Background Etiopathogenesis of preterm birth (PTB) is multifactorial, with a universe of risk factors interplaying between the mother and the environment. It is of utmost importance to identify the most informative factors in order to estimate the degree of PTB risk and trace an individualized profile. The aims of the present study were: 1) to identify all acknowledged risk factors for PTB and to select the most informative ones for defining an accurate model of risk prediction; 2) to verify predictive accuracy of the model and 3) to identify group profiles according to the degree of PTB risk based on the most informative factors. Methods The Maternal Frailty Inventory (MaFra) was created based on a systematic review of the literature including 174 identified intrauterine (IU) and extrauterine (EU) factors. A sample of 111 pregnant women previously categorized in low or high risk for PTB below 37 weeks, according to ACOG guidelines, underwent the MaFra Inventory. First, univariate logistic regression enabled p-value ordering and the Akaike Information Criterion (AIC) selected the model including the most informative MaFra factors. Second, random forest classifier verified the overall predictive accuracy of the model. Third, fuzzy c-means clustering assigned group membership based on the most informative MaFra factors. Results The most informative and parsimonious model selected through AIC included Placenta Previa, Pregnancy Induced Hypertension, Antibiotics, Cervix Length, Physical Exercise, Fetal Growth, Maternal Anxiety, Preeclampsia, Antihypertensives. The random forest classifier including only the most informative IU and EU factors achieved an overall accuracy of 81.08% and an AUC of 0.8122. The cluster analysis identified three groups of typical pregnant women, profiled on the basis of the most informative IU and EU risk factors from a lower to a higher degree of PTB risk, which paralleled time of birth delivery. Conclusions This study establishes a generalized methodology for building-up an evidence-based holistic risk assessment for PTB to be used in clinical practice. Relevant and essential factors were selected and were able to provide an accurate estimation of degree of PTB risk based on the most informative constellation of IU and EU factors.

No abstract

Link prediction is one of the central problems in graph mining. However, recent studies highlight the importance of higher-order network analysis, where complex structures called motifs are the first-class citizens. We first show that existing link prediction schemes fail to effectively predict motifs. To alleviate this, we establish a general motif prediction problem and we propose several heuristics that assess the chances for a specified motif to appear. To make the scores realistic, our heuristics consider -among others -correlations between links, i.e., the potential impact of some arriving links on the appearance of other links in a given motif. Finally, for highest accuracy, we develop a graph neural network (GNN) architecture for motif prediction. Our architecture offers vertex features and sampling schemes that capture the rich structural properties of motifs. While our heuristics are fast and do not need any training, GNNs ensure highest accuracy of predicting motifs, both for dense (e.g., 𝑘-cliques) and for sparse ones (e.g., 𝑘-stars). We consistently outperform the best available competitor by more than 10% on average and up to 32% in area under the curve. Importantly, the advantages of our approach over schemes based on uncorrelated link prediction increase with the increasing motif size and complexity. We also successfully apply our architecture for predicting more arbitrary clusters and communities, illustrating its potential for graph mining beyond motif analysis. CCS CONCEPTS• Information systems → Data mining.

Quality Control (QC) and Quality Assurance (QA) principles are essential for effective cervical cancer prevention. Being a crucial diagnostic step, colposcopy’s sensitivity and specificity improvements are strongly advocated worldwide since inter- and intra-observer differences are the main limiting factors. The objective of the present study was the evaluation of colposcopy accuracy through the results of a QC/QA assessment from a survey in Italian tertiary-level academic and teaching hospitals. A web-based, user-friendly platform based on 100 colposcopic digital images was forwarded to colposcopists with different levels of experience. Seventy-three participants were asked to identify colposcopic patterns, provide personal impressions, and indicate the correct clinical practice. The data were correlated with a panel of experts’ evaluation and with the clinical/pathological data of the cases. Overall sensitivity and specificity with the threshold of CIN2+ accounted for 73.7% and 87.7%, respectively, with minor differences between senior and junior candidates. Identification and interpretation of colposcopic patterns showed full agreement with the experts’ panel, ranging from 50% to 82%, in some instances with better results from junior colposcopists. Colposcopic impressions correlated with a 20% underestimation of CIN2+ lesions, with no differences linked to level of experience. Our results demonstrate the good diagnostic performance of colposcopy and the need for improving accuracy through QC assessments and adhesion to standard requirements and recommendations.

Link prediction is one of the central problems in graph mining. However, recent studies highlight the importance of higher-order network analysis, where complex structures called motifs are the first-class citizens. We first show that existing link prediction schemes fail to effectively predict motifs. To alleviate this, we establish a general motif prediction problem and we propose several heuristics that assess the chances for a specified motif to appear. To make the scores realistic, our heuristics consider -among others -correlations between links, i.e., the potential impact of some arriving links on the appearance of other links in a given motif. Finally, for highest accuracy, we develop a graph neural network (GNN) architecture for motif prediction. Our architecture offers vertex features and sampling schemes that capture the rich structural properties of motifs. While our heuristics are fast and do not need any training, GNNs ensure highest accuracy of predicting motifs, both for dense (e.g., k-cliques) and for sparse ones (e.g., k-stars). We consistently outperform the best available competitor by more than 10% on average and up to 32% in area under the curve. Importantly, the advantages of our approach over schemes based on uncorrelated link prediction increase with the increasing motif size and complexity. We also successfully apply our architecture for predicting more arbitrary clusters and communities, illustrating its potential for graph mining beyond motif analysis.