key: cord-0745920-lzhoj1tt
authors: Viteri-Nöel, Adrian; Martínez-Lacalzada, Miguel; Fabregate, Martin; Manzano, Luis
title: 'Predicting critical illness on initial diagnosis of COVID-19 based on easily-obtained clinical variables' – Author’s reply
date: 2022-02-10
journal: Clin Microbiol Infect
DOI: 10.1016/j.cmi.2022.02.007
sha: af35726bfe5322ac85d78044c2349d033275dd83
doc_id: 745920
cord_uid: lzhoj1tt

nan

Adrian Viteri-Nöel [1, 2] , Miguel Martínez-Lacalzada [ In our study we developed and validated a prediction model called PRIORITY to estimate the risk of critical illness in patients with COVID-19. However, unlike most of the available COVID-19 risk models, our approach was focused on developing a tool that could also be used in resource-limited or out-of-hospital settings. Thus, we only considered simple clinical features available on initial assessment, excluding any other additional test (e.g. imaging or J o u r n a l P r e -p r o o f laboratory). When developing our prediction model, we built on previous literature and expert opinion to select the predictors, rather than data-driven, so as to avoid biases that have affected previous studies [2] . Moreover, we assumed a number of design requirements (e.g.

inclusion/exclusion, merging, and number of predictors) that may limit the discriminative ability of the model, but met our ultimate purpose of maximizing the applicability of the model in the scenarios for which it was developed.

In our model age has a relevant impact on risk prediction and it is included in the subset of the nine best predictors of critical illness. However, here it is worth noting that we used fractional polynomials to better model the non-linear relationships between the outcome and continuous predictors (e.g. age and blood pressure). Thus, the interpretation of their odds ratios (ORs) should be carefully considered [3] , as the risk does not increase linearly across the whole range of the predictor. Therefore, we advise against interpreting the OR for the age-related term as 14.339 for every 10-year increment, since this OR refers to a 1-unit increment in the quadratic term (Age/100) 2 . We acknowledge that the interpretation of ORs for non-linear terms can be challenging, so we included in the manuscript the (a) footnote at For example, the OR for a 10-year increment in age from 40 to 50 years would be around 1.3, while for the same increment between 80 and 90 years the OR would rise up to 1.6.

Regarding the possibility of considering therapeutic limitation as a covariate in the predictive model, several points should be addressed. First, as previously mentioned, our model is mostly intended to be applied in a number of settings where it would be difficult to assess at presentation whether the patient would be a candidate for a complete therapeutic escalation (mainly ICU and mechanical ventilation). Likewise, therapeutic limitation criteria could vary across different pandemic scenarios and healthcare systems. Furthermore, increasing risk due to therapeutic limitation may be especially relevant in elderly patients. As discussed in J o u r n a l P r e -p r o o f detail above, we modeled the relationship between age and the outcome as a quadratic term to take into account a steep increase in risk for advanced age. Finally, the overall effect of therapeutic limitation on risk may be partially accounted for by predictors already included in the model, such as age, comorbidities or dependency.

National Early Warning Score 2 (NEWS2) is a standardized and widespread tool used to help prioritize severity, especially in emergency departments. NEWS2 includes as predictors easily obtainable variables, some of them also included in the PRIORITY model. As interestingly noted by Moretto et al., several studies have assessed NEWS2 for predicting critical outcomes in COVID-19 hospitalized patients. Even though promising, their findings should be considered in the light of some limitations, mainly due to small sample sizes, which may lead to lower performance on different cohorts [2] . Of note, Knight et al. [4] compared in a large external validation cohort the 4C Mortality Score against NEWS, reporting a C-statistic of 0.774 for 4C Mortality Score vs. 0.654 for NEWS, in predicting inhospital mortality. Additionally, Maves et al. [5] showed that the simple addition of age to NEWS appeared to improve the score's ability to predict progression to invasive ventilation or death among hospitalized COVID-19 patients. In summary, we agree that the comparison of PRIORITY and NEWS2 could be interesting due to their similarities, the prognostic value of predictors not included in NEWS2, such as age or previous comorbidities, should be considered.

Finally, as extensively discussed elsewhere, even though including imaging or laboratory tests may certainly improve the discriminative capacity of the model, it was beyond our objectives as it would limit its applicability. Notwithstanding, we thank you for the suggestion regarding the added prognostic value of CT-scan as an interesting approach for further research.

J o u r n a l P r e -p r o o f

Predicting critical illness on initial diagnosis of COVID-19 based on easily obtained clinical variables: development and validation of the PRIORITY model

Prediction models for diagnosis and prognosis of COVID-19 infection: systematic review and critical appraisal

The use of fractional polynomials to model continuous risk variables in epidemiology

Risk stratification of patients admitted to hospital with covid-19 using the ISARIC WHO Clinical Characterisation Protocol: development and validation of the 4C Mortality Score

Predictive Value of an Age-Based Modification of the National Early Warning System in Hospitalized Patients With COVID-19