A machine learning approach to high-risk cardiac surgery risk scoring

Development of a machine learning-based risk model for postoperative complications of lung cancer surgery

PURPOSE

To develop a comorbidity risk score specifically for lung resection surgeries.

METHODS

We reviewed the medical records of patients who underwent lung resections for lung cancer, and developed a risk model using data from 2014 to 2017 (training dataset), validated using data from 2018 to 2019 (validation dataset). Forty variables were analyzed, including 35 factors related to the patient's overall condition and five factors related to surgical techniques and tumor-related factors. The risk model for postoperative complications was developed using an elastic net regularized generalized linear model. The performance of the risk model was evaluated using receiver operating characteristic curves and compared with the Charlson Comorbidity Index (CCI).

RESULTS

The rate of postoperative complications was 34.7% in the training dataset and 21.9% in the validation dataset. The final model consisted of 20 variables, including age, surgical-related factors, respiratory function tests, and comorbidities, such as chronic obstructive pulmonary disease, a history of ischemic heart disease, and 12 blood test results. The area under the curve (AUC) for the developed risk model was 0.734, whereas the AUC for the CCI was 0.521 in the validation dataset.

CONCLUSIONS

The new machine learning model could predict postoperative complications with acceptable accuracy.

CLINICAL REGISTRATION NUMBER

2020-0375.

Survival Tree Provides Individualized Estimates of Survival After Lung Transplant

INTRODUCTION

Identifying contributors to lung transplant survival is vital in mitigating mortality. To enhance individualized mortality estimation and determine variable interaction, we employed a survival tree algorithm utilizing recipient and donor data.

METHODS

United Network Organ Sharing data (2000-2021) were queried for single and double lung transplants in adult patients. Graft survival time <7 d was excluded. Sixty preoperative and immediate postoperative factors were evaluated with stepwise logistic regression on mortality; final model variables were included in survival tree modeling. Data were split into training and testing sets and additionally validated with 10-fold cross validation. Survival tree pruning and model selection was based on Akaike information criteria and log-likelihood values. Estimated survival probabilities and log-rank pairwise comparisons between subgroups were calculated.

RESULTS

A total of 27,296 lung transplant patients (8175 single; 19,121 double lung) were included. Stepwise logistic regression yielded 47 significant variables associated with mortality. Survival tree modeling returned six significant factors: recipient age, length of stay from transplant to discharge, recipient ventilator duration post-transplant, double lung transplant, recipient reintubation post-transplant, and donor cytomegalovirus status. Eight subgroups consisting of combinations of these factors were identified with distinct Kaplan-Meier survival curves.

CONCLUSIONS

Survival trees provide the ability to understand the effects and interactions of covariates on survival after lung transplantation. Individualized survival probability with this technique found that preoperative and postoperative factors influence survival after lung transplantation. Thus, preoperative patient counseling should acknowledge a degree of uncertainty given the influence of postoperative factors.

Clinical Applications of Machine Learning

Objective

This review introduces interpretable predictive machine learning approaches, natural language processing, image recognition, and reinforcement learning methodologies to familiarize end users.

Background

As machine learning, artificial intelligence, and generative artificial intelligence become increasingly utilized in clinical medicine, it is imperative that end users understand the underlying methodologies.

Methods

This review describes publicly available datasets that can be used with interpretable predictive approaches, natural language processing, image recognition, and reinforcement learning models, outlines result interpretation, and provides references for in-depth information about each analytical framework.

Results

This review introduces interpretable predictive machine learning models, natural language processing, image recognition, and reinforcement learning methodologies.

Conclusions

Interpretable predictive machine learning models, natural language processing, image recognition, and reinforcement learning are core machine learning methodologies that underlie many of the artificial intelligence methodologies that will drive the future of clinical medicine and surgery. End users must be well versed in the strengths and weaknesses of these tools as they are applied to patient care now and in the future.

Machine Learning Analysis of Postlaparoscopy Hernias and "I'm Leaving You to Close" Strategy

INTRODUCTION

Contributing factors to postlaparoscopy hernia are unknown. We hypothesized that postlaparoscopy incisional hernias are increased when the index surgery was performed in teaching hospitals. Laparoscopic cholecystectomy was chosen as the archetype for open umbilical access.

MATERIALS AND METHODS

Maryland and Florida SID/SASD databases (2016-2019) wereused to track 1-year hernia incidence in both inpatient and outpatient settings, which was then linked to Hospital Compare, Distressed Communities Index (DCI), and ACGME. Postoperative umbilical/incisional hernia following laparoscopic cholecystectomy was identified using CPT and ICD-10. Propensity matching and eight machine learning modes were utilized including logistic regression, neural network, gradient boosting machine, random forest, gradient boosted trees, classification and regression trees, k nearest neighbors and support vector machines.

RESULTS

Postoperative hernia incidence was 0.2% (total = 286; 261 incisional and 25 umbilical) in 117,570 laparoscopic cholecystectomy cases. Days to presentation (mean ± SD) were incisional 141 ± 92 and umbilical 66 ± 74. Logistic regression performed best (AUC 0.75 (95% ci 0.67-0.82) and accuracy 0.68 (95% ci 0.60-0.75) using 10-fold cross validation) in propensity matched groups (1:1; n = 279). Postoperative malnutrition (OR 3.5), hospital DCI of comfortable, mid-tier, at risk or distressed (OR 2.2 to 3.5), LOS >1 d (OR 2.2), postop asthma (OR 2.1), hospital mortality below national average (OR 2.0) and emergency admission (OR 1.7) were associated with increased hernias. A decreased incidence was associated with patient location of small metropolitan areas with <1 million residents (OR 0.5) and Charlson Comorbidity Index-Severe (OR 0.5). Teaching hospitals were not associated with postoperative hernia after laparoscopic cholecystectomy.

CONCLUSIONS

Different patient factors as well as underlying hospital factors are associated with postlaparoscopy hernias. Performance of laparoscopic cholecystectomy at teaching hospitals is not associated with increased postoperative hernias.

ChatGPT's quiz skills in different otolaryngology subspecialties: an analysis of 2576 single-choice and multiple-choice board certification preparation questions

PURPOSE

With the increasing adoption of artificial intelligence (AI) in various domains, including healthcare, there is growing acceptance and interest in consulting AI models to provide medical information and advice. This study aimed to evaluate the accuracy of ChatGPT's responses to practice quiz questions designed for otolaryngology board certification and decipher potential performance disparities across different otolaryngology subspecialties.

METHODS

A dataset covering 15 otolaryngology subspecialties was collected from an online learning platform funded by the German Society of Oto-Rhino-Laryngology, Head and Neck Surgery, designed for board certification examination preparation. These questions were entered into ChatGPT, with its responses being analyzed for accuracy and variance in performance.

RESULTS

The dataset included 2576 questions (479 multiple-choice and 2097 single-choice), of which 57% (n = 1475) were answered correctly by ChatGPT. An in-depth analysis of question style revealed that single-choice questions were associated with a significantly higher rate (p < 0.001) of correct responses (n = 1313; 63%) compared to multiple-choice questions (n = 162; 34%). Stratified by question categories, ChatGPT yielded the highest rate of correct responses (n = 151; 72%) in the field of allergology, whereas 7 out of 10 questions (n = 65; 71%) on legal otolaryngology aspects were answered incorrectly.

CONCLUSION

The study reveals ChatGPT's potential as a supplementary tool for otolaryngology board certification preparation. However, its propensity for errors in certain otolaryngology areas calls for further refinement. Future research should address these limitations to improve ChatGPT's educational use. An approach, with expert collaboration, is recommended for the reliable and accurate integration of such AI models.

Recipient Survival after Orthotopic Liver Transplantation: Interpretable Machine Learning Survival Tree Algorithm for Patient-Specific Outcomes

BACKGROUND

Elucidating contributors affecting liver transplant survival is paramount. Current methods offer crude global group outcomes. To refine patient-specific mortality probability estimation and to determine covariate interaction using recipient and donor data, we generated a survival tree algorithm, Recipient Survival After Orthotopic Liver Transplantation (ReSOLT), using United Network Organ Sharing (UNOS) transplant data.

STUDY DESIGN

The UNOS database was queried for liver transplants in patients ≥18 years old between 2000 and 2021. Preoperative factors were evaluated with stepwise logistic regression; 43 significant factors were used in survival tree modeling. Graft survival of <7 days was excluded. The data were split into training and testing sets and further validated with 10-fold cross-validation. Survival tree pruning and model selection was achieved based on Akaike information criterion and log-likelihood values. Log-rank pairwise comparisons between subgroups and estimated survival probabilities were calculated.

RESULTS

A total of 122,134 liver transplant patients were included for modeling. Multivariable logistic regression (area under the curve = 0.742, F1 = 0.822) and survival tree modeling returned 8 significant recipient survival factors: recipient age, donor age, recipient primary payment, recipient hepatitis C status, recipient diabetes, recipient functional status at registration and at transplantation, and deceased donor pulmonary infection. Twenty subgroups consisting of combinations of these factors were identified with distinct Kaplan-Meier survival curves (p < 0.001 among all by log rank test) with 5- and 10-year survival probabilities.

CONCLUSIONS

Survival trees are a flexible and effective approach to understand the effects and interactions of covariates on survival. Individualized survival probability following liver transplant is possible with ReSOLT, allowing for more coherent patient and family counseling and prediction of patient outcome using both recipient and donor factors.