1
|
Sufriyana H, Wu YW, Su ECY. Human-guided deep learning with ante-hoc explainability by convolutional network from non-image data for pregnancy prognostication. Neural Netw 2023; 162:99-116. [PMID: 36898257 DOI: 10.1016/j.neunet.2023.02.020] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 07/11/2022] [Revised: 01/30/2023] [Accepted: 02/14/2023] [Indexed: 02/26/2023]
Abstract
BACKGROUND AND OBJECTIVE Deep learning is applied in medicine mostly due to its state-of-the-art performance for diagnostic imaging. Supervisory authorities also require the model to be explainable, but most explain the model after development (post hoc) instead of incorporating explanation into the design (ante hoc). This study aimed to demonstrate a human-guided deep learning with ante-hoc explainability by convolutional network from non-image data to develop, validate, and deploy a prognostic prediction model for PROM and an estimator of time of delivery using a nationwide health insurance database. METHODS To guide modeling, we constructed and verified association diagrams respectively from literatures and electronic health records. Non-image data were transformed into meaningful images utilizing predictor-to-predictor similarities, harnessing the power of convolutional neural network mostly used for diagnostic imaging. The network architecture was also inferred from the similarities. RESULTS This resulted the best model for prelabor rupture of membranes (n=883, 376) with the area under curves 0.73 (95% CI 0.72 to 0.75) and 0.70 (95% CI 0.69 to 0.71) respectively by internal and external validations, and outperformed previous models found by systematic review. It was explainable by knowledge-based diagrams and model representation. CONCLUSIONS This allows prognostication with actionable insights for preventive medicine.
Collapse
Affiliation(s)
- Herdiantri Sufriyana
- Graduate Institute of Biomedical Informatics, College of Medical Science and Technology, Taipei Medical University, 250 Wu-Xing Street, Taipei 11031, Taiwan; Department of Medical Physiology, Faculty of Medicine, Universitas Nahdlatul Ulama Surabaya, 57 Raya Jemursari Road, Surabaya 60237, Indonesia
| | - Yu-Wei Wu
- Graduate Institute of Biomedical Informatics, College of Medical Science and Technology, Taipei Medical University, 250 Wu-Xing Street, Taipei 11031, Taiwan; Clinical Big Data Research Center, Taipei Medical University Hospital, 250 Wu-Xing Street, Taipei 11031, Taiwan
| | - Emily Chia-Yu Su
- Graduate Institute of Biomedical Informatics, College of Medical Science and Technology, Taipei Medical University, 250 Wu-Xing Street, Taipei 11031, Taiwan; Clinical Big Data Research Center, Taipei Medical University Hospital, 250 Wu-Xing Street, Taipei 11031, Taiwan; Research Center for Artificial Intelligence in Medicine, Taipei Medical University, 250 Wu-Xing Street, Taipei 11031, Taiwan.
| |
Collapse
|
2
|
Iroz CB, Dahl CM, Cassimatis IR, Wescott AB, Miller ES. Prophylactic anticoagulation for preterm premature rupture of membranes: a decision analysis. Am J Obstet Gynecol MFM 2021; 3:100311. [PMID: 33493702 DOI: 10.1016/j.ajogmf.2021.100311] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 10/05/2020] [Revised: 01/05/2021] [Accepted: 01/10/2021] [Indexed: 11/29/2022]
Abstract
BACKGROUND The current standard of care in the setting of preterm premature rupture of membranes involves antenatal hospitalization until delivery. The reduced physical activity during this time compounds the heightened risk for venous thromboembolism in pregnancy. Prophylactic anticoagulation can decrease this risk of venous thromboembolism; however, this benefit must be balanced against the risks of precluding neuraxial analgesia or increasing the risk of postpartum hemorrhage. OBJECTIVE The objective of this study was to determine the optimal modality for venous thromboembolism prophylaxis during hospitalization for preterm premature rupture of membranes using a decision analysis model. STUDY DESIGN A decision-analytical Markov model was constructed using the TreeAge software comparing the use of unfractionated heparin, low-molecular-weight heparin or no anticoagulation in women with a singleton pregnancy who were hospitalized for preterm premature rupture of membranes after 24 weeks and remained hospitalized until delivery. Maternal outcomes examined included attainment of neuraxial analgesia (vs no analgesia for vaginal delivery or general anesthesia for cesarean delivery), venous thromboembolism, postpartum hemorrhage, and maternal death. Probabilities and utilities were derived from existing literature. Sensitivity analyses were performed to interrogate model assumptions, and a Monte Carlo probabilistic sensitivity analysis was performed to examine the robustness of the model. RESULTS In this decision-analytical model, no prophylactic anticoagulation maximized maternal utilities. Clinical outcomes among a theoretical cohort of 100,000 women are shown in the Table. The 1- and 2-way sensitivity analyses supported this conclusion. Monte Carlo probabilistic sensitivity analysis indicated that no prophylaxis was the preferred choice in 56% of simulations, unfractionated heparin in 34% of simulations, and low-molecular-weight heparin in 10% of simulations. CONCLUSION Our results do not support the routine use of prophylactic anticoagulation in women admitted to the hospital for preterm premature rupture of membranes. These findings can be used to inform clinical decisions when admitting low-risk singleton pregnancies to the hospital in the setting of preterm premature rupture of membranes.
Collapse
Affiliation(s)
- Cassandra B Iroz
- Feinberg School of Medicine, Northwestern University, Chicago, IL.
| | - Carly M Dahl
- Feinberg School of Medicine, Department of Obstetrics and Gynecology, Northwestern University, Chicago, IL
| | - Irina R Cassimatis
- Feinberg School of Medicine, Department of Obstetrics and Gynecology, Northwestern University, Chicago, IL
| | - Annie B Wescott
- Feinberg School of Medicine, Galter Health Sciences Library, Northwestern University, Chicago, IL
| | - Emily S Miller
- Feinberg School of Medicine, Division of Maternal Fetal Medicine, Department of Obstetrics and Gynecology, Northwestern University, Chicago, IL
| |
Collapse
|
3
|
Esin S, Tohma YA, Alay İ, Guden M, Colak E, Demirel N, Bas AY, Serdar Yalvac E, Kandemir O. Comparison of placental α microglobulin-1 protein assay (Amnisure) with speculum examination for the diagnosis of premature preterm rupture of membranes (PPROM): a clinical evaluation. J OBSTET GYNAECOL 2020; 41:703-707. [PMID: 32835549 DOI: 10.1080/01443615.2020.1789949] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/23/2022]
Abstract
In this study, we aimed to compare the clinical outcomes of Premature Preterm Rupture of Membranes (PPROM) cases diagnosed by classical speculum examination and by placental alpha microglobulin-1 protein (PAMG-1) assay. The medical records of all patients with singleton pregnancies that were diagnosed with PPROM were retrospectively reviewed. Singleton pregnancies with PPROM diagnosis that was confirmed either by direct visualisation of amniotic fluid leaking through the cervix or by placental alpha microglobulin-1 protein (PAMG-1) assay if no amniotic fluid leakage was documented were included in the study. Demographics, prenatal and postnatal characteristics were reviewed from the medical charts and were recorded. The study included 138 pregnancies with PPROM; 111 patients in clinical speculum examination group and 27 in PAMG-1 assay group. There were no significant differences in maternal and pregnancy characteristics between the clinical speculum examination and PAMG-1 assay groups. Foetal outcomes were comparable between clinical speculum examination and PAMG-1 assay groups. In the clinical speculum examination group, there were nine (8.1%) chorioamnionitis cases, however, there were no chorioamnionitis cases in the PAMG-1 assay group during the latency period (p = .21).Impact statementWhat is already known on this subject? Placental alpha microglobulin-1 protein assay uses immunochromatography method to detect trace amount of placental alpha microglobulin-1 protein in vaginal fluids and has high sensitivity and specificity for ROM diagnosis. However, to the best of our knowledge, the clinical outcome of ROM cases detected by classical speculum examination and by placental alpha microglobulin-1 protein assay has not been compared in the literature previously.What do the results of this study add? Although statistically insignificant, cases diagnosed by PAMG-1 assay had lower risk of chorioamnionitis during latency period.What are the implications of these findings for clinical practice and/or further research? Whether cases diagnosed by PAMG-1 assay represent a milder form of rupture of membranes than cases diagnosed by classical speculum examination group warrants further research.
Collapse
Affiliation(s)
- Sertac Esin
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - Yusuf Aytac Tohma
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - İsmail Alay
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - Mahmut Guden
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - Eser Colak
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - Nihal Demirel
- Department of Neoatology, Ankara Yıldırım Beyazıt University, Ankara, Turkey
| | - Ahmet Yagmur Bas
- Department of Neoatology, Ankara Yıldırım Beyazıt University, Ankara, Turkey
| | - Ethem Serdar Yalvac
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| | - Omer Kandemir
- Department of Obstetrics and Gynecology, Etlik Zübeyde Hanım Women's Health Teaching and Research Hospital, Ankara, Turkey
| |
Collapse
|
4
|
Baser E, Aydogan Kirmizi D, Ulubas Isik D, Ozdemirci S, Onat T, Serdar Yalvac E, Demirel N, Moraloglu Tekin O. The effects of latency period in PPROM cases managed expectantly. J Matern Fetal Neonatal Med 2020; 33:2274-2283. [PMID: 32089027 DOI: 10.1080/14767058.2020.1731465] [Citation(s) in RCA: 4] [Impact Index Per Article: 1.0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Indexed: 10/24/2022]
Abstract
Objective: Preterm premature rupture of membranes (PPROM), associated with prematurity, is an important obstetric complication that may cause neonatal mortality and morbidity. The optimal delivery time is controversial in cases with the expectant approach. The fetal effects of long-term exposure to PPROM are unknown. This study aimed to evaluate the maternal and fetal outcomes of expectantly-managed PPROM cases with different latency periods at 240/7-346/7 weeks of gestation.Material and method: The study group consisted of 206 patients at 240/7-346/7 weeks of gestation who met the inclusion criteria. Patients were divided into three groups according to their weeks of PPROM diagnosis as 240/7-286/7, 290/7-316/7, and 320/7-346/7. The period from membrane rupture to delivery was defined as the latency period and divided into three subgroups as 3-7 days, 8-13 days and ≥14 days. In addition to the demographic characteristics of the patients, maternal and obstetric complications, primary and secondary neonatal outcomes were compared between the groups. Primary neonatal outcomes were determined in terms of pathological Apgar scores (<5 at minute 1, <7 at minute 5), requiring resuscitation, admission to Neonatal Intensive Care Unit (NICU) and NICU length of stay. Secondary neonatal outcomes were determined in terms of respiratory distress syndrome, bronchopulmonary dysplasia, intraventricular hemorrhage, retinopathy of prematurity, necrotizing enterocolitis, patent ductus arteriosus, periventricular leukomalacia, and neonatal sepsis. In addition, for the prediction of morbidity and mortality, newborns were evaluated by SNAPPE II (Score for Neonatal Acute Physiology with Perinatal extension-II) consisting of the combination of biochemical and physiological parameters, using the parameters including mean blood pressure (mm/Hg), corporal temperature (°C), PO2/FiO2 ratio, lowest serum pH, multiple seizures, urine output (ml/kg/hr), Apgar score, birth weight, and small for gestational age. The higher the score of SNAPPE II, the higher the morbidity and mortality risk of neonates. For the statistical analysis, the Kruskal Wallis and one-way ANOVA tests were utilized for the numerical data. Categorical data were compared using the chi-square test. The receiver operating characteristic (ROC) test was used to determine the threshold value of the data affecting neonatal morbidity.Results: The mean PPROM week was found to be 29.7 ± 3.0 weeks and the mean delivery week was 31.8 ± 2.5 weeks. The mean latency period for all the patients was 15.1 ± 13.8 days. Clinic chorioamnionitis was observed in 17% of the cases. The lowest chorioamnionitis rate (8.6%) was in the 3-7-day latency period group. Total complications were significantly lower in the 290/7-316/7 week PPROM group in which the latency period was ≥14 days, compared to those in 3-7 days and 8-13 days (p = .001). Total complications were lower in the < 32 weeks PPROM groups in which the latency period was ≥14 days compared to those obtained in 3-7 days and 8-13 days. There was no significant difference between the latency period and total complications after 32 weeks (p = .422). The best discriminative cutoff value of SNAPPE-II for neonatal morbidity was 11.0 (sensitivity 82%, specificity 80%). In the present study, the optimal latency period for the best neonatal outcomes was found to be 34.5 days (sensitivity 70% and specificity 84%) between weeks 240/7-286/7, and 11.0 days between weeks 290/7-316/7 (sensitivity 68% and specificity 85%).Conclusions: Our findings indicated that a long latency period did not increase neonatal morbidity and there was no increase in neonatal complications after 32 weeks of the gestational period compared to those obtained before 32 weeks.
Collapse
Affiliation(s)
- Emre Baser
- Department of Obstetrics and Gynecology, Faculty of Medicine, Yozgat Bozok University, Yozgat, Turkey
| | - Demet Aydogan Kirmizi
- Department of Obstetrics and Gynecology, Faculty of Medicine, Yozgat Bozok University, Yozgat, Turkey
| | - Dilek Ulubas Isik
- Department of Neonatology, Etlik Zubeyde Hanim Women's Health Training and Research Hospital, Health Science University, Ankara, Turkey
| | - Safak Ozdemirci
- Department of Obstetrics and Gynecology, Etlik Zubeyde Hanim Women's Health Training and Research Hospital, Health Science University, Ankara, Turkey
| | - Taylan Onat
- Department of Obstetrics and Gynecology, Faculty of Medicine, Yozgat Bozok University, Yozgat, Turkey
| | - Ethem Serdar Yalvac
- Department of Obstetrics and Gynecology, Faculty of Medicine, Yozgat Bozok University, Yozgat, Turkey
| | - Nihal Demirel
- Department of Neonatology, Etlik Zubeyde Hanim Women's Health Training and Research Hospital, Health Science University, Ankara, Turkey
| | - Ozlem Moraloglu Tekin
- Department of Obstetrics and Gynecology, Etlik Zubeyde Hanim Women's Health Training and Research Hospital, Health Science University, Ankara, Turkey
| |
Collapse
|