Fluid, pH and electrolyte imbalance associated with COVID-19 mortality

The threat of COVID-19 has harried the world since early 2020. Risk of death from the infection is associated with age and pre-existing comorbidities such as diabetes, dementia, cancer, and impairment of immunological, hepatic or renal function. It still remains incompletely understood why some patients survive the disease, while others perish. Our univariate and multivariate analyses of real world data from U.S. electronic health records indicate that a priori diagnoses of �uid, pH and electrolyte imbalance are highly and independently associated with COVID-19 mortality. We propose that pre-existing homeostatic aberrations are magni�ed upon the loss of ACE2, which is a core component of the electrolyte management system as well as the entry point of internalizing SARS-CoV-2 viruses. Moreover, we also suggest such fragility of electrolyte homeostasis may increase the risk of plasma volume disturbances during the infection. Future interventional studies should investigate whether the risk of death can be alleviated by personalized management of the �uid and electrolyte balance of at-risk individuals before and during COVID-19.


Introduction
In 2020, the world ground to a halt owing to the COVID-19 pandemic that still continues on its course.Achieving durable universal sterilizing immunity, may that be through widespread primary infections or population-penetrating and transmission-halting vaccinations, remains a remote and possibly impermanent prospect.Instead, it seems a chilling possibility that a large proportion of the remaining worldwide COVID-19-naive population will contract the virus in the next few months.We anticipate that even slight improvements in fortifying the health of risk group individuals in anticipation of a potential infection, as well as improving the outcomes of patients succumbing to the severe form of COVID-19, may translate to vast savings in loss of life.
The novel SARS-CoV-2 virus that causes the COVID-19 infection enters human cells via the ACE2 receptor.The infection rst occurs in upper airways and at later stages may proceed to the lung, gastrointestinal tract, kidney, heart, or brain (Gupta et al. 2020).ACE2 and its homolog ACE are core enzymes of the reninangiotensin-aldosterone system (RAAS), which regulates electrolyte homeostasis, blood pressure and cardiovascular health (Patel et al. 2016), as well as restores balance upon volume disturbance of extracellular uid (Fountain and Lappin 2020;Navar 2014).The antagonistic effects of ACE and ACE2 are largely achieved by increasing or decreasing, respectively, the amount of circulating Angiotensin II and consequently, aldosterone, a hormone that is released from the adrenal cortex and regulates uid and electrolyte excretion or retainment by the kidneys.
The underlying mechanisms of infection and viral spread are not fully understood, and despite the advances in prevention and treatment of severe COVID-19, an unmet need remains to better understand the clinical course and the risk factors of severe disease and death.Here, we present an agnostic and data-driven analysis of real world data from U.S. electronic health records of 122,250 COVID-19 patients to identify factors that predict death during a COVID-19 infection.Our unbiased analyses reveal a group of novel, interconnected risk factors for COVID-19 mortality, in addition to con rming many that are already known.We present a working hypothesis of a link between these new risk factors and the reninangiotensin-aldosterone system, and suggest opportunities for better care and/or prevention of severe COVID-19 disease.

Study design and participants
We extracted COVID-19 cases collected by Optum® with a diagnosis date between 20 February and 1 July 2020.The Optum® de-identi ed COVID-19 EHR dataset contains patient-level medical and administrative records from hospitals, emergency departments, outpatient centers, and laboratories from across the United States.Data de-identi cation is performed in compliance with the HIPAA Expert Method and managed according to Optum® customer data use agreements.The COVID-19 EHR dataset sources clinical information from hospital networks that provide data meeting Optum's internal data quality criteria.We con rmed COVID-19 diagnosis either by documented ICD-10 codes (Supplementary Methods) or via positive PCR test result (Supplementary Table S1).Survival time was computed as the number of days between the date of COVID-19 diagnosis and last documented clinical activity (vitals, labs, medication, encounter, collected until 13 July 2020) or documented death.
We analyzed variables that were observed at least a month before the infection.We selected such a conservative buffer time period, since some patients might have had an undiagnosed COVID-19 infection for days before an opportunity to get tested.We also wanted to exclude any physiological changes potentially incurred by the virus during the incubation period, which may be up to 14 days (Lauer et al. 2020).We used the median value captured between 1 and 12 months before the initial COVID-19 diagnosis for vitals and laboratory measurements, and the entire past medical history (median length ~4 years) for variables with a long-term effect, such as diagnoses of chronic indications (Supplementary Methods).Prior inoculations were handled analogously.
We investigated all disease entities that are a part of the Charlson comorbidity index (Quan et al. 2011), the AHRQ (Agency for Healthcare Research and Quality 2018), or the former Elixhauser de nition (Elixhauser et al. 1998).A detailed description of assignment of ICD codes to disease entities can be found in Supplementary Table S2.
After quality control (Supplementary Methods), 249 variables were available for primary univariate analysis.For multivariate analysis, patients with an exaggerated proportion of missing data (Supplementary Methods) were removed, leaving 55,757 patients for analysis.Variables available for less than 10,000 patients were mean-imputed, resulting in a remaining missingness rate of 12.4% in total.The remaining missing values were imputed using the missForest R-package (Stekhoven and Buhlmann 2012).

Association analysis and model development
We originally set out to identify prognostic biomarkers that could identify patients at risk of COVID-19 mortality already before the onset of the disease.As primary analysis, we performed time-to-event analysis using Cox regression (Enderlein 1987).Univariate analysis was conducted using age, sex, ethnicity, race, insurance status, and US region/division as covariate parameters for adjustment.We applied a Bonferroni-correction with the number of variables (m=249) to account for multiple testing and required a signi cance level of =0.05/m=2*10 - .The univariate associations were calculated for the entire patient cohort, as well as separately for the age groups <50, 50-70, 70-80 and >80 years (Supplementary Table S3).
In order to allow comparison of hazard ratios (HRs) between different variables, we report the 2-standarddeviations hazard ratio "HR 2SD ".It is computed as HR 2SD =HR {2*SD} , where SD is the standard deviation of the respective variable.
As secondary analysis, multivariable modelling was performed.We pursued two approaches in parallel.First, we performed a backward selection procedure on the Cox regression model of all eligible variables.We iteratively removed the variable with least impact on model performance until all remaining parameters were signi cant at α 1 =0.05/249=2*10 -4 (Bonferroni-correction).By construction, the procedure controls the family-wise error rate at α=0.05.In parallel, we derived a regularized Lasso model (Simon et al. 2011).We tted a L1 (Lasso) regularized Cox-Proportional Hazards Model using glmnet version 3.02 (Simon et al. 2011), with the concordance index (C-index) (Steck, Krishnapuram, and Dehing-Oberije 2008) as the performance measure.The regularization parameter λ was optimized using ten-fold cross-validation.We selected λ such that we extracted the most regularized model with a C-index within one standard error of the best performing model.More details on model assumption checking, calibrations and performance measures can be found in Supplementary Methods.

Results
Real World Data source Our real world data set covers the de-identi ed electronic health records of 122,250 adult COVID-19 patients collected by Optum® from the U.S. healthcare providers and diagnosed between February 20 and July 1, 2020.The overall mortality in this data set is 5.5%, which is well in line with the Case-Fatality Ratio estimate of 5.89% for USA (Hoffmann and Wolf 2020).Full details on the data set are provided in the Methods and Supplementary Methods sections.

Univariate analysis of predisposition
We set out to identify prognostic biomarkers that could identify patients at risk of COVID-19 mortality already before the onset of the disease.We rst pursued this by a univariate analysis of 249 clinical variables that were observed at least a month before the infection.We selected a conservative buffer time period of one month, since we expected that some patients might have an undiagnosed COVID-19 infection for days before an opportunity to get tested.We also wanted to exclude the effect of any early physiological changes incurred by the virus during the incubation period, which may be up to 14 days (Lauer et al. 2020).We used the median value captured between 1 and 12 months before the initial COVID-19 diagnosis for vitals and laboratory measurements, and the entire past medical history (median length ~4 years) for variables with a long-term effect, such as diagnoses of chronic indications.The univariate associations were calculated for the entire data set, as well as for the age groups <50, 50-70, 70-80 and >80 years separately (Supplementary Table S3).
The univariate analysis revealed 127 variables signi cantly associated (P<0.0002) with mortality (Supplementary Table S3).As expected, age was the strongest prognostic factor, with a per-year hazardratio (HR) of 1.08 [1.077;1.084],and a two-standard-deviation hazard-ratio HR 2SD of 17.8 [16.3;19.5].The HR 2SD measure can re ect the risk increase between, for instance, patients of age 50 versus 90 (=2SD difference in age).Complementary to previous reports on biomarkers measured during COVID-19, we observed a priori decreases of albumin, hemoglobin, calcium, HDL cholesterol, lymphocyte proportion and lymphocyte to leukocyte ratio more frequently in deceased patients.Moreover, non-survivors had a higher likelihood of a history of elevated blood urea nitrogen, creatinine, blood glucose, respiratory rate, red cell distribution width, neutrophil percentage, neutrophil-to-lymphocyte ratio, and total white blood cell count.
An unexpected observation from the univariate analysis was that low, rather than high, measurements of diastolic blood pressure were associated with death (Supplementary Table S3, HR 2SD =0.711[0.655;0.771],P=1.43E-16; Figure 1A).This seemed to be in sharp contrast to earlier reports suggesting that hypertension was an important risk factor for COVID-19 (Zuin et al. 2020).Curiously, a priori diagnoses of both hypotension (ICD-10 I95.1, I95.9) and hypertension (I.10*, O.100.*,O.109.*) were more common among the deceased patients (HR=1.18[1.14;1.22],P=8.1.29], P=1.33E-09, respectively).We envisaged that these counterintuitive results might be associated with the natural decrease of DBP with age, or hypotension being a lagging comorbidity of heart failure, together with age and heart failure being risk factors for severe COVID-19 (Williamson et al. 2020).However, after excluding all patients with a prior diagnosis of heart failure and performing an age-group-speci c analysis, we still observed an association of low DBP and mortality (Figure 1B).Speci cally, patients with a hypertension diagnosis (ICD-10 I.10*) and older than 40 years only showed consistently higher mortality rates, if they had also experienced hypotensive levels of DBP (<60 mmHg) in the past year.In fact, even hypertensive median measurements of DBP (>90 mmHg) were associated with lower mortality than hypotensive ones in the age group 70-79 years (Figure 1C).Indeed, the large epidemiological OpenSAFELY study (Williamson et al. 2020) has also previously reported that the fully adjusted variable of hypertension/elevated blood pressure has a negative rather than positive correlation with COVID-19 mortality, speci cally in patients that are 70 years or older.
Another factor that could contribute to the hypotension observation is a decrease in serum albumin levels (Høstmark, Tomten, and Berg 2005).Decreased albumin levels were strongly associated with COVID-19 mortality in our data set, having the strongest effect for any laboratory measurement in the univariate analysis (HR 2SD =0.471 [0.442;0.503],P=5.10E-110).We also investigated the association of a priori measurements of hypotensive DBP and median albumin levels, and observed that patients with hypotensive DBP typically had low albumin (OR 2SD =0.18, P=7.01E-14).We conclude that many COVID-19 victims in our data set have a history of co-occurring hypotension and hypoalbuminemia.

Multivariate analysis ndings
To further dissect the predisposition landscape of COVID-19 mortality, we pursued a multivariate time-toevent analysis (Cox regression) of laboratory, vital, comorbidity, immunisation and demographic variables in 55,757 patients, for whom at least ten variables were available (see Supplementary Methods).We performed three multivariate analyses in parallel: one for comorbidities, one for laboratory measurements and vitals, and a combined main model for all variable groups as well as inoculations.Unless otherwise stated, we are discussing the combined multivariate model below.
The combined multivariate Cox model reaches a C-index of 0.853 (SE 0.003), a high level of prognostic power for survival (Supplementary Table S4).Both alternative models also showed good performances, with C-indices 0.841 (0.003) for the comorbidity model, and 0.851 (0.003) for the labs/vitals model.The combined model reveals a plausible combination of factors that are well aligned with previous studies (Williamson et al. 2020) and associated with age, male sex, renal impairment, diabetes, hypoxia, hematological challenge, dementia and cancer (Table 1).As repeatedly reported, the strongest predictors of mortality are age (HR 2SD =6.95 [6.11;7.91])and male sex.We also con rmed independent effects of African ethnicity and insurance status, as previously observed and discussed (Yehia et al. 2020).Furthermore, while increases in red blood cell distribution width (RDW) during COVID-19 had already previously been reported to be associated with mortality (Foy et al. 2020), our predisposition model shows that RDW aberrations precede the infection by at least a month.
The multivariate model indicates an independent protective effect for patients having received a booster vaccination for diphtheria, tetanus and pertussis (NDC code 58160084252).Furthermore, multiple other vaccinations were associated with better survival in the univariate analysis (Supplementary Table S3).
The most striking novel nding of the multivariate model is a comorbidity group of diagnoses associated with uid, pH and electrolyte imbalance (FPEI) (Table 1).The overall FPEI comorbidity group has not been previously reported to be associated with COVID-19 mortality to our best knowledge.However, given the fundamental role of ACE2 in maintaining electrolyte homeostasis, such a dependency seems intuitively credible.
Electrolyte and pH disturbances of any kind might involve impairment of renal function.That said, while the effect of renal comorbidities was also evaluated for the multivariate analysis, it did not show an independent effect in the combined model.An investigation of the co-dependence of different variables revealed that renal comorbidities were excluded from the model by correlated but stronger independent effects of age, albumin, and blood urea nitrogen measurements, as well as FPEI diagnoses and red blood cell distribution width (Supplementary Table S5).For comparison, our comorbidity-only multivariate model showed that the effects of FPEI comorbidities was independent to that of renal comorbidities (Supplementary Table S6).Interestingly, FPEI had the highest odds ratio and the lowest P value in the comorbidity multivariate model (HR=1.37[1.27;1.49],and P=1.71E-14).We conclude that while FPEI disorders in COVID-19 patients may often indicate suboptimal function of the kidneys, they do not always co-occur with a diagnosed renal comorbidity.
A priori levels of DBP did not make it to the multivariate model as an independent variable.However, another dissection of variable co-dependence showed that it was not the history of heart failure that sequestered DBP from the model.Instead, the largest correlating factors that reduced DBP's effects were age, albumin, red cell distribution width and blood urea nitrogen (Supplementary Table S5).Finally, the hypertension comorbidity group did not make it to any of the multivariate models, either.Instead, the most correlated variates that progressed to the model in its stead were age, albumin, blood urea nitrogen, FPEI and Hemoglobin A1c (Supplementary Table S5).

Fluid, pH and electrolyte imbalance (FPEI) co-morbidity group deep dive
To further characterize the FPEI nding, we performed additional analyses involving both the prior medical history and the observations during the COVID-19 infection.First, we inspected the overlap of patient cohorts with prior FPEI and/or renal comorbidities (Table 2, Figure 2).While the overlap was substantial, it was far from complete.Less than half of the FPEI patients in our data set also had a renal comorbidity diagnosis.Interestingly, having a history of both renal and FPEI comorbidities was associated with a very high mortality, 20%.In fact, a quarter of all non-survivors in the entire data set had a combination of FPEI and renal diagnoses.However, an additional 18% of all deceased patients had an FPEI diagnosis but no renal diagnoses prior to the COVID-19 infection.
We also investigated the frequency of disturbances in the levels of electrolytes or pH in the thirty days following the onset of COVID-19.We distinguished incidences of low or high potassium, sodium, chloride, or total CO2 and used thresholds for normal ranges from Healthline ("Healthline Basic Metabolic Panel" 2017).Non-survivors were more likely than survivors to have any of the eight aberrations in the non-FPEIdiagnosed cohort, and more likely to have any aberration bar low potassium and low chloride in the FPEIdiagnosed cohort (Figure 3A-H).Furthermore, high potassium as well as both low and high sodium, chloride, and total CO2 were more frequent in non-survivors with a prior FPEI diagnosis than in nonsurvivors without one (Figure 3A-H).For comparison, the difference in the electrolyte levels of the subcohorts with or without any observed electrolyte imbalance during the 30 days after diagnosis were signi cant in all eight categories for the non-FPEI-diagnosed patients, and in all bar low sodium and high total CO2 for the FPEI-diagnosed patients (Figure 3I-P) To summarize, our analysis shows that both elevated and decreased levels of basic electrolytes as well as pH-balance-associated total CO2 are more frequent in COVID-19 non-survivors than survivors.
Finally, we inspected the frequency of FPEI diagnoses assigned to the patients during the course of the COVID-19 infection.We plotted the proportion of patients having received an FPEI diagnosis on day -7… +30 relative to the original COVID-19 diagnosis date.We observed that the proportion of FPEI diagnosed patients throughout the followup period was signi cantly higher among the non-survivors (Figure 4A).Furthermore, we distinguished FPEI diagnoses to subgroups associated with volume/ uid depletion, uid overload, or aberrations of pH.Each of these subcategories is more often observed in non-survivors (Figure 4B-D).Hence FPEI diagnoses associated with physiological manifestations of uid depletion, overload, or pH balance disturbance alike were more frequent during COVID-19 in non-survivors than survivors.
Our observations prompted us to hypothesize that patients with a priori aberrances of uid, pH and electrolyte management have an elevated dependency on the RAAS, and the sequestration of ACE2 by COVID-19 further undermines their overall homeostatic balance.This is compatible with the prior observations of renal comorbidities as a risk factor for COVID-19, but it is important to note that most FPEI-de cient patients in our data set lack an actual renal diagnosis.To summarize, we propose that FPEI is a risk factor for COVID-19 mortality both in absence of and coupled with pre-existing renal comorbidities.

Discussion
Despite all advances in COVID-19 vaccines and treatments, there is still an urgent need to better understand the course of disease and the risk factors of severe COVID-19.We have analysed a rich real world data set of 122,250 COVID-19 patients and characterized the pre-existing clinical factors that put the patients at risk of death.We foresee that our ndings could generate novel hypotheses for COVID-19 treatment options and preventive medicine.The concordance of COVID-19 mortality prognostic factors implicated by our multivariate analysis and those from existing literature such as the large epidemiological OpenSAFELY study is high.This demonstrates that analysing real world data to dissect COVID-19 predisposition is a powerful approach, and con rms that risk factors in European and American patients have a considerable overlap.Moreover, some variables such as red blood cell distribution width, for which an association during the infection had previously been reported (Foy et al. 2020), were shown by our study to deviate from normal levels at least a month before the diagnosis (Table 2).
The multivariate and univariate models suggest multiple surprising associations.First, a previous inoculation with Diphtheria-Tetanus-Pertussis booster is independently associated with lower risk of mortality in the combined multivariate model.Curiously, herpes zoster vaccinations also show evidence of a protective effect in the univariate analysis, whereas a past diagnosis of herpes zoster is associated with lower mortality in the comorbidity multivariate model (Supplementary Tables S1, S5).Generally, it can be hypothesized that patients with active inoculation schedules have a keen interest in maintaining their health, and the nancial means to pursue this goal.However, compelling computational evidence has been presented that the DTP vaccine harbors a number of T cell epitopes with potential to induce cross-reactive immunity towards SARS-CoV-2 ( Reche 2020).Second, the multivariate model suggests an independent association of low height with death (Table 1).Indeed, prospective observational studies have previously indicated that height has an inverse correlation with mortality from respiratory disease (Smith et al. 2000).Intriguingly, the initial total volume administered to a patient according to the ARDS protocol is a direct derivative of height, with shorter patients receiving a smaller initial volume of supplementary air by default ("NIH NHLBI ARDS Clinical Network Mechanical Ventilation Protocol Summary " 2008).One could also hypothesize that taller patients receive a lower initial viral load when exposed to droplets during the transmission, assuming a vertical position.Further studies will be necessary to identify clinical implications of these associations.
The most prominent novel outcome of our multivariate and univariate analyses is that prior uid, pH and electrolyte imbalance (FPEI) is an independent predisposing factor for COVID-19 mortality.Out of various FPEI axes, those of metabolic acidosis seem strongest, complementing previous observations of diabetic ketoacidosis during COVID-19 (Li et al. 2020).However, all types of prior FPEI aberrations demonstrate an association with death in our univariate analysis and measurements of electrolyte levels during the infection.
Notably, ACE2 that is the entry point of SARS-CoV-2 into human cells is also an antagonist of ACE in the renin-angiotensin-aldosterone system (RAAS), which regulates electrolyte homeostasis, blood pressure and cardiovascular health (Patel et al. 2016), and restores balance upon volume disturbance of extracellular uid (Fountain and Lappin 2020;Navar 2014).A large proportion of active ACE2 normally resides in the lung alveoli, while abundant presence is also observed in the heart, gastrointestinal tract, nasopharyngeal region and vascular endothelium, and limited expression in many other tissues including brain, liver and kidney (Hamming et al. 2004;J. Chen et al. 2020).Such widespread expression of ACE2 may also partly explain the variety of COVID-19 clinical symptoms.
The antagonistic effects of ACE and ACE2 are largely achieved by increasing or decreasing the amount of circulating Angiotensin II, respectively.Ang II is a potent secretagogue of aldosterone, an adrenal cortex hormone that stimulates kidney principal cells and alpha intercalated cells to enhance renal reabsorption of sodium and water, excretion of potassium, and the maintenance of acid-base balance (Atlas 2007;Wagner 2014).The homeostatic regulatory function of Ang II is largely endocrine, but autocrine, paracrine and intracrine effects have also been identi ed (Fountain and Lappin 2020).Notably, the half-life of Ang II in the bloodstream is around 30 seconds (van Kats et al. 1997), so its direct effects are short-lived, whereas aldosterone can function for hours or days.
The internalization of SARS-CoV-2 viruses via ACE2 restricts the receptor bioavailability on the cell surface, whereas the infected cells also actively downregulate ACE2 expression, as well as initiate cleavage of the membrane-bound ACE2 to produce a soluble form that is released into the circulation (Groß et al. 2020).Interestingly, the soluble form of ACE2 has previously been shown to be associated with heart failure independently of COVID-19 (Epelman et al. 2008).Since COVID-19 patients may have a reduced presence of membrane-bound ACE2 in infected tissues and potentially an abundance of soluble ACE2, it can be envisaged that the balancing effect of ACE2 on Ang II will be locally and systemically undermined by high SARS-CoV-2 viral load, evoking uctuations in the levels of aldosterone, and consequently, plasma electrolytes.This molecular ambiguity of the collection of factors that we identify as FPEI has a single common denominator, ACE2 disturbance, with a straightforward and plausible connection with the disease.
We believe the role of all-round electrolyte and pH imbalance in COVID-19 could have gone partially unappreciated until now, owing to a focus on the median laboratory measurements (Lippi, South, and Henry 2020;D. Chen et al. 2020), which will obviously average out some of the effects of opposite extremities.However, recent reports have pointed out that both hypernatremia and hyponatremia are frequently present in COVID-19 patients (Hu et al. 2020).Our ndings implicate that fundamental instability of the renin-angiotensin-aldosterone system (RAAS), manifested as earlier FPEI diagnoses, and potentially as aberrations of blood pressure levels, may sensitize patients to unfavourable outcomes in COVID-19.
The role of ACE2 receptors in SARS-CoV-2 and SARS-CoV infections has been frequently studied.
Curiously, older individuals and Type II diabetics have lower baseline ACE2 levels despite being at higher risk of severe COVID-19 (J.Chen et al. 2020).It is tempting to hypothesize that a lack of ACE2 would manifest as an increase of Angiotensin II and aldosterone, and hence sodium and uid retention, increasing overall uid volume.However, our observations cover a comorbidity space much wider than that of hypernatremia.Rather, both elevated and reduced levels of sodium, potassium, chloride and total CO2 are signi cantly more frequently observed in COVID-19 non-survivors than survivors.The full consequences of the reduction and differential distribution of ACE2 in COVID-19 are yet to be understood.There are physiological feedback mechanisms counterbalancing the effects of RAAS based on the antidiuretic hormone and the natriuretic peptide, which may also become activated upon electrolyte or pH imbalance (Reid et al. 1983;Richards 1996) .A model whereby ACE2 disturbance always introduces an excess of uids is likely a simpli cation.Instead, we propose that various pre-existing, and in some cases mutually exclusive, homeostatic aberrations may be magni ed, when ACE2 and the RAAS are disordered by COVID-19.We also emphasize that an intact homeostatic regulation system is instrumental in preventing both excess and contracted plasma volumes, both of which are known risk factors for intensive care mortality (Perner 2009;Ogbu, Murphy, and Martin 2015).
Another striking nding of our predisposition models is that of hypoalbuminemia, which is also often associated with abnormally low levels of blood pressure and highly correlated with low diastolic blood pressure in our data set.Previous studies have indicated that a large majority of patients that are critically ill with COVID-19, and almost all non-survivors, either present with hypoalbuminemia at the start of the infection or develop it during the course of the disease (W.Huang et al. 2020).Low serum albumin is also a predictor of overall mortality in the elderly (Corti 1994).Patients with low albumin will lose colloidal osmotic (oncotic) pressure, and experience an extracellular uid shift from the intravascular space into the extravascular space (Gounden, Vashisht, and Jialal 2020).Notably, hypoalbuminemia coupled with excess extracellular uid may result in co-occurring oedema and hypovolemia (Bobkova et al. 2016).Such a mechanism has also been suggested to contribute to COVID-19 (J.Huang et al. 2020), and would be particularly disastrous in the lung.Indeed, compelling suggestions have been presented that volume contraction may be aggravating COVID-19 outcomes (Tsolaki et al. 2020).
While previous discussions on the effect of ACE2 in COVID-19 have often focused on its role in regulating vasodilation and dampening in ammation (Groß et al. 2020), our ndings complement this picture by showing that the RAAS-managed balance of uid, pH and electrolytes is more frequently disturbed in COVID-19 victims than survivors already before the onset of the infection, and associated with an increased risk of death independently of age and renal comorbidities.Future observational and interventional studies may show that careful and personalized correction of uid and electrolyte homeostasis early in the course of the infection correlates with better outcomes of COVID-19.

Limitations
Our analysis is based on a real world data resource and thus carries the burden of associated di culties and limitations.In contrast to controlled randomized settings, systematic bias and negative effects by data errors cannot be completely ruled out a priori.On the other hand, the large sample size available can help to overcome issues and to detect phenomena, which are otherwise overlooked.Indeed, the majority of risk factors we identify are consistent with previous ndings in the literature, thereby giving credibility to our novel ndings.Furthermore, results for comorbidities from patient history are consistent with those of biomarkers of the respective diseases, demonstrating internal consistency and plausibility of our data.In summary, while the exact risk effect size might be in uenced by the way the data was collected, it is likely that the qualitative conclusions we provide are correct.

Supplementary Files
This is a list of supplementary les associated with this preprint.Click to download. NahkurietalSupplementaryinformation.docx Declarations S. N. and A. B.-M. are Roche employees.A. B.-M. holds Roche stock options.T. B. and V. S. are contractors paid by Roche.S. M. is not paid by Roche.

Figures
Figures

Table 2 :
Mortality rates of patient cohorts with or without a prior history of FPEI and/or renal comorbidities.