Article 1
Use of minimally invasive surgery in emergency general surgery procedures. Arnold M, Elhage S, Schiffern L, Paton BL, Ross S, Matthews B, Reinke C. Surg Endoscopy. 2020 May; 34(5) :2258-2265.
The use of laparoscopy and minimally invasive techniques for emergency general surgery have been standard of care in many European countries for well over a decade with published guidelines. The use of minimally invasive surgery in emergency general surgery procedures in the Unites States, however, is still not well characterized. Emergency cases other than appendectomy and cholecystectomy are still commonly performed in an open fashion unless the surgeon is “skilled in advanced laparoscopy.” Surgical residents are now graduating with much more exposure to MIS techniques including robotic surgery and may go on to complete MIS fellowships though many acute care surgeons may not have the skillset. A plethora of data exists showing improved outcomes, lower infection rates, lower mortality and shorter length of stay when Minimally Invasive techniques are applied.
Arnold et al. sought to look at the utilization of MIS techniques as well as the outcomes for many EGS procedures over the last 10 years in the US. They used the ACS NSQIP database from 2007-2016 for outcomes based on AAST EGS ICD codes for four common EGS diagnoses including appendicitis, cholangitis/cholecystitis, peptic ulcer disease and small bowel obstruction. Only emergency cases were included. During the 10-year study period, 190,264 patients were included. The use of MIS increased over time during the study period from 71% in 2007 to 89% in 2016.
Procedure specific outcomes were analyzed. Mortality was decreased in all procedures by diagnosis. Cholecystectomy 0.6 % MIS vs 6.7% Open; Appendectomy 0.1% MIS vs 0.6% Open; Bowel obstruction 1.5% MIS vs 7.4% Open; Peptic ulcer disease 2.8% MIS vs 15.8% Open. Overall <0.01. Patients undergoing MIS also had lower rates of reoperation and readmission when compared to open. There was a significant decrease in hospital length of stay in the MIS patients. Infectious complications were also significantly lower in MIS group.
While many studies have demonstrated improved outcomes with the utilization of MIS techniques in elective, non-emergent procedures, the data is lacking regarding specific outcomes for EGS procedures by MIS techniques. This study demonstrates that common EGS procedures have improved outcomes when MIS approaches over open techniques.
There should be continued focus on the training of current and future EGS surgeons in minimally invasive techniques. This is also an opportunity for leading surgical societies to establish guidelines regarding the use of MIS in EGS patients.
Article 2
SAGES and EAES recommendations for minimally invasive surgery during COVID-19 pandemic. Francis N, Dort J, Cho E, Feldman L, Keller D, Lim R, Mikami D, Phillips E, Spaniolas K, Tsuda S, Wasco K, Arulampalam T, Sheraz M, Morales S, Pietrabissa A, Asbun H, Pryor A. Surgical Endoscopy. 2020 Jun; 34(6):2327–2331.
As we are several months into the unprecedented pandemic of COVID-19 healthcare practices and healthcare professionals have been significantly impacted. The novelty of this disease and as a result limited evidence, has led to many questions with respect to the safety of surgical practices especially during a time where minimally invasive surgery is standard for so many procedures in both elective and emergency general surgery. Though elective surgical and endoscopic procedures could be postponed, urgent and emergent procedures may not be able to be delayed.
Multiple societies including The Society of American Gastrointestinal and Endoscopic Surgeons (SAGES) and The European Association for Endoscopic Surgeons (EAES) have put together joint recommendations in order to support the surgical community in general and Minimally invasive Surgery (MIS).
The recommendations are summarized below:
- Suspension of non-essential surgical care during the immediate phases of the COVID-19 pandemic
- Testing all patients before surgery is desirable
- Consent discussion with patients to cover the risk of COVID-19 exposure
- Dedicated COVID-19 OR must be used during the pandemic with a minimum number of staff members during procedure
- All members of the OR/endoscopy staff should use PPE in all procedures during the pandemic regardless of known or suspected COVID status
- A closed smoke evacuation/filtration system with Ultra Low Particulate Air Filtration (ULPA) capability should be used during MIS
- Minimize the use of energy sources (risk of aerosolization) during surgery and endoscopy
- All pneumoperitoneum should be safely evacuated from the port attached to the filtration device before closure, trocar removal, specimen extraction, or conversion to open
- Since patients can present with gastrointestinal manifestations of COVID-19, all emergent endoscopic procedures performed in current environment should be considered high risk
- Advanced endoscopic procedures that require additional insufflation and or energy sources should be avoided
Practical considerations include that there is no evidence to indicate that laparoscopy can lead to aerosolization of COVID-19. Filtration of particles may be more difficult during open procedures. The proven benefits of MIS of reduced length of stay and complications and this should be strongly considered in these patients. The surgical management of patients with suspected or known COVID-19 requires specific consideration and safety measures. Although evidence to aerosol transmission during laparoscopy is lacking, every effort must be made to minimize risk during surgery.
Article 3
Evaluation of Machine-Learning Algorithms for Predicting Opioid Overdose Risk Among Medicare Beneficiaries With Opioid Prescriptions. Lo-Ciganic WH, Huang JL, Zhang HH, Weiss JC, Wu Y, Kwoh CK, Donohue JM, Cochran G, Gordon AJ, Malone DC , Kuza CC, Gellad WF. JAMA Netw Open. 2019 Mar 1;2(3):e190968.
Drug overdose continues to be a major cause of injury-related death in the United States with many health systems instituting new measures to combat this epidemic. There are many studies that aim to identify risk factors for overdose, but don’t make a prediction on an individual’s risk. Classical statistical analysis has not performed well in attempting to predict overdose, so the authors attempted to predict overdose among Medicare beneficiaries with at least one opioid prescription.
The authors used a 5% random sample of Medicare data between 2011 and 2015 to yield 186,686 patients. There were 3,188 opioid overdose deaths during this period. Death was the primary endpoint evaluated. There were five models used for prediction: multivariate logistic regression (MLR), least absolute shrinkage and selection operator-type regression (LASSO), random forest (RF), gradient boosting machine (GBM), and deep neural network (DNN). Important predictors were reported for GBM and RF. Each model had a C statistic, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio, number needed to evaluate to identify 1 overdose episode, and estimated rate of alerts. They also show the precision-recall curves for each model.
The C statistic reaches 0.91 with DNN (the highest) and 0.75 with MLR (the lowest). It appears as though DNN is a good model for making predictions on patients that may die of overdose. However, the C statistic does not take in to account the prevalence, so when looking at the other performance metrics (sensitivity and PPV) the performance is actually poor which is unsurprising for such a rare event. However, the specificity and NPV are quite good. The GBM and DNN are able to effectively risk stratify patients in to low, medium, and high risk. These two machine learning techniques have two important features: (1) it can effectively identify low risk patients and (2) risk stratify patients so there can be a step-wise allocation of resources to those that may be medium or high risk for drug overdose. The key to any predictive model is how it is used within the system’s workflow. In this case, a hospital could use this algorithm to aide a provider’s decision making for additional resource allocation. Thus, we should look at these prediction models as aides, and the final decision should be left to the provider.
Article 4
Comparison of Machine Learning Optimal Classification Trees With the Pediatric Emergency Care Applied Research Network Head Trauma Decision Rules. Bertsimas D, Dunn J, Steele DW, Trikalinos TA, Wang Y. JAMA Pediatr. 2019 Jul 1;173(7):648-656.
Pediatric patients suffering head trauma are a significant burden on emergency departments throughout the United States. Computed Tomography (CT) is currently the standard for diagnosing intracranial injury, but this imaging modality has more significant risk for the pediatric population when one considers the associated ionizing radiation and the possible need for sedation. In 2009, the Pediatric Emergency Care Applied Research Network (PECARN) developed and validated a simple tool to help clinicians make a better determination on the need for head CT. This study seeks to evaluate a machine learning model that improves upon the accuracy of the PECARN rules.
The authors used a dataset of a prospective cohort of 42,412 patients over a two year period from 2004 to 2006, which is the same dataset that the PECARN used to derive its rules. They also followed the original analysis by stratifying patients by age (cutoff was 2 years old) due to developmental changes that change how an evaluation would be performed (i.e. verbal versus nonverbal). However, the dataset that this study used was anonymized, so they could not use the same model development cohort and model validation cohorts. While many of the predictors they used were the same, they relied on additional variables that were available at the time of presentation (age, sex, loss of consciousness, etc.). They used optimal classification trees (OCTs), similar to classification and regression trees, as the machine learning algorithm to make predictions. The three risk classification categories were very low, low, and higher (the same as PECARN). They reported numerous accuracy metrics including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive and negative likelihood ratios (LRs), and odds ratios between OCTs and the PECARN rules.
The OCT algorithm outperformed the PECARN rules for predictive accuracy. While the sensitivity of both algorithms are excellent, the OCT algorithm had statistically significant improvement in the specificity. In children younger than 2 years old a 16% improvement in specificity was observed (odds ratio of 2, 95% confidence interval of 1.87-2.18) using the OCT algorithm. The improvements in prediction for older children were about half that of younger children (7%, OR 1.4, CI 1.36-1.46), but still significant. This type of improvement is precisely what machine learning hopes to accomplish, but showing a clinical and economic improvement will be more elusive. It would be reasonable for a hospital system to implement this decision support within the electronic health record with similar usage guidelines used for the PECARN rules.
