Effects of Continuous Versus Intermittent Glucose Monitoring in Intensive Care Unit Patients: A Systematic Review With Meta‐Analysis

ABSTRACT

Introduction

Glucose management in intensive care unit (ICU) patients often relies on point of care (POC) blood glucose measurements. An increasing number of randomized clinical trials (RCTs) have investigated continuous glucose monitoring (CGM) compared to POC, but effects on patient‐important outcomes are uncertain.

Methods

We systematically searched PubMed, Embase, CENTRAL, CINAHL and Web of Science. All reporting was done according to the PRISMA guideline. We included RCTs in ICU patients comparing the effects of CGM versus POC on glycemic and clinical outcomes. We performed meta‐analyses, Trial Sequential Analysis, and assessed the certainty of the evidence using GRADE.

Results

We identified 1271 records and included 18 RCTs comparing CGM versus POC with a total of 2027 participants; 15 trials with 1600 participants reported on mortality (relative risk 0.61, 95% CI 0.35–1.04; very low certainty evidence) and 14 trials with 1515 participants on hypoglycemia (relative risk 0.44, 95% CI 0.23 to 0.82; very low certainty evidence). None of our remaining secondary outcomes were reported in the trials. We identified potential benefits of CGM versus POC on glycemic process outcomes; however, we did not evaluate certainty of evidence.

Conclusions

CGM used for glucose management in ICU patients may reduce mortality and hypoglycemia, but the evidence is very uncertain. Sufficiently powered trials at low risk of bias are needed to confirm potential beneficial effects.

Editorial Comment

This meta‐analysis reports the available literature on continuous versus intermittent monitoring of blood glucose in critically ill patients. The authors found that the available evidence was very uncertain, although continuous monitoring might improve outcomes. It is likely that this is partially due to lack of standardization of measurements and different management strategies. It is possible that the question is best answered with a larger trial with a well‐defined treatment protocol.

1. Introduction

Glucose management during critical illness is inherently challenging due to metabolic stress, decompensated physiological mechanisms, and concomitant treatments [1]. Dysglycemia is associated with worsened clinical outcomes and thus guidelines recommend monitoring and stabilizing blood glucose to prevent complications [2, 3, 4]. Current standards in glucose monitoring rely on point of care (POC) measurements, which may be inadequate in providing optimal patient care as they leave patients unmonitored for longer time intervals.

Continuous glucose monitoring (CGM) could potentially resolve this problem by providing glucose measurements automatically every 1–5 min [5, 6, 7]. Since implementation, CGM has improved outpatient diabetes care, especially for patients with type 1 diabetes [8, 9]. Several studies have been conducted in hospital and intensive care unit (ICU) settings [10, 11], however, the results have been inconsistent. A meta‐analysis by Yao et al. from 2022 concluded that use of subcutaneous CGM in adult ICU patients may reduce incidences of hypoglycemia and mortality [12]. In this rapidly evolving field of research, new evidence is continuously emerging, and an updated synthesis is warranted. The aim of this systematic review with meta‐analysis and Trial Sequential Analysis (TSA) was to assess the effects of CGM versus POC glucose monitoring on pre‐defined clinical and glycemic outcomes in ICU patients.

2. Materials and Methods

This systematic review was reported in accordance with the Preferred Reporting Items for Systematic Review and Meta‐Analysis Protocols (PRISMA‐P) guidelines [13]. A protocol was drafted prior to initiation, registered in the international prospective register of systematic reviews (PROSPERO) (CRD420251048568), and published in a peer‐reviewed journal [14]. There were few deviations between the protocol and the conduct of the systematic review and meta‐analysis (Appendix A).

2.1. Eligibility Criteria

We screened all randomized clinical trials (RCTs) that allocated ICU patients to receive CGM versus POC glucose monitoring for glucose management. We applied no restriction on patient population, setting, type of CGM or type of POC device, and we allowed for any duration of trial and intervention, year of publication, and language. We also accepted any co‐interventions including any method of delivering and titrating insulin, if the co‐interventions were intended to be delivered in the same way to the allocation groups, or if the co‐interventions were intended to support the intervention to obtain the same target in both groups. We excluded quasi RCTs (defined as trials where treatment allocation was not truly random) and trials using automated insulin delivery (AID) systems.

2.2. Information Sources

We systematically searched PubMed (MEDLINE), Embase (OVID), Cochrane central register of controlled trials (CENTRAL), Cumulative Index to Nursing and Allied Health Literature (CINAHL), Web of Science Core Collection (CLARIVATE) from the 28th of March 2025 to 13th of August 2025. The searches were conducted with assistance from an information specialist. In addition, we screened reference lists of relevant literature and searched international trial registries defined in the protocol for unpublished results. The searches are presented in Appendix B.

2.3. Record Screening and Selection

We imported the records obtained from our searches into Covidence Systematic Review Tool (Veritas Health Innovation, Melbourne, Australia) [15] where titles, abstracts, and relevant records were screened by two independent reviewers (CLG, MGD). Any conflicts were resolved by discussion and consensus with consultation from a senior author (MHM).

2.4. Data Extraction

Data extraction was carried out independently by two reviewers (CLG, MGD). We extracted pre‐specified data points in relation to trial design, setting, population, intervention, outcomes, and additional information included funding sources and conflicts of interest. Disagreements were resolved by discussion and consensus.

2.5. Risk of Bias Assessment

Risk of bias was assessed independently by two authors (CLG, MGD) using the Cochrane tool for assessing risk of bias in randomized trials 2 (RoB 2 tool) [16]. Risk of bias was assessed for each of the domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported results. Each domain was classified as having low risk of bias, some concerns, or high risk of bias using the algorithm provided in the RoB 2 tool. Based on the risk of bias in each domain, an overall risk of bias on the outcome level was assigned for each trial in the following way: If all domains were assessed as low risk of bias, then the overall risk of bias was adjudicated as low risk of bias. If one domain was rated as some concerns, the overall risk of bias was adjudicated as some concerns. If one or more domains had high risk of bias or two or more domains had some concerns, the overall risk of bias was adjudicated as high risk of bias.

2.6. Data Synthesis

2.6.1. Outcomes and Variables

The primary outcome was:

• Mortality at the longest follow‐up.

The main secondary outcome was:

• Numbers of patients with hypoglycemia (as defined in the trials). In case of trials reporting more than one level of hypoglycemia, only the lowest level was used in the meta‐analysis.

We included five other secondary outcomes as defined in a core outcome set for general ICU patients [17].

• Days alive without life support.

• Days alive and out of hospital.

• Days alive without delirium.

• Health‐related quality of life.

• Cognitive function

In addition, we investigated four glycemic process outcomes:

• Time in range as defined in the included trials

• Time below range as defined in the included trials

• Time above range as defined in the included trials.

• Glycemic variability defined as the coefficient of variation, i.e., standard deviation of the glucose level distribution divided by mean glucose level.

We primarily used data from the intention‐to‐treat populations of the included trials and when this was not available, we used data from the per‐protocol populations.

2.6.2. Effect Measures

Dichotomous outcome variables were reported as risk ratios (RR) and for continuous outcome variables, mean differences were used. For both, we reported 95% confidence intervals (CI).

2.6.3. Heterogeneity Assessment

Heterogeneity was assessed by visual inspection of forest plots. This was supported by calculating I² statistics and τ ² statistics.

We used the Clinical Diversity In Meta‐analyses (CDIM) tool to assess clinical diversity among the included trials [18].

2.6.4. Meta‐Analysis

The meta‐analysis was conducted in accordance with the Cochrane Handbook [19]. We applied both a random effects model (Knapp‐Hartung‐Sidik‐Jonkmann estimator) and a fixed effects model (Mantel–Haenszel estimator). We reported the estimates of the most conservative model as planned [14]. In case the two models yielded comparable results, we reported the results of the best fitting model considering the included trials and their heterogeneity. The meta‐analysis was done in Rstudio V4.3.0 using the “meta” package [20]. The reported summary statistics were converted from median to mean when necessary.

2.6.5. Trial Sequential Analysis

To accommodate for type 1 (false positive results) and 2 errors (false negative results) we conducted a TSA for the primary and main secondary outcome—both of which were dichotomous [21, 22]. For both outcomes, alpha was set at 0.05, beta at 0.10, and the relevant relative risk reduction at 15%. From this and based on the accrued sample size, the required information size and boundaries for benefit and harm were constructed. We used diversity as suggested by the meta‐analysis.

2.6.6. Subgroup Analysis

To investigate the effect of potential effect modifiers, the following subgroup analyses were planned and conducted for the primary and main secondary outcome:

• High risk of bias or some concerns versus low risk of bias trials

• Trials assessing CGM with a co‐intervention (in the intervention group) versus those without

• Trials in an adult (≥18 years old) versus neonatal or pediatric population.

• Trials with ≥50% of patients with pre‐admission diabetes versus < 50%.

• Trials with target glucose 7.8–10 mmol/L or above versus lower.

The anticipated directionalities of the subgroup analyses are specified in the protocol [14].

Based on the results of the CDIM scoring, the following subgroups and meta‐ression were supplemented post hoc:

• Country: Chinese trials vs. non‐Chinese trials.

• Timing of event (mortality only): mortality at 28‐days or more versus in ICU or in‐hospital mortality

• Study year: We performed a meta‐regression analysis modelling the effect of study year on hypoglycemia and mortality.

The credibility of the subgroup analysis was assessed using the Instrument for assessing the Credibility of Effect Modification Analyses (ICEMAN) tool [23]. Credibility was assessed for each subgroup analysis and was ranked from very low to high based on set criteria relating to methodology, numbers, and outcomes.

2.6.7. Reporting Bias Assessment

We used funnel plots to assess reporting bias for the primary and main secondary outcomes. The funnel plots were visually inspected for asymmetry [24]. For statistical asymmetry testing for dichotomous outcomes, we planned to use the Harbord test if τ ² was equal to or below 0.1 and the Thompson test if τ ² was above 0.1 [25, 26]. For continuous outcomes, we planned to use the Egger test [27].

2.6.8. Sensitivity Analysis

To investigate the potential influence of missing data on our primary and main secondary outcome we performed a best/worst‐case scenario and a worst/best‐case scenario. In the best/worst‐case scenario, we assumed that all trial participants lost to follow‐up in the intervention group had the best possible outcome and that all participants in the control group had the worst possible outcome. The opposite was assumed in the worst/best‐case scenario.

2.7. Certainty of Evidence Assessment

We used the Grading of Recommendation Assessment, Development, and Evaluation (GRADE) to evaluate the certainty of the body of evidence [28]. The certainty of evidence was graded as either very low, low, moderate, or high based on an overall assessment of risk of bias, inconsistency, indirectness, imprecision, and publication bias. The results were displayed in a “summary of findings” table.

3. Results

We screened a total of 1271 records (Figure 1), among which 5 were obtained from citation searches and the remaining from structured searches from electronic databases (Appendix B). Subsequently, 59 records were retrieved for full‐text screening, of which 18 RCTs with a total of 2027 participants were included in this review. Pre‐defined trial registries were also screened, and authors of ongoing trials were contacted, but no additional trials were included.

FIGURE 1.

PRISMA flowchart showing in and exclusion process.

3.1. Trial Characteristics

Characteristics of the included trials are listed in Table 1. The trials were conducted from 2009 to 2025, and the number of patients randomized ranged from 24 to 180 (median 120; interquartile range (IQR) 77–144). Most trials were single center (89.0%), conducted in a mixed adult ICU population (both medical and surgical) (50.0%), and used subcutaneous CGM (94.0%). Two trials were conducted in neonatal ICU patients (11.1%) and one trial used intravenous GCM. Ten trials (55.6%) mentioned the inclusion of patients with diabetes.

TABLE 1.

Trial characteristics.

Authors	Year	Country	Design	ICU type	Patient number randomized	Age, years a	%‐with diabetes a	CGM modality	POC modality	Mortality definition b	Hypoglycemia definition, mmol/L c
Holzinger et al. [7]	2009	Austria	Single center	Medical	124	58/62	19/20	Subcutaneous	Arterial	In‐hospital	< 2.2
LV et al. [29]	2012	China	Single center	Neuro	117	60.8	0	Subcutaneous	Capillary	30‐days	< 3.9
Brunner et al. [30]	2012	Austria	Single center	Medical	174	59/62	19/18.9	Subcutaneous	Arterial	—	—
Qian et al. [31]	2013	China	Single center	Mixed	123	62.1/58.4	0	Subcutaneous	Arterial	28‐days	< 3.9
Kopecký et al. [32]	2013	Czech Republic	Single center	Cardio‐surgical	24	68.1/67.5	16.7/33.3	Subcutaneous	Arterial	—	< 2.9
Boom et al. [33]	2014	The Netherlands	Single center	Mixed	178	66.4/67.2	21/33	Subcutaneous	Arterial	In‐hospital	< 2.2
De Block et al. [34]	2015	Belgium	Multi center	Medical	35	62/68	31.3/15.8	Subcutaneous	Arterial	In‐hospital	< 3.3
Guan et al. [35]	2017	China	Single center	Mixed	130	52.1	—	Subcutaneous	Capillary	Not specified	Not specified
Galderisi et al. [36]	2017	Italy	Single center	Neonatal	50	30 weeks	—	Subcutaneous	Capillary	28‐days	< 2.6
Tian et al. [37]	2018	China	Single center	Mixed	161	60.7/64.6	21/17.5	Subcutaneous	Capillary	—	Not specified
Zhang K [38]	2018	China	Single center	Not specified	116	45.2/44.6	—	Subcutaneous	Capillary	Not specified	Not specified
Lu et al. [6]	2018	China	Single center	Mixed	144	50/49	27/27.1	Subcutaneous	Capillary	28‐days	< 2.2
Preiser et al. [39]	2018	Belgium	Cluster	Mixed	77	62/60	—	Intravenous	Combination	ICU‐mortality	< 2.2
Li et al. [40]	2019	China	Single center	Mixed	72	59.3/61.7	0	Subcutaneous	Capillary	28‐days	< 3.9
Beardsall et al. [41]	2021	UK, Spain and The Netherlands	Multi center	Neonatal	180	27.7/27.4 weeks	(6/8 maternal diabetes)	Subcutaneous	Capillary	Before week 36	< 2.2
Chu et al. [42]	2024	China	Single center	Mixed	110	70/69	72.9/60.4	Subcutaneous	Capillary	90‐days	< 2.2
Shang et al. [43]	2025	China	Single center	Medical	127	78.8/78.7	27/34	Subcutaneous	Capillary	28‐days	< 2.8
Franck et al. [44]	2025	US	Single center	Mixed	85	75/72	74.4/78.6	Subcutaneous	Combination	30‐days	—

^a All or intervention/control.

^b At longest follow‐up.

^c Lowest reported level.

Fourteen trials (77.8%) mentioned the use of a protocol to guide insulin administration, and two trials (11.1%) mentioned the use of an insulin protocol specifically for CGM data as a co‐intervention in the intervention group. Insulin infusion pumps were used in 12 trials (66.7%), and supported insulin delivery in both CGM and POC groups. The glycemic targets and outcome measurements were diverse and are displayed in Appendix C.

3.2. Risk of Bias

One trial had overall low risk of bias, three trials had some concerns, and the remaining had high risk of bias (Figure 2, Appendix L). The most common reasons for high risk of bias were concerns about the randomization process, measurement of the outcome, and incongruency with or absence of a statistical analysis plan.

FIGURE 2.

Risk of bias for mortality.

3.3. Mortality

We included 15 trials in the meta‐analysis of mortality (Figure 3a) with a total of 1600 trial participants and 288 events. We investigated mortality at the longest follow‐up reported, which varied from in‐ICU mortality to 90‐day mortality, and for some trials the follow‐up was not specified (Table 1). We found that CGM versus POC glucose measurements may reduce mortality (RR 0.61 [95% CI; 0.35, 1.04]), but the evidence was very uncertain. There was substantial inter‐trial variability and statistical heterogeneity (τ ² = 0.755, I ² = 68.8%). The required information size was not reached, and due to the small, accrued sample size compared with the required information size, TSA boundaries were ignored (Appendix E). We found moderate clinical diversity (CDIM score 17; Appendix D) which yielded an additional subgroup analysis for the mortality outcome (see methods).

FIGURE 3.

(a) Meta‐analysis of mortality. (b) Meta‐analysis of hypoglycemia.

Subgroup analysis according to country of origin suggested effect modification (test of interaction p < 0.05). Trials originating from China suggested a benefit of CGM on mortality (RR 0.39 [95% CI 0.21 to 0.72]) while trials not originating from China did not suggest a mortality benefit (RR 1.13 [95% CI 0.58–2.22]) (Appendix F). The remaining subgroup and meta‐regression analyses did not indicate effect modification. Using the ICEMAN framework, we judged the credibility of all subgroup analyses as low.

The sensitivity analysis involving best/worst and worst/best case scenarios increased uncertainty of the findings, as the best/worst scenario indicated a mortality benefit with CGM (RR 0.51 [95% CI 0.31–0.84]), and the worst/best scenario did not (RR 0.81 [95% CI; 0.46, 1.45]) (Appendix G).

There was no strong evidence of publication bias. The funnel plot showed slight asymmetry indicating potential small‐study effect and publication bias, but the results of the statistical test were non‐significant (p = 0.07) (Appendix K).

The results of the GRADE assessments are displayed in Table 2. The certainty of evidence for mortality was downgraded to very low due to inconsistency, imprecision, and indirectness.

TABLE 2.

Summary of findings table.

Author(s): Gantzel et al.
Date: 2025‐08‐15
Question: Should continuous glucose monitoring vs point of care glucose measurements be used for glucose management in intensive care unit patients?
Settings: Intensive care unit
Bibliography: See reference list
Quality assessment							No of patients		Effect		Quality	Importance
No of studies	Design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Continuous glucose monitoring	Point of care glucose measurements	Relative (95% CI)	Absolute
Mortality
15	Randomized trials	No serious risk of bias	Serious a	Serious b	Serious c	None	121/794 (15.2%)	167/806 (20.7%)	RR 0.61 (0.35 to 1.04)	81 fewer per 1000 (from 135 fewer to 8 more)	VERY LOW	CRITICAL
Hypoglycemia
14	Randomized trials	No serious risk of bias	Serious d	Serious e	Serious f	None	34/748 (4.5%)	82/767 (10.7%)	RR 0.44 (0.31 to 0.82)	60 fewer per 1000 (from 19 fewer to 74 fewer)	VERY LOW	CRITICAL
Days alive without life support—not reported
0	—	—	—	—	—	None	—	—	—	—
Days alive and out of hospital—not reported
0	—	—	—	—	—	None	—	—	—	—
Days alive without delirium—not reported
0	—	—	—	—	—	None	—	—	—	—
Health‐related quality of life—not reported
0	—	—	—	—	—	None	—	—	—	—
Cognitive function—not reported
0	—	—	—	—	—	None	—	—	—	—

Abbreviation: CI, confidence interval.

^a Substantial heterogeneity and studies without overlapping CIs.

^b Subgroup analysis revealed interaction for region of origin potentially unveiling unknown differences in study population.

^c CI includes no effect and the optimal study size was not reached.

^d Though most estimates pointed in the same direction, there was moderate statistical heterogeneity and p < 0.05.

^e Variations in patient population and outcome definitions justified downgrading one level.

^f Though the observed RRR > 15% and CI of the estimate did not include no effect the optimal information size was not reached and outcomes were rare.

3.4. Hypoglycemia

Fourteen trials were included in the meta‐analysis of hypoglycemia (Figure 3b) with a total of 1515 trial participants and 116 participants with hypoglycemia. Two trials reported on hypoglycemia but were not included in the meta‐analysis (of these, one trial reported 0.8 events per patient in the CGM group versus 3.3 events per patient in the POC group [36], and the other reported the cumulated number of hypoglycemia events, 16 versus 65 events respectively [43]).

We used the lowest reported glucose level for hypoglycemia which ranged from 2.2 to 3.9 mmol/L, and for some trials this was not specified (Table 1).

CGM versus POC may reduce hypoglycemia (RR 0.44 [95% CI; 0.23, 0.82 and TSA‐adjusted CI; 0.05, 4.48]) but the evidence was very uncertain. There was moderate‐to‐high inter‐trial variability and moderate statistical heterogeneity (τ ² = 0.601, I ² = 48.0%). In the TSA, the level of statistical significance was reached, but the boundary for benefit was not breached, nor was the required information size reached (Appendix E). We found moderate clinical diversity (CDIM score 15; Appendix D) which yielded an additional subgroup analysis for hypoglycemia (see methods).

The subgroup analysis comparing adult versus neonatal patients and the subgroup analysis comparing CGM with a co‐intervention versus without co‐intervention included the same trials, and the smallest group only consisted of one trial (Appendix H). The co‐intervention consisted of a CGM‐specific insulin titration algorithm. The subgroup analysis suggested an effect modification (test of interaction p < 0.05). Trials including an adult population and trials without a co‐intervention suggested a benefit on hypoglycemia (RR 0.35 [95% CI 0.19–0.65]) compared to a trial in a neonatal population and a trial with a co‐intervention (RR 2.07 [95% CI 0.79–5.43]). In post hoc meta‐regression analyses, increasing year of publication was associated with smaller treatment effect for hypoglycemia (p = 0.02) (Appendix H), suggesting that year of publication may partly explain the effects of CGM on hypoglycemia. This was not the case for mortality (p = 0.91) (Appendix F). The credibility of all subgroup analyses was low.

The overall result for hypoglycemia was not robust to sensitivity analysis using the worst/best case scenario (RR 0.89 [95% CI; 0.33, 2.36]) (Appendix I).

We found no evidence of potential publication bias as visual inspection of funnel plot and statistical testing revealed no evident asymmetry (p = 0.88) (Appendix K).

The overall certainty of evidence for hypoglycemia was very low (Table 2). It was downgraded due to inconsistency, imprecision, and indirectness.

3.5. Other Secondary Outcomes

None of the remaining secondary outcomes were reported in the included trials. One trial reported the number of patients with delirium and found no difference between the groups [44].

3.6. Glycemic Process Outcomes

Definitions of the glycemic process outcomes are displayed in Appendix C, and the results of the meta‐analysis in Appendix J. The meta‐analysis showed increased time in range by 8%‐points [95% CI; 2–13] and reduced time below range by 1%‐point [95% CI; 0–3] with CGM versus POC, but no difference in time above range (−5 [95% CI; −12–1]). Coefficient of variation was reduced with CGM versus POC by 1%‐point [95% CI; 0–3].

4. Discussion

In this systematic review with meta‐analysis, we investigated the clinical and glycemic outcomes from 18 RCTs comparing CGM versus POC in ICU patients. We found that CGM versus POC may reduce both mortality and hypoglycemia, but the certainty of the evidence was very low for both these outcomes. CGM versus POC may improve additional glycemic process outcomes; however, certainty of evidence was not assessed.

In 2022 Yao et al. published a systematic review comparing subcutaneous CGM with POC glucose monitoring in insulin‐treated adult ICU patients [12]. Our review yielded different results on several outcomes, including mortality where we did not identify a statistically significant benefit of using CGM and the certainty of evidence was very low. Importantly, our review and that by Yao et al. differed in several ways. First, our review included three RCTs that were not published in the 2022 review. At the population level, we included pediatric/neonatal patients, and the proportion of trials conducted in China was lower in our review. Especially the last point may be important in explaining the difference in results, as we identified country of origin as a potential effect modifier. At the intervention level, we excluded trials using AID systems, which is also a crucial difference between the two reviews. In addition to the differences in mortality effects, the RR for hypoglycemia was lower in the review by Yao et al. If hypoglycemia is believed to be a driver of mortality in the ICU population [45], this reduction in the RR for hypoglycemia could potentially explain the observed mortality differences between our review and the review by Yao et al.

We also found improvements in time in range, time below range, and coefficient of variation with the use of CGM vs. POC, which contrasts with Yao et al. A number of trials did not use a blinded CGM in the control group, which could affect outcome measurements between the groups and limit comparability. Finally, the two reviews also differ on the levels of outcome variables, data synthesis, risk of bias assessment, and assessment of the certainty of evidence.

4.1. Clinical Implications

The findings of this review indicate a glycemic and a potential clinical benefit for the use of CGM versus POC in ICU patients. However, the findings lack robustness due to missing outcome data and insufficient required information size. The overall certainty of the evidence was very low for both mortality and hypoglycemia which taken together raises considerable uncertainty about the true effect size of using CGM in ICU patients. Larger, well‐designed RCTs are warranted to inform about the potential benefits of CGM in ICU patients. Only two trials used a CGM specific insulin titration algorithm in the intervention group; however, these were also the only two trials of neonatal patients. Therefore, it was not possible to assess the potential effect of adding a CGM specific insulin algorithm in this review. Most trials in hospitalized patients using CGM alone have been unable to show a positive effect on glucose levels in general—apart from reducing number and duration of hypoglycemic events. In two recent RCTs, one of hospitalized non‐ICU patients with type 2 diabetes, and one of elective post cardiac surgery patients with type 2 and pre‐diabetes—CGM combined with an insulin algorithm specifically designed for CGM data improved overall glucose management and clinical outcomes [5, 46]. Whether this approach could provide benefit in acutely admitted ICU patients warrants further investigation, but it points to the importance of considering how CGM data are handled to provide clinical value. In general, we saw significant variability between the trials which again could be attributed to the use of different glucose management strategies and insulin titration protocols. We encourage trialists to take this into account when designing future trials. Adding to this, factors such as workflow, user satisfaction, clinical setting, and time required to use CGM and adjust blood glucose levels, none of which were assessed in this review, are important considerations when designing an implementation strategy [47, 48].

In recent years, an increasing focus has been put on patient important outcomes, also in the ICU¹⁷. This review set out to investigate the impact of CGM on several patient important outcomes; however, most outcomes were not reported in the included trials. Future trials on clinical interventions, such as CGM, should seek to report patient important outcomes [17].

4.2. Strengths and Limitations

The adherence to a predefined and published protocol along with support from an information specialist are strengths to this review. The use of subgroup analyses to uncover potential effect modifiers and the use of TSA are both strengths but with limitations, as all subgroup analyses had low credibility, and too little information was accrued for TSA on the primary outcome. We converted medians to means when the mean was not reported. This entails a risk of skewing the data which should be considered a limitation in the meta‐analyses of continuous outcomes. This review was not intended to include individual patient data; however, this could have increased validity. All the included trials were relatively small which limits precision. Further, some of the included trials lacked clear outcome definitions which reduced comparability and robustness in our results. Finally, we did not succeed in identifying trials that reported a number of our secondary outcomes which could also be considered a limitation.

5. Conclusions

CGM used for glucose management in ICU patients may reduce mortality and hypoglycemia, but the evidence is very uncertain. Sufficiently powered trials at low risk of bias are needed to confirm potential beneficial effects.

Author Contributions

Record screening and selection, data extraction, and risk of bias assessment were carried out by C.L.G. and M.G.‐D. with support from M.H.M. Statistical analysis including TSA was done by C.L.G. and M.M. C.L.G. wrote the first protocol draft. All authors contributed to the design of this review, editing, and approval of the final manuscript.

Funding

Milda Grigonyte‐Daraskeviciene was funded by a grant from the Novo Nordisk Foundation during this work. Christian Lange Gantzel received a research grant from the Research Foundation of North Zealand Hospital during this work.

Disclosure

The abstract of this research work has been accepted for presentation at Advanced Technologies and Treatments for Diabetes 2026 at the time of submission.

Conflicts of Interest

Ulrik Pedersen‐Bjergaard has served on advisory boards for Novo Nordisk, Sanofi‐Aventis, and Vertex and has received lecture fees from Novo Nordisk and Sanofi‐Aventis. Mikkel Thor Olsen has received speaker fees from Novo Nordisk and serves on the scientific advisory board for decide Clinical Software GmbH (GlucoTab). Kirsten Nørgaard serves as an adviser to Medtronic, Abbott, Convatec, Tandem, Pharmasens and Novo Nordisk; owns shares in Novo Nordisk; has received research grants to the institution from Novo Nordisk, Zealand Pharma, Dexcom and Medtronic; and has received fees for speaking from Medtronic and Novo Nordisk. Johan Mårtensson has received research support from Dexcom Inc. outside the submitted work. Milda Grigonyte‐Daraskeviciene, Anders Perner and Morten Hylander Møller are affiliated with Dept. of Intensive Care, Rigshospitalet, which receives funds for research from the Novo Nordisk Foundation, Sygeforsikring “danmark” and the Independent Research Foundation of Denmark. Morten Heiberg Bestle and Christian Lange Gantzel are affiliated with Dept. of Anesthesia and Intensive Care, Copenhagen University Hospital North—Zealand, which has received funding for other research activities from Sygeforsikring “danmark” and Svend Andersen Foundation. Morten Heiberg Bestle owns shares in Novo Nordisk and has received reimbursement for advisory board work from Biomerieux outside the submitted work.

Appendix A. See Table A1

TABLE A1.

Summary of deviations.

Remark	Reason
Two additional subgroup analyses have been completed (see methods)	We used the CDIM tool to assess clinical diversity according to our review protocol. As assumed, this created the need for additional subgroup analysis as potential effect modifiers were not addressed in our planned subgroup analysis. The one subgroup analysis “Chinese trials vs. non‐Chinese trials” was created to compare trials conducted in developing vs. developed countries as recommended by CDIM. As China was the only developing country in which trials where conducted we renamed this subgroup analysis for transparency reasons.
One additional meta‐regression analysis	We performed a meta‐regression analysis to investigate the effect of study year on hypoglycemia and mortality. This was done in the review process and in agreement with the CDIM tool.
Trial Sequential Analysis for mortality incomplete	Due to a low accrued sample size (< 5%) compared to the required information size, the Trial Sequential Analysis for mortality did not converge. Therefore, we were not able to construct a Trial Sequential Analysis adjusted confidence interval as intended. We were however able to use it for the GRADE assessment on imprecision as the analysis did compute a required information size.
Evaluation of imprecision	In our protocol we planned to use the Trial Sequential Analysis to guide evaluation of imprecision. As Trial Sequential Analysis could not be completed for mortality, we decided to use the GRADE approach instead to evaluate imprecision for both mortality and hypoglycemia.

Appendix B. Search Strategies

PubMed search:

(((("controlled trial*"[Title/Abstract]) OR (allocat*[Title/Abstract])) OR (random*[Title/Abstract])) OR ("Clinical Trial" [Publication Type])) AND ((((("shock"[MeSH Terms] OR "shock"[All Fields] OR "shocked"[All Fields] OR "shocking"[All Fields] OR "shocks"[All Fields] OR ((("sever"[All Fields] OR "severe"[All Fields] OR "severed"[All Fields] OR "severely"[All Fields] OR "severer"[All Fields] OR "severes"[All Fields] OR "severing"[All Fields] OR "severities"[All Fields] OR "severity"[All Fields] OR "severs"[All Fields]) AND ("illness"[All Fields] OR "illness s"[All Fields] OR "illnesses"[All Fields])) OR ("critical illness"[MeSH Terms] OR ("critical"[All Fields] AND "illness"[All Fields]) OR "critical illness"[All Fields] OR ("critically"[All Fields] AND "ill"[All Fields]) OR "critically ill"[All Fields]) OR ("critical illness"[MeSH Terms] OR ("critical"[All Fields] AND "illness"[All Fields]) OR "critical illness"[All Fields]) OR ("critical care"[MeSH Terms] OR ("critical"[All Fields] AND "care"[All Fields]) OR "critical care"[All Fields]) OR ("intensive care units"[MeSH Terms] OR ("intensive"[All Fields] AND "care"[All Fields] AND "units"[All Fields]) OR "intensive care units"[All Fields] OR "icu"[All Fields]) OR ("intensive care units"[MeSH Terms] OR ("intensive"[All Fields] AND "care"[All Fields] AND "units"[All Fields]) OR "intensive care units"[All Fields] OR ("intensive"[All Fields] AND "care"[All Fields] AND "unit"[All Fields]) OR "intensive care unit"[All Fields])))) OR (acute coronary syndrom)) OR (mechanical ventilation)) AND (((("CGM"[All Fields] OR (("continual"[All Fields] OR "continually"[All Fields] OR "continuance"[All Fields] OR "continuation"[All Fields] OR "continuations"[All Fields] OR "continue"[All Fields] OR "continued"[All Fields] OR "continuer"[All Fields] OR "continuers"[All Fields] OR "continues"[All Fields] OR "continuing"[All Fields] OR "continuities"[All Fields] OR "continuity"[All Fields] OR "continuous"[All Fields] OR "continuously"[All Fields]) AND ("glucose"[MeSH Terms] OR "glucose"[All Fields] OR "glucoses"[All Fields] OR "glucose s"[All Fields]) AND ("monitor"[All Fields] OR "monitor s"[All Fields] OR "monitorable"[All Fields] OR "monitored"[All Fields] OR "monitoring"[All Fields] OR "monitoring s"[All Fields] OR "monitorings"[All Fields] OR "monitorization"[All Fields] OR "monitorize"[All Fields] OR "monitorized"[All Fields] OR "monitors"[All Fields])) OR (("continual"[All Fields] OR "continually"[All Fields] OR "continuance"[All Fields] OR "continuation"[All Fields] OR "continuations"[All Fields] OR "continue"[All Fields] OR "continued"[All Fields] OR "continuer"[All Fields] OR "continuers"[All Fields] OR "continues"[All Fields] OR "continuing"[All Fields] OR "continuities"[All Fields] OR "continuity"[All Fields] OR "continuous"[All Fields] OR "continuously"[All Fields]) AND ("glucose"[MeSH Terms] OR "glucose"[All Fields] OR "glucoses"[All Fields] OR "glucose s"[All Fields]) AND ("measurability"[All Fields] OR "measurable"[All Fields] OR "measurably"[All Fields] OR "measure s"[All Fields] OR "measureable"[All Fields] OR "measured"[All Fields] OR "measurement"[All Fields] OR "measurement s"[All Fields] OR "measurements"[All Fields] OR "measurer"[All Fields] OR "measurers"[All Fields] OR "measuring"[All Fields] OR "measurings"[All Fields] OR "measurment"[All Fields] OR "measurments"[All Fields] OR "weights and measures"[MeSH Terms] OR ("weights"[All Fields] AND "measures"[All Fields]) OR "weights and measures"[All Fields] OR "measure"[All Fields] OR "measures"[All Fields])) OR (("continual"[All Fields] OR "continually"[All Fields] OR "continuance"[All Fields] OR "continuation"[All Fields] OR "continuations"[All Fields] OR "continue"[All Fields] OR "continued"[All Fields] OR "continuer"[All Fields] OR "continuers"[All Fields] OR "continues"[All Fields] OR "continuing"[All Fields] OR "continuities"[All Fields] OR "continuity"[All Fields] OR "continuous"[All Fields] OR "continuously"[All Fields]) AND ("glucose"[MeSH Terms] OR "glucose"[All Fields] OR "glucoses"[All Fields] OR "glucose s"[All Fields]) AND ("manage"[All Fields] OR "managed"[All Fields] OR "management s"[All Fields] OR "managements"[All Fields] OR "manager"[All Fields] OR "manager s"[All Fields] OR "managers"[All Fields] OR "manages"[All Fields] OR "managing"[All Fields] OR "managment"[All Fields] OR "organization and administration"[MeSH Terms] OR ("organization"[All Fields] AND "administration"[All Fields]) OR "organization and administration"[All Fields] OR "management"[All Fields] OR "disease management"[MeSH Terms] OR ("disease"[All Fields] AND "management"[All Fields]) OR "disease management"[All Fields])) OR (("continual"[All Fields] OR "continually"[All Fields] OR "continuance"[All Fields] OR "continuation"[All Fields] OR "continuations"[All Fields] OR "continue"[All Fields] OR "continued"[All Fields] OR "continuer"[All Fields] OR "continuers"[All Fields] OR "continues"[All Fields] OR "continuing"[All Fields] OR "continuities"[All Fields] OR "continuity"[All Fields] OR "continuous"[All Fields] OR "continuously"[All Fields]) AND ("glucose"[MeSH Terms] OR "glucose"[All Fields] OR "glucoses"[All Fields] OR "glucose s"[All Fields]) AND ("sensor"[All Fields] OR "sensor s"[All Fields] OR "sensoric"[All Fields] OR "sensorics"[All Fields] OR "sensoring"[All Fields] OR "sensorization"[All Fields] OR "sensorized"[All Fields] OR "sensors"[All Fields]))) OR (FGM or "Flash glucose monitoring" or isCGM OR rtCGM))) OR ("Continuous Glucose Monitoring"[Mesh])))

Embase search:

1.	CGM.mp.
2.	continuous glucose monitor*.mp.
3.	exp continuous glucose monitoring system/
4.	continuous glucose measur*.mp.
5.	continuous glucose management.mp.
6.	continuous glucose sensor*.mp.
7.	CGM.ab,ti.
8.	1 or 2 or 3 or 4 or 5 or 6 or 7
9.	ICU.ab,ti.
10.	Intensive care unit.ab,ti.
11.	critically ill.ab,ti.
12.	critical illness.ab,ti.
13.	critical care.ab,ti.
14.	severe illness.ab,ti.
15.	ICU.mp. or exp intensive care unit/
16.	Critical illness.mp. or exp critical illness/
17.	critically ill.mp. or exp critically ill patient/
18.	severe illness.mp.
19.	cardiogenic shock/or hypovolemic shock/or septic shock/or traumatic shock/or vasodilatory shock/
20.	shock.ti.
21.	9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20
22.	“critical care”.mp.
23.	21 or 22
24.	"Flash glucose monitoring".ab,ti.
25.	8 or 24
26.	23 and 25
27.	limit 26 to (embase or medline or "preprints (unpublished, non‐peer reviewed)")
28.	mechanical ventilation.mp. or exp artificial ventilation/
29.	exp acute coronary syndrome/
30.	acute coronary syndrom*.mp.
31.	23 or 28 or 29 or 30
32.	25 and 31
33.	randomized controlled trial.mp. or exp randomized controlled trial/
34.	clinical trial.mp. or exp clinical trial/
35.	random*.mp.
36.	allocat*.mp.
37.	33 or 34 or 35 or 36
38.	32 and 37

CINAHL search:

(MH "Intensive Care Units+/AM/CL/EC/ED/EI/EV/HI/LJ/LS/MT/OG/PF/SN/ST/TD/UT" OR (MH "Shock+" OR MH "Shock, Septic+" OR MH "Shock, Hemorrhagic" OR MH "Shock, Surgical" OR MH "Shock, Traumatic+" OR MH "Shock, Cardiogenic") OR (MH "Critical Illness/BL/CF/CI/CL/CO/DH/DI/DT/EC/ED/EH/EI/EM/EP/ET/EV/FG/HI/IM/LJ/ME/MI/MO/MT/NU/OG/PA/PC/PF/PP/PR/RA/RF/RH/RT/SS/ST/SU/TD/TH/TM/UR/US/UT" OR MM "Critically Ill Patients/CL/ED/EI/EV/HI/LJ/OG/PF/SN") OR MH "Acute Coronary Syndrome" OR MH "Critical Care+") AND (("continues glucose monitor*" OR CGM OR "flash glucose monitor*" OR FGM OR MH "Continuous Glucose Monitoring") AND (TI "clinical trial*" OR AB "clinical trial*" OR TI "controlled trial*" OR AB "controlled trial*" OR (MH "Randomized Controlled Trials+" OR MH "Clinical Trials+")))

CENTRAL (Cochrane) search:

ID	Search	Hits
#1	MeSH descriptor: [Continuous Glucose Monitoring] explode all trees	77
#2	MeSH descriptor: [Intensive Care Units] explode all trees	6275
#3	MeSH descriptor: [Critical Illness] explode all trees	3807
#4	MeSH descriptor: [Shock] explode all trees	3528
#5	(Continuous glucose monitor*):ti,ab,kw (Word variations have been searched)	5929
#6	(Continuous glucose measure*):ti,ab,kw (Word variations have been searched)	5684
#7	(Continuous glucose management):ti,ab,kw (Word variations have been searched)	2197
#8	(Continuous glucose sensor*):ti,ab,kw (Word variations have been searched)	1146
#9	(CGM):ti,ab,kw (Word variations have been searched)	2938
#10	#1 OR #5 OR #6 OR #7 OR #8 OR #9	9799
#11	(critical* ill*):ti,ab,kw (Word variations have been searched)	13752
#12	(critical* care):ti,ab,kw (Word variations have been searched)	19930
#13	(severe* ill*):ti,ab,kw (Word variations have been searched)	17479
#14	(intensive care unit*):ti,ab,kw (Word variations have been searched)	31435
#15	(ICU*):ti,ab,kw (Word variations have been searched)	21488
#16	MeSH descriptor: [Respiration, Artificial] explode all trees	9275
#17	(mechanic* ventilat*):ti,ab,kw (Word variations have been searched)	19173
#18	(shock*):ti,ab,kw (Word variations have been searched)	16133
#19	("acute coronary syndrome"):ti,ab,kw (Word variations have been searched)	9130
#20	MeSH descriptor: [Acute Coronary Syndrome] explode all trees	3121
#21	#2 OR #3 OR #4 OR #11 OR #12 OR #13 OR #14 OR #15 OR #16 OR #17 OR #18 OR #19 OR #20	105742
#22	#21 AND #10	780
#23	(Flash glucose monitor*):ti,ab,kw (Word variations have been searched)	364
#24	#10 OR #23	9954
#25	#24 AND #21	782

CLARIVATE (web of science) search:

Search

#5 AND #23

136

#16 AND #22

631

#17 OR #18 OR #19 OR #20 OR #21

19,498

"continuous glucose monitor*" (All Fields)

10,989

"cgm" (All Fields)

12,524

"flash glucose monitor*" (All Fields)

902

"continuous glucose manag*" (All Fields)

6

"continuous glucose measur*" (All Fields)

178

#6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15

1,260,112

"sever* ill*" (All Fields)

15,999

"critical* care" (All Fields)

288,535

"critical* ill*" (All Fields)

110,741

ALL=("intensive care unit*")

201,088

"ICU*" (All Fields)

160,341

"shock*" (All Fields)

609,510

"mechanic* ventitalion" (All Fields)

0

ALL=("mechanic* ventilat*")

91,433

ALL=("artifical ventilat*")

33

"acute coronary syndrom*" (All Fields)

63,902

#1 OR #2 OR #3 OR #4

3,702,211

"random* control* trial" (All Fields)

284,956

ALL=("random*")

2,789,080

"clinical trial*" (All Fields)

681,626

"allocat*" (All Fields)

600,117

Appendix C. See Table C1

TABLE C1.

CGM characteristics and glucose management practices.

Authors	CGM product name	Reports use of CGM alarms (yes/no)	Reports use of sensor calibrations (yes/no)	Reports use of any insulin protocol (yes/no)	Reports use of CGM specific insulin protocol (yes/no)	Target glucose level, mmol/L	Time in range definition, mmol/L	Time above range, mmol/L	Time below range, mmol/L	Additional hypoglycemia measurements
Holzinger et al. [7]	Guardian; Medtronic MiniMed, Northridge, CA	No	Yes	Yes	No	4.4–6.1	< 6.1	—	—	—
LV et al. [29]	RT‐CGMS system (Sari MediTech, China)	Yes	Yes	Yes	No	7.8–11.1	7.8–11.1	—	—	Low blood glucose index
Brunner et al. [30]	Guardian real time CGMS (Medtronic MiniMed)	No	Yes	Yes	No	4.44–6.11	4.44–6.11	—	—	Time below 2.22, 3.33, and 4.44 mmol/L, minutes/24 h
Qian et al. [31]	Lunen Real‐time Glucose Monitoring System (by Huida Shandong Medical Technology Co. Ltd.)	No	Yes	Yes	No	7.8–11.1	7.8–11.1	—	—	Duration of hypoglycemia
Kopecký et al. [32]	Guardian REAL‐Time CGMS (MiniMed Medtronic, Northridge, CA, USA)	No	Yes	Yes	No	4.4–6.1	4.4–6.1	> 6.1	< 4.4	—
Boom et al. [33]	FreeStyle Navigator, Abbott Diabetes Care, Alameda, CA, USA	Yes	Yes	Yes	No	5.0–9.0	5.0–9.0	> 9.0	2.2–5.0	Number of patients with 2.2–3.9 mmol/L
De Block et al. [34]	GlucoDayOS; A. Menarini Diagnostics, Florence, Italy	No	Yes	Yes	No	4.4–6.1	4.4–6.1	> 11.1	< 3.3	Low blood glucose index, time in hypoglycemia/24 h, and incidence/24 h
Guan et al. [35]	The CGMS‐2009 device (Zhuhai Princeton Medical Technology Co. Ltd.) (Zhuhai)	No	No	No	No	—	—	—	—	—
Galderisi et al. [36]	G4 Platinum CGM system (Dexcom Inc., San Diego, CA)	Yes	Yes	Yes	Yes	4.0–8.0	4.0–8.0	> 10	< 2.6	Time and number of episodes in 2.6–3.9 mmol/L
Tian et al. [37]	Meiqi RGMS‐I real‐time CGMS	No	Yes	No	No	—	—	—	—	—
Zhang [38]	Not specified	No	No	No	No	—	—	—	—	—
Lu et al. [6]	DGMS, San MediTech Medical Technology Co. Ltd, Huzhou, Zhejiang, China	Yes	Yes	Yes	No	8.0–10.0	8.0–10.0	> 10	< 8	< 4 mmol/L and duration of hypoglycemia
Preiser et al. [39]	GlucoClear, Edwards Lifesciences, Irvine, CA, USA.	Yes	Yes	Yes	No	5–8.3	5–8.3	—	< 3.9	Number of patients with < 3.9 mmol/L
Li et al. [40]	Not specified	No	No	Yes	No	Insulin initiated > 8.3 and target < 10	< 10	—	—	—
Beardsall et al. [41]	Enlite glucose sensor (Medtronic, Northridge, CA, USA)	No	Yes	Yes	Yes	4.0–8.0	2.6–10	> 15	< 2.6	Incidence of 2.2–2.6 mmol/L, and episodes < 2.6 for > 1 h
Chu et al. [42]	Guardian sensor 3	No	Yes	Yes	No	3.9–10	3.9–10	> 10	< 3.9	< 3.9
Shang et al. [43]	Freestyle libre; Abbott Diabetes Care	No	No	No	No	—	—	—	—	< 3.9
Franck et al. [44]	Dexcom G6, Dexcom Inc., San Diego, CA, USA	Yes	Yes	Yes	No	3.9–10	3.9–10	> 13.9	< 3.0	Time in 3.0–3.8 mmol/L

Appendix D. See Table D1

TABLE D1.

Clinical diversity in meta‐analyses assessments.

Outcome	Mortality	Hypoglycemia
Trials	Holzinger et al.	Holzinger et al.
	LV et al.	LV et al.
	Qian et al.	Qian et al.
	Boom et al.	Kopecky et al.
	De Block et al.	Boom et al.
	Guan et al.	De Block et al.
	Galderisi et al.	Guan et al.
	Zhang	Tian et al.
	Lu et al.	Zhang
	Preiser et al.	Lu et al.
	Li et al.	Preiser et al.
	Beardsall et al.	Li et al.
	Chu et al.	Beardsall et al.
	Shang et al.	Chu et al.
	Franck et al.	Shang et al.

CDIM scores
Setting (study year, country, treating unit)	2/2	2/2
Population (age, sex, inclusion criteria, baseline disease severity, comorbidities)	7/8	7/8
Intervention (intensity, strength, duration, timing, control, co‐interventions)	5/8	5/8
Outcome (definition, timing)	3/4	1/4
Total	17/22	15/22
Conclusion	Moderate clinical diversity	Moderate clinical diversity

Appendix E. See Figure E1

FIGURE E1.

Trial sequential analysis. (a) Trial sequential analysis—hypoglycemia. (b) Trial sequential analysis—mortality.

Appendix F. See Figure F1

FIGURE F1.

Subgroup analysis for mortality. (a) Subgroup analysis risk of bias. (b) Subgroup analysis adult/neonatal. (c) Subgroup analysis event timing. (d) Subgroup analysis diabetes. (e) Subgroup analysis: country of origin. (f) Subgroup analysis co‐intervention. (g) Subgroup analysis glucose target. (h) Meta‐regression on study year (test of interaction p = 0.91).

Appendix G. See Figure G1

FIGURE G1.

Sensitivity analysis mortality. (a) Best/worst case scenario for mortality. (b) Worst/best case scenario for mortality.

Appendix H. See Figure H1

FIGURE H1.

Subgroup analysis for hypoglycemia. (a) Subgroup analysis risk of bias. (b) Subgroup analysis: country of origin. (c) Subgroup analysis diabetes. (d) Subgroup analysis co‐intervention. (e) Subgroup analysis adult/neonatal. (f) Subgroup analysis glucose target range. (g) Meta‐regression analysis on study year (test of interaction p = 0.02).

Appendix I. See Figure I1

FIGURE I1.

Sensitivity Analysis for Hypoglycemia. (a) Best/worst case scenario for hypoglycemia. (b) worst/best case scenario for hypoglycemia.

Appendix J. See Figure J1

FIGURE J1.

Glycemic process outcomes. (a) Meta‐analysis for coefficient of variation. (b) Meta‐analysis for time above range. (c) Meta‐analysis for time below range. (d) Meta‐analysis for time in range.

Appendix K. See Figure K1

FIGURE K1.

Funnel plots. (a) Funnel plot for mortality. (b) Funnel plot for hypoglycemia.

Appendix L. See Figure L1

FIGURE L1.

Risk of Bias for Hypoglycemia and Glycemic Process Outcomes. (a) Risk of bias for hypoglycemia. (b) Risk of bias for glycemic process outcomes. (c) Risk of bias for the one included cluster randomized trial (similar for both mortality and glycemic outcomes).

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.