Using cluster analysis of cytokines to identify patterns of inflammation in hospitalized patients with community-acquired pneumonia: a pilot study

Pneumonia-Plus: a deep learning model for the classification of bacterial, fungal, and viral pneumonia based on CT tomography

OBJECTIVES

This study aims to develop a deep learning algorithm, Pneumonia-Plus, based on computed tomography (CT) images for accurate classification of bacterial, fungal, and viral pneumonia.

METHODS

A total of 2763 participants with chest CT images and definite pathogen diagnosis were included to train and validate an algorithm. Pneumonia-Plus was prospectively tested on a nonoverlapping dataset of 173 patients. The algorithm's performance in classifying three types of pneumonia was compared to that of three radiologists using the McNemar test to verify its clinical usefulness.

RESULTS

Among the 173 patients, area under the curve (AUC) values for viral, fungal, and bacterial pneumonia were 0.816, 0.715, and 0.934, respectively. Viral pneumonia was accurately classified with sensitivity, specificity, and accuracy of 0.847, 0.919, and 0.873. Three radiologists also showed good consistency with Pneumonia-Plus. The AUC values of bacterial, fungal, and viral pneumonia were 0.480, 0.541, and 0.580 (radiologist 1: 3-year experience); 0.637, 0.693, and 0.730 (radiologist 2: 7-year experience); and 0.734, 0.757, and 0.847 (radiologist 3: 12-year experience), respectively. The McNemar test results for sensitivity showed that the diagnostic performance of the algorithm was significantly better than that of radiologist 1 and radiologist 2 (p < 0.05) in differentiating bacterial and viral pneumonia. Radiologist 3 had a higher diagnostic accuracy than the algorithm.

CONCLUSIONS

The Pneumonia-Plus algorithm is used to differentiate between bacterial, fungal, and viral pneumonia, which has reached the level of an attending radiologist and reduce the risk of misdiagnosis. The Pneumonia-Plus is important for appropriate treatment and avoiding the use of unnecessary antibiotics, and provide timely information to guide clinical decision-making and improve patient outcomes.

CLINICAL RELEVANCE STATEMENT

Pneumonia-Plus algorithm could assist in the accurate classification of pneumonia based on CT images, which has great clinical value in avoiding the use of unnecessary antibiotics, and providing timely information to guide clinical decision-making and improve patient outcomes.

KEY POINTS

• The Pneumonia-Plus algorithm trained from data collected from multiple centers can accurately identify bacterial, fungal, and viral pneumonia. • The Pneumonia-Plus algorithm was found to have better sensitivity in classifying viral and bacterial pneumonia in comparison to radiologist 1 (5-year experience) and radiologist 2 (7-year experience). • The Pneumonia-Plus algorithm is used to differentiate between bacterial, fungal, and viral pneumonia, which has reached the level of an attending radiologist.

Performance of Machine Learning Algorithms for Predicting Adverse Outcomes in Community-Acquired Pneumonia

Background: The ability to assess adverse outcomes in patients with community-acquired pneumonia (CAP) could improve clinical decision-making to enhance clinical practice, but the studies remain insufficient, and similarly, few machine learning (ML) models have been developed.

Objective: We aimed to explore the effectiveness of predicting adverse outcomes in CAP through ML models.

Methods: A total of 2,302 adults with CAP who were prospectively recruited between January 2012 and March 2015 across three cities in South America were extracted from DryadData. After a 70:30 training set: test set split of the data, nine ML algorithms were executed and their diagnostic accuracy was measured mainly by the area under the curve (AUC). The nine ML algorithms included decision trees, random forests, extreme gradient boosting (XGBoost), support vector machines, Naïve Bayes, K-nearest neighbors, ridge regression, logistic regression without regularization, and neural networks. The adverse outcomes included hospital admission, mortality, ICU admission, and one-year post-enrollment status.

Results: The XGBoost algorithm had the best performance in predicting hospital admission. Its AUC reached 0.921, and accuracy, precision, recall, and F1-score were better than those of other models. In the prediction of ICU admission, a model trained with the XGBoost algorithm showed the best performance with AUC 0.801. XGBoost algorithm also did a good job at predicting one-year post-enrollment status. The results of AUC, accuracy, precision, recall, and F1-score indicated the algorithm had high accuracy and precision. In addition, the best performance was seen by the neural network algorithm when predicting death (AUC 0.831).

Conclusions: ML algorithms, particularly the XGBoost algorithm, were feasible and effective in predicting adverse outcomes of CAP patients. The ML models based on available common clinical features had great potential to guide individual treatment and subsequent clinical decisions.

Effective mathematical modelling of health passes during a pandemic

We study the impact on the epidemiological dynamics of a class of restrictive measures that are aimed at reducing the number of contacts of individuals who have a higher risk of being infected with a transmittable disease. Such measures are currently either implemented or at least discussed in numerous countries worldwide to ward off a potential new wave of COVID-19. They come in the form of Health Passes (HP), which grant full access to public life only to individuals with a certificate that proves that they have either been fully vaccinated, have recovered from a previous infection or have recently tested negative to SARS-Cov-2. We develop both a compartmental model as well as an epidemic Renormalisation Group approach, which is capable of describing the dynamics over a longer period of time, notably an entire epidemiological wave. Introducing different versions of HPs in this model, we are capable of providing quantitative estimates on the effectiveness of the underlying measures as a function of the fraction of the population that is vaccinated and the vaccination rate. We apply our models to the latest COVID-19 wave in several European countries, notably Germany and Austria, which validate our theoretical findings.

Machine Learning in Epidemiology and Health Outcomes Research

Machine learning approaches to modeling of epidemiologic data are becoming increasingly more prevalent in the literature. These methods have the potential to improve our understanding of health and opportunities for intervention, far beyond our past capabilities. This article provides a walkthrough for creating supervised machine learning models with current examples from the literature. From identifying an appropriate sample and selecting features through training, testing, and assessing performance, the end-to-end approach to machine learning can be a daunting task. We take the reader through each step in the process and discuss novel concepts in the area of machine learning, including identifying treatment effects and explaining the output from machine learning models.

[Analysis of Time-to-onset and Onset-pattern of Interstitial Lung Disease after the Administration of Monoclonal Antibody Agents]

The aim of this study has been to investigate the time-to-onset and onset-pattern of drug-induced interstitial lung disease (DILD) after the administration of monoclonal antibodies through the use of the spontaneous adverse reaction reporting system of the Japanese Adverse Drug Event Report database. DILD datasets for adalimumab, bevacizumab, cetuximab, denosumab, golimumab, infliximab, nivolumab, panitumumab, pembrolizumab, tocilizumab, and trastuzumab were used to calculate the median time-to-onset of DILD, as well as the Weibull distribution parameters. The median time-to-onset of DILD for pembrolizumab and infliximab was within 1 month. The median time-to-onset of DILD for cetuximab, nivolumab, panitumumab, bevacizumab, golimumab, trastuzumab, and tocilizumab ranged from 1 to 2 months. The median time-to-onset of DILD for denosumab and adalimumab was more than 2 months. Infliximab, trastuzumab and tocilizumab, and denosumab were estimated to fit the early failure type profile of the Weibull distribution parameters. Cetuximab, nivolumab, panitumumab, bevacizumab, golimumab, and adalimumab were estimated to fit the random failure type profile. Pembrolizumab was estimated to fit the wear out failure type profile. Cluster analysis was performed to classify the time-to-onset patterns of DILD. Hierarchical cluster analysis showed 3 clusters. The findings of this study established both the most likely time period and onset-pattern of DILD that can occur in patients after the administration of monoclonal antibody agents.

[Analysis of Time-to-onset of Interstitial Lung Disease after the Administration of Small Molecule Molecularly-targeted Drugs]

　The aim of this study was to investigate the time-to-onset of drug-induced interstitial lung disease (DILD) following the administration of small molecule molecularly-targeted drugs via the use of the spontaneous adverse reaction reporting system of the Japanese Adverse Drug Event Report database. DILD datasets for afatinib, alectinib, bortezomib, crizotinib, dasatinib, erlotinib, everolimus, gefitinib, imatinib, lapatinib, nilotinib, osimertinib, sorafenib, sunitinib, temsirolimus, and tofacitinib were used to calculate the median onset times of DILD and the Weibull distribution parameters, and to perform the hierarchical cluster analysis. The median onset times of DILD for afatinib, bortezomib, crizotinib, erlotinib, gefitinib, and nilotinib were within one month. The median onset times of DILD for dasatinib, everolimus, lapatinib, osimertinib, and temsirolimus ranged from 1 to 2 months. The median onset times of the DILD for alectinib, imatinib, and tofacitinib ranged from 2 to 3 months. The median onset times of the DILD for sunitinib and sorafenib ranged from 8 to 9 months. Weibull distributions for these drugs when using the cluster analysis showed that there were 4 clusters. Cluster 1 described a subgroup with early to later onset DILD and early failure type profiles or a random failure type profile. Cluster 2 exhibited early failure type profiles or a random failure type profile with early onset DILD. Cluster 3 exhibited a random failure type profile or wear out failure type profiles with later onset DILD. Cluster 4 exhibited an early failure type profile or a random failure type profile with the latest onset DILD.