key: cord-1004898-w0br1t79 authors: Alhamar, Ghadeer; Maddaloni, Ernesto; Al Shukry, Abdullah; Al‐Sabah, Salman; Al‐Haddad, Mohannad; Al‐Youha, Sarah; Jamal, Mohammed; Almazeedi, Sulaiman; Al‐Shammari, Abdullah A.; Abu‐Farha, Mohamed; Abubaker, Jehad; Alattar, Abdulnabi T.; AlOzairi, Ebaa; Alessandri, Francesco; D’Onofrio, Luca; Leto, Gaetano; Mastroianni, Carlo Maria; Mignogna, Carmen; Pascarella, Giuseppe; Pugliese, Francesco; Ali, Hamad; Al Mulla, Fahd; Buzzetti, Raffaella; Pozzilli, Paolo title: Development of a clinical risk score to predict death in patients with COVID‐19 date: 2022-03-22 journal: Diabetes Metab Res Rev DOI: 10.1002/dmrr.3526 sha: a6e0d868b44a404a4d30bf6af2867e4d563eaf2d doc_id: 1004898 cord_uid: w0br1t79 OBJECTIVE: To build a clinical risk score to aid risk stratification among hospitalised COVID‐19 patients. METHODS: The score was built using data of 417 consecutive COVID‐19 in patients from Kuwait. Risk factors for COVID‐19 mortality were identified by multivariate logistic regressions and assigned weighted points proportional to their beta coefficient values. A final score was obtained for each patient and tested against death to calculate an Receiver‐operating characteristic curve. Youden's index was used to determine the cut‐off value for death prediction risk. The score was internally validated using another COVID‐19 Kuwaiti‐patient cohort of 923 patients. External validation was carried out using 178 patients from the Italian CoViDiab cohort. RESULTS: Deceased COVID‐19 patients more likely showed glucose levels of 7.0–11.1 mmol/L (34.4%, p < 0.0001) or >11.1 mmol/L (44.3%, p < 0.0001), and comorbidities such as diabetes and hypertension compared to those who survived (39.3% vs. 20.4% [p = 0.0027] and 45.9% vs. 26.6% [p = 0.0036], respectively). The risk factors for in‐hospital mortality in the final model were gender, nationality, asthma, and glucose categories (<5.0, 5.5–6.9, 7.0–11.1, or 11.1 > mmol/L). A score of ≥5.5 points predicted death with 75% sensitivity and 86.3% specificity (area under the curve (AUC) 0.901). Internal validation resulted in an AUC of 0.826, and external validation showed an AUC of 0.687. CONCLUSION: This clinical risk score was built with easy‐to‐collect data and had good probability of predicting in‐hospital death among COVID‐19 patients. (39.3% vs. 20.4% [p = 0.0027] and 45.9% vs. 26 .6% [p = 0.0036], respectively). The risk factors for in-hospital mortality in the final model were gender, nationality, asthma, and glucose categories (<5.0, 5.5-6.9, 7.0-11.1, or 11.1 > mmol/L). A score of ≥5.5 points predicted death with 75% sensitivity and 86.3% specificity (area under the curve (AUC) 0.901). Internal validation resulted in an AUC of 0.826, and external validation showed an AUC of 0.687. This clinical risk score was built with easy-to-collect data and had good probability of predicting in-hospital death among COVID-19 patients. clinical risk score, comorbidities, COVID-19, glucose control, hyperglycemia, intensive care Severe acute respiratory syndrome coronavirus 2 (SARs-CoV-2), causing coronavirus disease 2019 (COVID- 19) , is currently the greatest public health threat in the world. 1 Since its emergence at the end of 2019, it has spread globally and resulted in the World Health Organization (WHO) categorising it as a worldwide pandemic in 2 March020. 2, 3 Although this is not the first coronavirus to infect the human population (SARs-CoV-1 and MERs), the velocity of SARs-CoV-2 transmission differentiates it from other viruses. Furthermore, its ability to result in fatal disease and acute respiratory distress syndrome (ARDS) necessitates the development for effective treatment and prevention strategies. [4] [5] [6] Individuals with COVID-19 vary from asymptomatic to critically severe cases that lead to ARDS, intensive care unit (ICU) admissions, invasive mechanical ventilation, and mortality. 7, 8 Across borders, severe cases of COVID-19 have been seen in patients who are predominantly male, older than 65 years, and have one or more comorbidities, with hypertension, diabetes, and cardiovascular disease (CVD) being the most pertinent. [9] [10] [11] [12] [13] Increased risk in individuals with comorbidities is possibly due to the mode in which SARs-CoV-2 infects cells and spreads within the body. SARs-CoV-2 binds via its Spike (S) protein to the angiotensin-converting enzyme 2 (ACE2) receptor, facilitating its entry into cells of the respiratory tract. It has been reported that SARs-CoV-2 has a 10-to 20-fold higher binding affinity to ACE2 than previous coronaviruses (namely SARs-CoV-1), hence increasing the uptake of SARs-CoV-2 and aiding in its increased pathogenicity. [14] [15] [16] While many studies have reported the clinical findings of hospitalised COVID-19 patients, hospitals and healthcare staff globally remain overwhelmed. Though the development of vaccines against COVID-19 17-20 signals hope for controlling the pandemic, tens of thousands of new cases are being reported every day and hospital staff need to be able to predict which patients are more likely to succumb to a severe form of COVID-19, ARDS, ICU admission, or even death. In this current study, we attempt to build a clinical risk score to aid clinicians in identifying patients more likely to develop critical cases of COVID-19 to better optimise care. Additionally, the development of new therapies with limited access to the general population, such as tocilizumab, might benefit from such a risk score. 21 The score was built using clinical data of a COVID-19 cohort from Kuwait and then validated with an external cohort of Italian COVID-19 patients (CoViDiab). To build the clinical risk score, we used data collected retrospectively from 417 consecutive patients positive for COVID-19 who were hospitalised at Jaber Al-Ahmad hospital (Kuwait) between February 24 th and 3 May 2020. 8 At a time when the local policy was to admit anyone with a positive COVID-19 real-time polymerase chain reaction (RT-PCR) of a nasopharyngeal swab regardless of symptom status. 22 Hence, the population included in the analysis ranged from asymptomatic to severe cases. Asymptomatic patients were defined as patients who had an RT-PCR positive for COVID-19, but who presented with no symptoms and did not require ICU treatment. On the other hand, symptomatic patients were those who had mild to moderate symptoms typical of COVID-19. These were patients who could still be treated in the wards and did not require ICU admission. Severe cases were characterised as those that require ICU admission, mechanical ventilation, and includes those that lead to mortality. Descriptive statistics are presented for categorical variables as numbers with proportions and for continuous variables as appropriate measures of central tendency and dispersion. Student's t-test was used to compare differences in continuous variables between groups, categorical variables were compared with a X 2 test. The primary outcome was defined as in-hospital death. Multivariate logistic regression models were then used to identify independent prognostic factors for the primary outcome. They were built step-bystep by adding or removing variables based on the results of previous models and retaining in the final model variables associated with the outcome of a nominal p-value <0.1. Before entering in the model, continuous variables (age and blood glucose) were converted into ordinal variables based on recognized cut-offs (age: <50, 51-70, and 70 > years of age; blood glucose: <5.5, 5.5-6.9, 7.0-11.1, and 11.1 > mmol/L). Briefly, we initially tested comorbidities such as hypertension, diabetes, malignancy, chronic renal disease, and asthma against the primary outcome via logistic regressions. Then we performed a similar logistic regression using demographic information. The predictive variables that proved to be significant at the nominal p-value <0.1 were carried out by performing additional regressions merging the data. The final model included gender, non-Kuwaiti national, asthma, and glucose categories as the predictive variables. Significant risk factors were assigned weighted points that were proportional to their beta regression coefficient values. The reference group of categorical variables were assigned 0 points, corresponding to a betacoefficient of zero. Receiver-operating characteristic (ROC) curve analyses were performed to assess the effectiveness of our risk score to predict death in patients hospitalised for COVID-19. In terms of disease prediction, typically an area under the curve (AUC) of 0.5 or less is considered insignificant, an AUC of 0.7-0.8 is considered an acceptable fit, meaning that the score can somewhat be able to predict patients more likely to proceed to the main outcome, and an AUC of 0.9 or above is a great fit for the score, indicating that the score is able to predict the outcome with a high degree of confidence. 23 The cut-off for death prediction was determined using The score was both internally and externally validated. Internal validation was performed by calculating the total clinical risk score for each patient within a separate COVID-19 cohort from Kuwait (admitted between May 4 th and 26 August 2020). Patient data was obtained retrospectively, and inclusion criteria in the validation cohort was based on the presence of admission data and availability of discharge information (either dead or alive). Patients lacking this information were excluded from the validation cohort. The respective scores were tested against the main outcome and analysed by an ROC, the "goodness of fit" of the score was determined by the AUC of the ROC curve. The cohort used for external validation of the score was composed of patients from the Italian CoViDiab cohort. As previously described, 24 CoViDiab is a multi-center observational study collecting data retrospectively from medical charts of patients hospitalized for COVID-19 in four academic hospitals located in the Lazio region of Italy up to 15 May 2020. Patients eligible for inclusion were aged ≥18 years old with a diagnosis of COVID-19 confirmed by at least one RT-PCR in agreement with the protocol set by the WHO. All the clinical data needed to calculate the proposed score and discharge information (either dead or alive) were available in 178 of the 354 patients originally enrolled in the CoViDiab study. Ethical approval was obtained from the standing committee for co- Nationality, gender, asthma, and blood glucose levels were the predictive variables independently associated with the primary outcome. Each variable was allocated a specific score based on the calculated beta coefficients of each predictive variable, as shown in Table 3 . The maximum allocated score was 12.5. The cut-off value to predict death was 5.5, which showed a sensitivity of 75% and specificity of 86.3% to predict the outcome (AUC 0.901). The score was internally and externally validated in order to assess its predictive potential (Table 4, Figure S1 ). Internal validation was performed on two cohorts from the COVID-19 population in Kuwait. This study developed a clinical risk score for the prediction of severe disease and death in COVID-19 patients. The score was developed retrospectively, utilising data from 417 consecutive patients hospitalised in one COVID-19 centre in Kuwait. 8 The score was based on assessing clinical and comorbid data from these patients, focussing on clinical data that would be routinely collected in any hospital or health centre internationally. The Kuwaiti COVID-19 cohort used to build the clinical risk score was symptomatically diverse. This was primarily due to the initial steps the Kuwaiti government had taken within the first Blood glucose ≥ 11.1 mmol/L 5.0 Note: This was calculated based on the Youden's index of the score (the point on the ROC curve that retains high sensitivity and 1-specificity). Abbreviation: ROC, receiver-operating characteristic. ALHAMAR ET AL. -5 of 8 to poorer outcome and increased mortality in COVID-19 patients. 31 Reports have suggested that these individuals are more at risk due to the pathophysiology of their underlying conditions. For instance, individuals with diabetes, especially diabetic patients with uncontrolled hyperglycaemia, have compromised innate and humoral immune systems. Diabetes can attribute to a proinflammatory state; thus, when diabetic patients contract COVID-19, it has been reported that they have a significant increase in systemic levels of C-reactive protein and interleukin-6 (IL-6). In addition, increased recruitment of T helper cells, triggering an already exacerbated inflammatory response and increased production of interferon gamma, results in a cytokine storm. 12, 32 The final model for the score included being male, non-Kuwaiti national, having asthma, blood glucose between 7.0 and 11.1 mmol/ L and glucose levels greater than 11.1 mmol/L. When tested within the primary cohort, the ROC curve had an AUC of 0.901 with an Negative predictive value (NPV) of 95.4%, indicating a great fit for the curve with a high probability to distinguish those without the primary outcome from those with the outcome. In our cohort, being non-Kuwaiti may attribute to the primary outcome due to the socioeconomic differences present. Non-Kuwaitis were predominantly of South Asian descent, these individuals are more likely to be male, laborers, and living in tight quarters. Hence, increasing their susceptibility to contracting SARs-CoV-2. 13 We saw that the addition of hyperglycaemia was a pushing factor for severe outcome regardless of the diabetic state. 35 Thus, suggesting that if a patient is identified as having a lower risk of succumbing to the main outcome with our developed score, we can have a good degree of confidence that this is a true negative. The retrospective nature of this study made it difficult to obtain data that was lacking, such as BMI information and HbA1C. This missing data may add critical information that may impact the development of severe COVID-19 and our clinical risk score. Furthermore, it is important to note that glucose management and treatment differences among our internal and external validation cohorts may also impact results. It is also important to note that information for Note: The Italian CoviDIAB (178 patients) cohort was used for external validation. %Sensitivity and %Specificity were derived from the ROC analysis using the Youden's index (5.5) . AUC represents the AUC, an AUC of 0.9-1.0 is an excellent fit for the model, 0.8-0.7 is a great fit, and 0.6 indicates a good fit. Using a cut-off score of 5.5, PPV and NPV were calculated based on the following formula: PPV = True positive/True Positive + False Positive, NPV = True Negative/True Negative + False Negative. Abbreviations: AUC, area under the curve; NPV, negative predictive value; PPV, positive predictive value; ROC, receiver-operating characteristic. asthma was missing from the CoViDiab population. Another limitation is the fact that the CoViDiab population is mainly focussed on diabetic patients unlike the general population admitted to hospitals. Nonetheless, given the high rate of diabetes, this population still constitutes a valid study population. Lastly, to deduce the true efficacy of the score, further external validation is required. The proposed risk score built with easy-to-collect clinical data had good performance for predicting in-hospital death among patients with COVID-19. I would like to acknowledge the department of endocrinology and diabetes at Campus Biomedico University of Rome for their support as well as the team from DDI and MOH in Kuwait, for their help and continued aid. The authors declare no competing interests. Ethical approval was obtained from the standing committee for coordination of health and medical research at the Ministry of Health in Kuwait (IRB 2020/1404). Ghadeer Alhamar was responsible for drafting the article, study design, data analysis and interpretation. Ernesto Maddaloni and Raffaella Buzzetti conceptualised the design of the study, were involved in data analysis and interpretation, and revising the manuscript. Paolo Pozzilli was involved in revision of the manuscript, as well as had final approval of the version to be published. The data that support the findings of this study are available from the corresponding author upon reasonable request. Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) and coronavirus disease-2019 (COVID-19): the epidemic and the challenges COVID-19 pathophysiology: a review An overview of COVID-19 ACE2 and COVID-19 and the resulting ARDS COVID-19: discovery, diagnostics and drug development The impact of strict public health measures on COVID-19 transmission in developing countries: the case of Kuwait The COVID-19 cytokine storm; what we know so far Clinical characteristics of coronavirus disease 2019 (COVID-19) patients in Kuwait Clinical characteristics of patients who died of coronavirus disease 2019 in China Transmission, Diagnosis, and Treatment of Coronavirus Disease 2019 (COVID-19): A Review Fasting blood glucose and COVID-19 severity: nonlinearity matters Impact of diabetes in patients diagnosed with COVID-19 Outcomes of COVID-19: disparities by ethnicity Angiotensin-converting enzyme 2 (ACE2), SARS-CoV-2 and the pathophysiology of coronavirus disease 2019 (COVID-19) ACE2 and FURIN variants are potential predictors of SARS-CoV-2 outcome: a time to implement precision medicine against COVID-19 SARS-CoV-2: possible recombination and emergence of potentially more virulent strains Robust antibody levels in both diabetic and non-diabetic individuals after BNT162b2 mRNA COVID-19 vaccination Previous COVID-19 infection and antibody levels after vaccination. Front Public Health Immunogenicity of BNT162b2 vaccine in patients with inflammatory bowel disease on infliximab combination therapy: a multicenter prospective study Serological response to BNT162b2 and ChAdOx1 nCoV-19 vaccines in patients with inflammatory bowel disease on biologic therapies Interleukin-6 Receptor Antagonists in Critically Ill Patients with Covid-19 Clinical characteristics and outcomes of COVID-19 patients with diabetes mellitus in Kuwait Receiver operating characteristic curve in diagnostic test assessment Cardiometabolic multimorbidity is associated with a worse Covid-19 prognosis than individual cardiometabolic risk factors: a multicentre retrospective study (CoViDiab II) Characteristics, risk factors and outcomes among the first consecutive 1096 patients diagnosed with COVID-19 in Kuwait Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19 Clinical risk score to predict in-hospital mortality in COVID-19 patients: a retrospective cohort study Advancing risk analysis of COVID-19 clinical predictors: the case of fasting blood glucose Obesity and diabetes mellitus in the Arab world Prevalence of coronary risk factors in healthy adult Kuwaitis COVID-19 and comorbidities: deleterious impact on infected patients COVID-19 and diabetes: the why, the what and the how Managing hyperglycemia in the COVID-19 inflammatory storm Admission hyperglycemia in non-diabetics predicts mortality and disease severity in COVID-19: a pooled analysis and meta-summary of literature Understanding and using sensitivity, specificity and predictive values