Learning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels.

Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathies (CAAs). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that correlated with gold-standard CERAD-like WSI scoring (p=0.07± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.

Learning fast and fine-grained detection of amyloid neuropathologies from coarse-grained expert labels.

Precise, scalable, and quantitative evaluation of whole slide images is crucial in neuropathology. We release a deep learning model for rapid object detection and precise information on the identification, locality, and counts of cored plaques and cerebral amyloid angiopathy (CAA). We trained this object detector using a repurposed image-tile dataset without any human-drawn bounding boxes. We evaluated the detector on a new manually-annotated dataset of whole slide images (WSIs) from three institutions, four staining procedures, and four human experts. The detector matched the cohort of neuropathology experts, achieving 0.64 (model) vs. 0.64 (cohort) average precision (AP) for cored plaques and 0.75 vs. 0.51 AP for CAAs at a 0.5 IOU threshold. It provided count and locality predictions that approximately correlated with gold-standard human CERAD-like WSI scoring (p = 0.07 ± 0.10). The openly-available model can quickly score WSIs in minutes without a GPU on a standard workstation.

Deep learning from multiple experts improves identification of amyloid neuropathologies.

Pathologists can label pathologies differently, making it challenging to yield consistent assessments in the absence of one ground truth. To address this problem, we present a deep learning (DL) approach that draws on a cohort of experts, weighs each contribution, and is robust to noisy labels. We collected 100,495 annotations on 20,099 candidate amyloid beta neuropathologies (cerebral amyloid angiopathy (CAA), and cored and diffuse plaques) from three institutions, independently annotated by five experts. DL methods trained on a consensus-of-two strategy yielded 12.6-26% improvements by area under the precision recall curve (AUPRC) when compared to those that learned individualized annotations. This strategy surpassed individual-expert models, even when unfairly assessed on benchmarks favoring them. Moreover, ensembling over individual models was robust to hidden random annotators. In blind prospective tests of 52,555 subsequent expert-annotated images, the models labeled pathologies like their human counterparts (consensus model AUPRC = 0.74 cored; 0.69 CAA). This study demonstrates a means to combine multiple ground truths into a common-ground DL model that yields consistent diagnoses informed by multiple and potentially variable expert opinions.

Trans-channel fluorescence learning improves high-content screening for Alzheimer's disease therapeutics.

In microscopy-based drug screens, fluorescent markers carry critical information on how compounds affect different biological processes. However, practical considerations, such as the labor and preparation formats needed to produce different image channels, hinders the use of certain fluorescent markers. Consequently, completed screens may lack biologically informative but experimentally impractical markers. Here, we present a deep learning method for overcoming these limitations. We accurately generated predicted fluorescent signals from other related markers and validated this new machine learning (ML) method on two biologically distinct datasets. We used the ML method to improve the selection of biologically active compounds for Alzheimer's disease (AD) from a completed high-content high-throughput screen (HCS) that had only contained the original markers. The ML method identified novel compounds that effectively blocked tau aggregation, which had been missed by traditional screening approaches unguided by ML. The method improved triaging efficiency of compound rankings over conventional rankings by raw image channels. We reproduced this ML pipeline on a biologically independent cancer-based dataset, demonstrating its generalizability. The approach is disease-agnostic and applicable across diverse fluorescence microscopy datasets.

Clopidogrel loading dose and bleeding outcomes in patients undergoing urgent coronary artery bypass grafting.

BACKGROUND: Coronary artery bypass grafting (CABG) performed within 5 days of clopidogrel administration is associated with increased bleeding. The impact of clopidogrel loading dose is unknown. We examined the effect of clopidogrel loading dose on bleeding outcomes in patients undergoing urgent CABG. METHODS: Clinical outcomes were examined retrospectively for 196 consecutive patients undergoing urgent CABG within 5 days of a clopidogrel loading dose between January 2003 and June 2009. Major bleeding was defined as a fall in hemoglobin > 5 g/dL, fatal or intracranial bleeding, or cardiac tamponade. RESULTS: One hundred forty-eight patients received 300 mg and 48 patients received ≥ 600 mg clopidogrel loading. Patients were predominantly male (78%) with a mean age of 66 ± 10 years. Mean duration from clopidogrel loading to CABG was 3.0 ± 1.5 and 3.0 ± 1.6 days for the 300 and 600 mg loading doses, respectively. Major bleeding occurred in 47% of patients receiving 300 mg and 73% of patients receiving ≥ 600 mg clopidogrel loading (P = .002). Compared with 300 mg, patients receiving ≥ 600 mg had greater 24-hour chest tube output (391 ± 251 vs 536 ± 354 mL, P = .01), stayed longer in surgical intensive care (4.3 ± 4.1 vs 5.0 ± 3.1 days, P = .0001), and trended toward greater reoperation for bleeding (5% vs 12%, P = .09). Following multivariate analysis, clopidogrel loading dose ≥ 600 mg (odds ratio 2.8, CI 1.2-6.6), preoperative hemoglobin (3.4, 2.7-5.0 per 1 g/dL increase), and female gender (2.9, 1.1-7.4) predicted major bleeding. CONCLUSIONS: Higher clopidogrel loading doses are associated with increased bleeding when administered within 5 days of CABG. The development of shorter-acting, reversible, oral antiplatelet agents may reduce perioperative bleeding in this population.

Robust Sequence Determinants of α-Synuclein Toxicity in Yeast Implicate Membrane Binding.

Protein conformations are shaped by cellular environments, but how environmental changes alter the conformational landscapes of specific proteins in vivo remains largely uncharacterized, in part due to the challenge of probing protein structures in living cells. Here, we use deep mutational scanning to investigate how a toxic conformation of α-synuclein, a dynamic protein linked to Parkinson's disease, responds to perturbations of cellular proteostasis. In the context of a course for graduate students in the UCSF Integrative Program in Quantitative Biology, we screened a comprehensive library of α-synuclein missense mutants in yeast cells treated with a variety of small molecules that perturb cellular processes linked to α-synuclein biology and pathobiology. We found that the conformation of α-synuclein previously shown to drive yeast toxicity-an extended, membrane-bound helix-is largely unaffected by these chemical perturbations, underscoring the importance of this conformational state as a driver of cellular toxicity. On the other hand, the chemical perturbations have a significant effect on the ability of mutations to suppress α-synuclein toxicity. Moreover, we find that sequence determinants of α-synuclein toxicity are well described by a simple structural model of the membrane-bound helix. This model predicts that α-synuclein penetrates the membrane to constant depth across its length but that membrane affinity decreases toward the C terminus, which is consistent with orthogonal biophysical measurements. Finally, we discuss how parallelized chemical genetics experiments can provide a robust framework for inquiry-based graduate coursework.

Prototype of running clinical trials in an untrustworthy environment using blockchain.

Monitoring and ensuring the integrity of data within the clinical trial process is currently not always feasible with the current research system. We propose a blockchain-based system to make data collected in the clinical trial process immutable, traceable, and potentially more trustworthy. We use raw data from a real completed clinical trial, simulate the trial onto a proof of concept web portal service, and test its resilience to data tampering. We also assess its prospects to provide a traceable and useful audit trail of trial data for regulators, and a flexible service for all members within the clinical trials network. We also improve the way adverse events are currently reported. In conclusion, we advocate that this service could offer an improvement in clinical trial data management, and could bolster trust in the clinical research process and the ease at which regulators can oversee trials.

Managing dissections of the thoracic aorta.

Thoracic aortic dissection is associated with substantial morbidity and mortality, and it requires timely and accurate diagnosis and treatment. Long-term antihypertensive therapy remains critical for the treatment of this disease. Surgical intervention, although still a formidable undertaking, has evolved to better address both acute and chronic dissection, and the results have improved. Basic and clinical research, as well as technological advances, have increased our understanding of this challenging disease state.

Ten-year follow-up of a province-wide cohort of surgical lung cancer patients in Nova Scotia.

BACKGROUND: After a diagnosis of lung carcinoma, survival is poor for all patients. We sought to assess 10-year survival and predictors of outcome after surgery for lung cancer in Nova Scotia. METHODS: We identified all patients n = 130) undergoing resection for lung cancer in Nova Scotia in 1994 from the Nova Scotia Cancer Registry and hospital charts and followed them prospectively for 10 years. We used Cox proportional hazards modelling to identify predictors of survival. RESULTS: The patients' mean age at operation was 67.7 (standard deviation [SD] 8.2) years, and 70% of the patients were men. Most of the operations n = 80, 61.5%) were performed in Halifax, and adenocarcinoma n = 55, 42.3%) was the most common histologic type. At the time of surgery, 66.9% of the cases were stage 1, 20.0% were stage 2 and 13.1% were stage 3. Survival at 5 and 10 years was 34% and 13%, respectively. Age of 70 years or older (hazard ratio [HR] 1.79, 95% confidence interval [CI] 1.20-2.68), large cell carcinoma (HR 2.27, 95% CI 1.31-3.94) and stage 3 cancer (HR 2.21, 95% CI 1.25-3.90) were significant independent predictors of survival. Hospital site was not associated with any difference in survival (p = 0.66), although there was a trend toward differential rates of lymph node sampling across sites (p = 0.06). The presence of node sampling was associated with improved survival in a separate multivariate model (HR 0.51, 95% CI 0.29-0.89). CONCLUSION: Actuarial survival after resection of lung carcinoma in Nova Scotia in 1994 was 34% at 5 years and 13% after 10 years. Age, stage and histology are independent predictors of survival; lymph node sampling was associated with greater survival.

Impact of cardiac intraoperative precursor events on adverse outcomes.

BACKGROUND: Although extensive study has been directed at the influence of patient factors and comorbidities on cardiac surgical outcomes, less attention has been focused on process. We sought to examine the relationship between intraoperative precursor events (those events that precede and are requisite for the occurrence of an adverse event) and adverse outcomes themselves. METHODS: Anonymous, prospectively collected intraoperative data was merged with database outcomes for 450 patients undergoing major adult cardiac operations. Precursor events were categorized by type, person most affected, severity, and compensation. Number and categories of precursor events were analyzed as predictors of a composite outcome combining death or near miss complications (DNM), using logistic regression. RESULTS: Precursor events occurred more frequently in cases with a DNM outcome than in those with no adverse event (2.7 +/- 2.4 vs 2.0 +/- 2.3/procedure, P = .005). After adjustment for other patient characteristics, the number of precursor events remained an independent predictor of DNM (RR, 1.14 per event [1.04 to 1.24]). Of 990 events, 35.6% related to management, 28.8% were technical, and 22.8% were environment-related. The surgeon was most affected in 40.8%, and 16.5% were of major severity. When categories of precursor events were analyzed, major severity events and those most affecting the surgeon were independent predictors of DNM. CONCLUSIONS: More detailed study of process in complex operations may lead to improved quality of care and patient safety. Special attention must be paid particularly to high risk patients and high risk precursor events.

Long-term Outcomes of Multiple Arterial Coronary Artery Bypass Grafting: A Population-Based Study of Patients in British Columbia, Canada.

Importance: Although the long-term survival advantage of multiple arterial grafting (MAG) vs the standard use of left internal thoracic artery (LITA) supplemented by saphenous vein grafts (LITA+SVG) has been demonstrated in several observational studies, to our knowledge its safety and other long-term clinical benefits in a large, population-based cohort are unknown. Objective: To compare the safety and long-term outcomes of MAG vs LITA+SVG among overall and selected subgroups of patients. Design, Setting, and Participants: In this population-based observational study, we included 20 076 adult patients with triple-vessel or left-main disease who underwent primary isolated coronary artery bypass grafting (MAG, n = 5580; LITA+SVG, n = 14 496) in the province of British Columbia, Canada, from January 2000 to December 2014, with follow-up to December 2015. We performed propensity-score analyses by weighting and matching and multivariable Cox regression to minimize treatment selection bias. Exposures: Multiple arterial grafting or LITA+SVG. Main Outcomes and Measures: Mortality, repeated revascularization, myocardial infarction, heart failure, and stroke. Results: Of 5580 participants who underwent MAG, 586 (11%) were women and the mean (SD) age was 60 (8.7) years. Of 14 496 participants who underwent LITA+SVG, 2803 (19%) were women and the mean (SD) age was 68 (8.9) years. The median (interquartile range) follow-up time was 9.1 (5.1-12.6) years and 8.1 (4.5-11.7) years for the groups receiving MAG and LITA+SVG, respectively. Compared with LITA+SVG, MAG was associated with reduced mortality rates (hazard ratio [HR], 0.79; 95% CI, 0.72-0.87) and repeated revascularization rates (HR, 0.74; 95% CI, 0.66-0.84) in 15-year follow-up and reduced incidences of myocardial infarction (HR, 0.63; 95% CI, 0.47-0.85) and heart failure (HR, 0.79; 95% CI, 0.64-0.98) in 7-year follow-up. The long-term benefits were coherent by all 3 statistical methods and persisted among patient subgroups with diabetes, obesity, moderately impaired ejection fraction, chronic obstructive pulmonary disease, peripheral vascular disease, or renal disease. Multiple arterial grafting was not associated with increased morbidity or mortality rates at 30 days overall or within patient subgroups. Conclusions and Relevance: Compared with LITA+SVG, MAG is associated with reduced mortality, repeated revascularization, myocardial infarction, and heart failure among patients with multivessel disease who are undergoing coronary artery bypass grafting without increased mortality or other adverse events at 30 days. The long-term benefits consistently observed across multiple outcomes and subgroups support the consideration of MAG for a broader spectrum of patients who are undergoing coronary artery bypass grafting in routine practice.

Prospective assessment of intraoperative precursor events during cardiac surgery.

OBJECTIVE: Increasing attention has been afforded to the ubiquity of medical error and associated adverse events in medicine. There remains little data on the frequency and nature of precursor events in cardiac surgery, and we sought to characterize this. METHODS: Detailed, anonymous information regarding intraoperative precursor events (which may result in adverse events) was collected prospectively from six key members of the operating team during 464 major adult cardiac surgical cases at three hospitals and were analyzed with univariable statistical methods. RESULTS: During 464 cardiac surgical procedures, 1627 reports of problematic precursor events were collected for an average of 3.5 and maximum of 26 per procedure. 73.3% of cases had at least one recorded event. One-third (33.3%) of events occurred prior to the first incision, and 31.2% of events occurred while on bypass. While 68.0% of events were regarded as minor in severity (e.g., delays and missing equipment), a substantial proportion (32.0%) was considered major and included anastomotic problems, pump failure, and drug errors. Most problems (90.4%) were reported as being compensated for, although many (30.9%) were never discussed among the team. Major events were more likely to be discussed (p<0.0001) and less likely to have been previously encountered (p=0.0005). Perceptions of the severity and compensation of events varied across the team, as did temporal patterns of reporting (p<0.0001). CONCLUSIONS: A wide range of problematic precursor events occurs during the majority of cardiac surgery procedures. Attention to causes and ways of preventing these precursor events could have an impact on the rate of significant errors and improve the safety of cardiac surgery.

Anastomotic complications after tracheal resection: prognostic factors and management.

OBJECTIVE: We sought to identify risk factors for anastomotic complications after tracheal resection and to describe the management of these patients. METHODS: This was a single-institution, retrospective review of 901 patients who underwent tracheal resection. RESULTS: The indications for tracheal resection were postintubation tracheal stenosis in 589 patients, tumor in 208, idiopathic laryngotracheal stenosis in 83, and tracheoesophageal fistula in 21. Anastomotic complications occurred in 81 patients (9%). Eleven patients (1%) died after operation, 6 of anastomotic complications and 5 of other causes (odds ratio 13.0, P = .0001 for risk of death after anastomotic complication). At the end of treatment, 853 patients (95%) had a good result, whereas 37 patients (4%) had an airway maintained by tracheostomy or T-tube. The treatments of patients with an anastomotic complication were as follows: multiple dilations (n = 2), temporary tracheostomy (n = 7), temporary T-tube (n = 16), permanent tracheostomy (n = 14), permanent T-tube (n = 20), and reoperation (n = 16). Stepwise multivariable analysis revealed the following predictors of anastomotic complications: reoperation (odds ratio 3.03, 95% confidence interval 1.69-5.43, P = .002), diabetes (odds ratio 3.32, 95% confidence interval 1.76-6.26, P = .002), lengthy (> or =4 cm) resections (odds ratio 2.01, 95% confidence interval 1.21-3.35, P = .007), laryngotracheal resection (odds ratio 1.80, 95% confidence interval 1.07-3.01, P = .03), age 17 years or younger (odds ratio 2.26, 95% confidence interval 1.09-4.68, P = .03), and need for tracheostomy before operation (odds ratio 1.79, 95% confidence interval 1.03-3.14, P = .04). CONCLUSIONS: Tracheal resection is usually successful and has a low mortality. Anastomotic complications are uncommon, and important risk factors are reoperation, diabetes, lengthy resections, laryngotracheal resections, young age (pediatric patients), and the need for tracheostomy before operation.

Open repair of thoracoabdominal aortic aneurysm in the modern surgical era: contemporary outcomes in 509 patients.

BACKGROUND: Recent technologic advances in endovascular devices have led to alternative approaches to thoracoabdominal aortic aneurysm (TAAA) repair; these innovative approaches must be compared with the "gold standard" of conventional open TAAA repair. To facilitate such comparisons, we evaluated contemporary outcomes of open TAAA repair. STUDY DESIGN: We retrospectively reviewed and analyzed data collected prospectively between May 2006 and October 2010 regarding 509 consecutive patients who underwent TAAA repair. Standard univariate statistical comparisons were performed, as well as multivariable modeling, to identify predictors of survival. RESULTS: A total of 305 patients (59.9%) had degenerative aneurysms without dissection, and 204 (40.1%) had aortic dissection. There were 104 (20.4%) urgent or emergent repairs and 26 (5.1%) ruptured aneurysms. Operative adjuncts were used selectively. Of the 290 patients (57.0%) who underwent extensive repairs (Crawford extents I and II), 282 (97.2%) had cerebrospinal fluid drainage, 257 (88.6%) had left heart bypass, and 213 (73.4%) had intercostal/lumbar artery reattachment. The overall operative survival rate was 92.1% (469 of 509), and survival was better after elective repairs (93.8% [380 of 405]) than after urgent or emergent operations (85.6% [89 of 104], p = 0.005). Renal failure necessitating hemodialysis at discharge developed in 30 patients (5.9%). Permanent paraplegia occurred in 13 patients (2.6%). Actuarial survival was 79.1% ± 2.0% at 2 years. CONCLUSIONS: Contemporary open TAAA repair is characterized by respectable early outcomes, particularly when repair is elective. Such results should be compared with those of evolving approaches, including endovascular and hybrid repairs.

Axillary artery cannulation in surgery for acute or subacute ascending aortic dissections.

BACKGROUND: Perfusion through the right axillary artery is an alternative to aortic or femoral artery cannulation during surgery for ascending aortic dissections. The results of this strategy, particularly beyond the immediate postoperative period, are not well described. METHODS: Eighty-three patients (median age, 58 years) with acute or subacute ascending aortic dissection underwent surgical repair with right axillary artery perfusion through an interposition Dacron graft. Sixty-five patients (78%) had DeBakey type I dissections. Procedures performed concomitantly with ascending aortic replacement included root replacement (n = 16; 19%), aortic valve repair or replacement (n = 51; 61%), and coronary artery bypass grafting (n = 13; 16%). Hypothermic circulatory arrest with antegrade cerebral perfusion was used in the majority of patients (n = 60; 72%). We retrospectively studied short-term and midterm outcomes, including survival and complications relating to the axillary cannulation. RESULTS: No patient incurred intraoperative axillary artery injuries or had arm ischemia. Fourteen patients (17%) died in the hospital or within 30 days of surgery, and 9 patients (11%) had strokes. Actuarial survival was 73% +/- 5% at 1 year and 64% +/- 6% at 3 years. Forty-six of the 57 surviving patients could be contacted by telephone; they reported few late complications related to the axillary artery cannulation site. These complications included 1 case each of right-arm weakness and right-arm numbness. CONCLUSIONS: Surgical repair of acute aortic dissection with right axillary artery perfusion can be performed safely, with a relatively low risk of stroke and a high probability of midterm survival.

Technical considerations to avoid pitfalls during transapical aortic valve implantation.

OBJECTIVE: Transapical aortic valve implantation is a recent therapeutic advance for aortic valvular disease. We sought to identify complications--and the relevant technical and management considerations--from our learning curve with this procedure. METHODS: We retrospectively reviewed perioperative complications during the first 60 transapical aortic valve implantations at a single institution, performed under compassionate release for patients who were candidates neither for conventional aortic valve replacement nor for transfemoral aortic valve implantation. Access was through a small left anterolateral thoracotomy. Particular attention was paid to securing the apical access site. Rapid ventricular pacing to reduce cardiac forward flow was used during balloon valvuloplasty and valve deployment. Careful positioning was guided by echocardiography and fluoroscopy. RESULTS: This was a select, high-risk (mean Society of Thoracic Surgeons score, 12.3% +/- 7.8% mortality) cohort. Mean age was 81.1 +/- 7.8 years. Technical success was achieved in 59 (98.3%) cases. One valve was malpositioned too far toward the ventricle, necessitating that a second device be implanted within it. In-hospital, 30-day mortality was 18.3% (11 deaths) overall, decreasing from 33.3% in the first 15 patients to 13.3% in the subsequent 45 patients. The only intraoperative death probably resulted from left main ostial obstruction by extensively calcified aortic cusps. Significant left ventricular apical bleeding occurred in 3 (5.0%) patients. Other complications included stroke in 2 (3.3%) patients and permanent atrioventricular block in 3 (5.0%). There were 4 (6.6%) cases of late pseudoaneurysm of the left ventricular apical access site. CONCLUSIONS: Important lessons have been learned from our early experience with transapical aortic valve implantation, and these may guide others as this technology is adopted more broadly.

Transapical transcatheter mitral valve-in-valve implantation in a human.

We describe a human transcatheter transapical mitral valve implant within a mitral bioprosthesis (valve-in-valve). A high-risk, 80-year-old man with symptomatic bioprosthetic mitral stenosis was positioned for anterior minithoracotomy. Left ventricular apical access was obtained. After balloon valvuloplasty, a cuffed, 26-mm Cribier-Edwards transcatheter valve (Edwards Lifesciences LLC, Irvine, CA) was deployed within the mitral xenograft, using rapid ventricular pacing. The transcatheter valve functioned properly postoperatively; however, the patient died of multiple organ dysfunction.

Learning in a new cardiac surgical center: an analysis of precursor events.

BACKGROUND: Few studies of learning in the health care sector have analyzed measures of process, as opposed to outcomes. We assessed the learning curve for a new cardiac surgical center using precursor events (incidents or circumstances required for the occurrence of adverse outcomes). METHODS: Intraoperative precursor events were recorded prospectively during major adult cardiac operations, categorized by blinded adjudicators, and counted for each case (overall and according to these categories). Trends in the number of precursor events were analyzed by hospital and by defining 10 equal-sized groups across time, as were trends in outcomes obtained from institutional databases. Results from the first 101 cases performed at a new cardiac surgical site (hospital A) were compared with 2 established centers. RESULTS: A steep reduction in the total number of precursor events over time was observed in the early experience of hospital A (9.2 +/- 4.9 to 2.0 +/- 1.2 events per case, from first to last decile of time, P(trend) < .0001) compared with qualitatively stable levels in the other hospitals; this reduction was driven largely by decreases in the minor severity (P(trend) < .0001), compensated (P(trend) < .0001), and environment (P(trend) < .0001) categories of precursor events. No detectable changes over time were observed in postoperative mortality and complications. No significant improvement was observed in patient comorbid conditions or medical status over time to explain the trend in hospital A. CONCLUSION: Analyzing and targeting specific kinds of process-related failures (precursor events) may provide a novel and sensitive means of tracking, deconstructing, and optimizing organizational learning in medicine.