Predicting Hepatocellular Carcinoma With Minimal Features From Electronic Health Records: Development of a Deep Learning Model

BACKGROUND

Hepatocellular carcinoma (HCC), usually known as hepatoma, is the third leading cause of cancer mortality globally. Early detection of HCC helps in its treatment and increases survival rates.

OBJECTIVE

The aim of this study is to develop a deep learning model, using the trend and severity of each medical event from the electronic health record to accurately predict the patients who will be diagnosed with HCC in 1 year.

METHODS

Patients with HCC were screened out from the National Health Insurance Research Database of Taiwan between 1999 and 2013. To be included, the patients with HCC had to register as patients with cancer in the catastrophic illness file and had to be diagnosed as a patient with HCC in an inpatient admission. The control cases (non-HCC patients) were randomly sampled from the same database. We used age, gender, diagnosis code, drug code, and time information as the input variables of a convolution neural network model to predict those patients with HCC. We also inspected the highly weighted variables in the model and compared them to their odds ratio at HCC to understand how the predictive model works.

RESULTS

We included 47,945 individuals, 9553 of whom were patients with HCC. The area under the receiver operating curve (AUROC) of the model for predicting HCC risk 1 year in advance was 0.94 (95% CI 0.937-0.943), with a sensitivity of 0.869 and a specificity 0.865. The AUROC for predicting HCC patients 7 days, 6 months, 1 year, 2 years, and 3 years early were 0.96, 0.94, 0.94, 0.91, and 0.91, respectively.

CONCLUSIONS

The findings of this study show that the convolutional neural network model has immense potential to predict the risk of HCC 1 year in advance with minimal features available in the electronic health records.

Development of a Web-Based System for Exploring Cancer Risk With Long-term Use of Drugs: Logistic Regression Approach

Background

Existing epidemiological evidence regarding the association between the long-term use of drugs and cancer risk remains controversial.

Objective

We aimed to have a comprehensive view of the cancer risk of the long-term use of drugs.

Methods

A nationwide population-based, nested, case-control study was conducted within the National Health Insurance Research Database sample cohort of 1999 to 2013 in Taiwan. We identified cases in adults aged 20 years and older who were receiving treatment for at least two months before the index date. We randomly selected control patients from the patients without a cancer diagnosis during the 15 years (1999-2013) of the study period. Case and control patients were matched 1:4 based on age, sex, and visit date. Conditional logistic regression was used to estimate the association between drug exposure and cancer risk by adjusting potential confounders such as drugs and comorbidities.

Results

There were 79,245 cancer cases and 316,980 matched controls included in this study. Of the 45,368 associations, there were 2419, 1302, 662, and 366 associations found statistically significant at a level of P<.05, P<.01, P<.001, and P<.0001, respectively. Benzodiazepine derivatives were associated with an increased risk of brain cancer (adjusted odds ratio [AOR] 1.379, 95% CI 1.138-1.670; P=.001). Statins were associated with a reduced risk of liver cancer (AOR 0.470, 95% CI 0.426-0.517; P<.0001) and gastric cancer (AOR 0.781, 95% CI 0.678-0.900; P<.001). Our web-based system, which collected comprehensive data of associations, contained 2 domains: (1) the drug and cancer association page and (2) the overview page.

Conclusions

Our web-based system provides an overview of comprehensive quantified data of drug-cancer associations. With all the quantified data visualized, the system is expected to facilitate further research on cancer risk and prevention, potentially serving as a stepping-stone to consulting and exploring associations between the long-term use of drugs and cancer risk.

Machine Learning Approach to Reduce Alert Fatigue Using a Disease Medication-Related Clinical Decision Support System: Model Development and Validation

Background

Computerized physician order entry (CPOE) systems are incorporated into clinical decision support systems (CDSSs) to reduce medication errors and improve patient safety. Automatic alerts generated from CDSSs can directly assist physicians in making useful clinical decisions and can help shape prescribing behavior. Multiple studies reported that approximately 90%-96% of alerts are overridden by physicians, which raises questions about the effectiveness of CDSSs. There is intense interest in developing sophisticated methods to combat alert fatigue, but there is no consensus on the optimal approaches so far.

Objective

Our objective was to develop machine learning prediction models to predict physicians’ responses in order to reduce alert fatigue from disease medication–related CDSSs.

Methods

We collected data from a disease medication–related CDSS from a university teaching hospital in Taiwan. We considered prescriptions that triggered alerts in the CDSS between August 2018 and May 2019. Machine learning models, such as artificial neural network (ANN), random forest (RF), naïve Bayes (NB), gradient boosting (GB), and support vector machine (SVM), were used to develop prediction models. The data were randomly split into training (80%) and testing (20%) datasets.

Results

A total of 6453 prescriptions were used in our model. The ANN machine learning prediction model demonstrated excellent discrimination (area under the receiver operating characteristic curve [AUROC] 0.94; accuracy 0.85), whereas the RF, NB, GB, and SVM models had AUROCs of 0.93, 0.91, 0.91, and 0.80, respectively. The sensitivity and specificity of the ANN model were 0.87 and 0.83, respectively.

Conclusions

In this study, ANN showed substantially better performance in predicting individual physician responses to an alert from a disease medication–related CDSS, as compared to the other models. To our knowledge, this is the first study to use machine learning models to predict physician responses to alerts; furthermore, it can help to develop sophisticated CDSSs in real-world clinical settings.

Prediction of fatty liver disease using machine learning algorithms

BACKGROUND AND OBJECTIVE

Fatty liver disease (FLD) is a common clinical complication; it is associated with high morbidity and mortality. However, an early prediction of FLD patients provides an opportunity to make an appropriate strategy for prevention, early diagnosis and treatment. We aimed to develop a machine learning model to predict FLD that could assist physicians in classifying high-risk patients and make a novel diagnosis, prevent and manage FLD.

METHODS

We included all patients who had an initial fatty liver screening at the New Taipei City Hospital between 1st and 31st December 2009. Classification models such as random forest (RF), Naïve Bayes (NB), artificial neural networks (ANN), and logistic regression (LR) were developed to predict FLD. The area under the receiver operating characteristic curve (ROC) was used to evaluate performances among the four models.

RESULTS

A total of 577 patients were included in this study; of those 377 patients had fatty liver. The area under the receiver operating characteristic (AUROC) of RF, NB, ANN, and LR with 10 fold-cross validation was 0.925, 0.888, 0.895, and 0.854 respectively. Additionally, The accuracy of RF, NB, ANN, and LR 87.48, 82.65, 81.85, and 76.96%.

CONCLUSION

In this study, we developed and compared the four classification models to predict fatty liver disease accurately. However, the random forest model showed higher performance than other classification models. Implementation of a random forest model in the clinical setting could help physicians to stratify fatty liver patients for primary prevention, surveillance, early treatment, and management.

Viral warts (Human Papilloma Virus) as a potential risk for breast cancer among younger females

INTRODUCTION

There have been several reports on the role of human papillomavirus (HPV) in the etiology of breast cancer. To our knowledge, this is first study to use disease-disease association data-mining approach to analyzing viral warts and breast cancer to be conducted in Taiwanese population.

MATERIALS AND METHODS

We analyzed the Taiwan's National Health Insurance database (NHIDM data comprising of 23 million patient data) to examine the association between viral warts and female breast carcinoma. The patients were categorized into three groups: breast cancer only, viral warts only, and those with both breast cancer and viral warts. The Cox proportion hazard regression analysis was used to measure the effect of HPV on the time to breast cancer diagnosis. Multivariable analyzes and stratified analyzes using hazard ratios (HRs) were presented with 95% confidence intervals (CIs) after adjusting for age, and CCI.

RESULT

Among 807,578 HPV population, we identified 6014 breast cancer cases. The HPV group was associated with a significantly higher risk of developing breast cancer (HR, 1.18; 95% CI, 1.15-1.21; p< 0.001) compared with the non-HPV group. HPV patients with age group 18-39 was slightly higher risk of breast cancer occurrence (HR, 1.07; 95% CI, 1.01-1.13; p<.05). The risk of breast cancer in 10-year incidence was 7% higher for females less than 40 years and 23% for over 40 year's patients when compared with non-HPV patients of the same age group.

CONCLUSION

Our study indicates that women who develop viral warts are at a significantly higher risk of developing breast cancer than women who have not diagnosed with viral warts. Thus, the presence of viral warts is a potential risk to breast cancer. Therefore, we suggest patients diagnosed with viral warts may get early screening for breast cancer.

Cancer-disease associations: A visualization and animation through medical big data

OBJECTIVE

Cancer is the primary disease responsible for death and disability worldwide. Currently, prevention and early detection represents the best hope for cure. Knowing the expected diseases that occur with a particular cancer in advance could lead to physicians being able to better tailor their treatment for cancer. The aim of this study was to build an animated visualization tool called as Cancer Associations Map Animation (CAMA), to chart the association of cancers with other disease over time.

METHODS

The study population was collected from the Taiwan National Health Insurance Database during the period January 2000 to December 2002, 782 million outpatient visits were used to compute the associations of nine major cancers with other diseases. A motion chart was used to quantify and visualize the associations between diseases and cancers.

RESULTS

The CAMA motion chart that was built successfully facilitated the observation of cancer-disease associations across ages and genders. The CAMA system can be accessed online at http://203.71.86.98/web/runq16.html.

CONCLUSION

The CAMA animation system is an animated medical data visualization tool which provides a dynamic, time-lapse, animated view of cancer-disease associations across different age groups and gender. Derived from a large, nationwide healthcare dataset, this exploratory data analysis tool can detect cancer comorbidities earlier than is possible by manual inspection. Taking into account the trajectory of cancer-specific comorbidity development may facilitate clinicians and healthcare researchers to more efficiently explore early stage hypotheses, develop new cancer treatment approaches, and identify potential effect modifiers or new risk factors associated with specific cancers.

A richly interactive exploratory data analysis and visualization tool using electronic medical records

Background

Electronic medical records (EMRs) contain vast amounts of data that is of great interest to physicians, clinical researchers, and medial policy makers. As the size, complexity, and accessibility of EMRs grow, the ability to extract meaningful information from them has become an increasingly important problem to solve.

Methods

We develop a standardized data analysis process to support cohort study with a focus on a particular disease. We use an interactive divide-and-conquer approach to classify patients into relatively uniform within each group. It is a repetitive process enabling the user to divide the data into homogeneous subsets that can be visually examined, compared, and refined. The final visualization was driven by the transformed data, and user feedback direct to the corresponding operators which completed the repetitive process. The output results are shown in a Sankey diagram-style timeline, which is a particular kind of flow diagram for showing factors’ states and transitions over time.

Results

This paper presented a visually rich, interactive web-based application, which could enable researchers to study any cohorts over time by using EMR data. The resulting visualizations help uncover hidden information in the data, compare differences between patient groups, determine critical factors that influence a particular disease, and help direct further analyses. We introduced and demonstrated this tool by using EMRs of 14,567 Chronic Kidney Disease (CKD) patients.

Conclusions

We developed a visual mining system to support exploratory data analysis of multi-dimensional categorical EMR data. By using CKD as a model of disease, it was assembled by automated correlational analysis and human-curated visual evaluation. The visualization methods such as Sankey diagram can reveal useful knowledge about the particular disease cohort and the trajectories of the disease over time.