key: cord-0254045-gzv2tadx
authors: Figueiredo, F. A.; Ramos, L. E. F.; Silva, R. T.; Pires, M. C.; Ponce, D.; Carvalho, R. L. R. d.; Schwarzbold, A. V.; Maurilio, A. d. O.; Alves Scotton, A. L. B.; Garbini, A. F.; Farace, B. L.; Garcia, B. M.; Silva, C. T. C. A.; Cimini, C. C. R. C.; de Carvalho, C. A.; Dias, C. d. S.; Silveira, D. V.; Manenti, E. R. F.; Cenci, E. P. d. A.; Anschau, F.; Aranha, F. G.; de Aguiar, F. C.; Bartolazzi, F.; Vietta, G. G.; Nascimento, G. F.; Noal, H. C.; Duani, H.; Vianna, H. R.; Guimaraes, H. C.; de Alvarenga, J. C.; Chatkin, J. M.; de Moraes, J. P. D.; Rugolo, J. M.; Ruschel, K. B.; Prado Martins, K
title: Development and validation of the MMCD score to predict kidney replacement therapy in COVID-19 patients
date: 2022-01-13
journal: nan
DOI: 10.1101/2022.01.11.22268631
sha: eb5ce70632df4ea29c64e89b29f37295c0e3c95e
doc_id: 254045
cord_uid: gzv2tadx

Background: Acute kidney injury (AKI) is frequently associated with COVID-19 and the need for kidney replacement therapy (KRT) is considered an indicator of disease severity. This study aimed to develop a prognostic score for predicting the need for KRT in hospitalized COVID-19 patients. Methods: This study is part of the multicentre cohort, the Brazilian COVID-19 Registry. A total of 5,212 adult COVID-19 patients were included between March/2020 and September/2020. We evaluated four categories of predictor variables: (1) demographic data; (2) comorbidities and conditions at admission; (3) laboratory exams within 24 h; and (4) the need for mechanical ventilation at any time during hospitalization. Variable selection was performed using generalized additive models (GAM) and least absolute shrinkage and selection operator (LASSO) regression was used for score derivation. The accuracy was assessed using the area under the receiver operating characteristic curve (AUCROC). Risk groups were proposed based on predicted probabilities: non-high (up to 14.9%), high (15.0 to 49.9%), and very high risk ([≥] 50.0%). Results: The median age of the model-derivation cohort was 59 (IQR 47-70) years, 54.5% were men, 34.3% required ICU admission, 20.9% evolved with AKI, 9.3% required KRT, and 15.1% died during hospitalization. The validation cohort had similar age, sex, ICU admission, AKI, required KRT distribution and in-hospital mortality. Thirty-two variables were tested and four important predictors of the need for KRT during hospitalization were identified using GAM: need for mechanical ventilation, male gender, higher creatinine at admission, and diabetes. The MMCD score had excellent discrimination in derivation (AUROC = 0.929; 95% CI 0.918-0.939) and validation (AUROC = 0.927; 95% CI 0.911-0.941) cohorts an good overall performance in both cohorts (Brier score: 0.057 and 0.056, respectively). The score is implemented in a freely available online risk calculator (https://www.mmcdscore.com/). Conclusion: The use of the MMCD score to predict the need for KRT may assist healthcare workers in identifying hospitalized COVID-19 patients who may require more intensive monitoring, and can be useful for resource allocation.

The use of the MMCD score to predict the need for KRT may assist healthcare workers in identifying hospitalized COVID-19 patients who may require more intensive monitoring, and can be useful for resource allocation.

Keywords: Acute kidney injury; COVID-19; kidney replacement therapy; score; risk factors; risk prediction.

Coronavirus disease 19 course is mild in most cases, but it can be severe and critical, with multiple organ dysfunction, septic shock and death [1] . Kidney disease among patients with COVID-19 can manifest as acute kidney injury (AKI), hematuria, or proteinuria, and it has been associated with an increased risk of mortality [2] .

The incidence of AKI among hospitalized patients with COVID-19 has shown to be variable, depending upon the severity of the disease and whether they are outpatient, in the ward or intensive care unit (ICU) environment. A recent systematic review, which included 30 studies and 18,043 patients with COVID-19, observed an overall incidence of AKI of 9.2% (95% confidence interval [CI] 4.6-13.9%), and 32.6% (95% CI 8.5-56.6%) in the ICU [3] .

All rights reserved. No reuse allowed without permission.

perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint Another systematic review from the beginning of the pandemic included 79 studies and 49,692 patients, and observed a significant variation in the incidence of AKI and kidney replacement therapy (KRT) and the risk of death in patients who develop AKI depending on the continent. The incidence of AKI, KRT requirement and death in patients with AKI was 4.3%, 1.4% and 33.3% in Asia, 11.6%, 5.7% and 29.4% in Europe and 22.6%, 4.0% and 7.4%

in North America, respectively [4] . There is a lack of studies from large cohorts in Latin America, which was severely hit by the pandemic.

Previous studies have explored the factors associated with AKI development in COVID-19 patients, including advanced age; black race;

underlying medical conditions such as diabetes mellitus, cardiovascular disease, chronic kidney disease and hypertension; COVID-19 severity; use of vasopressor medications and mechanical ventilation requirement [4, 5] .

However, most studies are limited to univariate analysis or have small sample sizes and there is a lack of studies analyzing independent risk factors for KRT requirement.

A risk score to predict KRT requirement during hospitalization, using clinical and laboratory data upon hospital presentation may be very useful aiming at a better allocation of health resources. However, there is a lack of evidence in this context. Fang et al [6] used a score created before the pandemic (UCSD-Mayo risk score) and analysed its efficiency in predicting hospital-acquired AKI in patients with COVID-19, but the performance of the score in patients in ICUs or under mechanical ventilation was not satisfactory.

All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; Therefore, we aimed to assess the incidence of AKI and KRT requirement in COVID-19 in-hospital patients, as well as to develop and validate a score to predict the risk of the need for KRT. (Table S1 ) and Prediction model Risk Of Bias Assessment Tool (PROBAST) [7, 8] .

All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; 

Data were extracted from the medical records in participant hospitals, including patient demographic information, comorbidities, laboratory results, treatments (including KRT) and outcomes. Data were collected by using a prespecified case report form applying Research Electronic Data Capture (REDCap) tools. Variables used in the risk score were obtained at admission, with the exception of the need for mechanical ventilation, which may have occurred at any time during the hospital stay, except in those patients in which it was initiated after KRT requirement.

The primary endpoint was KRT requirement. Secondary endpoints were the incidence of AKI and mortality in patients who required KRT.

All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint

In the descriptive analyses, categorical variables were described as absolute and relative frequency, and continuous variables by median and quartiles.

The dataset was split into development and validation, according to the date of hospital admission, using July 21, 2020 as the temporal cut (temporal validation).

All analyses were performed using R software version 4.0.2, with the mgcv, finalfit, mice, glmnet, pROC, rms, rmda, and psfmi packages. A p-value<0.05 was considered statistically significant for all analyses and 95% confidence intervals were reported.

Predictors were imputed if they had up to two thirds of complete values. Variables with a higher proportion of missing values than that were not included in the analysis. After analysing missing data patterns, multiple imputation with chained equations (MICE) was used to handle missing values on candidate variables, considering missing at random. Outcomes were not imputed. Predictive mean matching (PMM) method was used for imputation of continuous predictors and polytomous regression for categorical variables.

The results of ten imputed datasets, each with ten iterations, were then combined, following Rubin's rules [9] .

Predictor selection was based on clinical reasoning and literature review before modeling, as recommended [8] . The development cohort included patients admitted before July 21, 2020. All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint Variable selection was performed using generalized additive models (GAM), evaluating the relationships between KRT requirement and continuous (through penalized thin plate splines) and categorical (as linear components) predictors and calculating D1-(multivariate Wald test) and D2statistic (pools test statistics from the repeated analyses).

As our aim was to develop a score for easy application at bedside, continuous variables were categorized on cut-off points, based on evidence from an established score for sepsis [8, 10] .

Subsequently, least absolute shrinkage and selection operator (LASSO) logistic regression was used to derive the score by scaling the (L1 penalized) shrunk coefficients (Table S2 ). Ten-fold cross-validation methods based on mean squared error criterion were used to choose the penalty parameter λ in LASSO.

Lastly, risk groups were proposed based on predicted probabilities: non-high (up to 14.9%), high (15.0 -49.9%), and very high risk (≥ 50.0%).

The specific risks can be easily assessed using the developed MMCD risk score web-based calculator (https://www.mmcdscore.com), which is freely available to the public.

Patients who were admitted in participant hospitals from July 22, 2020 to September, 2020 were included as the external (temporal) validation cohort.

To assess model calibration, predicted dialysis probabilities were plotted against the observed values. To assess model discrimination, the area All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint under the receiver operating characteristic curve (AUROC) was calculated, with the respective confidence interval (95% CI), obtained through 2000 bootstrap samples. Positive and negative predictive values of the derived risk groups were also calculated. The Brier score was used to assess the overall performance [11] .

The derivation cohort included 3,680 COVID-19 patients admitted to the 35 participating hospitals, from March 1, 2020 to July 21, 2020. Those patients were from 159 cities in Brazil ( Figure 2 perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint Discrimination and model overall performance in derivation and validation cohorts for GAM, LASSO and MMCD score are shown in Table 4 .

Within the derivation cohort, the MMCD risk score showed excellent discrimination (AUROC = 0.929; 95% CI 0.918-0.939) a good overall performance (Brier score: 0.057) (figure 3a). preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted January 13, 2022. One in every five patients evolved with AKI and 9.3% required KRT. Among the analyzed predictors, four variables were related to progression to AKI and KRT requirement, including: the need for mechanical ventilation, sex, creatinine upon hospital presentation and diabetes mellitus.

occurs due to several factors. Direct renal injury by the SARS-CoV-2 to the renal endothelium, tubular epithelium and podocytes has been described, infecting the cell through angiotensin converting enzyme 2 receptors, which a b

All rights reserved. No reuse allowed without permission.

perpetuity.

preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in

The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint are abundantly present in the renal tissue [12] . In addition to this mechanism, there are numerous factors worsening the kidney injury, such as: cytokine storm with the release of several interleukins and cytokines, mainly interleukin-6 (IL-6) [3] ; cardiorenal syndrome, caused by right ventricular dysfunction secondary to pulmonary infection; hypercoagulable state with a coronary lesion, left ventricular dysfunction and consequent low renal output, worsened by hypovolemia; and release of nephrotoxic substances such as creatine phosphokinase secondary to rhabdomyolysis [2] .

The need for mechanical ventilation at any time during hospitalization was an important predictor of progression to AKI and the need for KRT, being the variable with the highest points in the risk score. Scoring mechanical ventilation only changed patients' category to "high risk" for evolving to AKI and KRT requirement. Another study from the USA (United States of America) with 5,449 patients, the need for mechanical ventilation was a significant risk factor for AKI and KRT requirement [5] . This finding confirms findings from studies carried out in other countries, such as a study in the USA from the beginning of the pandemic, in which mechanical ventilation was the strongest predictor for AKI (OR 10.7 [95% CI 6.81-16.70]) [5] . In another study from the United Kingdom, which included ICU patients, mechanical ventilation was also the strongest predictor, associated with a four-fold risk of progression of AKI in COVID-19 patients [13] . Study in patients hospitalized with COVID-19 in the city of Wuhan, China, with a significant difference in the need for mechanical ventilation between patients with and without AKI (54.4% vs. 13.7%) [14] .

Finally, a study developed in Brazil with patients hospitalized in a singlecenter with COVID-19 that described the need for mechanical ventilation All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint among patients without AKI, with AKI and with AKI associated with KRT needed were 17.0%, 69,3% and 100% [15] . There is a close relationship between alveolar and tubular damage (lung-kidney axis) in acute respiratory distress syndrome (ARDS), often progressing to different degrees of AKI [16] .

The relationship between mechanical ventilation (MV) and AKI has been widely recognized before the COVID-19 pandemic. Animal models suggest a causal relationship between MV and AKI through a reduction in renal blood flow, apparently by an impairment in intrarenal microcirculation due to hypoxemia and hypercapnia, as well as a drop in cardiac output due to changes in intrathoracic pressure with MV [17]. Additionally, the extrinsic positive end expiratory pressure (PEEP) seems to be associated with a redistribution of intra-renal blood flow, and "biotrauma" may be another cause.

This is a complex and not fully understood mechanism, in which inflammatory mediators are released by ventilated lungs into the systemic circulation [18] . Therefore, AKI in patients who require mechanical ventilation seems to be multifactorial, and it is difficult to define the specific role that each mechanism plays in the pathogenesis of AKI. They are usually observed simultaneously in critically ill patients, which limits the possibility to develop preventive strategies [19] .

In studies published by Chan L et al (n=3,993) [20] and Fisher M et al (n=3,345) [21] with hospitalized patients with COVID-19 in the USA, male sex was considered an independent predictor of progression to AKI and KRT requirement. In our study, the male sex was a risk predictor variable for the evolution of AKI and the need for KRT, being included in the risk score. Male sex has been previously observed to be associated with other adverse All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint outcomes in COVID-19 patients. In a recent meta-analysis with over three million COVID-19 cases, the authors observed no difference in the proportion of men and women who developed COVID-19, but men had almost three times the odds of requiring ICU admission (OR = 2.84; 95% CI = 2.06, 3.92) and higher odds of death (OR = 1.39; 95% CI = 1.31, 1.47) compared to women [22] .

Creatinine levels upon hospital presentation may be evidence of previous chronic kidney disease or an early manifestation of AKI caused by COVID-19 infection. Chronic kidney disease is a global health problem and a silent disease [23] . Serum creatinine levels were categorized according to the Sequential Organ Failure Assessment Score (SOFA) [10] to comply with TRIPOD guidelines, which advises not to use a data-driven method, to avoid model overfitting [8] . Our finding is consistent with a recent systematic review and meta-analysis with 22 studies (n=17,391), which observed an increased incidence of AKI in COVID-19 patients hospitalized in the USA who had abnormal baseline serum creatinine levels due to pre-existing chronic kidney disease [24] . Hansrivijit P et al [25] in their meta-analysis described abnormal basal serum creatinine levels as predictors of progression to AKI.

The association between diabetes mellitus and renal dysfunction is well known, in the form of diabetic nephropathy, as a result not only of intrarenal atherosclerosis and arteriosclerosis, but also non inflammatory glomerular damage [26, 27] . Among the predictor variables analyzed in this study, diabetes proved to be a predictor of risk of progression to AKI and KRT requirement in patients hospitalized with COVID-19. A meta-analysis published in 2020 with 26 studies (n=5,497) evaluated the incidence of AKI in patients with COVID-19 and showed that diabetes was a predisposing factor for progression to AKI [25] .

Meta-analyses showed that the association of patients diagnosed with COVID-19 who developed AKI had higher mortality, which was enhanced by the need for KRT [4, 28] . In Brazil, a country severely hit by the pandemic, there is lack of evidence on the association among AKI, need for KRT, mortality and COVID-19. The scarce existing studies are based in small databases. A study published with 200 ICU-patients showed a high incidence of AKI (about 50%) and 17% of patients requiring KRT, with significantly higher mortality in patients with AKI and needing KRT, in contrast to patients without AKI and KRT requirement [15] . In our study, the incidence of AKI and need for KRT in ICU-patients were lower (about 16% and 9%, respectively), although with higher in-hospital death in this group, similarly to finds in this article.

The MMCD model retrieved an AUROC of 0.96, which was classified as a excellent discrimination. An American study (n=2,256) developed prediction models of KRT using machine learning techniques, comparing L1penalized logistic regression (logistic L1), elastic-net logistic regression (logistic EN) and gradient boosted trees (GBT). Logistic L1 had the best accuracy in the validation cohort. However, the discrimination results were inferior than the one observed in the present analysis (0.847 [95% CI, 0.772-0.936]) and the study has several limitations: many risk predictor variables, hindering the applicability of the score and high incidence of missing variables [29] . All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint

Our study used a large patients database to develop a risk score to predict the need for KRT in patients admitted with COVID-19. A major strength of the MMCD score is its simplicity; the use of objective parameters, which may reduce the variability; and easy availability, even in underresourced settings. Then, the MMCD score may help clinicians to make a prompt and reasonable decision to optimize the management of COVID-19 patients with AKI and potentially reduce mortality. Additionally, our article strictly followed the TRIPOD recommendations [8] .

This study has limitations. Indication and timing of initiation of the KRT may differ according to institutional protocols, and we did not collect information on patients who did not perform dialysis due to limited resources.

Still, this has not affected the accuracy of the score. Additionally, as any other score, MMCD may not be directly generalized to populations from other countries. Furthermore, it was not possible to use the KDIGO (Kidney Disease: Improving Global Outcomes) classification for AKI due to the lack of data on previous serum creatinine of patients admitted to participating hospitals. Instead, we used the SOFA score, which has been widely used for ICU patients for years, and it is currently recommended to assess organ dysfunction in patients suspected of sepsis [30] . In patients with abnormal serum creatinine levels, it was not possible to define the causal factor (previous chronic kidney disease vs. COVID-19) due to the lack of data prior to the patient's hospitalization.

All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint

Using predictors available at baseline and within the first hours of the admission, we could objectively predict the probability of KRT of a COVID-19 patient with AKI. With an accurate prediction, it may help to organize resource allocation to patients who are at the highest risk of KRT requirement [29] , in addition to selecting patients who may benefit from renal protection strategies, close assessment and follow-up by a nephrologist [31] .

In conclusion, we developed and validated a clinical prediction score named MMCD, to predict the need for KRT in COVID-19 patients. This score used a few predictors available at baseline and mechanical ventilation anytime during hospital admission, and retrieved a good accuracy. This could be an inexpensive tool to predict the need for KRT objectively and accurately.

Additionally, it may be used to inform clinical decisions and the assignment to the appropriate level of care and treatment for COVID-19 patients with AKI. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for this this version posted January 13, 2022. ; https://doi.org/10.1101/2022.01.11.22268631 doi: medRxiv preprint

Characteristics of and Important Lessons From the Coronavirus Disease 2019 (COVID-19) Outbreak in China

Management of acute kidney injury in patients with COVID-19

Occurrence of acute kidney injury in adult patients hospitalized with COVID-19: A systematic review and meta-analysis

Risk factors and prognosis for COVID-19-induced acute kidney injury: a meta-analysis

Acute kidney injury in patients hospitalized with COVID-19

A validation study of UCSD

Mayo risk score in predicting hospital-acquired acute kidney injury in COVID-19 patients

PROBAST: A Tool to Assess the Risk of Bias and Applicability of Prediction Model Studies

Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): Explanation and Elaboration

Multiple imputation for nonresponse in surveys

Use of the SOFA score to assess the incidence of organ dysfunction/failure in intensive care units

Use of Brier score to assess binary predictions

Renal histopathological analysis of 26 postmortem findings of patients with COVID-19 in China

Acute kidney injury prevalence, progression and long-term outcomes in critically ill patients with COVID-19: a cohort study

Epidemiology and Outcomes of Acute Kidney Injury in COVID-19 Patients with Acute Respiratory Distress Syndrome: A Multicenter Retrospective Study

Renal function and intrarenal hemodynamics in acutely hypoxic and hypercapnic rats

Bench-to-bedside review: Ventilation-induced renal injury through systemic mediator release -just theory or a causal relationship

Acute Kidney Injury in Mechanically Ventilated Patients: The Risk Factor Profile Depends on the Timing of Aki Onset

AKI in Hospitalized Patients with COVID-19

AKI in Hospitalized Patients with and without COVID-19: A Comparison Study

Male sex identified by global COVID-19 meta-analysis as a risk factor for death and ITU admission

Global Prevalence of Chronic Kidney Disease -A Systematic Review and Meta-Analysis

Renal complications in COVID-19: a systematic review and meta-analysis

Incidence of acute kidney injury and its association with mortality in patients with COVID-19: a meta-analysis

Incidence of chronic kidney disease among people with diabetes: a systematic review of observational studies

Acute Kidney Injury in Diabetes Mellitus

High burden of acute kidney injury in COVID-19 pandemic: systematic review and meta-analysis

Development and validation of prediction models for mechanical ventilation, renal replacement therapy, and readmission in COVID-19 patients

Surviving sepsis campaign: international guidelines for management of sepsis and septic shock 2021

We would like to thank the hospitals, which are part of this