Predicting mother and newborn skin-to-skin contact using a machine learning approach (2025)

Background

Despite the known benefits of skin-to-skin contact (SSC), limited data exists on its implementation, especially its influencing factors. The current study was designed to use machine learning (ML) to identify the predictors of SSC.

Methods

This study implemented predictive SSC approaches based on the data obtained from the “Iranian Maternal and Neonatal Network (IMaN Net)” from January 2020 to January 2022. A predictive model was built using nine statistical learning models (linear regression, logistic regression, decision tree classification, random forest classification, deep learning feedforward, extreme gradient boost model, light gradient boost model, support vector machine, and permutation feature classification with k-nearest neighbors). Demographic, obstetric, and maternal and neonatal clinical factors were considered as potential predicting factors and were extracted from the patient’s medical records. The area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, and F_1 Score were measured to evaluate the diagnostic performance.

Results

Of 8031 eligible mothers, 3759 (46.8%) experienced SSC. The algorithms created by deep learning (AUROC: 0.81, accuracy: 0.75, precision: 0.67, recall: 0.77, and F_1 Score: 0.73) and linear regression (AUROC: 0.80, accuracy: 0.75, precision: 0.66, recall: 0.75, and F_1 Score: 0.71) had the highest performance in predicting SSC. Doula support, neonatal weight, gestational age, attending childbirth classes, and maternal age were the critical predictors for SSC based on the top two algorithms with superior performance.

Conclusions

Although this study found that the ML model performed well in predicting SSC, more research is needed to make a better conclusion about its performance.

Skin-to-skin contact (SSC) is commonly defined as placing the newborn directly on the mother’s bare chest immediately after birth and drying, while both mother and the newborn are wrapped in a warm blanket, for at least an hour or until finishing the first feed [1]. According to baby-friendly standards, SSC is valued and supported in hospitals. Numerous maternal and neonatal benefits have been determined for SSC. Maternal benefits include the promotion of breastfeeding [2], earlier placental expulsion [3], reducing postpartum bleeding [4], and reducing maternal stress [5], while neonatal benefits include postnatal neuro-physical adjustments, and regulation of body temperature, heart rate, and respiration; as well as gastrointestinal adaptation [6].

Despite the evidence-based benefits of SSC, limited data exist on its adoption and implementation. Despite the WHO recommendation for immediate or early skin-to-skin contact, separation of mother and newborn is common in most situations [7].

According to some studies, barriers to practicing SSC include a shortage in the number of healthcare providers and midwives in maternity wards, heavy workload, lack of knowledge, time constraints, difficulty in determining SSC eligibility, lack of social support, lack of guidelines, and policies; as well as cultural norms [8]. Little is known about the effect of mothers’ demographics, medical conditions, and obstetrical characteristics on SSC. Therefore, identifying the factors influencing hospital benchmark indicators, including SSC, is required for quality improvement interventions. Machine Learning (ML) models and explainable artificial intelligence techniques enable accurate predictions and explanations. This study aimed to use ML algorithms to predict SSC and explain the ML model’s behavior to aid decision-making.

Participants and design

We retrospectively assessed all singleton pregnant women who gave birth at a tertiary hospital in Bandar Abbas, Iran, for two years using electronic patient records. Extracted data included the mother’s personal characteristics, clinical variables, and variables related to SSC experience. Data extraction was done by trained staff of the “Iranian Maternal and Neonatal Network (IMaN Net),” which is a valid national database.

The inclusion criterion was immediate SSC, defined as SSC initiation within 10min after delivery. The exclusion criteria were unstable condition in either the mother or the newborn, including preterm labor ( delivery before 34th -week gestation), need for cardiovascular resuscitation, any infant who was not physiologically stable enough to tolerate handling SSC, COVID-19, postpartum hemorrhage, excessive vaginal rupture, eclampsia, and use of analgesics that impact the consciousness of mother.

Data collection

The potential predicting factors for SSC in this study were demographic, obstetric, maternal, and neonatal factors. Demographic factors included age, education, place of residency, attending birth classes, and access to prenatal care. Obstetric factors included parity, onset of labor, gestational age, and method of delivery. Maternal and neonatal clinical factors included a history of preeclampsia, gestational diabetes, chronic medical comorbidities, meconium amniotic fluid, fetal distress, intrauterine growth retardation, neonatal weight, newborn sex, and doula support during labor. All the data were extracted from the electronic medical records.

Statistical analysis

The Statistical Package for the Social Sciences (SPSS) version 19 (IBM Corp, Armonk, NY) was used to analyze the data. Categorical variables were presented as numbers and frequencies (%). A chi-square test was performed to identify predictors of SSC. All variables with P < 0.05 based on the Chi-square test were identified as predictors.

A predictive model was built using nine statistical learning models, including linear regression, logistic regression, decision tree classification, random forest classification, deep learning-feed forward, extreme gradient boost classification (XGBoost), light gradient boost (LGB), support vector machine (SVM), and permutation feature classification with k-nearest neighbors (KNN).

To report our findings, we followed the Guidelines for Developing and Reporting Machine Learning Predictive Models in Biomedical Research: A Multidisciplinary View [9]. The third iteration of Python, Python 3.0, was selected for developing the ML model. Scikit-learn, a machine-learning library written in Python, was used to implement the ML algorithm. It includes an extensive collection of cutting-edge machine-learning algorithms for both supervised, including the multi-output classification and regression algorithm, and unsupervised machine learning [10].

Internal validation was carried out using k-fold cross-validation. The cases were randomly assigned to either the “training set” (70%) or the “test set” (30%) using a train_test_split in Scikit. The original dataset kept the rate of SSC and non-SSC groups in the training and test set constant. Using the training set, we arranged the parameters of the prediction models and evaluated their performance using the “test set.” The average performance was calculated by repeating these steps ten times.

Metrics, including area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, and F_1 score, were used to assess the predictive power of the models. We used AUROC as the primary performance metric because it is a widely used index to describe the ML model’s ability to predict outcomes. The metrics were scaled from 0 to 1, with higher values indicating a better model [11]. The Accuracy metric calculates how often a model is correctly predicted across the entire dataset. Precision measures how many of the model’s “positive” predictions were correct. Recall estimates the number of correctly identified positive samples in the dataset. The F_1 score combines precision and recall by taking their harmonic mean and maximizing the F_1 score means maximizing both precision and recall simultaneously.

Of the 8030 eligible mothers, 3759 (46.8%) experienced SSC. Table1 shows the relationship between maternal demographic factors and SSC. SSC performance was more common among younger mothers with higher education, mothers living in rural areas, those who attended birth classes during pregnancy, and those with better access to prenatal care.

Table2 shows the relationship between SSC and obstetrical factors. SSC was more common among mothers who had spontaneous labor at 37th -week gestation or later; and those who had vaginal delivery without episiotomy.

Table3 shows the relationship between maternal and neonatal clinical factors and SSC. Preeclampsia, gestational diabetes, meconium amniotic fluid, fetal distress, chronic disease, and drug addiction were associated with lower SSC performance. Doula support during childbirth, female newborns, and babies weighing 2500–4000g were linked to a higher rate of SSC performance.

In this study, we attempted to evaluate parameters and feature selection based on performance parameters using various ML algorithms. Figure1 shows a plotted ROC chart.

Comparing performance metrics between algorithms showed that deep learning (AUROC: 0.81, accuracy: 0.75, precision: 0.67, recall: 0.77, and F_1 Score: 0.73) and linear regression (AUROC: 0.80, accuracy: 0.75, precision: 0.66, recall: 0.75, and F_1 Score: 0.71) had the highest performance in predicting SSC (Table4).

Table5 displays the critical predictors for SSC among the top two algorithms with the highest performance. According to the findings presented in Table5, factors such as doula support, neonatal weight, gestational age, attending childbirth classes, and maternal age were identified as significant predictors of SSC.

Despite strong recommendations from the WHO, SSC practices vary from 1 to 98% of infants worldwide, with a higher prevalence in high-income countries compared to low-middle-income countries, ranging from 8 to 74% [12]. The prevalence of SSC in our study was 46.8%, which was relatively higher than the reported prevalence in previous studies in Papua New Guinea (45.2%) [13], Gambia (35.7%) [14], Ethiopia (28.1%) [15], Nigeria (12.1%) [16], Bangladesh (28%) [17], and Southern Ethiopia (35.3%) [18]. However, the prevalence observed in our study was lower than that reported in Singapore (84.0%) [19]. Variations in time to initiate SSC and SSC duration can explain the extensive range of SSC in literature.

To the best of our knowledge, our study was the first study that used the ML approach to identify the clinical predictors of SSC. ML approaches have been applied to predict diseases and analyze risk factors based on large population datasets [20,21,22,23]. Comparing performance metrics between algorithms in our study showed that deep learning (AUROC: 0.81, accuracy: 0.75, and precision: 0.67) and linear regression (AUROC: 0.80, accuracy: 0.75, and precision: 0.66) had the highest performance in predicting SSC. Doula support, maternal education, neonatal weight, gestational age, childbirth class attendance, and maternal age were the highest-weighted factors in predicting SSC.

According to the findings of our study, SSC practice was more common among mothers with higher education. In our study, among mothers who experienced SSC, 2.5% were illiterate, while 47.9% had high school diplomas. We also found that mothers who attended birth classes during pregnancy and had better access to prenatal care had a higher rate of SSC practice compared to those who did not attend birth classes and had less access to prenatal care. For example, among 76 mothers who attended birth class, 68.4% experienced SSC in our study. This finding could be explained by the fact that mothers with some level of education could have been more informed and had adequate knowledge regarding the importance of SSC in the health outcome of their newborns. This finding was supported by previous studies [13, 16].

In our study, gestational age was a significant predictor of SSC practice. According to our study findings, newborns with a gestational age of at least 37 weeks were more likely to experience SSC than those with a gestational age of less than 37 weeks (95% vs. 5%). The reason for this finding might be related to the high risk of respiratory distress syndrome in premature babies that necessitates the use of oxygen therapy and ventilation assistance, which are the contraindications for SSc initiation [24].

In our study, the presence of a doula during childbirth was a significant predictor. Of 721 mothers who benefited from the support of a doula during labor, 67.4% practiced SSC. Giving birth in a health facility with the company of skilled and educated birth attendants ensures the provision of optimum maternal and neonatal practices, including SSC practice. This could justify the findings of our study. Furthermore, effectively Trained doulas are empowered to perform the critical care required to support cesarean births and facilitate SSC in the operating room [25].

The method of childbirth was another significant predictor of SSC in our study. While SSC should be the natural practice after vaginal delivery, it is not always feasible during cesarean section, especially in emergency cases [26]. The inappropriate temperature of the operation room, non-cooperating surgical and anesthesia team in performing SSC, crowded operation rooms, and the need for quick turnover of patients can be the barriers to conducting SSC in cesarean cases [27]. According to our study findings, mothers who had vaginal delivery, particularly those without episiotomy, had an increased likelihood of SSC practice. The rate of SSc practice among mothers who gave birth through vaginal delivery without episiotomy was 66.1%, and the rate of SSC performance among mothers who underwent vaginal delivery with episiotomy, cesarean section, and vacuum delivery was 30.5%, 3%, and 0.4%, respectively. One of the challenges of performing SSC is the pain and discomfort during episiotomy repair. Therefore, performing unnecessary episiotomies should be avoided to increase the chance of SSC.

Another factor that predicted SSC in our study was neonatal weight. According to our study findings, the prevalence of SSc practice among cases of neonatal weight less than 2500g was 5.5% compared to 92.3% among cases with neonatal weight between 2500 and 4000g, and 1.2% among cases with neonatal weight above 4000g. The benefits of immediate SSC in low birth weight infants include increased survival, increased breastfeeding, and better mother-infant bonding [28]. However, immediate SSC was not common in newborns weighing less than 2500g in our study, which requires further investigation.

In our study, SSC was more common among mothers who experienced spontaneous labor compared to those with induced labor (71.5% vs. 26.5%). We believe that induced labor is more painful and can be associated with other medical interventions, including oxytocin or prostaglandin augmentation, and is more frequently accompanied by cesarean section. These negative labor experiences reduce self-confidence and may reduce maternal desire to practice SSC.

In our study, all conditions that increase the risk of interventions, including preeclampsia, gestational diabetes, meconium amniotic fluid, fetal distress, chronic disease, and drug addiction, were associated with lower SSC.

We recommend managing the salient factors identified in our study through public education on the benefits and importance of SSC, encouraging women to practice SSC, avoiding the separation of mothers from infants, and encouraging relatives and medical staff to support SSC practice. In the case of cesarean section and complicated delivery, we recommend ensuring the longest possible SSC practice.

The use of the ML approach to identify the predictors of SSC practice was one of the strengths of our study. Our study attempted to simultaneously evaluate some predictors of SSC by using an appropriate sample size and statistical methods. Even though we used a large dataset with a lot of maternal and neonatal information, the database lacked important variables, including socioeconomic status, for the majority of the birth records, which prevented us from including these factors in our selection features. This was one of the limitations of our study. Furthermore, the APGAR score was not recorded in the database and therefore could not be evaluated in our study. It is recommended that further studies evaluate the relationship between these factors and SSC performance.

SSC:

Skin-to-skin contact

ML:

Machine learning

XGBoost:

Extreme gradient boost

LGB:

Light gradient boost

SVM:

Support vector machine

KNN:

K-nearest neighbors

AUROC:

Area under the receiver operating characteristic curve

Authors and Affiliations

1. Mother and Child Welfare Research Center, Hormozgan University of Medical Sciences, Bandar Abbas, Iran

   Sanaz Safarzadeh,Mohammadsadegh Vahidi Farashah,Nasibeh Roozbeh&Fatemeh Darsareh

2. Student research committee, Department of midwifery, School of Nursing and Midwifery, Shahid Beheshti University of Medical Sciences, Tehran, Iran

   Sanaz Safarzadeh

3. Department of Midwifery, Ardabil Branch, Islamic Azad University, Ardabil, Iran

   Nastaran Safavi Ardabili

Contributions

F.D. and S.S. designed the study. N.S. and S.S. contributed to data collection. M.V. and F.D. contributed to the manuscript draft preparation and editing, data analysis, and interpretation. N.R. evaluated and improved academic writing and critical appraisal of the manuscript. All authors read and approved the final manuscript draft and approved the submission.

Corresponding author

Correspondence to Fatemeh Darsareh.

Ethics approval and consent to participate

This study complied with the Declaration of Helsinki and was approved by the Ethics and Research Committee of the Hormozgan University of Medical Sciences (Code: IR.HUMS.REC.1403.061). Data analysis was performed on the records of patients who provided informed consent for using their data for research purposes. For participants who were under eighteen years old, informed consent was obtained from their guardians. Statistical analyses were performed on anonymized data following the university codes of ethics.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions