30 days mortality prediction and risk factor analysis of Asian patients with ACS using interpretable machine learning algorithm

Background

Thrombolysis in Myocardial Infarction (TIMI) is used to predict the mortality rate in patients with acute coronary syndrome (ACS). TIMI was developed with limited data on the Asian cohort and was based on the Western cohort. STEMI and NSTEMI have separate TIMI scores. There has been limited research on Asian ACS patients using interpretable machine learning (ML) algorithms.

Purpose

To construct a single 30-day mortality risk scoring system, as well as identify and analyse risk factors in ASIAN patients with ACS, that is applicable to both STEMI and NSTEMI patients, using an interpretable ML algorithm.

Methods

The National Cardiovascular Disease Database registry data of 9054 patients was used. 70% of the data was used for algorithm development, with the remaining 30% used for validation Fifty-four parameters were considered, demographics, cardiovascular risk, medications, and clinical variables. To provide better guidance and advice for clinical judgement, the gradient boosting algorithm (XGBoost) for classification analysis and SHapley Additive exPlanation (SHAP) value analysis graphs were used. Each indicator's SHAP value indicates the impact on model output (mortality) and was calculated using the XGBoost model. The performance evaluation metric was the area under the curve (AUC). The model was validated with a validation dataset and compared to the conventional score TIMI for STEMI and NSTEMI.

Results

The performance on validation dataset of the XGBoost algorithm using the top ten predictors from SHAP for; STEMI (AUC = 0.8534, 95% CI: 0.8226–0.8842, Accuracy: 0.8053, Sensitivity: 0.73125, Specificity: 0.81355) and NSTEMI (AUC = 0.8145, 95% CI: 0.77–0.8589, Accuracy: 0.7972, Sensitivity: 0.64356, Specificity: 0.81232) outperformed TIMI score (STEMI AUC = 0.785, NSTEMI AUC = 0.543). Killip class, age, heart rate, fasting blood glucose, ACEI, creatine kinase, systolic blood pressure, HDLC, cardiac catheterization, and oralhypogly are the top ten predictors chosen by the SHAP feature selection in ascending order. Cardiac catheterization and pharmacotherapy drugs as selected predictors improve mortality prediction in STEMI and NSTEMI patients compared to TIMI. The variable names are displayed on the y-axis in ascending order of importance. The average SHAP value is shown next to them. The SHAP value is shown on the x-axis. The colour represents the value of the feature, ranging from small to large, allowing comprehension of the distribution of the SHAP values for each feature (Figure 1). We can see that having a high killip class and being older are linked to a lower survival rate in ACS patients. Cardiac catheterization procedures, as well as the use of ACEI and OHA, both improve patient mortality (Figure 2).

Conclusions

A single algorithm would classify ACS patients better than TIMI, which requires two distinct scores. In order to better predict 30-day mortality in an ASIAN population, interpretable ML can be used.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1

Machine learning to predict in-hospital mortality risk among heterogenous STEMI patients with diabetes

Funding Acknowledgements

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): TECHNOLOGY DEVELOPMENT FUND 1

Background

Diabetes has become a major public health concern in Asia. In Malaysia, the prevalence of diabetes has escalated in adults above the age of 18, affecting 3.9  million individuals. Patients with diabetes and coronary heart disease have worse outcomes, compared with patients without diabetes who have coronary heart disease. Conventional Risk scores such as TIMI and GRACE were derived from a Western Caucasian cohort with limited data from  Asian countries, despite Asia hosting 60% of the world’s population.

Purpose

It is important to recognize the significant features associated with in-hospital mortality risk that is population-specific in Asian diabetes patients with STEMI to achieve a reliable and effective clinical diagnosis and improved outcome. Electronic health records contain large amounts of information on patients’ medical history and are becoming invaluable research tools that could be applied to cardiovascular disease risk prediction through machine learning (ML) algorithms. With the current success of ML  over conventional methods in STEMI mortality prediction, we aim to develop ML algorithms for in-hospital risk mortality in Asian patients diagnosed with DM that can be adopted for clinical predictions

Methods

We used registry data from the Malaysian National Cardiovascular Disease Database of 5783 patients diagnosed with DM from 2006 to 2016. Fifty parameters including demographics, cardiovascular risk, medications and clinical variables were considered. Four machine learning (ML) algorithms were constructed using a 70% registry dataset; Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Booster (XGB) and Logistic Regression (LR). Feature selections were done based on ML algorithms feature importance combined with Sequential Backward Selection (SBS). The area under the curve (AUC)  was used as the performance evaluation metric. All algorithms were validated using a 30 % validation dataset and compared to the conventional TIMI risk score for STEMI.

Results

The best model SVM (AUC = 0.90) outperformed other ML algorithms (Figure 1) and TIMI risk score (AUC = 0.83). The best SVM model consists of 11 predictors which are Killip class, fasting blood glucose, age, systolic blood pressure, heart rate, ACE inhibitor, beta-blocker, total cholesterol, diastolic blood pressure, lower density lipoprotein, and diuretic (Figure 2). Common predictors of SVM and TIMI risk score are Killip class, age, systolic blood pressure, and heart rate. We have shown that the population-specific data mining approach for the prediction of diabetes patients’ mortality post-STEMI outperformed conventional TIMI risk score.

Conclusion

In the Asian multiethnic population, combination of ML approaches with features selection demonstrated promising outcomes in patients with DM that may be used for better patient prognostic than the conventional method. Abstract Figure 1: ML Best Model Performance  Abstract Figure 2: Selected Predictors for ML

ACS mortality prediction in Asian in-hospital patients with deep learning using machine learning feature selection

Background

Thrombolysis in Myocardial infarction (TIMI) is used in predicting the mortality rate of the acute coronary syndrome (ACS) patients. TIMI was developed based on the Western cohort with limited data on the Asian cohort. There are separate TIMI scores for STEMI and NSTEMI. Deep learning (DL) and machine learning (ML) algorithms such as support vector machine (SVM) in population-specific dataset resulted in a higher area under the curve (AUC) to TIMI. The limitation of DL is selected features by the algorithm is unknown compared to ML algorithms.

Purpose

To construct a single in-hospital mortality risk scoring system that combines SVM feature importance and the DL algorithm in ASIAN patients with ACS that is applicable for both STEMI and NSTEMI patients. To investigate DL performance constructed using predictors selected from SVM feature extraction and DL using complete features and compare with TIMI risk score for STEMI and NSTEMI patients.

Methods

We constructed four algorithms: i) DL and SVM algorithm with feature selected from SVM variable importance, ii) DL and SVM algorithm without feature selection. SVM feature importance with the backward elimination method is used to select and rank important variables. We used registry data from the National Cardiovascular Disease Database of 13190 patient's data. Fifty-four parameters including demographics, cardiovascular risk, medications and clinical variables were considered. AUC was used as the performance evaluation metric. All algorithms were validated using validation dataset and compared to the conventional TIMI for STEMI and NSTEMI.

Results

Validation results in Figure 1 are by STEMI and NTEMI patients. Both DL algorithms outperformed ML and TIMI score on validation data. Similar performance is observed for DL and SVM algorithms using all predictors (54 predictors) with DL and SVM algorithm using selected predictors (14 predictors). Predictors selected by the SVM feature selection are: age, heart rate, Killip class, fasting blood glucose, ST-elevation, CABG, cardiac catheterization, angina episode, HDLC, LDC, other lipid-lowering agents, statin, anti-arrhythmic agent, oralhypogly. CABG and pharmacotherapy drugs as selected predictors improve mortality prediction compared to TIMI score. In DL, 25.87% of STEMI patients and 19.71% of NSTEMI patients are estimated as high risk (risk probabilities of >50%). TIMI underestimated the risk of mortality of high-risk patients (≥5 risk scores) with 13.08% from STEMI patients and 4.65% from NSTEMI patients (Figure 2).

Conclusions

In the ASIAN multi-ethnicity population, patients with ACS can be better classified using one single algorithm compared to the conventional method like TIMI which requires two different scores. Combining ML feature selection with DL allows the identification of distinct factors related to in-hospital mortality of ACS patients in a unique ASIAN population for better mortality prediction.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1 Figure 1. Performance resultsFigure 2. Analysis on the validation set

Investigating performance of deep learning and machine learning risk stratification of Asian in-hospital patients after ST-elevation myocardial infarction

Background

Machine learning (ML) algorithm support vector machine (SVM) performed better than Thrombolysis in Myocardial Infarction (TIMI) score for ASIAN STEMI patients. However, Deep Learning (DL) effectiveness in the multiethnic ASIAN population has yet to be determined. DL has automatic learning of the feature from a given dataset without the need to conduct feature selection. However, the selected features by the algorithm is black box. Identifying features associated with mortality is essential to recognize characteristics of patients with high risk for better patient management.

Purpose

To develop a DL algorithm for in-hospital mortality in multiethnic STEMI patients using predictors identified from the SVM algorithm. To investigate DL performance constructed using predictors from SVM feature extraction and expert-recommended predictors.

Methods

We constructed four algorithms; a) DL and SVM algorithms with predictors identified from the SVM variable importance b) DL and SVM using predictors based on expert recommendation. We used registry data from the National Cardiovascular Disease Database of 11397 patient's. Fifty parameters including demographics, cardiovascular risk, medications and clinical variables were considered. The Area under the curve (AUC) is the performance evaluation metric. Algorithms were validated against the TIMI and tested using the same validation data. SVM variable importance with backward elimination was used to select and rank important variables.

Results

DL algorithms outperform SVM and TIMI on the validation dataset; i) DL with SVM selected predictors (15 predictors, AUC = 0.97), ii) DL with expert-recommended predictors (16 predictors, AUC = 0.96), iii) SVM with selected predictors (15 predictors, AUC = 0.92), iv) SVM with expert-recommended predictors (AUC = 0.89) and TIMI (AUC = 0.82). Common predictors across SVM feature selection, expert-recommendation and TIMI are: age, heart rate, Killip class, fasting blood glucose, systolic blood pressure, comorbid diseases and ST-elevation. SVM feature selection also identified diuretics, PCI and pharmacotherapy drugs as predictors that improve mortality prediction in STEMI patients. Our findings suggest that the TIMI score underestimates patients risk of mortality. DL algorithm using selected predictors classified 35% of nonsurvival patients as high risk (risk probabilities >50%) compared to only 12.7% nonsurvival patients by TIMI (score >5) (Figure below).

Conclusions

In the ASIAN population, patients with STEMI can be better classified using the DL algorithm compared to the ML and TIMI score. Combining ML feature selection with DL allows the identification of distinct factors in a unique ASIAN population for better mortality prediction than relying solely on an expert recommendation as it is a very subjective approach. Continuous validation on population-specific algorithms using DL and ML is needed before implementing in a real clinical setting.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Technology Development Fund 1 TIMI performance on validation setDL performance on validation set

Predictors of difficult videolaryngoscopy with GlideScope® or C-MAC® with D-blade: secondary analysis from a large comparative videolaryngoscopy trial

BACKGROUND

Tracheal intubation using acute-angle videolaryngoscopy achieves high success rates, but is not without difficulty. We aimed to determine predictors of 'difficult videolaryngoscopy'.

METHODS

We performed a secondary analysis of a data set (n=1100) gathered from a multicentre prospective randomized controlled trial of patients for whom difficult direct laryngoscopy was anticipated and who were intubated with one of two videolaryngoscopy devices (GlideScope(®) or C-MAC(®) with D-blade). 'Difficult videolaryngoscopy' was defined as 'first intubation time >60 s' or 'first attempt intubation failure'. A multivariate logistic regression model along with stepwise model selection techniques was performed to determine independent predictors of difficult videolaryngoscopy.

RESULTS

Of 1100 patients, 301 were identified as difficult videolaryngoscopies. By univariate analysis, head and neck position, provider, type of surgery, and mouth opening were associated with difficult videolaryngoscopy (P<0.05). According to the multivariate logistic regression model, characteristics associated with greater risk for difficult videolaryngoscopy were as follows: (i) head and neck position of 'supine sniffing' vs 'supine neutral' {odds ratio (OR) 1.63, 95% confidence interval (CI) [1.14, 2.31]}; (ii) undergoing otolaryngologic or cardiac surgery vs general surgery (OR 1.89, 95% CI [1.19, 3.01] and OR 6.13, 95% CI [1.85, 20.37], respectively); (iii) intubation performed by an attending anaesthestist vs a supervised resident (OR 1.83, 95% CI [1.14, 2.92]); and (iv) small mouth opening (OR 1.18, 95% CI [1.02, 1.36]).

CONCLUSION

This secondary analysis of an existing data set indicates four covariates associated with difficult acute-angle videolaryngoscopy, of which patient position and provider level are modifiable.

Human papillomavirus type 18 and other risk factors for cervical cancer in Jakarta, Indonesia

Infection with human papillomavirus (HPV) has now been established as a necessary cause of cervical cancer. Indonesia is a country with a high cervical cancer incidence and with the world's highest prevalence of HPV 18 in cervical cancer. No information exists about the prevalence of HPV 18 or other HPV types in the Indonesian population. We conducted a hospital-based case-control study in Jakarta, Indonesia. A total of 74 cervical carcinoma cases and 209 control women, recruited from the gynecological outpatient clinic of the same hospital, were included. All women were HPV typed by the line probe assay, and interviews were obtained regarding possible risk factors for cervical cancer. HPV was detected in 95.9% of the cases and in 25.4% of the controls. In the control group, 13.4% was infected with a high-risk HPV type. HPV 16 was detected in 35% of the case group and in 1.9% of the control group and HPV 18 was identified in 28% of the case group and in 2.4% of the control group, suggesting that the oncogenic potentials of HPV 16 and HPV 18 in Indonesia are similar. In addition to HPV infection, young age at first intercourse, having a history of more than one sexual partner, and high parity were significant risk factors for cervical cancer. Within the control group, we did not identify determinants of HPV infection. We hypothesize that the high prevalence of HPV 18 in cervical cancer in Indonesia is caused by the high prevalence of HPV 18 in the Indonesian population.

Current management of malignant germ cell tumor of the ovary

Malignant germ cell tumors are an uncommon type of ovarian cancer which account for fewer than 5% of the total in Western countries and 20% in Japan. In females younger than 20, they represent approximately two-thirds of malignant ovarian tumors. Immature teratoma, endodermal sinus tumor, dysgerminoma and mixed type account for the majority (more than 80%), while embryonal carcinoma and polyembryoma are very few. The age of the patients ranges from 6 to 69 years with a median of 16-20 years. Clinically, these tumors are characterized by rapid growth and extensive intraabdominal spread. The symptoms and signs range from 1 day to 6 months with a median of 4 weeks, and the patients usually present with abdominal pain, palpable mass, abdominal distention and vaginal bleeding, and a very few with amenorrhea and precocious puberty. The size of tumors varies from 7 cm to 40 cm with a median of 15-16 cm. The tumor is rarely bilateral (12-19%) and never so in cases of endodermal sinus tumor. Diagnosis depends mainly on age, abdominal symptoms, size and consistency of the tumor, and tumor markers AFP and hCG. Surgery is the first step of management followed by adjuvant therapy, which depends on the histologic type. Dysgerminoma is very sensitive to radiation while other germ cell tumors are not. A combination chemotherapy currently used is VAC or VBP. Both are highly effective. The VBP regimen seems to have a stronger cancerocidal effect, while the VAC regimen is less toxic. VAC produces excellent results in stage I, while VBP is more effective for advanced disease. Conservative surgery and a combination chemotherapy (VAC, VBP) are appropriate for young patients who desire to retain their fertility. Second-look laparotomy is still controversial. As long as AFP or hCG or both can be used to monitor the disease in patients positive for these sensitive and reliable markers, or in an early stage with complete resection, second-look laparotomy is not useful. Survival is associated with prognostic factors, i.e., histologic type, clinical staging operation, lymph node and residual tumor. Patients with endodermal sinus tumor or mixed type tumor had a poorer outcome. The survival rate was higher in patients with earlier disease (stage I or II) and those who underwent primary surgery. Metastasis to the lymph node is not related to prognosis. The presence and size of residual tumors after surgery were closely related to the prognosis.

Epidemiology of gestational trophoblastic neoplasm at the Dr. Cipto Mangunkusumo Hospital, Jakarta, Indonesia

This retrospective research was conducted in the Department of Obstetrics and Gynaecology of the Dr. Cipto Mangunkusumo Hospital, Jakarta, covering the period between 1977 and 1981. The incidence of hydatidiform mole was 1 in 77 pregnancies. The incidence of malignant trophoblastic disease was 1 in 185 pregnancies. Of the 406 cases of hydatidiform mole, 22.9% became malignant. Patients of 24 years of age or younger had a higher risk of getting hydatidiform mole (P less than 0.05) compared to older patients. The risk of becoming malignant increased with age and became evident after 40 years of age. Parity 1 or less was associated with a higher risk of getting hydatidiform mole (P less than 0.05), but had no influence on hydatidiform mole becoming malignant. The influence of blood group was not so clear, although there was a tendency for moles to occur more frequently in patients with blood groups A or B. By contrast, there was a tendency for the change into malignancy to occur more frequently in women with blood groups B or O. Gestational age had no influence towards the change into malignancy or metastasis. Uterine size (greater than 20 weeks gestation) correlated with the progression of hydatidiform mole into malignancy. However, subsequent metastasis was not influenced by the size of the uterus. It was found that 76.4% of malignant trophoblastic diseases originated from hydatidiform moles, 12.4% from abortions, 9.5% from normal deliveries, and 1.2% from ectopic pregnancies. Non-hydatidiform moles had a slightly greater risk for metastasis, although this was not significant. Hydatidiform mole in histologic stages II or III (Hertig-Mansell classification) had a significantly greater tendency (P less than 0.05) to become malignant than in stage I.