key: cord-0252988-hi2r95o5
authors: Raschke, R. A.; Rangan, P.; Agarwal, S.; Uppalapu, S.; Sher, N.; Curry, S. c.; Hesie, C. W.
title: COVID-19 Time of Intubation Mortality Evaluation (C-TIME): A System for Predicting Mortality of Patients with COVID-19 Pneumonia at the Time They Require Mechanical Ventilation.
date: 2022-01-11
journal: nan
DOI: 10.1101/2022.01.09.22268977
sha: e70c5c193691ce3b8f5528262b56578d243f1dff
doc_id: 252988
cord_uid: hi2r95o5

Background: An accurate system to predict mortality in patients requiring intubation for COVID-19 could help to inform consent, frame family expectations and assist end-of-life decisions. Research objective: To develop and validate a mortality prediction system called C-TIME (COVID-19 Time of Intubation Mortality Evaluation) using variables available before intubation, determine its discriminant accuracy, and compare it to APACHE IVa and SOFA. Methods: A retrospective cohort was set in 18 medical-surgical ICUs, enrolling consecutive adults, positive by SARS-CoV 2 RNA by reverse transcriptase polymerase chain reaction or positive rapid antigen test, and undergoing endotracheal intubation. All were followed until hospital discharge or death. The combined outcome was hospital mortality or terminal extubation with hospice discharge. Twenty-five clinical and laboratory variables available 48 hours prior to intubation were entered into multiple logistic regression (MLR) and the resulting model was used to predict mortality of validation cohort patients. AUROC was calculated for C-TIME, APACHE IVa and SOFA. Results: The median age of the 2,440 study patients was 66 years; 61.6 percent were men, and 50.5 percent were Hispanic, Native American or African American. Age, gender, COPD, minimum mean arterial pressure, Glasgow Coma scale score, and PaO2/FiO2 ratio, maximum creatinine and bilirubin, receiving factor Xa inhibitors, days receiving non-invasive respiratory support and days receiving corticosteroids prior to intubation were significantly associated with the outcome variable. The validation cohort comprised 1,179 patients. C-TIME had the highest AUROC of 0.75 (95%CI 0.72-0.79), vs 0.67 (0.64-0.71) and 0.59 (0.55-0.62) for APACHE and SOFA, respectively (Chi2 P<0.0001). Conclusions: C-TIME is the only mortality prediction score specifically developed and validated for COVID-19 patients who require mechanical ventilation. It has acceptable discriminant accuracy and goodness-of-fit to assist decision-making just prior to intubation. The C-TIME mortality prediction calculator can be freely accessed on-line at https://phoenixmed.arizona.edu/ctime.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint 5 Introduction:

The Coronavirus disease 2019 (COVID- 19) pandemic raised concern that an overwhelming surge of critically-ill patients might require exclusion of patients with high predicted mortality from receiving mechanical ventilation (1). The majority of COVID-19 ventilator triage policies surveyed in 2020 incorporated the Sequential Organ Failure Assessment (SOFA) score to predict mortality (2). However, a subsequent study that collected SOFA score data 48 hours prior to intubation yielded a discriminant accuracy for mortality prediction of only 0.59 (95%CI: 0.55-0.63) (3). Although many other scoring systems have been developed to predict mortality in patients with , none focused on assessing the patient at the time of intubation, when patients, families and providers are forced to make critical decisions regarding life support. Although the capacity to provide mechanical ventilation to COVID-19 patients is improved, the need for ventilator triage is still possible in regional hotspots, and informed consent for endotracheal intubation should include discussion of prognosis. Our aim was to develop a mortality prediction system we called C-TIME (COVID-19 Time of Intubation Mortality Evaluation) using variables typically available in the 48 hours before intubation, in order to inform consent, frame family expectations and assist end-of-life planning. Our secondary aims were to validate C-TIME, determine its discriminant accuracy and compare it to SOFA and APACHE IVa mortality prediction models.

Study design. A retrospective cohort study, approved by our IRB, was set in 18 medical surgical ICUs in the Southwest United States between 6/1/2020 and 3/23/2021. June was chosen for cohort inception when preliminary results of the RECOVERY trial (28) were released, and administration of dexamethasone rapidly adopted in our study ICUs. We randomly split our cohort in half to compose model-development and validation cohorts.

Participants. Consecutive ICU patients were included based on the following eligibility criteria: >18 years of age; positive SARS-CoV 2 RNA by reverse transcriptase polymerase chain reaction or positive rapid antigen test; and undergoing endotracheal . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint 6 intubation >4 hours after admission. All patients were followed until hospital discharge or death.

The main outcome variable was hospital mortality or discharge to hospice after terminal extubation -henceforth this combined outcome is referred to as "mortality". We chose candidate predictor variables to use in model development based on previous literature (4-27) and hypotheses generated by our clinical research team.

We examined our clinical dataset and only selected candidate predictor variables missing in less than 10% of study patients. We made an exception for the partial pressure of arterial oxygen/fraction of inspired oxygen (PaO 2 /FiO 2 ) ratio, which we hypothesized would be a particularly important predictor (29) ; therefore we planned to impute missing PaO 2 /FiO 2 data (see statistics section below).

The following 25 candidate predictor variables, collected in the time period before intubation, were chosen to include in model development. Patient characteristics included: age, gender, body mass index (BMI), prior history of diabetes mellitus, hypertension, COPD, coronary artery disease, cancer or solid organ transplant.

Physical examination findings included maximum temperature, lowest mean arterial pressure and lowest Glasgow Coma scale in the 48 hours prior to intubation. Laboratory variables included the highest concentration of creatinine and bilirubin, and the lowest platelet count and PaO 2 /FiO 2 ratio in the 48 hours prior to intubation. Management variables comprised hospital days prior to intubation; hospital days receiving noninvasive respiratory support (high-flow nasal canula oxygen; continuous positive airway pressure or bilevel positive airway pressure) prior to intubation; hospital days receiving corticosteroids (dexamethasone, methylprednisolone or prednisone) prior to intubation; and administration of any of the following drugs: corticosteroids, therapeutic dose heparin/enoxaparin, oral Xa inhibitors, subcutaneous or intravenous insulin, or norepinephrine infusion. We also included intubation during surge conditions, defined as the time period(s) during which > 400 ventilators (>5.5 ventilators per 100,000 population) were in use by COVID-19 patients in the state of Arizona where most of our study hospitals were located. By this criteria, surge conditions occurred in our ICUs in . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint 7 the summer (6/23/2020 -8/7/2020) and winter (12/3/2020 -2/14/2021) (30) . Variables needed to calculate the SOFA score were also extracted in the 48 hours prior to intubation and used to derive the associated predicted mortality for each patient (31, 32) .

Variables used to calculate APACHE IVa predicted mortality were obtained in the first 24 hours after ICU admission, as the APACHE system is designed to operate. candidate predictor variables were entered into backwards, step-wise, multiple logistic regression (MLR) using the model-development cohort, with mortality as the dependent outcome variable. We retained all variables that remained in the model at P<0.05.

The MLR logistic equation from the model development cohort was then applied to calculate predicted mortality for each patient in the validation cohort and to calculate Nagelkerke's pseudo R 2 and Hosmer-Lemeshow goodness-of-fit tests. We compared goodness-of-fit and sensitivity for C-TIME versus SOFA and APACHE IVa in patients with very high predicted mortality, for whom prognostic information is most likely to affect end-of-life decisions. The three models were each used to identify clinically important patient subgroups with >75%, >90% and >95% predicted mortality, and the observed mortality proportion in each of the resulting subgroups was enumerated. The . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint 8 number of observed deaths in each subgroup over the total number of observed deaths in the validation cohort equaled the sensitivity of each model at each of the three predicted mortality cutoffs. We also looked at observed mortality proportion for patients C-TIME predicted to have <50% mortality. The area under the receiver operator characteristic curve (AUROC), a measurement of discriminant accuracy, for C-TIME, SOFA and APACHE IVa were calculated and compared using the Chi-squared statistic.

The Wilson method was used to calculate 95% confidence intervals for single proportions. We used STATA ® Version 17 (Statacorp, College Station, TX) for all statistical analyses.

We also calculated AUROCs for C-TIME and SOFA using only validation cohort patients for whom FiO 2 and PaO 2 were known (i.e. excluding patients with imputed values). These were compared to the AUROCs of our primary analysis by using z-tests on equality of proportions to test whether data imputation affected AUROC.

Results. Between 6/1/2020 and 3/23/2021, 18,431 patients with COVID-19 were admitted to study hospitals. Of these, 4,695 were admitted to the ICU and 2,440 were intubated >4 hours after admission. Characteristics of these 2,440 study patients are presented in the table 1. The median age was 66 years, 61.6 percent were men, and 50.5 percent were Hispanic, Native American or African American. Eighty-six percent of patients received corticosteroids. Eleven variables were significant in the final MLR model (see table 2 ). The validation cohort comprised 1,219 patients of whom 1,179 had complete data for analysis by MLR. Observed mortality in the validation cohort was 65.1%.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint Discussion. The C-TIME mortality prediction model, based on eleven easily obtained clinical and laboratory variables, has better discriminant accuracy than APACHE IVa with 145 variables (33) . Furthermore, the C-TIME model has acceptable "goodness of fit" and sensitivity in patients with high predicted mortality, in whom C-TIME may be helpful in making end-of-life decisions. Our study hospitals range from tertiary academic centers to community and critical access facilities serving a variety of persons from urban and rural communities with a wide diversity of racial/ethnic backgrounds and socioeconomic status, enhancing the external generalizability of our findings.

Well over one hundred prognostic systems, including general ICU systems (such as SOFA and APACHE), and novel systems specifically developed for COVID-19 patients have already been published to predict clinical outcomes in patients with COVID-19 (4).

These vary by target patient population, predictor variables and outcomes of interest. To provide context for C-TIME, we reviewed comparable scoring systems that were developed and validated specifically for hospitalized COVID-19 patients and which incorporated commonly available clinical and laboratory predictor variables, and which reported AUROCs for in-hospital mortality (5-27).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint Several features distinguish C-TIME from other validated COVID-19 mortality prediction systems we reviewed. 1) C-TIME is the only system that specifically evaluates patients with COVID-19 pneumonia just before they require mechanical ventilation. The discriminant accuracy of other prognostic models at this point in a patient's clinical course are unknown, due to spectrum effect: "When designing a predictive scoring system, the study cohort should be as similar as possible to the population in which the test is intended to be used" (34) .

2) The C-TIME study cohort had by far the highest reported mortality (65%) of any of the previous studies, as would be expected for intubated COVID-19 patients (35) . The mortality of the study cohort has a strong influence on the operating characteristics of associated mortality prediction systems (34) -another reason why previously reported mortality prediction scores are likely not reliable if used at the time of intubation. 3) Other mortality prediction systems utilized study cohorts that included patients admitted prior to 6/2020, when preliminary results of RECOVERY were released. The inclusion of significant numbers of patients who did not receive corticosteroids could limit their generalizability in relationship to current practice patterns. Eighty-six percent of our study patients received corticosteroids before intubation. 4) C-TIME is the only model that incorporates treatment variables. Days receiving corticosteroids and days receiving non-invasive respiratory support prior to intubation were associated with mortality in our model-development and validation cohorts, and were also significantly associated with surge conditions (p=0.0003 and 0.005, respectively). Surviving patients received a median of two days steroids and two days non-invasive respiratory support; non-survivors received a median of nine days steroids and eight days non-invasive respiratory support. A recent study showed that mortality increased significantly during the winter COVID-19 surge (36) however a metaanalysis concluded that delaying intubation does not influence mortality (37) . It is possible that the associations observed in our study might be due to prolonged efforts at non-invasive respiratory support and corticosteroid treatment of patients during surge conditions, selecting treatment non-responders for intubation. Inclusion of these treatment variables associated with surge conditions suggests that C-TIME might retain generalizability regardless of whether or not surge conditions prevail in the clinical setting in which it is used.

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint One particular C-TIME variable deserves brief comment. We tested pre-intubation antithrombotic therapy with heparin, enoxaparin and factor Xa inhibitors in model development based on the hypothesis that these drugs might prevent fatal thromboembolic complications related to COVID-19. However, therapeutic heparin/enoxaparin fell out of the model, and receiving factor Xa inhibitors was surprisingly associated with increased mortality. Analysis of a sample of these patients showed that pre-existing atrial fibrillation was the indication for factor Xa inhibitors in 80% of patients. It is possible that receiving a factor Xa inhibitor was a confounding variable representing pre-existing atrial fibrillation in our model. Several other mortality prediction systems with acceptable operating characteristics are available for prognostication at the time of admission to the hospital, rather than at the time of intubation. The 4C score is supported by the largest study cohort and has an AUROC (0.77) similar to C-TIME (5). We noted the highest AUROCs were reported for systems that prognosticated using variables from the time of admission in Hubei province early in the pandemic (8, 14, 18, 23, 25) . We feel these results are likely irreproducible outside the special circumstances under which they were reported. This contention is supported by a study using data from the Veterans Affairs Data Warehouse (11) that externally-validated several of these prediction scores (23, 25) Table 3 reveals that C-TIME has good fit when it predicts very high mortality, and that it is more sensitive than APACHE IVa or SOFA at identifying nonsurvivors. There were 399 fatal outcomes among patients predicted by C-TIME to have >75% predicted mortality, yielding a sensitivity of 399/794 (50.3%). The comparative sensitivity for APACHE IVa was 41/794 (5.2%) and for SOFA was 92/794 (11.6%).

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint Sensitivity quantifies the potential clinical impact that C-TIME could have should resuscitative measures be limited due to high predicted mortality.

Limitations of the study: Missing data was a major complication of our retrospective cohort design that limited us from including less-frequently-ordered predictor variables such as C-reactive protein, and led us to impute missing PaO 2 and FiO 2 data. Our sensitivity analysis showed that the later did not affect our AUROC estimates. Our EMR data source limited our ability to include variables not recorded as discrete data, such as COVID-19 vaccination status and pre-existing atrial fibrillation.

The discriminant accuracy achieved by C-TIME was modest, although similar to several other COVID-19 mortality prediction systems with AUROCs ranging 0.72-0.79 (5, 10, 17, 19, 21, 38) . We believe that it is inherently difficult to predict COVID-19 mortality 

C-TIME is the only currently available mortality predictive score specifically developed and validated for COVID-19 patients who require intubation. It has acceptable discriminant accuracy and goodness-of-fit to assist informed consent for intubation and . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint

ANDC: an early warning score to predict mortality risk for patients with Coronavirus Disease

The appropriateness of invasive ventilation in COVID-19 positive cancer patients: Proposal of a new prognostic score

A machine learning approach for mortality prediction in COVID-19 pneumonia: Development and evaluation of the Piacenza Score

Comparison of inhospital mortality risk prediction models from COVID-19

COVID-19@Spain and COVID@HULP Study Groups. Development and validation of a prediction model for 30-day mortality in hospitalised patients with COVID-19: the COVID-19 SEIMC score

Application of validated severity scores for pneumonia caused by SARS-CoV-2

A model to predict the risk of mortality in severely ill COVID-19 patients

An early warning tool for predicting mortality risk of COVID-19 patients using machine learning

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

An interpretable mortality prediction model for COVID-19 patients

A novel severity score to predict inpatient mortality in COVID-19 patients

Development and Validation of a

Prognostic Risk Score System for COVID-19 Inpatients: A Multi-Center Retrospective Study in China

Validation of pneumonia prognostic scores in a statewide cohort of hospitalised patients with COVID-19

Prognostic Modeling of COVID-19 Using Artificial Intelligence in the United Kingdom: Model Development and Validation

Troponin and other biomarker levels and outcomes among patients hospitalized with COVID-19: Derivation and validation of the HA2T2 COVID-19 Mortality Risk Score

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

Comparison of inhospital mortality risk prediction models from COVID-19

Medical Treatment Expert Group for COVID-19. Risk factors of fatal outcome in hospitalized subjects with Coronavirus Disease 2019 from a nationwide analysis in China

Real-Time Electronic Health Record Mortality Prediction During the COVID-19 Pandemic: A Prospective Cohort Study. MedRxiv

Scoring systems for predicting mortality for severe patients with COVID-19

Clinical characteristics, associated factors, and predicting COVID-19 mortality risk: A retrospective study in Wuhan, China

Clinical and Laboratory Predictors of In-hospital Mortality in Patients With Coronavirus Disease-2019: A Cohort Study in Wuhan, China

et al for the RECOVERY Collaborative Group. Effect of dexamethasone in hospitalized patients with COVID-19 -Preliminary report

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

The ARDS Definition Task Force*. Acute Respiratory Distress Syndrome: The Berlin Definition

Arizona Department of Health Services. COVID-19 Data Dashboard. Updated daily

On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure

Serial Evaluation of the SOFA Score to Predict Outcome in Critically Ill Patients

Acute Physiology and Chronic Health Evaluation (APACHE) IV: hospital mortality assessment for today's critically ill patients

The spectrum effect in tests for risk prediction, screening, and diagnosis

Case fatality rates for patients with COVID-19 requiring invasive mechanical ventilation. A Meta-analysis

Association Between Caseload Surge and COVID-19 Survival in 558

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

Effect of timing of intubation on clinical outcomes of critically ill patients with COVID-19: a systematic review and metaanalysis of non-randomized cohort studies

The COVID-GRAM Tool for Patients Hospitalized With COVID-19 in Europe

Predictive performance of SOFA and qSOFA for inhospital mortality in severe novel coronavirus disease

The prognostic value of the SOFA score in patients with COVID-19: A retrospective, observational study

International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity

Obtained funding: Curry.Supervision: Curry.

This research was performed with partial support from a Flinn Foundation grant, but the Flinn Foundation had no influence over the conduct of the study, interpretation of the data, preparation of the manuscript or the decision to submit for publication.

The copyright holder for this preprint this version posted January 11, 2022. ; https://doi.org/10.1101/2022.01.09.22268977 doi: medRxiv preprint