Development and Validation of a Deep Learning Based Diabetes Prediction System Using a Nationwide Population-Based Cohort

Background@#Previously developed prediction models for type 2 diabetes mellitus (T2DM) have limited performance. We developed a deep learning (DL) based model using a cohort representative of the Korean population. @*Methods@#This study was conducted on the basis of the National Health Insurance Service-Health Screening (NHIS-HEALS) cohort of Korea. Overall, 335,302 subjects without T2DM at baseline were included. We developed the model based on 80% of the subjects, and verified the power in the remainder. Predictive models for T2DM were constructed using the recurrent neural network long short-term memory (RNN-LSTM) network and the Cox longitudinal summary model. The performance of both models over a 10-year period was compared using a time dependent area under the curve. @*Results@#During a mean follow-up of 10.4±1.7 years, the mean frequency of periodic health check-ups was 2.9±1.0 per subject. During the observation period, T2DM was newly observed in 8.7% of the subjects. The annual performance of the model created using the RNN-LSTM network was superior to that of the Cox model, and the risk factors for T2DM, derived using the two models were similar; however, certain results differed. @*Conclusion@#The DL-based T2DM prediction model, constructed using a cohort representative of the population, performs better than the conventional model. After pilot tests, this model will be provided to all Korean national health screening recipients in the future.

Development and Validation of a Deep Learning Based Diabetes Prediction System Using a Nationwide Population-Based Cohort

Background@#Previously developed prediction models for type 2 diabetes mellitus (T2DM) have limited performance. We developed a deep learning (DL) based model using a cohort representative of the Korean population. @*Methods@#This study was conducted on the basis of the National Health Insurance Service-Health Screening (NHIS-HEALS) cohort of Korea. Overall, 335,302 subjects without T2DM at baseline were included. We developed the model based on 80% of the subjects, and verified the power in the remainder. Predictive models for T2DM were constructed using the recurrent neural network long short-term memory (RNN-LSTM) network and the Cox longitudinal summary model. The performance of both models over a 10-year period was compared using a time dependent area under the curve. @*Results@#During a mean follow-up of 10.4±1.7 years, the mean frequency of periodic health check-ups was 2.9±1.0 per subject. During the observation period, T2DM was newly observed in 8.7% of the subjects. The annual performance of the model created using the RNN-LSTM network was superior to that of the Cox model, and the risk factors for T2DM, derived using the two models were similar; however, certain results differed. @*Conclusion@#The DL-based T2DM prediction model, constructed using a cohort representative of the population, performs better than the conventional model. After pilot tests, this model will be provided to all Korean national health screening recipients in the future.

Development and External Validation of a Deep Learning Algorithm for Prognostication of Cardiovascular Outcomes

BACKGROUND AND OBJECTIVES: We aim to explore the additional discriminative accuracy of a deep learning (DL) algorithm using repeated-measures data for identifying people at high risk for cardiovascular disease (CVD), compared to Cox hazard regression.METHODS: Two CVD prediction models were developed from National Health Insurance Service-Health Screening Cohort (NHIS-HEALS): a Cox regression model and a DL model. Performance of each model was assessed in the internal and 2 external validation cohorts in Koreans (National Health Insurance Service-National Sample Cohort; NHIS-NSC) and in Europeans (Rotterdam Study). A total of 412,030 adults in the NHIS-HEALS; 178,875 adults in the NHIS-NSC; and the 4,296 adults in Rotterdam Study were included.RESULTS: Mean ages was 52 years (46% women) and there were 25,777 events (6.3%) in NHIS-HEALS during the follow-up. In internal validation, the DL approach demonstrated a C-statistic of 0.896 (95% confidence interval, 0.886–0.907) in men and 0.921 (0.908–0.934) in women and improved reclassification compared with Cox regression (net reclassification index [NRI], 24.8% in men, 29.0% in women). In external validation with NHIS-NSC, DL demonstrated a C-statistic of 0.868 (0.860–0.876) in men and 0.889 (0.876–0.898) in women, and improved reclassification compared with Cox regression (NRI, 24.9% in men, 26.2% in women). In external validation applied to the Rotterdam Study, DL demonstrated a C-statistic of 0.860 (0.824–0.897) in men and 0.867 (0.830–0.903) in women, and improved reclassification compared with Cox regression (NRI, 36.9% in men, 31.8% in women).CONCLUSIONS: A DL algorithm exhibited greater discriminative accuracy than Cox model approaches.TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT02931500

Development and External Validation of a Deep Learning Algorithm for Prognostication of Cardiovascular Outcomes

BACKGROUND AND OBJECTIVES@#We aim to explore the additional discriminative accuracy of a deep learning (DL) algorithm using repeated-measures data for identifying people at high risk for cardiovascular disease (CVD), compared to Cox hazard regression.@*METHODS@#Two CVD prediction models were developed from National Health Insurance Service-Health Screening Cohort (NHIS-HEALS): a Cox regression model and a DL model. Performance of each model was assessed in the internal and 2 external validation cohorts in Koreans (National Health Insurance Service-National Sample Cohort; NHIS-NSC) and in Europeans (Rotterdam Study). A total of 412,030 adults in the NHIS-HEALS; 178,875 adults in the NHIS-NSC; and the 4,296 adults in Rotterdam Study were included.@*RESULTS@#Mean ages was 52 years (46% women) and there were 25,777 events (6.3%) in NHIS-HEALS during the follow-up. In internal validation, the DL approach demonstrated a C-statistic of 0.896 (95% confidence interval, 0.886–0.907) in men and 0.921 (0.908–0.934) in women and improved reclassification compared with Cox regression (net reclassification index [NRI], 24.8% in men, 29.0% in women). In external validation with NHIS-NSC, DL demonstrated a C-statistic of 0.868 (0.860–0.876) in men and 0.889 (0.876–0.898) in women, and improved reclassification compared with Cox regression (NRI, 24.9% in men, 26.2% in women). In external validation applied to the Rotterdam Study, DL demonstrated a C-statistic of 0.860 (0.824–0.897) in men and 0.867 (0.830–0.903) in women, and improved reclassification compared with Cox regression (NRI, 36.9% in men, 31.8% in women).@*CONCLUSIONS@#A DL algorithm exhibited greater discriminative accuracy than Cox model approaches.TRIAL REGISTRATION: ClinicalTrials.gov Identifier: NCT02931500

Copycat Suicide Induced by Entertainment Celebrity Suicides in South Korea

OBJECTIVE: Throughout the past several years, there have been a number of entertainment celebrity suicides in South Korea. The aim of this study was to investigate the clustering of suicides following celebrities' suicides in South Korea from 2005 to 2008, particularly according to certain characteristics. METHODS: Seven celebrity suicides were examined and defined using the Korean Integrated Newspaper Database System (KINDS) and from these, we considered four affected periods occurring 28 days after each celebrity's suicide. A Poisson time-series autoregression model was used to estimate the relative risk of the total suicide number for each affected period from 2005 to 2008. Logistic regression analysis was performed to investigate whether there were specific increases in the numbers of suicides in subgroups matching each celebrity. RESULTS: There were significant increases in the risk of suicide during the affected periods. Remarkable increases were found in the subgroups matching each celebrity, especially in the group in which all factors (sex, age, and method) were similar. CONCLUSION: This study provides confirmation that a significant copycat effect was induced by these celebrities' suicides, especially among people who identified more with the celebrities. This implies that countermeasures for upright media coverage of celebrity suicides should be discussed and practiced properly in South Korea.

The Value of Elastic Modulus Index as a Novel Surrogate Marker for Cardiovascular Risk Stratification by Dimensional Speckle-Tracking Carotid Ultrasonography

BACKGROUND: Carotid intima media thickness (CIMT) and the presence of carotid plaque have been used for risk stratification of cardiovascular disease (CVD). To date, however, the association between multi-directional functional properties of carotid artery and CVD has not been fully elucidated. We sought to explore the multi-directional mechanics of the carotid artery in relation to cardiovascular risk. METHODS: Four hundred one patients who underwent carotid ultrasound were enrolled between January 2010 and April 2013. A high risk of CVD was defined as more than 20% of 10-year risk based on the Framingham risk score. Using a speckle-tracking technique, the longitudinal and radial movements were analyzed in the B-mode images. Peak longitudinal and radial displacements, strain and strain rate were also measured. Beta stiffness and elastic modulus index were calculated from the radial measurements. RESULTS: Of the overall sample, 13% (52) of patients comprised the high-risk group. In multivariate logistic regression, CIMT and elastic modulus index were independently associated with a high-risk of CVD {odds ratio (OR): 1.810 [95% confidence interval (CI) 1.249–2.622] and OR: 1.767 (95% CI: 1.177–2.652); p = 0.002, 0.006, respectively}. The combination of CIMT and elastic modulus index correlated with a high-risk of CVD more so than CIMT alone. CONCLUSION: The elastic modulus index of the carotid artery might serve as a novel surrogate marker of high-risk CVD. Measurement of the multi-directional mechanics of the carotid artery using the speckle tracking technique has potential for providing further information over conventional B-mode ultrasound for stratification of CVD risk.

Differential Prognostic Value of Coronary Computed Tomography Angiography in Relation to Exercise Electrocardiography in Asymptomatic Subjects

BACKGROUND: To explore the prognostic performance of coronary computed tomography angiography (CCTA) and exercise electrocardiography (XECG) in asymptomatic subjects. METHODS: We retrospectively enrolled 812 (59 +/- 9 years, 60.8% male) asymptomatic subjects who underwent CCTA and XECG concurrently from 2003 through 2009. Subjects were followed-up for major adverse cardiac events (MACE) including cardiac death, nonfatal myocardial infarction, unstable angina, and revascularization after 90 days from index CCTA. RESULTS: The prevalence of occult coronary artery disease (CAD) detected by CCTA was 17.5% and 120 subjects (14.8%) had positive XECG. During a mean follow-up of 37 +/- 16 months, nine subjects experienced MACE. In multivariable Cox-regression analysis, only the presence of CAD by CCTA independently predicted future MACE (p = 0.002). Moreover, CAD by CCTA improved the predictive value when added to a clinical risk factor model using the likelihood ratio test (p < 0.001). Notably, the prognostic value of CCTA persisted in the moderate-to-high-risk group as classified by the Duke treadmill score (p = 0.040), but not in the low-risk group (p = 0.991). CONCLUSION: CCTA provides incremental prognostic benefit over and above XECG in an asymptomatic population, especially for those in a moderate-to-high-risk group as classified by the Duke treadmill score. Risk stratification using XECG may prove valuable for identifying asymptomatic subjects who can benefit from CCTA.

Impact of family history on the presentation and clinical outcomes of coronary heart disease: data from the Korea Acute Myocardial Infarction Registry

BACKGROUND/AIMS: Family history (FHx) of coronary heart disease (CHD) is a well-known risk factor for CHD. However, the prognostic implication of FHx has not been established clearly in patients with acute myocardial infarction (AMI). METHODS: In total, 11,612 patients (8,132 males [70%], age 63 +/- 13 years) with first-onset AMI between November 2005 and June 2008 in a nationwide, prospective, multicenter, online registry (the Korea AMI Registry) were analyzed. Clinical characteristics and outcomes (cardiac death and major adverse cardiac events [MACEs]) were assessed according to the presence of FHx. RESULTS: The patients with FHx were younger and included more males. Male patients with FHx included more current smokers and individuals with poor lipid profiles. In all patients, after adjustment using the Cox proportional hazard model, FHx was related to the risk of MACEs (hazard ratio [HR], 1.41; p = 0.009) and cardiac death (HR, 1.56; p = 0.080). The poor prognostic implication of FHx was further augmented in females and a low risk subset of patients. A significant interaction was only found between male and female patients for composite MACEs (p for interaction = 0.057), and between patients with more risk factors (> or = 2 risk factors) and fewer risk factors for cardiac deaths (p for interaction = 0.008). CONCLUSIONS: FHx may be an independent prognostic predictor, especially in female patients and patients with low-risk profile.

Prognostic Estimation of Advanced Heart Failure With Low Left Ventricular Ejection Fraction and Wide QRS Interval

BACKGROUND AND OBJECTIVES: Cardiac resynchronization therapy (CRT) has been known to improve the outcome of advanced heart failure (HF) but is still underutilized in clinical practice. We investigated the prognosis of patients with advanced HF who were suitable for CRT but were treated with conventional strategies. We also developed a risk model to predict mortality to improve the facilitation of CRT. SUBJECTS AND METHODS: Patients with symptomatic HF with left ventricular ejection fraction 120 ms were consecutively enrolled at cardiovascular hospital. After excluding those patients who had received device therapy, 239 patients (160 males, mean 67+/-11 years) were eventually recruited. RESULTS: During a follow-up of 308+/-236 days, 56 (23%) patients died. Prior stroke, heart rate >90 bpm, serum Na or =1.5 mg/dL were identified as independent factors using Cox proportional hazards regression. Based on the risk model, points were assigned to each of the risk factors proportional to the regression coefficient, and patients were stratified into three risk groups: low- (0), intermediate-(1-5), and high-risk (>5 points). The 2-year mortality rates of each risk group were 5, 31, and 64 percent, respectively. The C statistic of the risk model was 0.78, and the model was validated in a cohort from a different institution where the C statistic was 0.80. CONCLUSION: The mortality of patients with advanced HF who were managed conventionally was effectively stratified using a risk model. It may be useful for clinicians to be more proactive about adopting CRT to improve patient prognosis.

Bioavailability of the amino acid-attached prodrug as a new anti-HIV agent in rats

The primary objective of this study was to compare thepharmacokinetics of a new anti-human immunodeficiencyvirus agent 1-(2-amino-pyridin-4-ylmethyl)-6-(3,5-dimethyl-benzoyl)-5-isopropyl-1H-pyrimidine-2,4-dione (VP-0502)with its amino acid prodrug alanine amide of VP-0502(VP-0502AL), following intravenous and oral administrationsto rats. The plasma concentrations of both analytes wereanalyzed via high-performance liquid chromatographycoupled with photodiode-array detection (HPLC-DAD).When VP-0502 was intravenously administered at 20mg/kg, the analyte appeared in low levels with an AUC of 0.3microg.h/ml, and C0 of 0.2microg/ml in plasma. However, boththe prodrug VP-0502AL and its metabolite VP-0502 appearedat comparatively higher levels following intravenousinjection of VP-0502AL at the same dose. VP-0502AL'spharmacokinetic parameters were Vd: 4.6 l/kg; AUC:3microg.h/ml; t1/2: 0.5h; C0: 6microg/ml; CLtot: 7l/h/kg; andMRT: 0.6h. Following oral administration of VP-0502(100mg/kg), it was not detectable in plasma (<50ng/ml),while after the oral administration of VP-0502AL, VP-0502 was quantitatively detected as an active metabolite forthe first 7h, with a maximum plasma concentration(Cmax) of 0.8microg/ml, and an area under the concentration-time curve (AUC) of 2microg.h/ml. The oral pharmacokineticparameters of VP-0502AL were calculated to be: maximumconcentration time (tmax) 2.7h; Cmax 0.2microg/ml; eliminationhalf-life (t1/2): 0.8h; and AUC 0.5microg.h/ml. Overall thefindings indicate that VP-0502AL has a favorable pharmaco-kinetic profile as a prodrug with rapid transformationinto the active metabolite, and that the attachment of theamino acid alanine to VP-0502 is an effective approach toimprove its oral bioavailability. VP-0502AL is predictedto become a new highly bioavailable anti-AIDS drugcandidate and/or lead compound.