Interpretability and fairness evaluation of deep learning models on MIMIC-IV dataset.

The recent release of large-scale healthcare datasets has greatly propelled the research of data-driven deep learning models for healthcare applications. However, due to the nature of such deep black-boxed models, concerns about interpretability, fairness, and biases in healthcare scenarios where human lives are at stake call for a careful and thorough examination of both datasets and models. In this work, we focus on MIMIC-IV (Medical Information Mart for Intensive Care, version IV), the largest publicly available healthcare dataset, and conduct comprehensive analyses of interpretability as well as dataset representation bias and prediction fairness of deep learning models for in-hospital mortality prediction. First, we analyze the interpretability of deep learning mortality prediction models and observe that (1) the best-performing interpretability method successfully identifies critical features for mortality prediction on various prediction models as well as recognizing new important features that domain knowledge does not consider; (2) prediction models rely on demographic features, raising concerns in fairness. Therefore, we then evaluate the fairness of models and do observe the unfairness: (1) there exists disparate treatment in prescribing mechanical ventilation among patient groups across ethnicity, gender and age; (2) models often rely on racial attributes unequally across subgroups to generate their predictions. We further draw concrete connections between interpretability methods and fairness metrics by showing how feature importance from interpretability methods can be beneficial in quantifying potential disparities in mortality predictors. Our analysis demonstrates that the prediction performance is not the only factor to consider when evaluating models for healthcare applications, since high prediction performance might be the result of unfair utilization of demographic features. Our findings suggest that future research in AI models for healthcare applications can benefit from utilizing the analysis workflow of interpretability and fairness as well as verifying if models achieve superior performance at the cost of introducing bias.

PolSIRD: Modeling Epidemic Spread Under Intervention Policies: Analyzing the First Wave of COVID-19 in the USA.

Epidemic spread in a population is traditionally modeled via compartmentalized models which represent the free evolution of disease in the absence of any intervention policies. In addition, these models assume full observability of disease cases and do not account for under-reporting. We present a mathematical model, namely PolSIRD, which accounts for the under-reporting by introducing an observation mechanism. It also captures the effects of intervention policies on the disease spread parameters by leveraging intervention policy data along with the reported disease cases. Furthermore, we allow our recurrent model to learn the initial hidden state of all compartments end-to-end along with other parameters via gradient-based training. We apply our model to the spread of the recent global outbreak of COVID-19 in the USA, where our model outperforms the methods employed by the CDC in predicting the spread. We also provide counterfactual simulations from our model to analyze the effect of lifting the intervention policies prematurely and our model correctly predicts the second wave of the epidemic.

Erratum: Self-made thoracic needled suspending device with a snare: An excellent aid for uniportal video-assisted thoracic lobectomy and segmentectomy for lung cancer.

[This corrects the article DOI: 10.3892/ol.2019.10030.].

Surgical management of organizing pneumonia: a retrospective study of 24 cases in a single Centre.

BACKGROUND: Organizing pneumonia (OP) is a rare disease that is often easily misdiagnosed as a malignancy. The diagnosis of OP can prove quite challenging. Patients typically receive treatment with high-dose corticosteroids. Relapse is common if corticosteroid treatment is reduced or stopped. However, given that long-term corticosteroid treatment often results in significant side-effects, the aim of this study was to discuss the diagnosis and surgical treatment of OP. MATERIAL AND METHODS: The medical records of 24 patients with pathologically diagnosed OP between October 2007 and January 2019 were retrospectively reviewed. All patients underwent thoracic computed tomography (CT) and transbronchial biopsy or CT-guided percutaneous needle aspiration. We analysed the clinical manifestations, radiological findings, diagnostic methods, treatment, and follow-up outcomes of all patients. RESULTS: In total, 24 patients with OP were identified. The study included 17 (70.8%) men and 7 (29.2%) women, and the mean age was 61.25 ± 11.33 years (range: 31-82). The most common symptom was cough (n = 16; 66.6%), and the most common radiological finding was consolidation (n = 13; 54.2%) on thoracic CT. The diagnosis of OP was made by transbronchial biopsy in 11 patients (45.8%), and percutaneous needle aspiration biopsy in 13 (54.2%). We performed 11 wedge resections, 9 segmentectomy, and 4 lobectomies. Twenty patients underwent video-assisted thoracoscopic surgery (VATS), and 4 underwent thoracotomy. Complete lesion resection was obtained in all patients, and all patients were discharged from the hospital between 5 and 11 days after surgery. The mean follow-up period was 59.1 ± 34.5 (range: 2-134) months. Residual lesions or local or distant recurrence were not observed. CONCLUSIONS: OP is a rare disease, and the exact aetiology remains unclear. Preoperative diagnosis is difficult to achieve despite the use of transbronchial biopsy or CT-guided percutaneous needle aspiration. Complete surgical resection represents an effective method for the treatment of OP.

Esophageal adenocarcinoma with leukemoid reaction: a case report.

BACKGROUND: Leukemoid reaction (LR) is defined as a reactive leucocytosis with WBC counts exceeding 50,000/mm3, and a significant increase in early neutrophil precursors. LR may be a paraneoplastic manifestation of various malignant tumors. Tumor-related LR is a kind of neoplastic syndrome, unrelated to an infection or other diseases. CASE PRESENTATION: A 74-year-old male visited a local doctor with a 20-day history of progressive dysphagia. The complete blood count revealed leucocytosis. Bone marrow aspirates and a biopsy confirmed LR and excluded chronic myelogenous leukemia. Following radical esophagectomy for an adenocarcinoma the WBC counts successively decreased to 10,450/mm3 and 8670/mm3 within 1 week and 1 month, respectively. CONCLUSION: We report a rare case of esophageal adenocarcinoma complicated with excessive leucocytosis caused by paraneoplastic LR; we also present a review of literature and an investigation of the clinical features. To our knowledge, this is the first report of LR associated with esophageal adenocarcinoma.

Self-made thoracic needled suspending device with a snare: An excellent aid for uniportal video-assisted thoracic lobectomy and segmentectomy for lung cancer.

Safety and feasibility of the self-made thoracic needled suspending device with a snare in the uniportal video-assisted thoracic lobectomy and segmentectomy for the treatment of non-small cell lung cancer were explored. In total, 80 pulmonary lung major resections (including lobectomy and segmental resections) with systematic mediastinal lymphadenectomy were retrospectively analyzed. Patients were randomly divided into an observation group and a control group. In the observation group, the device was used to hang affected lungs, left and right vagus nerve at the level of tracheal bifurcation, the arch of azygos vein, left phrenic nerve and left and right bronchus on the chest wall to offer a better exposure of the operation field. In the control group, the conventional uniportal video-assisted thoracic surgery was performed without using the self-made device. Systematic mediastinal lymphadenectomy was performed in both groups. Operation time, intraoperative blood loss, postoperative extubation time, hospital stay and perioperative complications in the early stage of patients in both groups were compared. The operation time 120.2±40.32 min, intraoperative blood loss 100.51±50.23 ml, and postoperative suction drainage volume 208±97.56 ml/day in the observation group were significantly different from those in the control group (P<0.05), and there were no significant differences in postoperative extubation time, hospital stay and perioperative complications between the two groups (P>0.05). The self-made thoracic needled suspending device with a snare is an excellent helper for uniportal video-assisted thoracic surgery, because it helps to expose surgical field and has no postoperative cicatrisation at puncture point on the wall of the chest. The device and its use are worthy of promotion.

Benchmarking deep learning models on large healthcare datasets.

Deep learning models (aka Deep Neural Networks) have revolutionized many fields including computer vision, natural language processing, speech recognition, and is being increasingly used in clinical healthcare applications. However, few works exist which have benchmarked the performance of the deep learning models with respect to the state-of-the-art machine learning models and prognostic scoring systems on publicly available healthcare datasets. In this paper, we present the benchmarking results for several clinical prediction tasks such as mortality prediction, length of stay prediction, and ICD-9 code group prediction using Deep Learning models, ensemble of machine learning models (Super Learner algorithm), SAPS II and SOFA scores. We used the Medical Information Mart for Intensive Care III (MIMIC-III) (v1.4) publicly available dataset, which includes all patients admitted to an ICU at the Beth Israel Deaconess Medical Center from 2001 to 2012, for the benchmarking tasks. Our results show that deep learning models consistently outperform all the other approaches especially when the 'raw' clinical time series data is used as input features to the models.