Two new mathematical models for prediction of early mortality risk in coronary artery bypass graft surgery

Machine Learning Methods for Predicting Long-Term Mortality in Patients After Cardiac Surgery

Objective:

This study aims to construct and validate several machine learning (ML) algorithms to predict long-term mortality and identify risk factors in unselected patients post-cardiac surgery.

Methods

The Medical Information Mart for Intensive Care (MIMIC-III) database was used to perform a retrospective administrative database study. Candidate predictors consisted of the demographics, comorbidity, vital signs, laboratory test results, scoring systems, and treatment information on the first day of ICU admission. Four-year mortality was set as the study outcome. We used the ML methods of logistic regression (LR), artificial neural network (NNET), naïve bayes (NB), gradient boosting machine (GBM), adapting boosting (Ada), random forest (RF), bagged trees (BT), and eXtreme Gradient Boosting (XGB). The prognostic capacity and clinical utility of these ML models were compared using the area under the receiver operating characteristic curves (AUC), calibration curves, and decision curve analysis (DCA).

Results

Of 7,368 patients in MIMIC-III included in the final cohort, a total of 1,337 (18.15%) patients died during a 4-year follow-up. Among 65 variables extracted from the database, a total of 25 predictors were selected using recursive feature elimination and included in the subsequent analysis. The Ada model performed best among eight models in both discriminatory ability with the highest AUC of 0.801 and goodness of fit (visualized by calibration curve). Moreover, the DCA shows that the net benefit of the RF, Ada, and BT models surpassed that of other ML models for almost all threshold probability values. Additionally, through the Ada technique, we determined that red blood cell distribution width (RDW), blood urea nitrogen (BUN), SAPS II, anion gap (AG), age, urine output, chloride, creatinine, congestive heart failure, and SOFA were the Top 10 predictors in the feature importance rankings.

Conclusions

The Ada model performs best in predicting 4-year mortality after cardiac surgery among the eight ML models, which might have significant application in the development of early warning systems for patients following operations.

Usefulness of artificial intelligence for predicting recurrence following surgery for pancreatic cancer: Retrospective cohort study

BACKGROUND

or Purpose: Pancreatic ductal adenocarcinoma (PDAC) is a leading cause of mortality in the world with the overall 5-year survival rate of 6%. The survival of patients with PDAC is closely related to recurrence and therefore it is necessary to identify the risk factors for recurrence. This study uses artificial intelligence approaches and multi-center registry data to analyze the recurrence of pancreatic cancer after surgery and its major determinants.

METHODS

Data came from 4846 patients enrolled in a multi-center registry system, the Korea Tumor Registry System (KOTUS). The random forest and the Cox proportional-hazards model (the Cox model) were applied and compared for the prediction of disease-free survival. Variable importance, the contribution of a variable for the performance of the model, was used for identifying major predictors of disease-free survival after surgery. The C-Index was introduced as a criterion for validating the models trained.

RESULTS

Based on variable importance from the random forest, major predictors of disease-free survival after surgery were tumor size (0.00310), tumor grade (0.00211), TNM stage (0.00211), T stage (0.00146) and lymphovascular invasion (0.00125). The coefficients of these variables were statistically significant in the Cox model (p < 0.05). The C-Index averages of the random forest and the Cox model were 0.6805 and 0.7738, respectively.

CONCLUSIONS

This is the first artificial-intelligence study with multi-center registry data to predict disease-free survival after the surgery of pancreatic cancer. The findings of this methodological study demonstrate that artificial intelligence can provide a valuable decision-support system for treating patients undergoing surgery for pancreatic cancer. However, at present, further studies are needed to demonstrate the actual benefit of applying machine learning algorithms in clinical practice.

Can machine learning improve mortality prediction following cardiac surgery?

OBJECTIVES

Interest in the clinical usefulness of machine learning for risk prediction has bloomed recently. Cardiac surgery patients are at high risk of complications and therefore presurgical risk assessment is of crucial relevance. We aimed to compare the performance of machine learning algorithms over traditional logistic regression (LR) model to predict in-hospital mortality following cardiac surgery.

METHODS

A single-centre data set of prospectively collected information from patients undergoing adult cardiac surgery from 1996 to 2017 was split into 70% training set and 30% testing set. Prediction models were developed using neural network, random forest, naive Bayes and retrained LR based on features included in the EuroSCORE. Discrimination was assessed using area under the receiver operating characteristic curve, and calibration analysis was undertaken using the calibration belt method. Model calibration drift was assessed by comparing Goodness of fit χ2 statistics observed in 2 equal bins from the testing sample ordered by procedure date.

RESULTS

A total of 28 761 cardiac procedures were performed during the study period. The in-hospital mortality rate was 2.7%. Retrained LR [area under the receiver operating characteristic curve 0.80; 95% confidence interval (CI) 0.77-0.83] and random forest model (0.80; 95% CI 0.76-0.83) showed the best discrimination. All models showed significant miscalibration. Retrained LR proved to have the weakest calibration drift.

CONCLUSIONS

Our findings do not support the hypothesis that machine learning methods provide advantage over LR model in predicting operative mortality after cardiac surgery.

Machine learning improves mortality risk prediction after cardiac surgery: Systematic review and meta-analysis

BACKGROUND

Interest in the usefulness of machine learning (ML) methods for outcomes prediction has continued to increase in recent years. However, the advantage of advanced ML model over traditional logistic regression (LR) remains controversial. We performed a systematic review and meta-analysis of studies comparing the discrimination accuracy between ML models versus LR in predicting operative mortality following cardiac surgery.

METHODS

The present systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis statement. Discrimination ability was assessed using the C-statistic. Pooled C-statistics and its 95% credibility interval for ML models and LR were obtained were obtained using a Bayesian framework. Pooled estimates for ML models and LR were compared to inform on difference between the 2 approaches.

RESULTS

We identified 459 published citations of which 15 studies met inclusion criteria and were used for the quantitative and qualitative analysis. When the best ML model from individual study was used, meta-analytic estimates showed that ML were associated with a significantly higher C-statistic (ML, 0.88; 95% credibility interval, 0.83-0.93 vs LR, 0.81; 95% credibility interval, 0.77-0.85; P = .03). When individual ML algorithms were instead selected, we found a nonsignificant trend toward better prediction with each of ML algorithms. We found no evidence of publication bias (P = .70).

CONCLUSIONS

The present findings suggest that when compared with LR, ML models provide better discrimination in mortality prediction after cardiac surgery. However, the magnitude and clinical influence of such an improvement remains uncertain.

Personalized Pancreatic Cancer Management: A Systematic Review of How Machine Learning Is Supporting Decision-making

This review critically analyzes how machine learning is being used to support clinical decision-making in the management of potentially resectable pancreatic cancer. Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines, electronic searches of MEDLINE, Embase, PubMed, and Cochrane Database were undertaken. Studies were assessed using the checklist for critical appraisal and data extraction for systematic reviews of prediction modeling studies (CHARMS) checklist. In total 89,959 citations were retrieved. Six studies met the inclusion criteria. Three studies were Markov decision-analysis models comparing neoadjuvant therapy versus upfront surgery. Three studies predicted survival time using Bayesian modeling (n = 1) and artificial neural network (n = 1), and one study explored machine learning algorithms including Bayesian network, decision trees, k-nearest neighbor, and artificial neural networks. The main methodological issues identified were limited data sources, which limits generalizability and potentiates bias; lack of external validation; and the need for transparency in methods of internal validation, consecutive sampling, and selection of candidate predictors. The future direction of research relies on expanding our view of the multidisciplinary team to include professionals from computing and data science with algorithms developed in conjunction with clinicians and viewed as aids, not replacement, to traditional clinical decision-making.

Trends, Predictors, and Outcomes of Stroke After Surgical Aortic Valve Replacement in the United States

BACKGROUND

Postoperative stroke is a devastating complication after aortic valve replacement (AVR). Our objective was to use a large national database to identify the incidence of and risk factors for stroke after AVR, as well as to determine incremental mortality, resource use, and cost of stroke.

METHODS

We identified 360,437 patients who underwent isolated surgical AVR between 1998 and 2011 from the National Inpatient Sample (NIH) database. Mean age was 66 ± 32 years. Multivariable regression and propensity matching were used to identify risk factors and the effect of stroke on outcomes. Patients were stratified according to the Elixhauser comorbidity score (ECS) into low- (0-5), medium- (6-15), and high-risk (16+) categories.

RESULTS

Stroke after AVR occurred in 5,092 (1.45%) patients. The incidence of stroke declined from 1.69% in 1999 to 0.94% in 2011 (p < 0.001). Increasing age and higher comorbidities were the main predictors of stroke (each p < 0.001). The highest-volume centers (>200 AVRs/y) had the lowest rate of stroke (1.2%). After multivariable adjustment, high-volume centers had lower odds of stroke in medium-risk (odds ratio [OR], 0.59; 95% confidence interval [CI], 0.37-0.94) and high-risk patients (OR, 0.39; 95% CI, 0.22-0.68) compared with the lowest-volume centers. For low-risk patients, volume was not associated with stroke. Patients who experienced stroke were hospitalized for 4 days longer, had an average of $10,496 higher costs, and had 2.74 (95% CI, 1.97-3.80) times higher odds of in-hospital mortality compared with those who did not experience stroke (all p < 0.001).

CONCLUSIONS

The incidence of stroke after AVR has decreased but remains a significant cause of morbidity in medium- and high-risk patients. Superior outcomes can be achieved in medium- to high-risk patients at high-volume centers.

Prognostic value of acute kidney injury after cardiac surgery according to kidney disease: improving global outcomes definition and staging (KDIGO) criteria

Objectives

The definition of acute renal failure has been recently reviewed, and the term acute kidney injury (AKI) was proposed to cover the entire spectrum of the syndrome, ranging from small changes in renal function markers to dialysis needs. This study was aimed to evaluate the incidence, morbidity and mortality associated with AKI (based on KDIGO criteria) in patients after cardiac surgery (coronary artery bypass grafting or cardiac valve surgery) and to determine the value of this feature as a predictor of hospital mortality (30 days).

Methods

From January 2003 to June 2013, a total of 2,804 patients underwent cardiac surgery in our service. Cox proportional hazard models were used to determine the association between the development of AKI and 30-day mortality.

Results

A total of 1,175 (42%) patients met the diagnostic criteria for AKI based on KDIGO classification during the first 7 postoperative days: 978 (35%) patients met the diagnostic criteria for stage 1 while 100 (4%) patients met the diagnostic criteria for stage 2 and 97 (3%) patients met the diagnostic criteria for stage 3. A total of 63 (2%) patients required dialysis treatment. Overall, the 30-day mortality was 7.1% (2.2%) for patients without AKI and 8.2%, 31% and 55% for patients with AKI at stages 1, 2 and 3, respectively. The KDIGO stage 3 patients who did not require dialysis had a mortality rate of 41%, while the mortality of dialysis patients was 62%. The adjusted Cox regression analysis revealed that AKI based on KDIGO criteria (stages 1–3) was an independent predictor of 30-day mortality (P<0.001 for all. Hazard ratio = 3.35, 11.94 and 24.85).

Conclusion

In the population evaluated in the present study, even slight changes in the renal function based on KDIGO criteria were considered as independent predictors of 30-day mortality after cardiac surgery.