Original Investigation | Emergency Medicine

Derivation and Validation of a Score for Predicting Poor Neurocognitive Outcomes

in Acute Carbon Monoxide Poisoning
Sung Hwa Kim, BS; Yoonsuk Lee, MD; Soo Kang, MD; Jin Hui Paik, MD; Hyun Kim, MD, PhD; Yong Sung Cha, MD

Abstract Key Points

Question Can a novel clinical scoring
IMPORTANCE Preventing neurocognitive sequelae is a major goal of treating acute carbon
system predict poor neurocognitive
monoxide (CO) poisoning. There is a lack of reliable score systems exist for assessing the probability
outcomes after acute carbon monoxide
of these sequelae.
poisoning?

OBJECTIVE To develop and validate a novel clinical scoring system for predicting poor Findings This prognostic study
neurocognitive outcomes after acute CO poisoning. developed and externally validated a
prediction model including 5 risk factors
DESIGN, SETTING, AND PARTICIPANTS This prognostic study included derivation and validation associated with poor neurocognitive
cohorts based on consecutive patient data prospectively collected at university hospitals from outcome at 1 month, creatine kinase
January 2006 to July 2021 in Wonju, Republic of Korea, and from August 2016 to June 2020 in level, hyperbaric oxygen therapy,
Incheon, Republic of Korea. Participants included individuals aged 16 years or older admitted with CO Glasgow Coma Scale score, age, and
poisoning. Data were analyzed from October 2021 to January 2022. shock (COGAS score), among patients
with carbon monoxide poisoning.
EXPOSURES Clinical and laboratory variables. COGAS score showed excellent
discrimination performance.
MAIN OUTCOMES AND MEASURES The outcome of interest was neurocognitive sequelae at 4
Meaning These findings suggest that
weeks after CO poisoning. Logistic regression models were used to identify predictors of poor
use of a reliable prediction model during
neurocognitive outcomes in the derivation cohort. Outcomes were assessed using the Global
the early phase of carbon monoxide
Deterioration Scale [GDS] at 1-month after CO exposure and classified as good (1-3 points) or poor
poisoning could help identify patients at
(4-7 points).
risk of poor neurocognitive sequelae.

RESULTS A total of 1282 patients (median [IQR] age, 47.0 [35.0-59.0] years; 810 [63.2%] men) were
assessed, including 1016 patients in the derivation cohort and 266 patients in the validation cohort. + Supplemental content
The derivation cohort included 126 patients (12.4%) with poor GDS scores. Among 879 patients in Author affiliations and article information are
the derivation cohort with 1-year follow-up data, 757 (86.1%) had unchanged GDS scores, 102 (11.6%) listed at the end of this article.

had improved GDS scores, and 20 (2.3%) had worsened GDS scores. In the final prediction model,
age older than 50 years (1 point), Glasgow Coma Scale score of 12 or less (1 point), shock (1 point),
serum creatine kinase level greater than 320 U/L at emergency department presentation (1 point),
and no use of hyperbaric oxygen therapy (1 point) remained factors significantly associated with
worse outcome; therefore, this scoring system was called COGAS (creatine kinase, hyperbaric oxygen
therapy, Glasgow Coma Scale, age, shock). Area under the receiver operating characteristic curve for
COGAS score was 0.862 (95% CI, 0.828-0.895) for the derivation cohort and 0.870 (95% CI,
0.779-0.961) for the validation cohort.

CONCLUSIONS AND RELEVANCE These findings suggest that assessing the COGAS score during
the early phase of CO poisoning may help identify patients at risk of poor neurocognitive sequelae.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552

Open Access. This is an open access article distributed under the terms of the CC-BY License.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 1/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Introduction
Annually, approximately 50 000 patients with carbon monoxide (CO) poisoning present to
emergency departments (EDs) of US hospitals, with 1500 deaths.1-3 Approximately 15 000
intentional CO poisonings annually account for more than two-thirds of reported deaths2,4,5 and
cause neurocognitive sequelae among survivors.6,7 Of patients with CO poisoning treated with
normobaric oxygen, nearly half develop cognitive sequelae after 6 weeks.6 Some possibly
permanent sequelae include gait and motor disturbances, peripheral neuropathy, hearing loss and
vestibular abnormalities, dementia, and psychosis.7 The societal costs (in direct hospital costs and
lost earnings) of accidental CO poisoning are estimated at more than $1.3 billion annually in the US.8
Hyperbaric oxygen (HBO) therapy within 24 hours after poisoning is recommended for symptomatic
patients with CO poisoning.1,6,9,10 In a 2002 study by Weaver,6 treatment with HBO reduced
neurocognitive sequelae incidence at 6 weeks by half and from 25% to 18% at the 1-year evaluation,
compared with no such treatment. However, neurocognitive sequelae developed in 18% of patients
even with HBO use in acute CO poisoning. It is important to be able to predict a poor neurocognitive
prognosis in acute CO poisoning to initiate rehabilitative interventions early.11
Clinical and laboratory variables suggested for predicting poor neurocognitive prognosis6,7,12-20
include older age, underlying cardiovascular disease, any interval loss of consciousness, acidosis,
Glasgow Coma Scale (GCS) score, serum creatine kinase level, longer CO exposure intervals, serum
lactate level, and HBO treatment. As there is no validated simple and accurate clinical scoring model
for stratifying patients according to their risk of poor neurocognitive prognosis in early phase, we
aimed to develop and validate a simple, accurate clinical scoring model for predicting poor
neurocognitive outcomes at 1 month after CO poisoning.

Methods
This prognostic study was approved by the institutional review boards at Wonju Severance Christian
Hospital and Inha University Hospital. In the derivation cohort, data after August 2020 were
collected with informed consent for a prospective cohort study. For data before August 2020 in the
derivation cohort and all data in the validation cohort, the requirement for informed consent was
waived because we analyzed retrospectively with prospectively collected registry data. This study
adhered to the ethical guidelines of the Declaration of Helsinki.21 This study followed the Transparent
Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis (TRIPOD)
reporting guideline.

Study Cohorts
The study data were derived from independent cohorts at 2 tertiary academic hospitals in the
Republic of Korea. The derivation cohort included those with acute CO poisoning enrolled between
January 2006 and July 2021 at Wonju Severance Christian Hospital, where a CO poisoning registry
was opened to prospectively collect consecutive patient data. Data from January 2006 to July 2020
were prospectively obtained from this registry; after August 2020, data were collected prospectively
after individuals or legal guardians provided their informed consent for the Carbon Monoxide
Intoxication in Korea: Prospective Cohort (CARE CO Cohort; ClinicalTrials.gov identifier:
NCT04490317). Patients were followed up until August 2021. The validation cohort included
registry data from patients with CO poisoning collected prospectively from August 2016 to June
2020 at the Inha University Hospital, Incheon. These institutions have previously interacted to
conduct various multicenter studies on CO poisoning; therefore, the patients were registered using
similar registration forms. Inha University Hospital has been implementing this since 2016. We
anonymized patients’ data before analyses.
At both hospitals, acute CO poisoning is diagnosed based on patient history and
carboxyhemoglobin (HbCO) level of greater than 5% (>10% for smokers; to convert to proportion of

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 2/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

1.0, multiply by 0.01). We treated patients with CO poisoning with 100% oxygen therapy through a
face mask with a reservoir bag. Patients with any loss of consciousness interval, neurocognitive
symptoms and signs, cardiovascular dysfunction, elevated cardiac enzymes, ischemic
electrocardiogram changes, severe acidosis, or HbCO of 25% or greater were treated with HBO.1
During the first HBO session, initial compression was performed to 2.8 atmospheres absolute for 45
minutes, followed by 2.0 atmospheres absolute for 60 minute, similar to the protocol described by
Thom et al.10

Data Extraction
For widespread applicability, we limited variables to known and potential risk factors6,7,12-20 and
commonly used variables in acute CO poisoning. We evaluated patient age, sex, poisoning intentions,
CO source (ie, charcoal, oil and gas, or fire), drug coingestion, GCS score at the site of rescue or ED
arrival, comorbidities (ie, diabetes, hypertension, cardiovascular disorder, and psychiatric disease),
current smoking status and alcohol coingestion, interval of loss of consciousness, shock, seizure, and
application of HBO (eAppendix 1 in the Supplement). We excluded CO exposure time from the main
scoring model, requiring only objective indicators, as it is difficult to measure CO exposure time
accurately in clinical settings for many patients. Laboratory variables assessed in the ED included
HbCO, bicarbonate, lactate, creatinine, creatine kinase, and troponin I levels. We excluded patients
with missing clinical and laboratory values.
CO-related neurocognitive outcomes were measured using the Global Deterioration Scale (GDS)
score (range, 1-7) (eAppendix 2 and eTable in the Supplement).22 For patients with GDS scores of 1
to 3 points, cognitive impairment ranged from none (GDS score, 1 point) to mild (GDS score, 3
points), objectively substantiated by a detailed interview by an experienced rehabilitation physician.
Patients with GDS scores of 4 to 7 points ranged from clear cognitive impairment (GDS score, 4
points) to loss of motor skills (including walking) and loss of all language skills except for inaudible,
unintelligible sounds (GDS score, 7 points). Patients who died within 1 month were assigned a GDS
score of 7 points, the most severe score. We investigated GDS stage at 1 month after CO exposure
through visiting the rehabilitation outpatient department. We assumed neurocognitive sequelae
caused by CO poisoning rarely develop after 4 to 6 weeks.6,7,10,23,24 For patients unable to visit
outpatient departments, their guardians were interviewed. We classified GDS scores as good (1-3
points) or poor (4-7 points)25 and investigated changes in the 1-month GDS score at 1 year after CO
poisoning.

Variable Selection and Score Construction

For the practical application of scores, a receiver operating characteristic curve analysis was
performed to determine the cutoff value for converting a continuous variable to a categorical
variable (eFigure 1 in the Supplement). Using MedCalc version 11.6.1 (MedCalc Software), the cutoff
values were obtained using the maximum sensitivity and specificity based on the Youden Index. To
minimize potential collinearity and overfitting of variables, multivariable logistic regression using
backward elimination included variables at P < .05 in the univariate analysis.26 Finally, 5 variables
were included in the scoring system, the accuracy of which was assessed by the area under the
receiver operating characteristic curve (AUC). Model calibration was assessed using the Hosmer-
Lemeshow goodness of fit test, and calibration plots were visually evaluated using 200 bootstrap
resamples.

Internal Validation
To reduce overfit bias, internal validation of the prediction model was assessed using 2 methods:
10-fold cross-validation and bootstrapping method (200 iterations). The 10-fold cross-validation was
used by dividing the training set into 10 mutually exclusive parts. Consequently, the predictive
performance of COGAS score was reassessed with 200 iterations of bootstrap resampling. The

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 3/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

bootstrapping method can be constructed and validated by 100% of all patients and is more efficient
than validation methods using split.27

External Validation
After developing the scoring system, external validation was performed on different data. The
observed and estimated probabilities of poor neurocognitive outcomes were compared in each
COGAS score. Then, the discrimination was assessed using the AUC, and calibration was assessed
using the calibration plot.

Statistical Analysis
Data are reported as means with SDs for continuous variables and numbers with percentages for
categorical variables. Differences between the derivation and validation cohorts were assessed using
the independent t test or Mann-Whitney U test for continuous variables and χ2 test or Fisher exact
test for categorical variables. DeLong test was used to confirm whether the AUCs of the 2 models
were statistically significantly different. All statistical analyses used SAS statistical software version
9.4 (SAS Institute) and R version 3.6.3 (R Project for Statistical Computing). A 2-sided P < .05
indicated statistical significance. Data were analyzed from October 2021 to January 2022.

Results
Study Population
A total of 1282 patients (median [IQR] age, 47.0 [35.0-59.0] years; 810 [63.2%] men) were assessed,
including 1016 patients in the derivation cohort and 266 patients in the validation cohort (eFigure 2
in the Supplement). The derivation cohort included 126 patients (12.4%) with poor GDS scores
(Table 1). The median (IQR) age in the derivation cohort was 48.0 (36.0-60.5) years, and 635
(62.5%) were men. Charcoal (758 patients [74.6%]) was the most common CO source, and the
median (IQR) GCS score at the scene or ED was 15.0 (12.0-15.0). Hypertension (197 patients [19.4%])
was the most common comorbidity, and HBO therapy was provided to 869 patients (85.5%).
Compared with the validation cohort, patients in the derivation cohort were younger, were less likely
to receive HBO therapy, and had lower creatine kinase levels, whereas they were more likely to have
intentional CO poisoning, drug coingestion, psychiatric disorders, higher lactate levels, and higher
troponin I levels (Table 1). Overall, 15 patients (1.2%) died within 1 month after CO poisoning while
hospitalized. Cause of death was uncontrolled metabolic acidosis, acute kidney injury, or profound
shock, except for 1 patient with aortic dissection and who was bedridden (GDS score, 7 points). We
investigated neurological impairment because GDS may be difficult to reflect it. In our study, among
9 patients with neurological symptoms (motor weakness, speech disturbance, and peripheral
neuropathy, such as foot drop), 3 had poor outcomes (with GDS scores of 4-6), 5 had good outcomes
(initial neurological symptoms resolved), and the remaining patient in the favorable outcome group
had initial persistent foot drop (peripheral neuropathy) at 1 month, which returned to normal at 10
months after CO poisoning.
We investigated whether the 1-month GDS score changed after 1 year (879 patients were
followed-up for 1 year) after CO poisoning in the derivation cohort. In 757 patients (86.1%) GDS scores
remained unchanged, 102 patients (11.6%) had improved GDS scores, and 20 patients (2.3%) had
worse GDS scores.

Factors Associated With Poor Neurocognitive Outcomes

In univariate analysis, all variables except sex, drug coingestion, cardiovascular disease, psychiatric
disorder, alcohol coingestion, and seizure were associated with poor neurocognitive outcomes. In
multivariable analysis, intentionality, CO source, diabetes, hypertension, current smoker status, loss
of consciousness, HbCO, and levels of serum bicarbonate, lactate, creatinine, and troponin I were not
statistically significant.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 4/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Model 1 (main model) included creatine kinase (laboratory value). Model 2 excluded the
laboratory variable of model 1, which may not be immediately available and allows a system without
laboratory measurements.
In model 1, age older than 50 years (odds ratio [OR], 4.06; 95% CI, 2.53-6.53), GCS score of 12
or less (OR, 6.06; 95% CI, 3.44-10.68), shock (OR, 3.65; 95% CI, 1.64-10.68), no HBO treatment (OR,
2.41; 95% CI, 1.32-4.40), and serum creatine kinase at the ED greater than 320 U/L (to convert to
microkatals per liter, multiply by 0.0167; OR, 5.06; 95% CI, 3.15-8.13) remained significant (Table 2).
We named this scoring system COGAS for the 5 elements: creatine kinase, hyperbaric oxygen
therapy, Glasgow Coma Scale score, age, and shock. The distribution of COGAS score in the cohorts
is presented in eFigure 3 in the Supplement). In multivariable analysis in model 2, poor
neurocognitive outcomes occurred in patients aged older than 50 years (OR, 4.27; 95% CI,
2.71-6.74), with GCS scores of 12 or less (OR, 10.49; 95% CI, 6.12-17.99), with shock (OR, 4.55; 95% CI,
2.09-9.90), and without HBO treatment (OR, 2.49; 95% CI, 1.42-4.36).

Table 1. Baseline Characteristics of Patients in the Derivation and Validation Cohorts

Patients, No. (%)

Total Derivation cohort Validation cohort
Variables (N = 1282) (n = 1016) (n = 266) P value
Age, median (IQR), y 47.0 (35.0-59.0) 48.0 (36.0-60.5) 44.0 (31.0-55.0) <.001
Sex
Women 472 (36.8) 381 (37.5) 91 (34.2)
Men 810 (63.2) 635 (62.5) 175 (65.8) .32
Intentionality 520 (40.6) 381 (37.5) 139 (52.3) <.001
CO source

Charcoal 952 (74.3) 758 (74.6) 194 (72.9)

Gas 167 (13.0) 126 (12.4) 41 (15.4) .40
Fire 163 (12.7) 132 (13.0) 31 (11.7)
Drug coingestion 124 (9.7) 73 (7.2) 51 (19.2) <.001
GCS score, median (IQR) 15.0 (12.0-15.0) 15.0 (12.0-15.0) 15 (13.0-15.0) .006
Comorbidities
Diabetes 136 (10.6) 111 (10.9) 25 (9.4) .47
Hypertension 238 (18.6) 197 (19.4) 41 (15.4) .14
Cardiovascular disease 50 (3.9) 43 (4.2) 7 (2.6) .23
Psychiatric disease 180 (14.0) 126 (12.4) 54 (20.3) .001
Alcohol coingestion 225 (17.6) 182 (17.9) 43 (15.8) .51
Current smoker 503 (39.2) 387 (38.1) 116 (43.6) .10
Symptoms at ED
Loss of consciousness 767 (59.8) 618 (60.8) 149 (56.0) .15
Shock 50 (3.9) 36 (3.5) 14 (5.3) .20
Seizure 17 (1.3) 13 (1.3) 4 (1.5) .77
Use of HBO 1070 (83.5) 869 (85.5) 201 (75.6) <.001
Laboratory findings, median (IQR)
HbCO, % 19.0 (7.5-31.4) 18.7 (7.4-30.4) 21.3 (8.3-35.2) .07 Abbreviations: CO, carbon monoxide; HbCO,
carboxyhemoglobin; ED, emergency department; GCS,
Bicarbonate, mEq/L 21.4 (18.8-23.4) 21.4 (18.8-23.3) 22.0 (18.0-24.0) .17
Glasgow Coma Scale; GDS, Global Deterioration Scale;
Lactate, mg/dL 20.72 (12.61-32.43) 18.92 (12.61-31.53) 23.42 (14.41-44.14) <.001
HBO, hyperbaric oxygen.
Creatinine, mg/dL 0.8 (0.7-1.0) 0.8 (0.7-1.0) 0.8 (0.7-1.1) .09
SI conversion factors: To convert bicarbonate to
Creatine kinase, U/L 146.5 (92.0-360.0) 149.0 (94.5-393.0) 131.5 (87.0-298.0) .02 millimoles per liter, multiply by 1; creatine kinase to
Troponin I, ng/mL 0.07 (0.02-0.32) 0.02 (0.02-0.33) 0.1 (0.10-0.26) <.001 microkatals per liter, multiply by 0.0167; creatinine to
GDS micromoles per liter, multiply by 76.25; lactate to
Good (1-3) 1129 (88.1) 890 (87.60) 239 (89.9) millimoles per liter, multiply by 0.111; HbCO to
.31 proportion of 1.0, multiply by 0.01; troponin I to
Poor (4-7) 153 (11.9) 126 (12.4) 27 (10.2)
micrograms per liter, multiply by 1.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 5/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Performance and Calibration of the COGAS Score

The AUC for COGAS score was calculated for both cohorts (Figure 1). For the derivation cohort, the
AUC was 0.862 (95% CI, 0.828-0.895), and for the validation cohort, the AUC was 0.870 (95% CI,
0.779-0.961), indicating excellent discriminatory performance (Figure 1). The Hosmer-Lemeshow
goodness of fit test yielded χ2 = 2.61 (P = .27). Internal validation of COGAS score demonstrated
good discrimination in the derivation cohort databased 10-fold cross-validation analysis (AUC, 0.864;
95% CI, 0.852-0.875) and the bootstrapping method (AUC, 0.862; 95% CI, 0.829-0.896).
Calibration plots of COGAS score for each cohort showed good agreement between the predicted
and observed outcomes, with close approximation (eFigure 4 in the Supplement). The original
intercept and slope of the calibration plots were 0 and 1, respectively.
We compared the models, and model 2 score still revealed a high discrimination (AUC, 0.812;
95% CI, 0.787-0.836) compared with model 1 (0.862). The performance of the models developed
with and without creatine kinase (a laboratory value) was significantly different (eFigure 5 in the
Supplement).

Scoring Model Validation

The observed 1-month probabilities of poor neurocognitive outcomes based on COGAS score were
similar to those of the derivation and validation cohorts (Table 3; eFigure 6 in the Supplement). The
estimated 1-month probabilities of COGAS score–based poor neurocognitive outcomes were: 0%
for a score of 0, 3.2% for a score of 1, 15.1% for a score of 2, 48.4% for a score of 3, 83.2% for a score

Table 2. Factors Associated With Poor Neurocognitive Outcomes in the Derivation Cohort

OR (95% CI)
Variables Unadjusted (model 1) Adjusted (model 2) Score
Age >50 y 3.53 (2.35-5.31) 4.06 (2.53-6.53) 1
Male sex 0.87 (0.59-1.27) NA NA
Intentionality 1.50 (1.03-2.18) NA NA
Source
Charcoal 1 [Reference] NA
Oil and gas 0.29 (0.12-0.67) NA NA
Fire 0.37 (0.18-0.78) NA NA
Drug coingestion 0.99 (0.48-2.05) NA NA
GCS score ≤12 11.12 (6.62-18.65) 6.06 (3.44-10.68) 1
Comorbidities
Diabetes 1.78 (1.06-2.98) NA NA
Hypertension 2.17 (1.44-3.28) NA NA
Cardiovascular disease 1.66 (0.75-3.66) NA NA
Psychiatric disease 1.40 (0.83-2.35) NA NA
Alcohol coingestion 1.03 (0.48-2.22) NA NA
Current smoker 0.59 (0.39-0.89) NA NA
Symptoms at ED
Loss of consciousness 6.63 (3.68-11.94) NA NA
Shock 11.67 (5.84-23.32) 3.65 (1.64-10.68) 1
Seizure 3.21 (0.97-10.58) NA NA Abbreviations: HbCO, carboxyhemoglobin; ED,
emergency department; GCS, Glasgow Coma Scale;
No use of HBO 2.07 (1.31-3.25) 2.41 (1.32-4.40) 1
HBO, hyperbaric oxygen; NA, not applicable; OR,
Laboratory findings
odds ratio.
HbCO >37.4% 1.80 (1.13-2.88) NA NA
SI conversion factors: To convert bicarbonate to
Bicarbonate ≤19.5 mEq/L 3.58 (2.44-5.24) NA NA millimoles per liter, multiply by 1; creatine kinase to
Lactate >18.02 mg/dL 2.35 (1.56-3.53) NA NA microkatals per liter, multiply by 0.0167; creatinine to
Creatinine >1.0 mg/dL 4.46 (3.03-6.56) NA NA micromoles per liter, multiply by 76.25; lactate to
Creatine kinase >320 U/L 9.84 (6.43-15.04) 5.06 (3.15-8.13) 1 millimoles per liter, multiply by 0.111; HbCO to
proportion of 1.0, multiply by 0.01; troponin I to
Troponin I >0.113 ng/mL 8.66 (5.58-13.44) NA NA
micrograms per liter, multiply by 1.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 6/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

of 4, and 96.3% for a score of 5 (Figure 2). The observed and estimated probabilities of poor
neurocognitive outcome were strongly correlated (r = 0.989; R2 = 0.978; P < .001). Similarly, model
2 was well calibrated and validated and, thus could be used without laboratory variables (eFigure 7
and eFigure 8 in the Supplement).

Figure 1. Area Under the Receiver Operating Characteristic Curve (AUC) of the Score Model in the Derivation
and Validation Cohorts

COGAS derivation
1.0
(AUC, 0.86; 95% CI,
0.83-0.90)

COGAS validation
0.8
(AUC, 0.87; 95% CI,
0.78-0.96)

0.6
Sensitivity

0.4

0.2

COGAS indicates the prediction score calculated on

0
1.0 0.8 0.6 0.4 0.2 0 creatine kinase, hyperbaric oxygen therapy, Glasgow
Specificity Coma Scale, age, and shock.

Table 3. Poor Outcomes Evidenced by COGAS Score in the Derivation and Validation Cohortsa

Derivation cohort (n = 1016) Validation cohort (n = 266)

Poor outcomes, Observed poor Poor outcomes, Observed poor
COGAS score No./total No. outcomes, % No./total No. outcomes, %
0 3/237 1.3 2/86 2.3
1 10/398 2.5 3/112 2.7
2 38/233 16.3 4/43 9.3 Abbreviation: COGAS, creatine kinase, hyperbaric
oxygen therapy, Glasgow Coma Scale, age, shock.
3 50/116 43.1 8/14 57.1
a
4 20/26 76.9 7/8 87.5 The probability of poor outcomes 1 month after
carbon monoxide exposure increased significantly
5 5/6 83.3 3/3 100.0
with increasing COGAS scores (P for trend < .001).

Figure 2. Estimated Probability of 1-Month Poor Neurocognitive Outcomes in Patients With Acute Carbon
Monoxide Poisoning

100
Estimated probability of poor outcomes, %

80

60

40

20

COGAS indicates the prediction score calculated on

0
0 1 2 3 4 5 creatine kinase, hyperbaric oxygen therapy, Glasgow
COGAS score Coma Scale, age, and shock.

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 7/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Additional Analysis
We compared the diagnostic power of the model including CO exposure time compared with model 1
in the derivation cohort. The explanatory power of model 1 (0.862) was similar to that of the model
with CO exposure time (AUC, 0.869; 95% CI, 0.836-0.902; P = .31) (eFigure 9 in the Supplement). In
addition, we analyzed the verification power of COGAS score in patients with poisoning by oil and
gas combustion, which are the main sources of CO poisoning in Western countries.5,28-30 Oil and gas
combustion was the source of CO poisoning for relatively few patients in the derivation cohort. The
COGAS model tested in the oil and gas combustion-exposed subpopulation had high discrimination
(eFigure 10 in the Supplement). We analyzed the verification power of COGAS score even if outcome
was classified based on GDS score of 2 points. Furthermore, we created a prognostic prediction
model based on two 2 GDS points similarly, and the results included all variables used to calculate
COGAS score (eFigure 11 and eFigure 12 in the Supplement).

Discussion
In this prognostic study, we first developed a new scoring model to stratify the risk of poor
neurocognitive outcomes based on 5 variables at the early phase of acute CO poisoning: older age
(>50 years), low GCS score (ⱕ12), shock, no HBO treatment, and creatine kinase (>320 U/L). The
COGAS score for CO poisoning was well validated externally with a separate cohort. So far, few studies
have evaluated the risk factors for poor neurocognitive prognosis.7,12,13,20,31 To our knowledge, no
study has created and validated an easily measurable scoring system with clinical and laboratory
parameters for patients with CO poisoning, regardless of severity and HBO therapy.
Previous studies had some limitations. In a multivariable logistic regression analysis conducted
by Weaver et al,12 patients aged 36 years or older treated with HBO had reduced 6-week
neurocognitive sequelae rates.12 Although the study was well conducted, the sample size of 75
patients was small; moreover, the study included patients who received HBO. Furthermore, Weaver
et al12 did not suggest a scoring system for predicting poor prognosis. A study by Pan et al13
investigated 634 patients and found that GCS score, serum blood urea nitrogen level, and intubation
days were associated with poor CO poisoning prognosis. However, Pan et al13 included only patients
who had received HBO treatment, and certain variables, such as the initial HbCO were not evaluated.
The poor outcome was not well defined, as patients either still had sequelae, were bedridden, or had
died, with no scoring model for prediction of poor outcomes. A study by Annane et al31 evaluated
risk factors for outcomes after only mild CO poisoning (those without loss of consciousness) using no
scoring model. Studies by Choi7 and Min20 included relatively small sample sizes and only focused
on estimating delayed neurocognitive sequelae. Moreover, the variables investigated were limited,
with no scoring model. Others who studied the prognosis of CO poisoning mainly analyzed risk factors
for mortality and not neurocognitive prognosis.32-34 Furthermore, Nakajima et al35 assessed risk
factors associated with in-hospital mortality, depressed mental status, and activities of daily living at
discharge.
The 5 variables adopted in COGAS score were age older than 50 years, GCS score of 12 or less,
presence of shock, no HBO treatment, and creatine kinase greater than 320 U/L. Old age is known to
be a risk factor associated with poor prognosis.7,12,20 Decreased mental status has been investigated
as a risk factor in previous studies,13,15,16,18,19,36 as in this study. However, any interval loss of
consciousness was not a risk factor included in this study, which may mean that patients with
prolonged rather than brief loss of consciousness were more likely to have a poor prognosis.
Treatment with HBO was a remarkable factor, consistent with the findings reported in previous
randomized clinical trials.1,6,9,10 In a few studies, serum creatine kinase18,37-40 was found to be a risk
factor, as similarly observed in our study. We compared model 1 with model 2 (ie, with vs without
laboratory values), and the score still had a high explanatory power of 0.812 (model 2), compared
with 0.862 (model 1).

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 8/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Previous studies12,16,17 have shown that longer CO exposure intervals are associated with
increased risk for neurocognitive sequelae. However, we did not use CO exposure time to construct
the main model because, for many patients, CO exposure time was difficult to accurately measure in
clinical settings. Addition of CO exposure time in the scoring model was not suitable, although easier
for the clinicians to use. We compared the diagnostic power of the model including CO exposure time
vs model 1 in the derivation cohort. There was no significant difference. In our analyses, there were
several variables that were not found to be risk factors associated with 1-month cognitive sequelae
after CO exposure, including sex, intentionality, loss of consciousness, seizure, initial HbCO level,
acidosis, lactate level, troponin I level, and history of cardiovascular or psychiatric disease. Early
symptoms could not be investigated in patients who were unconscious. Furthermore, cognitive
sequelae are not necessarily associated with physical expression at the time of poisoning.7,12,20,41-43
Although serum lactate and troponin I levels were reported as risk factors in previous studies,19,44,45
in this study, neither was associated with outcome. A few studies, such as a 2021 study by Sert et al,17
have suggested that acute brain damage detected with brain magnetic resonance imaging is
associated with poor prognosis, but we excluded brain magnetic resonance imaging, as it cannot be
performed early in all hospitals.

Limitations
This study had some limitations. First, bias may have occurred because of excluded patients, despite
the use of objective criteria. However, to our knowledge, this is the first study to create a screening
model with a large sample size (>1000 patients). Second, this study only included patients in the
Republic of Korea; therefore, COGAS score should be validated in different ethnic and regional
groups. Third, charcoal was the most common CO source in this study. However, in the oil and gas
subpopulation, COGAS score had high verification power. Fourth, although a few studies conducted
neurocognitive tests, usually equivalent to CO batteries,6,10 we only evaluated outcomes with GDS
scores. Many neurocognitive function tests may be difficult to perform in patients with sequelae.
GDS score has the advantage of identifying neurocognitive functions, such as memory and
concentration, as well as activities of daily living, and can be applied in all patients. We have
previously reported GDS as a neurocognitive outcome in a study related to CO poisoning,25,46,47 but
it may be insufficient to assess neurological symptoms, such as motor weakness, speech disturbance,
and peripheral neuropathy. However, a study by Choi et al7 reported neurologic impairment,
including hemiplegia, speech disturbance, and peripheral neuropathy in 3 to 20 of 549 patients. We
investigated neurological impairment, and the results did not change the outcome group of patients.
Fifth, a new, potentially permanent mild cognitive decline or impairment may be considered a good
outcome. However, our predictive model showed good performance even with GDS scores divided
into 2.

Conclusions
In this prognostic study a prediction model, dubbed COGAS score, using 5 factors associated with
poor neurocognitive outcome at 1 month among patients with CO poisoning was developed and
externally validated. This scoring system showed excellent capacity to predict poor neurocognitive
outcomes. Future studies are needed to validate COGAS score in various ethnic groups and other
clinical applications.

ARTICLE INFORMATION
Accepted for Publication: March 18, 2022.
Published: May 5, 2022. doi:10.1001/jamanetworkopen.2022.10552

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 9/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2022 Kim SH
et al. JAMA Network Open.
Corresponding Author: Yong Sung Cha, MD, Department of Emergency Medicine, Yonsei University Wonju
College of Medicine, Ilsan-ro 20, Wonju 26426, Republic of Korea (emyscha@yonsei.ac.kr).
Author Affiliations: Department of Biostatistics and Center of Biomedical Data Science, Yonsei University Wonju
College of Medicine, Wonju, Republic of Korea (S. H. Kim); Department of Emergency Medicine, Yonsei University
Wonju College of Medicine, Wonju, Republic of Korea (Lee, H. Kim, Cha); Research Institute of Hyperbaric Medicine
and Science, Yonsei University Wonju College of Medicine, Wonju, Republic of Korea (Lee, H. Kim, Cha);
Department of Emergency Medicine, Inha University College of Medicine, Incheon, Republic of Korea (Kang, Paik).
Author Contributions: Mr S.H. Kim and Dr Cha had full access to all of the data in the study and take responsibility
for the integrity of the data and the accuracy of the data analysis.
Concept and design: S.H. Kim, H. Kim, Cha.
Acquisition, analysis, or interpretation of data: S.H. Kim, Lee, Kang, Paik, Cha.
Drafting of the manuscript: S.H. Kim, Cha.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: S.H. Kim.
Obtained funding: H. Kim, Cha.
Administrative, technical, or material support: Lee, Kang, H. Kim, Cha.
Supervision: Lee, H. Kim, Cha.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by the National Research Foundation of Korea, which is funded by the
Korean government (Ministry of Science and Information and Communications Technology (grant No.
NRF-2019R1A2C208460413 and NRF-2021R1A2C200492211).
Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection,
management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and
decision to submit the manuscript for publication.

REFERENCES
1. Weaver LK. Carbon monoxide poisoning. In: Moon RE, ed. Hyperbaric Oxygen Therapy Indications. 14th ed. Best
Publishing Co; 2019:81-104.
2. Hampson NB. US mortality due to carbon monoxide poisoning, 1999-2014: accidental and intentional deaths.
Ann Am Thorac Soc. 2016;13(10):1768-1774. doi:10.1513/AnnalsATS.201604-318OC
3. Hampson NB, Weaver LK. Carbon monoxide poisoning: a new incidence for an old disease. Undersea Hyperb
Med. 2007;34(3):163-168.
4. Hampson NB, Bodwin D. Toxic CO-ingestions in intentional carbon monoxide poisoning. J Emerg Med. 2013;44
(3):625-630. doi:10.1016/j.jemermed.2012.08.033
5. Mott JA, Wolfe MI, Alverson CJ, et al. National vehicle emissions policies and practices and declining US carbon
monoxide-related mortality. JAMA. 2002;288(8):988-995. doi:10.1001/jama.288.8.988
6. Weaver LK, Hopkins RO, Chan KJ, et al. Hyperbaric oxygen for acute carbon monoxide poisoning. N Engl J Med.
2002;347(14):1057-1067. doi:10.1056/NEJMoa013121
7. Choi IS. Delayed neurologic sequelae in carbon monoxide intoxication. Arch Neurol. 1983;40(7):433-435. doi:
10.1001/archneur.1983.04050070063016
8. Hampson NB. Cost of accidental carbon monoxide poisoning: A preventable expense. Prev Med Rep. 2015;
3:21-24. doi:10.1016/j.pmedr.2015.11.010
9. Ducassé JL, Celsis P, Marc-Vergnes JP. Non-comatose patients with acute carbon monoxide poisoning:
hyperbaric or normobaric oxygenation? Undersea Hyperb Med. 1995;22(1):9-15.
10. Thom SR, Taber RL, Mendiguren II, Clark JM, Hardy KR, Fisher AB. Delayed neuropsychologic sequelae after
carbon monoxide poisoning: prevention by treatment with hyperbaric oxygen. Ann Emerg Med. 1995;25(4):
474-480. doi:10.1016/S0196-0644(95)70261-X
11. Hampson NB, Rudd RA, Hauff NM. Increased long-term mortality among survivors of acute carbon monoxide
poisoning. Crit Care Med. 2009;37(6):1941-1947. doi:10.1097/CCM.0b013e3181a0064f
12. Weaver LK, Valentine KJ, Hopkins RO. Carbon monoxide poisoning: risk factors for cognitive sequelae and the
role of hyperbaric oxygen. Am J Respir Crit Care Med. 2007;176(5):491-497. doi:10.1164/rccm.200701-026OC

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 10/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

13. Pan KT, Shen CH, Lin FG, et al. Prognostic factors of carbon monoxide poisoning in Taiwan: a retrospective
observational study. BMJ Open. 2019;9(11):e031135. doi:10.1136/bmjopen-2019-031135
14. Pepe G, Castelli M, Nazerian P, et al. Delayed neuropsychological sequelae after carbon monoxide poisoning:
predictive risk factors in the Emergency Department: a retrospective study. Scand J Trauma Resusc Emerg Med.
2011;19:16. doi:10.1186/1757-7241-19-16
15. Liao SC, Mao YC, Hung YM, Lee CH, Yang CC. Predictive role of QTc prolongation in carbon monoxide
poisoning-related delayed neuropsychiatric sequelae. Biomed Res Int. 2018;2018:2543018. doi:10.1155/2018/
2543018
16. Zhang Y, Lu Q, Jia J, Xiang D, Xi Y. Multicenter retrospective analysis of the risk factors for delayed neurological
sequelae after acute carbon monoxide poisoning. Am J Emerg Med. 2021;46:165-169. doi:10.1016/j.ajem.2020.
06.090
17. Sert ET, Kokulu K, Mutlu H. Clinical predictors of delayed neurological sequelae in charcoal-burning carbon
monoxide poisoning. Am J Emerg Med. 2021;48:12-17. doi:10.1016/j.ajem.2021.04.001
18. Han S, Choi S, Nah S, et al. Cox regression model of prognostic factors for delayed neuropsychiatric sequelae
in patients with acute carbon monoxide poisoning: a prospective observational study. Neurotoxicology. 2021;
82:63-68. doi:10.1016/j.neuro.2020.11.006
19. Jung JW, Lee JH. Serum lactate as a predictor of neurologic outcome in ED patients with acute carbon
monoxide poisoning. Am J Emerg Med. 2019;37(5):823-827. doi:10.1016/j.ajem.2018.07.046
20. Min SK. A brain syndrome associated with delayed neuropsychiatric sequelae following acute carbon
monoxide intoxication. Acta Psychiatr Scand. 1986;73(1):80-86. doi:10.1111/j.1600-0447.1986.tb02671.x
21. World Medical Association. World Medical Association Declaration of Helsinki: ethical principles for medical
research involving human subjects. JAMA. 2013;310(20):2191-2194. doi:10.1001/jama.2013.281053.
22. Reisberg B, Ferris SH, de Leon MJ, Crook T. The Global Deterioration Scale for assessment of primary
degenerative dementia. Am J Psychiatry. 1982;139(9):1136-1139. doi:10.1176/ajp.139.9.1136
23. Tibbles PM, Edelsberg JS. Hyperbaric-oxygen therapy. N Engl J Med. 1996;334(25):1642-1648. doi:10.1056/
NEJM199606203342506
24. Tibbles PM, Perrotta PL. Treatment of carbon monoxide poisoning: a critical review of human outcome studies
comparing normobaric oxygen with hyperbaric oxygen. Ann Emerg Med. 1994;24(2):269-276. doi:10.1016/
S0196-0644(94)70141-5
25. Kim SJ, Thom SR, Kim H, et al. Effects of adjunctive therapeutic hypothermia combined with hyperbaric
oxygen therapy in acute severe carbon monoxide poisoning. Crit Care Med. 2020;48(8):e706-e714. doi:10.1097/
CCM.0000000000004419
26. Leisman DE, Harhay MO, Lederer DJ, et al. Development and reporting of prediction models: guidance for
authors from editors of respiratory, sleep, and critical care journals. Crit Care Med. 2020;48(5):623-633. doi:10.
1097/CCM.0000000000004246
27. Harrell FE. Describing, resampling, validating, and simplifying the model. In: Harrell FE, ed. Regression
Modeling Strategies: With Applications to Linear Models, Logistic and Ordinal Regression, and Survival Analysis.
Springer International Publishing; 2015:103-126. doi:10.1007/978-3-319-19425-7_5
28. Hamilton-Farrell MR. British Hyperbaric Association carbon monoxide database, 1993-96. J Accid Emerg Med.
1999;16(2):98-103. doi:10.1136/emj.16.2.98
29. Lynch R, Laden G, Grout P. Carbon monoxide poisoning: correlation of neurological findings between accident
and emergency departments and a hyperbaric unit. Emerg Med J. 2001;18(2):95-98. doi:10.1136/emj.18.2.95
30. Weaver LK, Howe S, Hopkins R, Chan KJ. Carboxyhemoglobin half-life in carbon monoxide-poisoned patients
treated with 100% oxygen at atmospheric pressure. Chest. 2000;117(3):801-808. doi:10.1378/chest.117.3.801
31. Annane D, Chevret S, Jars-Guincestre C, et al. Prognostic factors in unintentional mild carbon monoxide
poisoning. Intensive Care Med. 2001;27(11):1776-1781. doi:10.1007/s00134-001-1127-z
32. Hampson NB, Hauff NM. Risk factors for short-term mortality from carbon monoxide poisoning treated with
hyperbaric oxygen. Crit Care Med. 2008;36(9):2523-2527. doi:10.1097/CCM.0b013e31818419d8
33. Huang CC, Ho CH, Chen YC, et al. Hyperbaric oxygen therapy is associated with lower short- and long-term
mortality in patients with carbon monoxide poisoning. Chest. 2017;152(5):943-953. doi:10.1016/j.chest.2017.
03.049
34. Rose JJ, Nouraie M, Gauthier MC, et al. Clinical outcomes and mortality impact of hyperbaric oxygen therapy
in patients with carbon monoxide poisoning. Crit Care Med. 2018;46(7):e649-e655. doi:10.1097/CCM.
0000000000003135

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 11/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022

 JAMA Network Open | Emergency Medicine Score for Predicting Poor Neurocognitive Outcomes in Acute Carbon Monoxide Poisoning

35. Nakajima M, Aso S, Matsui H, Fushimi K, Yasunaga H. Hyperbaric oxygen therapy and mortality from carbon
monoxide poisoning: a nationwide observational study. Am J Emerg Med. 2020;38(2):225-230. doi:10.1016/j.
ajem.2019.02.009
36. Grieb G, Simons D, Schmitz L, Piatkowski A, Grottke O, Pallua N. Glasgow Coma Scale and laboratory markers
are superior to COHb in predicting CO intoxication severity. Burns. 2011;37(4):610-615. doi:10.1016/j.burns.2010.
03.007
37. Kudo K, Otsuka K, Yagi J, et al. Predictors for delayed encephalopathy following acute carbon monoxide
poisoning. BMC Emerg Med. 2014;14:3. doi:10.1186/1471-227X-14-3
38. Lee H, Kang H, Ko BS, Oh J, Lim TH, Cho Y. Initial creatine kinase level as predictor for delayed neuropsychiatric
sequelae associated with acute carbon monoxide poisoning. Am J Emerg Med. 2021;43:195-199. doi:10.1016/j.
ajem.2020.02.054
39. Suzuki Y. Risk factors for delayed encephalopathy following carbon monoxide poisoning: importance of the
period of inability to walk in the acute stage. PLoS One. 2021;16(3):e0249395. doi:10.1371/journal.pone.0249395
40. Kitamoto T, Tsuda M, Kato M, Saito F, Kamijo Y, Kinoshita T. Risk factors for the delayed onset of
neuropsychologic sequelae following carbon monoxide poisoning. Acute Med Surg. 2016;3(4):315-319. doi:10.
1002/ams2.197
41. Winter PM, Miller JN. Carbon monoxide poisoning. JAMA. 1976;236(13):1502-1504. doi:10.1001/jama.1976.
03270140054029
42. Chambers CA, Hopkins RO, Weaver LK, Key C. Cognitive and affective outcomes of more severe compared to
less severe carbon monoxide poisoning. Brain Inj. 2008;22(5):387-395. doi:10.1080/02699050802008075
43. Garland H, Pearce J. Neurological complications of carbon monoxide poisoning. Q J Med. 1967;36(144):
445-455.
44. Moon JM, Shin MH, Chun BJ. The value of initial lactate in patients with carbon monoxide intoxication: in the
emergency department. Hum Exp Toxicol. 2011;30(8):836-843. doi:10.1177/0960327110384527
45. Kao HK, Lien TC, Kou YR, Wang JH. Assessment of myocardial injury in the emergency department
independently predicts the short-term poor outcome in patients with severe carbon monoxide poisoning receiving
mechanical ventilation and hyperbaric oxygen therapy. Pulm Pharmacol Ther. 2009;22(6):473-477. doi:10.1016/j.
pupt.2009.04.006
46. Lee Y, Cha YS, Kim SH, Kim H. Effect of hyperbaric oxygen therapy initiation time in acute carbon monoxide
poisoning. Crit Care Med. 2021;49(10):e910-e919. doi:10.1097/CCM.0000000000005112
47. Cho DH, Ko SM, Son JW, Park EJ, Cha YS. Myocardial injury and fibrosis from acute carbon monoxide
poisoning: a prospective observational study. JACC Cardiovasc Imaging. 2021;14(9):1758-1770. doi:10.1016/j.jcmg.
2021.02.020

SUPPLEMENT.
eAppendix 1. Definition of the Study Variables
eAppendix 2. Global Deterioration Scale Explanation
eTable. Global Deterioration Scale Definitions
eFigure 1. ROC Curve Analysis for Cutoff Value for Converting a Continuous Variable to a Categorical Variable
eFigure 2. Flowcharts of the Derivation and Validation Cohorts
eFigure 3. Histogram of the COGAS Score Model in the Derivation and Validation Cohorts
eFigure 4. Calibration Plots of the COGAS Score for Each Cohort
eFigure 5. Comparison of AUC Between Model 1 and Model 2
eFigure 6. Observed 1-Month Probability of Poor Neurocognitive Outcomes in the Derivation and Validation
Cohorts According to the COGAS Score
eFigure 7. Estimated 1-Month Probability of Poor Neurocognitive Outcomes by Model 2 and ROC Curve With the
AUC of Model 2
eFigure 8. Calibration Plots of Model 2 in the Derivation and Validation Cohorts
eFigure 9. Comparison Between the COGAS Score and the COGAS Score With the Presence of CO Exposure Time
eFigure 10. External Validation of the COGAS Score in the Oil and Gas Combustion-Exposed Subpopulation in the
Derivation and Validation Cohorts
eFigure 11. ROC Curve for the COGAS Scores as a Predictor When Neurocognitive Outcome Was Divided Based on
2 GDS Points
eFigure 12. Estimated 1-Month Probability of Poor Neurocognitive Outcomes and ROC Curve With the AUC When
Neurocognitive Outcome Was Divided Based on 2 GDS Points
eReferences

JAMA Network Open. 2022;5(5):e2210552. doi:10.1001/jamanetworkopen.2022.10552 (Reprinted) May 5, 2022 12/12

Downloaded From: https://jamanetwork.com/ UFBA by Vinicius Silva on 11/01/2022