Avoid common mistakes on your manuscript.
Dear Sirs,
Myasthenia gravis (MG) is a B-cell-mediated autoimmune disease and the most frequent disorder of the neuromuscular junction. Characteristic symptoms are muscle weakness, dysphagia, diplopia and fatigue that typically increase throughout continuous muscular activity [1, 2].
Diagnosis of MG typically involves clinical evaluation and diagnostic testing such as low-frequency repetitive nerve stimulation, blood tests to detect specific antibodies, and imaging studies of the thymus gland, which is often involved in the development of the disease. Treatment for MG includes the use of medications to improve muscle function and suppress the autoimmune response. Cholinesterase inhibitors and immunosuppressive medications are commonly used to manage MG symptoms. Living with MG can be challenging and within 2 years of the initial diagnosis, 15–20% of MG patients develop a potentially life-threatening myasthenic crisis (MC) [3]. MC has been defined as respiratory failure requiring mechanical ventilation (MV) due to respiratory muscle weakness or bulbar dysfunction. The mortality, even in our times, is around 12% and prolonged MV (defined as more than 15 days) occurs in 44.7% of the MG patients [4]. Analysis of invasive ventilated patients showed that in eligible patients, early tracheostomy can be associated with a shorter mechanical ventilation duration and a shorter stay in the intensive care unit [4, 5]. Weaning and extubation failure occur even in our times frequently [6]. The prediction of prolonged MV is still very challenging, since many factors such as precondition of the patient, treatment and characteristic of the MG play an important role. No prediction model has been developed to date due to the absence of large cohorts of patients with MC requiring MV.
In order to improve the individual treatment of the patients with MC and to estimate the risks for prolonged mechanical ventilation, we developed POLAR, a prediction model and an easy-to-use web-app for physicians.
We used the largest multicenter cohort available [4] that includes 149 features of 195 patients with invasive ventilation, where the exact number of days of mechanical ventilation is known. The features include baseline demographics, clinical features, medication, comorbidities, characteristics of the MG, treatment regimens during MC, complications of the MC, and outcome parameters like death and days of mechanical ventilation as published before [4, 5]. The main outcome parameter of interest in this analysis was a prolonged mechanical ventilation, defined as MV for more than 15 days.
First, we manually selected 11 features from our dataset. These features are complete (i.e., no missing values) and were selected based on expert experience and previous analyses [4, 5]. Next, we used the EFS (Ensemble Feature Selection) package in R for feature selection [7, 8]. EFS is based on multiple feature selection methods and combines their normalized outputs to quantitative ensemble importance, thereby reducing the individual biases of the methods and providing an unbiased biomarker discovery.
We used a logistic regression (LR) model. After training, it can be employed to predict the probability of the binary outcome for new input data. The logistic regression model uses the following equation:
where FM/MGFAII represents patients with first manifestation of MG during MC or patients with a MGFA score of II or less before MC; CPR represents patients requiring cardio-pulmonal resuscitation during MC; home represents patients who were living at home independent of help before MC.
For evaluation of the LR, we used repeated 10-times tenfold cross-validation. For calibration of the model, we employed GUESS from the R package CalibratR [9].
As an evaluation metric, we employed the Receiver Operating Characteristics curve (ROC) and the corresponding Area Under the Curve (AUC) using the pROC [10] package in R. A one-sided Student’s t test was used for statistical analyses.
We made the LR model available as a web-application using Shiny v.1.7. Shiny is a framework for R that facilitates the development of interactive web applications. Shiny leverages a feature known as reactivity to support interactivity in applications. This enables users to modify input values and have the displayed R objects automatically update in response.
EFS identified nine biomarkers (see Fig. 1) above our threshold of 0.4, namely the age of the patient, the prevalence of at least three chronic comorbidities affecting different organ systems (lung, heart, kidney, diabetes (pancreas) and/or a tumor disease [except thymoma]), late-onset MG, a high Myasthenia Gravis Foundation of America (MGFA) class (IVb) before crisis, implying a severe general course of MG, and delirium, pneumonia or cardio-pulmonary resuscitation during MC can be used to predict prolonged mechanical ventilation. A first manifestation through MC, as well as low MGFA class before MC or whether the patients were independent before MC were markers for a reduced likelihood for prolonged mechanical ventilation. Being a smoker before MC or having a sepsis during MC was no biomarkers with relevant impact (Fig. 1) for our prediction model.
The identified features were used to train a logistic regression model, as described earlier and all nine biomarkers with relevant impact according to the feature selection were included (see methods section).
Our prediction model is able to accurately (AUC = 0.78, p < 0.00001) (Fig. 2) predict the likelihood (in percentage) whether a prolonged mechanical ventilation is necessary for more than 15 days in a given MG patient, based on our cohort. The prediction model is available (in English and German) as a R/Shiny web-app at http://POLAR.heiderlab.de.
POLAR is a prediction model based on data of the largest retrospective cohort of ventilated MC patients [4]. The model comprises six biomarkers associated with prolonged ventilation including age, the prevalence of at least three chronic comorbidities, late-onset MG, a high MGFA class before crisis, delirium, pneumonia or cardio-pulmonary resuscitation during MC. Moreover, there were three biomarkers associated with a reduced ventilation including a first manifestation with MC, as well as low MGFA class and high independence before MC. Thus, our prediction model is able to predict the likelihood of prolonged MV (> 15 days), based on our cohort.
Like in other cohorts of MC patients, higher age is a reasonable parameter for prolonged MV or poor outcome and not specific for MG [11,12,13]. In addition, we found that late-onset MG is a risk factor for prolonged MV, probably overlapping with higher age as a factor, since most of our patients belonged to the late-onset group (80%). We found that a higher MGFA score, before MC increased the likelihood of prolonged mechanical ventilation in our cohort [4]. Other groups found similar results when analyzing Myasthenia Gravis Activities of Daily Living (MG-ADL) scales [12]. Lui et al. also confirmed that a higher number of comorbidities were associated with poor outcome which is well in line with our observation [12]. Another biomarker in our model was pneumonia, which was identified as factor associated with prolonged MV before [14], conforming our results. Sepsis during MC was associated with prolonged MV in another cohort [15]. Yet, in our cohort, the effect did not show a relevant impact in the feature selection and, therefore, did not make it into our prediction model. This surprising result could be explained by the fact, that sepsis was the main reason for death in our cohort and 57% of these patients died within 15 days and, therefore, did not belong to the prolonged MV group in our analysis. Most of the complications included in our score, such as delirium and pneumonia and, in some cases, even cardio-pulmonary resuscitation, occurred within the first 5 days of treatment at intensive care unit (ICU). Therefore, it is important to note that the score has a higher reliability between days 5 and 6 of treatment that in the first 48 h at ICU.
In this manuscript, we present for the first time a web-based prediction model to estimate prolonged mechanical ventilation in myasthenic crisis. POLAR can help to estimate prolonged mechanical ventilation and may enable an optimized personalized treatment, including decision making on intensive care unit, of patients with MC. However, our prediction model needs further validation in prospective cohorts and is, at the moment, only a first step towards a reliable prediction model for prolonged mechanical ventilation in myasthenic crisis. Nevertheless, POLAR can be seen as a good starting point for making probabilistic estimates for prolonged mechanical ventilation, which does not only help patients and their families, but also to improve medical care.
Data availability
Anonymized data will be made available upon reasonable request.
References
Gilhus NE, Verschuuren JJ (2015) Myasthenia gravis: subgroup classification and therapeutic strategies. Lancet Neurol 14(10):1023–1036. https://doi.org/10.1016/S1474-4422(15)00145-3
Melzer N, Ruck T, Fuhr P et al (2016) Clinical features, pathogenesis, and treatment of myasthenia gravis: a supplement to the Guidelines of the German Neurological Society. J Neurol 263(8):1473–1494. https://doi.org/10.1007/s00415-016-8045-z
Alshekhlee A, Miles JD, Katirji B, Preston DC, Kaminski HJ (2009) Incidence and mortality rates of myasthenia gravis and myasthenic crisis in US hospitals. Neurology 72(18):1548–1554. https://doi.org/10.1212/WNL.0b013e3181a41211
Neumann B, Angstwurm K, Mergenthaler P et al (2020) Myasthenic crisis demanding mechanical ventilation: a multicenter analysis of 250 cases. Neurology. 94(3):e299–e313. https://doi.org/10.1212/WNL.0000000000008688
Angstwurm K, Vidal A, Stetefeld H et al (2020) Early tracheostomy is associated with shorter ventilation time and duration of ICU stay in patients with myasthenic crisis—a multicenter analysis. J Intensive Care Med. https://doi.org/10.1177/0885066620967646
Neumann B, Angstwurm K, Dohmen C et al (2023) Weaning and extubation failure in myasthenic crisis: a multicenter analysis. J Neurol. https://doi.org/10.1007/s00415-023-12016-2
Neumann U, Riemenschneider M, Sowa J-P et al (2016) Compensation of feature selection biases accompanied with improved predictive performance for binary classification by using a novel ensemble feature selection approach. BioData Mining 9(1):36
Neumann U, Genze N, Heider D (2017) EFS: an ensemble feature selection tool implemented as R-package and web-application. BioData Min 10:21
Schwarz J, Heider D (2019) GUESS: projecting machine learning scores to well-calibrated probability estimates for clinical decision-making. Bioinformatics. 35(14):2458–2465
Robin X, Turck N, Hainard A et al (2011) pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinf 12:77
Seneviratne J, Mandrekar J, Wijdicks EFM et al (2008) Predictors of extubation failure in myasthenic crisis. Arch Neurol. 65(7):929–933. https://doi.org/10.1001/archneur.65.7.929
Liu F, Wang Q, Chen X (2019) Myasthenic crisis treated in a Chinese neurological intensive care unit: clinical features, mortality, outcomes, and predictors of survival. BMC Neurol. 19(1):172. https://doi.org/10.1186/s12883-019-1384-5
Thomas CE, Mayer SA, Gungor Y et al (1997) Myasthenic crisis: clinical features, mortality, complications, and risk factors for prolonged intubation. Neurology 48(5):1253–60. https://doi.org/10.1212/wnl.48.5.1253
Rabinstein AA, Mueller-Kronast N (2005) Risk of extubation failure in patients with myasthenic crisis. Neurocrit Care 3(3):213–5. https://doi.org/10.1385/NCC:3:3:213
Sivadasan A, Alexander M, Aaron S et al (2019) Comorbidities and long-term outcomes in a cohort with myasthenic crisis: experiences from a tertiary care center. Ann Indian Acad Neurol. 22(4):464–471. https://doi.org/10.4103/aian.AIAN_197_19
Acknowledgements
Collaborators of the German Myasthenic Crisis Study Group: Kornelius Fuchs MD, Amelie Vidal MD and Lee De-Hyung (Department of Neurology, University Medical Center Regensburg), Siegfried Kohler MD, Philipp Mergenthaler MD, Izabela Brachaczek, Jana Maidhof and Arno Wenke (Department of Neurology, Charité – Universitätsmedizin Berlin), Christian Dohmen MD (Department of Neurology, LRV Klinik Bonn), Silvia Schönenberger MD (Department of Neurology, Heidelberg University Hospital), Stefan Gerner MD, Hagen Huttner MD and Manuel Hagen MD (Department of Neurology, University Hospital Erlangen), Hauke Schneider MD, Heinz Reichmann MD, Jan Rahmig MD and Eik Schimmel MD (Department of Neurology, University Hospital, Technische Universität Dresden), Wolf Niesen MD and Hannah Fuhrerer MD (Department of Neurology, University of Freiburg), Christiane Schneider-Gold MD, Ingo Kleiter MD and Christine Fahrendorf MD (Department of Neurology, St. Josef‐Hospital, Ruhr‐University Bochum), Andrea Thieme MD and Andreas Steinbrecher MD (Department of Neurology, HELIOS Klinikum Erfurt), Juliane Dunkel MD and Christian Roth MD (Department of Neurology, DRK-Kliniken Nordhessen ind Kassel), Anke Alberty MD (Department of Neurology, Kliniken Maria Hilf GmbH Moenchengladbach), Jan Zinke MD (Hans Berger Department of Neurology, Jena University Hospital) for help with data collection.
Kornelius Fuchs, Amelie Vidal, Lee De-Hyung, Siegfried Kohler, Philipp Mergenthaler, Izabela Brachaczek, Jana Maidhof, Arno Wenke, Christian Dohmen, Silvia Schönenberger, Stefan Gerner, Hagen Huttner, Manuel Hagen, Hauke Schneider, Heinz Reichmann, Jan Rahmig, Eik Schimmel, Wolf Niesen, Hannah Fuhrerer, Christiane Schneider-Gold, Ingo Kleiter, Christine Fahrendorf, Andrea Thieme, Andreas Steinbrecher, Juliane Dunkel, Christian Roth, Anke Alberty, Jan Zinke, Bernhard Neumann
Funding
Open Access funding enabled and organized by Projekt DEAL. This research did not receive further specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
Author information
Authors and Affiliations
Consortia
Contributions
HS, MA, DH and BN: prepared this manuscript. BN and DH: made the concept and design of this study. HS and BN: organized this multicenter retrospective register. All the authors contributed substantially to analyzing and interpretation of data. BN, MA, and DH: drafted the manuscript, which then was critically revised by all the authors. All the authors have read and approved the final manuscript. Statistical analysis: conducted by MA and DH.
Corresponding author
Ethics declarations
Conflict of interest
Dominik Heider reports no disclosures; Henning R Stetefeld reports no disclosures; Andreas Meisel reports financial research support (paid to his institution) and personal fees from Alexion, AstraZeneca Rare Disease, financial research support (paid to his institution) and personal fees from argenx, personal fees from Grifols, personal fees from Hormosan, personal fees from Janssen, personal fees from Merck, personal fees from Novartis, financial research support (paid to his institution) from Octapharma, personal fees from UCB, outside the submitted work. He is chairman of the medical advisory board of the German Myasthenia Gravis Society; Julian Bösel reports personal fees from Medtronic, personal fees from Neuroptics, personal fees from TelaDoc, personal fees fomr PCORI and the DGNI outside the submitted work; Marie Artho reports no dicslosures; Ralf Linker reports no disclosures; Klemens Angstwurm reports travel support (until 2014) from Alexion, Bayer, BiogenIdec, MerckSerono, Novartis, Teva and honoria from Biogen und TEVA; moreover, he has been a principal investigator in studies in neuroimmunology sponsored by Alexion, Bayer, Biogen, Chugay, MerckSerono, Novartis, and Roche; all not related to the present study; Bernhard Neumann reports no disclosures. The authors declare no financial or other conflicts of interest related to the manuscript.
Ethical approval
All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Local ethic committees and institutional review boards of the participating centers approved the study based on the central vote of the ethics committee at University of Regensburg (No: 15-101-0259). Due to the retrospective character of the study, patient’s consent was not necessary according to the decisions of the ethic committees and institutional review boards.
Additional information
On behalf of the Initiative of German NeuroIntensive Trial Engagement (IGNITE) and with support from the German Neurocritical Care Society (DGNI).
Collaborators of the German Myasthenic Crisis Study Group was listed in the acknowledgements.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Heider, D., Stetefeld, H., Meisel, A. et al. POLAR: prediction of prolonged mechanical ventilation in patients with myasthenic crisis. J Neurol 271, 2875–2879 (2024). https://doi.org/10.1007/s00415-024-12208-4
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00415-024-12208-4