In-hospital Mortality of Critically Ill Patients with Interactions of Acute Kidney Injury and Acute Respiratory Failure in the Resource-limited Settings: Results from SEA- AKI Study

Nattachai Srisawat (  drnattachai@yahoo.com ) Chulalongkorn University, King Chulalongkorn Memorial Hospital Win Kulvichit Chulalongkorn University, King Chulalongkorn Memorial Hospital Kittipon Sarnvanichpitak Chulalongkorn University Sadudee Peerapornrattana Chulalongkorn University, King Chulalongkorn Memorial Hospital Somkanya Tungsanga Chulalongkorn University, King Chulalongkorn Memorial Hospital Nuttha Lumlertgul Chulalongkorn University, King Chulalongkorn Memorial Hospital Kearkiat Praditpornsilpa Chulalongkorn University, King Chulalongkorn Memorial Hospital Kriang Tungsanga Chulalongkorn University, King Chulalongkorn Memorial Hospital Somchai Eiam-Ong Chulalongkorn University, King Chulalongkorn Memorial Hospital John A Kellum University of Pittsburgh School of Medicine SEA-AKI study group SEA-AKI study group


Introduction
Acute kidney injury (AKI) and acute respiratory failure (ARF) are common associated complications which independently lead to higher mortality among critically ill patients. [1][2][3][4] When both conditions overlap, the incidence of in-hospital mortality is enhanced.
It has been demonstrated that AKI led to systemic uid congestion, retention of uremic toxins, and metabolic acidosis. A subsequent increase in work of breathing would cause or worsen lung injuries. 5,6 Conversely, ARF could affect gas exchange, and stimulate neurohormonal activities such as renin-angiotensin-aldosterone system, sympathetic activities, and release arginine vasopressin. These would result in salt and water retention, yet renal hypoxia and vasoconstriction. 7,8 The use of mechanical ventilation could also cause right ventricular dysfunction from increased intrathoracic pressure and lead to impaired hemodynamic status, systemic and renal congestion. Moreover, the injury in either renal tubular, or alveolar cells, or biotrauma from mechanical ventilation would contribute to a release of systemic in ammatory mediators causing further injury in the other organ, as in vicious cycle. [9][10][11][12] While the theory of kidney-lung interorgan cross-talk has been postulated for several years, the causal relationship remained unclear. 13 During the pandemic of COVID-19 infection, more studies report strong relationship between AKI and ARF in critically ill patients with COVID-19 infection. 14 However, the clinical outcomes from each patterns of simultaneous kidney and lung injuries has not been clearly demonstrated. Thus, we aimed to explore the interaction between AKI and ARF, and their clinical outcomes among critically ill patients.

Patients and study design
This was asubanalysis ofthe prospective,multicenter,multinational,observational study, Southeast Asia AKI (SEA-AKI) study. 15 In brief, this studywas conducted in 23 ICUs throughout 22 hospitals from various countries across SEA region including Thailand, Laos, and Indonesia. The study sites in this study included all level of hospitals ranging from university hospitals, regional hospitals, and provincial hospitals. All available data from the cohort were retrieved to maximize the power and generalizability of the results.We excluded patients with prior history of end stage renal disease (ESRD) and patients who had missing important classi cation data i.e. daily ventilator status, daily urine output, and daily serum creatinine and hence their AKI-ARF pattern cannot be de ned.The study protocol was reviewed by the Institutional Review Board at each participating site and the need for informed consent was waived. For this study, the Institutional Review Board of Faculty of Medicine, Chulalongkorn University approved the study (IRB 440/63).This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. 16

Study outcomes
Our primary outcome was in-hospital mortalityof each AKI-ARF pattern. Secondary outcomes includedICU length of stay, hospital length of stay, total ventilator day, ventilator free day, renal recovery at discharge,the factors associated with inhospital mortalityand the factors associated withthe development of AKI after ARF and ARF after AKI.

Study de nitions
We de ned AKI as stage 2-3 AKI using the Kidney Disease Improving Global Outcomes (KDIGO) classi cation. 17 Thisde nition of AKI aimedto avoid misclassification of minor changes in renal function or influences of fluid balance. KDIGO stage 2 AKI is de ned as a 2.0-2.9 times baseline increase in serum creatinine (SCr). KDIGO stage 3 AKI is de ned as an increase in SCr to ≥3.0 times baseline or an increase in SCr to ≥4 mg/dL. Stages were determined based on SCr and urine output(UO)data. ARF was de ned based on the need for endotracheal intubation and mechanical ventilatory support for at least 24 hours. 18-20 For patients with known preadmission SCr, the most recent value in the last 12 months prior to hospitalization was used as baseline SCr. For patients without preadmission SCr, baseline SCr was estimated either using the Modi cation of Diet in Renal Disease (MDRD) equation back calculation or the lower SCr from hospital admission, as recommended by the Acute Disease Quality Initiative. 21 Pattern of AKI and ARF of each patient were categorized based on the onset of development of AKI compared to ARF. The "Concurrent AKI-ARF" was de ned by the development of AKI and ARF on the same day (within 24 hours apart). The "AKI rst" was de ned by the development of AKI before ARF (at least 24 hours apart). The "ARF rst" was de ned by the development of ARF before AKI (at least 24 hours apart). The "AKI alone" was de ned by the development of AKI without ARF during ICU admission. The "ARF alone" was de ned by the development of ARF without AKI during ICU admission. The "no AKI/ARF" was de ned by neither AKI nor ARF was observed.Renal recovery at discharge was de ned by the absence of AKI by either SCr or UO criteria.

Statistical analysis
Categorical data were presented as counts and percentages. Continuous data were presented as mean and standard deviation (SD) if normally distributed or median with inter-quartile range(IQR)if non-normally distributed. The Chisquareasymptotictest of independence was used to compare the proportions of different patterns of patients, whereas as Kruskal-Wallis one-way analysis of variance by ranks was used to compare continuousvariables.
Kaplan-Meier survival curves were also generatedstrati ed byAKI and ARFpattern.Log-rank test was used to compare survival of each pattern. Cox proportional hazards model was also done to graphically describe the covariate-adjusted survival at 14 day after ICU admission. The rationale of showing only Cox-adjusted survival curve of 14 days was to preserve constant proportional hazards assumption which was found to be violated beyond this time point.
As this was a multi-center study with centers representing clusters, multivariable logistic mixed effect regression was usedtoidentify risk factors for in-hospital mortality, AKI development after ARF, ARF development after AKI.To determine the interactions of AKI-ARF pattern and considered risk factors on in-hospital mortality, we use AKI-ARF pattern as a main effect and accounting for interactions between AKI-ARF pattern and all other risk factors.
Mixed effect model allowed us to adjust for the site speci c random effect. We investigated both the random intercept and random coe cients model. Choosing the random intercept model as it had a lower Akaike's Information Criteria and therefore a lower propensity to over t to the sample. We also run multicollinearity diagnostics as part of our modeling process, variance in ation factor above 4 was of concern. Factors with p-value below 0.20 in univariable analysis were retained in the multivariable analysis. Regression coe cients of each factor were reported as odds ratios (OR). All analysis was conducted using the R statistical package (V4.0.3, R core team, 2020) and the mixed effect modeling using the R library, Ime4. A P-value of less than 0.05 was considered to be statistically signi cant. To avoid false discovery of multiple pairwise comparison, Benjamini-Hochberg correction was also applied.

ICU characteristics
This cohort included 6993 eligible patients from 23 centers. After excluding 270 records of ESRD patients and 1255 records of patients with missing classi cation data regarding mechanical ventilation status and AKI status, 5468 patients were available for the nal analysis ( Figure 1). The majority of the patients were from Thailand (59.4%) followed by Laos (26.5%), and Indonesia (14.1%). Mixed ICU and medical ICU were the most common type of ICU in this cohort (49.8% and 44.5% respectively). In terms of hospital types, there were 3207 patients from university hospital (58.7%), 1109 patients from regional hospital (20.3%), and 1152 patients from provincial hospital (21.1%). The de nitions of each hospital type were provided in the previously published study. 25 Patient characteristics by pattern of AKI and ARF Of the analyzable cases, 1392 patients had AKI (25.5%) and 2680 patients had ARF (49%). Of all included AKI cases, 562 had AKI stage 3 (40.4%) and 830 patients had stage 2 (59.6%) at the time of diagnosis. After categorizing patients into each pattern of AKI and ARF, the most common pattern observed was "No AKI/ARF" (N=2156, 39.4%), followed by "ARF alone" (N=1920 patients, 35.1%), "AKI alone" (N=632 patients, 11.6%), "ARF rst" (N=658 patients, 12%), "AKI rst" (N=79 patients, 1.4%), and "Concurrent AKI-ARF" (N=23 patients, 0.4%).
The average age of the entire cohort was 59.3 ± 8.7 years and 41.2% were female patients. The main primary diagnosis for ICU admission were cardiovascular disease, infectious disease, pulmonary disease, and surgical-related disease. All patient baseline characteristics strati ed by pattern of AKI and ARF are provided in Table 1. Regarding the etiology of AKI, sepsis and renal hypoperfusion were the leading causes of AKI across all patterns of AKI and ARF (Table S1). As for ARF, hypoxemic ARF was the most predominant type in all patterns (Table S2).

Patient outcomes
Overall in-hospital mortality of this cohort was 25.1 percent. Based on AKI and ARF pattern, "No AKI/ARF" had in-hospital mortality of 12.4% similar to 14.6% of those in "AKI alone" while "ARF alone" had mortality of 31.5%. As expected, inhospital mortality was higher in patients with dual organ failure (53% in "ARF rst" and 54.4% in "AKI rst"). Among those with dual organ failure, "Concurrent AKI-ARF" had higher in-hospital mortality of 69.6% than the other two groups.
As shown in Figure 2, survival is also significantly different between AKI-ARF pattern with median survival time of 47 days for "ARF alone", 59 days for "AKI alone", 31 days for "ARF rst", 15 days for "AKI rst", and 21 days for "Concurrent AKI-ARF" (P=0.001; log-rank test). Median survival time was not reached in "No AKI/ARF". The same trends observed in unadjusted analyses were also seen in the adjusted model, in that "No AKI/ARF" does best follows by "AKI alone", "ARF alone", "ARF rst", "AKI rst", and "Concurrent AKI-ARF", accordingly ( Figure 3).
To explore whether risk factors for in-hospital mortality would change between patterns of AKI and ARF, separate multivariable logistic regression models were built for each of the pattern of AKI and ARF pattern (Table S3). We also discovered signi cant interactions between patterns of AKI and ARF and the other risk factors on in-hospital mortality including hospital type and vasopressor use (Table 4).
Factors associated with the development of the secondary organ failure Multivariable logistic regression models were also performed to explore the risk factors associated with the development of secondary organ failure i.e. secondary ARF (ARF after AKI) and secondary AKI (AKI after ARF).

Discussion
We have displayed that the incidence of patients with either AKI or ARF in resource-limited settings were as high as 60.6% of the total ICU population, 22.9% of which had both AKI and ARF. The in-hospital mortality between different groups of patients varied from 12.4% to 69.6%. (Table 2) Patients with double organ failure had much poorer outcomes than those with single organ failure, especially those with "Concurrent AKI-ARF". (Table 2, Figure 2) Our study also showed that the onset-based classi cation of AKI and ARF interplay has shown different risk pro les. (Table 3, 4) In our study, the in-hospital mortality rates of patients in each group were higher than those previously reported in highresource-settings. 4,[26][27][28][29] For example, an observational cohort by Darmon et al. showed that the in-hospital mortality of AKI with acute respiratory distress syndrome (ARDS) were 42.3%. These distinct ndings are probably due to the different risk associated with the resource-limited settings. One of the main factors is the delayed ICU admission which could be re ected by the earlier onset of AKI and ARF in the ICUs (median time to organ failure onset 1 day for ARF and 2 days for AKI). Among patients with single organ failure, we observed that the in-hospital mortality of "ARF alone" appeared to be higher than "AKI alone". This may be due to the higher severity of acute illness which were re ected by higher APACHE II score and SOFA score in "ARF alone". After adjusting severity scores as confounders, the adjusted odds ratio of "AKI alone" and "ARF alone" on in-hospital mortality were not statistically different (Table 3). This might imply that the de nition of AKI by KDIGO stage 2 and the de nition of ARF by mechanical ventilation status re ect similar mortality outcomes.
Another interesting result of our analysis is that the outcome of AKI-ARF interaction depends not only on the additive effect of the total number of organ failure but also the timing of organ failure which results in the highest in-hospital mortality among those with "Concurrent AKI-ARF". Though the higher severity of concurrent occurrence of organ failure may be confounded by the shorter time of organ failure evolution, contradicting evidence were described by Sakr et al. that the time required for ICU patients to reach the highest degree of organ failure was shorter in survivors. 29 Thus, the different pattern of AKI and ARF interaction may be regarded as different clinical entity.
From our analysis, uid overload, de ned by positive uid balance of more than 10% of the actual bodyweight, was a major independent risk factor for the development of secondary organ failure in both AKI after ARF and ARF after AKI. This nding was consistent with many studies in the past. [30][31][32] This signi es the importance of conservative uid management to prevent further organ failure after an occurrence of either AKI or ARF as uid overload is a modi able risk factor.
While higher maximum AKI staging was a risk factor for ARF after AKI, renal recovery was an independent protective factor. These results suggest that the patients with severe and/or persistent AKI are at higher risk of subsequent development of ARF. The dose-response relationship further validates the causality between AKI and ARF. Therefore, providing proper AKI management may also improve pulmonary outcomes. 13 Unsurprisingly, vasopressor use and higher APACHE II score were both independent risk factors for AKI after ARF. Though this may merely re ect the higher severity of illness, the independent relationship between vasopressor use and AKI development after ARF may also suggest unstable hemodynamics as a pathophysiological mechanism of AKI in patients with ARF. Moreover, obesity was also found to increase the risk of AKI development after ARF. This result was corroborated by the study by Soto et al. which observed obesity as an independent risk factor for development of AKI in 751 patients with acute respiratory distress syndrome. 33 However, the underlying mechanism is still yet to be elucidated.
To our knowledge, this is one of the rst few studies that demonstrates the patterns of AKI and ARF and their relevance to the clinical outcomes. A key strength of our study is the large database of patients and its multicenter, including a wide coverage of population in Southeast Asia and a wide variety of ICUs. It also contains the data regarding the process of care i.e., reimbursement data and levels of hospital. This data are very important and highly relevant to the treatment decision especially in the limited-resource settings. 34 Another strong point is that our AKI de nition and staging adhere to the KDIGO criteria which incorporates both serum creatinine and urine output data. Furthermore, the availability of data permitted the detailed analysis of risk factors of AKI after ARF and ARF after AKI, something that has previously received little attention.
We acknowledge a number of limitations of our study. First, because the parent study, the SEA-AKI study, mainly focused on risk factors for AKI, we lack some of the important ARF-related data including the etiology of ARF, blood gas, and mechanical ventilator settings. However, a large observational study by Panitchote et al. reported that ventilator-speci c variables had no impact on the development of AKI in patients with ARDS. 27 Therefore, including those variables may not substantially alter the ndings of our study. Second, our study covers only short-term outcomes, namely, the inhospital mortality. The longer-term outcomes still needed to be further elucidated. Third, the number of patients in "Concurrent AKI-ARF" is relatively low because most patients who simultaneously met criteria of AKI and ARF on the day of ICU admission could not be determined whether AKI and/or ARF was developed prior to the ICU admission and, hence, need to be excluded from our analysis. In the resource-limited settings, large number of patients are mechanically ventilated outside the ICUs due to short supply of ICU beds. 35 Therefore, we could not determine the true onset of those who developed ARF since the rst day of the ICU admission. Fourth, true prior baseline serum creatinine levels were determined only in 22% of patients and we chose the reference creatinine from the lowest value between the rst serum creatinine on the day of admission and the MDRD formula assuming a GFR of at least 75 mL/min/1.73 m 2 . If a patient had a history of CKD, then we used the rst available serum creatinine as the baseline serum creatinine, as MDRD back calculation assumes that patients do not have CKD. Lastly, the amount of urine output was recorded on the 24-hour basis. Therefore, the modi ed urine output criteria were used for the diagnosis of AKI. This approach would have delayed the diagnosis in some of the patients with stage 2 AKI by urine output which may result in misclassi cation of AKI-ARF pattern. However, we allowed the time span of 24-48 hours between each pattern in order to compensate the risk of misclassi cation.

Conclusion
AKI and ARF are very prevalent in the ICU. The outcomes of the interaction of AKI and ARF are severe, especially in the resource-limited settings. Based on the sequence of organ failure occurrence, different patterns of AKI and ARF and their distinct clinical outcomes were identi ed. While the risk factors of in-hospital mortality of each pattern of AKI and ARF were almost similar, the risk factors for the development of secondary organ failure varied by the condition of the primary organ failure. Fluid optimization may be the key to prevent the development of secondary organ failure in the AKI and ARF interaction. The results of our study should foster further research into the clear understanding of lungkidney interactions in critically ill patients.

Declarations
Ethical approval and consent to participate The study protocol was reviewed by the Institutional Review Board at each participating site and the need for informed consent was waived. For this study, the Institutional Review Board of Faculty of Medicine, Chulalongkorn University approved the study (IRB 440/63).

Consent for publication
Not applicable.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests
The authors declare that they have no competing interests. Authors' contributions WK, SP, NL, and NS were responsible for the study concept and design. WK, SP, NL, and NS were responsible for the data acquisition. WK and KS were responsible for statistical analysis. WK, SP, ST, NL, JK and NS were responsible for data interpretation. WK, SP, ST, NL, and NS were responsible for drafting the main manuscript. WK prepared the gures. All authors reviewed the manuscript. 34. Sengthavisouk N, Lumlertgul N, Keomany C, Banouvong P, Senavong P, Sayyaphet S, et al. Epidemiology and shortterm outcomes of acute kidney injury among patients in the intensive care unit in Laos: a nationwide multicenter, prospective, and observational study. BMC Med 2020;18:1-9.  Tables   Table 1 Baseline characteristics strati ed by acute kidney injury and acute respiratory failure patterns