This is an exploratory, prospective, open-label, parallel, single-center RCT. Eligible participants were recruited from Shanghai Sixth People’s Hospital Affiliated to Shanghai Jiao Tong University School of Medicine at Lingang Campus and were randomly divided into the intervention group (wearing a CGM device) and the control group (routine fingertip measurements). The allocation ratio was 1:1. This trial was registered at the Chinese Clinical Trial Registry (ChiCTR2200059733) before the first recruitment of the participants.

This trial was approved by the Institutional Review Board of Shanghai Sixth People’s Hospital (approval 2022-KY-054). All participants provided written informed consents, and all data collected were only used in this study. All data collected in this study have been anonymized to ensure participant privacy and confidentiality. Personal identifiers have been removed, and unique codes were used to manage data without any direct link to participants’ identities. No compensation was provided to participants in this study. The research was conducted voluntarily, and participants were informed of this during the consent process.

The recruitment process involved identifying potential participants from hospital ICU registries. Hospital databases were systematically reviewed to locate patients meeting the initial eligibility criteria, including age, confirmed COVID-19 diagnosis, and indicators of critical illness. Once potential participants were identified, their medical records were thoroughly examined to verify clinical details, including diagnostic results from real-time, reverse transcription polymerase chain reaction (RT-PCR) assays and assessments of respiratory function, oxygen saturation levels, and pulmonary imaging findings. Following this initial review, clinical evaluations were conducted to confirm the presence of the inclusion criteria. In parallel, exclusion criteria were assessed by reviewing the patients’ mental health status and cognitive function to identify any clinically significant psychosis or cognitive impairment that might interfere with their ability to participate in or comply with the study protocol.

Eligible participants were frail older adults aged 65 years or older who were diagnosed with COVID-19 and critical illness. COVID-19 cases were confirmed by real-time RT-PCR assay of nasal and pharyngeal swab specimens. Critical illness was defined as patients with one of the following conditions: shortness of breath, respiratory rate ≥30 times/min; oxygen saturation ≤93% in a resting state; arterial oxygen partial pressure/oxygen inhalation concentration ≤300 mm Hg; pulmonary imaging showing that the lesions have significantly progressed >50% within 24-28 hours; any occurrence of respiratory failure and requiring mechanical ventilation; or shock, combined with other organ failure. Patients were excluded if they had any clinically significant psychosis or cognitive impairment; they were also excluded if they were unlikely to comply with or complete the study.

The patients in the intervention group were instructed on the use of an intermittently scanned CGM (isCGM) device (Freestyle libre; Abbott Diabetes Care). The sensor was implanted in a site with sufficient subcutaneous fat on the back of the upper arm, and this site can remain flat (without bends or wrinkles) during the participants’ activities. Scars, moles, obesity lines, or lumps were avoided. One sensor lasted no longer than 14 days. If the sensor was expired or malfunctioned, a new one would be immediately implanted until discharge or death. Data were extracted and processed by using a scanner (by swipes) and auxiliary software. The frequency of the swipes followed the routine care during ICU stays at 4 hours intervals or whenever necessary. Data extracted from the CGM device included the mean glucose levels, coefficient of variance (CV), and times of severe hypoglycemia and hypoglycemia (<2.8 and <3.9 mmol/L, respectively) every day. The patients in the control group received a routine fingertip measurement (point-of-care testing [POCT]) for glucose levels. The frequency was consistent with the CGM swipes, ranging from every 1 hour to 4 hours when necessary. If a patient encountered hypoglycemia, more frequent testing of swipes and POCT were done. The mean glucose levels and CV were calculated based on the fingertip glucose profiles.

According to a previous case-control study [11], the composite adverse COVID-19 outcome (defined as progression to critical illness or death) occurred in 46% of patients in the isCGM group compared with 66% in the POCT group. Therefore, this study can achieve 80% statistical power at the level of α=.05 by including 50 patients in each group to achieve the superiority end point.

The participants were randomly assigned to the intervention and control groups in a 1:1 ratio at the time of admission to the ICU. The randomization sequence was generated by the staff. No blinding was applied during the study period. Anthropometric and clinical data were collected from the patients’ self-reporting or the electronic medical records at baseline and at the end of their ICU stay or death after the intervention. These data included sex, height, weight, laboratory measurements, and history of diseases and medications. BMI was calculated as the weight in kilograms divided by the squared height in meters. The treatments of COVID-19 among critically ill patients were prescribed according to the diagnosis and treatment guideline for COVID-19 (ninth version) in China. For all patients received intravenous glucocorticoids, insulin was initiated when the glucose level was persistently higher than 13.9 mmol/L. The average daily dosages of glucocorticoids (converted to the dosages of prednisone) and insulin were also collected.

The primary outcome was 28-day mortality in the ICU. The secondary outcome included the length of stay in the ICU, as well as the occurrence of hypoglycemia and severe hypoglycemia events (<3.9 and <2.8 mmol/L, respectively).

The analysis was performed according to the intention-to-treat principle. The data are reported as the mean (SD) unless otherwise stated. One‐way ANOVA was used to analyze differences in population characteristics from baseline to the end of study. A multivariate generalized estimating equation model adjusted for sex and age was used to evaluate influential factors on in-hospital mortality. The Kaplan-Meier survival analysis was used to compare the survival rate between patients in the isCGM and POCT groups. All data were analyzed using R (version 4.0.3; R Foundation for Statistical Computing). A 2‐tailed P<.05 was defined as significant.

Of the 130 participants who underwent screening in this study, 127 (97.6%) were enrolled and randomly assigned. In all, 64 were in the isCGM group while 63 were in the POCT group. Of these 127 participants, a total of 124 (97.6%) completed the study. A total of 2 participants from the isCGM group and 1 participant from the POCT group discontinued the study because they lost the sensor during the study period (Figure 1).

The baseline characteristics are shown in Table 1. The mean age was 78.3 (SD 11.5) years. The mean fasting glucose level at baseline was 8.12 (SD 1.54) mmol/L. The mean hemoglobin A_1c level was 7.2% (SD 0.8%). The percentage of participants with diabetes was 30.6% (38/124). The baseline characteristics were well balanced and comparable at baseline.

There was no difference in fasting glucose levels observed between the isCGM and the POCT groups at baseline (mean 8.10 SD 1.35 vs 8.13, SD 1.67; P=.43). When compared with the baseline value of fasting glucose, both groups showed significant decreases at the end of the study period (isCGM: mean 8.10, SD 1.35 vs 6.07, SD 0.63; P=.001; and POCT: mean 8.13, SD 1.67 vs 7.76, SD 0.62; P=.04). At the end of study period, the mean value of fasting glucose in the isCGM group and the POCT group was 6.07 (SD 0.63) mmol/L and 7.76 (SD 0.62) mmol/L, respectively, showing a significant intergroup difference (Table 2; P=.01). The daily mean glucose levels (6.93, SD 0.72 vs 8.43, SD 0.52 mmol/L; P=.02) and the CV (25.2%, SD 1.23% vs 33.7%, SD 1.15%; P=.01) were also found to be significantly lower in the isCGM group than those in the POCT group at the end of the trial. Multivariate generalized linear model defined the reduction of fasting glucose levels as a dependent variable and the control group as the reference. After adjustment of age, sex, and BMI, the results showed that at the end of the study period, the intervention (isCGM) was associated with a larger reduction of fasting glucose levels (–1.69, 95% CI –1.65 to –1.73; P<.001).

A total of 15 death events occurred during the study period, with 2 in the isCGM group and 13 in the POCT group. The corresponding hazard ratio of mortality in the ICU in the isCGM group when compared with the POCT group was 0.18 (95% CI 0.04-0.79; Figure 2). The average length of ICU stay was 10.0 (SD 7.57) days in the isCGM group and 14.0 (SD 6.86) days in the POCT group, with a significant P value for comparison (P=.02).

A total of 207 hypoglycemia events (<3.9 mmol/L) was detected, with 43 in the isCGM group (0.0006 times per patient per day) and 164 in the POCT group (0.002 times per patient per day; P<.001). A total of 81 severe hypoglycemia events (<2.8 mmol/L) was detected, with 16 in the isCGM group (0.0002 times per patient per day) and 65 in the POCT group (0.0008 times per patient per day; P<.001).

The major adverse event in this study was bleeding in the puncture site, with a total of 6 occurrences in the isCGM group. During the follow-up, none of the participants dropped out because of bleeding in the puncture site.