Background: Severe acute kidney injury (AKI) among critically ill patients is sometimes treated with renal replacement therapy (RRT), and in Sweden continuous RRT (CRRT) is the dominant modality used in this population.

• The optimal timing of renal replacement therapy (RRT) initiation in critically ill patients with acute kidney injury (AKI) is unknown
• No consensus to guide clinical practice on this issue
• Lack of consistency regarding outcome measurements; should we look at morbidity or mortality?
• Wide variability in the timing of RRT initiation in the intensive care unit (ICU) population

Hypothesis: This is an important knowledge gap in the support of critically ill patients with AKI and we hypothesize that early initiation of RRT is beneficial.

Methods: The present study aims to test this hypothesis by using a large scale high resolution intensive care database, the Clinisoft repository. In this database, we have information on >60 000 patients from three different hospitals and five ICUs, during the years 2005 up until today. The repository will be crossmatched, using the unique Swedish national ID number, with hospital records; to gather information on preexisting illnesses, chronic medication and post-ICU outcomes. It is likely that over 5%, more than 3000 patients, have been treated with RRT. We will categorize these patients into "early" and "late" groups using both biomarker data and clinical data. Importantly, early and late RRT can be categorized using biomarkers, like urea and creatinine; using degree of fluid accumulation, by level of pH in blood and just by using hours-days after ICU admission. All possible definitions of early/late RRT initiation can be tested in this study.

Outcomes: Our primary outcome is 90 day mortality. Secondary outcomes include: mortality at 30, 60, 180 and 365 days. Two- and three year mortality.

Morbidity, measured as end-stage renal disease (ESRD) for 90-day survivors. ICU length of stay, hospital length of stay.

Background

• Optimal timing of renal replacement therapy (RRT) initiation in critically ill patients with acute kidney injury (AKI) is unknown
• No consensus guides clinical practice on this issue
• Conflicting results from randomized controlled trials
• Lack of consistency regarding outcome; should we look at morbidity or mortality?
• Wide variability in the timing of RRT initiation in this population
• This is an important knowledge gap in the support of critically ill patients with AKI

Most previous studies have pointed to benefits of early RRT:

1. The SOAP study (Payen 2008) enrolled> 3100 patients and 278 patients required RRT - in these patients, "early RRT" was defined as RRT initiation within two days of ICU admission while "late RRT" was defined as RRT initiation occurring thereafter. The early RRT group was significantly sicker, including higher SAPS II/SOFA scores, greater need for mechanical ventilation, and lower urine output compared to the late RRT group. Despite this (and without adjustment for these clinical differences at baseline), both ICU and hospital mortality along with ICU length of stay were significantly lower in the early RRT group.

2. A secondary analysis of the FINNAKI (Vaara 2014) studied 239 critically ill patients with acute kidney injury treated with renal replacement therapy. The exposure was timing relative to evidence of developing ≥ 1 "conventional" indications, and three groups were defined. A) "Pre-emptive" - no conventional indication. B) "Classic - urgent" - renal replacement therapy started < 12 hr of indication. C) "Classic - delayed" - renal replacement therapy started ≥ 12 hr after indication. The analysis was stratified by group + propensity-matched analysis of "pre-emptive" to non-renal replacement therapy treated. 90-Day Mortality results were quite striking.

   Pre-emptive renal replacement therapy vs. Classic renal replacement therapy 30% vs. 49%; OR 2.05; 95%, CI 1.0-4.1 Classic Urgent vs. Classic Delayed 39% vs. 68%; OR 3.85; 95% CI, 1.5-10.2 Pre-emptive vs. no renal replacement therapy (67% matched) 27% vs. 49% (diff 22.4%; 95% CI, 7.5-35.9)

3. Karvellas (2011) performed a meta-analysis comparing early vs late initiation of renal replacement therapy. 15 unique studies (2 randomised, 4 prospective cohort, 9 retrospective cohort) out of 1,494 citations. The overall methodological quality was low. Early, compared with late therapy, was associated with a significant improvement in 28-day mortality (odds ratio (OR) 0.45; 95% confidence interval (CI), 0.28 to 0.72).

4. Leite (2013) performed a study of 150 patients with AKIN stage 3. Mortality was lower in the early renal replacement therapy group (51.5 vs. 77.9%, P = 0.001). After achieving balance between the groups using a propensity score, there was a significant 30.5 (95% confidence interval [CI] 14.4 to 45.2%, P = 0.002) relative decrease of mortality in the early renal replacement therapy group. Moreover, patients on the early renal replacement therapy group had lower duration of mechanical ventilation, time on renal replacement therapy and a trend to lower intensive care unit (ICU) length of stay.

5. Shiao (2012) looked at 648 postoperative renal replacement therapy cases. Definitions were: early (EG, ≦1 day), intermediate (IG, 2-3 days), and late (LG, ≧4 days) groups. The in-hospital mortality rate censored at 180 day was defined as the endpoint. A U-shaped curve was found, indicating that intermediate start of renal replacement therapy was beneficial in this cohort.

6. The two RCTs; the AKIKI and ELAIN-studies (2016) from NEJM and JAMA report conflicting results. AKIKI showed no benefit with early initiation of renal replacement therapy but ELAIN did. Complicating matters were the fact that AKIKI dosing was unknown, and IHD was used in around 50% of those patients.

There are ongoing randomized controlled trials, but STARRT-AKI (see #2) have released feasibility data.

1. I.D.E.A.L.-I.C.U. (NCT01682590) (start Mar 2012): - RCT of "early" (within 12 hrs of RIFLE-F) vs. "deferred" (48-60 hr) on 90-day mortality in septic AKI (planned recruitment ~ 824) - PI: J.P. Quenot (Dijon) - completion March 2015.
2. STARRT-AKI (NCT01557361) (May 2012-October 2013). "Accelerated" (n=48) vs "Standard" (n=52). 90-day mortality was similar, 38 and 37%. Surviving patients requiring renal replacement therapy at Day 90 were only found in the standard arm, where 3.8% were dialysis dependent. Median ICU length of stay and number of renal replacement therapy sessions were insignificantly lower in the accelerated group.

It is clear that more knowledge regarding outcome concerning mortality, morbidity (end-stage renal disease) as well as ICU- and hospital care consumption would be beneficial. The investigators have highly detailed data from three hospitals based on a cohort of over 50 000 ICU patients from Stockholm. The investigators believe that the sheer number of patients combined with the subjective nature* of renal replacement therapy initiation allow for useful information, potentially adding to the knowledge that the RCTs hopefully will bring.

*Initiation of renal replacement therapy (= CRRT) is based on a spectrum of clinical information and provider bias. Clinicians always initiate renal replacement therapy when confronted with life-threatening complications (like hyperkalemia or pulmonary edema). However, wide variations in the minimum severity of indications prompting renal replacement therapy initiation exist. Factors influencing decisions include: age, comorbidity, responsiveness to diuretic challenge; illness severity (predicted mortality); prescribing service and time of day.

The fact that renal replacement therapy is so "unregulated" is beneficial for a cohort study, as it is likely to create a wide variety of patients in early and late groups. It creates a "natural experiment". If it is good for the patients is another matter.

Hypothesis: early renal replacement therapy is beneficial compared to late renal replacement therapy.

Methods: this is an observational cohort study of critically ill patients treated in three hospitals (Karolinska Solna, Karolinska Huddinge and Södersjukhuset) and in five different ICUs in Stockholm, Sweden, between 2007 and 2017 using the Clinisoft database.

Registry: The Clinisoft database (Centricity Clinisoft, General Electric) is based on a patient database management system (PDMS), implemented 2005 and initially only recorded data from Karolinska University Hospital Solna (Central medicosurgical ICU and neurosurgical/neurological ICU, NIVA) and Karolinska University Hospital Huddinge (medicosurgical ICU, IVA). From 2006 to 2008 the other ICUs (Cardiothoracic ICU, Karolinska Solna and Södersjukhusets two ICUs, one medical and one surgical) started using Clinisoft. Essentially, the investigators will have complete data coverage from 2007/2008-2017. The neurosurgical/neurological ICU will not participate in this study; their patients with need of CRRT are treated in other ICUs.

The investigators will collect demographic data, age, sex and comorbidities. Moreover, The investigators will collect ICU specific data; illness severity, mechanical ventilation, need for vasoactive therapy, reason for ICU admission, data on initiation of renal replacement therapy. Post-ICU mortality and mortality data at different time points (30, 60, 90, 180, 365 and when applicable 2 and 3 year mortality) will be reported. Lastly, The investigators will investigate post-ICU morbidity, including new onset CKD diagnosis and ESRD for 90 day survivors.

Early renal replacement therapy can be defined in many ways, based on:

1. Blood urea nitrogen level
2. Grade of RIFLE/AKIN/KDIGO stage
3. Hours from ICU admission
4. Degree of fluid overload
5. Level of pH at initiation

The Clinisoft registry, as mentioned, based on the Centricity Critical Care suite of programs is run by General Electric. This registry has data based on intensive care monitors and machines, ranging from medical pumps, respirators to invasive monitoring data. The registry also has laboratory variables automatically imported from the electronic medical records (EMR) system, Take Care. Moreover, illness severity scores, central lines, body weight, urinary output and medications are entered manually into the system on a regular basis. Data is regularly audited as part of the hospital standard operating procedure. There are data sum checks ensuring that predefined rules and ranges are followed. The investigators have source data verification, as some laboratory data from the electronic medical records (EMR) system Take Care is exported to the Clinisoft registry. Coding information and a data dictionary exists.

Matching: The Standard Operating Procedures for patient recruitment, data collection, data management and data analysis is as follows. For each case of early CRRT, The investigators will create a sequentially propensity matched control patient that a) either will never receive CRRT or b) will go on to get late CRRT based on absolute indications.

Absolute indications:

1. hyperkalemia (serum potassium≥6 mEq/L),
2. severe acidosis (pH≤7.15),
3. plasma urea>36 mmol/L (equals BUN=100.8 mg/dl),
4. oliguria or anuria (urine output<0.3 ml/kg per hour for ≥24 hours or anuria for ≥12 hours), and
5. fluid overload with pulmonary edema as defined by the presence of all the following factors: (a) >10% fluid accumulation (cumulative fluid balance/baseline weight>10%), (b) oliguria (urine output<0.5 ml/kg per hour for ≥12 hours), and (c) severely impaired oxygenation (PaO2/FiO2<200 indicated by respiratory Sequential Organ Failure Assessment [SOFA] score≥3) A patient where initiation of renal replacement therapy is started without the absolute indications ("early renal replacement therapy") will be matched based on age, sex, chronic kidney disease, admission diagnosis group, presence of severe sepsis and cumulative urine output. The investigators will try to find matched pairs going on to get "late renal replacement therapy" and never renal replacement therapy; i.e. two matched patients per early renal replacement therapy-patient.

Sample size assessment: first off, out of the total number of patients - around 50 000 patients collected during the years 2007-2017 - the investigators anticipate at least 5% treated with CRRT, i.e. 2500 patients.

value for p1: 0.35 value for p2: 0.45 value for α: .05 value for power (default is .80): .80 sample size (for each sample separately) is: 376

This power analysis is based on regular alpha and power (0.5 and 0.8) and expects a 10% (a 35% mortality and 45% for early vs late renal replacement therapy) difference in mortality at six months - significantly smaller difference than in the Vaara, Karvellas and Leite studies.

This gives us a sample size of 375 patients in the two groups, 750 patients in total.

The investigators are likely to greatly exceed the numbers needed to find a difference in mortality if such a mortality difference exists. It is unclear if the investigators will exceed the numbers needed to find a difference in ESRD-morbidity, since the field lacks adequate data to allow for a power analysis.

Plan for missing data: our trial design seeks to limit the likelihood of missing data. If variables are missing, unavailable, non-reported or uninterpretable the investigators will use imputation by last observation. This is, admittedly, based on unscientific assumptions, therefore sensitivity analyses will be conducted to assess the robustness of findings when missing data has been an issue.

Statistical analysis plan: The investigators will report continuous data as medians with interquartile ranges and categorical data as counts and percentages. The investigators will use the Mann-Whitney U test to compare continuous data and the Fisher's exact test to compare categorical data. All statistics is two-tailed and a P value of less than 0.05 will be considered significant.

CRRT initiation will be categorised according to early and late start according to several different criteria i.e. time from ICU admission, biochemical data at initiation and percentage of body weight increase. The investigators will then perform separate univariate and multivariate analysis for each definition, using Cox and Poisson regression models for primary and secondary outcomes and adjust for clinically relevant confounders and independent factors with observed differences between patients with early and late renal replacement therapy initiation.

A competing risks polynomial logistic regression will be performed to identify predictors of primary and secondary outcome at 6 months. This risks model will include four-level polytomous outcomes defined as death, ESRD, ESRD and death or no negative outcome. The area under the receiver operating characteristic curve (AUC) will be used to assess discrimination of each of our CRRT timing models.

Analysis will be performed using Stata version 12 (StataCorp LP, College Station, Tx, USA).

Inclusion Criteria:

*Critically ill patients admitted to intensive care units in Stockholm at: Karolinska University Hospital (Solna and Huddinge) and at Södersjukhuset.

*Patients over 18 years of age

Exclusion Criteria:

• Patients <18 years
• Patients with DNAR (do not attempt resuscitation)-orders
• Patients dying within 12 hours of commencing renal replacement therapy.

ICU patients treated in Stockholm between 2007 and 2017 in the following five ICUs from three hospitals: Karolinska University Hospital in Solna - central medico-surgical ICU; Karolinska University Hospital in Solna - cardiothoracic ICU; Karolinska University Hospital Huddinge - medico-surgical ICU; Södersjukhuset Hospital - medical ICU; Södersjukhuset Hospital - surgical ICU.

We plan to share the database, once anonymized, with other researchers.