Role of Inflammatory Markers and Doppler Parameters in Late-Onset Fetal Growth Restriction: A Machine Learning Approach

doi:10.21203/rs.3.rs-4864163/v1

Download PDF

Research Article

Role of Inflammatory Markers and Doppler Parameters in Late-Onset Fetal Growth Restriction: A Machine Learning Approach

https://doi.org/10.21203/rs.3.rs-4864163/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 17 Oct, 2024

Read the published version in American Journal of Reproductive Immunology →

Version 1

posted

You are reading this latest preprint version

Objectives

This study evaluates the association of novel inflammatory markers and Doppler parameters in late-onset FGR, utilizing a machine learning approach to enhance predictive accuracy.

Materials and methods

A retrospective case-control study was conducted at the Department of Perinatology, Ministry of Health Etlik City Hospital, Ankara, from 2023 to 2024. The study included 240 patients between 32–37 weeks of gestation, divided equally between patients diagnosed with late-onset FGR and a control group. We focused on novel inflammatory markers—systemic immune-inflammation index (SII), systemic inflammatory response index (SIRI), and neutrophil-percentage-to-albumin ratio (NPAR)—and their correlation with Doppler parameters of umbilical and uterine arteries. Machine learning algorithms were employed to analyze data collected, including demographic, neonatal, and clinical parameters, to develop a predictive model for FGR.

Results

The machine learning model, specifically the Random Forest algorithm, effectively integrated the inflammatory markers with Doppler parameters to predict FGR. NPAR showed a significant correlation with FGR presence, providing a robust tool in the predictive model. In contrast, SII and SIRI, while useful, did not achieve the same level of predictive accuracy. The model highlighted the potential of combining ultrasound measurements with inflammatory markers to improve diagnostic accuracy for late-onset FGR.

Conclusions

This study illustrates the efficacy of integrating machine learning with traditional diagnostic methods to enhance the prediction of late-onset FGR. Further research with a larger cohort is recommended to validate these findings and refine the predictive model, which could lead to improved clinical outcomes for affected pregnancies.

The take-home message:

This study demonstrates that combining novel inflammatory markers, particularly the neutrophil-percentage-to-albumin ratio (NPAR) and the systemic immune-inflammation index (SII), with Doppler ultrasound parameters can significantly improve the prediction accuracy of late-onset fetal growth restriction (FGR) using a machine learning approach. This integration of machine learning with traditional diagnostic methods provides a more robust and cost-effective tool for the early diagnosis and management of FGR in clinical settings.

Fetal growth restriction

inflammatory markers

Doppler ultrasound

machine learning

predictive model

late-onset FGR

Fetal growth restriction, also known as intrauterine growth restriction, is a common pregnancy complication associated with a number of unfavorable perinatal outcomes. The terminology, etiology, and diagnostic criteria for fetal growth restriction are not universally recognized, and the optimal treatment and timing of delivery for the growth-restricted fetus remain unclear (Anon 2021).

The pathogenesis is multifactorial and includes genetic factors, maternal inflammation, immune system abnormalities, infections, oxidative stress, defects in remodeling of the spiral arteries and endocrine causes (Hracsko et al. 2008, Wixey et al. 2017, Zhang et al. 2023). As the underlying pathophysiological mechanisms and the methods of diagnosis and treatment are different, it is analyzed under 2 subheadings. FGR that begins below 32 weeks of gestation is considered early-onset FGR and FGR that begins above 32 weeks is considered late-onset FGR (Gordijn et al. 2016).

Although the risk of perinatal complications is lower compared to early-onset FGR, late-onset FGR is associated with an increased risk of short- and long-term adverse outcomes, including hypoxemic events and mild neurodevelopmental delay (Rizzo et al. 2020). In both groups, the diagnosis of FGR is based on ultrasound-guided fetal biometric measurements and evaluation of doppler parameters of the umbilical artery, middle cerebral artery and uterine artery (Gordijn et al. 2016, Rizzo et al. 2020, Anon 2021). In contrast to early-onset FGR, however, there is no consensus in the literature on the diagnosis and treatment of late-onset FGR. The difficulties in the diagnosis and management of late-onset FGR are also related to the fact that the underlying pathophysiological mechanisms are not clear. Evaluating the role of inflammatory processes in the development of FGR could make the disease more understandable. To this end, the aim of this study was to evaluate this entity with new markers such as the systemic immune inflammation index (SII), the systemic inflammatory response index (SIRI) and the neutrophil-to-albumin ratio (NPAR), which have prognostic value in the evaluation of the immune system. Changes in these inflammatory markers, which can be determined using simple peripheral blood parameters, have been shown to be associated with diseases such as malignancies, cardiovascular disease, inflammatory and autoimmune diseases and some obstetric pathologies (Tanacan et al. 2020, Wang et al. 2021, He et al. 2022, Hrubaru et al. 2022, Lan et al. 2023, Lv et al. 2023, Wu et al. 2023).

However, advances in current diagnostic methods now require the use of advanced technical ultrasound equipment, which ironically limits the applicability of these diagnostic criteria in low-income countries. The fact that access to artificial intelligence, the latest product of technology, only requires the use of the internet, makes artificial intelligence cost-effective and not a major expense. The use of predictive models created using machine learning is becoming inevitable in the medical field. This study aimed to investigate whether inflammatory processes are effective in patients with late-onset FGR by using NPAR, whose effectiveness has never been studied in patients with fetal growth retardation, together with inflammatory markers such as SII and SIRI. Our results were evaluated in a predictive model using machine learning algorithms, and we aimed to investigate the value of inflammatory markers in predicting FGR.

Study Design

This study, which is designed as a retrospective case-control study, was conducted between 2023 and 2024 in the Department of Perinatology, Ministry of Health, Etlik City Hospital, (Ankara/Turkey). The Ethics Committee of our center approved the study protocol (Protocol ID: AEŞH-BADEK-2024-43) and all participating patients provided written informed consent. The study was conducted in accordance with the universal ethical standards of the Declaration of Helsinki. In addition, the study received clinical trial approval with the number NCT06372938.

Participants

The study included 240 patients between 32–37 weeks of gestation. Crown-rump length in the first trimester was used to confirm gestational age. Of the patients included in the study, 120 patients were diagnosed with late-onset FGR and 120 patients were included in the control group of developing fetuses according to gestational age.

Demographic, neonatal and clinical data were obtained from the medical records (age, body mass index, parity, smoking status, FGR diagnosis week, gestational age at delivery, birth weight, cesarean section, APGAR score at 1 and 5 minutes). The diagnosis of FGR was made according to the following Delphi criteria (5). (26909664). EFW < 3rd percentile or EFW < 10th percentile with Doppler evidence of placental dysfunction (Umbilical artery Doppler (UA) pulsatility index (PI) > 95th percentile, absence of umbilical artery end-diastolic flow (UAEDF) or reverse UAEDF and/or cerebroplacental ratio (CPR) < 5th percentile). Patients with multiple pregnancies, gestational hypertension and diabetes mellitus, preterm labor and chorioamnionitis, fetuses with fetal chromosomal abnormalities and/or structural anomalies, acute or chronic inflammatory diseases, chronic kidney and liver diseases, immunologic and hematologic disorders, genetic diseases that may affect serum biomarker levels, and patients taking anti-inflammatory or cortisone-containing medications were excluded.

Ultrasound Measurements and Doppler Parameters

All ultrasound measurements were performed transabdominally with a convex 3.5 MHz transducer on a Voluson S10 Expert sonography unit (GE Healthcare, Milwaukee, WI) by a fetal and maternal medicine expert with 10 years of experience. Fetal doppler and biometric parameters were measured according to the standards of the International Society of Ultrasound in Obstetrics and Gynecology and the Institute of Ultrasound in Medicine. These measurements include estimated fetal weight (EFW), biparietal diameter (BPD), head circumference (HC), maximum vertical pouch (MVP) and doppler parameters such as umbilical artery (UA), middle uterine artery (Ut-A) and middle cerebral artery (MCA). The UA Doppler values were obtained from the free loops of the umbilical cord. Uterine artery Doppler values were measured at the point where the uterine artery crossed the external iliac artery after color flow mapping by placing the probe longitudinally at a medial angle to the lower lateral quadrant of the iliac artery. MCA Doppler measurements were performed at the point where the artery crossed the sphenoidal wing near the Willis' circulus. The cerebral-placental ratio (CPR) was calculated as a simple ratio between the MCA PI and the UA PI (Lees et al. 2020).

Laboratory Tests and Inflammatory Markers

The laboratory values were measured at the time of FGR diagnosis (between 32 and 37 weeks of pregnancy). After evaluation of hemoglobin (g/dl), leukocytes (103/µL), monocytes (103/µL), lymphocytes (103/µL), neutrophils (103/µL), platelets (103/µL) and albumin (g/dl), the inflammation values were calculated as follows:

SII = Absolute platelet count (APC)* Absolute neutrophil count (ANC) / Absolute lymphocyte count (ALC),
SIRI = Absolute monocyte count (AMC) * ANC/ ALC,
NPAR = Proportion of neutrophils (in total leukocytes) (%) × 100/albumin (g/dL).

The indices obtained and recorded from the hemogram values were calculated according to these formulas.

Statistical analysis

Statistical analyzes were performed using Statistical Package for Social Sciences (SPSS) software version 27.0 (IBM Corp., Armonk, NY, USA) for standard statistical tests and Python software (version 3.12.2) for advanced data analysis, including machine learning. The distribution of variables was assessed using the Kolmogorov-Smirnov test. Numerical variables were presented as mean ± standard deviation for normally distributed data and as median (min-max range) for non-normally distributed data. Differences between groups for parametric and non-parametric variables were analyzed using the t-test for independent samples and the Mann-Whitney U-test, respectively. To investigate the predictive power of the fetal growth restriction (FGR) indices studied, an analysis of the receiver operating characteristic (ROC) curve was performed. A p-value of less than 0.05 was considered statistically significant.

In addition to traditional statistical methods, Python software was used to implement the Random Forest algorithm as part of our machine learning analysis. This approach allowed us to integrate and analyze complex data sets and further evaluate the relationship between inflammatory markers and FGR. The random forest model, a sophisticated ensemble learning method that constructs multiple decision trees during the training phase and outputs the mode of classes (classification) or mean prediction (regression) of each tree, was specifically chosen for its robustness and ability to deal with imbalanced datasets, which is common in medical research. This machine learning analysis provided valuable insights into the predictive accuracy of the indices studied and contributed to a deeper understanding of their potential utility in predicting FGR.

By integrating SPSS for conventional statistical analysis with the machine learning capabilities of Python, a comprehensive analytical approach was used in this study to investigate the predictive value of various indices related to FGR, demonstrating the synergy between traditional statistical methods and advanced data analysis techniques.

Features of the machine learning model

Pre-processing of the data

When preparing our data set for analysis, we first determined the dependent and independent variables. Then our data set was divided into a training set and a test set (test size 0.3). It was ensured that the class distributions in the training and test data sets were as close as possible to the distribution in the original data set.

Creating a model

In creating our model, a classification model was developed using the RandomForestClassifier class. To make the model work more effectively, a pipeline with feature scaling and RandomForestClassifier operations was created.

Model training

In the pipeline created, the model was trained with the training data using the fit function. This process allows the model to learn from the training set using examples and use this information to predict future data.

Performance evaluation

In the performance evaluation phase of the model, predictions were made for the test set and the class probabilities were calculated. Based on these predictions, performance metrics such as the accuracy rate of the model and the AUC value were calculated. In addition, key performance indicators of the model such as sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were determined using the confusion matrix.

Determination of the feature significance

In the final phase, the significance of the features identified by the RandomForestClassifier in the pipeline created was investigated. The influence of each of these features on the model was transferred to a DataFrame, sorted and analyzed. The findings were recorded.

The demographic and clinical characteristics as well as the perinatal outcomes of the study groups are shown in Table 1. There were no statistically significant differences between the groups in terms of maternal age, parity, gestational age during the study and C/S rate (p: 0.294; p: 0.301; p: 0.080, p: 0.796; respectively). BMI was higher in the control group than in the FGR group (p: 0.030). Estimated fetal weight, amniotic fluid volume, gestational age at birth, birth weight, Apgar score at 1 minute and Apgar score at 5 minutes, middle cerebral artery PI, CPR, and abdominal circumference centiles were lower in the FGR group than in the control group (p: < 0.001, p: 0.001, p: < 0.001, p: < 0.001, p: < 0.001, p: < 0.001, p:0.011, p: <0.001, p: <0.001; respectively). NICU admission and umbilical artery PI were higher in the FGR group than in the control group (p: < 0.001, p: <0.001; respectively)

Table 1

Maternal characteristics and perinatal outcomes of the participants
	Control Group n: 120	FGR n: 120	P
Maternal age (year) (mean ± SD)	27.4 ± 5.2	26.7 ± 5.7	0.294
Diagnosis week (week) (mean ± SD)	35.4 ± 1.7	35.8 ±1.7	0.08
Parity (n,%)			0.301
Nulliparous	53 (44.2%)	61 (50.8%)
Multiparous	67 (55.8%)	59 (49.2%)
Cesarean section (n,%)	56 (46.7%)	58 (48.3%)	0.796
Smoker (n,%)	0 (0.0%)	5 ( 4.2%)	0.024
BMI (kg/m²) (mean ± SD)	28.7 ± 4.9	26.9 ± 4.5	0.03
Gestational age at delivery (week) (mean ± SD)	38.7 ± 1.3	37.0 ± 1.1	< 0.001
Birth weight (g) (mean ± SD)	3253 ± 425	2345 ± 340.5	< 0.001
APGAR Score at 1st minute (mean ± SD)	8 ± 0.8	8 ± 1.1	< 0.001
APGAR Score at 5th minute (mean ± SD)	9 ± 0.6	9 ±0.9	< 0.001
NICU admission (n,%)	14 (11.7%)	45 (37.5%)	< 0.001
Umbilical artery PI	0.83 ±0.17	0.98 ±0.32	< 0.001
Middle cerebral artery PI	1.60 ± 0.82	1.39 ± 0.69	0.011
CPR	2.25 ± 0.84	1.70 ±0.70	< 0.001
Estimated fetal weight (g)	2772 ± 550	2112 ± 418	< 0.001
SDVP (mm)	50 ± 10.9	45 ± 13.1	< 0.001
Abdominal circumference (centile)	46 ± 25.6	1.6 ± 1.2	< 0.001
BMI: Body mass index, NICU: Neonatal intensive care unit, PI: Pulsatility index, CPR: cerebro-placental ratio, SDVP: Single deepest vertical pocket

A comparison of the inflammatory parameters and the ratios of their indices between the study groups is shown in Table 2. There were no statistically significant differences between the groups in terms of WBC, neutrophils, monocytes, lymphocytes, platelets and hemoglobin (p: 0.996; p: 0.360; p: 0.054, p: 0.075, p: 0.651, p: 0.141; respectively). İmmature granulocyte, albumin were higher in the control group than in the FGR group (p: 0.017; p: 0.011; respectively). The neutrophil/WBC ratio, NPAR and SII ratio values were statistically higher in the FGR group than in the control group (p: 0.002; p: <0.001; p: 0.042; respectively). There were no statistically significant differences in the SIRI ratio between the groups (p: 0.904).

Table 2

Relationship between inflammation parameters in fetuses with fetal growth restriction and control group
mean ± SD	Control group n: 120	FGR n: 120	P
*WBC (10³/mm**³)	9930 ± 1849	10076 ± 2652	0.996
*Neutrophil (10³/mm**³)	7082 ± 1553	7532 ±2268	0.360
*Monocyte (10³/mm**³)	704 ± 191	652 ± 194	0.054
*Lymphocyte (10³/mm³) (mean ± SD)**	2061 ± 816	1882 ± 489	0.075
*Platelet (10³/mm**³)	241 ± 56	246 ± 66	0.651
İmmature granulocytes	98.7 ± 68.0	88.6 ± 75.6	0.017
Hemoglobin (g/dL)	12.2 ± 6.3	11.9 ± 1.2	0.141
Albumin (g/dL)	28.7 ± 4.9	26.9 ± 4.5	0.011
Neutrophil/WBC Ratio	0.71 ± 0.057	0.79 ± 0.630	0.002
IG percentage	0.0097 ± 0.00639	0.0103 ± 0.02454	0.006
NPAR	1.908 ± 0.197	2.168 ± 1.731	< 0.001
SII	890.6 ± 321.4	1048.3 ± 505.7	0.042
SIRI	2618 ± 1111	2820 ± 1554	0.904
WBC: White blood cell count, IG: immature granulocytes, NPAR: Neutrophil-Percentage-to-Albumin Ratio, SII: Systemic immune inflammation index, SIRI: Systemic inflammation response index

Receiver operating characteristic (ROC) analysis to assess the utility of the NPAR, SII, SIRI and CPR ratios as well as sensitivity, specificity, positive predictive value and negative predictive value are shown in Table 3. NPAR, SII and CPR were significantly associated with FGR in predicting fetal growth restriction (p:<0.001, p:0.006, p:<0.001, respectively). SIRI showed no statistically significant difference in FGR prediction (p: 0.250). The area under the curve (AUC) values obtained as a result of the ROC curve analysis are shown in Fig. 1.

Table 3

Logistic regression analysis and ROC analysis for each variable
	Sensivity	Specificity	PPV	NPV	AUC	OR	CI 95%	P-value
NPAR	60%	63%	62%	61%	0.642	11.644	3.060–44.301	< 0.001
SII	50%	58%	54%	53%	0.576	1.001	1.000–1.002	0.006
SIRI	39%	59%	48%	49%	0.505	1.000	1.000–1.000	0.250
CPR	74%	59%	64%	69%	0.710	0.392	0.266–0.576	< 0.001
NPAR: Neutrophil-Percentage-to-Albumin Ratio, SII: Systemic immune inflammation index, SIRI: Systemic inflammation response index, CPR: Cerebro-placental ratio, PPV: Positive predictive value, NPV: Negative predictive value, CI: Confidence interval, AUC: Area under curve OR: Odds ratio

The accuracy rates and area under the curve (AUC) for NPAR, umbilical artery doppler PI centile, MCA doppler PI centile and uterine artery doppler PI centilewere assessed in the prediction model when assessing the performance metrics of the machine learning algorithm used for FGR prediction. The corresponding values were 44% (AUC: 0.516), 62% (AUC: 0.704), 55% (AUC: 0.477) and 66% (AUC: 0.731), respectively. Subsequently, the accuracy rates and area under the curve (AUC) for the inflammatory markers NPAR, SII, and SIRI were reassessed in the prediction model that combined the uterine artery doppler PI centile with the umbilical artery doppler PI centile (Table 4). The resulting values were 77% (AUC: 0.851), 75% (AUC: 0.818) and 73% (AUC: 0.793), respectively. In these prediction models, the coefficients of the NPAR, SII and SIRI variables in the model were 0.34, 0.32 and 0.33, respectively. The results are shown in Fig. 2.

Table 4

Performance metrics of the machine learning algorithm used for FGR prediction
	Sensivity	Specificity	PPV	NPV	Accuracy	AUC
NPAR	43%	45%	45%	43%	44%	0.516
Umbilical A. PI Centile	64%	60%	63%	61%	62%	0.704
MCA PI Centile	59%	51%	56%	54%	55%	0.477
Uterin A. PI Centile	81%	51%	63%	72%	66%	0.731
NPAR + Umbilical A. PI Centile + Uterin A. PI Centile	75%	80%	80%	75%	77%	0.851
SII + Umbilical A. PI Centile + Uterin A. PI Centile	81%	68%	73%	77%	75%	0.818
SIRI + Umbilical A. PI Centile + Uterin A. PI Centile	81%	65%	71%	76%	73%	0.793
PI: Pulsatiliy İndex, A: Artery, MCA: Middle cerebral artery, NPAR: Neutrophil-Percentage-to-Albumin Ratio, SII: Systemic immune inflammation index, SIRI: Systemic inflammation response index

In the present study, we showed that NPAR and SII were significantly higher in women with a fetus with late-onset FGR compared to the control group (p: <0.001, p: 0.042; respectively). Both ROC analyzes and a machine learning prediction model showed that the combination of these inflammatory markers with umbilical and uterine artery doppler PI percentiles strongly predicted FGR as a secondary outcome.

Pregnancy is a complex and dynamic process in which pro-inflammatory and anti-inflammatory mechanisms are in a delicate balance. Disturbances in inflammatory mechanisms can cause obstetric complications such as FGR and pre-eclampsia by interfering with natural processes such as spiral artery remodeling and trophoblast invasion. Many studies have focused on the immunological changes in pregnancy. Ariyakumar et al. have shown that suppression of pro-inflammatory Th1/Th17 immunity is crucial for a successful pregnancy. The p65 + + Th1/Th17 suppression is absent in patients with FGR and FasL expression is decreased, revealing a process in which the inflammatory response dominates (Ariyakumar et al. 2021). Zanno et al. showed that Th2 and IL-4 were increased in rats with FGR and that this increase disrupted postnatal myelination in the rat model and that the development of myelination increased in rat fetuses in which the IL-4 response was neutralized (Zanno et al. 2019). In another study, CD163 + macrophages were shown to be higher in rats with FGR, and in the earlier study by Vesce et al. the absolute value of NK cells was much higher in the FGR group (Vesce et al. 2014, Sharps et al. 2020).

Several studies in the literature show that inflammatory processes are directly related to pregnancy complications (Kalagiri et al. 2016, Wixey et al. 2017). This role of inflammation has been observed in animal studies, notably by Cotechini et al. using low doses of lipopolysaccharide, and these effects have been shown to be associated with certain features of FGR (Cotechini et al. 2014). Greer et al. studied 10,204 placentas to investigate the immunologic basis of placental insufficiency and showed that pregnancies in which lymphohistiocytic infiltration caused low-grade and high-grade chronic villous inflammation were associated with a hypoplastic placenta and FGR (Greer et al. 2012).

The quantification of a proinflammatory environment is important in this context and new markers are used in the literature. In the study by Fıratlıgil et al. on SII, one of these inflammatory markers, high ΔSII levels in maternal blood indicate an inflammatory process causing FGR. They showed that it can be used as a screening test with a specificity of 90%. (Firatligil et al. 2024). Ağaoğlu et al. showed that SII is a useful parameter both in the diagnosis of FGR and in the prediction of NICU admission (Ağaoğlu et al.). In our study, we showed that SII was significantly higher in patients with FGR (p: 0.042). In contrast to SII, SIRI, another inflammatory marker, showed no significant difference between patients with FGR and the control group (p: 0.904). Our results are also consistent with the literature. There are almost no studies on NPAR, another inflammatory marker that we examined in our study, from an obstetric perspective. Lv et al. reported that NPAR was an independent predictor of poor outcome in stroke-related pneumonia and predicted poor outcome in intracranial hemorrhage (Lv et al. 2023). Taner et al. evaluated the creatinine levels of renal transplant patients 5 years later and found that creatinine levels were strongly correlated with NPAR (Taner et al. 2023). In our study, NPAR was statistically significantly increased in the FGR group and proved to be a stronger marker than all other inflammatory markers. NPAR performed better than other inflammatory markers in predicting FGR with a sensitivity of 60% and a specificity of 63%, suggesting that it may be an effective indicator for predicting FGR (p: <0.001 AUC: 0.642). However, the sensitivity and specificity of SII and SIRI were lower, with AUC values of 0.576 and 0.505, respectively. However, it should be kept in mind that all these markers should not be used alone for the diagnosis of FGR, but in combination with ultrasonographic evaluation and doppler examinations.

In this context, we have developed a machine learning algorithm using ultrasonographic parameters and inflammatory markers. The random forest algorithm we used in our study works by combining the predictions of a set of decision trees and is an ideal algorithm for our study as it works well when there is an imbalance between classes in the datasets, is resistant to overfitting and can be used in small datasets (Han et al. 2021). While the sensitivity and specificity of the ML prediction model for NPAR alone were 43% and 45%, respectively, when combined with umbilical artery PI percentile and uterine artery PI percentile, the performance of the model increased significantly and the AUC value increased to 0.851, suggesting that combining inflammatory markers with ultrasonographic measurements may provide a more powerful and effective model for predicting FGR. In the prediction models of SII and SIRI with these doppler parameters, the sensitivity values were slightly higher than for NPAR, but NPAR proved to be more successful in the overall model (AUC NPAR: 0.851 sensitivity 75%; AUC SII: 0.818 sensitivity 81%; SIRI: AUC 0.793 sensitivity 81%, respectively). Feature Importance values for NPAR, SII and SIRI of 0.34, 0.32 and 0.33, respectively, were calculated for the 3 inflammatory markers. These coefficients reflect the predictive power of the markers in our model and help us to understand the importance and contribution of the markers in the model. In these analyzes, NPAR was the inflammatory parameter with the highest coefficient in the model. In conclusion, our machine learning model and statistical analyzes provide valuable insights into the utility of inflammatory markers in the early diagnosis of FGR and the potential efficacy of combining these markers with Doppler parameters. This information can be validated in future studies in larger patient populations and used in the development of FGR management strategies.

The exclusion of pregnant women with obstetric conditions such as pre-eclampsia and gestational diabetes mellitus, as well as patients who had taken anti-inflammatory medications and had comorbidities that would affect inflammatory parameters, was a limitation as it reduced the number of patients included in the study, but made the study strong as it ensured randomization. However, to better understand the mechanisms of inflammation, studies should be conducted in larger patient populations. Increasing the sample size both for a better functioning of the artificial intelligence algorithms and for the creation of subgroups according to gestational week will determine the future direction of this study. Being aware of these limitations, our study aims to direct future studies to investigate inflammatory processes in more detail and to use artificial intelligence algorithms in this area.

Building on the literature detailing the association of NPAR, SII and SIRI with FGR, this study investigated the role of previously understudied inflammatory markers in this condition and used machine learning algorithms to assess how effective they are for the diagnosis and prediction of fetal growth restriction. These inflammatory markers alone should not be expected to predict fetal growth restriction diagnosed by ultrasound and vascular Doppler examinations. However, the results of this study should improve our understanding of the role of the inflammatory process and its markers in the pathophysiology of FGR and provide new perspectives for clinical practice.

Our study used machine learning to investigate possible associations between late-onset fetal growth restriction (FGR) and inflammatory markers. The combination of markers such as NPAR, SII and SIRI with ultrasound measurements improved the accuracy of FGR prediction. The Random Forest algorithm was effective in dealing with the imbalance of medical datasets. This study emphasizes the need for research on large patient populations and a larger sample size. Our results provide a deeper understanding of inflammatory processes to improve management strategies for FGR.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Author Contribution

CO Ulusoy: Manuscript writing, Data analysis, Protocol management

A Kurt: Data collection, Manuscript editing

Z Seyhanli: Data analysis

B Hizli: Data collection

M Bucak: Data analysis

RT Agaoglu: Data collection

Y Oguz: Protocol development

KY Yucel: Protocol management, Manuscript editing

Ağaoğlu RT et al The significance of systemic immune inflammation index (sii) and systemic inflammation response index (siri) in fetuses with fetal growth restriction. Acad J HEALTH 1, 37–41
Anon (2021) Fetal growth restriction: Acog practice bulletin, number 227. Obstet Gynecol 137:e16–e28
Ariyakumar G et al (2021) Nf-κb regulation in maternal immunity during normal and iugr pregnancies. Sci Rep 11:20971
Cotechini T et al (2014) Inflammation-induced fetal growth restriction in rats is associated with altered placental morphometrics. Placenta 35:575–581
Firatligil FB et al (2024) Evaluation of systemic immune-inflammation index for predicting late-onset fetal growth restriction. Arch Gynecol Obstet 310:433–439
Gordijn SJ et al (2016) Consensus definition of fetal growth restriction: A delphi procedure. Ultrasound Obstet Gynecol 48:333–339
Greer LG et al (2012) An immunologic basis for placental insufficiency in fetal growth restriction. Am J Perinatol 29:533–538
Han S et al (2021) Improving random forest predictions in small datasets from two-phase sampling designs. BMC Med Inf Decis Mak 21:322
He X et al (2022) The neutrophil percentage-to-albumin ratio is related to the occurrence of diabetic retinopathy. J Clin Lab Anal 36, e24334
Hracsko Z et al (2008) Evaluation of oxidative stress markers in neonates with intra-uterine growth retardation. Redox Rep. 13, 11 – 6
Hrubaru I et al (2022) The predictive role of maternal biological markers and inflammatory scores nlr, plr, mlr, sii, and siri for the risk of preterm delivery. J Clin Med 11
Kalagiri RR et al (2016) Inflammation in complicated pregnancy and its outcome. Am J Perinatol 33:1337–1356
Lan CC et al (2023) Predictive role of neutrophil-percentage-to-albumin, neutrophil-to-lymphocyte and eosinophil-to-lymphocyte ratios for mortality in patients with copd: Evidence from nhanes 2011–2018. Respirology 28:1136–1146
Lees CC et al (2020) Isuog practice guidelines: Diagnosis and management of small-for-gestational-age fetus and fetal growth restriction. Ultrasound Obstet Gynecol 56:298–312
Lv XN et al (2023) Neutrophil percentage to albumin ratio is associated with stroke-associated pneumonia and poor outcome in patients with spontaneous intracerebral hemorrhage. Front Immunol 14:1173718
Rizzo G et al (2020) Role of doppler ultrasound at time of diagnosis of late-onset fetal growth restriction in predicting adverse perinatal outcome: Prospective cohort study. Ultrasound Obstet Gynecol 55:793–798
Sharps MC et al (2020) Increased placental macrophages and a pro-inflammatory profile in placentas and maternal serum in infants with a decreased growth rate in the third trimester of pregnancy. Am J Reprod Immunol 84:e13267
Tanacan A et al (2020) A cut-off value for systemic immune-inflammation index in the prediction of adverse neonatal outcomes in preterm premature rupture of the membranes. J Obstet Gynaecol Res 46:1333–1341
Taner S et al (2023) Role of systemic inflammatory markers in pediatric kidney transplantation. Transplant Proc. 55, 1152–1155
Vesce F et al (2014) The behaviour of the peripheral natural killer cells in the foetal growth restriction. Eur Rev Med Pharmacol Sci 18:2248–2252
Wang X et al (2021) Association between the neutrophil percentage-to-albumin ratio and outcomes in cardiac intensive care unit patients. Int J Gen Med 14:4933–4943
Wixey JA et al (2017) Review: Neuroinflammation in intrauterine growth restriction. Placenta 54:117–124
Wu CC et al (2023) Association between neutrophil percentage-to-albumin ratio (npar), neutrophil-to-lymphocyte ratio (nlr), platelet-to-lymphocyte ratio (plr) and long-term mortality in community-dwelling adults with heart failure: Evidence from us nhanes 2005–2016. BMC Cardiovasc Disord 23:312
Zanno AE et al (2019) Reducing th2 inflammation through neutralizing il-4 antibody rescues myelination in iugr rat brain. J Neurodev Disord 11:34
Zhang Y et al (2023) Neutrophil infiltration leads to fetal growth restriction by impairing the placental vasculature in denv-infected pregnant mice. EBioMedicine 95:104739

Download PDF

Journal Publication

published 17 Oct, 2024

Read the published version in American Journal of Reproductive Immunology →

Version 1

posted

You are reading this latest preprint version

Role of Inflammatory Markers and Doppler Parameters in Late-Onset Fetal Growth Restriction: A Machine Learning Approach

Status:

Journal Publication

Version 1

Abstract

Objectives

Materials and methods

Results

Conclusions

The take-home message:

Figures

Introduction

Materials and Methods

Study Design

Participants

Ultrasound Measurements and Doppler Parameters

Laboratory Tests and Inflammatory Markers

Statistical analysis

Features of the machine learning model

Pre-processing of the data

Creating a model

Model training

Performance evaluation

Determination of the feature significance

Results

Discussion

Conclusion

Declarations

Declaration of interests

Author Contribution

References

Status:

Journal Publication

Version 1