Glycaemic variability is associated with all-cause mortality in COVID-19 patients with ARDS, a retrospective subcohort study | Scientific Reports -…

Study population

106 patients with laboratory-confirmed COVID-19 who were admitted to the ICU for treatment of ARDS at University Hospital Aachen, Germany, were recruited to this retrospective subcohort study. 59 of these patients were published8,9 previously in respect to a single-centre cohort study, COVAS. Patients in the present study were admitted between February 24, 2020, until May 15, 2021 and fulfilled the following criteria (Fig.1). Inclusion criteria for the subcohort were a positive respiratory SARS-CoV-2 PCR result and admission to the ICU requiring mechanical ventilation due to COVID-19 and ARDS. Exclusion criteria of this study were lack of consent, positive PCR result or age of majority, as well as pregnancy or inability to legally give consent. In order to calculate the variability of FPG levels, we excluded patients who did not have at least 3 days of consecutive glucose measurements during their admission to the ICU.

Flowchart of patient enrolment. The register population represents all patients included in the COVAS cohort. A total of 106 out of 271 patients were enrolled in this subgroup analysis. ARDS acute respiratory distress syndrome; DGV daily glycaemic variability.

Since University Hospital Aachen is designated a tertiary care facility, patients with minor or mild severity were triaged by the emergency services towards other regional hospitals. Thus, the present study includes a significant number of patients with severe clinical course from other regional hospitals, who were either previously screened for ECMO or other high-end treatment methods.

All patients gave their written informed consent before participating in the COVAS study, which complies with the Declaration of Helsinki. Study approval was acquired by the ethics committee at the Faculty of Medicine of RWTH Aachen University (EK080/20). This trial has been retrospectively registered in the German Clinical Trials Register (DRKS00027106).

Based on national guidelines10 and internal standards of operation, all patients with an FPG above 180mg/dl were titrated to a FPG target of 150mg/dl using continuous insulin infusion during ICU admission.

ARDS was defined according to the Berlin definition11. Comorbidities were defined as conditions that were known before hospital admission. Likewise, previous medication included any medication prescribed before admission to our hospital.

Baseline vital parameters are characterized as the first available measurements after ICU admission. Respiratory disease was defined as a composite of bronchial asthma, chronic obstructive pulmonary disease, obstructive sleep apnoea syndrome and pulmonary malignancy. Moreover, composite heart disease is a composite of arterial hypertension, atrial fibrillation, coronary artery disease, heart failure and previous myocardial infarction. History of T2D was specified either by a previously known T2D diagnosis, diabetes medication at time of hospital admission or HbA1c at admission of6.5% (48mmol/mol).

For outcome measures, the primary endpoint was defined as all-cause mortality during ICU admission. As exposures, high and low glycaemic variabilities during ICU admission were defined as daily glycaemic variability (DGV)25.5mg/dl and DGV<25.5mg/dl. To determine this cut-off for DGV, we fitted a regression tree model (25.5mg/dl) and compared it to a cut-off based on a hazard ratio of 1 derived from a Cox-PH model, which was adjusted for age, sex and history of T2D (31mg/dl). The regression tree-based cut-off demonstrated a higher AUC (0.729 vs. 0.689) in 30-day survivalROC curves, therefore we used the cut-off DGV value of 25.5mg/dl in further models and testing rather than the Cox-PH based cut-off of DGV 31mg/dl.

We collected symptoms on admission, co-morbidities and previous medication either per interview/questionnaire in alert patients or per admission/discharge documents from our emergency department and previous hospitals. Vital parameters were acquired immediately on the first day of ICU admission. On subsequent days, we recorded the worst daily value, in the context of shock and/or respiratory failure. All data was manually retrieved from our EHR software, which automatically transfers ventilation parameters at set intervals from the ventilator to the patients electronic health record. In order to reduce confounders in ventilation parameters due to this automated process and initially extreme ventilation parameters, we intentionally omitted the first four hours of ventilation parameters after admission and intubation to allow the staff to properly configure the ventilator according to the patients requirements at the time.

PCR results were acquired by quantitative real-time polymerase-chain-reaction (PCR). Diagnosis of COVID-19 was established by positive respiratory PCR from either a throat swab or tracheal fluid in awake patients and bronchoalveolar lavage (BAL) in intubated patients. Respiratory PCR was repeated on days 7 and 14 of admission. Additionally, BAL, serum, stool and urine samples were tested for bacterial, fungal and viral pathogens, including Legionella pneumophila and Streptococcus pneumoniae antigens as well as SARS-CoV-2. All patients received daily routine laboratory tests including glucose levels between 03:00 05:00 AM.

All statistical analysis was performed in R version 4.1.212 using packages ggplot2 (version 3.3.5)13 for scatter plots, tangram (0.7.1)14 for tables and Rmarkdown for text. The characteristics were described as median (IQR) for continuous and percentages for categorical variables. Categorical parameters were compared by Fishers Exact Test and continuous parameters by KruskalWallis test. Statistical significance was determined as a p-value below 0.05. We opted not to do any parameter imputation for missing values.

In order to select a suitable metric to evaluate fasting plasma glucose variability, we first compared established parameters: standard deviation (SD), Neumans (root) mean square of successive differences (MSSD and rMSSD), bias corrected coefficient of variation (CoefVar) and median of the absolute difference between successive values (DGV, daily glycaemic variability). In order to compute DGV, we first calculated the absolute differences of FPG (FPG) for consecutive days, where FPGday represents the fasting plasma glucose of the current day and FPGday+1 the fasting plasma glucose of the following day (Eq.1):

$$Delta FP{G}_{day}=left|FP{G}_{day}-FP{G}_{day+1}right|$$

(1)

Then, we calculated DGV as the median of all (FPG) values for each patient.

We calculated MSSD and rMSSD using the psych package (version 2.1.9)15 and CoefVar using the implementation provided by the DescTools package (version 0.99.44)16.

The cut-off DGV was estimated by regression tree analysis using rpart (version 4.1.16)17. Through rms (version 6.2.0)18, smooth hazard ratios and survival analysis were examined in Cox-proportional-hazard (Cox-PH) regression models, which were compared by likelihood ratio (LR) test, Akaike information criterion (AIC) and Concordance Index (C-Index).

All Cox-PH models were tested for the proportional hazard assumption as well as for collinearity utilising the variance inflation factor (vif function rms18 package). While recommendations vary, in accordance with most studies, we defined a VIF<5 as acceptable.

Before analysis and based on clinical judgment we selected the following confounders for adjustment of our final Cox-PH models: age, sex, BMI, history of type 2 diabetes (T2D), dialysis during admission, dexamethasone treatment, median procalcitonin (PCT) and FPG during ICU admission. To reduce overadjustment of the models, we removed BMI and dialysis during admission from the final model. For this model we additionally used Firths penalized maximum likelihood bias reduction method, provided by the coxphf (version 1.13.1)19 package.

To evaluate the accuracy of the outcome-based cut-offs, we compared the AUC in 30-day survival models using the implementation of survivalROC (version 1.0.3)20.

Utilizing survminers (version 0.4.9)21 ggforest function, forest plots were created. Furthermore, KaplanMeier estimator was calculated with the survival (version 3.2.13)22 package and plotted with survminers (version 0.4.9)21 ggsurvplot function, which compared survival curves and computed p-values using the log-rank test.

Study approval was acquired by the ethics committee at the Faculty of Medicine of RWTH Aachen University (EK080/20). The present research complies with the Declaration of Helsinki.

View original post here:

Glycaemic variability is associated with all-cause mortality in COVID-19 patients with ARDS, a retrospective subcohort study | Scientific Reports -...