key: cord-0811499-lm2v1qr6
authors: Kabootari, Maryam; Habibi Tirtashi, Reza; Hasheminia, Mitra; Bozorgmanesh, Mohammadreza; Khalili, Davood; Akbari, Hamideh; Roshandel, Gholamreza; Hadaegh, Farzad
title: Clinical features, risk factors and a prediction model for in-hospital mortality among diabetic patients infected with COVID-19: data from a referral centre in Iran
date: 2021-11-17
journal: Public Health
DOI: 10.1016/j.puhe.2021.11.007
sha: def57a8a663e5e6b8a5f7dc0a03f9ebdbaef62cb
doc_id: 811499
cord_uid: lm2v1qr6

Objectives The aim of this study was to identify risk factors of in-hospital mortality among diabetic patients infected with COVID-19. Study design Retrospective cohort study. Methods Using logistic regression analysis, the independent association of potential prognostic factors and COVID-19 in-hospital mortality was investigated in three models. Model 1 included demographic data and patient history; model 2 consisted of model 1, plus vital signs and pulse oximetry measurements at hospital admission; and model 3 included model 2, laboratory test results at hospital admission. The odds ratios (ORs) and 95% confidence intervals (95% CIs) were reported for each predictor in the different models. Moreover, to examine the discriminatory powers of the models, a corrected area under the receiver-operating characteristic curve (AUC) was calculated. Results Among 560 patients with diabetes (men = 291) who were hospitalised for COVID-19, the mean age of the study population was 61.8 (standard deviation [SD] 13.4) years. During a median length of hospitalisation of 6 days, 165 deaths (men = 93) were recorded. In model 1, age and a history of cognitive impairment were associated with higher mortality; however, taking statins, oral anti-diabetes drugs and beta-blockers were associated with a lower risk of mortality (AUC = 0.76). In model 2, adding the data for respiratory rate (OR 1.07 [95% CI 1.00–1.14]) and oxygen saturation (OR 0.95 [95% CI 0.92–0.98]) slightly increased the AUC to 0.80. In model 3, the data for platelet count (OR 0.99 [95% CI 0.99–1.00]), lactate dehydrogenase (OR 1.002 [1.001–1.003]), potassium (OR 2.02 (95% CI 1.33–3.08]) and fasting plasma glucose (OR 1.04 [1.02–1.07]) significantly improved the discriminatory power of the model to AUC 0.86 (95% CI 0.83–0.90). Conclusions Among patients with type 2 diabetes, a combination of past history and pulse oximetry data, with four non-expensive laboratory measures, was significantly associated with in-hospital COVID-19 mortality.

Diabetes is one of the most frequent comorbidities in patients who are hospitalised for coronavirus 37 disease 2019 1 . Previous systematic reviews have demonstrated that diabetes is a risk 38 factor for severe disease and is associated with an approximately 2-3 fold increased mortality rate 39 from COVID-19 compared with patients without diabetes 2-5 . Results of studies among patients 40 with diabetes have shown that some phenotypic characteristics, radiological and laboratory 41 parameters have been associated with the severity of COVID-19 6, 7 ; however, diabetes is a 42 heterogeneous disease, and specific phenotypes associated with poorer outcome are inconsistent 43 among studies. In addition, model development studies that predict outcomes among patients with 44 diabetes are sparse due to insufficient sample sizes 8, 9 . 45 In Iran, more than 4,580,000 confirmed cases of COVID-19 and 100,255 deaths had been reported 46 (until 20 August 2021), according to the World Health Organisation (WHO) report 10 . Furthermore, 47 compared with other countries in the Middle East and North Africa (MENA) region, Iran has the 48 highest total number of COVID-19 deaths (as of 20 August 2021) 11 . A multi-centre, cross-49 sectional study conducted in 19 hospitals in Tehran, Iran, showed a case fatality rate (CFR) of 50 10.05% among 16,000 cases of COVID-19 12 . In that study, the highest rate of mortality was 51 observed in patients with diabetes. In another single-centre study including 2968 Iranian patients 52 who were hospitalised with COVID-19, patients with diabetes had significantly higher rates of 53 CFR compared with patients who had no comorbidities (9.73% vs 7.61%) 13 . 54

Globally, in 2017, the MENA region had the second highest prevalence of type 2 diabetes (10.8%), 55

with an increasing trend of 1.5-2 times in the past three decades 14 . Hence, it was expected that 56 during the COVID-19 pandemic, patients with diabetes in this region would be greatly impacted. 57

As the burden of disease due to diabetes 15 and COVID-19 10 increases in Iran, the current study 58 aims to: (1) describe the clinical and laboratory characteristics of patients with diabetes and 59 COVID-19; (2) identify the risk factors of in-hospital mortality among these patients; and (3) 60 develop a predictive model for in-hospital mortality among Iranian adult patients with type 2 61 diabetes who were hospitalised for COVID-19. optimism-corrected AUC was estimated using 1000 bootstrap resamples for every underlying 129 model. The difference between the original and the mean AUC of the 1000 replicates was used as 130 a correction factor and subtracted from the original AUC. This bias-corrected AUC was used as a 131 measure for internal validation. 132

In order to evaluate the calibration, which shows agreement between the observed (actual) 133 outcomes and predictions, we used observed to predicted ratios, the Hosmer-Lemeshow goodness-134 of-fit test and a calibration plot. The calibration plot shows predicted in-hospital death probabilities 135 (x-axis) against the observed outcomes (y-axis) in deciles of the predicted probabilities. Using the 136 LOWESS (locally weighted scatter plot smoothing) line, we smoothed the calibration plot. Perfect 137 predictions are on the 45° line (y=x). Validation of the goodness of fit of each underlying model 138 was determined by the Hosmer-Lemeshow test in deciles based on the predicted risk. A non-139 significant test implied that the observed outcome did not differ significantly from the predicted 140 mortality risk. 141

To encourage the integration of the prognostic model into everyday clinical situations, the 142 mathematical formula of the prognostic algorithm obtained from logistic regression modelling was 143 also incorporated into a nomogram. The nomogram developed herein serves as a graphical 144 representation of our prognostic algorithm, incorporating significant prognostic factors as 145 continuous variables to predict the risk of in-hospital mortality from COVID-19. Except for the 146 variable selection, P <0.05 was considered significant. Statistical analyses were performed with 147 SPSS 22 (SPSS Inc., Chicago, IL, USA) and STATA 14 (StataCorp, college station, TX, USA). 148 in-hospital deaths were recorded (men = 93). 161

The baseline characteristics of patients who survived and those who died are compared in Table  162 1. The prevalence of overweight, obesity and CKD were 37.2%, 38.6% and 44.5%, respectively. 163

A medical history of hypertension, coronary artery disease (CAD), stroke and pulmonary disease 164 was observed in 78.9%, 42.7%, 10.4% and 18.6% of the participants, respectively. The mean level 165 of plasma glucose at the time of hospital admission was 231.4 (114.6) mg/dl, and the level of 166 HbA1c (only for 70 patients) was 9.4%. The most common glucose-lowering medications were 167 metformin, followed by insulin, sulfonylurea and other oral glucose-lowering agents. Moreover, 168 ACE inhibitors and/or ARBs, beta-blockers, statins and antiplatelet drugs were used by 44.5%, 169

13.6%, 19.8% and 22.1% of the participants, respectively. 170

The most common signs of COVID-19 on admission were dyspnoea, cough, fever, fatigue, 171 gastrointestinal disorders, cognitive impairment and anosmia, hyposmia and ageusia. Thoracic 172 computed tomography (CT) imaging was performed for all patients at hospital admission and did 173 not reveal any abnormality in 25% of patients. Details of other results are shown in Table 1 . 174

Patients who died compared with those who survived were older, more likely to have a history of 175 stroke, and present with gastrointestinal symptoms and cognitive impairment. Moreover, they were 176 less likely to be taking metformin and statins. In-hospital mortality was more likely in individuals 177 who initially presented (i.e. at hospital admission) with significantly lower systolic blood pressure 178 (SBP), diastolic blood pressure (DBP), oxygen saturation (SpO2), and lower levels of 179 lymphocytes, platelet, albumin and eGFR, but higher levels of respiratory rate, WBC, neutrophils, 180 LDH, CPK, creatinine, potassium, AST, ALT and FPG compared with patients who survived (all 181 P-values <0.05). 182 Table 2 shows multivariate prediction models for in-hospital mortality. In model 1, age (OR 1.02 183 [95% CI 1.00-1.04]) and a history of cognitive impairment (OR 3.17 (95% CI 1.77-5.68]) were 184 associated with a significantly higher risk of in-hospital mortality. Moreover, prior use of OADs, 185 beta-blockers and statins were associated with the significant 55%, 51% and 49% lower risks of 186 mortality, respectively. In model 2, age and history of cognitive impairment were independently 187 associated with a higher risk of mortality, while the use of statins, beta-blockers, OADs, lower 188 respiratory rate (OR 1.07 [95% CI 1.00-1.14]) and higher oxygen saturation (OR 0.95 were found to be independently associated with the risk of death. OAD treatment gets the score of (1.5 + 1 + 1 + 2 + 3.5 + 4 + 3 + 0 + 1 + 1.2 + 1 = 19.2) and will 205 have a 95% probability of mortality. 206

207

The current study was conducted in a large tertiary centre in the North East of Iran during the first 209 half of 2020. Our findings, among 560 patients with diabetes who were hospitalised for COVID-210 treated obstructive sleep apnoea, and microvascular and macrovascular complications, to be 222 independent predictors of death on day 7 (the current study observed a higher risk of mortality, 223 and the only common predictor of death with our population was age). The other study among 224 patients with diabetes who were hospitalised with COVID-19 was in the US 6 and showed a 225 mortality rate of 33.1%, which is comparable with results obtained in our study. Furthermore, the 226 US study showed that HbA1c was not associated with mortality events, while insulin treatment 227 was a strong predictor of mortality. In the current study, we found that a high level of plasma 228 glucose at hospital admission, as a proxy for the level of diabetes control, 26 was significantly 229 associated with an increased risk of mortality. Moreover, in contrast to the Agarwal et al. study, 230

we demonstrated that a history of using OADs and insulin was significantly associated with a 231 lower risk of in-hospital mortality. Importantly, according to a national study, 27 despite some 232 improvement in the knowledge and screening of diabetes in Iran, 24.7% of patients with diabetes 233

were not aware of their disease. In addition, these newly diagnosed patients exhibited a coronary 234 heart disease (CHD) risk comparable to patients without diabetes with a prior CHD event 28 . 235

Results of a systematic review and meta-analysis of 14 studies showed that male sex, age, history 236 of cardiovascular disease (CVD), CKD, chronic obstructive pulmonary disease (COPD), high 237 plasma glucose at hospital admission and chronic insulin use were associated with a high risk of 238 death for patients with diabetes who also had COVID-19 25 . 239

Among patients with diabetes, a few studies found no association between statin use and poor 240 outcome 7, 22, 25, 29 . Several studies 30, 31 with larger sample sizes among patients without diabetes 241 showed that previous statin use in patients hospitalised with COVID-19 was associated with lower 242 in-hospital mortality, which might be related to the immunomodulatory action or by preventing 243 cardiovascular damage in addition to their lipid-lowering activity 32 . These results are consistent 244 with our study that also found a 63% lower risk of death among patients who used statins, despite 245 the low number of lipid-lowering medication users in our study population. In another study, low 246 statin use was also observed among patients with diabetes in Iran 33 . It is interesting that the current 247 study found that using beta-blockers was significantly associated with a lower risk of in-hospital 248 mortality. The results of other studies among patients with diabetes who were infected with 249 COVID-19 were inconsistent; one study showed a 19% higher risk, and another study found a 33% diabetes is limited to those with acute coronary syndrome or who are experiencing heart failure 253 (HF) 34 . Systematic reviews revealed that using beta-blockers was associated with improved 254 outcomes among patients with HF, regardless of diabetes status 35 . Hence, in the current study 255 population, with a prevalence of CVD >40%, we speculated that a large number of patients had 256 some degree of existing HF and would benefit from beta-blockers. Unfortunately, the data of 257 ejection fraction were not available for our patients. 258

To the best of our knowledge, no study has investigated the association between potassium level 259 and COVID-19 mortality among patients with diabetes; however, in studies conducted in the 260 general population, potassium was not associated with COVID-19 mortality risk 36 or a higher risk 261 37 of COVID-19 infection. In the current study, each 1 mEq/L higher potassium level was 262 associated with a two-fold higher risk of in-hospital mortality. These findings are in line with 263 previous studies among ICU patients showing that hyperkalaemia is an independent risk factor for 264 death, even at a moderate increase above normal levels 38, 39 . Importantly, the significant risk of 265 higher potassium level, which was found in the current study, was independent of several 266 important confounding factors, including eGFR and ACE inhibitor/ARB use. However, pH value 267 is a strong confounder for potassium level as metabolic acidosis can cause a potassium shift from 268 the intracellular to the extracellular space; unfortunately, we did not have data on pH levels. The 269 most important mechanism that is caused by hyperkalaemia is a lowering of the resting membrane The current findings need to be interpreted in light of the study limitations. First, this study did 285 not validate the model externally; however, the model showed a reasonable internal validity as 286 examined by the bootstrapping method. Second, there were no data on HbA1c; however, we 287 adjusted for plasma glucose as a surrogate of HbA1c level 26 . Third, a large number of patients 288

were probable cases of COVID-19, and the cases were not confirmed with PCR. However, 289

previous studies have shown that because of difficulty in obtaining reliable nasopharyngeal swab 290 specimens, the timing of detection and limited detection capacity during the early stages of the 291 outbreak, false-negative results are often seen in the PCR method 40 . Moreover, CT imaging of the 292 chest is a more reliable, feasible and rapid method to diagnose and assess COVID-19 compared 293

with RT-PCR, especially in epidemic regions such as Iran 41-43 . Fourth, we did not categorise 294 patients as those with previous diabetes and those who had increased blood glucose due to COVID-295 19 infection, which might overestimate the number of patients with diabetes. Finally, data 296 regarding in-hospital treatment was not included in our analysis. 297

In conclusion, approximately one-third of patients with diabetes who were hospitalised for 298 COVID-19 in this large referral centre located in the North East of Iran died within 1 week of 299 admission. A simple and non-expensive risk score consisting of 11 variables, including age, history 300 of cognitive impairment, use of statins, OADs, insulin, beta-blockers, SpO2, platelet count, LDH, 301 potassium and FPG levels, demonstrated excellent prediction for in-hospital mortality among 302 patients with diabetes. This simple risk score may help physicians in emergency departments to 303 assess the prognosis of patients with diabetes. J o u r n a l P r e -p r o o f 

Vital signs on admission SBP (mmHg), mean ± SD

SPO2 (%)

Prothrombin time (s), median (IQR) 283

HbA1c (%), median (IQR) 70

LDH (U/L), median (IQR)

Creatinine (µmol/L), median (IQR) 559