Figure 1.
Traditional and cutting-edge technologies used in predicting preterm birth
| Technology | Study (Year) | Key Predictive Factors | Sample Size | Sensitivity (%) | Specificity (%) | AUC |
|---|---|---|---|---|---|---|
| Risk Scoring Systems (RSS) | Ferreira et al. (2023) | Maternal age, smoking, cervical length | 56 studies (Meta-analysis) | 4.2–92.0 | Varies | 0.59–0.95 |
| Vaginal Progesterone | Romero et al. (2018) | Cervical length, prior PTB | Meta-analysis | 41% reduction in PTB | — | — |
| Obstetric Pessary | Jin et al. (2019) | Cervical insufficiency | Meta-analysis | Reduced cesarean rates | — | — |
| Cervical Cerclage | Brown et al. (2019) | Short cervix (<25 mm) | Systematic review | Effective in high-risk women | — | — |
| QUiPP Mobile App | Lee et al. (2020) | Cervicometry, fibronectin | Application-based cohort | High | High | — |
| Machine Learning (ANN) | Chen et al. (2011) | Age, BMI, smoking, bleeding | 910 (Taiwan cohort) | 80–100 | — | — |
| SVM (AI Model) | Vovsha et al. (2014) | Socioeconomic status, maternal age | 2,929 (USA) | 60 | 75 | 0, 70 |
| Naive Bayesian Classifier | Malea et al. (2010) | Smoking, age >35, low education | 546 (Romania) | 88 | 82 | 0, 85 |
| Ensemble ML Models | Koivu & Sairanen (2020) | Diabetes, hypertension, IVF history | 16 million (CDC) | — | — | 0, 64 |
Overview of Studies on Artificial Intelligence and Machine Learning Methods in Preterm Birth Prediction
| Study | Authors | Sample Size | Methods | Key Risk Factors Identified | Prediction Accuracy | Findings/Results |
|---|---|---|---|---|---|---|
| Risk of Preterm Birth Prediction | I. Vovsha et al. Columbia University (USA) | 2,929 women | AI algorithms (logistic regression vs. SVM) | Socioeconomic status, race, maternal age | SVM: 60% sensitivity | SVM outperformed logistic regression in sensitivity for PTB prediction. |
| Predicting PTB with Machine Learning | Polytechnic University of Timisoara (Romania) | 546 records | Machine learning (naive Bayesian classifier) | Smoking, maternal age >35, low education level | 88% accuracy | Achieved high prediction accuracy using a naive Bayesian classifier. |
| Evaluating Machine Learning Algorithms for PTB | A. Koivu & M. Sairanen University of Turku (Finland) | 16 million data points | ML algorithms (ANN, decision trees, ensembles) | Diabetes, hypertension, PTB history, infertility treatment, and marital status | AUC 0.64 | ML methods showed superior predictive ability, significantly outpacing traditional statistical methods. |
| Assessing PTB Risk Factors with ANN | H.-Y. Chen et al. National Taiwan University | 910 mother-child pairs | ANN, decision tree | Multiple pregnancies, antepartum bleeding, maternal age, and gynecological conditions | 80–100% accuracy (ANN) | Identified 15 influential risk factors for PTB, with ANN providing high accuracy in prediction across various risk factors. |
Major Risk Factors for Preterm Birth
| Category | Specific Risk Factors |
|---|---|
| Maternal Factors | Advanced maternal age, high/low BMI, smoking, previous PTB, infections (e.g., H. pylori, systemic lupus erythematosus), chronic diseases (hypertension, diabetes). |
| Fetal Factors | Multiple gestations, fetal growth restriction, and high fetal fibronectin levels. |
| Cervical/Uterine Factors | Short cervical length, cervical insufficiency, history of cervical surgeries (e.g., cone biopsy), uterine anomalies (fibroids, adenomyosis). |
| Socioeconomic & Environmental | Low socioeconomic status, poor nutrition, high stress levels, and exposure to pollutants. |
Current and Emerging PTB Prediction Methods
| Method Type | Examples | Limitations |
|---|---|---|
| Clinical History-Based | Prior PTB, maternal age, obstetric history | Limited predictive power, does not account for new risk factors. |
| Biomarker-Based | Cervical length measurement, fetal fibronectin, and inflammatory markers | Requires standardisation, variability in results. |
| AI/Machine Learning | Predictive algorithms (QUiPP, CLEOPATRA), deep learning models | Data quality issues, algorithm biases, ethical concerns. |