Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
We thank Arnaud and Devilliers1 for their correspondence on our article.2 They raise the important topic of risk prediction models for estimating the risk of COVID-19-related death, and the potential development of a risk prediction tool using data from our article.
We agree with the authors that patients using immunosuppressive/immunomodulatory agents have different risks, depending on factors such as their age, chronic conditions and specific drug therapies. However, we also recognise that there are significant limitations in using the estimates published in our manuscript to develop a COVID-19-related death risk prediction model to guide vaccination strategies. These limitations include:
While robust models that predict the prognosis of COVID-19 are desirable to support decisions about shielding, hospital admission, treatment and population level interventions such as COVID-19 vaccination, this was not the primary aim of the published study. Importantly, it should be noted that owing to the voluntary nature of the registry there is an inherent selection bias, with an over-representation of severe cases, as discussed in our manuscript. Any model developed in a specific dataset will only reflect the risk for a particular patient under similar circumstances and receiving similar care. Therefore, in the same way that the hospitalisation and death rates reported in our article cannot be extrapolated to the entire population of patients with rheumatic diseases, a risk model developed using the reported data may lack generalisability.3
Given the rapid and dynamic evolution of COVID-19, static risk prediction models are likely to rapidly become obsolete. COVID-19 warrants the need to develop ‘living’ risk prediction models which can be updated regularly as our understanding of COVID-19 increases and more data becomes available. For example, the performance and generalisability of any risk prediction model will depend heavily on contextual and environmental time-dependent factors, such as the underlying burden of infection and immunity levels in the population of interest. Temporal trends can be dictated by improved testing capacity, vaccination efforts and increased ability to better treat patients with COVID-19 (eg, pharmacological treatment with glucocorticoids and remdesivir, changes in invasive/non-invasive ventilation strategies, prone positioning and prophylaxis/prevention of complications such as thromboembolic events). They can also arise from typically unmeasured time-dependent factors such as adherence to shielding and other infection control measures (that can affect the likelihood of exposure to SARS-CoV-2) and healthcare resource availability, which has not only varied significantly over time, but also between and within countries and regions at the same point in time.4
A systematic review of published risk prediction models for COVID-19 found that most models are subject to a high risk of bias with optimistic reported performance, raising concern that these models may be unreliable when applied in practice.5 Indeed, development of risk prediction models should follow a robust and standardised approach, as outlined in the Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis statement.6 This recommended standardised approach includes aspects such as assessment of model performance, calibration and internal/external validation. All these steps are essential to be undertaken before a risk prediction model can be used to support population risk stratification in relation to public health interventions such as vaccine utilisation.
All these elements being clarified, and for reasons of data transparency, we provide here the estimate of the intercept of the model ().
We appreciate the comment from Arnaud and Devilliers,1 and agree that risk prediction models have the potential to help patients and doctors reach a shared understanding of risk, and help stratify risk in populations for public health purposes. However, developing a risk prediction model for risk of COVID-19-related death in patients with rheumatic diseases should constitute a separate effort, undertaken with a larger sample size, and taking the above considerations into account.
Handling editor Josef S Smolen
Twitter @pedrommcmachado, @carmona_loreto, @philipcrobinson
Contributors PMM drafted the first version of the manuscript. All authors revised the manuscript and approved the final version.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Disclaimer The views expressed here are those of the authors and participating members of the COVID-19 Global Rheumatology Alliance, and do not necessarily represent the views of the American College of Rheumatology, the European Alliance of Associations for Rheumatology, the (UK) National Health Service, the National Institute for Health Research, or the (UK) Department of Health, or any other organisation.
Competing interests PMM has received consulting/speaker’s fees from AbbVie, BMS, Celgene, Eli Lilly, Janssen, MSD, Novartis, Orphazyme, Pfizer, Roche and UCB, all unrelated to this manuscript, and is supported by the National Institute for Health Research (NIHR), University College London Hospitals, Biomedical Research Centre (BRC). MS has nothing to disclose. AS reports personal fees from lectures for AbbVie, MSD, Roche, BMS and Pfizer, all outside the submitted work. LG reports personal consultant fees from AbbVie, Amgen, BMS, Biogen, Celgene, Gilead, Janssen, Lilly, Novartis, Pfizer, Samsung Bioepis, Sanofi-Aventis, UCB, and grants from Amgen, Lilly, Janssen, Pfizer, Sandoz, Sanofi, Galapagos, all unrelated to this manuscript. MAG reports grants from National Institutes of Health, NIAMS, outside the submitted work. SLT has nothing to disclose. SLT has nothing to disclose. EFM reports that LPCDR received support for specific activities: grants from Abbvie, Novartis, Janssen-Cilag, Lilly Portugal, Sanofi, Grünenthal S.A., MSD, Celgene, Medac, Pharmakern, GAfPA; grants and non-financial support from Pfizer; non-financial support from Grünenthal GmbH, outside the submitted work. LC has not received fees or personal grants from any laboratory, but her institute works by contract for laboratories among other institutions, such as AbbVie Spain, Eisai, Gebro Pharma, Merck Sharp & Dohme España, S.A., Novartis Farmaceutica, Pfizer, Roche Farma, Sanofi Aventis, Astellas Pharma, Actelion Pharmaceuticals España, Grünenthal GmbH, and UCB Pharma. KLH reports she has received non-personal speaker’s fees from AbbVie and grant income from BMS, UCB, and Pfizer, all unrelated to this manuscript, and is supported by the NIHR Manchester BRC. PCR reports personal fees from AbbVie, Eli Lilly, Gilead, Janssen, Novartis, Pfizer, Roche and UCB, non-financial support from BMS, research funding from Janssen, Novartis, Pfizer and UCB, all outside the submitted work.JY reports consulting fees from Astra Zeneca and Eli Lilly, and grants from Pfizer, outside the submitted work.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Patient consent for publication Not required.
Provenance and peer review Commissioned; internally peer reviewed.