Comparative Effectiveness of Continuous Glucose Monitoring Versus Blood Glucose Monitoring in Pregnancy With Diabetes: Systematic Review and Meta-analysis With Trial Sequential Analysis

Abstract

OBJECTIVE

The value of adding continuous glucose monitoring (CGM) to blood glucose monitoring (BGM) during pregnancy is unclear.

PURPOSE

To clarify CGM's role in pregnancy using meta-analysis, Grading of Recommendations Assessment, Development and Evaluation (GRADE) ratings, harmonized neonate size definitions, and trial sequential analysis (TSA).

DATA SOURCES

We performed bibliographic database and hand searches. We selected clinically relevant randomized controlled trials (RCTs) comparing CGM and BGM in pregnancy.

DATA EXTRACTION

Neonate size measures were primary; additional maternal, obstetric, and neonatal outcomes data were also extracted.

DATA SYNTHESIS

Seventeen RCTs (n = 2,672 participants) met inclusion criteria. Incidence of macrosomia was reduced overall (risk ratio [RR] 0.67 [95% CI 0.48, 0.94]; strength of evidence [SOE], moderate) and for gestational diabetes mellitus (GDM) (RR 0.55 [95% CI 0.35, 0.86]; SOE, high); incidence of large for gestational age was also reduced overall (RR 0.77 [95% CI 0.61, 0.98]; SOE, very low) and for GDM (RR 0.65 [95% CI 0.45, 0.96]; SOE, very low). Incidence of small for gestational age (SGA) was elevated with CGM but not statistically significant (RR 1.37 [95% CI 0.96, 1.95]; SOE, moderate). The TSA showed no neonate size outcome met the required information size (RIS) to draw firm conclusions, and no cumulative Z curve crossed the adjusted monitoring boundary. Summary statistical significance may fluctuate with the publication of new studies.

CONCLUSIONS

CGM benefits mothers with preexisting diabetes and GDM and their neonates for varying outcomes, including reducing large size at birth. Our TSA found that the RIS for neonate size outcomes has not been reached. Literature should be monitored particularly for RCTs reporting SGA in GDM, which the TSA shows cannot be conclusively ruled out as a risk.

Introduction

Maternal hyperglycemia in pregnancy increases risks for both mother and neonate, including large neonate size at birth (1), which raises risks of complications including shoulder dystocia, birth trauma, and cesarean delivery (2), as well as long-term offspring complications, including obesity, type 2 diabetes, and cardiovascular disease (3,4).

Continuous glucose monitoring (CGM) may offer incremental benefit when added to blood glucose monitoring (BGM) and HbA_1c during pregnancy to inform glucose management, because it delivers real-time data on glucose fluctuations, including overnight and short-term excursions (5). It measures mean glucose, time spent in (and above and below) target glucose ranges, and glucose variability—metrics that enable timely adjustments of therapy (6).

Four 2026 systematic reviews comparing CGM and BGM reached different conclusions about neonatal size at birth (Supplementary Table 1), which is the primary outcome of this review, given the links of CGM and BGM to other neonatal outcomes. Suvarna and Shetty, in five gestational diabetes mellitus (GDM) randomized controlled trials (RCTs), found reduced incidence of macrosomia but no effect on the incidence of large for gestational age (LGA) or small for gestational age (SGA) (7). Gautam et al., analyzing 11 GDM RCTs, found no significant macrosomia reduction and did not pool LGA or SGA, due to definitional heterogeneity across studies (8). Lai et al. reported no difference in LGA in 17 GDM and preexisting diabetes RCTs but found a significantly greater risk of SGA with CGM (9). Dutta et al., in 17 GDM and preexisting diabetes RCTs, found that effects depended on type of diabetes and CGM (10).

The present study addresses methodological sources of divergence among previous systematic reviews to clarify the role of CGM in pregnancy. Distinguishing features include prespecified stratification by diabetes type and analysis of a greater array of factors related to CGM use; application of standardized definitions for neonate size at birth (macrosomia ≥4,000 g, LGA >90th percentile, and SGA <10th percentile); trial sequential analysis (TSA) for primary outcomes (neonate size) to determine whether the cumulative evidence has reached the required information size (RIS) for firm conclusions; Grading of Recommendations Assessment, Development and Evaluation (GRADE) certainty assessment for all outcomes; and no English-language restriction.

Research Design and Methods

This evidence synthesis aligns with the National Academy of Medicine Committee on Standards for Systematic Reviews of Comparative Effectiveness Research (11), reports in accordance with the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (12), and was prospectively registered with PROSPERO (identifier CRD42024549551).

Literature Search

We used free-text and controlled vocabulary targeting the terms “continuous glucose monitoring” and “pregnancy” in PubMed and Cochrane Central Register of Controlled Trials (because it searches multiple databases including Embase) on 29 April 2026 (see Supplementary Table 2 for the full strategy). We searched relevant bibliographies and the internet for additional RCTs and gray literature; no additional RCTs were identified. We selected RCTs allocating pregnant people with diabetes to CGM or BGM and reporting maternal glycemic, obstetric, and/or neonatal outcomes in adequate detail for data extraction and critical appraisal, to select the highest-quality evidence available.

Study Selection

All database records were screened for retrieval in blinded duplicate review by three reviewers (M.N., K.F., and T.C.S.) using Rayyan software (Rayyan, Cambridge, MA), and retrieved texts were evaluated in blinded duplication by the same reviewers (M.N., K.F., and T.C.S.) with adjudication by a third party as needed (R.R.B., G.A., or S.P.). We tracked studies in preformatted and pilot-tested forms in Microsoft 365 applications (Microsoft, Redmond, WA) and managed citations in EndNote 21 (Clarivate, Philadelphia, PA).

Data Extraction

Data were extracted by one reviewer (M.N.) and checked by a second (K.F. and T.C.S.) via a consensus process. Primary outcomes are macrosomia (birth weight ≥4,000 g), LGA (birth weight ≥90th percentile), and SGA (birth weight <10th percentile). We reclassified trial data by these definitions as needed, such that birth weights ≥90th centile labeled “macrosomia” (13,14) were reclassified as LGA. Alternative definitions were excluded (15). One study labeled >90th and ≤95th centile as LGA and >95th centile macrosomia so we added events and reclassified as LGA (16). Another study reported macrosomia and LGA combined and was excluded from those outcomes (17). One study defined SGA as ≤5th centile and was excluded from that outcome (14).

Data Analysis

Pairwise random-effects meta-analysis (18) was performed for outcomes with data reported in a form that is statistically compatible or could be converted into compatibility from studies without excessive clinical or methodological heterogeneity. We calculated mean differences and risk ratios (RRs). Median and interquartile range were converted to mean and SD (19). We interpreted substantial heterogeneity as greater than would be expected due to chance alone, I² > 50% (20) and defined significant heterogeneity as P < 0.10.

We performed subgroup analysis by type of diabetes (preexisting, GDM). We would have preferred to separate type 1 diabetes and type 2 diabetes, but studies enrolling both populations reported combined data. We also assessed CGM type (intermittently scanned CGM [isCGM], real-time CGM [rtCGM], or professional), total application time (<1 month or ≥1 month), whether the device was self-managed (isCGM or rtCGM) versus professionally managed, and the application of CGM (continuous, episodic, single session). Meta-regression was considered appropriate for outcomes addressed by ≥10 RCTs with I² > 50%. We assessed robustness using cumulative meta-analysis by date and one-study-removed analyses. For outcomes with 10 or more studies, a trim-and-fill test (21) was performed to detect potential publication bias along with the visual inspection of Funnel plots.

For macrosomia, LGA, and SGA outcomes, we performed TSA to determine if the evidence base has achieved RIS to avoid false conclusions, because they are our primary outcome measures and because the findings of previous systematic reviews are inconsistent (22,23). The RIS was calculated from the control groups of studies included in this review (total events/total live births), assuming a base case of 20% relative risk reduction with sensitivity analyses at 15% and 25%. The diversity-adjusted required information size was computed as RIS/(1 − I²) to account for between-trial heterogeneity (I²) in random-effects meta-analyses (23).

Critical Appraisal

We used Cochrane’s Risk of Bias 2 (RoB 2) instrument in blinded duplication by two experienced systematic reviewers (M.N., K.F., or T.C.S.) (24). We assessed certainty in the cumulative evidence in alignment with the GRADE approach (25).

Data and Resource Availability

The full search strategy used to generate the findings of this study is available in Supplementary Table 2.

Results

Evidence Base

Bibliographic searches yielded 189 nonduplicate citations, of which 17 RCTs met inclusion criteria (Fig. 1 and Supplementary Table 3). Discussion of excluded studies of note (26–33) is available in the Supplementary Material.

Figure 1.

PRISMA 2020 flow diagram.

Twelve RCTs (15,16,34–43) allocated 1,892 people with GDM and four RCTs (13,17,44,45) allocated 480 pregnant people with preexisting diabetes to either CGM with BGM or BGM only. One additional RCT (14) reported data for 191 people with preexisting diabetes and 109 with GDM separately for three outcomes (preeclampsia, gestational hypertension, and LGA) and combined for others.

Studies were published between 2007 and 2026 and conducted worldwide. From 40 (17) to 375 (38) participants were allocated to one of two study arms per study, and most studies were small. Participants were generally aged in their early 30s (years). Individuals with preexisting diabetes were typically enrolled in the first trimester of pregnancy, and individuals with GDM typically enrolled in late second to early third trimester. Types of CGM device and duration of CGM use varied.

Aside from CGM use in the intervention arm, background obstetric and diabetes management was equivalent for the CGM and BGM arms within each trial, though these management approaches varied across trials. Such variation may contribute to between-study heterogeneity; however, it does not confound the within-study between-arm comparisons that form the basis of this meta-analysis. In five RCTs, blinded CGMs were administered in the control-group study period (16,35,38,41,44), but in only one were the data unmasked and used to inform medication treatments or other decisions during the study (35).

Data on diabetes medication use were too sparse and metrics too heterogenous to analyze changes; therefore, they are reported here to characterize the population. One GDM study excluded data on individuals with diabetes using insulin or metformin (38), and two others enrolled individuals using insulin (14,40), but most did not describe prestudy medication use. The proportion of insulin use by the end of the studies was generally higher in the CGM arm, reaching statistical significance in two studies (15,42) but not in others (16,36–38). Two additional studies reported similar proportions of insulin users between groups (34,39). All studies addressing preexisting diabetes included data on administered insulin but none quantified comparative changes.

Outcomes are described in the following sections and are summarized in Table 1. The Supplementary Material contains forest plots for meta-analyses, details of subgroup analyses for each quantitatively analyzed outcome, adjunctive content, and a full GRADE assessment table with our evaluation rubric.

Table 1.

Summary of meta-analyses with GRADE assessment

Outcome	Population (reference)	K analyzed	I/C, n	Effect (95% CI)	I2 (%)	Favored method**	Certainty (strength of evidence)
Macrosomia (≥4,000 g)	Overall (13,17,37,38,41–44)*	9	738/713	RR 0.67 (0.48–0.94)†	2	CGM	Moderate
	GDM (34,36–39,41–43)	8	638/613	RR 0.55 (0.35–0.86)†	0	CGM	High
	Preexisting (44)	1	100/100	RR 0.85 (0.53–1.38)	NA	NS	Very low
LGA (>90th centile)	Overall (13,14,16,35–45)	14	1,195/1,107	RR 0.77 (0.61–0.98)†	49§	CGM	Very low
	GDM (14,16,35–43)	11	893/811	RR 0.65 (0.45–0.96)†	43§	CGM	Very low
	Preexisting (13,14,44,45)	4	302/296	RR 0.89 (0.66–1.21)	63	NS	Low
SGA (<10th centile)	Overall (13,17,37,38,41–44)	8	611/578	RR 1.37 (0.96–1.95)	0	NS	Moderate
	GDM (37,38,41–43)	5	453/429	RR 1.36 (0.94–1.95)	0	NS	Moderate
	Preexisting (13,17,44)	3	158/149	RR 1.54 (0.39–6.14)	0	NS	Low
Preterm birth (<37 weeks)	Overall (13–15,17,34–38,40,41,43–45)	14	1,056/979	RR 1.00 (0.82–1.22)	0	NS	Moderate
	GDM (13,17,37,38,41–44)	9	679/608	RR 1.10 (0.73–1.67)	14	NS	Moderate
	Preexisting (13,17,44,45)	4	234/224	RR 0.98 (0.74–1.29)	0	NS	Moderate
	Mixed (14)	1	143/147	RR 0.94 (0.61–1.44)	NA	NS	Very low
Neonatal hypoglycemia	Overall (13–17,34–42,44,45)*	16	1,196/1,100	RR 0.87 (0.74–1.02)	0	NS	Moderate
	GDM (15,16,34–42)	11	819/731	RR 0.83 (0.65–1.06)	0	NS	Moderate
	Preexisting (13,17,44,45)	4	234/222	RR 0.71 (0.51–0.98)†	0	CGM	Moderate
	Mixed (14)	1	143/147	RR 1.06 (0.81–1.40)	NA	NS	Very low
NICU admission	Overall (13,15,16,34–38,40,41,43,44)*	12	971/878	RR 0.74 (0.60–0.91)†	0	CGM	Moderate
	GDM (15,16,34–38,41,43)	9	809/723	RR 0.75 (0.58–0.98)†	0	CGM	Moderate
	Preexisting (13,40,44)	3	162/155	RR 0.70 (0.49–1.00)	0	NS	Low
Maternal fasting glucose, mg/dL	GDM (34,36,40,41,43)	5	304/266	MD −3.56 (−6.68, −0.43)†	83§	CGM	Very low
TIR, %	Overall (16,17,35,37,38,41,44)	7	600/431	MD 2.28 (−0.99, 5.55)	92§	NS	Very low
	GDM (16,35,37,38,41)	5	502/335	MD 0.94 (−2.71, 4.59)	93§	Null	Very low
	Preexisting (17,44)	2	98/96	MD 5.68 (3.10–8.26)†	0	CGM	Moderate
Time above range, %	Overall (16,17,35,37,38,41,44)	7	600/431	MD −1.01 (−3.09, 1.06)	84§	Null	Very low
	GDM (16,35,37,38,41)	5	502/335	MD −0.45 (−2.86, 1.96)	87§	Null	Very low
	Preexisting (17,44)	2	98/96	MD −2.85 (−5.72, 0.03)	11	NS	Low
TBR, %	Overall (16,17,35,38,41,44)*	6	538/369	MD −0.52 (−1.05, 0.02)	83§	NS	Very low
	GDM (16,35,38,41)	4	440/273	MD −0.05 (−0.24, 0.15)	0	Null	High
	Preexisting (17,44)	2	98/96	MD −2.47 (−4.68, −0.25)†	82§	CGM	Very low
Maternal weight gain, kg	Overall (15–17,36,38–45)	12	902/847	MD −0.37 (−0.94, 0.20)	34	Null	High
	GDM (15,16,36,38–43)	9	702/651	MD −0.33 (−0.98, 0.31)	39	Null	High
	Preexisting (17,44,45)	3	200/196	MD −0.53 (−2.17, 1.11)	43	NS	Moderate
Cesarean delivery	Overall (13–17,34–45)	17	1,315/1,229	RR 0.94 (0.87–1.02)	0	NS	Moderate
	GDM (15,16,34–43)	12	934/853	RR 0.95 (0.85–1.05)	0	NS	Moderate
	Preexisting (13,17,44,45)	4	238/229	RR 0.95 (0.80–1.12)	20	NS	Moderate
	Mixed (14)	1	143/147	RR 0.93 (0.70–1.23)	NA	NS	Very low
Pregnancy-related hypertension	Overall (14,36–38,41,43)	6	618/584	RR 0.85 (0.52–1.38)	40	NS	Moderate
	GDM (14,36–38,41,43)	6	529/491	RR 1.01 (0.63–1.60)	22	NS	Moderate
	Preexisting (14)	1	89/93	RR 0.44 (0.25–0.77)†	NA	CGM	Very low
Preeclampsia	Overall (13,14,16,35,44,45)	6	595/538	RR 0.73 (0.44–1.21)	21	NS	Moderate
	GDM (14,16,35)	3	289/235	RR 0.79 (0.28–2.22)	44	NS	Low
	Preexisting (13,14,44,45)	4	306/303	RR 0.65 (0.36–1.19)	15	NS	Moderate

*One additional study reported this outcome but was dropped from the analysis due to zero events in each group or data reporting issues. See text for details.

†Statistically significant effect (P < 0.05) favoring CGM.

§Statistically significant heterogeneity (P < 0.10). Voormolen et al. (14) reported data for GDM and preexisting diabetes groups separately for LGA, pregnancy-related hypertension, and preeclampsia, and these data were included in both population subgroup analyses for these outcomes.

**Null denotes a confidence interval that crosses the null but lies entirely within the clinically trivial range (equivalence pattern); NS denotes a confidence interval that crosses the null but includes clinically meaningful magnitudes. NA, not applicable.

Neonatal Outcomes

The incidence of macrosomia and LGA was significantly reduced with CGM (Fig. 2). Nine studies with 1,451 births were analyzed for macrosomia (birth weight ≥4,000 g) (34,36–39,41–44); one additional study (40) was dropped due to zero events. An approximately 30% reduction, overall, favoring CGM use was observed (RR 0.67; 95% CI 0.48–0.94) (Fig. 2A). Notably, only one RCT addressed a population with preexisting diabetes (44), which did not favor CGM when considered individually, although the eight GDM studies considered collectively did (P = 0.03) (Supplementary Fig. 1 and Supplementary Table 4). Heterogeneity was not substantial overall, but subgroup analyses suggested potential advantages with isCGM, patient use of CGM (isCGM or rtCGM), episodic use, and use for less than 1 month (Supplementary Table 4).

Figure 2.

Meta-analyses of neonatal size outcomes by diabetes type. Red diamonds indicate study effects sub-grouped (pooled) by type of diabetes. See the Supplementary Material for an expanded version with counts and for all other outcomes.

Fourteen studies with 15 groups assessing 2,302 births reported LGA (birth weight ≥90th percentile) (13,14,16,35–45). Incidence of LGA was approximately 25% lower with CGM use overall (RR 0.77 [95% CI 0.61–0.98]; P = 0.04) and 35% lower in the GDM studies (RR 0.65 [95% CI 0.45–0.96]; P = 0.03). The effect was not significant in the preexisting diabetes studies considered separately (Fig. 2B and Supplementary Fig. 2). Additional subgroup analyses suggest potential advantages with continuous application of CGM and use for less than a month (Supplementary Table 5). Incidence of LGA was the only outcome in this analysis that met prespecified criteria for meta-regression. We assessed feasibility of performing it for multiple maternal baseline factors, but due to data reporting limitations, we were only able to run models for mean HbA_1c at study entry and year of study publication. Neither was significant.

Eight studies with 1,189 births were assessed for SGA (<10th percentile birth weight) (13,17,37,38,41–44). One additional study was dropped from the analysis due to zero events in either group (40). Incidence of SGA was higher with CGM and approached, but did not achieve, statistical significance (RR 1.37; 95% CI 0.96–1.95). There was no significant heterogeneity and no statistically significant findings in subgroup analyses including diabetes type (Supplementary Fig. 3 and Supplementary Table 6).

There were no significant findings for preterm birth overall or in any subgroup (Supplementary Fig. 4 and Supplementary Table 7). Incidence of neonatal hypoglycemia was reduced with CGM for the preexisting diabetes studies only (RR 0.71 [95% CI 0.51–0.98]; 4 RCTs, n = 456 participants) (Supplementary Fig. 5). Heterogeneity was not observed (I² = 0%) despite variation among studies in how neonatal hypoglycemia was defined (Supplementary Table 8), and no significant findings upon subgroup analysis were noted (Supplementary Table 9). Neonatal intensive care unit (NICU) admission was approximately 25% lower in the CGM group overall (RR 0.74 [95% CI 0.60–0.91]; 12 RCTs, n = 1,849 participants). It was lower with CGM for GDM (P = 0.04) and marginal for preexisting diabetes (P = 0.05) (Supplementary Fig. 6). Subgroup analyses suggest reduction in NICU admission may be associated with use of rtCGM, self CGM use, and continuous application (Supplementary Table 10). Data on neonatal death and related outcomes (Supplementary Table 11) do not suggest differences between CGM and BGM groups.

Maternal Outcomes

Modest benefits for maternal glycemia were observed. Five RCTs (n = 570 participants) (34,36,40,41,43) found a reduction of −3.56 mg/dL (95% CI −6.68, −0.43) in fasting glucose among individuals with GDM (Supplementary Fig. 7). In two studies conducted with participants with preexisting diabetes (17,44), CGM was associated with improved time in range (TIR) (5.68%; 95% CI 3.10–8.26) (Supplementary Fig. 8) and time below range (TBR) (−2.47%; 95% CI −4.68, −0.25) (Supplementary Fig. 9), but benefit was not observed in the pooled GDM studies or overall. Notably, GDM data composed the majority of the overall evidence, and the control groups in the GDM groups had very high TIR (median or mean ≥87% for all), creating a mathematical ceiling that constrained the possibility of CGM benefit, indicating that TIR is a poor outcome metric of glycemia for GDM rather than evidence of no effect. The TBR was not significantly improved for GDM, likely also due to high TIR (Supplementary Fig. 10). Subgroup analyses were not meaningful for any of the maternal glucose metrics (Supplementary Tables 12–15).

Maternal total weight gain was not associated with benefits for CGM overall or by diabetes type, but subgroups suggested small benefits (∼1 kg) with professional CGM, use in a single episode, and use for less than a month (Supplementary Fig. 11 and Supplementary Table 16). Cesarean ldelivery was not different overall or in any subgroup (Supplementary Fig. 12 and Supplementary Table 17). One study reported significant benefit for preexisting diabetes for reduction of pregnancy-related hypertensive disorders (14), but this was not observed in six GDM studies (14,36–38,41,43) (Supplementary Fig. 13 and Supplementary Table 18). No benefit was observed for preeclampsia overall or in any subgroup (Supplementary Fig. 14 and Supplementary Table 19).

Evidence was too sparse for meta-analysis of other maternal outcomes. Severe hypoglycemic episodes were infrequent in both arms of two RCTS of preexisting diabetes (13,44). Among GDM trials, two reported no episodes in either arm (16,40) and one (39) reported infrequent CGM-detected episodes without parallel control-arm data; pooling was not feasible. Feig et al. (44) reported on the only trial that presented quality-of-life data. They found no significant differences in the 12-Item Short Form Health Survey between arms and also reported satisfaction that increased over time and favored CGM. Elkind-Hirsch et al. (35) reported high CGM satisfaction without a BGM comparator, and participants in the study by Zorko et al. (43) expressed interest in adjunctive isCGM for future pregnancies. No maternal deaths were reported.

Critical Appraisal

The RoB 2 assessment identified some concerns for RoB, but the evidence base was generally low risk (Supplementary Fig. 15). RCTs were free from substantial baseline imbalances and used objective measures and data collection methods. Participants were not blinded to treatment, but the objective nature of the outcomes minimizes potential related bias. Furthermore, we do not believe study participants would risk serious harm to their fetuses by not monitoring their blood glucose, due to group allocation dissatisfaction. Concealment of allocation was explicitly reported in a minority of trials; however, all but two initiated treatment as assigned (14,40) and these studies were not influential in one-study-removed sensitivity analyses. Elkind-Hirsch et al. (35) used blinded CGM to inform treatment for the BGM control group, but while this may influence the purity of the BGM control desired for this review, it is not a departure from their planned protocol. Importantly, this study was not impactful in one-study-removed analyses.

We could not locate a published protocol or record of registry for three studies (15,34,42), but those studies reported having prospective ethics committee approval. Two RCTs referenced HbA_1c in their abstract (15) or methods (36) without reporting it among their results, but this had no analytic impact in our study, because we did not assess HbA_1c. No other sources of concern were identified regarding missing outcome data, outcome measurement, or selection of reported results. The studies were too similar for RoB domains to be used as moderators for subgroup analyses.

Funnel plots and statistical tests were applied to outcomes addressed by ≥10 RCTs (LGA, neonatal hypoglycemia, cesarean delivery, NICU admission, preterm birth, maternal weight gain, and subgroups of those outcomes with ≥10 RCTs). Neither visual inspection nor formal testing suggested overestimation of CGM’s benefit, except for LGA. For LGA, trim-and-fill analysis suggested two potential missing studies and an adjusted random-effects estimate overall (RR 0.80; 95% CI 0.62–1.01) and for GDM (RR 0.70; 95% CI 0.48–1.02). While similar in direction and magnitude to the observed estimate (RR 0.77; 95% CI 0.61–0.98), this suggests statistical significance is sensitive to adjustment. One-study-removed analyses and cumulative analyses applied to the overall evidence bases were stable for all outcomes.

Strength of Evidence

The GRADE assessment yielded variable strength of evidence (SOE) or certainty for the primary outcomes measuring neonate size. Confidence in macrosomia was moderate certainty of decreased incidence overall due to imprecision and very low for preexisting diabetes due to imprecision and inconsistency, yet high for CGM (no downgrades required). LGA overall had very low certainty of decreased incidence, because the upper confidence limits sit close to the null, suggesting potential instability of statistical significance with the publication of additional trials, as well as publication bias and inconsistency. The GDM subgroup was rated very low due to imprecision and publication bias, and the preexisting diabetes subgroup was rated low due to inconsistency and imprecision. SGA was rated moderate certainty of no effect in the overall and GDM population strata, due to imprecision because the lower confidence limit lies within 0.10 of null on the harm side, but with some reservation because the total sample size is small, especially in consideration of SGA’s low control event rate. In preexisting diabetes, it was rated as low due to extreme imprecision. Across other outcomes, CGM was associated with a reduction in the incidence of NICU admission (moderate SOE) and neonatal hypoglycemia in pregestational diabetes (moderate SOE), with moderate- or high-certainty findings of no clinically meaningful effect on the incidence of cesarean delivery or the amount of gestational weight gain or TBR in GDM. Glycemic metrics showed substantial heterogeneity, with moderate certainty of CGM benefit on TIR only in pregestational diabetes. High levels of TIR in the control groups of GDM groups very likely limit the utility of TIR as an outcome for GDM studies. Single-trial subgroup estimates were rated as very low certainty due to lack of replication. The full GRADE table, rubric, and footnotes are provided in Supplementary Table 20 and Supplementary Material.

Trial Sequential Analysis

The TSA found no neonate size outcome reached RIS to draw firm conclusions, and no cumulative Z curve crossed the adjusted monitoring boundary (Fig. 3). For macrosomia, analyzed RCTs provided only approximately 16% of evidence needed to confirm a 20% relative effect (and 12% in GDM; preexisting diabetes was not applicable, because there was only one applicable study). The RIS was higher for LGA, at approximately 55% overall and 52% with preexisting diabetes, but only 23% in the GDM studies. For SGA, all strata were markedly underinformed (overall ∼15%; GDM 14%; preexisting diabetes <1%), and the cumulative Z curve drifted in the direction of more SGA in CGM arms (approximately −1.75) without reaching statistical significance. This means that a clinically meaningful increase in SGA risk cannot be excluded. Sensitivity analyses did not alter these conclusions. Although GRADE considers sample size in imprecision, it does not integrate the TSA specifically, and we did not elect to apply it, because it would apply an outsized and inconsistent consideration of inconsistency to neonate size outcomes.

Figure 3.

TSA of neonatal size.

Conclusions

CGM is a useful adjunct to BGM in pregnancies complicated by diabetes, with the most consistent benefits being lower rates of macrosomia (overall SOE moderate, GDM SOE high), NICU admission (overall and GDM SOE moderate), neonatal hypoglycemia (preexisting diabetes SOE moderate), and improved maternal glycemia (preexisting TIR SOE moderate). The incidence of LGA also is reduced (overall SOE very low, GDM SOE very low). Current evidence shows no effect on the incidence of SGA (SOE moderate overall and GDM, low preexisting diabetes). However, the TSA confirms the suspected unachieved RIS for all three measures of neonatal size, which means, especially when coupled with the marginal upper confidence limits for large neonatal size and low confidence limits for SGA, that subsequent meta-analyses may return differing findings for statistical significance despite the high and moderate strength GRADE ratings. Clinicians may be better served to look at overall trends while the RIS accumulates to futility than to rely on P values.

Our data indicate CGM does not meaningfully change incidence of cesarean delivery (SOE moderate overall) or preterm birth (SOE moderate overall), clinically meaningful amounts of gestational weight gain (SOE high overall), pregnancy-related hypertensive disorders (SOE moderate overall), or preeclampsia (SOE moderate overall). We found the evidence on TIR for GDM to be of little to no value as a metric, due to the very high TIR in the control groups.

Our review is distinguished by harmonized neonate-size definitions, TSA applied to those outcomes, stricter RoB exclusions, and comprehensive GRADE across 14 outcomes (n = 40 pooled estimates)—the broadest 2026 synthesis, to our knowledge (Supplementary Table 1). These methodologic choices account for most differences from concurrent reviews. Our findings align directionally with those of Lai et al. (9) on GDM LGA reduction and SGA elevation with CGM, but our overall LGA reached significance and, for SGA, although we recognize directionality toward increase, our strict GRADE yielded moderate certainty of no significant effect, whereas Lai et al. (9) concluded increased risk. Other reviews diverged primarily through narrower stratification (10), smaller evidence pools (7) (k = 4 vs. our k = 10), or because LGA and SGA data were not pooled (8). Minimal macrosomia heterogeneity (I² = 2%) supported our pooling; LGA heterogeneity was significant, but CGM-type subgroups were not.

Future Research

RIS for neonatal outcomes is high and not met by the current body of RCTs. Our TSA indicated the RIS for LGA is most mature and therefore likely to be achieved first; the literature should likewise be monitored for data on SGA, which is currently the most underinformed outcome in this context. Research is needed to determine whether the trend toward higher SGA (particularly in recent GDM trials) is a true effect. However, the number of pregnant people randomized would need to be quite large to achieve RIS, particularly for preexisting diabetes, for which the incidence of SGA was very low. Because people with type 1 diabetes have widespread access to CGM (where it is available) and people with type 2 diabetes increasingly do, it may make sense to allocate experimental resources to GDM, which would require under base case estimates of approximately 10,700 people for macrosomia (or 6,600 for LGA, or 6,160 for SGA). In the ongoing absence of RCTs that are adequately powered to detect SGA, real-world evidence, such as registries or large claims-based cohorts, may be useful to close the research gap.

Future research should also address whether TIR targets should differ for GDM and type 2 diabetes versus type 1 diabetes. The data presented in this review clearly show a ceiling effect of high control-group TIR in GDM participants, making it a poor choice of research metric of CGM efficacy in GDM; other measures of glycemia should be considered. Additional research topics include the optimal duration of the pregnancy-specific TIR; the CGM metrics most predictive of adverse pregnancy outcomes; which metrics best apply in early versus late gestation; and how ambulatory glucose profiles should be during pregnancy (46). Research should include RCTs with detailed CGM metric reporting (mean glucose, glucose variability, TIR, TBR, and time above range) and pre-and-post data per group to minimize baseline imbalance bias. Trials should clarify how the method of glucose measurement informed clinical decision-making, collect patient-reported outcomes (treatment satisfaction, fear of hypoglycemia, quality of life), and screen participants with GDM at baseline to exclude (or separately analyze) those with prediabetes or preexisting type 2 diabetes; two trials in this synthesis (34,42) reported baseline HbA_1c higher than expected for GDM, with an unclear effect on generalizability.

Discussion

Strengths of this review include analysis of the latest RCTs; harmonization of neonate size outcomes; broader outcome coverage and meta-analytic depth than previous reviews; formal publication bias testing; and use of TSA for primary outcomes. Review limitations include inclusion of older CGM devices and restriction to RCTs, which increases the internal validity of the findings but may diminish generalizability. Evidence-base limitations include pooled reporting across preexisting diabetes types, and sparse data on quality of life and treatment satisfaction.

The findings of our study support the continued use of adjunctive CGM for people with pregestational diabetes and its continued provisional, studied use in people with GDM. When considering the results of this review, clinicians may consider interpreting directional signals across neonate size outcomes as more informative than threshold-based P values and should remain attentive to the possibility that tighter glycemic control may not only decrease macrosomia in neonates of people with GDM but possibly increase SGA, with further research needed. No clear benefit for a particular type of CGM emerged, so factors such as availability, cost, and patient preference should be considered.

This article contains supplementary material online at https://doi.org/10.2337/figshare.33030929.