The Not-So-Distant Future or Just Hype? Utilizing Machine Learning to Predict 30-Day Post-Operative Complications in Laparoscopic Colectomy Patients.

BACKGROUND: Complex machine learning (ML) models have revolutionized predictions in clinical care. However, for laparoscopic colectomy (LC), prediction of morbidity by ML has not been adequately analyzed nor compared against traditional logistic regression (LR) models. METHODS: All LC patients, between 2017 and 2019, in the National Surgical Quality Improvement Program (NSQIP) were identified. A composite outcome of 17 variables defined any post-operative morbidity. Seven of the most common complications were additionally analyzed. Three ML models (Random Forests, XGBoost, and L1-L2-RFE) were compared with LR. RESULTS: Random Forests, XGBoost, and L1-L2-RFE predicted 30-day post-operative morbidity with average area under the curve (AUC): .709, .712, and .712, respectively. LR predicted morbidity with AUC = .712. Septic shock was predicted with AUC ≤ .9, by ML and LR. CONCLUSION: There was negligible difference in the predictive ability of ML and LR in post-LC morbidity prediction. Possibly, the computational power of ML cannot be realized in limited datasets.

Using deep learning to identify bladder cancers with FGFR-activating mutations from histology images.

BACKGROUND: In recent years, the fibroblast growth factor receptor (FGFR) pathway has been proven to be an important therapeutic target in bladder cancer. FGFR-targeted therapies are effective for patients with FGFR mutation, which can be discovered through genetic sequencing. However, genetic sequencing is not commonly performed at diagnosis, whereas a histologic assessment of the tumor is. We aim to computationally extract imaging biomarkers from existing tumor diagnostic slides in order to predict FGFR alterations in bladder cancer. METHODS: This study analyzed genomic profiles and H&E-stained tumor diagnostic slides of bladder cancer cases from The Cancer Genome Atlas (n = 418 cases). A convolutional neural network (CNN) identified tumor-infiltrating lymphocytes (TIL). The percentage of the tissue containing TIL ("TIL percentage") was then used to predict FGFR activation status with a logistic regression model. RESULTS: This predictive model could proficiently identify patients with any type of FGFR gene aberration using the CNN-based TIL percentage (sensitivity = 0.89, specificity = 0.42, AUROC = 0.76). A similar model which focused on predicting patients with only FGFR2/FGFR3 mutation was also found to be highly sensitive, but also specific (sensitivity = 0.82, specificity = 0.85, AUROC = 0.86). CONCLUSION: TIL percentage is a computationally derived image biomarker from routine tumor histology that can predict whether a tumor has FGFR mutations. CNNs and other digital pathology methods may complement genome sequencing and provide earlier screening options for candidates of targeted therapies.

Preventable Morbidity and Mortality Among Non-trauma Emergency Surgery Patients: The Role of Personal Performance and System Flaws in Adverse Events.

BACKGROUND: Preventable morbidity and mortality among emergency surgery patients is not adequately analyzed. We aim to describe and classify preventable complications and deaths in this population. METHODS: The medical records and quality control documents of patients with emergency, non-trauma, surgical disease admitted between September 1, 2006, and August 31, 2018, and recorded to have a preventable or potentially preventable morbidity and mortality were reviewed. The primary outcome was a classification of the complications and deaths by a panel of experts, as attributable to issues of personal performance or system deficiencies. RESULTS: One hundred and fifty patients were identified (127 complications and 23 deaths). The most commonly encountered preventable complications were surgical-site infection (17%), bleeding (13%), injury to adjacent structures (12%), and anastomotic leak (8%). The majority of complications seemed to stem from personal performance (97%), due to either technical or judgment issues, and only 3% were linked with system flaws, either in the form of communication or inadequate protocols. Alcohol use disorder and duration of operation were different between patients with preventable adverse events related to technical issues and patients related to judgment issues; furthermore, more patients who experienced judgment issues died during hospital stay (p <0.05). CONCLUSION: Among emergency surgery patients, who suffer preventable complications and deaths, issues related to personal performance are more frequent than system flaws. Whereas the effort to improve systems should be unwavering, the emphasis on the surgeon's personal responsibility to avoid preventable complications should not be derailed.

It still hurts! Persistent pain and use of pain medication one year after injury.

BACKGROUND: Given the scarce literature data on chronic post-traumatic pain, we aim to identify early predictors of long-term pain and pain medication use after major trauma. METHODS: Major trauma patients (Injury Severity Score ≥ 9) from three Level I Trauma Centers at 12 months after injury were interviewed for daily pain using the Trauma Quality of Life questionnaire. Multivariate logistic regression models identified patient- and injury-related independent predictors of pain and use of pain medication. RESULTS: Of 1238 patients, 612 patients (49%) felt daily pain and 300 patients (24%) used pain medication 1 year after injury. Of a total of 8 independent predictors for chronic pain and 9 independent predictors for daily pain medication, 4 were common (pre-injury alcohol use, pre-injury drug use, hospital stay ≥ 5 days, and education limited to high school). Combinations of independent predictors yielded weak predictability for both outcomes, ranging from 20% to 72%. CONCLUSIONS: One year after injury, approximately half of trauma patients report daily pain and one-fourth use daily pain medication. These outcomes are hard to predict.