Limited generalizability of deep learning algorithm for pediatric pneumonia classification on external data.

PURPOSE: (1) Develop a deep learning system (DLS) to identify pneumonia in pediatric chest radiographs, and (2) evaluate its generalizability by comparing its performance on internal versus external test datasets. METHODS: Radiographs of patients between 1 and 5 years old from the Guangzhou Women and Children's Medical Center (Guangzhou dataset) and NIH ChestXray14 dataset were included. We utilized 5232 radiographs from the Guangzhou dataset to train a ResNet-50 deep convolutional neural network (DCNN) to identify pediatric pneumonia. DCNN testing was performed on a holdout set of 624 radiographs from the Guangzhou dataset (internal test set) and 383 radiographs from the NIH ChestXray14 dataset (external test set). Receiver operating characteristic curves were generated, and area under the curve (AUC) was compared via DeLong parametric method. Colored heatmaps were generated using class activation mapping (CAM) to identify important image pixels for DCNN decision-making. RESULTS: The DCNN achieved AUC of 0.95 and 0.54 for identifying pneumonia on internal and external test sets, respectively (p < 0.0001). Heatmaps generated by the DCNN showed the algorithm focused on clinically relevant features for images from the internal test set, but not for images from the external test set. CONCLUSION: Our model had high performance when tested on an internal dataset but significantly lower accuracy when tested on an external dataset. Likewise, marked differences existed in the clinical relevance of features highlighted by heatmaps generated from internal versus external datasets. This study underscores potential limitations in the generalizability of such DLS models.

A supervised machine learning model for identifying predictive factors for recommending head and neck cancer surgery.

BACKGROUND: New patient referrals are often processed by practice coordinators with little-to-no medical background. Treatment delays due to incorrect referral processing, however, have detrimental consequences. Identifying variables that are associated with a higher likelihood of surgical oncological resection may improve patient referral processing and expedite the time to treatment. The study objective is to develop a supervised machine learning (ML) platform that identifies relevant variables associated with head and neck surgical resection. METHODS: A retrospective cohort study was conducted on 64 222 patient datapoints from the SEER database. RESULTS: The random forest ML model correctly classified patients who were offered head and neck surgery with an 81% accuracy rate. The sensitivity and specificity rates were 86% and 71%. The positive and negative predictive values were 85% and 73%. CONCLUSIONS: ML modeling accurately predicts head and neck cancer surgery recommendations based on patient and cancer information from a large population-based dataset. ML adjuncts for referral processing may decrease the time to treatment for patients with cancer.

Selective serotonin reuptake inhibitors may lead to improved cataract surgery outcomes in patients with amblyopia.

PURPOSE: To evaluate the effects of selective serotonin reuptake inhibitor (SSRI)/serotonin norepinephrine reuptake inhibitor (SNRI) medications in combination with cataract surgery in treating amblyopia in adult patients. PATIENTS AND METHODS: A retrospective chart review study was conducted on patients who had undergone cataract surgery at the Johns Hopkins Hospital Wilmer Eye Institute. Six inclusion criteria were used to assess patient eligibility: 1) >18 years of age, 2) diagnosis of amblyopia, 3) diagnosis of cataract and treatment with surgery, 4) electronic medical record contains pre-surgery and post-surgery visual acuity (VA) measurements, 5) electronic medical record contains information on whether the patient was ever prescribed a SSRI/SNRI and the treatment duration, and 6) interocular VA difference of two lines or more on Snellen chart prior to cataract surgery. From each record, preoperative VA, postoperative VA, date of surgery, date at which postoperative VA was measured, and age at surgery were collected. RESULTS: A total of 237 patients were included, with 38 of them being on SSRI/SNRI. The mean improvement in VA after surgery was not significantly greater in patients on SSRI/SNRI (SSRI/SNRI: -0.276 logMAR, control: -0.192 logMAR, p=0.15). Multivariable regression was subsequently performed and while holding all other variables constant, demonstrated a statistically significant improvement in VA in patients on SSRI/SNRI (95% CI: -0.194, -0.0116, p=0.03). The regression analysis further demonstrated that advanced age has an adverse effect on the change in post-op VA (CI: 3.34×10-3 logMAR, 9.77×10-3 logMAR, p<0.005). Worse baseline VA is associated with a greater improvement in post-op VA (95% CI: -0.659 logMAR, -0.463 logMAR, p<0.005) but adverse effect on the absolute post-op VA (95% CI: 0.341 logMAR, 0.544 logMAR, p<0.005). CONCLUSION: This study suggests that patients with amblyopia undergoing cataract surgery may potentially have a greater visual improvement when treated with SSRI/SNRIs.

Antimicrobial Susceptibility, Biotypes and Phylotypes of Clinical Cutibacterium (Formerly Propionibacterium) acnes Strains Isolated from Acne Patients: An Observational Study.

INTRODUCTION: The aim of this study was to investigate the distribution of antimicrobial susceptibility, biotypes and phylotypes of clinical Cutibacterium acnes (C. acnes, formerly Propionibacterium acnes) isolates as well as the relationship among demographic factors, C. acnes biotypes and phylotypes. METHODS: Cutibacterium acnes was collected from the skin lesions of acne patients who visited the dermatologic department of Huashan Hospital in Shanghai from October 2016 to March 2017. The agar dilution method was conducted to determine the minimum inhibitory concentrations (MICs) of C. acnes, the fermentation test to identify biotypes and then multiplex touchdown polymerase chain reaction (PCR) to identify phylotypes. RESULTS: Of the 63 C. acnes strains we isolated, 18 (28.6%), 31 (49.2%) and 4 (6.3%) strains were resistant to clindamycin, erythromycin and moxifloxacin, respectively; no strains were resistant to tetracycline, minocycline, fusidic acid or ß-lactam, while metronidazole was completely resisted; 3 strains showed multidrug resistance (MDR). Biotype III (BIII) was the major biotype (50.8%) followed by BI and BV (both 15.9%), BII (12.7%) and lastly BIV (4.8%). IA1 was the predominant phylotype (71.4%) followed by IA2 (19.0%), II (4.8%), IB (3.2%) and IC (1.6%), while III was not detected. Significant differences were observed in the severity of disease: different degrees of acne severity reflected different biotype and phylotype distributions, and the biotype distribution of mild acne was different from that of moderate acne; the phylotype distribution of moderate acne varies from that of severe acne, too. Additionally, there was no significant difference in the distribution of biotypes or phylotypes between resistant and susceptible strains. CONCLUSION: Erythromycin and clindamycin resistances are the most common in clinical C. acnes strains; BIII is the predominant biotype and IA1 is the major phylotype of C. acnes, which are mainly related to disease severity.

Interconnectedness of global hemostasis assay parameters in simultaneously evaluated thrombin generation, fibrin generation and clot lysis in normal plasma.

BACKGROUND: Fluorogenic thrombin generation (TG) assays and turbidity-based fibrin generation (FG)- and fibrinolysis (FL)-resistance assays have been sought to assess bleeding and clotting disorders. Theoretically, TG, FG and FL tests should provide overlapping information because thrombin is responsible for FG and induces protection from FL. The relationships between TG, FG and FL parameters remain poorly investigated, partly because existing experimental systems do not permit simultaneous detection of both TG and FG in the same sample of plasma, and are instead tested in separate experiments. OBJECTIVES AND METHODS: We evaluated the potential benefits of a combined TG/FG/FL assay by testing responses of normal plasma to a wide range of tissue factor (TF) and tissue plasminogen activator (tPA) concentrations. Correlations between multiple parameters extracted from the TG and FG/FL curves were also compared. RESULTS: Rate of FG correlated well with TG peak height at all TF concentrations, but correlations between TG and FL parameters depended on the TF concentration. Without thrombomodulin, all FG/FL parameters at high TF could be predicted from TG parameters and no FL protection was observed. With thrombomodulin and high TF, TF-dependent FL protection did not correlate with TF-dependent TG. The fluorogenic thrombin substrate did not interfere with optical density readings, and meaningful tPA concentrations did not interfere with TG readings. CONCLUSIONS: In normal plasma, TG, FG and FL parameters may provide interchangeable information. Evaluation of FL-resistance may provide additional data under special assay conditions, but the value of this information should be studied under disease conditions.