Application of Bayesian network modeling to pathology informatics

Using Bayesian networks with Tabu-search algorithm to explore risk factors for hyperhomocysteinemia

Hyperhomocysteinemia (HHcy) is a condition closely associated with cardiovascular and cerebrovascular diseases. Detecting its risk factors and taking some relevant interventions still represent the top priority to lower its prevalence. Yet, in discussing risk factors, Logistic regression model is usually adopted but accompanied by some defects. In this study, a Tabu Search-based BNs was first constructed for HHcy and its risk factors, and the conditional probability between nodes was calculated using Maximum Likelihood Estimation. Besides, we tried to compare its performance with Hill Climbing-based BNs and Logistic regression model in risk factor detection and discuss its prospect in clinical practice. Our study found that Age, sex, α1-microgloblobumin to creatinine ratio, fasting plasma glucose, diet and systolic blood pressure represent direct risk factors for HHcy, and smoking, glycosylated hemoglobin and BMI constitute indirect risk factors for HHcy. Besides, the performance of Tabu Search-based BNs is better than Hill Climbing-based BNs. Accordingly, BNs with Tabu Search algorithm could be a supplement for Logistic regression, allowing for exploring the complex network relationship and the overall linkage between HHcy and its risk factors. Besides, Bayesian reasoning allows for risk prediction of HHcy, which is more reasonable in clinical practice and thus should be promoted.

A Bayesian network model for prediction of low or failed fertilization in assisted reproductive technology based on a large clinical real-world data

STUDY QUESTION

To construct prediction models based on the Bayesian network (BN) learning method for the probability of fertilization failure (including low fertilization rate [LRF] and total fertilization failure [TFF]) in assisted reproductive technology (ART) treatment. A BN model was developed to predict TFF/LFR. The model showed relatively high calibration in external validation, which could facilitate the identification of risk factors for fertilization disorders and improve the efficiency of in vitro fertilization/intracytoplasmic sperm injection (IVF/ICSI) treatment.

WHAT IS KNOWN ALREADY

The prediction of TFF/LFR is very complex. Although some studies attempted to construct prediction models for TFF/LRF, most of the reported models were based on limited variables and traditional regression-based models, which are unsuitable for analyzing real-world clinical data. Therefore, none of the reported models have been widely used in routine clinical practice. To date, BN modeling analysis is a prominent and increasingly popular machine learning method that is powerful in dealing with dynamic and complex real-world data.

STUDY DESIGN, SIZE, DURATION

A retrospective study was performed with 106,640 fresh embryo IVF/ICSI cycles from 2009 to 2019 in one of China's largest reproductive health centers.

PARTICIPANTS/MATERIALS, SETTING, METHODS

A total of 106, 640 cycles were included in this study, including 97,102 controls, 4,339 LFR cases, and 5,199 TFF cases. Twenty-four predictors were initially included, including 13 female-related variables, five male-related variables, and six variables related to IVF/ICSI treatment. BN modeling analysis with tenfold cross-validation was performed to construct the predictive model for TFF/LFR. The receiver operating characteristic (ROC) curves and the corresponding area under the curves (AUCs) were used to evaluate the performance of the BN model.

MAIN RESULTS AND THE ROLE OF CHANCE

All twenty-four predictors were first organized into seven hierarchical layers in a theoretical BN model, according to prior knowledge from previous literature and clinical practice. A machine-learning BN model was generated based on real-world clinical data, containing a total of eighteen predictors, of which the infertility type, ART method, and number of retrieved oocytes directly influence the probabilities of LFR/TFF. The prediction accuracy of the BN model was 91.7%. The AUC of the TFF versus control groups was 0.779 (95% CI: 0.766-0.791), with a sensitivity of 71.2% and specificity of 70.1%; the AUC of of TFF versus LFR groups was 0.807 (95% CI: 0.790-0.824), with a sensitivity of 49.0% and specificity of 99.0%.

LIMITATIONS, REASON FOR CAUTION

First, our study was based on clinical data from a single center, and the results of this study should be further verified by external data. In addition, some critical data (e.g., the detailed IVF laboratory parameters of the sperm and oocytes used for insemination) were not available in this study, which should be given full consideration when further improving the performance of the BN model.

WIDER IMPLICATIONS OF THE FINDINGS

Based on extensive clinical real-world data, we developed a BN model to predict the probabilities of fertilization failures in ART, which provides new clues for clinical decision-making support for clinicians in formulating personalized treatment plans and further improving ART treatment outcomes.

STUDY FUNDING/COMPETING INTEREST(S)

Dr. Y. Wang was supported by grants from the Beijing Municipal Science & Technology Commission (Z191100006619086). We declare that there are no conflicts of interest.

TRIAL REGISTRATION NUMBER

N/A.

Medical implementation practice and its medical performance evaluation of a giant makeshift hospital during the COVID-19 pandemic: An innovative model response to a public health emergency in Shanghai, China

Introduction

In confronting the sudden COVID-19 epidemic, China and other countries have been under great pressure to block virus transmission and reduce fatalities. Converting large-scale public venues into makeshift hospitals is a popular response. This addresses the outbreak and can maintain smooth operation of a country or region's healthcare system during a pandemic. However, large makeshift hospitals, such as the Shanghai New International Expo Center (SNIEC) makeshift hospital, which was one of the largest makeshift hospitals in the world, face two major problems: Effective and precise transfer of patients and heterogeneity of the medical care teams.

Methods

To solve these problems, this study presents the medical practices of the SNIEC makeshift hospital in Shanghai, China. The experiences include constructing two groups, developing a medical management protocol, implementing a multi-dimensional management mode to screen patients, transferring them effectively, and achieving homogeneous quality of medical care. To evaluate the medical practice performance of the SNIEC makeshift hospital, 41,941 infected patients were retrospectively reviewed from March 31 to May 23, 2022. Multivariate logistic regression method and a tree-augmented naive (TAN) Bayesian network mode were used.

Results

We identified that the three most important variables were chronic disease, age, and type of cabin, with importance values of 0.63, 0.15, and 0.11, respectively. The constructed TAN Bayesian network model had good predictive values; the overall correct rates of the model-training dataset partition and test dataset partition were 99.19 and 99.05%, respectively, and the respective values for the area under the receiver operating characteristic curve were 0.939 and 0.957.

Conclusion

The medical practice in the SNIEC makeshift hospital was implemented well, had good medical care performance, and could be copied worldwide as a practical intervention to fight the epidemic in China and other developing countries.

Optimization of anesthetic decision-making in ERAS using Bayesian network

Enhanced recovery after surgery (ERAS) can accelerate patient recovery. However, little research has been done on optimizing the ERAS-related measures and how the measures interact with each other. The Bayesian network (BN) is a graphical model that describes the dependencies between variables and is also a model for uncertainty reasoning. In this study, we aimed to develop a method for optimizing anesthetic decisions in ERAS and then investigate the relationship between anesthetic decisions and outcomes. First, assuming that the indicators used were independent, the effects of combinations of single indicators were analyzed based on BN. Additionally, the impact indicators for outcomes were selected with statistical tests. Then, based on the previously selected indicators, the Bayesian network was constructed using the proposed structure learning method based on Strongly Connected Components (SCC) Local Structure determination by Hill Climbing Twice (LSHCT) and adjusted according to the expert’s knowledge. Finally, the relationship is analyzed. The proposed method is validated by the real clinical data of patients with benign gynecological tumors from 3 hospitals in China. Postoperative length of stay (LOS) and total cost (TC) were chosen as the outcomes. Experimental results show that the ERAS protocol has some pivotal indicators influencing LOS and TC. Identifying the relationship between these indicators can help anesthesiologists optimize the ERAS protocol and make individualized decisions.

Identifying Associations in Minimum Inhibitory Concentration Values of Escherichia coli Samples Obtained From Weaned Dairy Heifers in California Using Bayesian Network Analysis

Objective

Many antimicrobial resistance (AMR) studies in both human and veterinary medicine use traditional statistical methods that consider one bacteria and one antibiotic match at a time. A more robust analysis of AMR patterns in groups of animals is needed to improve on traditional methods examining antibiotic resistance profiles, the associations between the patterns of resistance or reduced susceptibility for all isolates in an investigation. The use of Bayesian network analysis can identify associations between distributions; this investigation seeks to add to the growing body of AMR pattern research by using Bayesian networks to identify relationships between susceptibility patterns in Escherichia coli (E. coli) isolates obtained from weaned dairy heifers in California.

Methods

A retrospective data analysis was performed using data from rectal swab samples collected from 341 weaned dairy heifers on six farms in California and selectively cultured for E. coli. Antibiotic susceptibility tests for 281 isolates against 15 antibiotics were included. Bayesian networks were used to identify joint patterns of reduced susceptibility, defined as an increasing trend in the minimum inhibitory concentration (MIC) values. The analysis involved learning the network structure, identifying the best fitting graphical mode, and learning the parameters in the final model to quantify joint probabilities.

Results

The graph identified that as susceptibility to one antibiotic decreases, so does susceptibility to other antibiotics in the same or similar class. The following antibiotics were connected in the final graphical model: ampicillin was connected to ceftiofur; spectinomycin was connected with trimethoprim-sulfamethoxazole, and this association was mediated by farm; florfenicol was connected with tetracycline.

Conclusions

Bayesian network analysis can elucidate complex relationships between MIC patterns. MIC values may be associated within and between drug classes, and some associations may be correlated with farm of sample origin. Treating MICs as discretized variables and testing for joint associations in trends may overcome common research problems surrounding the lack of clinical breakpoints.

PI Prob: A risk prediction and clinical guidance system for evaluating patients with recurrent infections

Background

Primary immunodeficiency diseases represent an expanding set of heterogeneous conditions which are difficult to recognize clinically. Diagnostic rates outside of the newborn period have not changed appreciably. This concern underscores a need for novel methods of disease detection.

Objective

We built a Bayesian network to provide real-time risk assessment about primary immunodeficiency and to facilitate prescriptive analytics for initiating the most appropriate diagnostic work up. Our goal is to improve diagnostic rates for primary immunodeficiency and shorten time to diagnosis. We aimed to use readily available health record data and a small training dataset to prove utility in diagnosing patients with relatively rare features.

Methods

We extracted data from the Texas Children’s Hospital electronic health record on a large population of primary immunodeficiency patients (n = 1762) and appropriately-matched set of controls (n = 1698). From the cohorts, clinically relevant prior probabilities were calculated enabling construction of a Bayesian network probabilistic model(PI Prob). Our model was constructed with clinical-immunology domain expertise, trained on a balanced cohort of 100 cases-controls and validated on an unseen balanced cohort of 150 cases-controls. Performance was measured by area under the receiver operator characteristic curve (AUROC). We also compared our network performance to classic machine learning model performance on the same dataset.

Results

PI Prob was accurate in classifying immunodeficiency patients from controls (AUROC = 0.945; p<0.0001) at a risk threshold of ≥6%. Additionally, the model was 89% accurate for categorizing validation cohort members into appropriate International Union of Immunological Societies diagnostic categories. Our network outperformed 3 other machine learning models and provides superior transparency with a prescriptive output element.

Conclusion

Artificial intelligence methods can classify risk for primary immunodeficiency and guide management. PI Prob enables accurate, objective decision making about risk and guides the user towards the appropriate diagnostic evaluation for patients with recurrent infections. Probabilistic models can be trained with small datasets underscoring their utility for rare disease detection given appropriate domain expertise for feature selection and network construction.

Demystifying machine learning: a primer for physicians

Machine learning is a tool for analysing digitised data sets and formulating predictions that can optimise clinical decision-making. It aims to identify complex patterns in large data sets and encode them into models that can then classify new unseen cases or make predictions on new data. Machine learning methods take several forms and individual models can be of many different types. More than 50 models have been approved for use in routine healthcare, and the numbers continue to grow exponentially. The reliability and robustness of any model depends on multiple factors, including the quality and quantity of the data used to develop the models, and the selection of features in the data considered most important to maximising accuracy. In ensuring models are safe, effective and reproducible in routine care, physicians need to have some understanding of how these models are developed and evaluated, and to collaborate with data and computer scientists in their design and validation. This narrative review introduces principles, methods and examples of machine learning in a way that does not require mastery of highly complex statistical and computational concepts.

Personalized Medicine and Cervical Screening: Development of Individualized Quantitative Risk Assessments for Cervical Adenocarcinoma and Adenocarcinoma in situ

INTRODUCTION

Cervical screening has decreased the incidence of cervical carcinoma around the world primarily by preventing cervical squamous carcinoma, with significantly less measurable protective benefits in prevention of cervical adenocarcinoma. In this study, we apply Bayesian modeling of cervical clinical, screening, and biopsy data from a large integrated health system to explore the feasibility of calculating personalized risk assessments on screened system patients for subsequent histopathologic diagnoses of invasive cervical adenocarcinoma (AdCa) or cervical adenocarcinoma in situ (AIS).

MATERIALS AND METHODS

Diagnoses of cervical AIS or AdCa rendered between 2005 and 2018 were identified in our large health system database with 1,053,713 cytology results, 354,843 high-risk (hr) human papillomavirus (HPV) test results, and 99,012 cervical histopathologic results. Using our continuously updated Bayesian cervical cancer screening model which includes clinical data, cervical screening results, and cervical biopsy results, we projected quantitative estimates of patients' 5-year cumulative risk for cervical AIS or AdCa.

RESULTS

161 patients were identified with AIS (ages 17-75, mean 37 years), and 99 patients had diagnoses of cervical AdCa (ages 26-91, mean 48 years). Quantitative Bayesian 5-year cumulative risk projections for diagnoses of cervical AdCa or AIS in patients with different cervical screening test and biopsy histories were determined. The highest patient risk projections for subsequent cervical AdCa and/or AIS histopathologic diagnoses were associated with prior cervical screening test results of HPV-positive atypical glandular cells. Prior squamous cytologic abnormalities were associated with lower risk estimates. Prior histopathologic diagnoses of squamous abnormalities also influenced quantitative risk. A prior histopathologic diagnosis of AIS was associated with a very low risk of subsequent AdCa, consistent with effective excisional treatment. AdCa risk was greatest in women aged 30-65 years with prior CIN3 biopsy results, whereas AIS risk was greatest in women <30.

CONCLUSION

Prevention of cervical AdCa in screened patients remains a major challenge for cervical screening. Individualized risk projections for cervical glandular neoplasia reflecting patient age, prior cervical screening test results, and prior cervical biopsy history are feasible using Bayesian modeling of health system data.

Are CIN3 risk or CIN3+ risk measures reliable surrogates for invasive cervical cancer risk?

•Discuss ASCCP guideline. •CIN3 reliable surrogates for cervical cancer? •The Pittsburgh Cervical Cancer Screening Model.

A decision support tool to find the best cyclosporine dose when switching from intravenous to oral route in pediatric stem cell transplant patients

PURPOSE

Managing the pharmacokinetic variability of immunosuppressive drugs after pediatric hematopoietic stem cell transplantation (HSCT) is a clinical challenge. Thus, the aim of our study was to design and validate a decision support tool predicting the best first cyclosporine oral dose to give when switching from intravenous route.

METHODS

We used 10-years pediatric HSCT patients' dataset from 2008 to 2018. A tree-augmented naïve Bayesian network model (method belonging to artificial intelligence) was built with data from the first eight-years, and validated with data from the last two.

RESULTS

The Bayesian network model obtained showed good prediction performances, both after a 10-fold cross-validation and external validation, with respectively an AUC-ROC of 0.89 and 0.86, a percentage of misclassified patients of 28.7% and 35.2%, a true positive rate of 0.71 and 0.65, and a false positive rate of 0.12 and 0.14 respectively.

CONCLUSION

The final model allows the prediction of the most likely cyclosporine oral dose to reach the therapeutic target specified by the clinician. The clinical impact of using this model needs to be prospectively warranted. Respecting the decision support tool terms of use is necessary as well as remaining critical about the prediction by confronting it with the clinical context.

Individualized Bayesian Risk Assessment for Cervical Squamous Neoplasia

Background:

Cervical screening could potentially be improved by better stratifying individual risk for the development of cervical cancer or precancer, possibly even allowing follow-up of individual patients differently than proposed under current guidelines that focus primarily on recent screening test results. We explore the use of a Bayesian decision science model to quantitatively stratify individual risk for the development of cervical squamous neoplasia.

Materials and Methods:

We previously developed a dynamic multivariate Bayesian network model that uses cervical screening and histopathologic data collected over 13 years in our system to quantitatively estimate the risk of individuals for the development of cervical precancer or invasive cervical cancer. The database includes 1,126,048 liquid-based cytology test results belonging to 389,929 women. From-the-vial, high risk human papilloma virus (HPV) test results and follow-up gynecological surgical procedures were available on 33.6% and 12% of these results (378,896 and 134,727), respectively.

Results:

Historical data impacted 5-year cumulative risk for both histopathologic cervical intraepithelial neoplasia 3 (CIN3) and squamous cell carcinoma (SCC) diagnoses. The risk was highest in patients with prior high grade squamous intraepithelial lesion cytology results. Persistent abnormal cervical screening test results, either cytologic or HPV results, were associated with variable increasing risk for squamous neoplasia. Risk also increased with prior histopathologic diagnoses of precancer, including CIN2, CIN3, and adenocarcinoma in situ.

Conclusions:

Bayesian modeling allows for individualized quantitative risk assessments of system patients for histopathologic diagnoses of significant cervical squamous neoplasia, including very rare outcomes such as SCC.