key: cord-0984558-kk5d8za2
authors: Vizcaychipi, Marcela P.; Shovlin, Claire L.; McCarthy, Alex; Godfrey, Andrew; Patel, Sheena; Shah, Pallav L.; Hayes, Michelle; Keays, Richard T.; Beveridge, Iain
title: Increase in COVID-19 inpatient survival following detection of Thromboembolic and Cytokine storm risk from the point of admission to hospital by a near real time Traffic-light System (TraCe-Tic)
date: 2020-08-18
journal: Braz J Infect Dis
DOI: 10.1016/j.bjid.2020.07.010
sha: 024a21728f372cd4717d25699261c95962b6ede0
doc_id: 984558
cord_uid: kk5d8za2

INTRODUCTION: Our goal was to evaluate if traffic-light driven personalized care for COVID-19 was associated with improved survival in acute hospital settings. METHODS: Discharge outcomes were evaluated before and after prospective implementation of a real-time dashboard with feedback to ward-based clinicians. Thromboembolic categories were “medium-risk” (D-dimer >1,000 ng/mL or CRP >200 mg/L); “high-risk” (D-dimer >3,000 ng/mL or CRP > 250 mg/L) or “suspected” (D-dimer >5,000 ng/mL). Cytokine storm risk was categorized by ferritin. RESULTS: 939/1039 COVID-19 positive patients (median age 69ys, 563/939 (60%) male) completed hospital encounters to death or discharge by 21(st) May 2020. Thromboembolic flag criteria were reached by 568/939 (60.4%), including 238/275 (86.6%) of the patients who died, and 330/664 (49.7%) of the patients who survived to discharge, p < 0.0001. Cytokine storm flag criteria were reached by 212 (22.5%) of admissions, including 80/275 (29.0%) of the patients who died, and 132/664 (19.9%) of the patients who survived, p < 0.0001. The maximum thromboembolic flag discriminated completed encounter mortality (no flag: 37/371 [9.97%] died; medium-risk: 68/239 [28.5%]; high-risk: 105/205 [51.2%]; and suspected thromboembolism: 65/124 [52.4%], p < 0.0001). Flag criteria were reached by 535 consecutive COVID-19 positive patients whose hospital encounter completed before traffic-light introduction: 173/535 (32.3% [95% confidence intervals 28.0, 36.0]) died. For the 200 consecutive admissions after implementation of real-time traffic light flags, 46/200 (23.0% [95% confidence intervals 17.1 - 28.9]) died, p = 0.013. Adjusted for age and sex, the probability of death was 0.33 (95% confidence intervals 0.30 - 0.37) before traffic light implementation, 0.22 (0.17 - 0.27) after implementation, p < 0.001. In subgroup analyses, older patients, males, and patients with hypertension (p ≤0.01), and/or diabetes (p = 0.05) derived the greatest benefit from admission under the traffic light system. CONCLUSION: Personalized early interventions were associated with a 33% reduction in early mortality. We suggest benefit predominantly resulted from early triggers to review/enhance anticoagulation management, without exposing lower-risk patients to potential risks of full anticoagulation therapy.

Human infection due to the novel severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) 1 (commonly referred to as COVID-19 disease) has had unprecedented global impact on society and healthcare provision. In the first seven months of 2020, there have been more than 600,000 fatalities, with three countries (USA, Brazil and UK) each reporting more than 40,000 deaths. 2 Current management protocols have generally emerged from China, where risk factors for in-hospital death included older age, higher Sequential Organ Failure Assessment (SOFA) score, and D-dimer greater than 750 ng/ml on admission. 3 Cytokine patterns [3] [4] [5] were recognized as compatible with a "cytokine storm" of dysregulated inflammation and hyperpyrexia. 6-9 Adult respiratory distress syndrome (ARDS), acute kidney, cardiac and liver failure, and a cytokine storm driving later deteriorations have been foci of research and clinical initiatives, alongside development of vaccines and antiviral therapeutics. 10 Autopsy studies [11] [12] [13] [14] provided a slightly different picture as the risk of fatal thromboembolic disease emerged, in keeping with the initial reports of elevated D-dimers. 3 Thromboembolic disease is now estimated to affect high proportions of patients with COVID-19. [15] [16] [17] [18] COVID-19 displays heterogenous presentation patterns and outcomes. Patient comorbidities including hypertension, diabetes, obesity, and heart disease are associated with higher risk of severe COVID-19 disease. 3, 4, [19] [20] [21] while patients with additional common diseases including asthma and chronic obstructive lung disease (COPD) are advised to take extra precautions in the USA. 22 Reported COVID-19 inpatient mortality rates vary by city/region, 23 between countries, 1, 3, 4, 16, 17, 19, [23] [24] [25] and by disease severity. 3, 4, [23] [24] [25] [26] Differential survival rates can therefore reflect a multitude of confounders from patients, geography, and health care systems, superimposed on pathophysiological variability that is not yet understood.

A near real-time decision tool was introduced early in the pandemic in our institution, as we were concerned about the high published mortality rates. This tool refreshes in the background every 10 min and updates a clinical dashboard accordingly. Generated on the electronic patient record, traffic lights were used to alert clinicians when patients met criteria for increased risk of the three most common causes of death in COVID-19thromboembolism, cytokine storm, and respiratory deterioration/ARDS.

The goal of the current study was to evaluate whether prospective application during April and May 2020 had been associated with any change in patient early mortality across completed hospital episodes.

All patients were consecutive admissions to the emergency department. A COVID-19 battery of tests was developed in electronic medical records and introduced into clinical practice in early March 2020. The COVID pathology order was requested for all patients admitted to the emergency department with suspected and or confirmed COVID-19 infection, and subsequently requested by physicians throughout patients' stay to monitor trends and adjust treatments accordingly. This standardized COVID-19 careset spanned complete blood count (CBC); coagulation including prothrombin time, activated partial thromboplasin time (APTT), fibrinogen, and D-dimer; electrolytes (sodium, potassium); renal function (urea, creatinine); liver function (bilirubin, alkaline phosphatase, albumin, aspartate transaminase, alanine transaminase, gamma-glutamyltransferase (GGT), lactate dehydrogenase (LDH); iron studies (serum iron, transferrin saturation index, ferritin), and additional acute phase response markers including C-reactive protein (CRP).

A near real-time decision support tool was developed and set up on each electronic patient record. 16 This was created for use with COVID-19 patients in March 2020 (Supplementary material 1). In initial stages of the COVID-19 pandemic, a composite of variables which included first vital signs and the COVID-19 battery of test results on admission to the emergency department was used to model the disease. In brief, our clinical dataset was passed through a Neural Network algorithm contained in SQL Server Analysis Services (SSAS) Data Mining Package. The input variables were age; sex; first recorded systolic blood pressure (BP); first recorded diastolic BP; first recorded respiratory rate; first recorded temperature; first recorded oxygen saturation (SpO2); first recorded fractional inspired oxygen concentration (FiO2); first recorded heart rate, and first recorded c-reactive protein (CRP). We rapidly learnt that the CRP showed highly predictive value when patients presented with D-dimer <3000 ng/mL. In view of the lack of J o u r n a l P r e -p r o o f randomized controlled trial data, and post-mortem communications predating references, 11-14 the authors felt it was prudent to include the minimal D-dimer cut off of 750 ng/mL published by Chinese colleagues at the end of January 2020. 3 For simplicity the value was rounded to 1,000 ng/mL. Between 20 th March 2020 and 12 th April 2020, a daily report was issued on all COVID-19 positive patients in the institution, and sent to an experienced critical care clinician for review. Results were interpreted and fed back to the clinical teams to review, and modify pathways as appropriate, with data from the first phase defining patterns of disease and biomarkers. There were recommended anticoagulation protocols and guidelines for management of cytokine storm (Table 1 ) which the treating physician adopted and tailored to patient's response to treatment when indicated. The traffic light system went live across the full institution on 12 th April 2020, updated every 10 minutes, for implementation into prospective pathways by all clinicians. Clinicians were also updated by whatsApps, emails, and phone calls, following electronic medical record monitoring of prescription compliance on a regular basis. The most effective way of communication was by utilization of whatsApps groups with updates of the institutional hematology guidelines. The date of introduction (12 th April 2020) was used to cohort patients in the study, as described further below.

Data processing was authorized by the Clinical Information Governance committees. Access to the data and authorization for the present study was jointly granted by both the Institution's Data Protection Office and by the Institution's Analytical Unit, under the General Data Processing Regulations (processing under public authority for purposes in the area of public health). To ensure compliance with General Data Protection Regulations, data were extracted from pseudonymized datasets into aggregate reports only for the outcomes of interest. The corresponding author had full access to all the data in the study and the final responsibility to submit for publication.

The study cohort represented all COVID-19 positive patients who had completed their hospital encounter by 21 st May 2020. Patients were either discharged home alive (including to temporary homes and/or residential care homes), or had died. At the date of reporting, patients who remained in the admitting hospital, or had been transferred to other hospitals, were considered to have an "incomplete" encounter and were not included in the analysis.

Patients with completed encounters were categorized into three cohorts, according to timing in relation to 12 th April 2020 when the traffic light system went live across the full institution. The "Pre" cohort completed their hospital encounter (to home discharge or death) before 12 th April 2020. The "Post" cohort were admitted on or after 12 th April 2020, so their admission was fully managed by the real-time system. A third group was admitted before, but discharged after 12 th April 2020 (the "Bridge" cohort).

The comorbidities of asthma, COPD, diabetes, and hypertension were defined where listed as a Diagnosis or Problem on the electronic patient record or historic coding. A prior pharmacy record of salbutamol prescription also placed patients in the asthma category, and a requested glycated hemoglobin (HbA1c) level placed patients in the diabetic category. Patients could have one, several or none of the comorbidities.

Anonymized data were analyzed using Stata IC version 15 (Statacorp, Texas) across the full cohort and across all time periods. Comparisons were performed using the non-parametric Kruskal-Wallis equality-ofpopulations rank test, with Dunn's multiple comparison test used to derive the multiple comparisonadjusted, two-tailed p-values. To adjust for differences in age, sex and other variables between the time periods, multivariate logistic regression was performed for the binary outcome of death ('1') versus home discharge ('0'). The probabilities of death in the "Pre" and "Post" cohorts were calculated by post-test marginal comparisons after logistic regression across the full 939 cases using model covariates of age, sex and a 3-cohort variable distinguishing the "Pre", "Bridge" and "Post" cohorts. For illustration of summary statistics and heat maps, data were exported to Graph Pad Prism 8. infected patients), or whether to wait for patients to develop end-stage disease.

We hypothesized that:

(1) Blood test results in dynamic reassessments would discriminate patients at or becoming at higher risk of complications;

(2) If rapidly communicated to clinicians, higher risk patients could be targeted for early review and implementation of accelerated enhanced treatments;

(3) Early implementation of accelerated enhanced treatments, targeted at individual level at the point of entry, would translate to improved outcome demonstrated by more patients surviving to be discharged home alive.

A total of 1039 COVID-positive patients had been admitted by 21 st May 2020. Of those, 889 (85.6%) were managed on general wards, and 150/1039 (14.4%) were admitted to adult intensive care for part of their inpatient stay. Of the 1039 patients, 664 had been discharged home, either to their usual residence (n=614), a temporary home (n=16), or a residential care home (n=34), and 275 had died. A further 100 patients were still in the hospital, or of unknown outcome following transfer to a different hospital.

The 939/1039 who had completed their hospital encounter by 21 st May 2020 constituted the current study cohort. Patients' median age was 67 years (interquartile range, IQR 54-81years) i.e. more than 25% of admitted patients were at least 81 years old, and 563/939 (60%) were males. The population comprised 62% white (370/597 recorded), and 38% black and minority ethnic (227/597 recorded). Comorbidities were common: 500/939 (53%) patients were hypertensive; 354/939 (38%) were in the diabetic category, 226/939 (24%) were defined as having asthma, and 95/939 (10%) with COPD ( Table 2) . (15.2%) the highest cytokine storm category reached was a "medium-risk" flag, 34/939 (3.6%) received a "high-risk" flag, and 35 (3.7%) were flagged with "suspected" cytokine storm. Mortality differed overall between the four categories (Kruskal Wallis p <0.0001), predominantly attributable to higher mortality in the 69 patients who were flagged in the two higher risk categories (Figure 1 ).

Of the 939 patients with a completed encounter, 535 (57%) were in the "Pre" cohort admitted and discharged or died before/on 12 th April 2020. 204 (22%) patients were in the "Bridge" cohort, and 200 (21%) were in the "Post" cohort admitted and discharged or died after 12 th April 2020 (Supplementary The median age of the "Pre" cohort was 67 years (IQR 52-80). The median age of the "Post" cohort was 72 years (IQR 56, 82.5), p=0.037 (Figure 2A) . In crude analyses, 173/535 patients in the "Pre" cohort died (32.3% [95% confidence interval 28.0 to 36.0]), compared to 46/200 patients in the "Post" cohort (23.0%

[95% confidence interval 17.1 to 28.9]), Dunn's test p=0.013 ( Figure 2B ). An intermediate mortality was observed in the 204 patients whose admission spanned real-time traffic light introduction on 12 th April 2020 of these, 56/204 died ( Figure 2B ). This was not significantly different to that of the "Pre" or "Post" cohorts (Wald test p-values 0.39 and 0.65, respectively).

Full comparisons of the cohorts are provided in Supplementary Table A.2. Trends for the "Post" cohort to have higher proportions flagged at moderate-risk (p=0.27) or high-risk (p=0.23) of thromboembolism did not reach statistical significance, though a higher proportion of the "Post" cohort were flagged with "suspected" thromboembolism compared to the "Pre" cohort (p=0.019, Figure 3A -C). There was no difference between the "Pre" and "Post" subgroups in the smaller proportions with moderate-risk (p=0.67), high-risk (p=0.96) or suspected (p=0.86) cytokine storm flags ( Figure 3D -F).

In logistic regression analyses, adjusted odds ratios for mortality were smaller (i.e. more favorable) than the crude odds ratio for the "Post" cohort following adjustment for age, sex, and all combinations of variables in Table 1 (data not shown).

Across all 939 patients, and adjusted for age and sex, the probability of death was 0.33 (95% confidence intervals 0.30 to 0.37) in the "Pre" cohort, and 0.22 (95% confidence intervals 0.17 to 0.27) in the "Post" cohort, p<0.001 (Figure 4 ).

Death-rates across the three time periods were examined for key patient subgroups. Compared to the "Pre" cohort, death rates were lower in the "Post" cohort for males (p=0.0014), for patients in the upper three age quartiles and above (≥54 years, p=0.0014), particularly the third age quartile (68-79 years, p=0.0014); for patients with hypertension (p=0.0072), and for patients with diabetes (p=0.05) ( Figure 5 ).

We have shown that early introduction of near real time traffic light-driven, personalized care was associated with an overall reduction in crude mortality from 32.3% (95% confidence interval 28.0, 36.0), to 23.0% (95% confidence interval 17.1, 28.9). The overall age and sex-adjusted probability of mortality There were several study limitations with many shared by other studies in emergency-based assessments of this new human disease with very high mortality. Interventions prioritized clinical care delivery, and the main purpose of this manuscript was to stratify the disease and introduce early medical management, rather than detailed expositions of population characteristics and symptomatology. Patients with very prolonged hospitalization did not have completed encounters and were not included: responses in this important subgroup will need evaluation in future prospective randomised controlled trials. Additionally, capturing mortality to end-of-hospital admission may have missed subsequent deaths in the community, and these may have differed in proportions over the study period. However, for crude mortality in the "Post" cohort to reach that of the "Pre" cohort, an 86% mortality rate would be required in the inpatients admitted on/after 12 th April 2020, when their current length of stay IQR already exceeds the IQR for length of stay before J o u r n a l P r e -p r o o f death (Supplementary Table A.1) . Therefore, this reduction in early mortality is considered to be robust, and important to communicate.

The study was conducted at a two-site institution, therefore it is difficult to compare baseline mortalities with other institutions. 28 With the very high worldwide mortality 2 and strong steers towards risk factors from the earliest reports from China, 1,3 it was not considered appropriate to randomize patients. Where lethality is high, as in COVID-19, the paucity of sequential analysis techniques has been noted with concern. 28 The current study employed a sequential analysis technique, within a dual centre, single institution, and demonstrated that the introduction of full traffic light-driven care on 12 th April 2020 was associated with a 33% reduction in age and sex-adjusted inpatient mortality. We recognize that a limitation of a sequential analysis is that it is not possible to fully standardize variables, and acknowledge that other factors could have contributed. It is feasible that in the current study, the baseline "Pre" group were already benefiting from the first phase of traffic light implementation, 16 and different approaches to care. 29 The survival signal was robust to age and sex adjustment; adjustment for patient severity indices would have made the association stronger, and other major treatment protocols did not change in the study period. Clinical expertise and pathways did have extra-time to embed, though we note unchanged mortality figures in major country-wide audits over the same period. 30 It is therefore difficult to conclude that any other factor could have been responsible for generating a completed encounter inpatient mortality rate that at 23.0% [17.1 to 28.9] is now lower than in New York, 24 and lower than the rates reported during the Middle East Respiratory Syndrome (MERS) epidemic. 31 Further, this completed encounter mortality rate was achieved with more than 25% of the patient population aged 81 years or over at the time of admission, and with advanced disease at presentation as Despite more severe disease in the "Bridge" cohort, survival rate was no worse than for the "Pre" cohort.

This suggests that the "Bridge" cohort may have also benefitted in the first phase of traffic-light implementation.

Numerically, the key differences compared to our initial practice were earlier, enhanced anticoagulation for patients identified as at higher risk of thromboembolic disease. A smaller group of patients flagged at risk of cytokine storm based on serum ferritin also had enhanced attention to fluid balance and intravenous fluid delivery. For thrombosis, the flags helped to identify the stratified risk of thromboembolic event, and the intervention was based on the risk regardless of the radiological findings. In COVID-19, thrombosis is commonly microvascular and difficult to detect using current imaging, with reports tending to focus on the presence or absence of major thrombus. The observed benefit from enhanced anticoagulation is in keeping with the limited published evidence in COVID-19. 26, 33 The prothrombotic COVID-19 state is currently considered less to reflect disseminated intravascular coagulation (DIC) caused by the systemic generation of thrombin, than hypercoagulability in the setting of a severe inflammatory state, 34 and particularly in situ thrombus formation. 18, 35, 36 The findings would also be supported by previous evidence on multi-organ failure in sepsis where inhibiting thrombin generation by anticoagulation may have benefit in reducing mortality. 37 There was early evidence that the presence of an elevated D-dimer was associated with a poor prognosis in 38 It has been suggested that high D-dimer values may not reliably predict the presence of a thrombus that has been degraded by fibrinolysis, but instead represent a marker of poor overall outcome in However, data from the current study targeting D-dimer-identified subgroups, and from two other groups, 26, 33 indicate that enhanced anticoagulation can improve survival, in advance of full mechanistic understanding. 40 As proposed in 40 , we also examined fibrinogen initially, and unexpectedly found that CRP (which was not part of the Chinese investigative sets), 3,4 was more discriminatory. CRP and D-dimer not only identified higher risk categories for enhanced treatments, but also a low risk subgroup (all D-dimer ≤1000 ng/mL, and all CRP ≤200 mg/L). The ability to identify subgroups with better prognosis is generally helpful in establishing where risks of intensive therapies may be more difficult to justify. Since therapeutic anticoagulation carries hemorrhagic concerns, particularly for diseases with abnormal vasculature as in COVID-19, [11] [12] [13] [14] 18 we speculate that restricting enhanced thromboembolic prophylaxis to patients in subgroups with higher thrombotic risks may have contributed to the overall favorable survival figures in the traffic lights-managed cohort.

In conclusion, early recognition of thromboembolic risk based on CRP and or D-dimer, and impending cytokine storm predominantly based on serum ferritin, enabled hospital survival for completed encounters of 77.0% (95% confidence intervals, 71.1 to 82.8%) when more than 25% of admitted patients were at least 81 years old. Many COVID-19 at-risk groups benefitted including older patients, males, and patients with concurrent hypertension and diabetes. The blood measurements are already part of most hospitals' COVID-19 datasets, and interventions for thromboembolic disease, and fluid management are widely available and inexpensive. Thus broad application seems feasible, and wider implementation is encouraged. 

The code and guidance is freely available on request from …[withheld identifying] and is to be published to …[withheld identifying] in due course. Fully anonymised data will be available post peer review publication on reasonable request, in accordance with institutional protocols above. A data dictionary defining each field in the data set will also be made available to others. Legend: * Therapeutic a/c, therapeutic anticoagulation unless contraindications; Vit C, ascorbic acid 1g bd; iv fluids, fluid management adjusted to replace insensible losses (1ml.kg -1 .h -1 for every degree above 38° C). If there was no response to initial fluid management and diastolic blood pressure remained <35mmHg, or oxygen requirement increased, then the patients were referred to the AICU for further management. 

A Novel Coronavirus from Patients with Pneumonia in China

COVID-19 Dashboard by the Center for Systems Science and Engineering

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China

Pathological findings of COVID-19 associated with acute respiratory distress syndrome

TH17 responses in cytokine storm of COVID-19: An emerging target of JAK2

Epub ahead of print

Engineering (Beijing)

HLH Across Speciality Collaboration, UK. COVID-19: consider cytokine storm syndromes and immunosuppression

Advances in the research of cytokine storm mechanism induced by Corona Virus Disease

Epub ahead of print

Vander Heide RS. Pulmonary and Cardiac Pathology in Covid-19: The First Autopsy Series from New Orleans

Complement associated microvascular injury and thrombosis in the pathogenesis of severe COVID-19 infection: A report of five cases

Autopsy Findings and Venous

Thromboembolism in Patients With COVID-19: A Prospective Cohort Study

Pulmonary Vascular Endothelialitis, Thrombosis, and Angiogenesis in Covid-19

Incidence of thrombotic complications in critically ill ICU patients with COVID-19

Epub ahead of print

Factors associated with hospital admission and critical illness among 5279 people with coronavirus disease

Comorbidity and its impact on 1590 patients with COVID-19

China: a nationwide analysis

Obesity as a risk factor for COVID-19: an overview

Metabolic syndrome and COVID-19: An update on the associated comorbidities and proposed therapies

Clinical characteristics and outcomes of hospitalised patients with COVID-19 treated in Hubei (epicenter) and outside Hubei (non-epicenter): A Nationwide Analysis of China

Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized With COVID-19 in the

Features of 20 133 UK patients in hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: prospective observational cohort study

Anticoagulant treatment is associated with decreased mortality in severe coronavirus disease 2019 patients with coagulopathy

The world is waiting, use sequential analysis and get us the evidence-based treatment we need for COVID-19

among hospital patients with Middle East respiratory syndrome coronavirus (MERS-CoV) infection

Association of Treatment Dose Anticoagulation with In-Hospital Survival Among Hospitalized Patients with COVID-19

Hypercoagulability of COVID-19 patients in Intensive Care Unit

Thromboelastography Findings and other Parameters of Hemostasis

COVID-19 and haemostasis: a position paper from Italian Society on Thrombosis and Haemostasis (SISET)

Is thromboprophylaxis with high-dose enoparin really necessary for COVID-19 patients? A new "prudent" randomized clinical trial

Diagnosis and management of sepsis-induced coagulopathy and disseminated intravascular coagulation

Abnormal coagulation parameters are associated with poor prognosis in patients with novel coronavirus pneumonia

Disease: Implications for Prevention, Antithrombotic Therapy, and Follow-up

ISTH interim guidance on recognition and management of coagulopathy in COVID-19

factors: Total datasets Median (IQR)/ N-% 1%-99%

Legend: *The most common single symptoms were dyspnea (77.4%), chest pain (19.9%) and collapse (6.1%). ULS: upper limit of scale.