Abstract
Aims
Diabetic ketoacidosis (DKA) is a serious complication of diabetes, requiring intravenous (IV) insulin until resolution and subsequent transition to subcutaneous insulin. Currently, clinical guidelines vary regarding the timing of long‐acting subcutaneous insulin initiation, with some advocating early administration during IV insulin infusion, while others recommend delaying until DKA resolution. We aimed to evaluate the efficacy and safety of concurrent versus sequential initiation of long‐acting subcutaneous insulin in paediatric and adult patients with DKA already receiving regular insulin.
Materials and methods
A systematic search of five databases (inception to January 2026) identified eligible studies. Early initiation was defined as administration of long‐acting insulin before resolution of DKA, while late initiation occurred after resolution of DKA. Primary outcomes included time to DKA resolution, total IV insulin and fluid requirements, and risks of hypoglycaemia, hypokalaemia, and rebound hyperglycaemia. Pooled effect sizes were calculated using random‐effects models.
Results
Nine randomised control trials encompassing 652 patients were included. Early long‐acting insulin was associated with a shorter time to DKA resolution (SMD: −0.61; 95% CI: −0.83 to −0.38) and was associated with lower total insulin and fluid requirements. Available evidence was insufficient to rule out an increased risk of hypoglycaemia (RR: 0.81; 95% CI: 0.52–1.27) or hypokalaemia (RR: 1.21; 95% CI: 0.90–1.63).
Conclusions
Early initiation of long‐acting insulin during IV insulin infusion in DKA likely shortens time to resolution based on moderate certainty evidence and may reduce total insulin and fluid requirements. Evidence for rebound hyperglycaemia and recurrent DKA outcomes remains limited and imprecise.
Keywords: basal insulin, diabetes complications, insulin glargine, meta‐analysis, systematic review
1. INTRODUCTION
Diabetic ketoacidosis (DKA) is a life‐threatening complication of both type 1 (T1DM) and type 2 diabetes (T2DM), characterised by hyperglycaemia, ketosis, and metabolic acidosis. 1 The current standard of care involves intravenous (IV) fluids for circulatory restoration, correction of electrolyte imbalances, and continuous IV insulin infusion until resolution of ketoacidosis, followed by transition to subcutaneous (SC) insulin.2, 3, 4 In practice, IV insulin is continued until closure of the anion gap before initiating long‐acting SC insulin as part of the transition to routine therapy. 2
Clinical guidelines differ in how they recommend transitioning from IV to SC insulin. The American Diabetes Association recommends initiating long‐acting insulin after DKA resolution, with an overlap of 2–4 h before discontinuing IV insulin to prevent rebound hyperglycaemia and recurrence of DKA. 2 Diabetes Canada guidelines for DKA do not discuss or recommend initiating long‐acting insulin at the same time as IV insulin. 5 In contrast, the Joint British Diabetes specifically suggest continuing long‐acting insulin or starting long‐acting insulin concurrently, based on emerging evidence of improved glycaemic outcomes. 6
Recent studies have explored whether earlier initiation of long‐acting insulin has clinical benefit. Randomised control trials (RCTs) and retrospective analyses in adults have demonstrated that early administration of long‐acting insulin, typically at 0.3 units/kg within 1–3 h of starting IV insulin, can shorten time to DKA resolution and hospital stay without increasing adverse events such as hypoglycaemia or hypokalemia.6, 7, 8, 9 However, not all findings are consistent, with some studies showing no reduction in resolution time but improvements in IV insulin duration and fluid requirements. 10 In a paediatric population, a recent double‐blind RCT demonstrated that early glargine initiation did not accelerate resolution but did reduce treatment failure (defined as failure to reduce BG by 18 mg/dL/h for two consecutive hours or persistent acidosis) and rebound hyperglycaemia, supporting its safety in children. 11 Together, these studies suggest potential but mixed benefits of early long‐acting insulin use across age groups.
The only prior meta‐analysis addressing early versus late initiation of long‐acting insulin in DKA was published in 2016 and included a limited number of RCTs, fewer outcomes, and no age stratification. 12 Since then, several additional RCTs and large cohort studies have been published in both paediatric and adult populations.5, 6, 7, 8, 9, 10, 12, 13 Thus, this systematic review and meta‐analysis aims to evaluate the efficacy and safety of early versus late initiation of long‐acting insulin in both paediatric and adult DKA patients based on updated evidence.
2. MATERIALS AND METHODS
This systematic review and meta‐analysis followed the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines 14 and the Meta‐analysis of Observational Studies in Epidemiology (MOOSE) guidelines. 15 It was prospectively registered on the International Prospective Register of Systematic Reviews (PROSPERO) (CRD420251137318) (Data S1, Supporting Information).
2.1. Study selection and data extraction
A systematic search of PubMed, MEDLINE, Embase, Scopus, and CENTRAL was conducted from database inception through January 2026 (Data S2). Reference lists of included studies and prior reviews were screened for additional eligible trials.
Eligible studies included English‐language RCTs of paediatric or adult patients with DKA. The intervention of interest was early initiation of long‐acting insulin, defined as administration concurrent with IV insulin infusion and prior to DKA resolution (as defined by the respective included studies). Comparators were trials in which long‐acting insulin was initiated late, typically after discontinuation of IV insulin or closure of the anion gap. Outcomes of interest included risk of hypoglycaemia, hypokalaemia, and rebound hyperglycaemia, as well as time to resolution of DKA (hours), total IV insulin administered, and total volume of IV fluids administered. Observational studies, editorials, reviews, conference proceedings without full text, and studies with fewer than 10 patients were excluded.
Two independent reviewers (BN and SQ) screened titles/abstracts and assessed full texts in Covidence (Veritas Health Innovation, Melbourne, Australia). Discrepancies were resolved through discussion with a third reviewer (BD). Two reviewers (BN and BD) extracted study characteristics, risk ratios (RRs) with 95% confidence intervals (CIs) (calculated if not available), as well as mean and standard deviation values for each outcome of interest.
2.2. Statistical analysis
Dichotomous outcomes (hypoglycaemia, hypokalaemia, rebound hyperglycaemia) were summarised using pooled RRs with corresponding 95% CIs. Continuous outcomes (time to resolution, total insulin units, total IV fluids) were analysed using pooled standardised mean differences (SMDs; Hedges' g) to account for differences in measurement scales across studies. For each study, effect sizes and their standard errors were derived from the reported summary statistics (event counts or mean ± standard deviation, and sample sizes). Mean difference analyses were also performed to supplement SMDs.
All pooled estimates were calculated using random‐effects models (restricted maximum likelihood, REML) to account for expected between‐study heterogeneity. Statistical heterogeneity was assessed with the Cochran Q test with corresponding p values, the between‐study variance estimate (τ 2), and the I 2 statistic. Forest plots were generated to visually summarise study‐specific and pooled effect estimates, with the line of no effect at RR = 1 for dichotomous outcomes and SMD = 0 for continuous outcomes (i.e., time to DKA resolution in hours, total administered insulin units, and total IV fluids in mL).
We assessed potential small‐study effects using visual inspection of funnel plots and Egger's regression test (exploratory if <10 studies per outcome). All analyses were performed using the metafor (version 4.8.0) package in R (version 4.3.1, R Foundation for Statistical Computing, Vienna, Austria). Sensitivity analyses were performed based on patient age group (i.e., adult, paediatric), low risk of bias, and for T1DM‐focused trials only.
We also conducted univariate meta‐regression analyses for effect modifiers that were sufficiently reported across trials, including baseline pH, mean anion gap, and paediatric versus adult population. Statistically significant moderators were reported.
2.3. Risk of bias assessment
The Cochrane Risk of Bias (RoB) 2.0 Tool 16 was used to assess the quality of RCTs. A traffic light plot was generated using the robvis tool. 17
2.4. Certainty of evidence assessment
The certainty of the evidence was quantified using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) scale. 18
3. RESULTS
A total of nine RCTs were included (Figure S1), encompassing 652 patients with diabetic ketoacidosis9, 11, 19, 20, 21, 22, 23, 24, 25 (Table 1). All studies were single‐centre trials conducted between 2012 and 2025 across diverse geographic regions. Five studies were conducted in adult populations, three in paediatric cohorts, and one included both adults and children. Most trials enrolled patients with T1DM, although several also included individuals with T2DM. Sample sizes ranged from 40 to 108 participants, with intervention arms (long‐acting insulin) ranging from 20 to 72 patients and comparator arms (non‐long‐acting insulin) ranging from 20 to 50 patients. The duration of SC and IV insulin overlap as well as definitions of DKA resolution, rebound DKA, and rebound hyperglycaemia used by each included study are presented in Table S1.
TABLE 1.
Summary of the design, data sources, and sample sizes of included studies.
| Study | Study design | Country | Study period | Setting | Population | Timing of long‐acting insulin | Baseline severity markers | Diabetes type | Total (N) | Long‐acting group (N) | Non‐long‐acting group (N) |
|---|---|---|---|---|---|---|---|---|---|---|---|
| El Hawary et al. 22 | RCT | Egypt | 2020–2021 | Single centre | Paediatric | Within 6 h of admission | pH ~7.08–7.09; HCO3 − ~9.5 mEq/L; corrected Na+ ~142 mEq/L; K+ ~4.0 mEq/L | T1DM | 100 | 50 | 50 |
| Doshi et al. 19 | RCT | USA | 2012–2013 | Single centre | Adult | Within 2 h of diagnosis | pH ~7.1–7.2; HCO3 − ~12–13 mEq/L; anion gap ~18–20 mEq/L; glucose ~540–640 mg/dL | T1DM and T2DM | 40 | 20 | 20 |
| El Feky et al. 20 | RCT | Egypt | 2024–2025 | Single centre | Adult | Within 3 h of diagnosis | pH 7.13–7.18; HCO3 − 6.43–8.33 mEq/L; corrected Na+ 137.0–140.2 mEq/L; K+ 4.65–5.00 mEq/L; glucose 498–559 mg/dL | T1DM | 90 | 60 | 30 |
| Bansal et al. 11 | RCT | India | 2022–2023 | Single centre | Paediatric | Within 1 h of starting low‐dose insulin infusion | Glucose 472 ± 127 vs. 503 ± 140 mg/dL; pH 7.12 (7.0–7.2) vs. 7.05 (6.95–7.14); HCO3 − 8.3 (6.4–10.3) vs. 7.8 (5.6–9.5) mEq/L; anion gap 22.4 ± 6.0 vs. 23.8 ± 7.1 mEq/L; corrected Na+ 136 (133–142) vs. 140 (136–145) mmol/L; K+ 4.4 (3.7–4.8) vs. 4.3 (4.07–4.8) mmol/L | T1DM | 82 | 42 | 40 |
| Thammakosol et al. 9 | RCT | Thailand | 2020–2021 | Single centre | Adult | Within 3 h of diagnosis | pH 7.20 ± 0.15 vs. 7.31 ± 0.16; HCO3 − 10.3 ± 5.1 vs. 12.8 ± 5.2 mEq/L; anion gap 26.2 ± 5.6 vs. 24.0 ± 5.6 mEq/L; glucose 601.5 ± 210.2 vs. 554.6 ± 193.5 mg/dL; K+ 5.05 ± 1.14 vs. 4.50 ± 0.92 mEq/L | T1DM and T2DM | 60 | 30 | 30 |
| Houshyar et al. 23 | RCT | Iran | 2013–2014 | Single centre | Adult and children | Within 3 h of initiation of IV insulin infusion | Glucose 30.0 ± 11.6 vs. 27.63 ± 5.7 mmol/L; pH 7.09 ± 0.15 vs. 7.09 ± 0.14; HCO3 − 6.51 ± 3.34 vs. 6.37 ± 3.49 mmol/L; Na+ 136.95 ± 3.59 vs. 137.10 ± 4.73 mmol/L; K+ 4.65 ± 0.74 vs. 4.59 ± 0.59 mmol/L; BUN 7.38 ± 2.46 vs. 7.29 ± 3.40 mmol/L | N/A | 40 | 20 | 20 |
| Ammar et al. 24 | RCT | Egypt | 2021–2022 | Single centre | Adult | Within 2 h of admission | Not reported | T1DM and T2DM | 52 | 26 | 26 |
| Saffari et al. 25 | RCT | Iran | N/A | Single centre | Paediatric | Simultaneous start with IV insulin | Glucose 485.5 ± 134.6 vs. 538.0 ± 129.7 vs. 478.5 ± 134.9 mg/dL; Na+ 132.8 ± 3.5 vs. 131.6 ± 3.4 vs. 133.4 ± 3.3 mmol/L; K+ 4.4 ± 0.7 vs. 4.4 ± 0.6 vs. 4.2 ± 0.6 mmol/L; anion gap 23.9 ± 2.5 vs. 23.9 ± 2.8 vs. 24.1 ± 2.5 mmol/L | T1DM | 108 | 72 | 36 |
| Thammakosol et al. 21 | RCT | Thailand | 2023–2024 | Multicentre | Adult | Within 3 h of diagnosis | Glucose 504.4 ± 148.3 vs. 512.9 ± 210.0 mg/dL; β‐hydroxybutyrate 7.3 ± 3.0 vs. 6.7 ± 2.9 mmol/L; HCO₃− 10.6 ± 5.0 vs. 11.1 ± 4.9 mEq/L; anion gap 22.8 ± 5.2 vs. 23.2 ± 6.2 mEq/L; pH 7.24 ± 0.15 vs. 7.28 ± 0.13; K+ 4.77 ± 0.74 vs. 4.61 ± 1.03 mEq/L | T1DM and T2DM | 80 | 40 | 40 |
3.1. Standardised mean differences: Time to DKA resolution
For the time to resolution of DKA, long‐acting insulin was associated with a significantly faster recovery compared to non‐long‐acting insulin based on a pooled analysis of seven trials.11, 19, 20, 21, 23, 24, 25 The pooled SMD was −0.61 (95% CI: −0.83 to −0.38, p < 0.001), corresponding to an approximate reduction of ~3–5 h in time to DKA resolution based on pooled trial means (Figure 1). Visual inspection of the funnel plot showed no obvious asymmetry, although some dispersion was observed among smaller studies (Figure S2). Egger's regression test was not statistically significant (z = 0.460, p = 0.6457). Heterogeneity was low to moderate (I 2 = 28.6%, Q (6) = 8.18, p = 0.23).
FIGURE 1.
Forest plot of standardised mean differences (Hedges' g) for time to resolution (h) comparing long‐acting versus non‐long‐acting insulin in the management of DKA. Negative values favour long‐acting insulin.
When stratifying by age group, adults19, 20, 21, 23, 24 (five studies) showed an SMD of −0.63 (95% CI: −0.96 to −0.29; p = 0.00027; I 2 = 49%), while paediatric studies11, 25 (two studies) showed an SMD of −0.55 (95% CI: −0.87 to −0.23; p = 0.00082; I 2 = 0%), both favouring long‐acting insulin.
A sensitivity analysis of three T1DM‐only studies11, 20, 25 reported a pooled mean difference of −4.18 (95% CI: −5.86 to −2.49, p < 0.001), suggesting a shorter time to resolution with early long‐acting insulin (Figure S3). Heterogeneity was absent (Q (2) = 1.17, p = 0.56; I 2 = 0%).
Finally, a sensitivity analysis excluding Houshyar et al., which was rated at high risk of bias, still demonstrated a statistically significant decreased time to DKA resolution (SMD −0.62, 95% CI: −0.87 to −0.36) (Figure S4).
3.2. Pooled risk ratios: Hypoglycaemia
For hypoglycaemia, meta‐analysis of nine studies9, 11, 19, 20, 21, 22, 23, 24, 25 demonstrated no significant difference between long‐acting and non‐long‐acting regimens (pooled RR: 0.81, 95% CI: 0.52–1.27, p = 0.36; Figure 2). Visual inspection of the funnel plot did not suggest marked asymmetry, with study estimates distributed broadly around the pooled effect size (Figure S5). Egger's regression test was not significant (z = 1.101, p = 0.2709). Heterogeneity across studies was low to moderate (Q (8) = 8.15, p = 0.42; I 2 = 20.1%).
FIGURE 2.
Forest plot of the pooled relative risk of hypoglycaemia with long‐acting versus non‐long‐acting regimens. Individual study estimates with 95% confidence intervals are shown alongside the random‐effects pooled estimate.
In age‐stratified analyses, adults9, 19, 20, 21, 24 (five studies) showed no effect (RR: 1.06, 95% CI: 0.53–2.13, p = 0.87; I 2 = 0%), while paediatric studies11, 22, 25 (three studies) also showed no statistically significant difference (RR: 0.65; 95% CI: 0.29–1.47; p = 0.30).
A sensitivity analysis of three type 1 diabetes mellitus (T1DM)‐only studies11, 20, 25 suggested no significant difference between long‐acting and non‐long‐acting regimens, with a pooled RR of 0.83 (95% CI: 0.29–2.41, p = 0.74) (Figure S6). Heterogeneity was moderate (Q (2) = 3.88, p = 0.14; I 2 = 48.9%).
Based on univariate meta‐regression, baseline pH was a statistically significant moderator for hypoglycaemia (QM = 4.857, p = 0.028), with the baseline pH coefficient on the risk ratio (RR) scale RR = 504.895 (95% CI: 1.992–127996.148, p = 0.028) (Figure S7). Under this model, the intercept was RR = 0.880 (95% CI: 0.592–1.309, p = 0.529), and residual heterogeneity was τ 2 = 0.0000 with I 2 = 0.0%.
3.3. Pooled risk ratios: Hypokalaemia
For hypokalaemia, eight studies were included,9, 11, 20, 21, 22, 23, 24, 25 yielding a pooled RR of 1.21 (95% CI: 0.90–1.63, p = 0.22; Figure 3) with no significant difference between treatment groups. The funnel plot for hypokalaemia appeared visually symmetric, with no clear clustering of smaller studies on one side of the pooled estimate (Figure S8). Egger's test was not significant (z = 1.166, p = 0.2435). Moderate heterogeneity was observed (Q (7) = 6.87, p = 0.44; I 2 = 22.0%).
FIGURE 3.
Forest plot of the pooled relative risk of hypokalaemia with long‐acting versus non‐long‐acting regimens. Individual study estimates with 95% confidence intervals are shown alongside the random‐effects pooled estimate.
In subgroup analyses, adult studies9, 20, 21, 24 (four studies) showed a nonsignificant trend toward higher risk with long‐acting insulin (RR: 1.41, 95% CI: 0.95–2.10, p = 0.09; I 2 = 0%), whereas paediatric studies11, 22, 25 (three studies) showed no difference (RR: 1.06, 95% CI: 0.68–1.66, p = 0.79; I 2 = 42%).
A sensitivity analysis of four T1DM‐only studies11, 20, 22, 25 showed no significant difference in risk between long‐acting and non‐long‐acting regimens (pooled RR = 1.06, 95% CI: 0.72–1.56, p = 0.77) (Figure S9). Heterogeneity was low (Q (3) = 3.34, p = 0.34; I 2 = 28.0%).
3.4. Pooled risk ratios: Rebound hyperglycaemia
For rebound hyperglycaemia, analysis of six studies9, 11, 20, 21, 23, 24 did not demonstrate a statistically significant difference between long‐acting and non‐long‐acting regimens (pooled RR: 0.87; 95% CI: 0.72–1.07; p = 0.18; Figure 4). Heterogeneity was low (Q (5) = 8.36, p = 0.14; I 2 = 28.9%).
FIGURE 4.
Forest plot of the pooled relative risk of rebound hyperglycaemia with long‐acting versus non‐long‐acting regimens. Individual study estimates with 95% confidence intervals are shown alongside the random‐effects pooled estimate.
In stratified analyses, adult studies9, 20, 21, 24 (four studies) showed no significant difference (RR: 0.99, 95% CI: 0.79–1.24, p = 0.95; I 2 = 0%). As only one paediatric‐focused study was available, subgroup analysis was not possible. 11 A sensitivity analysis of two T1DM‐only studies11, 20 suggested a lower risk of rebound hyperglycaemia with early long‐acting insulin, yielding a pooled RR of 0.63 (95% CI: 0.42–0.93, p = 0.019) (Figure S10). Heterogeneity was absent (Q (1) = 0.42, p = 0.51; I 2 = 0%).
Based on univariate meta‐regression, adult versus paediatric age was a statistically significant moderator (QM = 4.156, p = 0.041), with adult indicator RR = 0.574 (95% CI: 0.337–0.979, p = 0.041) and intercept RR = 0.993 (95% CI: 0.792–1.244, p = 0.949) (Figure S11). Heterogeneity was τ 2 = 0.0000 and I 2 = 0.0%.
3.5. Pooled risk ratios: Rebound diabetic ketoacidosis
For rebound diabetic ketoacidosis, three studies9, 20, 21 were included. No statistically significant difference was observed between long‐acting and non‐long‐acting regimens (pooled RR: 0.33; 95% CI: 0.08–1.38; p = 0.13) (Figure S12). Heterogeneity was absent (I 2 = 0%; Q (2) = 0.74, p = 0.69). An age‐stratified analysis was not possible, as the studies all included adults only.
3.6. Standardised mean differences: Total insulin units
For total insulin requirements (IV and SC insulin combined), the use of long‐acting insulin was associated with a significantly lower insulin dose overall across four trials.20, 21, 23, 24 The pooled SMD was −0.62 (95% CI: −1.15 to −0.09, p = 0.022), corresponding to a moderate reduction in total insulin exposure, which translated to approximately several tens of insulin units lower in contributing trials, though absolute differences varied substantially by protocol and population. Heterogeneity was substantial (I 2 = 73.7%, Q (3) = 11.0, p = 0.012), suggesting variability in effect sizes across studies (Figure S13). Among adult studies9, 20, 24 (three studies), the SMD was −0.82 (95% CI: −1.31 to −0.34; p = 0.00086; I 2 = 56%). No paediatric‐focused data was available and therefore could not be pooled.
When excluding Houshyar et al., which was at high risk of bias, there was no statistically significant difference in total IV insulin given (SMD −0.68, 95% CI: −1.40 to 0.04) (Figure S14).
3.7. Standardised mean differences: Intravenous fluids
For IV fluid administration, long‐acting insulin was associated with a modest reduction in total fluid volumes compared with non‐long‐acting insulin across three trials.20, 21, 24 The pooled SMD was −0.33 (95% CI: −0.62 to −0.05, p = 0.022), corresponding to a small absolute reduction in IV fluid volume, generally on the order of several hundred millilitres in contributing studies. No statistical heterogeneity was observed (I 2 = 0.0%, Q (2) = 1.76, p = 0.41) (Figure S15).
3.8. Risk of bias assessment
The Cochrane RoB 2.0 tool was used to evaluate the risk of bias across the included RCTs (Figure S16). The assessment revealed variable quality, with five studies rated overall as having some concerns,19, 20, 21, 22, 25 three as low risk,9, 11, 24 and one as high risk of bias. 23 The most common issues were related to the randomisation process (Domain 1), deviations from intended intervention (Domain 2), and potential selection of reported results (Domain 5), which were frequently judged as having some concerns. For Domain 5, we applied an outcome‐level assessment rather than a study‐level judgment, explicitly evaluating whether prespecified outcomes (e.g., rebound diabetic ketoacidosis) were omitted, incompletely reported, or selectively presented across trials. One study was judged to have a high overall risk of bias, primarily due to concerns in randomisation and measurement of outcomes. 23 In contrast, Bansal et al. demonstrated consistently low risk across all domains. 11 The presence of outcome‐level selective reporting was considered when interpreting pooled estimates for risk of bias, particularly for safety outcomes with fewer contributing studies.
3.9. Certainty of evidence assessment
Because all included studies were RCTs, evidence for each outcome started at high. Evidence was downgraded one level for risk of bias when contributing studies were judged at high risk or had important concerns across multiple RoB 2 domains. For the outcome of time to DKA resolution, certainty was additionally downgraded for indirectness due to variability in outcome definitions across trials, with some studies defining resolution using biochemical criteria alone and others requiring both biochemical and clinical improvement (Table S1). For dichotomous safety outcomes (hypoglycaemia, hypokalaemia, rebound hyperglycaemia), confidence intervals frequently crossed the line of no effect, leading to downgrading for imprecision. Substantial statistical heterogeneity (I 2 ≥ 50%) led to downgrading for inconsistency where present. Funnel plots for hypoglycaemia and hypokalaemia did not suggest marked small‐study effects, though other outcomes had fewer than six studies and were not assessed for publication bias. Overall, certainty ranged from moderate (e.g., time to resolution, IV fluids) to low (e.g., hypoglycaemia, hypokalaemia, rebound hyperglycaemia, total insulin units) with the overall results in the study with corresponding certainty of evidence for each outcome summarised in Table S2.
4. DISCUSSION
In this systematic review and meta‐analysis of nine RCTs encompassing 652 patients, early initiation of long‐acting insulin during IV insulin infusion for DKA was associated with shorter time to DKA resolution and reduced treatment intensity, in the context of variable trial quality and non‐uniform definitions of DKA resolution.
Our findings are consistent with contemporary RCTs and observational studies showing that adding long‐acting insulin (e.g., glargine or degludec) early, typically within the first 1–3 h of diagnosis and concurrent with IV insulin, shortens time to resolution and streamlines care processes.8, 9, 10, 18, 19, 20 Compared with the earlier pooled synthesis that included fewer trials and outcomes, our analysis incorporates more recent randomised data and aligns with the overall direction of benefit previously suggested. 12 These findings also align with the physiological rationale that establishing a basal depot may buffer glycaemic variability during transitions and reduce the need for prolonged IV insulin titration.21, 26, 27, 28
The safety profile observed also mirrors individual trials in which early basal coverage did not increase hypoglycaemia or hypokalaemia despite protocolised overlap with IV insulin.8, 9, 10, 18, 19, 20 Trials similarly did not show a signal for increased serious adverse outcomes, including in‐hospital mortality, when early long‐acting insulin was used.8, 10, 20 Rebound hyperglycaemia appeared lower in some studies and trended down in the pooled analysis, although statistical significance was borderline, suggesting that any benefit for this endpoint may depend on other factors. 9 Given low certainty and limited contributing studies for rebound hyperglycaemia, and even fewer for rebound DKA, these findings should be interpreted as exploratory as we cannot conclude that early basal insulin prevents recurrent hyperglycemia or DKA recurrence. In addition, long‐acting insulin pharmacology may modify treatment effects, as insulin degludec has an ultra‐long, flatter, and less day‐to‐day variable pharmacokinetic and pharmacodynamic profile than insulin glargine (U100/U300), which could plausibly influence transition stability and safety endpoints such as hypoglycaemia or rebound hyperglycaemia in early‐overlap protocols. 26
Renal function varied across the included trials and is an important consideration for interpreting applicability. Approximately two‐thirds of studies excluded patients with advanced or progressive renal failure or end‐stage kidney disease, while permitting inclusion of patients with transient acute kidney injury during DKA or mild‐to‐moderate chronic kidney disease (CKD). Three trials explicitly reported inclusion of patients with baseline CKD, including two that enrolled patients with CKD stages III–V and one in which renal impairment was the target population with basal insulin dosing adjusted by estimated glomerular filtration rate. Because renal dysfunction alters insulin clearance and prolongs insulin action, the pooled findings primarily reflect patients without severe renal disease.
Across the three studies that reported hospital length of stay (LOS) and the two that reported ICU outcomes, early coadministration of a long‐acting insulin alongside IV insulin showed mixed and largely nonsignificant effects on overall hospitalisation, but potential reductions in ICU utilisation in selected populations. In Doshi et al., Houshyar et al., and Thammakosol et al., there was no statistically significant difference in length of hospital stay (LOS) in the concurrent long‐acting insulin group versus IV insulin alone.19, 21, 23 Ammar et al. found a statistically significant shorter ICU LOS in the long‐acting insulin group; however, in Doshi et al., there was no such difference.19, 24
From an implementation perspective, the clinical relevance of an average 3–5 h reduction in trial‐defined time to DKA resolution must be interpreted in the context of real‐world workflow and care delivery. In some settings, modest reductions in resolution time may translate into earlier ICU‐to‐ward transfer, shorter duration of continuous IV insulin titration, or reduced monitoring burden, which could offset the additional coordination required for early basal insulin administration. However, in other contexts, particularly high‐acuity units with frequent staff turnover or less standardised transition protocols, introducing early long‐acting insulin may increase nursing workload, risk communication errors during handoffs, or create confusion around overlapping insulin orders. As such, whether the observed time savings justify adoption of early overlap protocols is likely to be context and institution dependent.
Strengths of this study include the exclusive inclusion of RCTs with evaluation of heterogeneity and small‐study effects. Limitations include clinical heterogeneity in basal insulin formulation and dosing/timing of overlap protocols, variation in timing and dose of long‐acting insulin and overlap with IV insulin, and heterogeneous definitions of DKA resolution. Sensitivity analyses of paediatric trials and T1DM‐only trials were underpowered to support conclusions regarding efficacy in these subgroups. Sensitivity analyses were also considered to explore the impact of heterogeneity in DKA resolution definitions (biochemical‐only versus biochemical plus clinical criteria). However, exclusion of studies based solely on definitional differences would have resulted in small and underpowered subsets. Safety estimates remained imprecise for some endpoints, and publication‐bias assessments were feasible only where study counts permitted. Meta‐regression or protocol‐level subgroup analyses by basal insulin formulation, dose, or timing of overlap could not be performed due to lack of data. Thus, we could not determine which specific basal insulin strategy (agent/dose/timing) drives benefit or whether effects differ meaningfully across protocols.
In conclusion, based on low–moderate certainty evidence across adult and paediatric RCTs, early initiation of long‐acting insulin during IV insulin therapy for DKA was associated with faster biochemical resolution and lower treatment intensity. Future work may prioritise multicentre RCTs with standardised timing, standardised resolution criteria, and stratified analyses by age and diabetes type. In addition, larger paediatric trials are needed before drawing practice‐changing conclusions for children, as current paediatric evidence is limited and contributes substantial imprecision.
FUNDING INFORMATION
This was an unfunded study.
CONFLICT OF INTEREST STATEMENT
The authors declare no conflicts of interest.
ETHICS STATEMENT
As this project involved only publicly available data, ethics approval was not required. The study adheres to the principles of the Helsinki Declaration.
Supporting information
Data S1. Supporting Information.
Data S2. Supporting Information.
Data S3. Supporting Information.
ACKNOWLEDGEMENTS
Not applicable.
DATA AVAILABILITY STATEMENT
The R statistical code and data underlying this study will be available on reasonable request to the corresponding author.
REFERENCES
- 1. Fayfman M, Pasquel FJ, Umpierrez GE. Management of hyperglycemic crises. Med Clin North Am. 2017;101(3):587‐606. doi: 10.1016/j.mcna.2016.12.011 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. American Diabetes Association Professional Practice Committee , ElSayed NA, McCoy RG, et al. Diabetes care in the hospital: standards of care in diabetes—2025. Diabetes Care. 2025;48(Suppl 1):S321‐S334. doi: 10.2337/dc25-S016 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Dhatariya KK, Glaser NS, Codner E, Umpierrez GE. Diabetic ketoacidosis. Nat Rev Dis Primers. 2020;6(1):40. doi: 10.1038/s41572-020-0165-1 [DOI] [PubMed] [Google Scholar]
- 4. Veauthier B, Levy‐Grau B. Diabetic ketoacidosis: evaluation and treatment. Am Fam Physician. 2024;110(5):476‐486. [PubMed] [Google Scholar]
- 5. Malcolm J, Halperin I, Miller DB, et al. In‐hospital management of diabetes. Can J Diabetes. 2018;42:S115‐S123. doi: 10.1016/j.jcjd.2017.10.014 [DOI] [PubMed] [Google Scholar]
- 6. Dhatariya D, Umpierrez GE. Management of diabetes and hyperglycemia in hospitalized patients. In: Feingold KR, Ahmed SF, Anawalt B, eds. Endotext [Internet]. MDText.com, Inc; 2024. [Google Scholar]
- 7. Barski L, Brandstaetter E, Sagy I, Jotkowitz A. Basal insulin for the management of diabetic ketoacidosis. Eur J Intern Med. 2018;47:14‐16. doi: 10.1016/j.ejim.2017.08.025 [DOI] [PubMed] [Google Scholar]
- 8. Mohamed A, Ploetz J, Hamarshi MS. Evaluation of early administration of insulin glargine in the acute management of diabetic ketoacidosis. Curr Diabetes Rev. 2021;17(8):e030221191986. doi: 10.2174/1573399817666210303095633 [DOI] [PubMed] [Google Scholar]
- 9. Thammakosol K, Sriphrapradang C. Effectiveness and safety of early insulin glargine administration in combination with continuous intravenous insulin infusion in the management of diabetic ketoacidosis: a randomized controlled trial. Diabetes Obes Metab. 2023;25(3):815‐822. doi: 10.1111/dom.14929 [DOI] [PubMed] [Google Scholar]
- 10. Do MM, Fleury JA, Morgan GP, et al. Early versus late administration of long‐acting insulin in adult diabetic ketoacidosis. Ann Pharmacother. 2025;59(6):549‐553. doi: 10.1177/10600280241278371 [DOI] [PubMed] [Google Scholar]
- 11. Bansal D, Jayashree M, Nallasamy K, Dayal D, Kakkar A. Insulin glargine supplementation during early management phase in children with diabetic ketoacidosis: a double‐blinded randomised controlled trial. Arch Dis Child. 2025;110(12):983‐987. doi: 10.1136/archdischild-2024-327912 [DOI] [PubMed] [Google Scholar]
- 12. Andrade‐Castellanos CA, Colunga‐Lozano LE. Systematic review with meta‐analysis: subcutaneous insulin glargine coadministration for diabetic ketoacidosis. Gac Med Mex. 2016;152(6):761‐769. [PubMed] [Google Scholar]
- 13. Gilbert BW, Murray DS, Cox TR, He J, Wenski AM, Qualls KE. What are we waiting for? A review of early basal insulin therapy in diabetic ketoacidosis management. Am J Emerg Med. 2023;72:147‐150. doi: 10.1016/j.ajem.2023.07.049 [DOI] [PubMed] [Google Scholar]
- 14. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. doi: 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Brooke BS, Schwartz TA, Pawlik TM. MOOSE reporting guidelines for meta‐analyses of observational studies. JAMA Surg. 2021;156(8):787. doi: 10.1001/jamasurg.2021.0522 [DOI] [PubMed] [Google Scholar]
- 16. Sterne JAC, Savović J, Page MJ, et al. RoB 2: a revised tool for assessing risk of bias in randomised trials. BMJ. 2019;366:l4898. doi: 10.1136/bmj.l4898 [DOI] [PubMed] [Google Scholar]
- 17. McGuinness LA, Higgins JPT. Risk‐of‐bias VISualization (robvis): an R package and shiny web app for visualizing risk‐of‐bias assessments. Res Synth Methods. 2021;12(1):55‐61. doi: 10.1002/jrsm.1411 [DOI] [PubMed] [Google Scholar]
- 18. Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924‐926. doi: 10.1136/bmj.39489.470347.AD [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Doshi P, Potter AJ, De Los Santos D, Banuelos R, Darger BF, Chathampally Y. Prospective randomized trial of insulin glargine in acute management of diabetic ketoacidosis in the emergency department: a pilot study. Acad Emerg Med. 2015;22(6):657‐662. doi: 10.1111/acem.12673 [DOI] [PubMed] [Google Scholar]
- 20. El Feky AY, Abdou MS, Naguib R, Abdelsalam SF, Hamdy M, Elrahmany SM. Effectiveness and safety of early insulin glargine U100 and glargine U300 administration in the management of diabetic ketoacidosis in adults with type 1 diabetes mellitus: a randomized clinical trial. Endocr Pract. 2025;31:1554‐1559. doi: 10.1016/j.eprac.2025.08.006 [DOI] [PubMed] [Google Scholar]
- 21. Thammakosol K, Jantarapootirat M, Traiwanatham S, Sriphrapradang C. Early insulin degludec with continuous intravenous insulin infusion in the management of diabetic ketoacidosis: a randomized controlled trial. Diabetes Obes Metab. 2025;27:7020‐7029. doi: 10.1111/dom.70101 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. El Hawary A, Sobh A, Elsharkawy A, Gamal G, Awad MH. Assessment of the use of long‐acting insulin in management of diabetic ketoacidosis in pediatric patients: a randomized controlled trial. Ann Pediatr Endocrinol Metab. 2025;30(2):95‐101. doi: 10.6065/apem.2448086.043 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Houshyar J, Bahrami A, Aliasgarzadeh A. Effectiveness of insulin glargine on recovery of patients with diabetic ketoacidosis: a randomized controlled trial. J Clin Diagn Res. 2015;9(5):OC01‐OC05. doi: 10.7860/JCDR/2015/12005.5883 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Ammar MA, Aboubakr E, Abdelmoneim W. The outcome of using long‐acting insulin glargine with regular insulin infusion in diabetic ketoacidosis patients with renal impairment: a randomized clinical trial. J Cell Mol Anesth. 2023;8(3):160‐167. doi: 10.22037/jcma.v8i2.39294 [DOI] [Google Scholar]
- 25. Saffari F, Homaei A, Chegini V, Javadi A, Chegini V. Evaluation of the effects of incorporating long‐acting subcutaneous insulin into the standard treatment protocol for diabetic ketoacidosis in children. Int J Endocrinol Metab. 2024;22(1):e139684. doi: 10.5812/ijem-139684 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Owens DR, Bailey TS, Fanelli CG, Yale JF, Bolli GB. Clinical relevance of pharmacokinetic and pharmacodynamic profiles of insulin degludec (100, 200 U/mL) and insulin glargine (100, 300 U/mL)—a review of evidence and clinical interpretation. Diabetes Metab. 2019;45(4):330‐340. doi: 10.1016/j.diabet.2018.11.004 [DOI] [PubMed] [Google Scholar]
- 27. Goldman J, Kapitza C, Pettus J, Heise T. Understanding how pharmacokinetic and pharmacodynamic differences of basal analog insulins influence clinical practice. Curr Med Res Opin. 2017;33(10):1821‐1831. doi: 10.1080/03007995.2017.1335192 [DOI] [PubMed] [Google Scholar]
- 28. Hsia E, Seggelke S, Gibbs J, et al. Subcutaneous administration of glargine to diabetic patients receiving insulin infusion prevents rebound hyperglycemia. J Clin Endocrinol Metab. 2012;97(9):3132‐3137. doi: 10.1210/jc.2012-1244 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data S1. Supporting Information.
Data S2. Supporting Information.
Data S3. Supporting Information.
Data Availability Statement
The R statistical code and data underlying this study will be available on reasonable request to the corresponding author.