Isotonic fluids have been established as the preferred fluid choice since a pivotal study in 1958 compared with hypotonic, isotonic, and hypertonic fluids in patients with severe DKA. These authors found that hypertonic fluids may be detrimental due to worsening of hyperosmolarity, hypernatraemia, and hyperchloraemia, while hypotonic fluids were noted to cause diuresis (3).

In 2013, a Cochrane review of 78 randomised controlled trials (RCTs) assessed the effect of colloids compared to crystalloids for fluid resuscitation in critically unwell patients with trauma, burns or in the post-operative period (15). Crystalloids were preferred as colloids were considerably more expensive and offered no survival benefit over crystalloids (15). However, specifically hydroxyethyl starch crystalloids were associated with a significant increase in mortality and acute kidney injury, and it was concluded that they should be avoided (14–19). It is important to note that none of the included studies specifically considered these fluids in the setting of DKA.

The optimal crystalloid solution to use in DKA is unclear (Table 2). Among crystalloids, normal saline (0.9% sodium chloride) has come under scrutiny due to concerns raised over hyperchloraemic metabolic acidosis, as well as a possible propensity for oliguria, prolonged acidosis, and coagulopathy (1, 5, 14, 16–18, 20–25).

During the recovery phase of DKA, hyperchloraemia tends to develop because of preferential excretion of ketones during rehydration and improved renal perfusion, resulting in a raised anion gap metabolic acidosis (14). It is proposed that rehydration with normal saline may contribute to hyperchloraemia and a hyperchloraemic metabolic acidosis with a persisting base deficit and may cause renal vasoconstriction and decreased glomerular filtration rate (14, 27). Alternatively, this acidosis may represent a physiological response to resolving DKA rather than a result of the hydration fluid itself (17, 20–23, 28).

In the DKA-specific setting, a small, randomised, double–blinded, prospective trial (n = 45) compared the effects of fluid resuscitation in patients with DKA using Plasma-Lyte (a balanced electrolyte solution) and normal saline and concluded that balanced electrolyte solutions resulted in lower serum chloride and higher bicarbonate levels consistent with prevention of hyperchloraemic metabolic acidosis (16). However, there was no difference in clinical outcomes (16) (Table 2).

A similar conclusion was drawn in a small three centre retrospective study (n = 23) which compared Plasma-Lyte to normal saline as fluid resuscitation in DKA (14). This study was limited by its patient numbers with difficulty matching cases to controls.

The only other RCT was also small, studied different crystalloids in DKA, comparing patients who received Ringer’s lactate (a balanced electrolyte solution) (n = 28) with patients who received normal saline (n = 29). In this study, Ringer’s lactate offered no significant superiority in time to normalisation of pH and took a significantly longer time to achieve a blood glucose concentration ≤14mmol/L (26). A proposed mechanism for this delay in control of blood glucose level was that lactate from Ringer’s solution provided excess substrate for ongoing gluconeogenesis (26). Using the 2006 (and then the 2009 post hoc) American Diabetes Association (ADA) definitions for resolution of DKA, fluid selection made no statistically significant difference in time to resolution (26).

Another concern regarding normal saline is saline-induced oliguria. In the DKA context, a RCT comparing Plasma-Lyte to normal saline found cumulative urine output to be lower at 4–6 h in the normal saline group (14).

Balanced electrolyte solutions have insufficient potassium content when used alone and are not available with adequate premixed potassium (7). For this reason, the United Kingdom National Institute for Health Care Excellence (NICE) and the ADA guidelines favour continued use of normal saline (29).

From the available evidence, we conclude that there is no evidence for significant superiority between crystalloid and colloid solutions, while noting that hydroxyethyl starch solutions are best avoided due to risk of mortality and acute kidney injury.

There is currently insufficient evidence for superiority amongst crystalloids, with very weak evidence to suggest that balanced electrolyte solutions may reduce the risk of hyperchloremic metabolic acidosis, though this has not translated to any clinical significance in patient outcomes in DKA. Normal saline is cheap and familiar to all clinicians and therefore can be used as the fluid of choice (Figure 3).

The rationale for intensive hydration in DKA is to avoid hypoperfusion, correct marked hyperglycaemia and hyperosmolarity, and improve responses to insulin therapy (1). Alternatively, the rationale for slower rehydration is to reduce the risks associated with correcting the hyperosmolar state too rapidly (30, 31) thus decreasing the potential risk of cerebral oedema (31, 32) as reported in children (32, 33).

Most studies of hydration rate take place in the intensive care setting making use of central venous pressure monitoring as opposed to general ward-based management. The latter is more common in the Australian context (22, 28, 34, 35).

Conservative hydration in relatively small numbers of open label studies revealed no adverse events, although the selection of patients was tightly regulated and may not be generalisable (22, 23, 28, 35). In 1989, a prospective analysis (n = 28) assessed aggressive resuscitation at 1,000 mL/h for 4 h followed by 500 mL/h for 4 h, compared to rehydration at half this rate (22). The study concluded that a slower rate of rehydration still led to prompt recovery, lack of any harmful effects, and had a significant reduction in the overall cost of medical therapy (22). This principle was reiterated by another, also prospective randomised controlled study with limited numbers (n = 27), comparing the effect of hydration using a rate of 1,000 mL/h versus a rate of 500 mL/h of normal saline, which concluded there was no significant difference in rate of resolution of biochemical derangements (36).

The major concern with rapid hydration is cerebral oedema which is seen in the paediatric population (33) and rarely reported in adult literature. There have been limited numbers of prospective trials, although given the rarity of this outcome, these have been inadequately powered (16, 20–23, 28, 35). Others have argued that rapid hydration is not a concern as cerebral oedema, as demonstrated by CT imaging of the brain, may be present prior to hydration and relates to the severity of initial acidosis rather than medical intervention (31, 34).

Pulmonary venous congestion is another potential complication of DKA management, although there is minimal evidence for this occurrence. Assessing extravascular lung water and colloid-osmotic pressure in a case series from 1988 did not reveal any tendency towards pulmonary venous congestion, despite large total infused volumes of 6 ± 1.01 L (35).

The ADA consensus statement in 2009 recommended that in the absence of cardiac compromise, normal saline is to be infused at a rate of 15–20 mL/kg body weight or at 1–1.5 L during the first hour (1). Determination of subsequent rates should be dependent on patient haemodynamic and hydration status and serum electrolyte levels (1). Neither studies nor guidelines addressed the issue of fluid replacement in obese patients with DKA.

In patients with normal or high corrected serum sodium levels, the recommended rehydration fluid is 1/2 normal saline (0.45% sodium chloride) given at 250–500 mL/h, whilst the recommended choice of rehydration fluid for patients with low corrected serum sodium is still normal saline, infused at the same rate (1). In either instance, fluid deficits should be corrected within the initial 24-h period (1). NICE guidelines recommend hydration at rates which are “not too rapid” except in cases of circulatory collapse (29).

The currently available evidence is very weak and suggests slow hydration rates result in equal management outcomes, with no consensus regarding optimal rates.

In 1987, Storms et al established that regular human insulin was superior to porcine insulin due to faster resolution of DKA (49), and for decades regular human insulin has remained the insulin of choice in DKA management. This convention has been challenged by the emergence of short and ultrashort-acting insulin analogues, which has now become much more familiar to clinicians than regular human insulin in general diabetes management.

Insulin analogs glulisine, aspart and lispro have been reported to have equal efficacy and in vivo potency compared to regular insulin in animals and humans, attributable to their similar receptor binding affinity and receptor mediated clearance (50–53), although only studies comparing the role of intravenous glulisine infusion as an alternative to intravenous infusion of regular human insulin has been performed (54). In spite of their more rapid onset, studies comparing the pharmacokinetics and pharmacodynamics of intravenous glulisine (an ultrashort-acting insulin analogue) to intravenous regular insulin (short-acting insulin) have found glulisine demonstrates equivalent glucose utilisation and disposal; and a similar distribution and elimination profile to regular insulin (50, 54).

Non-inferiority was demonstrated in a small, multicentre, open-label RCT (n = 68), comparing IV infusions of glulisine to regular human insulin. This study found both insulin types to be equally effective during the acute treatment of DKA, with no statistical difference in mean treatment duration (15.7 ± 4.5 vs. 20.5 ± 12 h) or amount of insulin infused until resolution of DKA (70 ± 33 vs. 76 ± 46 units) for glulisine and regular human insulin, respectively (54). Furthermore, the rate of decline of blood glucose concentrations, changes in bicarbonate and anion gap were similar, and neither groups experienced mortality or recurrence of DKA.

Due to the lack of superiority and the higher cost of insulin analogs, IV infusion of regular insulin is still considered the mainstay over insulin analogs in ADA consensus statement, though has not been specified in the NICE guidelines (1).

The attraction of subcutaneous insulin is to avoid the need for intensive care unit admission for IV insulin infusions and its associated high costs, although the concern is that it may not be as effective as IV insulin.

Studies comparing treatment of DKA using subcutaneous or intramuscular insulin against conventional IV insulin infusions found slower plasma glucose reductions at 1 h (failure to reduce by 10%) (55) and at 3 h (56) and slower ketone clearance (56).

Additionally, a retrospective study (n = 59) comparing bolus insulin injections in intravenous, subcutaneous, and intramuscular routes against continuous IV infusions found a higher rate of hypoglycemic episodes amongst the bolus insulin injection group (8 out of 30 patients), compared to the intravenous infusion group (1 out of 29 patients) (p = 0.03) (57).

Importantly, none of the abovementioned studies assessed longer term outcomes such as time in intensive care or length of hospital stay. As such, it is unclear whether rapid correction of hyperglycemia and ketonaemia necessarily leads to better outcomes.

A Cochrane review in 2016 analysed five small RCTs (total n = 201) comparing subcutaneous injections of insulin analogues (lispro and aspart) to regular insulin IV infusions in patients with mild to moderate DKA (Table 4) and revealed no statistical difference in time to resolution of DKA or rates of hypoglycaemia (42). These findings were based on very low to low-quality evidence and were limited by the number of included trials and participants. Other than hypoglycaemia, there was no analysis of adverse outcomes, and no deaths occurred in these studies (42).

The benefits of administering subcutaneous insulin injections hourly and 2 hourly outside of the intensive care setting must also be weighed against the increased demands on nursing staff on medical wards, increased variability with dose administration times, time to onset of action, peak effect, and duration of effect (58).

An alternative management strategy is the initiation of both an intravenous insulin and continuation of the patient’s subcutaneous long-acting insulin during the initial management of DKA. This aims to decrease insulin infusion requirements, provide background insulin once intravenous insulin is discontinued, and reduce the risk of hypoglycaemia postcessation of insulin infusion. This was recommended in the NICE guidelines 2016 but was not present in the ADA 2009 consensus statement (1). This concept is supported by a prospective randomised study, which concluded that once daily subcutaneous insulin glargine administered during intravenous insulin infusions, safely prevented rebound hyperglycaemia without risk of hypoglycaemia (43).

Conversely, a pilot prospective RCT concluded that this model of treatment was feasible, although it did not find a significant benefit in time to closure of anion gap, or number of hypoglycaemic events (44) (Table 5).

Since the first trials in the 1970s, continuous subcutaneous insulin infusions delivered via insulin pumps have been used in patients with type 1 diabetes, and it is well established that DKA can occur as a complication of pump failure (59–62). Currently, there is no evidence surrounding the continuation or discontinuation of pump therapy during management of DKA. Current practice involves cessation of insulin pump therapy during management of DKA, then transition back to pump therapy after resolution of DKA (60). Conversely, given the potential roles of both continuous intravenous insulin and subcutaneous insulin injections, it could be postulated that insulin pumps may have a role in DKA management, as it is a convenient and pre-existing route of administration, although no studies have explored this potential.

The ADA 2009 consensus statement asserts that insulin therapy is effective regardless of route of administration, although both the ADA and NICE guidelines advocate for intravenous infusion (1, 29). The current evidence is weakly suggestive that there is no difference in outcome between administration of insulin as IV infusion or SC injections and is conflicting regarding impact of combining SC long-acting insulin with IV insulin infusion. Morbidity and safety data are limited with hypoglycemia being the only reported adverse outcome, and no deaths occurred in studies to date.

An initial intravenous insulin bolus has been considered in multiple small trials.

A small prospective analysis in 1995 found patients who received an insulin bolus had a significantly higher incidence of hypoglycemia but no difference in hypokalaemia (57).

In contrast, a prospective observational cohort study in 2007 (n = 157) found no statistical difference in the incidence of hypoglycaemia, rate of glucose reduction, or length of hospital stay (63) and concluded that an initial bolus of intravenous insulin offered no significant clinical benefit or harm. The study was limited by lack of randomisation; standardised treatment protocols and used administration of 50% dextrose as a surrogate definition for hypoglycemia rather than serum glucose levels.

In 2008, a small prospective randomised protocol established that an initial bolus of insulin avoided the need for supplemental insulin doses if the insulin infusion rate was less than 0.14 units/kg/h (1, 64). The study compared patients (n = 12) who received a loading dose of 0.07 units/kg of regular insulin, followed by 0.07 units/kg/h insulin infusion, group 2 (n = 12) insulin infusion at 0.07 units/kg/h without loading insulin and group 3 (n = 13) insulin infusion at 0.14 units/kg/h, without loading. It found that group 2 patients required supplemental insulin doses, whilst other patients did not. Otherwise, there was no significant difference in duration taken to reach a serum resolution of DKA, length of hospital stay, hypokalaemia, or other complications including death (64).

Despite these findings, use of an initial bolus insulin dose has been incorporated into the ADA 2009 consensus statements but not in the NICE guidelines (29). The limited available evidence suggests use of bolus insulin results in no difference in outcomes.

Three decades ago, a small prospective RCT (n = 48) established the preference for low-dose insulin over high-dose insulin in management of hyperglycaemic emergencies (65). The study concluded there was no difference in correction rate of biochemical derangements, though 25–50% of patients who received high-dose insulin developed hypoglycaemia or hypokalaemia, compared to only 0–4% rate amongst patients who received low-dose insulin (3, 5, 65).

The intravenous insulin rates remain contentious, with fixed weight-based insulin rates being advocated by national authorities (1, 5, 24), to account for changing patient demographic, including obesity and higher insulin-resistance states such as pregnancy (5).

The Joint British Diabetes Society guidelines recommend insulin infusion at a rate that achieves a reduction in blood ketones by at least 0.5 mmol/L/h or alternatively, a venous bicarbonate increase by 3 mmol/L/h and capillary blood glucose reduction by 3 mmol/L/h (5). Similarly, the 2009 ADA consensus statement recommend an insulin infusion rate adjusted to target a serum glucose reduction rate of 3–4 mmol/L/h with no suggestions for target ketone reduction (1), once serum glucose levels reach 11.1 mmol/L, the insulin infusion rate should be reduced to 0.02–0.05 units/kg/h, coinciding with the addition of 5% dextrose infusion (1).

Whilst weak evidence has supported the traditional preference of low-dose over high-dose insulin, there is currently a lack of head to head studies guiding choice between fixed weight-based dosing over BGL-based sliding scale dosing.

On resolution of DKA and cessation of intravenous insulin infusion, rebound hyperglycaemia is a major issue (43, 66). Once an intravenous infusion is ceased its duration of action is only minutes, whilst the onset of basal subcutaneous insulin is 1–2 h and thus overlap of the two is an integral component of successful transition.

In 2007, a small study (n = 75) investigating the optimal insulin dose for transition from IV insulin infusion to subcutaneous insulin glargline, randomized patients to receive either 40%, 60%, or 80% of their total daily insulin requirement, calculated from the rate of insulin infused during the last 6 h of their insulin infusion, as insulin glargine (67). The study concluded that insulin glargine administered at 80% of the total daily insulin requirement resulted in the highest percentage of capillary blood glucose level within target glycaemic range of 4.4–8.3 mmol/L within the initial 24 h after transition (67).

A prospective randomised trial compared multidose insulin regimens demonstrated that split mixed insulin (NPH plus regular insulin) and basal bolus insulin regimen (glargine and glulisine) offered similar glycaemic control though basal bolus regimen was associated with a lower rate of hypoglycaemic events (15 vs. 41%) (1).

The ADA 2009 consensus statement recommend patients normally on subcutaneous insulin should be recommenced on their usual dose if this was typically adequate for glycaemic control, whilst insulin naive patients should be commenced on either a multidose insulin of 2–3 boluses daily or basal bolus regimen (1).

Current limited evidence suggest that transitioning from insulin infusion to subcutaneous insulin using 80% of the total insulin dose offers optimal glycaemic control, with nil evidence to guide recommendations for either a spit mixed or basal bolus regime.