Time to resolution of diabetic ketoacidosis in children with type 1 diabetes: a survival analysis of clinical predictors

Abstract

Background

Diabetic ketoacidosis (DKA) is one of the most serious acute complications of diabetes mellitus in children, often leading to severe dehydration, altered consciousness, and death if not promptly managed. The burden of DKA is increasing globally and in Ethiopia, placing substantial strain on pediatric emergency and inpatient care services. Despite its clinical and public health importance, evidence on time to resolution of DKA and its predictors remains limited in Ethiopia, particularly in the northeastern region. This study aimed to assess the time to resolution of DKA and its predictors in children with type 1 diabetes.

Methods

A retrospective follow-up study was conducted using 494 medical records of children with type 1 diabetes mellitus treated at Dessie Comprehensive Specialized Hospital between January 1, 2020, and December 31, 2024. Patient charts were selected using a simple random sampling technique. Data were extracted through a structured checklist based on registry and medical chart reviews. Kaplan–Meier survival analysis was employed to estimate time to resolution from DKA, and differences in survival distributions across categories of explanatory variables were assessed using the log-rank test. Cox proportional hazards regression was applied to identify predictors of time to resolution of DKA. Variables with a p-value < 0.25 in the bivariable analysis were included in the multivariable Cox regression model, adjusted hazard ratio (AHR) with its 95% confidence interval and p-value ≤ 0.05 in the multivariable analysis were considered statistically significant.

Results

A total of 487 children were followed for 12,279 person-hours of observation. Of these, 406 children recovered, yielding a resolution proportion of 83.37% (95% CI: 80.06–86.67), while 81 (16.63%) were censored during the follow-up period. The overall incidence rate of resolution from DKA was 3.30 per 100 person-hours (95% CI: 2.99–3.64), with a median time to resolution of 22 h (95% CI: 18.32–25.67). In the multivariable Cox regression analysis, baseline random blood sugar (RBS) levels > 500 mg/dL (AHR = 0.77; 95% CI: 0.62–0.96), presence of infection (AHR = 0.65; 95% CI: 0.47–0.90), newly diagnosed diabetes mellitus (AHR = 0.79; 95% CI: 0.63–0.99 and DKA duration ≥ 24 h (AHR = 0.08; inverse of < 24 h) were associated with a longer time to resolution of DKA. Conversely, mild DKA (AHR = 1.37; 95% CI: 1.01–1.84) and DKA duration < 24 h (AHR = 11.96; 95% CI: 7.71–18.55) were significantly associated with a shorter time to resolution of DKA.

Conclusion and recommendations

The study identified a relatively prolonged resolution time of DKA among children in the study area. Baseline random blood sugar level > 500 mg/dL, presence of infection and newly diagnosed diabetes had negative relationship (delayed resolution) while mild DKA severity, and duration of DKA < 24 h had positive relationship (faster resolution). These findings highlight the need for healthcare providers and caregivers to address these factors to accelerate resolution and improve clinical outcomes.

Trial registration

Clinical trial number: Not applicable.

Introduction

Diabetes mellitus (DM) refers to a group of metabolic conditions marked by chronic hyperglycemia resulting from disturbances in glucose regulation, including altered gluconeogenesis, glycogenolysis, and impaired cellular glucose utilization [1]. The disease is broadly categorized into type 1 diabetes mellitus (T1DM), type 2 diabetes mellitus (T2DM), gestational diabetes, and other less common specific forms [2]. Type 1 diabetes mellitus develops through autoimmune-mediated destruction of pancreatic β-cells, which ultimately causes absolute insulin deficiency. In contrast, T2DM arises from a complex interaction between inadequate insulin secretion and peripheral insulin resistance. This resistance, particularly in hepatic and muscle tissues, contributes to increased endogenous glucose production, enhanced mobilization of free fatty acids from adipose tissue, and progressive deterioration of β-cell function over time [3, 4].

Diabetic ketoacidosis (DKA) represents an acute and serious metabolic emergency associated with diabetes. It occurs most frequently in individuals with T1DM but is increasingly recognized in those with T2DM as well [5]. Clinically, DKA can lead to profound dehydration, impaired consciousness, and, if untreated, death [6]. The underlying pathophysiology involves a shift toward catabolism, resulting in the breakdown of glycogen stores, lipids, and muscle protein [7].

Pediatric diagnostic criteria outlined by the International Society for Pediatric and Adolescent Diabetes (ISPAD) define DKA by the coexistence of hyperglycemia (glucose > 200 mg/dL), metabolic acidosis (venous pH < 7.3 or serum bicarbonate < 15 mEq/L), and evidence of ketosis, demonstrated either by elevated blood beta-hydroxybutyrate (> 3 mmol/L) or ketonuria [8]. The American Diabetes Association (ADA) further grades the condition according to the severity of acidosis and neurological status, describing mild, moderate, and severe forms based on arterial pH ranges (7.25–7.30; 7.00–7.24; <7.00) together with the patient’s level of alertness [9].

More broadly, widely applied clinical criteria include marked hyperglycemia (> 250 mg/dL), acidemia (pH < 7.3), reduced bicarbonate (< 15 mEq/L), an elevated anion gap (> 12), and significant ketonemia, typically indicated by beta-hydroxybutyrate concentrations above 3 mmol/L [10]. Severity may also be stratified biochemically using venous pH and bicarbonate levels, with progressively lower values corresponding to mild, moderate, and severe DKA [11].

Mortality associated with DKA differs markedly by setting, with reported case-fatality rates of approximately 2–5% in high-income countries compared with 6–24% in resource-limited settings, particularly where diagnosis and treatment are delayed [12]. Episodes are frequently triggered by acute infections or inadequate adherence to insulin therapy, but other stressors such as myocardial infarction, cerebrovascular events, pancreatitis, trauma, burns, surgical procedures and psychiatric conditions, including depression and eating disorders, and the intentional omission of insulin for weight control or self-harm may also precipitate the condition [13].

Appropriate management centered on prompt intravenous fluid replacement, continuous insulin administration, and careful correction of electrolyte disturbances substantially reduces the likelihood of adverse outcomes and supports resolution [14]. Resolution of DKA is generally considered achieved when blood glucose falls below 200 mg/dL in conjunction with improvement in metabolic parameters, typically reflected by at least two of the following: serum bicarbonate ≥ 15 mEq/L, venous pH > 7.3, an anion gap ≤ 12 mEq/L, and blood ketone levels < 0.6 mmol/L [15, 16].

Studies have shown that consistent application of standardized treatment protocols shortens resolution time and lowers complication rates [17, 18]. Under optimal clinical care, most pediatric patients are expected to recover within roughly 24 h [7, 16, 19]. Conversely, prolonged metabolic derangement increases susceptibility to serious complications, including hypoglycemia, hypokalemia, renal impairment, cerebral edema, cardiac arrhythmias, pulmonary edema, and mortality [20–23].

Diabetic Ketoacidosis continues to represent a major reason for hospital admission among children with diabetes worldwide. It is reported at initial presentation in as many as 70% of children newly diagnosed with T1DM, while 1–10% of those with established disease experience at least one episode each year [11]. Considerable geographic variation has been documented. Among newly diagnosed children, DKA has been reported in 35.3% of cases in the United States [24] and 28.7% in India [25]. Even greater heterogeneity is observed within Ethiopia, where studies indicate a wide range of occurrence depending on region and population group. The prevalence of DKA in Ethiopia varies widely, ranging from 21.7% to 78.7%, with higher proportions generally observed among newly diagnosed cases compared to those with established type 1 diabetes, and notable variation across regions [26–31].

Beyond its frequency, DKA contributes substantially to illness severity, risk of death, and financial strain on health systems. Mortality due to DKA varies widely, remaining below 1% in high-income countries but reaching up to 15% or higher in low- and middle-income settings, including Ethiopia. This disparity reflects differences in healthcare access, delayed presentation, limited insulin and diagnostic capacity, and challenges in managing comorbidities. The economic burden also differs substantially; hospital charges per admission in the United States range from approximately $21,000 to $36,000, whereas in Ethiopia, resource constraints significantly affect access to care and treatment outcomes [32–38].

While clinical guidelines suggest that most pediatric DKA cases resolve within approximately 24 h [7, 16, 19, 35, 38], reports from several African settings describe considerably longer resolution periods, including averages of 21 h in South Africa [39] and 59 h in Kenya [40]. Variations in resolution duration have been linked to both disease-related and patient-related factors. These include the severity of acidosis at presentation, serum potassium levels, coexisting illnesses, admission pH, age, timing of hospital presentation, diabetes type, baseline bicarbonate and blood glucose concentrations [39, 41, 42]. Close biochemical monitoring throughout treatment, together with consistent implementation of established hospital management protocols, has been emphasized as an important approach to accelerating metabolic correction and minimizing the risk of complications [41].

Despite numerous studies on prevalence and determinants of DKA in Ethiopia, limited evidence exists on median resolution time and its predictors among children. Therefore, this study aimed to estimate the median time to resolution of DKA and identify its predictors in children with T1DM.

Methods

Study settings

The study was conducted at Dessie Comprehensive Specialized Hospital (DCSH), located in Dessie City Administration, South Wollo Zone, Amhara Region, Ethiopia. Dessie city is situated approximately 401 km northeast of Addis Ababa, the capital city of Ethiopia. Dessie Comprehensive Specialized Hospital serves as a major referral hospital for health facilities in South and North Wollo zones and surrounding regions. The hospital provides preventive, curative, and rehabilitative services to an estimated population of more than 10 million people and is staffed by over 865 employees, including administrative and healthcare professionals. Data collection was carried out from March 20 to April 20, 2025.

Study design

An institution-based retrospective follow-up study design was employed.

Population

Study population

The study population included all children who were diagnosed with T1DM, admitted and managed for DKA at DCSH between January 1, 2020, and December 31, 2024.

Source population

The source population comprised all children who were diagnosed with T1DM, admitted and treated for DKA at DCSH.

Eligibility criteria

Inclusion

All children aged 1 month old to 15 years old who were diagnosed with T1DM, admitted and treated with DKA at DCSH. Participants were eligible if they were within the defined pediatric age range and had a confirmed clinical diagnosis of T1DM based on medical records.

Exclusion

Medical charts with incomplete information like age, sex, clinical diagnosis of T1DM, blood glucose levels, ketone status, or treatment outcomes for children aged 1 month old to 15 years old admitted for DKA management were excluded.

Study variables

Dependent variable

Time to resolution of DKA in children with T1DM. The time interval, measured in hours, from initiation of vascular fluid resuscitation for DKA management in the emergency department or intensive care unit to documented resolution of DKA.

Independent variables

Socio-demographic factors: age (1 month to 5 years, 6–10 years, 11–15 years), sex (Female, Male), residence (urban, rural); biochemicalfactors (admission pH, potassium level, blood urea, serum creatinine, hemoglobin level, initial blood glucose level, urine ketone); clinicalcharacteristics: severity of DKA (mild, moderate, severe), presence of preceding infection (yes, no), comorbidities (yes, no), duration ofdiabetes mellitus in years (newly diagnosed, ≤ 5, >5), vital signs at presentation, frequency of diabetes follow-up); and treatment-relatedfactors (treatment regimen, amount of intravenous fluid administered, potassium supplementation, initiation time of management, and useof glucocorticoid medications).

Operational definitions

Event

Children who recovered from DKA during the study period before hospital discharge.

Censored

Children who were referred to another facility, died, or were discharged for any reason before resolution from DKA during the study period.

Diabetic ketoacidosis

A clinical diagnosis of DKA defined by random blood glucose > 250 mg/dL, urine ketones ≥ + 2, arterial pH < 7.3, and serum bicarbonate < 15 mEq/L [43].

Resolution of DKA

Achievement of blood glucose < 200 mg/dL with a negative urine ketone dipstick on at least two consecutive measurements taken two hours apart following treatment initiation [44].

DKA resolution time

The time interval, measured in hours, from initiation of vascular fluid resuscitation for DKA management in the emergency department or intensive care unit to documented resolution of DKA.

Incomplete data

Medical records lacking documentation of negative urine ketone results and/or initiation time of DKA management.

Severity of DKA

DKA severity was classified according to the International Society for Pediatric and Adolescent Diabetes (ISPAD) criteria: Mild (pH 7.20–7.29, bicarbonate 10–15 mmol/L); Moderate (pH 7.10–7.19, bicarbonate 5–9.9 mmol/L); Severe (pH < 7.10, bicarbonate < 5 mmol/L).

Sampling technique, procedure and sample size determination

The sample size for the primary objective was calculated using the Cox proportional hazards model in STATA version 14, considering a hazard ratio of 0.429 for increased PCO₂ from a study conducted in pediatric intensive care units in Turkey [45], probability of event of 0.5, withdrawal proportion of 0.1, 80% power, and a 95% confidence level, yielding a minimum sample size of 98.

For predictors, the sample size was determined using the double population proportion formula in Epi Info version 7.2.3.1, assuming a two-sided significance level of 5%, 80% power, 95% confidence level, and a 1:1 ratio of exposed to non-exposed groups. Various predictors were considered, including severity of DKA and duration of diabetes mellitus [46–48]. The largest calculated sample size was 494, which was taken as the final sample size.

Medical records of children admitted with DKA at DCSH from January 1, 2020, to December 31, 2024, were identified using medical registration numbers obtained from registration books. Charts meeting the inclusion criteria were listed, while missing and incomplete records were excluded. The final sample of 494 medical charts was selected using a simple random sampling technique generated through EpiInfo software version 7.2, and data were extracted from the selected charts.

Data collection procedure and quality assurance

Data were collected using a structured extraction checklist adapted from previously published studies [5, 47, 48]. The tool captured socio-demographic, biochemical, clinical, and treatment-related variables. Patient records were retrieved using unique registration numbers. Data extraction was performed by two trained BSc nurses and supervised by one MSc nurse.

The data collection tool was pretested on 5% (25 charts) of the calculated sample size among patients admitted before January 1, 2020, at the emergency department of the same hospital. Necessary modifications were made based on the pretest findings. Data collectors received training on study objectives and procedures. Completed extraction forms were checked daily for completeness and consistency by data collectors, supervisors, and the principal investigator.

Data processing and analysis

Data were entered, cleaned, and analyzed using STATA 14 version statistical software. Descriptive statistics were used to summarize continuous variables using median and interquartile range for skewed data and mean and standard deviation for symmetric data while frequencies and proportions were used for categorical variables. The outcome variable was dichotomized as resolution (event = 1) or censored (event = 0). Time to resolution of DKA was calculated in hours from initiation of DKA management to resolution.

Kaplan–Meier survival analysis was used to estimate time to resolution from DKA, and differences between groups were assessed using the log-rank test. Model selection was guided by Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC). Multicollinearity was assessed using variance inflation factor (VIF) with a threshold of < 10 acceptable level of collinearity. All variables included in the final model were yielded VIF value below the threshold. Bivariable and multivariable Cox proportional hazards regression models were fitted to identify predictors of time to DKA resolution. Variables with p-values < 0.25 in bivariable analysis were included in the multivariable model. This relatively liberal cutoff was intentionally used to avoid excluding potentially important predictors and confounders at an early stage, as variables that are not statistically significant in bivariable analysis may become significant after adjustment. This approach is widely recommended in regression modeling.

Model assumptions were assessed using log-minus-log plots, Cox–Snell residuals, and global tests. Adjusted hazard ratios (AHR) with 95% confidence intervals and p-values ≤ 0.05 were used to determine statistical significance. Results were presented using text, tables, and figures.

Ethical considerations

Ethical approval was obtained from the Ethical Review Committee of Zemen Postgraduate College of Public Health Department. An official letter of cooperation was secured from Dessie Comprehensive Specialized Hospital. As the study was retrospective, informed consent was waived. Confidentiality of patient information was strictly maintained throughout the study. Furthermore, our research was conducted in accordance with the Declaration of Helsinki.

Results

Socio-demographic characteristics

A total of 494 medical were selected initially. Of these, 7 records were excluded due to incomplete or missing key variables, including essential information on clinical characteristics and outcome measures required for the analysis. After exclusion, 487 records with complete information were included in the final analysis. More than half of the participants were female (284, 58.3%). The mean (SD) age of the children was 6.71 (3.93) years, with ages ranging from 1 to 14 years. Nearly half of the study participants (242, 49.7%) were urban residents. The most common clinical presentations among children with DKA were Kussmaul respiration (382, 78.4%), vomiting (330, 67.8%), and nausea (270, 55.4%). The mean (SD) random blood sugar level of the participants was 499.67 (81.11) mg/dL (Table 1).

Table 1.

Socio-demographic and baseline clinical characteristics among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Variables		Frequency	Percentage
Sex	Female	284	58.3
	Male	203	41.7
Age (years)	< 5 years	114	23.4
	6–10 years	202	41.5
	11–15 years	171	35.1
	Mean age (SD)	6.71 (3.93)
Residence	Urban	242	49.7
	Rural	245	50.3
Nausea	Yes	217	44.6
	No	270	55.4
Abdominal pain	Yes	153	31.4
	No	334	68.6
Fatigue	Yes	231	47.4
	No	256	52.6
Classic osmotic symptoms including polyuria, polydipsia and polydipsia	Yes	348	71.5
	No	139	28.5
Kussmaul respiration	Yes	382	78.4
	No	105	21.6
Loss of consciousness	Yes	100	20.5
	No	387	79.5
Vomiting	Yes	330	67.8
	No	157	32.2
RBS	> 500	169	34.7
	< 500	318	65.3
	Mean (SD)	499.67 (81.11)
WBC	> 10 × 103	62	12.7
	< 10 × 103	425	87.3
	Mean (SD)	8237.5(4122.98)

Among the total 487 study participants only 83 (17%) had comorbidity. The most common comorbidity of child DKA patients had acute kidney injury (AKI) 38 (7.8%), HTN 13 (2.7%), and Asthma 12 (2.5%) followed by cardiovascular disease (CVD) 10 (2.1%) and chronic kidney disease (CKD) 10 (2.1%) respectively (Fig. 1).

Fig. 1.

Types of DM comorbidities among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Time to resolution of children from DKA

Children with a diabetes duration of more than 5 years had a median resolution time from DKA of 19 h. The Kaplan–Meier estimator curve (Table 4) shows that children with longer diabetes duration experienced faster resolution. The log-rank test indicated that this difference in DKA resolution time was statistically significant (p < 0.001) (Table 2).

Table 4.

Goodness-of-fit test is assessing proportional hazards assumption

Predictors	rho	chi2	Df	P-value
1b.Age
2.Age	0.11221	5.48	1	0.1930
3.Age	0.08781	3.40	1	0.0652
1b.Baseline RBS
2.Baseline RBS	0.05665	1.40	1	0.2362
1b.Diabetes duration
2.Diabetes duration	0.00825	0.03	1	0.8640
3.Diabetes duration	-0.03526	0.53	1	0.4660
1b.Presence of Infection
2.Presence of Infection	-0.03203	0.42	1	0.5147
1b.Comorbidity of DM
2.Comorbidity of DM	0.01611	0.12	1	0.7336
1b.Severity of DKA
2.Severity of DKA	0.01293	0.07	1	0.7889
3.Severity of DKA	0.03590	0.53	1	0.4654
1b.Previous Hx DM
2.Previous Hx DM	0.04018	0.66	1	0.4182
1b.Take Corticosteroids
2.Take Corticosteroids	-0.04702	0.91	1	0.3413
1b.Hx of drug non compliance
2.Hx of drug non compliance	0.09437	3.72	1	0.0538
1b.Duration of DKA
2.Duration of DKA	0.02326	0.22	1	0.6370
3.Duration of DKA	0.00922	0.04	1	0.8498
4.Duration of DKA	0.07807	2.57	1	0.1086
global test		16.69	13	0.2142

Table 2.

The median time to resolution, and comparison of DKA resolution time among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Variables	Category	Survival status of DKA		Median resolution time in hour	Chi-square	Log-rank test p-Value
		Recovered No (%)	Censored No (%)
Diabetes Duration	Newly Dx	123(25.3)	25(5.1)	28	21.59	< 0.001
	≤ 5 years	57(11.7)	12(2.5)	28
	> 5 years	226(46.4)	44(9.0)	19
Comorbidity of DM	Yes	50(10.3)	33(6.8)	42.420	7.61	0.006
	No	356(73.1)	48(9.9)	30.188
history of DKA episodes	Yes	208(42.7)	48(9.9)	32.23	0.08	0.77
	No	198(40.7)	33(6.8)	31.53
Severity of DKA	Mild	210(43.1)	30(6.2)	27.03	15.61	< 0.001
	Moderate	148(30.4)	31(6.4)	33.96
	Severe	48(9.9)	20(4.1)	41.56
Duration of DKA	<=24 h	176(36.1)	23(4.7)	16.051	274.21	< 0.001
	25–48 h	104(21.4)	21(4.3)	38.538
	49–96 h	59(12.1)	18(3.7)	43.915
	≥ 97 h	67(13.8)	19(3.9)	51.403
Infection	Yes	46(9.4)	16(3.3)		6.36	0.012
	No	360(73.9)	65(13.3)

The incidence rate of resolution from DKA

A total of 487 children were followed for 12,279 person-hours of risk time, of whom 406 (83.37%, 95% CI: 80.06–86.67) recovered, while the remaining 81 (16.63%) were censored during the study period. The incidence rate of resolution from DKA was 3.30 per 100 person-hours of observation (95% CI: 2.99–3.64), with a median resolution time of 22 h (95% CI: 18.32–25.67). The minimum and maximum lengths of follow-up were 4 and 92 h, respectively. The DKA-free survival time was estimated using the Kaplan–Meier survival curve, which showed a rapid decline within the first 22 h, indicating that most children recovered from DKA within this period (Fig. 2).

Fig. 2.

Overall Kaplan –Meier estimation of time to resolution from DKA among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Comparison of survival status

The log-rank test was used to compare resolution time from DKA across categories of different predictors. This analysis revealed that survival times differed significantly between groups based on baseline RBS level, presence of infection, severity of DKA, and previous history of diabetes mellitus (Figs. 3, 4, 5 and 6).

Fig. 3.

Kaplan-Meier estimator curve for the DKA resolution time on baseline RBS among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Fig. 4.

Kaplan-Meier estimator curve for the DKA resolution time on presence of infection among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Fig. 5.

Kaplan-Meier estimator curve for the DKA resolution time on severity of DKA among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Fig. 6.

Kaplan-Meier estimator curve for the DKA resolution time on previous DM history among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Predictors of resolution time from DKA

In the multivariable Cox proportional hazards analysis, five of these ten variables remained significantly associated with time to resolution from DKA at the 5% level of significance. The significant predictors were: baseline RBS [AHR = 0.77; 95% CI: 0.62–0.96], presence of infection [AHR = 0.65; 95% CI: 0.47–0.90], severity of DKA [AHR = 1.37; 95% CI: 1.01–1.84], previous history of diabetes mellitus [AHR = 0.79; 95% CI: 0.63–0.99], and duration of DKA [AHR = 11.96; 95% CI: 7.71–18.55] (Table 3).

Table 3.

predictors of time to resolution from DKA among children with type 1 diabetes at DCSH, Amhara Region, Ethiopia, 2025 (n = 487)

Variables		Survival status of DKA		CHR(95%CI)	AHR(95%CI)
		Recovered	Censored
Age	< 5year	137	34	0.96(0.73,1.25)	1.21(0.91,1.61)
	6–10 year	174	28	0.84(0.65,1.08)	1.02(0.78,1.33)
	11-15year	95	19	1	1
Baseline RBS	≥ 500	143	26	0.85(0.69,1.05)	0.77(0.62,0.96)*
	< 500	263	55	1	1
Diabetes duration	Newly Dx	123	25	0.64(0.51,0.80)	0.76(0.57,1.00)
	≤ 5 years	57	12	0.63(0.47,0.84)	0.83(0.60,1.15)
	> 5 years	226	44	1	1
Presence of Infection	Yes	46	16	0.67(0.49,0.92)	0.65(0.47,0.90)*
	No	360	65	1	1
Comorbidity of DM	Yes	50	33	0.66(0.49,0.89)	0.99(0.72,1.36)
	No	356	48	1	1
Severity of DKA	Mild	210	30	1	1.37(1.01,1.84)*
	Moderate	148	31	0.72(0.58,0.89)	1.14(0.83,1.55)
	Severe	48	20	0.71(0.52,0.98)	1
Previous Hx DM	No	198	33	0.76(0.62,0.93)	0.79(0.63,0.99)*
	Yes	208	48	1	1
Take Corticosteroids	Yes	102	35	0.82(0.66,1.03)	1.07(0.79,1.43)
	No	304	46	1	1
Hx of drug non compliance	No	233	26	1.43(1.17,1.75)	1.10(0.89,1.36)
	Yes	173	55	1	1
Duration of DKA	<=24 h	176	23	12.82(8.43,19.50)	11.96(7.71,18.55)**
	24–48 h	104	21	1.65(1.15,2.37)	1.58(1.08,2.31)*
	49–96 h	59	18	1.06(0.73,1.54)	1.03(0.70,1.52)
	> 97 h	67	19	1	1

Model fitness statistics

The overall model fitness was assessed using the Cox proportional hazards regression model. In the present study, the hazard function closely followed the 45° line, indicating that the Cox model fits the data reasonably well. The Cox–Snell residuals test further confirmed the goodness of fit, demonstrating that the model adequately describes the data (Fig. 7).

Fig. 7.

Cox Snell residual test showing overall goodness of fit of the Cox proportional hazards model

Test of proportional-hazards assumption

The findings indicated that all covariates included in the model satisfied the proportional hazards (PH) assumption (p > 0.05), indicating that the model provided an adequate fit to the data (Table 4).

Discussion

This study was conducted to assess the time to resolution of DKA and its predictors among children with T1DM at DCSH. A total of 487 children were followed for 12,279 person-hours of observation, of whom 406 (83.37%) (95% CI: 80.06–86.67) recovered during the study period, while 81 (16.63%) were censored. Baseline random blood sugar (RBS), presence of infection, severity of DKA, previous history of diabetes mellitus, and duration of DKA prior to treatment initiation were identified as significant predictors of time to resolution of DKA.

The incidence rate of resolution of DKA was 3.30 per 100 person-hours of observation (95% CI: 2.99–3.64), with a median resolution time of 22 h (95% CI: 18.32–25.67). This finding is consistent with a study conducted in Iran, which reported a median resolution time of 21 h [49]. However, resolution time in the present study was shorter than reports from India (26 h) [50], Indonesia (28.8 h) [51], and China (41.72 h) [52]. In contrast, it was longer than findings from the United States (8.4 h) [53], Japan (10–11 h) [54], and Turkey (14.30 h) [45]. These discrepancies may be attributed to variations in case ascertainment methods, timing of DKA management initiation, and methodological differences such as study design, sample size, and study period. Socioeconomic and cultural differences between study populations may also contribute to the observed variation in resolution times. Additionally, the observed variation may also be explained by differences in healthcare system capacity, including availability of trained healthcare professionals, timely diagnosis, access to essential medications such as insulin, and adequacy of monitoring and supportive care during DKA management. In resource-limited settings, constraints in these areas may delay treatment initiation and compromise optimal management, potentially leading to poorer outcomes. Evidence suggests that mortality and resolution outcomes from diabetic ketoacidosis are substantially better in well-resourced settings due to standardized protocols and advanced supportive care, whereas limited infrastructure and resource shortages contribute to worse outcomes in low- and middle-income countries. These findings highlight the need to strengthen health system capacity, improve access to standardized treatment protocols, and ensure availability of essential supplies to enhance the management and prognosis of children with diabetic ketoacidosis.

Baseline hyperglycemia was significantly associated with delayed resolution time of DKA. Children presenting with baseline RBS levels greater than 500 mg/dL had a 23% lower rate of resolution compared with those with RBS levels below 500 mg/dL. This finding is supported by a study conducted in Turkey, which identified initial blood glucose level as a significant predictor of rapid DKA resolution [45]. Extremely high blood glucose levels may lead to severe dehydration, hypotension, and impaired central nervous system function, including hyperglycemic coma, thereby prolonging resolution from DKA [44]. Healthcare professionals should closely monitor key clinical and biochemical factors associated with prolonged resolution from DKA and provide continuous patient education on the importance of timely healthcare seeking during illness. Early identification of high-risk patients and the implementation of targeted interventions are essential to improve resolution outcomes.

Severity of DKA was also an important predictor of resolution time of DKA. Children with mild DKA recovered 1.37 times faster than those with severe DKA. This finding is consistent with studies conducted in the United States [53] and China [55, 56]. The delayed resolution observed among patients with severe DKA may be explained by greater metabolic derangements and electrolyte abnormalities compared with mild forms of the disease [19]. This is further supported by evidence from Israel, which reported a negative correlation between higher admission levels of bicarbonate, potassium, phosphorus, and pH and resolution time from DKA [57]. Patients with diabetes should seek immediate medical care when they become ill or experience symptoms suggestive of metabolic decompensation, to facilitate early diagnosis and management of DKA.

The presence of infection significantly delayed resolution of DKA. Children with concurrent infections had a 35% lower hazard of resolution compared with those without infection. This finding aligns with studies conducted in Japan [54] and India [58, 59]. Infection-related inflammation, proinflammatory cytokine release, and increased secretion of counter-regulatory hormones can exacerbate insulin resistance and metabolic deterioration, thereby prolonging DKA resolution [28].

Additionally, resolution time was delayed by 21% among children with newly diagnosed diabetes mellitus compared with those with established diabetes. This finding is consistent with studies from Colombia [60], Japan [54], Turkey [45], and Israel [57]. Children with newly diagnosed diabetes may have limited awareness of diabetic symptoms and disease management [47], leading to delayed presentation and more severe metabolic derangements. Prolonged hyperglycemia may also result in increased insulin resistance and glucose toxicity, further extending the resolution period [48, 61].

Finally, the duration of DKA prior to initiation of treatment was a critical predictor of resolution time. Patients who received treatment within ≤ 24 h and 24–48 h recovered 11.96 and 1.58 times faster, respectively, compared with those who initiated treatment after ≥ 97 h. This finding is consistent with a study conducted in Kenya, which reported that delays in the initiation of management contributed to prolonged DKA resolution times [40]. Delayed recognition of hyperglycemic symptoms increases the risk of severe DKA and prolongs resolution [61]. Moreover, prolonged hyperglycemia may exacerbate insulin resistance and glucose toxicity, further delaying resolution [43]. Timely diagnosis and early initiation of appropriate DKA management have been shown to be associated with faster resolution [7].

Recent advances in diabetes monitoring and prediction have increasingly emphasized the importance of glucose dynamics and continuous glucose monitoring (CGM)-derived metrics in identifying early metabolic deterioration. Emerging evidence suggests that CGM-derived indicators, including glycemic variability and entropy-based glucose dynamics measures, may provide additional insights beyond conventional static glucose measurements by capturing subtle fluctuations in metabolic status [62]. Furthermore, data-driven and machine learning approaches have shown promising potential for predicting metabolic deterioration using longitudinal glucose patterns, oral glucose tolerance test trajectories, and CGM-derived temporal profiles. These approaches may improve early identification of high-risk patients and support more individualized diabetes management strategies [63]. Although such advanced technologies and analytical methods were beyond the scope and resource setting of the current study, particularly in low-resource settings such as Ethiopia, their growing clinical relevance warrants consideration in future research and diabetes care frameworks.

Limitations and strengths of the study

This study has some limitations that should be acknowledged. First, the retrospective nature of the study, which relied on medical chart reviews, resulted in missing data due to incomplete or inadequate documentation. This may have affected the completeness and accuracy of some variables. Second, the study did not assess certain important clinical parameters, particularly admission arterial pH and serum bicarbonate levels, which are known potential predictors of resolution outcomes in DKA. The omission of these variables was mainly due to limitations in the availability of advanced clinical and laboratory investigations in the study setting. These limitations should be considered when interpreting the findings of the study.

This study has several notable strengths. First, it employed a relatively large sample size with an adequate follow-up period, allowing for a robust estimation of time to resolution from diabetic ketoacidosis (DKA) and enhancing the statistical power of the analysis. Second, the use of time-to-event (survival) analysis enabled a more appropriate assessment of resolution time and its predictors while accounting for censoring, thereby providing more reliable and clinically meaningful estimates. Third, the study identified multiple important clinical and contextual factors associated with DKA resolution, offering comprehensive insight into determinants of prolonged resolution in pediatric patients. Finally, the use of routinely collected clinical data reflects real-world practice, increasing the generalizability of the findings to similar healthcare settings, particularly in resource-limited contexts.

We also acknowledge the important limitations of urine ketone testing. First, urine ketone levels may not accurately reflect the patient’s current metabolic status because they represent ketone accumulation in the bladder since the last void. Second, severe dehydration may delay sample collection. Third, false-positive or false-negative results may occur due to medication interference, highly acidic urine, or degraded strips. Furthermore, during treatment, β-hydroxybutyrate is converted back to acetoacetate, which may paradoxically increase urine ketone positivity despite clinical improvement.

Accordingly, we recognize that using urine ketone negativity as a criterion for DKA resolution may limit direct comparability with studies that used standard biochemical criteria and may influence the interpretation of recovery time. We have now clarified this rationale and added the corresponding limitation in the revised manuscript.

Conclusion

This study demonstrated that the median time to resolution from diabetic ketoacidosis (DKA) was longer than that reported in most previous studies. Several clinically important factors were identified as being significantly associated with prolonged resolution time, including higher baseline random blood sugar levels, the presence of infection, greater severity of DKA at admission, a previous history of diabetes mellitus, and longer duration of DKA prior to presentation. These findings highlight the need for early detection, prompt management of precipitating infections, and close monitoring of high-risk patients to improve resolution outcomes from DKA.

Health facilities should strengthen the implementation of standardized DKA management protocols, promote patient-centered diabetes education, and establish structured follow-up programs. These measures may contribute to reducing DKA-related morbidity and mortality.

Future research

Further studies, particularly prospective follow-up designs, are recommended to incorporate additional potential predictors such as admission arterial pH and serum bicarbonate levels. This would enhance understanding of factors influencing resolution from DKA and inform evidence-based clinical practice.

Abbreviations

ADA

American Diabetes Association

AHR

Adjusted hazard ratio

BG

Blood Glucose

CE

Cerebral Edema

CI

Confidence Interval

CKD

Chronic Kidney Disease

CVD

Cardio Vascular Disorder

DKA

Diabetic Ketoacidosis

DM

Diabetes Mellitus

HCO3

Bicarbonate

HTN

Hypertension

ISPAD

International Society for Pediatric and Adolescent Diabetes

IV

Intravenous

LMICs

Low- and Middle-Income Countries

NaCl

Sodium Chloride

NT1D

Newly Diagnosed Type 1 Diabetes

PT1D

Previously Diagnosed Type 1Diabetes

RBS

Random Blood Sugar

T1DM

Type One Diabetes Mellitus

T2DM

Type Two Diabetes Mellitus

TTR

Time to Resolution

USA

United States of America

VBG

Venous Blood Gas

WHO

World Health Organization

Author contributions

All authors contributed substantially to the conception, execution, and completion of this study. AKA, AMM, RAM and AM: contribute to the conception or design of the work; AKA, AMM, RAM and AM: contribute to the acquisition, analysis and interpretation of data; AKA, AMM, RAM and AM: have drafted the work or substantively revised it. All authors read it, revised it critically for important intellectual content, and gave final approval to the submitted manuscript version. All authors agreed to each contribution and the integrity of any part of the work.

Funding

The authors declare that no financial funding was obtained to support the conduct of this research or the publication of this article.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. Data sharing is subject to ethical approval and institutional regulations to protect patient confidentiality.

Declarations

Ethical approval and consent to participate

Ethical approval was obtained from the Ethical Review Committee of Zemen Postgraduate College of Public Health Department. An official letter of cooperation was secured from Dessie Comprehensive Specialized Hospital. As the study was retrospective, informed consent was waived. Confidentiality of patient information was strictly maintained throughout the study. Furthermore, our research was conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable as there is no image or other confidentiality related issues.

Competing interests

The authors declare no competing interests.