Development and validation of a nomogram for predicting comorbid anxiety and depression in patients with gestational diabetes mellitus

Abstract

BACKGROUND

Gestational diabetes mellitus (GDM) is a common metabolic complication of pregnancy closely associated with elevated risks of maternal perinatal anxiety and depression. Psychological comorbidities worsen glycemic control and adversely affect maternal and fetal outcomes. Currently, standardized risk prediction tools for anxiety and depression among women with GDM remain scarce, and most existing prediction tools ignore easily accessible sociodemographic and psychosocial predictors. Therefore, a visual prediction tool is needed to screen high-risk populations, and age, miscarriage history, education, and social support are hypothesized as independent risk factors.

AIM

To validate a nomogram for predicting anxiety-depression risk in GDM patients and evaluate its discrimination, calibration, and net clinical benefit.

METHODS

This retrospective study retrospectively enrolled 238 GDM patients diagnosed at our hospital between February 2022 and January 2024, categorized into comorbid (38 cases) and control (200 cases) groups based on Self-Rating Anxiety Scale (SAS) and Self-Rating Depression Scale (SDS) scores. General data were collected using a custom-made questionnaire. SAS and SDS scores were analyzed and univariate analysis performed on general data of the groups. Multivariate logistic regression was used to identify risk factors, and a prediction model was developed accordingly.

RESULTS

The comorbidity group had significantly older age, higher rates of miscarriage history, lower education (junior high school or below), and poorer social support than controls. Multivariate regression identified age, miscarriage history, and low education and social support as independent risk factors for anxiety and depression in GDM patients. Receiver operating characteristic analysis yielded areas under the curve of 0.718, 0.635, 0.666, 0.667, and 0.853 for age, miscarriage history, low education, social support, and the combined predictor, respectively, with corresponding sensitivities of 0.579, 0.395, 0.553, 0.684, and 0.816. The model exhibited favorable screening efficacy for anxiety-depression in GDM patients. Bootstrap internal validation confirmed well-calibrated curves; decision curve analysis showed positive net clinical benefit across threshold probabilities of 0.02-0.93, superior to reference curves.

CONCLUSION

The identified risk factors independently elevate anxiety-depression risk in GDM patients. The constructed prediction model exhibits robust discrimination and calibration, enabling early identification of high-risk GDM patients for psychological intervention.

Key Words: Gestational diabetes mellitus; Anxiety; Depression; Risk prediction model; Model performance

Core Tip: This study explored the factors associated with combined anxiety and depression in pregnant women with gestational diabetes mellitus (GDM) and constructed a targeted risk prediction nomogram. The results emphasized that older maternal age, prior miscarriage, lower educational attainment, and limited social support were key contributors to psychological comorbidities. The prediction model showed good diagnostic performance. Therefore, it may serve as a practical tool for early detection and timely intervention of emotional disorders in GDM patients. Future research should aim for external validation and further refinement of this model to enhance its accuracy and clinical applicability.

INTRODUCTION

Many perinatal women with gestational diabetes mellitus (GDM) experience distress. GDM is one of the most common pregnancy metabolic complications and can affect the health of the mother and child during pregnancy. Recent studies have reported that GDM can lead to maternal psychological problems such as anxiety and depression[1,2]. Women with GDM face multiple challenges. Physiological changes caused by impaired glucose metabolism, blood sugar control lifestyle adjustment, and the stress of pregnancy render them vulnerable to persistent negative emotions. Emotional issues can influence perinatal health and perinatal outcomes[3].

Anxiety and depression are more common in women with GDM than in the normal general pregnant population. Anxiety and depression may lead to a poor quality of life and can affect fetal growth and long-term development[4,5]. Anxiety and depression screening and early treatment are thus important for the better health of the mother and fetus.

Managing these emotional disorders in GDM patients can be challenging. Stress and psychological changes during pregnancy can make it difficult for patients to express how they feel, which can delay a correct diagnosis. In addition, there are few specialized tools for predicting the risk of anxiety and depression in GDM patients. Without these tools, clinicians will have difficulty assessing individual risk before it becomes evident[6-8]. Existing predictive models mainly focus on glycemic and obstetric indicators while neglecting modifiable psychosocial factors, including social support and educational background.

To fill this clinical gap, this retrospective observational study enrolled hospitalized GDM patients; sociodemographic, obstetric, and laboratory data were collected; and independent risk factors were screened via logistic regression. We constructed a visual nomogram prediction model and comprehensively evaluated its predictive efficiency using receiver operating characteristic (ROC) curves, C-index, Brier score, calibration curves, and decision curve analysis (DCA). This study aimed to provide a simple, practical screening tool for antenatal care providers to identify high-risk GDM women and deliver targeted psychological intervention early.

MATERIALS AND METHODS

General information

In total, 238 patients diagnosed with GDM at our hospital between February 2022 and January 2024 were retrospectively enrolled in this study. Participants were categorized into a comorbidity group (38 cases) and a control group (200 cases) based on their Self-Rating Anxiety Scale (SAS) and Self-Rating Depression Scale (SDS) scores. The inclusion criteria for patients were as follows: (1) Met the diagnostic criteria of Guideline No. 393-Diabetes in Pregnancy[9]; (2) Registered and delivered at our hospital and consented to participate in the study; (3) Had singleton pregnancies; (4) Had SAS scores of ≥ 50 points and SDS scores of ≥ 53 points; and (5) Had complete clinical data.

The exclusion criteria were as follows: (1) Presence of severe infectious or immune system diseases; (2) Severe dysfunction of vital organs such as the heart, lungs, liver, or kidneys; (3) A history of diabetes prior to pregnancy; (4) Other pregnancy-related syndromes; and (5) Mental and cognitive disorders.

Sample size consideration

This study followed the events per variable (EPV) rule (10-15 events per predictor variable) proposed by Peduzzi et al[10]. The study included four candidate predictors with 38 positive outcome events, yielding an EPV of 9.5, which was slightly below the recommended minimum threshold. Bootstrap internal validation with 1000 resamples was adopted to partially correct potential overfitting bias; a small single-center sample size without external multicenter validation is explicitly listed as a major limitation of this research.

Data collection methods

Collection of general information: A hospital-designed questionnaire was used to collect general information on early pregnancy reactions, family history of diabetes, other pregnancy-related syndromes, education level, history of abortion, family monthly income, occupation type, and social support.

Laboratory index collection: All participants fasted overnight for 8-10 hours before venous blood collection. Fasting blood glucose (FBG), glycated hemoglobin (HbA1c), triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), uric acid (UA), creatinine (Cr), brain natriuretic peptide (BNP), and insulin resistance (IR) levels were detected by enzyme-linked immunosorbent assay following kit standard operating procedures [AbK (Shanghai) Trading Co., Ltd., China].

SAS: SAS was adopted to assess subjective anxiety and anxiety severity of the research participants[11]. Composed of 20 self-assessment items, this scale adopts a 4-point scoring standard, in which the scoring criteria are defined as follows: 1 point for “none of the time”, 2 points for “a little of the time”, 3 points for “a good deal of the time” and 4 points for “most of the time”. Notably, five specific items (items 5, 9, 13, 17, and 19) are scored in a reverse manner; that is, the scores are recorded in a sequence of 4 to 1 instead of the conventional 1 to 4. Clinically, an SAS score of no less than 50 points is regarded as a sign of anxiety symptoms, and the severity of anxiety is positively correlated with this score; higher scores indicate more severe anxiety symptoms.

SDS: SDS was used to evaluate the presence of depressive symptoms and their corresponding severity[12]. First developed by Zung[12] in 1965, this scale is a self-administered assessment tool consisting of 20 items and adopts a 4-point scoring system consistent with the SAS. An SDS score of 53 points or higher is considered to indicate the presence of depressive symptoms. Similar to the SAS, the score is positively associated with the severity of depression; higher scores represent more severe depressive symptoms.

Research methods

This study was conducted between February 2022 and July 2024. Three on-duty nurses with professional training were responsible for distributing the questionnaires and guiding the participants. Before handing out the questionnaires, the research objectives and specific details of the survey were explained to each patient and their family members. Questionnaires were distributed only after written consent was obtained from the patients and their families. During the completion of the surveys, the nurses addressed all questions raised by the participants to ensure that the information provided was accurate.

A total of 252 questionnaires were distributed, and 238 valid responses were collected, resulting in a response rate of 94.44%. After data collection, appropriate measures were taken to protect the privacy of participants and prevent information leakage.

Statistical analysis

Statistical analyses were performed using SPSS 22.0 and RStudio 4.2.1 software. Measurement data conforming to normal distribution were expressed as mean ± SD and compared via the independent samples t-test; count data were presented as n (%) and compared via the χ² test. Non-normally ranked data were analyzed using the rank-sum test. Univariate logistic regression was used to screen candidate predictors; variables with P < 0.05 were included in multivariate stepwise logistic regression to identify independent risk factors. The rms package in R was used to construct the nomogram, calculate C-index, and generate bias-corrected bootstrap calibration curves (1000 resamples). The rmda package was used to perform DCA for net clinical benefit evaluation. The Brier score was calculated to quantify model calibration error. ROC curves were plotted to calculate the area under the curve (AUC), sensitivity, specificity, and Youden’s index. Statistical significance was defined as a two-tailed P < 0.05. All the statistical analyses were reviewed and certified by a professional biomedical statistician.

RESULTS

SAS and SDS scores of GDM patients

A total of 38 GDM patients with SAS scores ≥ 50 points and SDS scores ≥ 53 points were classified into the comorbidity group. The remaining 200 patients constituted the control group. The SAS and SDS scores were significantly higher in the comorbidity group than in the control group (both P < 0.001) (Table 1).

Table 1 Self-Rating Anxiety Scale and Self-Rating Depression Scale scores of gestational diabetes mellitus patients, mean ± SD.

Group	n	SAS	SDS
Comorbidity group	38	57.39 ± 4.88	58.50 ± 4.23
Control group	200	41.84 ± 4.73	44.64 ± 4.65
t value		18.489	18.230
P value		< 0.001	< 0.001

GDM: Gestational diabetes mellitus; SAS: Self-Rating Anxiety Scale; SDS: Self-Rating Depression Scale.

Comparison of clinical data between the two groups

No significant intergroup differences were observed in gestational age, pre-pregnancy body mass index, gravidity, parity, family history of diabetes, early pregnancy reactions, gestational complications, household income, occupation, residence, planned pregnancy, FBG, HbA1c, UA, Cr, BNP, TG, TC, HDL-C, LDL-C, or IR (all P > 0.05). Patients in the comorbidity group were significantly older, with higher proportions of prior miscarriage, junior high school or lower education, and low/moderate social support (all P < 0.001) (Table 2).

Table 2 Comparison of clinical and laboratory data between the two groups, mean ± SD.

Factor		Comorbidity group (n = 38)	Control group (n = 200)	χ2/t value	P value
Age (years)		30.24 ± 2.31	28.60 ± 2.49	4.809	< 0.001
Gestational age (weeks)		25.92 ± 2.76	26.55 ± 2.19	-1.552	0.122
Pre-pregnancy BMI (kg/m2)		22.21 ± 1.84	22.02 ± 2.08	0.520	0.604
Pregnancy times (times)		3.31 ± 1.00	3.51 ± 1.10	-1.022	0.308
Number of births (times)		1.45 ± 0.39	1.54 ± 0.42	-1.186	0.237
Early pregnancy symptoms (n)	Yes	18	104	0.274	0.601
	No	20	96
Family history of diabetes (n)	Yes	7	40	0.050	0.823
	No	31	160
Other pregnancy syndrome (n)	Yes	10	36	1.416	0.234
	No	28	164
Previous abortion (n)	Yes	15	30	16.617	< 0.001
	No	23	170
Educational level (n)	Junior high school and below	21	49	11.819	0.001
	High school and above	17	151
Monthly household income (yuan) (n)	≤ 3000	4	14	1.565	0.457
	3001-6000	14	60
	> 6000	20	126
Occupational type (n)	Mental work	29	138	0.816	0.366
	Physical work	9	62
place of residence (n)	Towns	17	84	0.098	0.754
	Village	21	116
Planned pregnancy (n)	Yes	16	102	1.011	0.315
	No	22	98
Social support (n)	Low/moderate	26	75	12.489	< 0.001
	Satisfaction	12	125
FBG (mmol/L)		5.58 ± 1.23	5.27 ± 0.98	1.691	0.092
HbA1c (%)		7.36 ± 2.42	7.01 ± 1.53	0.858	0.396
UA (mmol/L)		282.25 ± 55.11	279.46 ± 54.57	0.289	0.773
Cr (μmol/L)		96.81 ± 9.54	94.97 ± 9.69	1.077	0.283
BNP (pg/mL)		100.38 ± 22.44	96.52 ± 21.21	1.020	0.309
TG (mmol/L)		1.50 ± 0.25	1.45 ± 0.20	1.066	0.292
TC (mmol/L)		4.21 ± 0.83	4.43 ± 0.72	-1.668	0.097
HDL-C (mmol/L)		1.32 ± 0.35	1.29 ± 0.37	0.429	0.668
LDL-C (mmol/L)		2.88 ± 0.51	2.75 ± 0.50	1.412	0.159
IR		2.23 ± 0.40	2.17 ± 0.47	0.759	0.449

BMI: Body mass index; FBG: Fasting blood glucoses; HbA1c: Glycated hemoglobin; UA: Uric acid; Cr: Creatinine; BNP: Brain natriuretic peptide; TG: Triglycerides; TC: Total cholesterol; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; IR: Insulin resistance.

Univariate logistic screening of risk factors

Univariate analysis indicated that age, history of abortion, education level, and social support might affect anxiety and depression in GDM patients (P < 0.001) (Table 3).

Table 3 Single-factor screening of gestational diabetes mellitus patients with anxiety and depression disorder.

Item	c	SE	Wald χ2	P value	OR (95%CI)
Age	0.375	0.087	18.646	< 0.001	1.455 (1.227-1.726)
Previous abortion	1.518	0.395	14.794	< 0.001	4.565 (2.106-9.897)
Educational level	1.477	0.368	16.090	< 0.001	4.380 (2.128-9.013)
Social support	1.392	0.379	13.482	< 0.001	4.024 (1.914-8.460)

OR: Odds ratio; CI: Confidence interval.

Multivariate logistic regression analysis of comorbid anxiety and depression in GDM patients

Dependent variable: Group (1 = comorbidity group, 0 = control group). Independent variable assignment was standardized; all risk exposure factors were coded as 1 and reference protective factors as 0 (Table 4). Multivariate regression confirmed advanced age, miscarriage history, junior high school or lower education, and low/moderate social support as independent risk predictors (all P < 0.05) (Table 5).

Table 4 Group assignment of participants.

Variable	Name	Assignment
Dependent variable	Group	1 = comorbidity group, 0 = control group (reference)
Independent variable	Age	Measurement data are included according to actual value
	Previous abortion	1 = yes (risk factor), 2 = no (reference)
	Educational level	1 = junior high school and below (risk factor), 2 = high school and above(reference)
	Social support	1 = low/moderate social support (risk factor); 0 = satisfactory social support (reference)

Table 5 Multivariate logistic regression analysis of comorbid anxiety and depression in gestational diabetes mellitus patients.

Item	β	SE	Wald χ2	P value	OR (95%CI)
Age	0.338	0.093	13.166	< 0.001	1.403 (1.168-1.684)
Previous abortion	1.621	0.469	11.927	0.001	5.057 (2.016-12.686)
Educational level	1.391	0.427	10.626	0.001	4.018 (1.741-9.272)
Social support	1.358	0.430	9.955	0.002	3.889 (1.673-9.042)

OR: Odds ratio; CI: Confidence interval.

Nomogram construction

A nomogram prediction model for evaluating the risk of anxiety and depression in GDM patients was developed using the risk factors listed in Table 5. The analysis showed that each additional year of age resulted in a 12.3-point increase in the nomogram model score. In GDM patients with a history of abortion, the score increased by 29.8 points. When the educational level of GDM patients was junior high school or below, the score increased by 25 points. A general sense of social support among GDM patients was associated with a 24.3-point increase in the model score (Figure 1).

Figure 1  Nomogram of the risk of comorbidity anxiety and depression disorders in patients with gestational diabetes mellitus.

Comprehensive model performance evaluation

ROC analysis: Single indicators yielded AUCs ranging from 0.635 to 0.718, and the combined nomogram model achieved an AUC of 0.853, sensitivity of 0.816, and specificity of 0.785 (Table 6 and Figure 2).

Figure 2 Receiver operating characteristic curve analysis. AUC: Area under the curve.

Table 6 Predictive value of receiver operating characteristics analysis.

Test variable	AUC	SE	P value	95%CI	Cut-off value	Youden index	Sensitivity	Specificity
Age	0.718	0.046	< 0.001	0.629-0.808	30.500	0.354	0.579	0.775
Previous abortion	0.635	0.054	0.008	0.530-0.740	1.500	0.270	0.395	0.875
Educational level	0.666	0.051	0.001	0.566-0.766	1.500	0.333	0.553	0.780
Social support	0.667	0.048	0.001	0.573-0.761	1.500	0.334	0.684	0.650
Joint prediction	0.853	0.033	< 0.001	0.789-0.917	0.859	0.601	0.816	0.785

AUC: Area under the curve; CI: Confidence interval.

Calibration performance: C-index = 0.847 and Brier score = 0.091; The bootstrap bias-corrected calibration curve closely overlapped with the ideal reference line (Figure 3), demonstrating good calibration.

Figure 3  Calibration curves of the nomogram risk prediction model.

DCA results: The model maintained a positive net clinical benefit within a threshold probability range of 0.02-0.93 (Figure 4). However, threshold probabilities above 0.90 are clinically impractical for routine psychological intervention in antenatal clinics, as clinicians cannot implement universal mental intervention for patients with extremely high predicted risk.

Figure 4  Decision curves of the nomogram risk prediction model.

DISCUSSION

Women with GDM face substantially elevated risks of concurrent anxiety and depression, which creates a bidirectional vicious cycle: Negative emotions activate sympathetic excitation, elevate glucagon secretion, and worsen IR, while poor glycemic control further aggravates psychological stress[13-15]. Early risk identification and targeted psychological intervention are critical to break this cycle.

Patients with comorbidities in this study were older than those in the control group. They also had a higher rate of previous abortion and lower educational level (junior high school or below), with a tendency to develop anxiety and depression (P < 0.05), as reported by Salimi et al[16]. Older maternal age is related to poor cell quality and an increased rate of miscarriage, a higher chance of which shows a “U”-shaped curve. Aging is also related to poor islet function, high blood placental prolactin levels, and risk of diabetes and other complications[17]. A history of abortion may cause long-term mental trauma of emotions such as guilt and blame, and cause anxiety and depression.

Lower education levels may be related to poor knowledge about GDM, its severity, treatment options, and consequences, which may increase fear and worry[18]. These patients may also have difficulty with self-care and blood glucose control, which additionally increases their anxiety and worry[19].

Social support was another factor. Patients with low perceived social support were prone to fear and depression. The nomogram model revealed that low social support contributed 24.3 points to the risk score. A shortage of support from family, friends, or community aggravates feelings of isolation and incompetence. They may also result in misunderstandings about the disease, its treatment, and prognosis, thereby aggravating psychological stress. Rational social support can assist people in overcoming adversity and being compliant with treatment, while deficient social support can worsen the situation and induce a vicious circle[20]. The combined predictive model, including age, abortion history, education, and social support, obtained an AUC of 0.853 and good discrimination.

FBG, HbA1c, lipid profiles, and IR indices were comparable between the two groups, indicating that comorbid anxiety and depression in GDM patients were primarily driven by psychosocial and sociodemographic factors rather than glycemic or metabolic disturbances[21,22]. This finding provides a novel clinical insight: Psychological risk assessment should be independent of routine glycemic monitoring and integrated into standardized antenatal care for all women with GDM. The moderate overall glycemic control of this single-center cohort may narrow the metabolic indicator variation and limit the ability to detect weak correlations between metabolism and mood disorders[16].

Most existing prediction models for perinatal mental disorders have been developed for general pregnant populations or depend merely on biochemical indicators. In contrast, the present nomogram incorporates four readily obtainable nonlaboratory predictors and does not require additional blood tests, enabling convenient and rapid bedside screening in primary obstetric clinics[23-25]. DCA demonstrated the stable net clinical benefit of this model across a wide range of threshold probabilities. Although extremely high-risk thresholds are clinically unfeasible for routine practice, clinicians can combine their professional experience to establish individualized cutoff values for standardized clinical screening and targeted intervention. This study had several limitations. It adopted a single-center retrospective design with a limited sample size, including only 38 positive events, corresponding to an event per variable value of 9.5, which was slightly lower than the recommended standard threshold of 10-15. Bootstrap internal validation cannot replace external multicenter cohort validation for verifying the generalizability of the model. In addition, social support was assessed using a self-designed and unvalidated questionnaire rather than standardized psychometric scales, which may have compromised the reliability of outcome measurement. The DCA threshold range in this study was relatively wide, and the high-risk thresholds above 0.90 lacked practical value for clinical intervention. Furthermore, no external validation cohort was included, and no sensitivity analyses, such as least absolute shrinkage and selection operator regression or leave-one-out cross-validation, were performed to further improve model robustness[26].

Future large-sample, multicenter, prospective cohort studies should be conducted to externally validate and refine the nomogram. Standardized social support rating scales should be adopted to replace self-designed questionnaires and enhance the validity of variable measurements. Moreover, machine learning algorithms can be applied in subsequent research to compare model performance and optimize risk cutoff values for clinical psychological intervention, thereby providing more accurate evidence for the screening and targeted intervention of anxiety and depression comorbidity among pregnant women.

CONCLUSION

Advanced maternal age, a history of spontaneous miscarriage, low educational level, and insufficient social support are independent risk factors for comorbid anxiety and depression in women with GDM. The internally validated logistic regression nomogram exhibits good discrimination and calibration performance, enabling rapid and early risk stratification in antenatal clinics. Psychosocial factors outweigh metabolic indicators in predicting mood comorbidities in GDM patients. Multicenter external validation and standardized psychosocial measurement tools are required to improve the clinical applicability of the model.