Risk factors for hyperglycemia in COVID‐19 patients treated with remdesivir

Abstract

The primary objective of this study was to investigate the factors contributing to hyperglycemic adverse events (AEs) associated with the administration of remdesivir in hospitalized patients diagnosed with coronavirus disease 2019 (COVID‐19). Furthermore, the study aimed to develop a risk score model employing various machine learning approaches. A total of 1262 patients were enrolled in this investigation. The relationship between covariates and hyperglycemic AEs was assessed through logistic regression analysis. Diverse machine learning algorithms were employed for the purpose of forecasting hyperglycemia‐related complications. After adjusting for covariates, individuals with a body mass index ≥23 kg/m², those using proton pump inhibitors, cholinergic medications, or individuals with cardiovascular diseases exhibited approximately 2.41‐, 2.73‐, 2.65‐, and 1.97‐fold higher risks of experiencing hyperglycemic AEs (95% CI 1.271–4.577, 1.223–6.081, 1.168–5.989, and 1.119–3.472, respectively). Multivariate logistic regression, elastic net, and random forest models displayed area under the receiver operating characteristic curve values of 0.65, 0.66, and 0.60, respectively (95% CI 0.572–0.719, 0.640–0.671, and 0.583–0.611, respectively). This study comprehensively explored factors associated with hyperglycemic complications arising from remdesivir administration and, concurrently, leveraged a range of machine learning methodologies to construct a risk scoring model, thereby facilitating the tailoring of individualized remdesivir treatment regimens for patients with COVID‐19.

Study Highlights.

• WHAT IS THE CURRENT KNOWLEDGE ON THE TOPIC?

Remdesivir, the first drug approved by the US FDA as a treatment for COVID‐19, underwent fast track and priority review processes. Clinical trial results have reported grade 3/4 hyperglycemia in more than 10% of cases.

• WHAT QUESTION DID THIS STUDY ADDRESS?

Using real‐world data from electronic health records, we sought to identify risk factors of hyperglycemia and determine those most significant for patient care.

• WHAT DOES THIS STUDY ADD TO OUR KNOWLEDGE?

This study identified risk factors of hyperglycemia in remdesivir users using a machine learning approach. Body mass index ≥23 kg/m², proton pump inhibitor use, cholinergic medication use, and cardiovascular disease were associated with hyperglycemic adverse events in hospitalized COVID‐19 patients receiving remdesivir.

• HOW MIGHT THIS CHANGE CLINICAL PHARMACOLOGY OR TRANSLATIONAL SCIENCE?

Identification of risk factors of hyperglycemia can assist healthcare providers in managing patients and the results of this study can be used to design individually tailored remdesivir treatments for COVID‐19 patients in order to achieve optimal outcome.

INTRODUCTION

Remdesivir, a nucleotide prodrug of an adenosine analog, received Food and Drug Administration (FDA) approval on October 22, 2020, as the inaugural medication for coronavirus disease 2019 (COVID‐19) in the United States. This decision was based on the outcomes of the NIAID‐ACTT‐1 trial, a randomized controlled study involving 1063 hospitalized patients with COVID‐19. The results of this trial indicated a notable reduction in hospitalization or treatment duration and a significant improvement in patients' recovery time. ¹ Furthermore, remdesivir's efficacy against various variants, including sublineages BQ.1.1 and XBB, has sustained its usage for over two‐and‐a‐half years. ² , ³ , ⁴ Additionally, a meta‐analysis revealed an exceptionally low pooled COVID‐19 mortality rate associated with its use, and another study demonstrated its potential to reduce mortality among COVID‐19 patients requiring supplemental oxygen therapy. ⁵ , ⁶

Despite its clinical benefits, the administration of remdesivir has been associated with the development of hyperglycemia, a condition characterized by elevated blood glucose levels both during and after therapy. Previous studies have established an association between remdesivir and elevated blood glucose levels in COVID‐19 patients. ⁷ , ⁸ It is noteworthy that hyperglycemia has been positively correlated with unfavorable outcomes in COVID‐19 patients, ⁹ although the exact pathophysiological mechanisms responsible for this effect remain unclear, whether it is attributable to the disease itself or the medication administered. Therefore, the assessment of risk factors contributing to elevated blood glucose levels in individuals undergoing remdesivir treatment is of paramount importance. Regrettably, the available body of evidence regarding these factors remains insufficient. Moreover, the study seeks to go beyond mere characterization by harnessing machine learning techniques to develop a prediction score.

The aim of this study was to identify the clinical characteristics that contribute to hyperglycemia in COVID‐19 patients treated with remdesivir and, concurrently, to develop a predictive score through the training of machine learning models. With the purpose of providing clinicians with a valuable and easily applicable tool in clinical practice, this predictive model was designed to aid in the early identification of patients at risk of developing hyperglycemia while undergoing remdesivir therapy. In this context, this study aids healthcare professionals in better understanding the complexities of remdesivir treatment and provides a proactive approach to reduce associated risks.

METHODS

Study patients and data collection

This observational cohort study adhered to ethical research principles, considering the unique circumstances of COVID‐19 patients receiving remdesivir treatment. Notably, this study did not assign different weights to the imbalanced group sizes, primarily due to the standard practice nature in the study's context. The treatment under investigation represented the standard of care for COVID‐19 patients during the study period, widely accepted and uniformly applied to all eligible patients. In cases where the treatment aligns with the standard practice, it is customary in the field not to apply differential weighting to imbalanced group sizes. This approach is in line with the notion that treatment assignment remains uninfluenced by the research design, rendering equal weighting a reasonable choice for evaluating treatment outcomes within an observational cohort study framework.

During the study period from January 2021 to June 2022, all hospitalized COVID‐19 patients receiving remdesivir treatment at Bestian Hospital (Cheongju‐si, Korea) were included. Exclusions comprised patients under 18 years of age, those with incomplete data, and individuals diagnosed with diabetes and prescribed diabetes medications, aiming to eliminate their influence on hyperglycemia outcomes.

Baseline data encompassing patient characteristics were collected on their initial hospitalization date. Information included sex, age, height, weight, duration of hospitalization, smoking and alcohol history, COVID‐19 severity assessed via multi‐detector computed tomography (MDCT) chest X‐ray, comorbidities, and concurrent medications. Hyperglycemic adverse events (AEs) were categorized per the Common Terminology Criteria for Adverse Events (CTCAE), version 5.0, ¹⁰ with Grade 1 hyperglycemia indicating abnormal glucose levels exceeding the baseline (above 200 mg/dL in this study) without medical intervention. Grade 2 signified a change in daily management from baseline for diabetes patients, including the initiation of oral antiglycemic agents. Grade 3 indicated the initiation of insulin therapy with hospitalization required, while Grade 4 denoted life‐threatening consequences necessitating urgent intervention, and Grade 5 was defined as a fatal outcome.

This study received approval from the Institutional Review Board of Bestian Hospital, a dedicated COVID‐19 center in Korea, under approval number 2022‐05‐001‐001. All procedures involving human participants adhered to the principles outlined in the Declaration of Helsinki.

Statistical analysis and machine learning methods

Categorical variables were compared between patients with hyperglycemic complications and those without complications (control group) using either the Chi‐square test or Fisher's exact test. To identify independent risk factors for hyperglycemic outcomes, a multivariable logistic regression analysis was conducted. The variables with a p‐value <0.05 in the univariate analysis, together with sex as a clinically relevant confounder, were included in the multivariable analysis. The odds ratios were obtained from the univariate analysis, while the adjusted odds ratios were derived from the multivariate analysis. The goodness of fit of the model was assessed using the Hosmer–Lemeshow test. A p‐value <0.05 was considered statistically significant. Univariate analysis was performed using IBM SPSS Statistics, version 20 software (IBM Corp., Armonk, NY, USA).

A random forest‐based classification approach was employed to assess the significance of various variables associated with hyperglycemic AEs. A total of 66 variables were included in the machine learning model as clinically relevant predictors. To predict hyperglycemic complications, the following machine learning methods were utilized: multivariate logistic regression, elastic net, random forest, and support vector machines (SVM). All these methods were implemented using the caret R package. The capability to predict hyperglycemia in each machine learning model was assessed by calculating the area under the receiver operating curve (AUROC), area under the precision recall curve (AUPRC), and its 95% confidence interval (CI). Machine learning analyses were conducted using R software version 3.6.0 (R Foundation for Statistical Computing, Vienna, Austria).

The performance of the machine learning models was evaluated through internal validation. The dataset was randomly divided during the prediction process for model development and evaluation. Using one data sample split into five subsets, one subset was selected for model validation, while the other subsets were used to train machine learning models. To assess the prediction power of the machine learning models, fivefold cross‐validation was repeated five times. A risk‐scoring system, based on multivariable and machine learning analyses, was developed. In the logistic regression model, each coefficient was divided by its smallest value and rounded to generate the risk scores. The risk scoring system, developed from multivariate logistic regression, predicts the probability of hyperglycemic AEs based on multiple risk factors. Coefficients derived from the regression quantify each risk factor's impact on the outcome. By assigning points to these coefficients and calculating a cumulative score for individuals, the system categorizes risk levels, aiding healthcare decisions and tailored interventions.

RESULTS

Baseline characteristics of remdeisivir cohort

Among the patients enrolled in this study (n = 1262), 331 patients were excluded due to age less than 18 years (n = 51) or comorbidities of diabetes or prescription of diabetes medications (n = 280). Consequently, data on 931 patients who were receiving remdesivir were used for the analysis. This group included 542 (58.2%) males, and 403 (43.3%) patients were aged 65 years or older. The average treatment duration for these patients was 10 days (standard deviation [SD] = 6.19, range 1–86). Additionally, the mean vial usage for a single dose of 100 mg/vial was 7.4 (SD = 2.4, range 2–11). Among these patients, 59 (6.3%) experienced hyperglycemic AEs after receiving remdesivir. The outcomes of patients who received remdesivir included 75 deaths, 46 severe outcomes, and 809 remissions.

Hyperglycemic complications were more common in patients aged 65 years and older compared to their younger counterparts (p = 0.043). Patients with a body mass index (BMI) of 23 kg/m² or higher were associated with more AEs than those with BMI < 23 kg/m² (p = 0.009). Significantly higher rates of hyperglycemic complications were observed in patients taking specific medications compared to those not receiving the medications mentioned below: antiarrhythmics (p = 0.006), anticoagulants (p = 0.021), calcium channel blockers (p = 0.021), cholinergic medications (p < 0.001), gout medications (p = 0.004), isosorbides (p = 0.002), and proton pump inhibitors (PPIs) (p < 0.001). Patients with cardiovascular diseases also had more hyperglycemic complications than those who did not have the comorbidity (p = 0.002) (Table 1; refer to Table S2 for a full list of concurrent medications and comorbidities, and Table S3 for a list of comorbidities).

TABLE 1.

Baseline characteristics of COVID‐19 patients receiving remdesivir.

Characteristic	Hyperglycemia (n = 59)	Control (n = 872)	P‐value
Sex			0.713
Female	26 (44.1)	363 (41.6)
Male	33 (55.9)	509 (58.4)
Age (years)			0.043*
≥65	33 (55.9)	370 (42.4)
<65	26 (44.1)	502 (57.6)
Body mass index (kg/m2)			0.009*
≥23	46 (78)	526 (60.9)
<23	13 (22)	338 (39.1)
Duration of hospitalization (days)			0.325
≥10	30 (50.8)	386 (44.3)
<10	29 (49.2)	486 (55.7)
Smoking history			0.100
Yes	7 (11.9)	55 (6.3)
No	52 (88.1)	814 (93.7)
Alcohol history			0.277
Yes	7 (11.9)	151 (17.4)
No	52 (88.1)	719 (82.6)
MDCT X‐ray			0.484
Severe	24 (40.7)	311 (35.7)
Mild	35 (59.3)	561 (64.3)
Concurrent medication
Alzheimer's disease medications			0.928
Yes	5 (8.5)	71 (8.1)
No	54 (91.5)	801 (91.9)
Anticoagulants			0.021*
Yes	4 (6.8)	18 (2.1)
No	55 (93.2)	854 (97.9)
Antiplatelets			0.119
Yes	10 (16.9)	91 (10.4)
No	49 (83.1)	781 (89.6)
Antipsychotics			0.942
Yes	4 (6.8)	57 (6.5)
No	55 (93.2)	815 (93.5)
ARBs			0.115
Yes	15 (25.4)	151 (17.3)
No	44 (74.6)	721 (82.7)
Benzodiazepines			0.164
Yes	7 (11.9)	61 (7)
No	52 (88.1)	811 (93)
Calcium channel blockers			0.021*
Yes	17 (28.8)	148 (17)
No	42 (71.2)	724 (83)
Cholinergic medications			<0.001*
Yes	10 (16.9)	42 (4.8)
No	49 (83.1)	830 (95.2)
Diuretics			0.130
Yes	3 (5.1)	18 (2.1)
No	56 (94.9)	854 (97.9)
Ezetimibe			0.723
Yes	3 (5.1)	36 (4.1)
No	56 (94.9)	836 (95.9)
Gastirc			0.119
Yes	2 (3.4)	82 (9.4)
No	57 (96.6)	790 (90.6)
Glucocorticosteroids			0.368
Yes	3 (5.1)	26 (3)
No	56 (94.9)	846 (97)
Gout medications			0.004*
Yes	3 (5.1)	8 (0.9)
No	56 (94.9)	864 (99.1)
HMG‐CoA reductases			0.838
Yes	10 (16.9)	139 (15.9)
No	49 (83.1)	733 (84.1)
Isosorbides			0.002*
Yes	3 (5.1)	7 (0.8)
No	56 (94.9)	865 (99.2)
NSAIDS			0.758
Yes	3 (5.1)	37 (4.2)
No	56 (94.9)	835 (95.8)
PPIs			<0.001*
Yes	10 (16.9)	46 (5.3)
No	49 (83.1)	826 (94.7)
SSRIs			0.386
Yes	4 (6.8)	38 (4.4)
No	55 (93.2)	834 (95.6)
Thiazides			0.826
Yes	3 (5.1)	39 (4.5)
No	56 (94.9)	833 (95.5)
α Blockers			0.318
Yes	5 (8.5)	47 (5.4)
No	54 (91.5)	825 (94.6)
β‐Blockers			4.574
Yes	10 (16.9)	40 (4.6)
No	49 (83.1)	832 (95.4)
Comorbidities
Cancer			0.663
Yes	7 (11.9)	88 (10.1)
No	52 (88.1)	784 (89.9)
Cardiovascular disease			0.002*
Yes	35 (59.3)	337 (38.6)
No	24 (40.7)	535 (61.4)
Neuropsychiatric disorders			0.787
Yes	9 (15.3)	122 (14)
No	50 (84.7)	750 (86)
Respiratory disease			0.223
Yes	2 (3.4)	67 (7.7)
No	57 (96.6)	805 (92.3)
Urinary disorders			0.386
Yes	4 (6.8)	38 (4.4)
No	55 (93.2)	834 (95.6)
Others			0.589
Yes	5 (8.5)	58 (6.7)
No	54 (91.5)	814 (93.3)

Note: Statistical significance (P‐value < 0.05) is denoted by an asterisk (*).

Abbreviations: ARB, angiotensin receptor blocker; HMG‐CoA reductase, 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A reductase; MDCT X‐ray, multi‐detector computed tomography X‐ray; NSAID, non‐steroidal anti‐inflammatory drug; PPI, proton pump inhibitor; SSRI, selective serotonin reuptake inhibitor.

Among COVID‐19 patients without hyperglycemic AEs, 68 of 872 patients (7.80%) unfortunately passed away during their hospitalization, whereas 7 of 59 patients (11.86%) with hyperglycemic AEs experienced the same outcome. Ventilator usage was required for 133 of 872 patients (15.28%) without hyperglycemic AEs, whereas 18 of 59 patients (30.51%) with AEs required mechanical ventilation. In terms of discharge, 763 of 872 patients (87.56%) without hyperglycemic AEs were successfully discharged from the hospital after recovering from COVID‐19, compared to 46 of 59 patients (77.97%) with hyperglycemic AEs who achieved discharge. Additionally, among the COVID‐19 patients without hyperglycemic AEs, 263 of 872 (30.14%) had a baseline severity level greater than 2 according to MDCT X‐ray grading. In contrast, among the COVID‐19 patients who experienced hyperglycemic AEs, 24 of 59 (40.68%) had a baseline severity level greater than 2 based on MDCT X‐ray grading. This comparison underscores the association between baseline severity and the occurrence of hyperglycemic AEs, with a higher proportion of patients in the hyperglycemic AEs group exhibiting a higher baseline severity.

Hyperglycemia risk score using multivariable logistic regression

The multivariable analysis (Table 2) incorporated sex together with factors with p < 0.05 in the univariate analysis (age, BMI, antiarrhythmics, anticoagulant medications, calcium channel blockers, cholinergic medications, gout medications, isosorbides, PPIs, and cardiovascular diseases). Following adjustment for covariates, patients with a BMI ≥23 kg/m² exhibited an approximately 2.41‐fold higher incidence of hyperglycemic AEs compared to patients with a BMI < 23 kg/m². Furthermore, patients receiving PPIs and cholinergic medications had approximately 2.73 and 2.65 times more complications than those not taking these medications. Patients with cardiovascular diseases had about a 1.97‐fold increased risk of hyperglycemia compared to patients without such conditions. The Hosmer–Lemeshow test demonstrated that the fitness of the multivariable analysis model was satisfactory (χ ² = 0.920, 4 degrees of freedom, p = 0.922). The variance inflation factor values were close to 1 (range 1.004–1.066), indicating no collinearity among the independent variables.

TABLE 2.

Multivariable analysis to identify predictors for hyperglycemia‐related adverse events in COVID‐19 patients receiving remdesivir.

Variable	Crude OR (95% CI)	P‐value	Adjusted OR (95% CI)	P‐value	Score
BMI ≥23 kg/m2	2.28 (1.210–4.272)	0.011	2.41 (1.271–4.577)	0.007	1
PPIs	3.67 (1.745–7.697)	0.001	2.73 (1.223–6.081)	0.014	1
Cholinergic medications	4.03 (1.910–8.516)	<0.001	2.65 (1.168–5.989)	0.020	1
Cardiovascular disease	2.32 (1.353–3.961)	0.002	1.97 (1.119–3.472)	0.019	1

Note: Adjusted for sex age, BMI, antiarrhythmics, anticoagulant medications, calcium channel blockers, cholinergic medications, gout medications, isosorbides, PPIs, and cardiovascular diseases.

Abbreviations: BMI, body mass index; CI, confidence interval; OR, odds ratio; PPI, proton pump inhibitor.

To establish a risk‐scoring system, BMI ≥23 kg/m² (1 point), PPIs (1 point), cholinergic medications (1 point), and cardiovascular diseases (1 point) were incorporated into the analysis (Table 2). The scores indicated the approximate risk of hyperglycemic complications as follows: 0, 2.0%; 1, 4.7%; 2, 10.3%; 3, 21.3%; and 4, 38.9%. With a risk score of 0, 2.0% of patients exhibited hyperglycemic complications, while with a risk score of 4, 38.9% displayed an increased risk of hyperglycemia. The logistic regression curve, derived by mapping the scores to risk scores, is presented in Figure 1.

FIGURE 1.

Rate of hyperglycemic complication according to the risk score calculated by logistic regression.

Risk factor validation using machine learning feature selection

This study identified the 20 most important variables through feature selection using the fivefold cross‐validated random forest approach. These variables included the duration of hospitalization, sex, age, disease severity, BMI, smoking history, calcium channel blockers, other medications, statins, alcohol history, angiotensin receptor blockers (ARBs), antiplatelets, cardiovascular diseases, gout medications, cancer, neuropsychiatric disorders, beta‐blockers, cholinergic medications, PPIs, and thiazide. The primary reason for employing a machine learning‐based feature selection approach was to address the potential inclusion of type 1 errors (false‐positives) when selecting features that were statistically significant in univariate analysis. Therefore, it became imperative to validate the features from univariate analysis using machine‐learning‐based classification. This study revealed that all features included in the multivariable analysis were incorporated within the top 20 feature selections using a machine learning algorithm (Figure 2). This validation procedure served as a crucial step in confirming the significance and relevance of the identified variables.

FIGURE 2.

Top 20 variables whose feature importance was estimated using the random forest approach to predict hyperglycemia‐related adverse events in COVID‐19 patients receiving remdesivir. BMI, body mass index; HMG‐CoA, 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A.

Table 3 shows the average AUROC and AURPC values after fivefold cross‐validation of logistic regression, elastic net, random forest, and SVM models across five random iterations. The AUROC values for multivariate logistic regression (0.65; 95% CI 0.572–0.719), elastic net (0.66; 95% CI 0.640–0.671), and random forest (0.60; 95% CI 0.583–0.611) indicated acceptable model performance. ¹¹ Detailed information regarding the parameters used for training the models is provided in Table S1.

TABLE 3.

Comparisons of area under the curve for logistic regression, elastic net, random forest, and support vector machine models.

Method	AUROC (95% CI)	AUPRC (95% CI)
Logistic regression	0.65 (0.572–0.719)	0.24 (0.161–0.312)
Elastic net	0.66 (0.640–0.671)	0.23 (0.214–0.244)
Random forest	0.60 (0.583–0.611)	0.20 (0.182–0.210)
SVM (linear)	0.55 (0.524–0.575)	0.18 (0.157–0.197)
SVM (radial)	0.54 (0.515–0.558)	0.14 (0.135–0.152)

Note: Machine learning using variables of sex age, body mass index, antiarrhythmics, anticoagulant medications, calcium channel blockers, cholinergic medications, gout medications, isosorbides, proton pump inhibitors, and cardiovascular diseases.

Abbreviations: AUPRC, area under the precision recall curve; AUROC, area under the receiver operating curve; CI, confidence interval; SVM, support vector machine.

DISCUSSION

This study reveals that 6.3% of patients treated with remdesivir experienced hyperglycemic AEs. Its primary finding highlights the association between hyperglycemic AEs in COVID‐19 patients receiving remdesivir and specific factors, including a BMI ≥23 kg/m², the use of PPIs, cholinergic medications, and the presence of cardiovascular diseases. The risk of hyperglycemia was notably higher for these patient groups, with 2.41 times higher risk for those with a BMI ≥23 kg/m², 2.0 times higher for patients with cardiovascular diseases, and 2.73 times higher for those taking PPIs, while patients with cholinergic medications faced a 2.65 times higher risk compared to their counterparts without these conditions. Furthermore, the study employed various machine learning models, including multivariate logistic regression, elastic net, and random forest, all of which demonstrated acceptable AUROC values ranging from 0.60 to 0.66.

Hyperglycemia is one of the complications indicated in the FDA‐labeled drug monograph for remdesivir. A randomized, controlled trial conducted by the National Institute of Allergy and Infectious Disease (NIAID) revealed elevated blood glucose levels in 12% of subjects receiving remdesivir. ¹ Another randomized, open‐label, multicenter clinical trial involving hospitalized COVID‐19 patients treated with remdesivir also documented cases of hyperglycemia. ¹² Given that elevated blood sugar is associated with higher levels of inflammatory mediators, ¹³ it can be speculated that hyperglycemia may lead to severe outcomes in COVID‐19 patients. A study demonstrated that COVID‐19 patients with elevated blood glucose levels faced significantly higher risks of death, admission to the intensive care unit (ICU), intubation, acute kidney injury, and severe sepsis/septic shock compared to patients with normal blood glucose levels. ¹⁴ In this context, hyperglycemia‐related complications emerge as an important safety concern when managing remdesivir therapy.

This study demonstrated that patients with a BMI ≥23 kg/m² faced an increased risk of hyperglycemic AEs after receiving remdesivir therapy. Hyperglycemia is recognized as being more prevalent among overweight or obese individuals, and a high BMI is considered one of the leading causes of type 2 diabetes. ¹⁵ This may be due to altered glucose homeostasis resulting from changes in signal transduction. In the presence of obesity, insulin signaling is often associated with a chronic inflammatory state, and obese patients receiving remdesivir have reduced production of anti‐inflammatory adipocytokines. ¹⁶ As a result, a high BMI can contribute to an inflammatory state, which, in turn, can lead to elevated blood glucose levels. ¹⁷ In this context, the altered activity of insulin signaling may be a contributing factor to increased glucose levels in obese COVID‐19 patients receiving remdesivir.

PPI use was identified as a risk factor for hyperglycemic complications associated with remdesivir in this study. From an efficacy perspective, PPIs have been studied for their potential to enhance the antiviral effectiveness of remdesivir by inhibiting viral replication. ¹⁸ Bojkova et al. revealed a 10‐fold increase in remdesivir efficacy due to omeprazole. ¹⁹ To optimize pharmaceutical care, it is crucial to evaluate the benefits and risks of PPI use, especially in its potential association with elevated blood glucose levels. While hyperglycemia is listed in the drug monograph as an AE of remdesivir, it is possible that the enhanced activity of remdesivir may lead to blood glucose abnormalities in the presence of PPIs. Further studies are necessary to comprehend the mechanism of action of PPIs and remdesivir and to confirm the association between these two medications.

This study revealed an association between cholinergic medications and hyperglycemic AEs in COVID‐19 patients undergoing remdesivir treatment. Cholinergic medications exert their effects on the neurotransmitter acetylcholine, which plays a role in regulating the release of insulin and glucagon through peripheral muscarinic acetylcholine receptors. ²⁰ Given that COVID‐19 patients may experience insulin resistance, ²¹ the concomitant administration of medications affecting these pathways may lead to alterations in glycemic control. Considering the potential of cholinergic medications to interfere with insulin secretion, their use may influence the occurrence of hyperglycemic complications during remdesivir therapy.

Additionally, this study identified cardiovascular disease as a risk factor for hyperglycemic AEs. A meta‐analysis has indicated that COVID‐19 can induce a hyperglycemic state, possibly through mechanisms mediated by the angiotensin‐converting enzyme‐2 (ACE‐2) receptor. ²² Furthermore, remdesivir has been shown to significantly increase liver enzyme activity and blood glucose levels. ⁸ An interconnected mechanism involving ACE‐2 and components of the renin‐angiotensinogen system (RAS) has been reported. ²³ In this context, alterations in the RAS in COVID‐19 patients with cardiovascular diseases may have contributed to the dysregulation of blood glucose levels. It is plausible that hyperglycemic AEs were exacerbated in patients receiving remdesivir with such comorbidities.

The application of machine learning algorithms for predicting adverse drug events in patient care remains a relatively uncharted territory. These models hold great promise for clinical settings, offering the potential to predict and manage complications. This study employed the random forest algorithm, which utilizes bootstrap‐aggregated binary classification trees to combat overfitting. ²⁴ Multiple machine learning models were trained to predict AEs, and the models derived from this study successfully identified risk factors for predicting hyperglycemic complications in COVID‐19 patients treated with remdesivir.

This study does have certain limitations, including its single‐center design and the need for further elucidation of detailed mechanisms. In addition, a limitation of our analysis is the potential omission of baseline blood glucose and glycated hemoglobin (HbA1c) levels as covariates. These variables could influence the development of hyperglycemic AEs, and not accounting for them may introduce bias into our results. Ethical constraints led us to opt for an observational approach, given the recognized efficacy of remdesivir as standard practice, making a placebo‐controlled trial ethically impractical. Although there was no placebo group, this approach harnessed the strengths of observational research to assess real‐world treatment outcomes. To address potential confounding factors, comprehensive statistical adjustments and sensitivity analyses were undertaken. Despite the constraints of a relatively small sample size, this study design provided valuable insights to the scientific and clinical communities in an ethical manner. Additionally, the AUROC values, while not exceptionally high, reflect the intricacies and variability in patient responses within our specific medical context. The risk scores we developed may not yield high AUROC values, but they offer a practical tool for early patient identification and monitoring. This can assist clinicians in prioritizing interventions, ultimately enhancing patient outcomes and safety.

Despite these limitations, to the best of our knowledge this study stands as the first exploration of factors associated with hyperglycemia in hospitalized COVID‐19 patients treated with remdesivir. Moreover, this study provided valuable features and risk scores based on machine learning algorithms, including logistic regression, elastic net, random forest, and SVMs. Effective management of blood glucose levels is paramount for improving the prognosis of critically ill patients. With the risk‐scoring system derived from this study, early identification of patients at risk of hyperglycemia during remdesivir treatment becomes feasible. This, in turn, enables close monitoring of these individuals, forming an integral part of the treatment approach to optimize therapeutic outcomes.

AUTHOR CONTRIBUTIONS

W.K. and G.W.L. wrote the manuscript. N.R., K.H.M., J.H.M., and K.E.L. designed the research. K.H.M., J.H.K., J.Y.G., S.Y.K., and J.M.H. performed the research and analyzed the data.

FUNDING INFORMATION

This study was supported by the Regional Innovation Strategy (2021RIS‐001) and by the Medical Research Center Program (2017R1A5A2015541) of the National Research Foundation funded by the Korean Government Ministry of Education and Ministry of Science and Information and Communication Technology (ICT).

CONFLICT OF INTEREST STATEMENT

All the authors have declared that they have no competing interests.

Supporting information

Table S1

Table S2

Table S3

Kim W, Lee GW, Rhee N, et al. Risk factors for hyperglycemia in COVID‐19 patients treated with remdesivir. Clin Transl Sci. 2024;17:e13684. doi: 10.1111/cts.13684