Limited Number of Cases May Yield Generalizable Models, a Proof of Concept in Deep Learning for Colon Histology.

BACKGROUND: Little is known about the effect of a minimum number of slides required in generating image datasets used to build generalizable machine-learning (ML) models. In addition, the assumption within deep learning is that the increased number of training images will always enhance accuracy and that the initial validation accuracy of the models correlates well with their generalizability. In this pilot study, we have been able to test the above assumptions to gain a better understanding of such platforms, especially when data resources are limited. METHODS: Using 10 colon histology slides (5 carcinoma and 5 benign), we were able to acquire 1000 partially overlapping images (Dataset A) that were then trained and tested on three convolutional neural networks (CNNs), ResNet50, AlexNet, and SqueezeNet, to build a large number of unique models for a simple task of classifying colon histopathology into benign and malignant. Different quantities of images (10-1000) from Dataset A were used to construct >200 unique CNN models whose performances were individually assessed. The performance of these models was initially assessed using 20% of Dataset A's images (not included in the training phase) to acquire their initial validation accuracy (internal accuracy) followed by their generalization accuracy on Dataset B (a very distinct secondary test set acquired from public domain online sources). RESULTS: All CNNs showed similar peak internal accuracies (>97%) from the Dataset A test set. Peak accuracies for the external novel test set (Dataset B), an assessment of the ability to generalize, showed marked variation (ResNet50: 98%; AlexNet: 92%; and SqueezeNet: 80%). The models with the highest accuracy were not generated using the largest training sets. Further, a model's internal accuracy did not always correlate with its generalization accuracy. The results were obtained using an optimized number of cases and controls. CONCLUSIONS: Increasing the number of images in a training set does not always improve model accuracy, and significant numbers of cases may not always be needed for generalization, especially for simple tasks. Different CNNs reach peak accuracy with different training set sizes. Further studies are required to evaluate the above findings in more complex ML models prior to using such ancillary tools in clinical settings.

Optimal use of fresh frozen plasma.

Fresh frozen plasma contains a number of therapeutically useful substances, most notably coagulation factors. As with any transfusion, there are risks associated with plasma transfusion. Ironically, the risk of viral transmission (human immunodeficiency virus or hepatitis), although widely publicized, is extremely small. On the other hand, less well-known, noninfectious complications are common. Indeed, these noninfectious complications are the most significant cause of morbidity and mortality following transfusion. Although certain patients undeniably benefit from plasma transfusion, the benefit for many patients is less clear. This review will discuss indications for plasma transfusion, the associated risks, and special considerations for plasma administration.

Implementing and validating transcutaneous bilirubinometry for neonates.

Noninvasive, transcutaneous bilirubin (TcB) measurement is an attractive option for neonates, but opinions about its usefulness vary among studies. We collected paired measurements of TcB and serum bilirubin (SB) in 343 term neonates using the BiliCheck meter (SpectRx, Norcross, GA) and 3 different SB methods. Correlations between SB and TcB were similar for all laboratory methods and TcB measurement sites. However, TcB bias varied depending on the comparison SB method and TcB measurement site. TcB bias also varied with race when measurements were done on the forehead but not when they were done on the sternum. Several factors must be considered before implementing TcB measurement: (1) Each laboratory instrument has its own unique relationship to TcB. (2) The chosen measurement site affects the relationship. (3) Race can affect TcB bias when the measurement is taken on the forehead. Properly used, TcB measurement, especially when taken from the sternum, can be a useful screening method for neonatal jaundice.

Suboptimal effect of a three-factor prothrombin complex concentrate (Profilnine-SD) in correcting supratherapeutic international normalized ratio due to warfarin overdose.

BACKGROUND: Plasma transfusion is standard therapy for urgent warfarin reversal in the United States. "Four-factor" prothrombin complex concentrate (PCC), available in Europe, has advantages over plasma therapy for warfarin reversal; however, only "three-factor" PCCs (containing relatively low Factor [F]VII) are available in the United States. STUDY DESIGN AND METHODS: The efficacy of a three-factor PCC for urgent warfarin reversal was evaluated in 40 patients presenting with supratherapeutic international normalized ratio (ST-INR > 5.0) with bleeding (n = 29) or at high risk for bleeding (n = 11). In 13 patients, pre- and posttherapy vitamin K-dependent factors were assayed. Historical controls (n = 42) treated with plasma alone were used for rate of ST-INR correction comparison. RESULTS: Treatment with plasma alone (mean, 3.6 units) lowered the INR to less than 3.0 in 63 percent of historical controls. Low-dose (25 U/kg) and high-dose (50 U/kg) PCC alone lowered INR to less than 3.0 in 50 and 43 percent of patients, respectively. Additional transfusion of a small amount of plasma (mean, 2.1 units) increased the rate of achieving an INR of less than 3.0 to 89 and 88 percent for low- and high-dose PCC therapy, respectively. FII, F IX, and FX increments were similar for PCC-treated patients with or without supplemental plasma; FVII was significantly higher in the PCC plus plasma group compared to the PCC-only group (p = 0.001). CONCLUSION: Three-factor PCC does not satisfactorily lower ST-INR due to low FVII content. Infusion of a small amount of plasma increases the likelihood of satisfactory INR lowering.

Should plasma be transfused prophylactically before invasive procedures?

PURPOSE OF REVIEW: Plasma transfusion to correct abnormal coagulation test results prior to an invasive procedure is a common clinical practice; however, there are no evidence-based guidelines. This review aims to analyze the most recent publications to either support or disprove such practice. RECENT FINDINGS: Due to heightened awareness of transfusion-related acute lung injury and volume overload in susceptible patients, clinicians are increasingly questioning the validity of prophylactic plasma transfusion. Recently, several articles, reviews and clinical studies (although small and poorly designed) have shown no benefit of prophylactic plasma transfusion in either correcting abnormal coagulation tests or reducing perceived risk of hemorrhage. SUMMARY: The use of sensitive reagents (especially for prothrombin time) has resulted in increased incidence of abnormal preprocedure coagulation screening test results - tests that are not designed to assess risk of bleeding in patients without a history of bleeding. Transfusion of plasma prior to an invasive procedure to correct mild to moderate abnormal test results neither corrects the abnormality nor reduces the perceived bleeding risk.

Toward rational fresh frozen plasma transfusion: The effect of plasma transfusion on coagulation test results.

Numerous published guidelines encourage appropriate use of fresh frozen plasma (FFP). However, adherence is documented as poor. Therefore, we sought to determine the laboratory effect of FFP administration to patients with an international normalized ratio (INR) less than 1.6 (prothrombin time < 1.6 times normal). We found minimally prolonged INRs decreased with treatment of the underlying disease alone. Adding FFP to the treatment failed to change the decrease in INR over time. In addition, we observed that the change in the INR per unit of FFP transfused can be predicted by the pretransfusion INR (INR change = 0.37 [pretransfusion INR] - 0.47; r2 = 0.82). With an observed analytic variation of 3.2%, a significant amount of change in the INR following FFP transfusion is expected at an INR of more than 1.7. Indeed, only 50% of patients with an INR of 1.7 showed a significant change in INR with FFP transfusion. Therefore, transfusion for patients not meeting current FFP guidelines does not reliably reduce the INR and exposes patients to unnecessary risk.

Total laboratory automation can help eliminate the laboratory as a factor in emergency department length of stay.

We obtained data on laboratory turnaround time (TAT) and emergency department (ED) length of stay (LOS). We correlated potassium test TAT outlier percentage (TAT-OP) with ED LOS and found that for each outlier percentage (potassium result > 40 minutes), a projected impact on ED LOS was approximately 2.8 additional minutes (ED LOS = 2.79 TAT-OP + 78.77). To address this issue, we began implementation of a totally automated chemistry system to decrease TAT-OPs. Our TAT means did not change substantially with automation (potassium, 28 to 27 minutes); however, TAT-OPs decreased substantially (potassium, 18% to 5%). Preautomation average ED LOS correlated best with the TAT-OP (r(2) = 0.98; P = .01), but this relationship weakened substantially after automation (r(2) = 0.29; P > .05), suggesting the laboratory was no longer a factor in ED LOS. The postautomation ED LOS correlated best with ED patient volume (r(2) = 0.88; P = .06). Although laboratories have focused on TAT means for performance assessment, our study suggests TAT-OPs are more clinically relevant benchmarks. Furthermore, our findings suggest that total laboratory automation can effectively improve overall laboratory service reliability and help eliminate the laboratory as a factor in ED LOS.

Medical students' perceptions of pathology and the effect of the second-year pathology course.

To explore the perceptions of medical students regarding pathology and the effect of the sophomore pathology course, a questionnaire was given to second-year students from 5 different medical schools at the beginning and again at the completion of their pathology course. The questionnaire was given to students in the class of 1995 and then again for those in the class of 2002, with nearly 1,500 surveys collected over the study period. The survey included questions that sought to determine students' affinity for pathology, their understanding of the typical duties of a pathologist, and how they viewed pathology relative to other specialties and perceived positives and negatives of being a pathologist. Overall, the second-year pathology course had little effect on medical students' perceptions of pathology but did provide some increase in their understanding of pathology as a profession. Responses were in general stable when comparing the different classes. Those students interested in pathology emphasized the intellectual aspects of pathology although being less deterred by negative factors such as limited patient contact. In our sample of medical schools, the sophomore pathology course was ineffective at influencing students' perceptions of pathology. Furthermore, those students who are interested in pathology are drawn because of a perceived fit between their personalities and the perception of pathology as a solely scholarly and isolated specialty. Better education about the practice of pathology in the second year, and more importantly, continuing into the clinical years, is necessary to combat these misconceptions about pathology.

The transverse humeral ligament: a separate anatomical structure or a continuation of the osseous attachment of the rotator cuff?

BACKGROUND: No study to date has isolated the anatomical nature of the transverse humeral ligament and its relationship to the biceps tendon and the anterosuperior portion of the rotator cuff. HYPOTHESIS: There is no separate identifiable transverse humeral ligament, but rather the fibers covering the intertubercular groove are composed of a sling formed by fibers from the subscapularis and supraspinatus tendons. STUDY DESIGN: Descriptive laboratory study. METHODS: A total of 14 shoulder examinations were performed on 7 matched pairs of fresh-frozen cadaveric shoulders. Magnetic resonance imaging scans were performed, followed by gross and microscopic anatomical dissection. RESULTS: In the location of the transverse humeral ligament, magnetic resonance imaging and gross dissection revealed the continuation of superficial fibers of the subscapularis tendon from the tendon body across the intertubercular groove to attach to the greater tuberosity, whereas deeper fibers of the subscapularis tendon inserted on the lesser tuberosity. Longitudinal fibers of the supraspinatus tendon and the coracohumeral ligament were also noted to travel the length of the groove, deep to the other interdigitating fibers but superficial to the biceps tendon. Histologic studies confirmed these gross dissection patterns of fiber attachment and also revealed the absence of elastin fibers, which are more commonly seen in ligamentous structures and are typically absent from tendinous structures. CONCLUSION: There is no identifiable transverse humeral ligament, but rather the fibers covering the intertubercular groove are composed of a sling formed mainly by the fibers of the subscapularis tendon, with contributions from the supraspinatus tendon and the coracohumeral ligament. CLINICAL RELEVANCE: According to our findings, dislocations of the long head of the biceps must disrupt at least the deep fibers of the annular sling created mainly by the subscapularis tendon insertion. This finding provides anatomical support for the findings of a positive biceps tendon subluxation or dislocation and subscapularis tear during glenohumeral arthroscopy with a normal-appearing subscapularis during open surgery or subacromial arthroscopy.

Reducing laboratory turnaround time outliers can reduce emergency department patient length of stay: an 11-hospital study.

Poor core laboratory performance that causes delays in diagnosis and treatment is an impediment to optimal patient care, particularly in high-volume patient care areas such as the emergency department (ED). To evaluate the impact of laboratory performance on patient care outcomes, we obtained data from 11 hospitals related to laboratory test turnaround time (TAT) parameters and ED patient throughput. We observed that the average length of stay (LOS) in the ED correlated significantly with the percentage of total laboratory outliers (R2 = 0.75; P < .01) and to a lesser extent the TAT means (R2 = 0.66; P < .01). Furthermore, improvements in laboratory performance during the study were associated with concurrent decreases in ED LOS. Although in the past, laboratories have focused on TAT means for performance assessment, our observations suggest that a more appropriate method of benchmarking might be to aggressively set clinically driven TAT targets and assess performance as the percentage of results achieving this goal.