Assessment of Fetal Cardiac Function and Pregnancy Outcomes in Well‐Controlled Gestational Diabetes Mellitus Using Novel Ultrasound Technology: A Prospective Case–Control Study

Abstract

Objectives

To investigate the changes in fetal cardiac function in pregnant women with well‐controlled gestational diabetes mellitus (GDM) using a novel ultrasound technology named HOLO‐PW. Secondly, we aim to evaluate the predictive ability of cardiac function parameters for adverse perinatal outcomes and establish an individualized nomogram.

Methods

This study included 122 pregnant women with well‐controlled GDM and 256 pregnant women with normal blood glucose. Fetal cardiac function parameters were extracted based on the HOLO‐PW technology. The differences between the 2 groups were analyzed. Subgroup analysis was performed in the GDM group according to the pregnancy outcome and fetal growth status. The effectiveness of cardiac function parameters in predicting adverse pregnancy outcomes in GDM was evaluated by receiver operating characteristic (ROC) curves. Independent predictors were identified through logistic regression with LASSO variable selection, and a nomogram was developed. The model's performance was evaluated with ROC analysis, calibration curves, and decision curve analysis (DCA).

Results

Compared with the control cohort, the myocardial performance index (MPI), K index (KI), isovolumic contraction time (ICT), and isovolumic relaxation time (IRT) of left and right ventricles were increased in the fetuses of GDM cohort (p < .05). Compared with the appropriate for gestational age (AGA), the fetuses with abnormal growth in the GDM cohort presented significant differences in the cardiac function parameters (p < .05). The incidence of adverse pregnancy outcomes in the GDM cohort was higher than that in the control cohort. The cardiac function parameters of fetuses with adverse and normal perinatal outcomes significantly differed in the GDM cohort (p < .05). We also evaluated the predictive capacity of each heart function parameter for adverse pregnancy outcomes in GDM. The LMPI showed the strongest ability to predict adverse pregnancy outcomes with an AUC of 0.951 (95% CI: 0.909–0.993). A nomogram constructed with the 3 key predictors selected by LASSO regression (LMPI, LKI, and RIRT) demonstrated excellent discrimination, with an AUC of 0.960 (95% CI: 0.924–0.996). The model was well calibrated, and DCA indicated clinical utility.

Conclusion

Even under well‐controlled glycemic conditions, fetal cardiac function is altered in GDM pregnancies. In this cohort, a model based on fetal cardiac function parameters showed good predictive performance for composite adverse perinatal outcomes. However, external validation is required before clinical implementation.

Abbreviations

AC

abdominal circumference

AGA

appropriate for gestational age

BPD

biparietal diameter

CI

confidence interval

CPR

cerebroplacental ratio

DCA

decision curve analysis

EPV

events per variable

ET

ejection time

FL

femur length

FT

filling time

GDM

gestational diabetes mellitus

HC

head circumference

IADPSG

International Association of Diabetes and Pregnancy Study Groups

ICC

intraclass correlation coefficient

ICT

isovolumic contraction time

IDF

International Diabetes Federation

IGFBP

insulin‐like growth factor‐binding protease

IQR

interquartile range

IRT

isovolumic relaxation time

ISUOG

International Society of Ultrasound in Obstetrics and Gynecology

KI

K index

LASSO

Least Absolute Shrinkage and Selection Operator

LGA

large‐for‐gestational age

LP

linear predictor

MCA‐PI

middle cerebral artery pulsatility index

MPI

myocardial perfusion imaging

NICU

neonatal intensive care unit

OR

odds ratios

PAPP‐A

pregnancy‐associated plasma protein‐A

PSM

propensity score matching

PW

pulse wave

RDS

respiratory distress syndrome

ROC

receiver operating characteristic

SGA

small‐for‐gestational age

SMD

standardized mean difference

STE

speckle tracking echocardiography

UA‐PI

umbilical artery pulsatility index

VIF

variance inflation factor

Gestational diabetes mellitus (GDM) is a common endocrine disease during pregnancy, which is associated with poor neonatal outcomes. Based on the definition of International Association of Diabetes and Pregnancy Study Groups (IADPSG) criteria, the International Diabetes Federation (IDF) estimates that 14% of all newborns worldwide suffer from gestational diabetes. ¹ Gestational diabetes may cause fetal developmental abnormalities, metabolic disturbance, and long‐term damage, such as large‐for‐gestational age (LGA), small‐for‐gestational age (SGA), premature delivery, and fetal hypoglycemia. Previous study found that pregnant women with GDM exhibited changes in fetal cardiac structure and function. ² However, the study on GDM cases with optimal glycemic control is few and limited.

GDM usually develops due to abnormal β‐cell activity with persistent insulin resistance in whole pregnancy. β‐cell is responsible for sensing the blood glucose level and secreting insulin. ³ High maternal glucose environment in pregnancy would cause excessive stimulation of pancreatic β‐cell and increased insulin secretion. ⁴ High levels of insulin could promote the proliferation and hypertrophy of cardiomyocyte by stimulating the protein and fat synthesis, which would finally result in fetal cardiac hypertrophy. ⁴ In addition to morphological changes, fetal echocardiography also found that maternal diabetes would cause impairment of fetal cardiac performance in terms of fetal myocardial systolic and diastolic functions. ⁵ , ⁶ The fetal myocardial perfusion imaging (MPI) is an indicator of global cardiac function, and the KI is an indicator of ventricular diastolic function. ⁷ To our knowledge, there are few reports to investigate the fetal diastolic function in expectant mothers with GDM by using KI. M‐mode ultrasound imaging, pulse wave Doppler (PW), speckle tracking echocardiography (STE), and MPI are often used to evaluate fetal cardiac function. MPI can be measured from the blood flow spectrum of the atrioventricular valve inflow and outflow tracts by PW. The HOLO‐PW technology has 2 sampling lines and can display 2 Doppler waveforms at different positions simultaneously. This technology greatly reduced the measurement error of measuring different cardiac cycles and improved the accuracy of cardiac function assessment. However, there are few reports to measure MPI in fetuses utilizing this technique.

This study aimed to investigate the effects of GDM with optimal glycemic control on fetal cardiac function by using the HOLO‐PW. We also wanted to explore the changes in fetal cardiac function induced by adverse pregnancy outcomes in GDM and establish a multi‐index predictive model for poor neonatal outcomes in GDM with well‐managed blood glucose levels.

Methods

Study Design and Population

This prospective case–control study was carried out in Beijing Obstetrics and Gynecology Hospital, Capital Medical University, China from August 2024 to September 2025. The diagnosis of GDM was made at 24–28 weeks' gestation. Well‐controlled GDM was defined according to the IADPSG criteria (75 g oral glucose tolerance test with thresholds: fasting ≥5.1 mmol/L, 1‐hour ≥10.0 mmol/L, 2‐hour ≥8.5 mmol/L). Glycemic control was achieved through dietary modification (78% of GDM participants) or insulin therapy (22%) if dietary targets were not met. Good glycemic control was defined as 70% or greater of reported normal blood sugar levels (FBG <5.3 mmol/L and 1‐h PBG <7.8 mmol/L, 2‐h PBG <6.7 mmol/L). Inclusion criteria included singleton pregnancies in women aged 18–35 years with clinically diagnosed GDM and fasting blood glucose <5.3 mmol/L after dietary restriction or insulin therapy between 28⁺⁰–40⁺⁰ weeks of gestation. The exclusion criteria were multiple births, fetal structural abnormalities, and mental illness that limited maternal involvement in the study. The control group comprised randomly selected gravida who were not clinically diagnosed with GDM. The gestational weeks of both groups were matched, with the 2 groups of pregnant women studied in a 1:2 ratio.

Maternal and Fetal Characteristics

Maternal characteristics, including age, obstetric history, blood pressure, parity, laboratory tests, and pre‐pregnancy body mass index, were recorded. Additional information regarding pregnancy outcomes, such as adverse outcomes, gestational age at delivery, and birth weight, was obtained from maternal hospital records. Fetal growth was evaluated and documented in accordance with the International Society of Ultrasound in Obstetrics and Gynecology (ISUOG) practice guidelines. ⁸

Ultrasound Examination

All echocardiographic examinations were conducted by the same echocardiographer (L.Y.) under the supervision of a highly experienced expert in Obstetrics and Gynecology Ultrasound (Q.W.). A Wisonic Lotus ultrasound machine (Wisonic Medical Systems, Shenzhen, China) equipped with a 2–5 MHz convex probe and the HOLO‐PW dedicated software package was utilized. Pulsed Doppler ultrasound was employed to measure the middle cerebral artery pulsatility index (MCA‐PI) and umbilical artery pulsatility index (UA‐PI), from which the cerebroplacental ratio (CPR, MCA‐PI/UA‐PI) was calculated. Additionally, the most commonly used fetal biometric parameters—biparietal diameter (BPD), head circumference (HC), abdominal circumference (AC), and femur length (FL)—were measured to evaluate fetal growth.

Fetal echocardiography was performed with HOLO‐PW. The Doppler cursor was placed in the 2~4 mm range and the angle of insonation <30° for the measurement. For the assessment of the left ventricular function, a clear view of the fetal 5‐chamber heart view including mitral and aortic valve was required. The HOLO‐PW software was initiated, placing a Doppler sample gate in the left ventricular inflow tract and another 1 in the left ventricular outflow tract. For the assessment of the right ventricular function a short axis view of the fetal aorta should be obtained. Followed by the initiation of the HOLO‐PW software and placement of Doppler sampling gates in the right ventricular inflow tract and outflow tract. All time intervals were measured in seconds (s). Isovolumic contraction time (ICT), isovolumic relaxation time (IRT), ejection time (ET), and filling time (FT) of 2 ventricles were recorded. MPI and KI were calculated according to the following formulas: MPI = (ICT + IRT)/ET, KI = (ICT + IRT)/FT. Evaluation of ventricular diastolic function was performed by measuring E/A ratio (early filling and active filling) waves of the mitral and tricuspid valves by measuring the peak filling velocities of the E (early filling) and A (active filling) waves of the mitral and tricuspid valves. Each subject was measured 3 times and the mean value of each parameter was used for the analysis (Figure 1).

Figure 1.

Ultrasound measurement of cardiac function parameters. a, Filling time (FT); b, isovolumic contraction time (ICT); c, ejection time (ET); d, isovolumic relaxation time (IRT).

Statistical Analysis

Data analysis was performed using SPSS version 22.0 software (SPSS Inc., Chicago, IL, USA) and the R programing language (http://www.Rproject.org). Continuous variables with normally distribution were expressed as mean ± SD and group comparison using independent sample t‐test. Continuous variables with non‐normally distribution were expressed as median (interquartile range) [M (IQR)] and group comparison using Mann–Whitney U‐test. Categorical variables were expressed as frequencies (percentages) [n (%)] and the count data using chi‐square test or Fisher's exact test. To further address potential confounding due to baseline imbalance, we conducted a sensitivity analysis using propensity score matching (PSM). Propensity scores were estimated using logistic regression including pre‐pregnancy BMI, maternal age, gestational age at scan, systolic and diastolic blood pressure, CPR, gravidity, parity, and cholestasis. We then performed nearest‐neighbor matching with a caliper width of 0.2 standard deviations of the logit of the propensity score, allowing up to 2 controls per GDM case (without replacement). Covariate balance was assessed using standardized mean differences (SMD); values close to 0 indicate good balance. The correlation analysis of fetal cardiac function parameters and adverse perinatal outcome were analyzed by univariate and multivariate logistic regression analysis and showed as odds ratios (OR) and 95% confidence intervals (CI).

A prediction model for adverse perinatal outcomes within the GDM cohort was developed using the following rigorous approach. First, multicollinearity among candidate predictors was assessed using variance inflation factor (VIF) analysis, with VIF > 10 indicating severe collinearity, and Pearson correlation coefficients. To address overfitting given the modest event count, variable selection was performed using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression implemented. Ten‐fold cross‐validation was used to determine the optimal regularization parameter (λ), with the λ.1se (1 standard error rule) chosen to favor a more parsimonious and generalizable model. All continuous predictors were standardized to z‐scores before modeling. Internal validation was performed using bootstrap resampling with 500 replicates to estimate optimism and calculate bias‐corrected performance metrics. The shrinkage factor (calibration slope) was derived from bootstrap validation to assess and correct for overfitting.

The discriminatory ability of ultrasound parameters and predictive models for adverse perinatal outcome in GDM patients was analyzed by building receiver operating characteristic (ROC) curves and calculating the C‐statistic (area under the receiver operating characteristic curve, AUC) with 95% CI. Independent predictors were selected by multivariate regression analysis and by clinical significance. Finally, a nomogram prediction model was established by R programing language. A nomogram was constructed based on the final model coefficients. There were no missing data for variables included in the model. The comparison between the nomogram's predicted probability and observed probability was made by a Calibration curve, and intercept and slope were calculated for evaluation. The Prediction model's clinical Net Benefit was assessed by decision curve analysis (DCA).

Intra‐ and inter‐observer reproducibility for all Doppler‐derived indices (ICT, IRT, ET, FT) and calculated indices (MPI, KI, E/A) were assessed in a random subset of 20 participants. Intraclass correlation coefficients (ICC) and Bland–Altman limits of agreement were calculated. All ultrasound operators were blinded to the maternal GDM status and subsequent pregnancy outcomes during image acquisition and offline analysis. Statistical significance was set at p < .05.

Results

Patient Characteristics

This study included a total of 378 pregnant women; 122 were patients with GDM and 256 pregnant women with normal blood glucose levels. Clinical demographic data, hematological results and adverse perinatal outcomes are shown in Table 1 in all patients. Maternal age, pre‐pregnancy BMI, gravidity, systolic pressure, cholestasis and CPR differed between the GDM and control groups (p < .05). And there were no significant differences between the 2 cohorts in terms of maternal parity, diastolic pressure, GA at scan and FHR (p > .05). In addition, GDM patients showed significantly higher levels of UA, TG, LDL, WBC, HGB, and PLT when compared to the control cohort (p < .05). When all fetuses were divided into appropriate for gestational age (AGA), SGA, and LGA groups according to biometry, it was determined that GDM consisted of 73 cases (60.3%) AGA, 28 cases (23%) LGA and 20 cases (16.4%) SGA, while control group included 193 cases (75.4%) AGA, 34 cases (13.3%) LGA and 29 cases (11.3%) SGA. AGA and LGA showed significant differences between the 2 groups (p < .05) while SGA did not show any significant difference (p > .05). After PSM, 94 participants with GDM were matched to 132 controls (matched cohort n = 226). Baseline comparability improved substantially, with marked reductions in SMDs for major confounders, including BMI (0.78–0.139), maternal age (0.398–0.092), systolic blood pressure (0.283–0.093), and gravidity (0.292–0.001). CPR imbalance was also reduced (0.653–0.207) and was no longer statistically different between groups after matching (p = .131), indicating improved group comparability overall (Table S1).

Table 1.

Clinical Demographic Data, Hematological Results, and Adverse Perinatal Outcomes of 378 Participants

	Total (n = 378)	Controls (n = 256)	GDM (n = 122)	p‐Value
Maternal age (years), mean ± SD	31.36 ± 3.04	30.96 ± 2.86	32.18 ± 3.24	<.001
Pre‐pregnancy BMI (kg/m2), mean ± SD	22.49 ± 3.24	21.69 ± 2.76	24.16 ± 3.53	<.001
Gravidity, n (%)				.007
Primigravida	261 (69.05)	188 (73.44)	73 (59.84)
Multigravida	117 (30.95)	68 (26.56)	49 (40.16)
Parity, n (%)				.069
Primipara	330 (87.30)	229 (89.45)	101 (82.79)
Multipara	48 (12.70)	27 (10.55)	21 (17.21)
Systolic pressure (mmHg), mean ± SD	119.93 ± 9.38	119.03 ± 8.18	121.82 ± 11.29	.007
Diastolic pressure (mmHg), mean ± SD	72.89 ± 8.16	72.88 ± 7.74	72.91 ± 9.01	.976
Cholestasis, n (%)				.038
No	371 (98.15)	254 (99.22)	117 (95.9)
Yes	7 (1.85)	2 (0.78)	5 (4.1)
Maternal UA (μmol/L), mean ± SD	222.55 ± 54.56	211.75 ± 52.82	245.19 ± 51.30	<.001
Maternal CRE (μmol/L), mean ± SD	48.04 ± 5.84	48.31 ± 5.82	47.46 ± 5.86	.188
Maternal TC (mmol/L), mean ± SD	4.54 ± 2.30	4.59 ± 2.76	4.42 ± 0.61	.52
Maternal TG (mmol/L), mean ± SD	1.12 ± 0.49	1.05 ± 0.49	1.27 ± 0.43	<.001
Maternal HDL (mmol/L), mean ± SD	1.47 ± 0.28	1.50 ± 0.27	1.40 ± 0.27	<.001
Maternal LDL (mmol/L), mean ± SD	2.33 ± 0.66	2.28 ± 0.74	2.43 ± 0.44	.039
Maternal TBIL (μmol/L), mean ± SD	12.60 ± 7.11	13.06 ± 8.05	11.62 ± 4.42	.064
Maternal DBIL (μmol/L), mean ± SD	4.17 ± 1.53	4.35 ± 1.69	3.79 ± 1.03	<.001
Maternal WBC count (×109/L), mean ± SD	8.60 ± 4.05	8.31 ± 4.74	9.21 ± 1.80	.044
Maternal RBC count (×1012/L), maternal, mean ± SD	4.57 ± 2.92	4.65 ± 3.54	4.40 ± 0.34	.43
Maternal HGB(g/L), mean ± SD	129.65 ± 8.92	128.78 ± 8.87	131.49 ± 8.78	.006
Maternal PLT count (×109/L), mean ± SD	253.40 ± 52.97	247.06 ± 51.71	266.70 ± 53.32	<.001
GA at scan (weeks), mean ± SD	32.4 ± 3.0	32.4 ± 3.0	32.4 ± 3.1	.955
FHR (bpm), mean ± SD	145.3 ± 7.2	145.5 ± 7.4	144.9 ± 6.6	.456
CPR, mean ± SD	1.9 ± 0.6	2.1 ± 0.6	1.7 ± 0.4	<.001
SGA, n (%)				.17
No	329 (87.0)	227 (88.7)	102 (83.6)
Yes	49 (13.0)	29 (11.3)	20 (16.4)
LGA, n (%)				.018
No	316 (83.6)	222 (86.7)	94 (77)
Yes	62 (16.4)	34 (13.3)	28 (23)
AGA, n (%)				.003
No	111 (29.4)	63 (24.6)	48 (39.7)
Yes	266 (70.6)	193 (75.4)	73 (60.3)
Adverse perinatal outcomes, n (%)				<.001
No	306 (81.0)	236 (92.2)	70 (57.4)
Yes	72 (19.0)	20 (7.8)	52 (42.6)
Premature delivery	8 (11.1)	3 (15.0)	5 (9.6)
Neonatal hypoglycemia	24 (33.3)	8 (40.0)	16 (30.8)
Umbilical cord arterial pH<7.1	11 (15.3)	4 (20.0)	7 (13.5)
NICU admission	22 (30.6)	3 (15.0)	19 (36.5)
Respiratory distress syndrome	7 (9.7)	2 (10.0)	5 (9.6)

Data are given as mean ± SD, median (25% quantile, 75% quantile), or n (%). For the specific components of adverse perinatal outcomes (premature delivery, neonatal hypoglycemia, etc.), the percentages are calculated based on the number of participants who experienced the composite adverse outcome in each group (total: n = 72; control: n = 20; GDM: n = 52).

BMI, body mass index; FHR, fetal heart rate; SGA, small for gestational age; LGA, large for gestational age; AGA, appropriate for gestational age; NICU, neonatal intensive care unit; UA, urine acid; CRE, creatinine; TC, total cholesterol; TG, triglyceride; HDL, high density lipoprotein; LDL, Low density lipoprotein; TBIL, total bilirubin; DBIL, direct bilirubin; WBC, white blood cell; RBC, red blood cell; HGB, hemoglobin; PLT, blood platelet.

Fetal Cardiac Function Parameters

Table 2 shows fetal cardiac function test results in GDM and control groups. Compared to the control group, the GDM group showed an increase in MPI, ICT, IRT, and KI for both left and right parameters (p < .05) and a decrease in LET, LFT, RET, and RFT (p < .05). Additionally, the E/A ratio of tricuspid and mitral diastolic parameters showed a significant decrease in the GDM group (p < .05). In the matched cohort, between‐group comparisons of fetal cardiac function parameters remained consistent with the primary analysis. The GDM group continued to show higher MPI, ICT, IRT, and KI indices and lower ET, FT, and E/A ratios compared with controls, and most between‐group differences remained statistically significant. Notably, RET showed attenuation and was no longer statistically significant after matching (p = .071), whereas the overall pattern and direction of effects remained unchanged (Table S2, Figure S1).

Table 2.

Cardiac Function Parameters of GDM Group and Control Group

	Total (n = 378)	Controls (n = 256)	GDM (n = 122)	p‐Value
LMPI, mean ± SD	0.392 ± 0.079	0.357 ± 0.061	0.465 ± 0.061	<.001
LICT (s), mean ± SD	0.030 ± 0.008	0.027 ± 0.007	0.036 ± 0.007	<.001
LIRT (s), mean ± SD	0.039 ± 0.024	0.036 ± 0.008	0.047 ± 0.041	<.001
LET (s), mean ± SD	0.175 ± 0.011	0.177 ± 0.012	0.170 ± 0.009	<.001
LFT (s), mean ± SD	0.173 ± 0.015	0.175 ± 0.016	0.168 ± 0.013	<.001
LKI, mean ± SD	0.398 ± 0.084	0.362 ± 0.064	0.473 ± 0.069	<.001
T E/A, mean ± SD	0.593 ± 0.128	0.654 ± 0.087	0.466 ± 0.104	<.001
RMPI, mean ± SD	0.417 ± 0.086	0.386 ± 0.072	0.481 ± 0.079	<.001
RICT (s), mean ± SD	0.033 ± 0.009	0.030 ± 0.008	0.038 ± 0.009	<.001
RIRT (s), mean ± SD	0.040 ± 0.007	0.038 ± 0.007	0.045 ± 0.007	<.001
RET (s), mean ± SD	0.177 ± 0.012	0.179 ± 0.011	0.174 ± 0.012	<.001
RFT (s), mean ± SD	0.171 ± 0.014	0.173 ± 0.013	0.165 ± 0.015	<.001
RKI, mean ± SD	0.435 ± 0.097	0.399 ± 0.075	0.510 ± 0.096	<.001
M E/A, mean ± SD	0.590 ± 0.137	0.665 ± 0.081	0.432 ± 0.088	<.001

Table 3 displays subgroup analyses comparing cardiac function tests for adverse versus normal pregnancy outcomes among the GDM cohort. LMPI, LICT, LFT, LKI, RMPI, RICT, RIRT, RFT, and RKI exhibited significant differences between the groups (p < .05).

Table 3.

Comparison of Cardiac Function Parameters Between Normal and Adverse Perinatal Outcomes in GDM Group

	Total (n = 122)	Normal Perinatal Outcomes (GDM) (n = 70)	Adverse Perinatal Outcomes (GDM) (n = 52)	p‐Value
LMPI, mean ± SD	0.465 ± 0.061	0.426 ± 0.048	0.517 ± 0.033	<.001
LICT (s), mean ± SD	0.036 ± 0.007	0.032 ± 0.006	0.041 ± 0.004	<.001
LIRT (s), mean ± SD	0.047 ± 0.041	0.041 ± 0.005	0.054 ± 0.062	.077
LET (s), mean ± SD	0.170 ± 0.009	0.171 ± 0.008	0.169 ± 0.009	.098
LFT (s), mean ± SD	0.168 ± 0.013	0.170 ± 0.013	0.165 ± 0.011	.048
LKI, mean ± SD	0.473 ± 0.069	0.434 ± 0.060	0.527 ± 0.038	<.001
T E/A, mean ± SD	0.466 ± 0.104	0.464 ± 0.108	0.467 ± 0.100	.881
RMPI, mean ± SD	0.481 ± 0.079	0.433 ± 0.042	0.545 ± 0.072	<.001
RICT (s), mean ± SD	0.038 ± 0.009	0.034 ± 0.007	0.044 ± 0.009	<.001
RIRT (s), mean ± SD	0.045 ± 0.007	0.042 ± 0.005	0.048 ± 0.007	<.001
RET (s), mean ± SD	0.174 ± 0.012	0.175 ± 0.011	0.172 ± 0.012	.105
RFT (s), mean ± SD	0.165 ± 0.015	0.168 ± 0.012	0.160 ± 0.016	<.001
RKI, mean ± SD	0.510 ± 0.096	0.452 ± 0.048	0.589 ± 0.086	<.001
M E/A, mean ± SD	0.432 ± 0.088	0.430 ± 0.082	0.435 ± 0.095	.769

Table 4 presents the subgroup comparison findings of cardiac function tests for AGA, SGA, and LGA fetuses within the GDM cohort. Fetal LET, LFT, TE/A, and ME/A results exhibited no significant differences across the 3 subgroups. Nevertheless, AGA fetuses showed significantly lower values than the other subgroups for fetal LMPI, LICT, LIRT, LKI, RMPI, RICT, RIRT, and RKI (p < .05), while LET, LFT, RET and RFT values were higher compared to the other subgroups (p < .05).

Table 4.

Comparison of Cardiac Function Parameters Between Fetuses with Abnormal Growth and Normal Growth in the GDM Group

	Total (n = 122)	AGA (n = 74)	SGA (n = 20)	LGA (n = 28)	p‐Value
LMPI, mean ± SD	0.465 ± 0.061	0.437 ± 0.056	0.510 ± 0.045	0.507 ± 0.040	<.001
LICT (s), mean ± SD	0.036 ± 0.007	0.033 ± 0.007	0.040 ± 0.006	0.040 ± 0.005	<.001
LIRT (s), mean ± SD	0.047 ± 0.041	0.041 ± 0.005	0.046 ± 0.006	0.061 ± 0.084	.095
LET (s), mean ± SD	0.170 ± 0.009	0.172 ± 0.008	0.169 ± 0.007	0.168 ± 0.010	.09
LFT (s), mean ± SD	0.168 ± 0.013	0.169 ± 0.013	0.164 ± 0.014	0.166 ± 0.010	.178
LKI, mean ± SD	0.473 ± 0.069	0.445 ± 0.063	0.525 ± 0.061	0.512 ± 0.050	<.001
T E/A, mean ± SD	0.466 ± 0.104	0.482 ± 0.104	0.448 ± 0.098	0.437 ± 0.104	.108
RMPI, mean ± SD	0.481 ± 0.079	0.444 ± 0.052	0.539 ± 0.103	0.536 ± 0.062	<.001
RICT (s), mean ± SD	0.038 ± 0.009	0.035 ± 0.009	0.044 ± 0.008	0.043 ± 0.009	<.001
RIRT (s), mean ± SD	0.045 ± 0.007	0.043 ± 0.006	0.047 ± 0.008	0.047 ± 0.006	.014
RET (s), mean ± SD	0.174 ± 0.012	0.177 ± 0.010	0.170 ± 0.013	0.170 ± 0.012	.004
RFT (s), mean ± SD	0.165 ± 0.015	0.169 ± 0.013	0.161 ± 0.011	0.156 ± 0.016	<.001
RKI, mean ± SD	0.510 ± 0.096	0.467 ± 0.072	0.566 ± 0.097	0.585 ± 0.083	<.001
M E/A, mean ± SD	0.432 ± 0.088	0.440 ± 0.092	0.432 ± 0.094	0.411 ± 0.071	.335

Adverse Perinatal Outcomes

The primary outcome was a composite of adverse perinatal events, defined as the occurrence of any of the following: (1) Preterm delivery (<37 weeks' gestation); (2) Neonatal hypoglycemia (plasma glucose <2.2 mmol/L within 24 hours after birth requiring intervention); (3) Umbilical artery cord pH <7.10 at delivery; (4) Admission to the neonatal intensive care unit (NICU) for >24 hours for any reason other than routine observation; (5) respiratory distress syndrome (RDS) diagnosed by clinical signs and requiring respiratory support. The incidence of these adverse outcomes was significantly higher in the GDM cohort than in the control group (p < .05, Table 1). Among the participants who experienced the composite outcome, the distribution of each component in the GDM group (n = 52) versus the control group (n = 20) was as follows: premature delivery (5 cases [9.6%] versus 3 cases [15.0%]), neonatal hypoglycemia (16 cases [30.8%] versus 8 cases [40.0%]), umbilical cord arterial pH <7.1 (7 cases [13.5%] versus 4 cases [20.0%]), NICU admission (19 cases [36.5%] versus 3 cases [15.0%]), and respiratory distress syndrome (5 cases [9.6%] versus 2 cases [10.0%]). Within the GDM cohort (among the 52 patients with the composite outcome), the distribution of these components was: neonatal hypoglycemia (16 cases, 30.8% of GDM adverse outcomes), NICU admission (19 cases, 36.5%), premature delivery (5 cases, 9.6%), respiratory distress syndrome (5 cases, 9.6%), and umbilical cord arterial pH <7.1 (7 cases, 13.5%).

The predictive performance for adverse outcomes in GDM was evaluated for each ultrasound parameter, with the left and right MPI, ICT, IRT, and KI demonstrating the most elevated AUC values (Figure 2).

Figure 2.

ROC for predicting adverse pregnancy outcomes using various cardiac function parameters.

Model Performance Metrics with Internal Validation

Among the 122 GDM pregnancies included in the analysis, 52 (42.6%) experienced adverse perinatal outcomes. With 3 predictors in the final model, the events per variable (EPV) were 17.3, exceeding the recommended minimum of 10. Assessment of multicollinearity revealed substantial collinearity among cardiac function parameters, with RKI (VIF = 124.7) and RMPI (VIF = 119.8) showing severe collinearity. High correlations were observed between LMPI‐LKI (r = 0.87), RMPI‐RKI (r = 0.92), and RMPI‐RICT (r = 0.74) (Figure 3). LASSO regression with 10‐fold cross‐validation was employed for variable selection to address overfitting and multicollinearity. The analysis selected 3 variables at the λ.1se (0.098) point, favoring parsimony and stability: LMPI, LKI, RIRT (Figure 4, A and B). As shown in Table 5, after adjustment for other variables, LMPI was the strongest predictor of adverse outcome, with each standard deviation increase associated with an approximately 8.73‐fold increase in risk. RIRT also showed a statistically significant positive association with risk (OR 2.35, 95% CI: 1.45–5.62). Although LKI was retained in the final model by LASSO regularization, its confidence interval crossed unity (OR 2.52, 95% CI: 0.67–37.89), indicating a less precise estimate. This suggests that while LKI contributes to the overall predictive performance of the model, its independent effect should be interpreted with caution. The final model demonstrated excellent discrimination with an apparent C‐statistic (AUC) of 0.960 (95% CI: 0.924–0.996) (Figure 5). Bootstrap internal validation (B = 500 replicates) revealed minimal optimism (0.008), yielding a bias‐corrected C‐statistic of 0.952. The model showed good calibration with a calibration slope of 1.044 and calibration‐in‐the‐large (intercept) of −0.018 (ideal values: slope = 1, intercept = 0). The Brier score was 0.062, indicating good overall accuracy (Table S3). The model equation is as follows: linear predictor (LP) = −0.765 + 2.167 × LMPI _z  + 0.923 × LKI _z  + 0.856 × RIRT _z .

$$\text{Predicted Probability}=1/\left(1+{\mathrm{e}}^{\left(-\mathrm{LP}\right)}\right)$$

where, subscript z indicates z‐score standardized values [(value − mean)/SD].

Figure 3.

Correlation matrix of cardiac function parameters.

Figure 4.

Lasso coefficient path (A); Lasso cross‐validation performance (B).

Table 5.

Final Prediction Model Coefficients and Odds Ratios

Variable	Coefficient (β)	OR	95% CI
Intercept	−0.765	—	—
LMPI (per 1 SD)	2.167	8.73	2.38–67.35
LKI (per 1 SD)	0.923	2.52	0.67–37.89
RIRT (per 1 SD)	0.856	2.35	1.45–5.62

LKI was retained in the final model by LASSO regularization; its wide confidence interval indicates less precision and its independent effect should be interpreted cautiously.

OR, odds ratio; CI, confidence interval; SD, standard deviation; LMPI, left ventricular myocardial performance index; LKI, left ventricular Tei index; RIRT, right ventricular isovolumic relaxation time.

Figure 5.

ROC curve for adverse perinatal outcomes prediction.

A nomogram was constructed based on the final model coefficients to facilitate clinical visualization of the prediction model (Figure 6). The calibration curve demonstrates a strong agreement between the predicted probabilities in the nomogram and the observed probabilities (Figure 7). DCA demonstrated that the model provided net clinical benefit across a wide range of threshold probabilities, supporting its potential clinical utility (Figure 8).

Figure 6.

Nomogram prediction of adverse pregnancy outcomes in well‐controlled GDM.

Figure 7.

Calibration curve for nomogram of adverse perinatal outcomes in well‐controlled GDM.

Figure 8.

Decision curve for nomogram of adverse perinatal outcomes in well‐controlled GDM.

Intra‐ and Inter‐Observer Reproducibility

The reproducibility of HOLO‐PW‐derived fetal cardiac function measurements was excellent. For intra‐observer reliability, the ICCs ranged from 0.934 to 0.979 for all key parameters. Inter‐observer reliability was also high, with ICCs ranging from 0.886 to 0.938 (Table 6). Bland–Altman analysis confirmed good agreement, with narrow limits of agreement and minimal systematic bias (Figure S2).

Table 6.

Intra‐ and Inter‐Observer Reliability for Measuring the LMPI, LICT, LIRT, LET, LFT, LKI, T E/A, RMPI, RICT, RIRT, RET, RFT, RKI, M E/A, Described with Intraclass Correlation Coefficients (ICCs), and 95% Confidence Intervals (CI)

Measurement	ICC (95% CI)
	Intra‐Observer	Inter‐Observer
LMPI	0.964 (0.904–0.992)	0.926 (0.870–0.969)
LICT, s	0.972 (0.932–1.000)	0.935 (0.874–0.972)
LIRT, s	0.969 (0.901–0.993)	0.907 (0.775–0.949)
LET, s	0.958 (0.882–0.974)	0.911 (0.815–0.954)
LFT,s	0.947 (0.874–0.968)	0.919 (0.926–0.962)
LKI	0.973 (0.916–0.991)	0.914 (0.813–0.947)
T E/A	0.941 (0.863–0.982)	0.892 (0.735–0.945)
RMPI	0.965 (0.897–0.989)	0.907 (0.796–0.964)
RICT, s	0.979 (0.921–0.994)	0.923 (0.864–0.962)
RIRT, s	0.972 (0.913–0.998)	0.938 (0.852–0.981)
RET, s	0.952 (0.878–0.981)	0.908 (0.783–0.963)
RFT, s	0.953 (0.884–0.986)	0.911 (0.825–0.965)
RKI	0.966 (0.898–0.987)	0.903 (0.768–0.946)
M E/A	0.934 (0.846–0.977)	0.886 (0.713–0.928)

Discussion

This study, fetal cardiac function is examined by ultrasound in women with well‐controlled GDM and found an important finding that the adverse effects of GDM on fetal cardiac function still existed even when maternal blood glucose was maintained within the clinically acceptable range. In our study, the main manifestations were elevated MPI and KI, prolonged ICT and IRT, shortened ET and FT, and decreased T E/A and M E/A. These results were in line with the results of previous studies. ⁹ , ¹⁰ , ¹¹ The common belief is that strict blood sugar control can prevent the adverse effects of GDM on the fetus. However, our results are consistent with the conclusions of recent studies that challenge this idea. ¹² , ¹³ Even when fasting blood glucose is maintained within the normal range, problems may still exist in the maternal environment in some cases associated with GDM. In a study by Ambrosi et al, ¹² fetal diastolic function was examined in GDM cases with well‐controlled glycemic levels. The RFT was significantly shorter in the GDM group than in the control group (p < .05). Although the LFT was also shorter in the GDM group than in the control group, the variance was not statistically significant. Ma et al ¹³ found fetal right ventricular systolic dysfunction in GDM cases with good glycemic control. Our study is more perfect. We examined both systolic and diastolic ventricular function and observed the changes in fetal systolic and diastolic function in diabetic women with well‐controlled glycemia. Importantly, to address potential confounding from baseline imbalances, we performed a rigorous sensitivity analysis using propensity score matching. The persistence of significant fetal cardiac function alterations in the well‐balanced matched cohort greatly strengthens the inference that these changes are independently associated with GDM status itself, rather than being driven by differences in maternal age, BMI, blood pressure, or obstetric history.

Pregnant women with diabetes, although treated, may have more intense blood glucose fluctuations in amplitude and duration than healthy pregnant women. This prolonged hyperglycemia can cause endothelial dysfunction, oxidative stress, and inflammatory dysregulation through the placenta, directly damaging the mitochondrial function and calcium ion handling ability of myocardial cells, leading to cardiac systolic and diastolic dysfunction. ¹⁴ Fetal hyperinsulinemia and insulin‐like growth factor‐I can induce cardiomyocyte hypertrophy and lead to reduced cardiac function. ¹⁵

Moreover, in our study, 28 cases of LGA (22.95%) and 20 cases of SGA (16.39%) were also observed in the GDM group. This may be due to the fact that the fetus was exposed to hyperinsulinemia for a long time. Hyperglycemia could cause the fetus to present with protein and fat synthesis enhancement and fat decomposition inhibition, which would make the fetus gain excessive weight and growth. ¹⁵ , ¹⁶ In addition, pregnancy‐associated plasma protein‐A (PAPP‐A) could function as insulin‐like growth factor‐binding protease (IGFBP). ¹⁷ When the content of PAPP‐A was low, it would cause the dysfunction of the placenta, contributing to SGA. ¹⁸ , ¹⁹ , ²⁰ Our results indicated that cardiac function parameters of LGA and SGA differ from those of AGA. LGA may be due to an abnormal fetal metabolic state with myocardial metabolic disorder and hypertrophy, which can depress diastolic function and increase MPI. SGA may be due to increased fetal cardiac afterload and a compensatory cardiac response, which also leads to increased MPI.

We also investigated cardiac function of GDM pregnant women in relation to a negative pregnancy outcome. The systolic and diastolic functions of poor pregnant women in GDM differed significantly from those of the good ones (p < .05). We used cardiac function parameters to develop ROC curves for predicting adverse pregnancy outcomes in GDM pregnant women. To directly address concerns regarding overfitting and collinearity, we employed LASSO regression for variable selection and conducted bootstrap internal validation. This rigorous approach yielded a more parsimonious and stable model. A nomogram was created to visualize the association of adverse fetal outcomes in pregnant women with gestational diabetes. We developed an internally assessed multivariable nomogram that stratified the risk of a composite adverse perinatal outcome within the GDM cohort. Given the modest event count and lack of external validation, the model should be considered exploratory and hypothesis‐generating rather than a clinically deployable tool. The results obtained using the ROC curves yielded an AUC value of 0.960 (95% CI: 0.924–0.996), which provides better predictive power than individual indices. This finding, along with the good calibration and net benefit shown in DCA, underscores the model's potential for risk stratification within similar populations. Our study has 2 main innovations. Firstly, we used the HOLO‐PW method to show 2 Doppler waveforms in 1 cardiac cycle. Measurement error resulting from using different cardiac cycles was minimized. Fetal motion may cause measurement error. In the fetal circulation, the right heart is dominant over the left and accounts for 60% to 70% of the total output. ²¹ However, assessments of right ventricular function are particularly vulnerable to interference, as they cannot capture both outflow and inflow spectra within the same cardiac cycle. Consequently, this technique proves to be more effective for accurately evaluating right heart function. To our knowledge, this study is the first to utilize the HOLO‐PW technique to assess fetal cardiac function in cases of GDM. Second, fetal diastolic function is of particular importance, and we employed a parameter that has not been widely utilized in clinical settings to evaluate diastolic function. Ahsan et al ⁷ constructed a new index to evaluate ventricular diastolic function, and they called it KI. They studied 57 normal fetuses and 14 fetuses with congenital heart disease. They found that the KI value was significantly reduced in the latter group. In our study, we used 256 normoglycemic fetuses and 122 fetuses with GDM. Our results showed that left and right KI values were significantly increased in the GDM group when compared with the normal group. These results were associated with the increase in ICT and IRT and the decrease in FT. However, KI was not used in clinical practice at present. Much more clinical experiences are needed to validate it.

The main limitations of this study include its single‐center design and modest sample size. Despite these limitations, our internally validated nomogram—incorporating LMPI, LKI, and RIRT—demonstrated excellent discrimination (C‐statistic 0.960) and calibration for predicting adverse perinatal outcomes. However, it is important to note that while LASSO regression selected these 3 variables to optimize predictive performance and mitigate overfitting, the wide confidence interval for LKI (OR 2.52, 95% CI: 0.67–37.89) reflects uncertainty regarding its precise effect size. This underscores that the model's strength lies in the combination of these parameters, rather than the independent predictive power of each component. Consequently, the model should be considered exploratory and hypothesis‐generating, and external validation in independent, multicenter cohorts is required before clinical application. We will increase the number of cases of gestational diabetes in women with optimal glyceric control in multicenter studies. We will try to understand more about the fetal cardiac function parameters in women with optimal glyceric control. We will try to establish more reliable nomogram to predict adverse pregnancy outcomes in women with optimal glyceric control. We attempted to evaluate fetal cardiac function in a relaxed state; however, we could not standardize or control these variables completely. There were still some confounding variables. Although propensity score matching effectively balanced the observed covariates between groups, residual confounding from unmeasured factors (eg, glycemic variability, detailed dietary composition, physical activity levels, or genetic predisposition) cannot be ruled out. We evaluated fetal cardiac function only by conventional Doppler parameters. We did not try to use 2‐dimensional speckle tracking imaging to validate the cardiac function alterations.

Conclusions

For women with GDM who have good glycemic control, the fetal cardiac performance index can be a highly specific and reliable indicator of subclinical fetal cardiac insufficiency that may go undetected in routine management. This association was robust to adjustment for potential confounders using propensity score matching. In addition, the cardiac function parameters can reflect the participation of the fetal cardiac function in GDM and are also good predictors for perinatal poor outcomes. In this cohort, a multivariable model incorporating these parameters demonstrated good discriminative ability for predicting a composite adverse outcome. This model, visualized as a nomogram, represents a novel strategy for individualized risk assessment. However, its clinical utility remains to be established through external validation in independent populations. Future research should focus on validating these findings and exploring their integration into clinical management pathways to potentially improve perinatal prognosis.

Supporting information

Figure S1. Comparison of fetal cardiac function parameters between groups in the matched cohort.

Figure S2. Bland–Altman plots for intra‐ and inter‐observer reproducibility.

Table S1. Baseline characteristics before and after propensity score matching.

Table S2. Fetal cardiac function parameters in the propensity score‐matched cohort.

Table S3. Model Performance Metrics with Internal Validation.

Data Availability Statement

Research data are not shared.