Marcelo Ochoa, MD‡‡‡ month Extended Glasgow Outcome Scale for 332 participants: average Extended Glasgow
Rafael Davila, MD§§§ Outcome Scale score (SD) = 3.6 (2.6) nonprotocol, 4.7 (2.8) protocol (GEE and confounder-
Jacobo Mora, MD|||||| adjusted and lost to follow-up–adjusted difference [95% CI] = 1.36 [.55, 2.17], P = .001).
Luis Bustamante, MD¶¶¶ CONCLUSION: ICUs managing patients with sTBI using the ICE protocol had better
Carlos Perez, MD### functional outcome than those not using a protocol. ICUs treating patients with sTBI
Jorge Leiva, MD**** without ICP monitoring should consider protocolization. The ICE protocol, tested here and
Carlos Carricondo, MD†††† previously, is 1 option.
Ana Maria Mazzola, MD‡‡‡‡
KEY WORDS: Intracranial hypertension, Intracranial pressure monitoring, Severe traumatic brain injury, Neu-
Juan Guerra, MD§§§§ rocritical care, Global health, Protocols
‡
University of Washington, Harborview
Medical Center, Seattle, Washington, USA; Neurosurgery 92:472–480, 2023 https://doi.org/10.1227/neu.0000000000002251
(Continued on next page)

*Randall M. Chesnut, Nancy Temkin, ABBREVIATIONS: BEST TRIP, Benchmark Evidence from South America Trial of Intracranial Pressure; EDH, epidural
Walter Videtta, Silvia Lujan, and Jason hematoma; GEE, generalized estimating equation; GOAT, Galveston orientation and amnesia test; GOSE, Extended
Barber contributed equally to this work. Glasgow Outcome Scale; HICs, high-income countries; ICE, imaging and clinical examination; ICP, intracranial
pressure; ICU, intensive care unit; ISS, injury severity scale; LMICs, low- or middle-income countries; LOS, length of
Correspondence: stay; PC, protocol center; NPC, nonprotocol center; PSI, processing speed index; RAVLT, Rey Auditory Verbal
Randall M. Chesnut, MD, Learning Test; SOAT, el Seguro Obligatorio de Accidentes de Tránsito; SWLS, Satisfaction With Life Scale; STBI, severe
Harborview Medical Center,
traumatic brain injury; TBI, traumatic brain injury.
University of Washington,
325 Ninth Ave, Supplemental digital content is available for this article at neurosurgery-online.com.
Seattle, WA 98104-2499, USA.
Email: [email protected]

Received, December 2, 2021. CNS Journal Club Podcast and CME Exams available at cns.org/podcasts by scanning this QR code
Accepted, August 30, 2022. using your mobile device.
Published Online, December 12, 2022.

© Congress of Neurological Surgeons

2022. All rights reserved.

472 | VOLUME 92 | NUMBER 3 | MARCH 2023 neurosurgery-online.com

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

 PATIENTS WITH sTBI PATIENTS WITHOUT ICP MONITORING: ICE PROTOCOL

T
raumatic brain injury (TBI) is disproportionately common standard, ICU-based supportive care, including mechanical ventilation, se-
in low- and-middle-income countries (LMICs) where re- dation, analgesia, and aggressive resuscitation and management of general non-
source limitations compound its negative impact. The TBI neurologic ICU issues. Neither PCs nor NPCs changed management for Phase
literature, principally generated from high-income countries (HICs), 1. All eligible patients were considered, and families arriving at hospital within
often has limited applicability in LMICs as exemplified by the absence 96 hour of injury were asked to consent to data collection (48 hour for the first
6 months). Enrollment spanned January 2014 to July 2015, when the planned
of publications on management strategies for centers without access to
sample size was reached; follow-up was completed in January 2016. The study
intracranial pressure (ICP) monitoring. When designing the BEST was approved by an institutional review board and ethics committees at all
TRIP randomized trial1 comparing management based on ICP- study centers.
monitor data with that without such monitoring, the absence of
published treatment algorithms necessitated ad hoc development of a
management paradigm for the nonmonitored group based on Imaging Variables
and Clinical Examination (the ICE protocol2,3) alone. Ipso facto, our Potential confounders are identified in Supplemental Statistical
Methods, http://links.lww.com/NEU/D512,6-8 and listed in Table 1.
Latin American coinvestigators created the first published severe TBI
Data were collected by similarly trained staff at each site using study-
care protocol. When the nonmonitored and monitored groups had provided case report forms. They obtained acute information from the medical
comparable outcomes, we questioned the stand-alone utility of the ICE record or by direct observation. Examiners obtained outcome information by
protocol vs alternatives such as the Hawthorne Effect4,5 and site se- testing or interviewing the participant or, if too impaired, interviewing family
lection. Because most patients with TBI globally are managed under members to obtain the Glasgow Outcome Scale-Extended (GOSE).9
similar resource-limited conditions, the importance of the question Primary outcome was a composite score10 of 7 data elements: survival,
prompted a prospective, observational trial comparing protocolized care Galveston Orientation and Amnesia Test (GOAT) at discharge and 6-
against business-as-usual management. We hypothesized that patients month postinjury, GOSE,11 Satisfaction With Life Scale (SWLS),12 Rey
treated in intensive care units (ICUs) using the ICE protocol would Auditory Verbal Learning Test (RAVLT),13 and WAIS IV Processing
have lower mortality and better neurobehavioral functioning than those Speed Index14 at 6 months after injury. Higher values of the composite
treated in ICUs using no protocol. We here report phase I of that study. indicate better outcome. To calculate the composite, each observation
was replaced by the rank (from worst to best) corresponding to the
individual’s score using midranks for tied observations. These were
METHODS converted to percentiles, and the composite score was calculated as the
average of the percentiles observed for that participant. For the GOAT,
Study Design SWLS, RAVLT, and Processing Speed Index, participants who died
before an assessment were assigned the lowest (worst) rank and those alive
This is phase I of a multicenter observational study investigating the in-
but too neurologically impaired to test were ranked between deaths and
fluence of protocolized management on sTBI outcome at hospitals without
the lowest score of anyone who performed the test. We used Gehan’s
routinely available ICP monitoring. It compares 6-month outcomes and
method15 to rank the sometimes-censored survival. We secondarily
management process variables from 2 patient groups, older than 13 years,
looked at individual measures. We prospectively collected ICU hourly
presenting with nonpenetrating sTBI within 24 hour of injury or deteriorating
clinical and treatment data as process variables. Treatments directed at
to that level within 48 hour of injury. We defined sTBI as a Glasgow Coma
suspected intracranial hypertension (SICH) were defined as “brain-
Scale (GCS) score ≤8 or a GCS motor (GCSm) score ≤5 if intubated. Patients
specific treatments.” These included hyperosmotics, pressors, hyper-
were excluded if they presented with a GCS of 3 with bilateral fixed and dilated
ventilation, furosemide, paralytics, CSF drainage, high-dose barbiturates,
pupils or injuries otherwise considered nonsurvivable. Phase I involved 2
and craniectomy but excluded ventilation, sedation, and analgesia.
hospital groups: protocol centers (PCs) and nonprotocol centers (NPCs). PCs
Duration of therapy was defined as days from injury until the final brain-
were 5 hospitals in Bolivia and Colombia participating in prior investigations
specific treatment. To integrate brain-specific treatments, we summed
with the Global Neurotrauma Research Group and thus familiar with the ICE
treatments per hour over the treatment interval.
no-ICP protocol.1,2 These hospitals continued to manage patients using this
protocol. NPCs included 14 hospitals in Colombia, Ecuador, Venezuela, and
Argentina new to our research, wherein ICP monitoring was not part of Statistical Methods
routine sTBI management and no formal management protocol was used. We derived the sample size of 390 using simulations to provide 80%
They managed patients without study-specific instruction. All patients received power to detect an average improvement of 10 percentage points on the

(Continued from previous page)

§
Medicina Intensiva, Hospital Nacional Professor Alejandro Posadas, Buenos Aires, Argentina; ||Hospital Emergencia, Dr Clemente Alvarez, Rosario, Argentina; ¶San Diego, California,
USA; #Centro de Informatica e Investigacion Clinica, Rosario, Argentina; **Medicina Intensiva, Hospital San Juan de Dios, Santa Cruz de la Sierra, Bolivia; ††Medicina Intensiva,
Hospital Japones, Santa Cruz de la Sierra, Bolivia; ‡‡Medicina Intensiva, Hospital Videma, Cochabamba, Bolivia; §§Medicina Intensiva, Hospital San Juan de Dios, Tarija, Bolivia;
||||
Medicina Intensiva, Hospital Obrero No 1, La Paz, Bolivia; ¶¶Medicina Intensiva, Fundacion Clinica Campbell, Barranquilla, Colombia; ##Medicina Intensiva, Clinica Universitaria
Rafael Uribe, Cali, Colombia; ***Medicina Intensiva, Hospital Erasmo Meoz ICU No 1, Cucuta, Colombia; †††Medicina Intensiva, Hospital Erasmo Meoz ICU No 2, Cucuta, Colombia;
‡‡‡
Medicina Intensiva, Hospital José Carrasco Artega, Cuenca, Ecuador; §§§Medicina Intensiva, Hospital Luis Razetti, Barinas, Venezuela; ||||||Medicina Intensiva, Hospital Luis Razetti,
Barcelona, Venezuela; ¶¶¶Medicina Intensiva, Delicia Conception Hospital Masvernat, Concordia, Entre Rı́os, Argentina; ###Medicina Intensiva, Hospital Justo José de Urquiza,
Concepción del Uruguay, Entre Rı́os, Argentina; ****Medicina Intensiva, Hospital Córdoba, Córdoba, Argentina; ††††Medicina Intensiva, Hospital Central, Mendoza, Argentina;
‡‡‡‡
Medicina Intensiva, Hospital San Felipe, San Nicolás, Buenos Aires, Argentina; §§§§Medicina Intensiva, Hospital COSSMIL Militar, La Paz, Bolivia

NEUROSURGERY VOLUME 92 | NUMBER 3 | MARCH 2023 | 473

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

CHESNUT ET AL

TABLE 1. Demographics and Injury Characteristics

Unweighted Weighteda

Nonprotocol Protocol P-valueb Nonprotocol Protocol P-valueb

Subjects 236 178 236 178

Age, mean (SD) 35.7 (16.5) 33.7 (14.7) .363 34.9 (16.1) 34.7 (14.3) .809
Sex, male 206 (87%) 163 (92%) .147 211 (89%) 161 (91%) .746
Cause of injury
A: Car 27 (12%) 9 (5%) .009 19 (8%) 9 (5%) .749
B: Motorcycle 127 (54%) 85 (48%) 119 (52%) 99 (56%)
C: Bicycle 3 (1%) 5 (3%) 6 (3%) 3 (2%)
D: Pedestrian 14 (6%) 28 (16%) 25 (11%) 18 (10%)
E: Fall 35 (15%) 25 (14%) 33 (14%) 21 (12%)
F: Assault 7 (3%) 7 (4%) 7 (3%) 5 (3%)
G: Other 21 (9%) 18 (10%) 22 (10%) 22 (12%)
Admission—direct to study hospital 130 (55%) 77 (43%) .022 120 (51%) 91 (51%) 1.000
Time to study hospital (h)
Median (IQR) 3.0 (1.5, 7.4) 2.9 (1.0, 6.2) .169 3.0 (2.0, 7.0) 3.0 (1.0, 7.0) .087
Enrolled because of late deteriorationc: yes 31 (13%) 32 (18%) .213 34 (14%) 25 (14%) 1.000
Enrollment GCSmd: mean (SD) 3.4 (1.4) 3.9 (1.4) <.001 3.7 (1.4) 3.6 (1.4) .097
ISS nonheade
Mean (SD) 11.7 (17.4) 5.4 (8.8) <.001 9.1 (14.7) 10.3 (15.8) .527
≥16 54 (23%) 12 (7%) 43 (18%) 31 (17%) .897
First ICU pupil reactivity
Abnormal (at least 1 nonreactive) 82 (36%) 53 (30%) .242 83 (36%) 51 (29%) .137
Marshall classificationf
1: Diffuse injury I 5 (2%) 1 (1%) .010 4 (2%) 1 (0%) .202
2: Diffuse injury II 70 (31%) 40 (22%) 64 (28%) 54 (30%)
3: Diffuse injury III 67 (29%) 50 (28%) 60 (26%) 55 (31%)
4: Diffuse injury IV 10 (4%) 3 (2%) 11 (5%) 3 (2%)
5: Evacuated mass lesion 64 (28%) 78 (44%) 79 (34%) 60 (34%)
6: Nonevacuated Mass Lesion 12 (5%) 6 (3%) 11 (5%) 4 (2%)
Mesencephalic cisternsf
Compressed/absent 128 (56%) 124 (70%) .006 130 (57%) 113 (63%) .186
Midline shiftf ≥5 mm 34 (15%) 37 (21%) .147 41 (18%) 30 (17%) .794
CT signs of increased ICPf: yes 180 (79%) 144 (81%) .709 181 (79%) 138 (78%) .718
Neurosurgeries
EDH 31 (13%) 42 (24%) .006 41 (17%) 31 (17%) 1.000
SDH 42 (18%) 47 (26%) .040 53 (22%) 38 (21%) .812
Contusion 5 (2%) 5 (3%) .751 8 (3%) 3 (2%) .364
ICH 2 (1%) 2 (1%) 1.000 1 (1%) 1 (1%) 1.000
Craniectomy 54 (23%) 61 (34%) .011 65 (28%) 45 (25%) .654
Total number (mean) 0.6 (0.9) 0.9 (0.9) <.001 0.8 (1.1) 0.7 (0.9) .170
EDH, epidural hematoma; GCSm, motor component of Glasgow Coma Scale score; ICP, intracranial pressure; ICU, intensive care unit; ISS, injury severity scale; SD, standard deviation.
a
Weights are inversely proportional to the predicted probability of belonging to the subject’s actual group (protocol or nonprotocol) and are calculated by inverting this probability
then scaling within each group so that the original group N’s are preserved.
b
Statistical significance by Fisher Exact (unweighted), chi-square (weighted), or Mann-Whitney as appropriate.
c
Admission if GCS >8 and motor >5, postadmission GCS met inclusion criteria.
d
Admission motor if <6, otherwise postadmission motor.
e
Only AIS regions were collected; converted to ISS by putting AIS spine into ISS thorax.
f
Based on the first CT.

percent with good recovery or moderate disability on the GOSE ac- used logistic regression to construct a propensity model that estimated the
companied by similar reduction in deficit on the other outcome measures. likelihood of being enrolled from a PC16 based on all characteristics listed in
We compared protocol site enrollees with patients enrolled at non- Table 1 (Supplemental Digital Content, Supplemental Statistical
protocol sites regarding participant and injury characteristics using Fisher Methods, http://links.lww.com/NEU/D512). To illustrate the ability of
exact and Mann–Whitney tests as appropriate. Because neither sites nor the propensity model to reduce confounding, we calculated weighted
patients were randomly assigned and the groups differed considerably, we descriptives using inverse probability weighting17 based on the probabilities

474 | VOLUME 92 | NUMBER 3 | MARCH 2023 neurosurgery-online.com

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

 PATIENTS WITH sTBI PATIENTS WITHOUT ICP MONITORING: ICE PROTOCOL

from the propensity model and compared them between groups using propensity model for group (columns labeled “Weighted”),
weighted chi-square or Mann–Whitney tests. characteristics were balanced between groups.
For the primary analyses, we evaluated differences in outcome and Table 2 presents neurobehavioral outcome data. The overall
management process variables using generalized estimating equation follow-up rate for the 6-month GOSE was 80% (78% for the
(GEE) regression modeling, accounting for correlation among partici-
nonprotocol group and 85% for the protocol group); those lost to
pants from the same hospital based on an exchangeable correlation matrix
and used model-based estimates and robust standard errors. Details of
follow-up had data collected through hospital discharge. The results
the primary analyses and sensitivity analyses are available in the Sup- were significantly superior for the protocol group for all neuro-
plemental Digital Content, http://links.lww.com/NEU/D512. We used behavioral outcomes except mortality and SWLS. Figure 2 displays
linear regression for continuous and ordinal outcomes and ranks, logistic the Kaplan-Meier cumulative mortality (100% cumulative survival)
regression for dichotomous outcomes, and Cox regression for survival. We by study group. There was a strong but nonsignificant trend to-
treated quantitative variables as having a linear effect for both propensity ward survival advantage to the protocol group (hazard ratio 0.69,
estimation and regression modeling. We used case-wise deletion for missing P = .118). Figure 3 displays the GOSE at 6 months for the 2 groups,
predictors and outcomes. Mortality and the composite score were available for illustrating significantly better functional outcome in the protocol
all participants. A significance level of .05 was used for GOSE and the group (43% vs 59% with favorable outcome, ie, GOSE 5 or higher,
composite; the Holm-Bonferroni18 method accounted for multiple com- P = .018 comparing percent favorable, P = .001 comparing mean
parisons for other measures. All analyses were performed using SPSS version
GOSE, both from adjusted GEE models.)
26 (IBM) and SAS version 9.4 (SAS Analytics Software and Solutions).
Table 3 presents primary analyses of management process
variables. The increase in CT scans obtained by the protocol
RESULTS group was the only significant difference. Despite some trends
toward increased management efficiency associated with protocol
There were 236 patients in the nonprotocol group and 178 use (shorter ICU length of stay and lower number of mechanical
patients in the protocol group. Figure 1 displays the participant ventilation days, sedation days, hyperventilation days, days with
flow diagram. Table 1 lists group demographics and injury therapy directed at SICH, and total number of brain-specific
characteristics (columns labeled “Unweighted”). The groups treatments), none of the individual variables reached significance,
differed significantly in multiple ways. After weighting using the nor did the trend toward fewer systemic complications.

FIGURE 1. Participant flow diagram.

NEUROSURGERY VOLUME 92 | NUMBER 3 | MARCH 2023 | 475

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

CHESNUT ET AL

TABLE 2. Neurobehavioral Outcomes

Measure N Nonprotocol (N = 236) Protocol (N = 178) Protocol effecta P-valueb P-valuec P-valued

Composite outcomee,f 414 46.8 (24.0) 56.9 (24.5) +12.2 (+4.6, +19.8) .002 .003 —
GOS-E 6 mo 332 3.6 (2.6) 4.7 (2.8) +1.36 (+0.55, +2.17) .001 .001 —
processing speed index
Mortality 6 mo 414 36% 25% HR = 0.69 (0.43, 1.10) .118 — .138
GOS-E ≥5 332 43% 59% OR = 2.48 (1.27, 4.83) .008 — .018
8 8% 19%
7 10% 23%
6 14% 8%
5 10% 9%
4 3% 5%
3 9% 3%
2 2% 3%
1 44% 29%
GOAT dischargef 343 5.5 (31.1) 24.5 (38.9) +24.3 (+10.1, +38.5) .001 .000 .005
GOAT 6 mof 334 33.5 (44.3) 51.0 (44.8) +18.6 (+5.9, +31.3) .004 .003 .014
SWLSf 336 14.7 (11.0) 15.9 (9.4) +2.53 (-1.11, +6.18) .173 .115 .173
RAVLTf 289 14.3 (18.9) 19.7 (18.9) +8.50 (+1.29, +15.71) .021 .018 .029
PSIf 279 57.2 (14.3) 63.3 (17.2) +7.55 (+1.76, +13.34) .011 .002 .019

EDH, epidural hematoma; GOAT, Galveston Orientation and Amnesia Test; PSI, processing speed index; RAVLT, Rey Auditory Verbal Learning Test; SWLS, Satisfaction With Life Scale.
a
Unstandardized regression coefficient reported for GOSE and other continuous measures (linear model); odds ratio reported for dichotomous GOSE; hazard ratio reported for
cumulative mortality (Cox model).
b
Statistical significance by linear/logistic/Cox regression.
c
Statistical significance by rank regression.
d
CNS untestables were assigned a value of one point less than the sample minimum, with deaths one point less than that.
e
The composite outcome consists of 7 measures of outcome (survival, GOAT discharge, GOAT 6 mo, GOSE 6mo, SWLS, RAVLT, PSI), where each score is converted to a sample
percentile (high corresponding to a better outcome) and the resulting percentiles averaged (regardless of the number of missing outcomes).
f
Parametric significance after adjusting for multiple comparisons (Benjamini-Hochberg, m = 7) for secondary outcomes.
g
Analyses adjusted for all variables imbalanced between groups in the unweighted analyses in Table 1 (injury cause, transfer, GCS motor, ISS nonhead, Marshall, cisterns, EDH
procedure, SDH procedure, craniectomy, number of neurosurgeries) as fixed effects, site as clustering variable (GEE). Time to admit is also added as a fixed effect to the GOSE, RAVLT,
and PSI models because of its correlation with missingness.
Regression analyses adjusting for variables imbalanced between groupg using generalized estimating equations (GEE) to account for correlations within hospital. P-values are given
for linear regression (primary analysis), rank regression (sensitivity analysis) for continuous or ordinal outcomes, and P-values adjusting for multiple comparisons for secondary
outcomes. Higher values indicate better outcome for all measures except mortality.

To address the possibility that the analytic methods used biased and nonprotocol groups (Table 1). The differences likely reflect the
our analyses, we include the results of sensitivity analyses on the location (Bolivia) of 4 of the 5 protocol centers. Bolivia has national
ranks of the measures (Tables 2 and 3, second column from left) insurance (el Seguro Obligatorio de Accidentes de Tránsito [SOAT])
and of using regression analysis accounting for propensity scores covering road traffic accidents. Routinely, insurance-funded patients
(eTables 1 and 2, http://links.lww.com/NEU/D512). The only are brought initially to private clinics and transferred to public
P-value shift across the significance threshold was for the sec- hospitals (ie, study centers) when SOAT funding ends (generally after
ondary outcome RAVLT. the inclusion time window has expired). SOAT influencing only
road traffic accidents (RTAs) skews the Bolivian injury mechanism
DISCUSSION profile (and associated cranial and systemic injury types).
Explaining discrepancies does not control for their possible
After accounting for sizeable baseline variable imbalances using influence. It is possible to balance out these differences as il-
covariate adjustment and GEE to account for correlations within lustrated by the propensity-adjusted demographics profile
hospital, this observational study found significantly better (Table 1), which eliminated the significance of all disparate
functional and overall outcome at sites using the ICE protocol. variables. We used GEE to account for correlations within
The trend toward lower mortality did not reach significance. hospital (center effects) in our analyses of outcomes (Table 2)
Differences were in the direction of beneficial efficiency-of-care and process variables (Table 3) that used regression modeling
effects, although these did not reach significance. There was a to minimize the likelihood that differences in demographic or
small but significant increase in CT scan frequency. injury characteristics accounted for our principal findings. Our
The major interpretation issues lie in explaining and controlling sensitivity analyses against our conclusions being an artifact
for the demographic and injury discrepancies between the protocol of our analytic technique include rank regression using the

476 | VOLUME 92 | NUMBER 3 | MARCH 2023 neurosurgery-online.com

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

 PATIENTS WITH sTBI PATIENTS WITHOUT ICP MONITORING: ICE PROTOCOL

FIGURE 2. Kaplan-Meier mortality plot (100% estimated % surviving) showing the cumulative mortality rate over 6 months among patients
managed at nonprotocol vs protocol sites. The inset shows the hazard ratio for death (0.69) and significance level by GEE Cox regression accounting for
variables imbalanced in Table 1 (P = .118). Curves are unadjusted. GEE, generalized estimating equation.

same GEE approach (Table 2) and a separate analysis adjusting frequently impeded by financial and logistic obstructions. Our
for propensity scores (eTable 1, http://links.lww.com/NEU/ impression is that protocolization facilitated overcoming such
D512). There were only minor changes in significance. obstacles to reach a minimal CT imaging frequency.
We believe we have reasonably isolated the influence of having Of course, unmeasured confounders cannot be ruled out.
a protocol on sTBI outcome (Table 2) and on the frequency of Neurotrauma management studies suggest treatment protocol
patient care treatment interventions (Table 3). Patient outcomes implementation is often associated with improved outcomes,19,20
at PCs were improved (Figures 2 and 3 and Table 2). Our primary decreased resource utilization,20 and lower costs.21 Studies ad-
outcome (Composite Score) and 6-month GOSE (continuous dressing protocol adoption based on the Brain Trauma Foundation
and dichotomized) were significantly superior, although the fa- TBI Guidelines,22,23 indexed by ICP monitoring usage, are often
vorable mortality trend did not reach significance. Composite associated with improved outcomes and decreased resource utili-
Score component measures were all significantly better except for zation,24-27 even when a causal relationship with ICP monitoring
SWLS. All process variables, but acute hospital length of stay, were per se is not supported.28 Attributing changes in resource utilization
in the direction of increased efficiency within the PC group for or outcomes to specific aspects of sTBI protocols (eg, ICP moni-
fewer days of mechanical ventilation, hyperventilation, and seda- toring) or their overall focus (eg, “aggressiveness of care”) should be
tion; fewer days involving TBI-specific treatment (not including viewed with caution.29-31 Such caveats are relevant to interpreting the
continued ICU stay for other reasons); and fewer total brain-specific significant improvements in outcome and trends in resource utili-
treatments. Increased efficiency trends were not associated with zation reported here. The PCs had been in the BEST TRIP trial and
neuroworsening. The number of CTs was significantly higher in an overlapping prospective observational trial for more than 5 years. A
the PC group, although the absolute difference was small. nonspecific Hawthorne Effect4,5 would be expected in these centers
The slight but significant increase in CT scanning at PCs is as well as more specific study influences such as vigorous oversight of
worthy of comment. We believe it is actually a benefit of the protocol adherence, adverse events, and outcomes. All PCs had also
protocol. At LMIC centers, obtaining desired CT imaging is had the experience of using ICP monitoring in half of their study

NEUROSURGERY VOLUME 92 | NUMBER 3 | MARCH 2023 | 477

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

CHESNUT ET AL

FIGURE 3. Stacked bar plots of GOSE at 6 months comparing functional outcome in those treated at nonprotocol and protocol sites.
Stacked bars are unadjusted. Comparison is from GEE linear regression accounting for variables imbalanced in Table 1 or imbalanced
regarding missing outcome. GEE, generalized estimating equation.

patients and providing the data suggesting that their nonmonitored group of naive centers new to this study and research in general. In
approach produced a very satisfactory outcome. All such factors addition, unforeseen demographic differences between PCs and
could contribute to PC’s improved outcomes compared with a NPCs could influence outcome. We attempted to account for the

TABLE 3. Management Process Variables

Measure N Nonprotocol (N = 236) Protocol (N = 178) Protocol effecta,b P-valuec P-valued P-valuee

ICU LOS 407 15.9 14.8 1.37 ( 6.08, +3.34) .570 .418 .633
Acute LOS 404 27.1 28.3 0.09 ( 8.76, +8.58) .984 .606 .984
Mechanical vent. days 396 7.84 5.19 1.29 ( 3.49, +0.92) .253 .512 .521
Sedation days 396 3.39 1.31 0.81 ( 2.74, +1.12) .411 .926 .530
Hyperventilation days 351 3.00 1.59 1.26 ( 3.51, +0.99) .274 .397 .521
Neuroworsened 407 16% 15% OR = 0.65 (0.23, 1.87) .424 --- .530
CT’s 409 2.07 2.73 +0.67 (+0.32, +1.03) <.001 <.001 .002
Systemic complication 408 39% 27% OR = 0.71 (0.36, 1.38) .312 — .521
Brain-specific ICU daysf 393 8.50 5.44 1.33 (-3.85, +1.18) .299 .604 .521
Brain-specific tx’sg 396 287 117 116 ( 290, +57) .190 .876 .521
a
Analyses adjusted for all predictors of protocol group (injury cause, transfer, GCS motor, ISS nonhead, Marshall, cisterns, EDH procedure, SDH procedure, craniectomy, number of
neurosurgeries) as fixed effects, site as repeated measure (GEE). Time to admit is also added as a fixed effect to the hyperventilation model because of its correlation with
missingness.
b
Regression coefficient reported for continuous measures (linear model); odds ratio reported for dichotomous measures.
c
Statistical significance by linear or logistic regression.
d
Statistical significance by rank regression.
e
Parametric significance after adjusting for multiple comparisons (Benjamini-Hochberg, m = 10).
f
Defined as the aggregate number of days of brain-specific therapy (after excluding non-BST entries and gaps in the TIL record).
g
Defined as the aggregate number of hourly brain-specific treatments (analogous to AUC), with high-dose therapeutic intensity levels counting double.
Regression analyses adjusting for variables imbalanced between group using generalized estimating equations (GEE) to account for correlations within hospital. P-values are given
for linear regression (primary analysis), rank regression (sensitivity analysis) for continuous or ordinal outcomes, and P-values for linear or logistic regression adjusting for multiple
comparisons.

478 | VOLUME 92 | NUMBER 3 | MARCH 2023 neurosurgery-online.com

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

 PATIENTS WITH sTBI PATIENTS WITHOUT ICP MONITORING: ICE PROTOCOL

latter by adjusting for covariates and using GEE regression models increase in treatment efficiency than those managing them
to control for center effects. This did not eliminate favorable as- without a protocol. Comparable ICUs treating patients with sTBI
sociations between PC status and outcome, differences that were without invasive ICP monitoring should consider protocolizing
not reversed in our sensitivity analyses (Table 2 and eTable 1, care. Adopting the ICE protocol is an option.
http://links.lww.com/NEU/D512).
Those confounding influences potentially related to prior re- Funding
search study involvement would likely combine to look like a National Institute of Health, NINDS—ICE protocol developed with funding
favorable effect of protocolization. Indeed, they may actually be from NIH Grant R01-NS-058302. Dr Chaddock reports funds from Fundación
important aspects to protocolization regarding decreasing practice ALAS. Dr Hendrickson reports funding from Research Councils UK (RCUK). Dr
variability and management uncertainty. If so, then their existence Alanis reports the following: En lo económico, Por cada paciente nos cancelaban,
como también los pasajes para las reuniones en Buenos Aires y la Estadı́a. Dr Merida
does not detract from interpreting our findings as reflecting benefits
reports funds from LABIC foundation.
from initiating and using a protocol. The major problem is dif-
ferentiating whether our findings result from this specific protocol
or are primarily related to protocolization in general. Disclosures
At this point, we cannot speculate on this differentiation. Our The authors have no personal, financial, or institutional interest in any of the
drugs, materials, or devices described in this article.
data suggest caregivers managing patients with sTBI without ICP
monitoring should not expect outcomes equivalent to either the
BEST TRIP group without a formal treatment protocol. We References
support considering the ICE protocol when managing patients with
1. Chesnut RM, Temkin N, Carney N, et al. A trial of intracranial-pressure monitoring
sTBI in resource-limited environments, in light of evidence of its in traumatic brain injury. New Engl J Med. 2012;367(26):2471-2481.
adequacy and the lack of alternate published protocols. We cannot, 2. Chesnut RM, Temkin N, Dikmen S, et al. A method of managing severe traumatic
however, suggest that the specifics of this protocol are necessary to brain injury in the absence of intracranial pressure monitoring: the imaging and
produce similar results as long as some protocol is established. clinical examination protocol. J Neurotrauma. 2018;35(1):54-63.
3. Chesnut RM, Temkin N, Carney N, et al. Supplementary Appendix. A Trial of
We address issues of specific vs general protocolization in phase II Intracranial-Pressure Monitoring in Traumatic Brain Injury; 2012. Accessed October
of our study design, wherein both groups (PC/NPC) adopt a new, 23, 2022. https://www.nejm.org/doi/suppl/10.1056/NEJMoa1207363/suppl_file/
more detailed and comprehensive consensus-derived treatment nejmoa1207363_appendix.pdf.
4. Roethlisberger FJ, Dickson WJ. Management and the Worker. Harvard University
algorithm32 generated for this study. Center effects will be inter- Press; 1939.
nally controlled by this design. Effects of protocolization will be 5. Landsberger H. Hawthorne Revisited. Cornell University Press; 1958.
minimized when comparing results within the current protocol 6. Baker SP, ONeill B. The injury severity score: an update. J Trauma Inj Infect Crit
group but will confound the protocol-specific effects in the current Care. 1976;16(11):882-885.
7. Baker SP, ONeill B, Haddon WJ, Long WB. The injury severity score: a method
nonprotocol group. We will try to isolate protocol-specific effects by for describing patients with multiple injuries and evaluating emergency care.
comparing absolute and relative changes in outcomes and resource J Trauma Inj Infect Crit Care. 1974;14(3):187-196.
utilization within and across the 2 cohorts for both phases. 8. Marshall LF, Marshall SB, Klauber MR, et al. A new classification of head
injury based on computerized tomography. J Neurosurg. 1991;75(suppl):
These results do not promote sTBI management without ICP
S14-S20.
monitoring. Although BEST TRIP demonstrated similar outcomes for 9. Levin HS, O’Donnell VM, Grossman RG. The Galveston Orientation and Amnesia Test. A
patients managed with an ICP monitor–based protocol vs aggressive practical scale to assess cognition after head injury. J Nervous Ment Dis. 1979;167(11):
SICH management without monitoring,1 one of the recommended 675-684.
10. O’Brien PC. Procedures for comparing samples with multiple end points. Bio-
interpretations was that centers currently using ICP monitoring should metrics. 1984;40(4):1079-1087.
not change their practice.33 However, nonmonitored management 11. Wilson JL, Pettigrew LE, Teasdale GM. Structured interviews for the Glasgow
predominates in LMICs, and little relevant literature exists to guide Outcome Scale and the extended Glasgow Outcome Scale: guidelines for their use.
treatment of most patients with sTBI worldwide. Our study should be J Neurotrauma. 1998;15(8):573-585.
12. Diener E, Emmons RA, Larsen RJ, Griffin S. The satisfaction with Life scale.
readily generalizable to at least the middle-income spectrum of these J Personal Assess. 1985;49(1):71-75.
LMICs where we would suggest protocolization is likely to improve 13. Rey A. L’examen clinique en psychologie. Presses Universitaires de France; 1964.
outcome (although the effect magnitude cannot be predicted). Al- 14. Climie EA, Rostad K. Test review: wechsler adult intelligence scale. J Psychoeducational
Assess. 2011;29(6):581-586.
though a sizeable percentage of patients with sTBI meeting the 15. Gehan EA. A generalized wilcoxon test for comparing arbitrarily singly-censored
guidelines-based monitoring criteria are treated without monitoring in samples. Biometrika. 1965;52(1/2):203-223.
HICs,34-38 we caution against the readiness of generalization of our 16. Rosenbaum PR, Rubin DB. The central role of the propensity score in obser-
data to nonmonitored patients treated in HICs. vational studies for causal effects. Biometrika. 1983;70(1):41-55.
17. Horvitz DG, Thompson DJ. A generalization of sampling without replacement
from a finite universe. J Am Stat Assoc. 1952;47(260):663-685.
18. Holm S. A simple sequentially rejective multiple test procedure. Scand J Stat. 1979;6(2):
CONCLUSION 65-70.
19. Clayton TJ, Nelson RJ, Manara AR. Reduction in mortality from severe head injury
LMIC ICUs managing patients with sTBI using the ICE following introduction of a protocol for intensive care management. Br J Anaesth.
protocol had better functional outcome and trended toward an 2004;93(6):761-767.

NEUROSURGERY VOLUME 92 | NUMBER 3 | MARCH 2023 | 479

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.

CHESNUT ET AL

20. Spain DA, McIlvoy LH, Fix SE, et al. Effect of a clinical pathway for severe 37. Sahjpaul R, Girotti M. Intracranial pressure monitoring in severe traumatic brain
traumatic brain injury on resource utilization. J Trauma Inj Infect Crit Care. 1998; injury--results of a Canadian survey. Can J Neurol Sci. 2000;27(2):143-147.
45(1):101-105; discussion 104-105. 38. Stocchetti N, Penny KI, Dearden M, et al. Intensive care management of head-
21. Faul M, Wald MM, Rutland-Brown W, Sullivent EE, Sattin RW. Using a cost- injured patients in Europe: a survey from the European brain injury consortium.
benefit analysis to estimate outcomes of a clinical treatment guideline: testing Intensive Care Med. 2001;27(2):400-406.
theBrain Trauma Foundation guidelines for the treatment of severe traumatic brain
injury. J Trauma Inj Infect Crit Care. 2007;63(6):1271-1278.
22. Bullock R, Chesnut RM, Clifton G, et al. Guidelines for the management of severe Acknowledgments
head injury. J Neurotrauma. 1996;13(11):643-678.
This work was supported by NIH Grant R01-NS080648. The ICE protocol
23. Carney N, Totten AM, O’Reilly C, et al. Guidelines for the management of severe
traumatic brain injury, fourth edition. Neurosurgery. 2017;80(1):6-15.
was developed with funding from NIH Grant R01-NS-058302. The funders had
24. Arabi YM, Haddad S, Tamim HM, et al. Mortality reduction after implementing a no role in the conduct or reporting of the study (ClincialTrials.gov #44679). The
clinical practice guidelines-based management protocol for severe traumatic brain authors would like to express their thanks to Dr Juan Antonio Guerra Garcia for his
injury. J Crit Care. 2010;25(2):190-195. contributions to this study and this report, in the light of his having passed on 30
25. Bulger EM, Nathens AB, Rivara FP, et al. Management of severe head injury: May, 2020. The CBG Study Contributors: Erik Garcia, MD, Marı́a del Carmen
institutional variations in care and effect on outcome. Crit Care Med. 2002;30(8): Valverde, Elva Nogal, Hospital San Juan de Dios, Santa Cruz de la Sierra, Bolivia;
1870-1876. Carla Mérida, MD, Marı́a del Carmen Valverde, Hospital Japonés, Santa Cruz de
26. Fakhry SM, Trask AL, Waller MA, Watts DD. Management of brain-injured patients la Sierra, Bolivia; Marı́a Pérez, MD, Jesusa Torrez, Hospital Viedma, Cochabamba,
by an evidence-based medicine protocol improves outcomes and decreases hospital
Bolivia; Isabel Navajas, MD, Fernando Justiniano, Rafael Angel Castillo, Hospital
charges. J Trauma Inj Infect Crit Care. 2004;56(3):492-500; discussion 499-500.
27. Stein SC, Georgoff P, Meghan S, Mirza KL, El Falaky OM. Relationship of
San Juan de Dios, Tarija, Bolivia; Alexander Rodriguez, MD, Karen Perales,
aggressive monitoring and treatment to improved outcomes in severe traumatic Robinson Barrios, Juan de la Torre, Fundación Clı́nica Campbell, Barranquilla,
brain injury. J Neurosurg. 2010;112(5):1105-1112. Colombia; Javier Campos, MD, Carmen Montoya, Clı́nica Universitaria Rafael
28. Haddad S, Aldawood AS, Alferayan A, Russell NA, Tamim HM, Arabi YM. Uribe, Cali, Colombia; Marco Camacho, MD, Maria Fernandez, Tania Guzmán,
Relationship between intracranial pressure monitoring and outcomes in severe Hospital Erasmo Meoz ICU #1, Cúcuta, Colombia; Jorge Alberto Castro, MD,
traumatic brain injury patients. Anaesth Intensive Care. 2011;39(6):1043-1050. Carlos Ibarra, Hospital Erasmo Meoz ICU #2, Cúcuta, Colombia; Soraya Puertas,
29. Mauritz W, Steltzer H, Bauer P, Dolanski-Aghamanoukjan L, Metnitz P. Monitoring MD, Katty Trelles, Hospital José Carrasco Artega, Cuenca, Ecuador; Veruska
of intracranial pressure in patients with severe traumatic brain injury: an Austrian Zambrano, MD, Yannely Carrero, MD, Rosangel Leal, Dany Giraldo, Hospital
prospective multicenter study. Intensive Care Med. 2008;34(7):1208-1215.
Luis Razetti, Barinas, Venezuela; Germary León, MD, Gregolina Rojas
30. McKinley BA, Parmley CL, Tonneson AS. Standardized management of intra-
cranial pressure: a preliminary clinical trial. J Trauma Inj Infect Crit Care. 1999;
Hernández, Hospital Luis Razetti, Barcelona, Venezuela; Julher Rivera, MD,
46(2):271-279. Selene Espina, Hospital Dr Patrocinio Penuela Ŕuiz, San Cristóbal, Venezuela;
31. Schirmer-Mikalsen K, Moen KG, Skandsen T, Vik A, Klepstad P. Intensive care Marcia Carrara, Pablo Carrara, Hospital Delicia Concepción Masvernat, Con-
and traumatic brain injury after the introduction of a treatment protocol: a cordia, Entre Rı́os, Argentina; Fernando Cerrudo, MD, Alicia Román, Hospital
prospective study. Acta Anaesthesiologica Scand. 2013;57(1):46-55. Justo José de Urquiza, Concepción del Uruguay, Entre Rı́os, Argentina; Marı́a
32. Chesnut RM, Temkin N, Videtta W, et al. Consensus-Based Management Pro- Filippi, MD, Alejandro Gamboa, MD, Santiago Izza, Hospital Dr Jose Maria
tocol (CREVICE Protocol) for the treatment of severe traumatic brain injury based Cullen, Santa Fe, Argentina; Marta Beatriz Heredia, MD, Hospital Córdoba,
on imaging and clinical examination for use when intracranial pressure monitoring Córdoba, Argentina; Guillermo Soria, MD, Mariana Vera, Hospital Central,
is not employed. J Neurotrauma. 2020;37(11):1291-1299.
Mendoza, Argentina; Virginia Lopez, MD, Guillermina Arellano, Hospital San
33. Chesnut RM, Bleck TP, Citerio G, et al. A consensus-based interpretation of the
benchmark evidence from South American trials: treatment of intracranial pressure
Felipe, San Nicolás, Buenos Aires, Argentina; Bismarck Ibáñez, MD, Silvia
trial. J Neurotrauma. 2015;32(22):1722-1724. Fernandez, Hospital COSSMIL Militar, La Paz, Bolivia.
34. Alali AS, Fowler RA, Mainprize TG, et al. Intracranial pressure monitoring in severe
traumatic brain injury: results from the American College of Surgeons Trauma Supplemental digital content is available for this article at neurosurgery-online.com.
Quality Improvement Program. J Neurotrauma. 2013;30(20):1737-1746.
Supplemental Statistical Methods: a. Regression modeling; b. sensitivity analyses.
35. Cnossen MC, Scholten AC, Lingsma HF, et al. Adherence to guidelines in adult
patients with traumatic brain injury: a living systematic review. J Neurotrauma. eTable 1. Sensitivity analyses for neurobehavioral outcomes.
2021;38(8):1072-1085. eTable 2. Sensitivity analysis for management process variables.
36. Myburgh JA, Cooper DJ, Finfer SR, et al. Epidemiology and 12-month outcomes
from traumatic brain injury in Australia and New Zealand. J Trauma Inj Infect Crit CNS Journal Club Podcast and CME Exams available at cns.org/podcasts.
Care. 2008;64(4):854-862.

480 | VOLUME 92 | NUMBER 3 | MARCH 2023 neurosurgery-online.com

© Congress of Neurological Surgeons 2022. Unauthorized reproduction of this article is prohibited.