Incidence and predictors of diabetic ketoacidosis among adults with type 1 diabetes in Northwest, Ethiopia, 2023: a retrospective cohort study

Abstract

Background

Diabetic ketoacidosis (DKA) is a serious, potentially fatal diabetes complication, particularly in low-resource settings, marked by hyperglycemia, ketoacidosis, and ketonuria. Mortality from DKA remains high in developing countries, including Ethiopia. Despite its impact, data on DKA’s incidence and predictors are limited in Ethiopia.

Objectives

To determine the incidence and predictors of DKA among adults with type 1 diabetes in Amhara regional state hospitals, 2023.

Methods

A five-year retrospective cohort study was conducted, analyzing records from September 11, 2017, to September 10, 2022. Data extraction occurred from April 12 to May 28, 2023, using simple random sampling to select 634 participant records. Data were entered and analyzed with EPI data version 4.6.0.6 and STATA version 17.

Results

Among participants, 240 (39.8%) developed DKA, with an incidence rate of 15.96 per 100 person-years (95% CI: 14.04, 18.12). Significant predictors of DKA included middle age (AHR: 0.35, 95% CI: 0.21, 0.58), sex (AHR: 0.72, 95% CI: 0.55, 0.95), family history of diabetes (AHR: 1.45, 95% CI: 1.01, 2.11), long disease duration (AHR: 13.30, 95% CI: 9.09, 19.46), poor glycemic control (AHR: 2.29, 95% CI: 1.70, 3.08), recent acute illness (AHR: 1.41, 95% CI: 1.06, 1.87), comorbidities (AHR: 1.78, 95% CI: 1.34, 2.38), and medication noncompliance (AHR: 2.35, 95% CI: 1.76, 3.15).

Conclusions

DKA incidence remains high in this cohort. Interventions should focus on improving glycemic control, medication compliance, and managing comorbidities to reduce DKA risk. Additionally, education targeting those with long disease duration and a family history of diabetes is recommended to prevent DKA.

Introduction

Diabetes Mellitus (DM) is the collective term for heterogeneous metabolic disorders, and its main finding is chronic hyperglycemia, and the cause is either impaired insulin secretion or action or usually both [1, 2]. Type 1 diabetes (T1DM) and type 2 diabetes (T2DM) are the two main types [3]. T1DM is a chronic autoimmune disease marked by inadequate T cell-mediated β cell destruction in the pancreas, leading to an inadequate supply of insulin and hyperglycemia [4, 5]. According to the WHO report, in 2014 about 422 million people worldwide had DM, the majority living in low- and middle-income countries, and in 2019, DM was the direct cause of 1.5 million deaths, and 48% of all deaths due to DM occurred before the age of 70 years, and in 2017 there were 9 million people with T1DM [6].

Diabetic ketoacidosis (DKA) is the most common acute complication form for T1DM and is a serious and potentially fatal complication of DM, characterized by hyperglycemia, ketoacidosis, and ketonuria [7, 8, 9]. It is associated with high mortality rates in the developing world [10]. A global systematic review showed incidence rates (IR) ranging from 0/1000 to 263/1000 person-years in eight different countries [11]. Some national economies are experiencing a health care cost crisis as a result of the burden of DKA [12, 13]. DKA mortality is high in Sub-Saharan Africa [14], particularly in Ethiopia, where there is an 11%-12% cumulative mortality rate among hospitalized DM patients [15, 16]. In Ethiopia, access to insulin and specialized diabetes care remains limited compared to other regions, which may contribute to the higher incidence of DKA [17]. A shortage of endocrinologists and diabetologists further complicates diabetes management, emphasizing the need for a greater understanding of local risk factors for DKA development [17].

Poor public health care, inadequate health education, and poor self-glycemic control are all major contributors to the expanding DKA burden in Sub-Saharan nations, including Ethiopia [18, 19]. Generally, socio-demographic, behavioral, clinical, and therapeutic aspects constitute DKA’s possible predictors [19, 20]. It has not been fully understood why some T1DM individuals develop DKA while others do not; it is also uncertain if the development of DKA results from delayed diagnosis, treatment, and other predictors [21].

However, the incidence and predictors of DKA among adult T1DM patients are not well studied in Ethiopia [22]. Understanding DKA predictors, early detection, and therapy is crucial for reducing its incidence. This knowledge can contribute to developing effective treatments and decreasing the rate of DKA in low-income countries like Ethiopia, where limited understanding of the condition persists. This study investigated the incidence and predictors of DKA in the Amhara regional state’s comprehensive specialized hospitals.

Methods and materials

Study design, period and setting

A five-year institution-based retrospective cohort study of patients with Type 1 diabetes seen between September 11, 2017, and September 10, 2022, was conducted, with data extraction taking place from April 12, 2023, to May 28, 2023. The study was carried out in eight comprehensive, specialized hospitals located in the Amhara regional state of Ethiopia. These hospitals include the University of Gondar Comprehensive Specialized Hospitals (UoGCSH), Felegehiwot Comprehensive Specialized Hospitals (FHCSH), Tibebegion Comprehensive Specialized Hospitals (TGCSH), Debre Markos Comprehensive Specialized Hospitals (DMCSH), Debre Tabor Comprehensive Specialized Hospitals (DTCSH), Dessie Comprehensive Specialized Hospitals (DCSRH), Debrebrihan Comprehensive Specialized Hospitals (DBCSH), and Woldiya Comprehensive Specialized Hospitals (WCSH). These hospitals offer a range of medical, surgical, pediatric, maternal, and other types of patient care services.

Source and study population

The source population consisted of all adults (age ≥ 18 years old) diagnosed with T1DM and receiving T1DM follow-up care in the comprehensive specialized hospitals of the Amhara regional state. The study population included each adult diagnosed with T1DM who had T1DM follow-up in the selected comprehensive specialized hospitals of the Amhara regional state.

Inclusion and exclusion criteria

All adults aged ≥ 18 years old diagnosed with T1DM and receiving T1DM follow-up care in the selected comprehensive specialized hospitals of the Amhara regional state between September 11, 2017, and September 10, 2022, were included. However, individuals who presented with DKA at the time of their initial T1DM diagnosis were fully excluded from the study to focus on incident cases of DKA among established T1DM patients.

Sample size determination and sampling procedure

The sample size was calculated by using the single population proportion formula with considerations of the following statistical assumptions: a 50% proportion of DKA (p = 0.5 and q = 0.5), a 95% confidence level, and a 5% margin of error.

Where; ni = initial sample size, Z = 1.96 the corresponding Z-score for the 95% CI, α = confidence interval (95%).

p = proportions of DKA, 0.5.

ni = ((1.96)². (0.5) (0.5))/ (0.05)²= 384.

n _total = ni×Design effect (1.5) = 576 and by considering lost and incomplete patient records, 10% of the initial sample size was added and the final sample size was 634.

Sampling procedure

Four hospitals (UoGCSH, FHCSH, DMCSH, and DTCSH) were randomly selected from eight hospitals. The total caseload of T1DM patients from September 11, 2017, to September 10, 2022, was assessed by reviewing the follow-up forms in the discharge catalog of the inpatient wards, emergency department, and outpatient department of each hospital. Proportional allocation was then performed for each hospital based on the caseload. The identification numbers of T1DM patients who met the inclusion criteria were collected to create a sampling frame. Study participants’ medical charts were selected using a computer-generated random number technique. Please refer to (See Fig. 1 for sampling schema) for a schematic representation of the sampling procedure.

Fig. 1.

Schematic presentation of the sampling procedure of adult type 1 diabetic patients in Amhara regional state comprehensive specialized hospitals, Ethiopia, 2023

Variables of the study

The dependent variable of this study was the incidence of DKA. The independent variables age, sex, health insurance, residence, and family history of T1DM were included as sociodemographic factors. Presence of recent acute illness, glycemic control, T1DM duration, presence of comorbidity, and presence of chronic diabetic complications were included as clinical related factors. Finally, follow up frequency, treatment started duration, type of insulin used, and medication non-compliance was included as treatment related factors.

Operational definition

Survival: Time measured between the date of T1DM diagnosis and to the date of developing DKA or date censored [23].

Censored: Adult diagnosed with T1DM and who did not develop DKA during follow up study (Referred, died, end of the study or discharged cases for any reason before the development of DKA during the study period) [23, 24].

DKA: In the current study, DKA was defined as a written diagnosis of diabetic ketoacidosis based on the decision of the physician on the patient card with in the follow-up time. Where available, laboratory findings such as hyperglycemia, acidosis, and ketonuria were used to support the diagnosis.

Established T1DM: Diabetic duration of at least one week [25]. In the current study, established T1D was defined as a written diagnosis of T1DM based on the decision of the physician on the patient card with in the follow-up time.

Type 1 diabetic duration: After established T1DM, calculated as the difference in the age at diagnosis and the age at last follow up date [23].

Co-morbidities: Adult with T1DM and having other chronic medical or psychiatric diseases [23].

Psychiatric disorders: Diagnosed with depression, anxiety, Schizophrenia, substance abuse behaviors, dementia, and epilepsy and eating disorders [23].

Chronic cardiovascular disorders: Congestive heart failure, hypertension, and stroke [23].

Chronic respiratory disorders: Chronic obstructive pulmonary disease, asthma, and pulmonary hypertension [23].

Chronic liver diseases: Chronic hepatitis, liver cirrhosis, hepatocellular carcinoma, alcoholic liver disease, and non-alcoholic fatty liver disease [23].

Event: Incidence of DKA from T1DM diagnosis to end of the study [23, 24].

Entry date: First date for each observation within the study period at the date of T1DM diagnosis [23, 24].

End date: Last date for each observation within the study period that the patient observed last [23, 24].

Incomplete Records: Charts with missing data for the variables of the diagnostic date for T1DM, the date the event occurred, or the date of the last observation or the last follow-up [23, 24].

Recent acute illness: Diagnosis of (acute febrile illness, respiratory infection, cardiac infarction, urinary tract infection, and gastrointestinal infection) prior to two weeks before the onset of DKA or final follow-up observation that is documented on the patient’s medical card [23].

Glycemic control: The term “controlled” refers to an average three-month FBS level of 70 to 130 mg/dl or an HgA1c of less than 7%. Moreover, uncontrolled blood sugar is defined as an average FBS of 70 or less or > 130 mg/dl [26]. Due to limited HbA1c testing availability, glycemic control was assessed via fasting blood glucose, which may not reflect long-term control.

Recent six-month glycemic control: History of glycemic control (poor or good) over the roughly six-month follow-up intervals prior to developing DKA or being censored [23].

Medication noncompliance/missed insulin dose: Medication noncompliance/ missed insulin dose history noted in patients’ medical records [23]. It was determined based on medical record review, including documented missed doses, interruptions in therapy, or clinician notes indicating poor adherence.

Data collection tool and procedures

The study considered various predictors, including socio-demographic factors (age, sex, community health insurance, residence, family history of T1DM), clinical factors (recent acute illness, glycemic control, T1DM duration, comorbidity, chronic diabetic complications), and treatment factors (follow-up frequency, medication type, medication non-compliance, treatment duration). These predictors were identified from relevant literature sources [19–21, 27, 28]. The study utilized a systematic data extraction tool, adapted from the WHO DKA follow sheet forms. As the records were written in English and the data collectors were proficient in English, there was no need to translate the data extraction tool into Amharic. Access to medical records was based on each hospital’s unique medical registration number. Trained BSc nurses (four) from DM or non-communicable disease clinics or follow-up centers were responsible for data collection, while two MSc nurses supervised the overall process.

Data quality control

To ensure data quality, supervisors and data collectors received two-day training from the principal investigator. The training covered the study’s objectives, data gathering methods, the designated data collection period, proper completion of checklists, and ethical considerations. To ensure data quality, supervisors and data collectors received two-day training from the principal investigator. The training covered the study’s objectives, data gathering methods, the designated data collection period, proper completion of checklists, and ethical considerations. This standardized the data collection process and guided the information to be gathered from the data sources. Experts and advisors reviewed the data extraction form for accuracy and missing data before data collection was commenced. A pretest was conducted on 20 medical charts in UoGCSH, resulting in modifications such as excluding variables like smoking status, alcohol consumption, HbA1c, and BMI, while adding the variable of family history of T1DM. Throughout the data collection period, the completeness of the data was verified on-site daily. Regular communication was maintained between the principal investigator, supervisors, and data collectors. Prior to analysis, the principal investigator carefully entered and cleaned the data.

Data processing and analysis

Descriptive and summary statistics were computed, taking into account the data distribution assessed through histograms and the Shapiro-Wilk test. Categorical variables were presented as frequencies and percentages. The outcome of study participants was categorized as either developing DKA or being censored. Missing data were managed using mean and mode imputations, assuming the missing data occurred completely at random. The incidence of DKA was calculated for the entire follow-up period. The Kaplan-Meier survival curve was used to estimate the median survival time, and the log rank test was employed to compare survival rates among different groups of independent variables. Multicollinearity was assessed using the variance inflation factor (VIF), with an overall value of 1.14 and no variable exceeding 10. Survival analysis was performed to estimate time to development of diabetic ketoacidosis. Parametric survival models were compared, and the Gompertz model was selected based on Akaike Information Criterion (AIC) and the observed hazard pattern. Binary variables with a p-value < 0.2 were considered for multiple variables analysis. A backward stepwise selection method was employed to calculate the hazard ratio (HR) and its 95% confidence interval (CI). Predictors with a p-value ≤ 0.05 in the multiple variables analysis were considered statistically significant predictors for the incidence of DKA among T1DM adult patients. Statistical analysis was performed using STATA version 17.0.

Results

Sociodemographic characteristics

Of 634 sampled records, 603 (95.1%) were included for analysis. The remaining 31 (4.89%) had incomplete records or lost medical charts. The majority were young adults (459, 76.12%); more than half were females (350, 58.04%); and more than half resided in urban areas (405, 67.16%) (Table 1).

Table 1.

Sociodemographic and clinical characteristics of T1DM adult patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Variables	Categories	n (%)
Age at onset of T1DM	Young age(18–44 years)	459 (76.12)
	Middle age (45- 64years)	81 (13.43)
	Old age (≥ 65 years)	63 (10.45)
Sex	Male	253 (41.96)
	Female	350 (58.04)
Residence	Urban	405 (67.16)
	Rural	198 (32.84)
Community health insurance	Yes	380 (63.02)
	No	223 (36.98)
Family history of T1DM	Yes	69 (11.44)
	No	534 (88.56)
T1DM duration	< 3 years	337 (55.89)
	≥ 3years	266 (44.11)
Recent six month glycemic control	Controlled	345 (57.21)
	Uncontrolled	258 (42.79)
Acute illness	Yes	193 (32.01)
	No	410 (67.99)
Comorbidities	Yes	204 (33.83)
	No	399 (66.17)
Chronic diabetic complications	Yes	3 (0.50)
	No	600 (99.50)

Clinical characteristic

Out of the total T1DM adult patients, approximately one-third (204, 33.83%) had comorbidities, with 30 (14.7%) having more than one comorbidity. The most common comorbidities were cardiac diseases (71, 34.8%), hypertension (50, 24.51%), liver diseases (41, 20.09%), and asthma (30, 14.71%). Other comorbidities included renal diseases (20, 9.8%), HIV/AIDS (14, 6.86%), COPD (7, 3.43%), psychiatric diseases (3, 1.47%), stroke (2, 0.98%), and gynecologic disorders (2, 0.98%). About 193 (32.01%) had experienced acute recent illnesses, with 19 (9.85%) having more than one acute recent illness. The most common acute illnesses were respiratory infections (87, 45.07%), gastrointestinal tract infections (41, 21.24%), acute febrile illness (35, 18.13%), urinary tract infections (31, 16.06%), pancreatitis (7, 3.63%), myocardial infarction (1, 0.52%), and other infections (10, 5.18%). Approximately 258 (42.79%) of the patients had poor recent six-month glycemic control (Table 1).

Treatment related characteristic

29% of the T1DM adult patients had a history of medication noncompliance. Additionally, approximately 80% of the T1DM adult patients used Neutral Protamine Hagedorn (NPH) insulin, while less than 1% used a combination of regular insulin and NPH insulin during their follow-up. The majority of the patients initiated their treatment immediately after the initial T1DM diagnosis (Table 2).

Table 2.

Treatment related characteristics of T1DM adult patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Variables	Categories	n (%)
Treatment duration	Immediate	586 (97.18)
	Not immediate	17 (2.82)
Insulin type	NPH insulin	481 (79.77)
	Regular insulin	113 (18.74)
	Lenten insulin	5 (0.83)
	Combination	4 (0.66)
Follow up frequency	Every 2 to 3 weeks	2 (0.33)
	Monthly	24 (3.98)
	Every 2 to 3 months	276 (45.77)
	>=4 month	181 (30.02)
	Irregular follow- up	120 (19.90)
Medication non compliance	Yes	175 (29.02)
	No	428 (70.98)

Incidence of DKA among T1DM adult patients

A total of 603 T1DM adult patients were followed for five years, resulting in 18,032 person-month observations. The mean follow-up time was 30 months (95% CI: 29, 31). Among the patients included in the analysis, approximately 40% developed DKA during the follow-up period. The overall incidence of DKA in the cohort was 15.96 per 100 person-years (95% CI: 14.04–18.12) (1.33 per 100 person-months (95% CI: 1.17–1.51)). The median survival time for the entire cohort was 45 months (95% CI: 42, 47). The highest incidence of DKA occurred between 40 and 54 months of follow-up (Table 3).

Table 3.

Life table showing the survival to develop DKA among T1DM adult patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Interval(in months)	Patients at risk	DKA cases	Censored	Cumulative survival	SD error	95% CI
						L	U
[0–10]	603	29	42	0.9502	0.0090	0.9291	0.9651
(10–20]	532	30	79	0.8923	0.0133	0.8631	0.9156
(20–30]	423	21	81	0.8433	0.0163	0.8083	0.8724
(30–40]	321	60	66	0.6676	0.0240	0.6183	0.7121
(40–50]	195	74	59	0.3691	0.0290	0.3125	0.4257
(50–60]	62	26	36	0.1510	0.0298	0.0983	0.2143

Among the 240 T1DM patients who developed DKA, the proportion of DKA cases from FHCSH, UOGCSH, DMCSH, and DTCSH was 88 (36.66%), 82 (34.16%), 38 (15.83%), and 32 (13.33%), respectively. Of the 240 T1DM patients who developed DKA, more than half (126, 52.5%) reported nausea/vomiting as the presenting symptom, followed by abdominal pain (85, 35.42%), polyuria (85, 35.42%), altered sensorium (46, 19.17%), generalized weakness (36, 15%), polydipsia (31, 12.92%), polyphagia (22, 9.17%), and weight loss (17, 7.08%). Only 9 (3.75%) reported fever as a presenting symptom. The mean blood glucose level at the time of DKA diagnosis was 388.35 mg/dl with a standard deviation of ± 3.94 mg/dl. Among the censored observations (363), the majority of the patients (344, 94.77%) continued their T1DM follow-up, 14 (3.86%) transferred to other health facilities, 3 (0.83%) died from other causes, and approximately 2 (0.55%) discontinued their follow-up.

Median survival time

Survival analysis is very important to visualize the overall median survival time among T1DM adult patients, which was 45 months with a median interquartile range (37–53). Regarding the variables, median survival time is summarized in Table 4 below.

Table 4.

Median survival time of incidence of DKA among T1DM adult patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Variables	Categories	Median (IQR)	Chi square value	Log rank test
Overall		45(37–53)
Age	Young	45(37–52)	18.05	< 0.01
	Middle	58(42–59)
	Old	39(36–44)
Sex	Male	40(31–51)	21.55	< 0.01
	Female	47(40–58)
Residence	Urban	47(39–55)	8.59	0.0034
	Rural	40(32–52)
Health insurance	Yes	45(37–54)	0.13	0.7178
	No	43(37–53)
Family history of T1DM	Yes	42(37–50)	2.44	0.1182
	No	46(37–55)
T1DM duration	≥ 3 year	49(41–56)	163.59	< 0.01
	< 3 year	32(23–35)
Glycemic control	Controlled	51(42–58)	45.78	< 0.01
	Uncontrolled	41(32–49)
Acute illness	Yes	41(35–49)	24.12	< 0.01
	No	49(38–58)
Comorbidities	Yes	41(34–50)	11.61	0.0007
	No	47(39–56)
Chronic diabetic complications	Yes	38(3–40)	5.95	0.0147
	No	45(37–54)
Treatment duration	Immediate	45(37–54)	0.25	0.6198
	Not- Immediate	37(37–48)
Insulin type	Lenten	41(40- .)	5.15	0.1608
	NPH insulin	46(37–55)
	Regular insulin	41(34–49)
	Combination	36(32–42)
Medication noncompliance	Yes	40(31–48)	59.90	< 0.01
	No	50(40–59)
Follow up frequency	2 to 3 Week	38(38–38)	44.84	< 0.01
	Monthly	41(41-.)
	2 to 3 months	47(39–58)
	≥ 4 months	49(39–55)
	Irregular	39(25–47)

Overall survival function

The overall Kaplan- Meier estimate showed that the probability of survival of T1DM adult patients without developing DKA was high during the first months of observation (99.34%). The probability of survival without developing DKA relatively falls as the follow-up month increases, and the lowest survival probability (8.46%) was seen in the last months of observation (Fig. 2).

Fig. 2.

The overall Kaplan-Meier survival curve of survival estimate of T1DM adult patients without developing DKA in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Survival function and Comparison of different categorical variables

The estimate of survival among several groupings of predictors is shown on the Kaplan-Meier (estimator) curve for comparison. The log-rank test provided statistical support for the observed difference. For categorical variables, separate graphs of the estimations of the Kaplan-Meier survivor functions were created (Figs. 3, 4, 5, 6, 7, 8, 9, and 10). In general, the group indicated by the upper curve has greater survival than the group defined by the lower curve when one survivorship function is positioned above another. The occurrence of any statistically significant variations in survival across distinct categorical predictors was examined using the log-rank test. Patients who met the following criteria had greater survival than the equivalent categories: being female, no family history of T1DM, regulated six-month glycemic control, no history of medication noncompliance, duration ≥ 3 years, middle age, type 1 diabetic patients without comorbidities, and no history of an acute recent illness fared better than their counterparts in terms of avoiding DKA. The survival plot for incidence of DKA by age is shown in (Fig. 3).

Fig. 3.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on age at diagnosis in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 4.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on sex in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 5.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on family history of T1DM in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 6.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on T1DM duration in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 7.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on recent six months glycemic control in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 8.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on recent acute illness in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 9.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on comorbidities in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

Fig. 10.

The Kaplan-Meier survival curves comparing survival time of T1DM adult patients based on medication noncompliance in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603). Shaded regions indicate 95% confidence bands for the survival probabilities

The survival plot for incidence of DKA by sex is shown in Fig. 4. The median survival time for males and females was 40 months and 47 months, respectively. The plot indicates that the risk of developing DKA is similar for both males and females at the beginning. The result of the log rank test also revealed the same idea (p < 0.01), meaning this difference is statistically significant.

The survival plot for incidence of DKA by family history of T1DM is shown in Fig. 5. The median survival time for patients who had a family history of T1DM and patients who hadn’t a family history of T1DM was 42 months and 46 months, respectively (p = 0.1182).

The survival plot of incidence DKA by T1DM duration is given below in Fig. 6. The median survival time for ≥ 3 years and < 3 years was 49 months and 32 months, respectively. The result of the log rank test also supported the existence of a significant difference (p = < 0.01).

The survival plot for incidence of DKA by recent six months glycemic control is shown in Fig. 7. The median survival time for controlled and uncontrolled was 51 months and 41 months, respectively. The plot indicates that the risk of developing DKA is similar for both controlled and uncontrolled at the beginning. The result of the log rank test also revealed the same idea (p = < 0.01), meaning this difference is statistically significant.

The survival plot of incidence of DKA by recent acute illness is shown in Fig. 8. The median survival time for who had not recent acute illness and T1DM who had recent acute illness was 41 months and 49 months, respectively. The result of the log rank test also revealed the same idea (p = < 0.01), meaning this difference is statistically significant.

The survival plot for incidence of DKA by comorbidities is shown in Fig. 9. The median survival time for T1DM who had comorbidities and who had not comorbidities was 41 months and 47 months, respectively. The result of the log rank test also revealed the same idea (p = 0.0007), meaning this difference is statistically significant.

The survival plot for incidence of DKA by medication noncompliance is shown in Fig. 10. The median survival time for noncompliance and compliance was 40 months and 50 months, respectively. The plot indicates that the risk of developing DKA is similar for both noncompliance and compliance at the beginning. However, the difference becomes visible after some points on the curve and comes closer at the end. The result of the log rank test also revealed the same idea (p = < 0.01), meaning this difference is statistically significant.

Bivariable and multivariable analysis

This study utilized bivariable analysis to examine the effect of each predictor on the incidence of DKA among individuals with T1DM before proceeding to multivariable analysis. The bivariable analyses were conducted for each predictor at a p-value of 0.2. However, the use of Cox regression was found to be inappropriate for the data due to the violation of the proportional hazard assumption, as indicated by the Schoenfeld test for both the global and rank tests. The proportional-hazards assumption is not satisfied by predictors including residence (p = 0.011), T1DM duration (p = 0.0021), glucose control (p = 0.0270), and acute recent illness (p = 0.0018), as evidenced by their p-values being less than 0.05. Their impact on DKA risk varies throughout time. According to the global test (Chi2 = 33.07, df = 10, p = 0.0003), the proportional-hazards assumption for the model is generally violated, indicating that a proportional hazards Cox model might not be entirely appropriate for this dataset. More findings that are accurate would be obtained with a different model that considers time-varying effects. Additionally, selecting the appropriate function of survival time for incorporation into a time-varying Cox model posed challenges. The temporal distribution in the data did not fulfill the assumption of a time-varying Cox model, as shown in Fig. 11. Consequently, the utilization of parametric survival models was deemed more preferable.

Fig. 11.

The kernel-smoothed hazard estimate of DKA among T1DM adult patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

To analyze the incidence of DKA data, various multivariable models, including the proportional hazard (PH), exponential, Weibull, log-logistic, Gompertz, and lognormal distributions, were fitted. These models incorporated all the predictors that were found to be significant in the bivariable analysis at a significance level of 0.2. The efficiency of different models was compared using the Akaike Information Criterion (AIC), which is commonly used for model selection. The model with the minimum AIC value was considered the preferred choice. Among the given alternatives, the Gompertz model was selected because the hazard function demonstrated a monotonic increase over time, consistent with Gompertz assumptions, and it yielded the lowest AIC (AIC = 463.44) among the tested parametric models. Subsequently, the backward elimination technique was employed to remove predictors that became insignificant in the multivariable analysis. Consequently, residence and chronic diabetic complications were excluded from the final model. Age, sex, family history of T1DM, T1DM duration, glycemic control, recent acute illness, comorbidities, and medication noncompliance remained as predictors in the final model (Table 5).

Table 5.

Bivariable and multivariable analysis of incidence of DKA among T1DM patients in Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Variables	Category	Censored	Developed DKA	CHR (95% CI)	AHR (95% CI)	P value
Age	Young age	262	197	1
	Middle age	61	20	0.42(0.27, 0.67)	0.35(0.21, 0.58)	< 0.01***
	Older age	40	23	1.29(0.83,1.99)	0.84(0.52, 1.35)	0.485
Sex	Male	127	126	1
	Female	236	114	0.57(0.44, 0.73)	0.72(0.55, 0.95)	0.019***
Residence	Urban	251	154	1
	Rural	112	86	1.44(1.10, 1.88)	1.15(0.86, 1.54)	0.322
T1DM duration	≥ 3years	123	143	1
	< 3years	240	97	7.51(5.34, 10.57)	13.30(9.09,19.46)	< 0.01***
Family history of T1DM	No	330	204	1
	Yes	33	36	1.33(0.93,1.89)	1.45(1.01, 2.11)	0.049***
Recent six month glycemic control	Controlled	274	71	1
	Uncontrolled	89	169	2.46(1.86, 3.24)	2.29(1.70, 3.08)	< 0.01***
Recent acute illness	No	291	119	1
	Yes	72	121	1.84(1.43, 2.37)	1.41(1.06, 1.87)	0.016***
Comorbidities	No	271	128	1
	Yes	92	112	1.53(1.19, 1.97)	1.78(1.34, 2.38)	< 0.01***
Chronic diabetic complications	No	363	237	1
	Yes	0	3	3.35(1.07, 10.49)	2.65(0.80, 8.70)	0.108
Medication noncompliance	No	322	106	1
	Yes	41	134	2.57(1.99, 3.32)	2.35(1.76, 3.15)	< 0.01***

CHR Crude hazard rate, AHR Adjusted hazard rate, HR 1 is a reference category group, CI Confidence interval

*** = significant at 95% level of confidence

Model diagnosis

Cox snell residual plot

One technique for assessing the goodness of fit of the model to the data is to examine the Cox-Snell residuals. In Fig. 12, the fit of the Gompertz residuals model to the data is demonstrated using maximum likelihood estimation and cumulative hazard functions. The plot displays the cumulative hazard function of residuals plotted against the Cox-Snell residuals. Ideally, if the model fits the data well, this plot should roughly follow a straight line with a slope of one. In the graphic, straight lines are drawn through the origin of the Gompertz baseline distribution, indicating that it is a suitable model for the incidence of the DKA dataset.

Fig. 12.

Cumulative hazard versus Cox Snell residual plots for Gompertz model diagnosis of T1DM adult patients data Amhara regional state comprehensive specialized hospitals, Northwest Ethiopia 2023(n = 603)

Discussion

The current study aimed to determine the incidence and predictors of DKA among adult T1DM patients who have a diabetic follow-up in Amhara regional state comprehensive specialized hospitals, Ethiopia. Current research has emphasized aspects such as insulin administration and the impact of acute diseases, as well as growing patterns and risk factors for DKA across various settings and patient demographics [29–31]. By identifying important factors in the Ethiopian healthcare system, where scarce resources may increase the incidence of DKA, our findings contribute to this body of knowledge. The low prevalence of chronic diabetic complications (0.5%) observed in this cohort may be attributable to data collection challenges or a relatively short duration of disease among participants. Further studies with extended follow-up periods may help clarify this finding. The current study at hand found that about 240(39.80%) developed DKA with the DKA incidence of 15.96 per 100-person year observation (95%CI: 14.04, 18.12) (1.33 per 100-person months (95%CI: 1.17–1.51)). This finding was lower than studies done in Bahir Dar, Ethiopia (36 per 100 person year) [23], Nigeria (93.6%) [32], Congo (79.3%) [33, 34], China (26.3 per 100 person year) [11], and Arab countries (46.7%) [35]. However, this finding is higher than studies done in Israel (0 event per 1000 person-year) [11], United States of America (USA) (61.6 per 10,000 person year) [36], Denmark (13 per 100,000 person year) [37], England (4.1 per 100 person year) [38], Finland (1.8 per 100 person year) [39], Australia (178.6 (85.7 to 328.5)/10 000 person-years) [40], Northern Sweden (5.9 per 100 000 person year) [41], and USA (55.5 per 1000 person year) [42]. The substantially elevated incidence (10–50 fold high) observed in this setting compared to high-income countries likely reflects underlying systemic disparities, including limited access to essential diabetes care (e.g., insulin and glucose monitoring), delayed presentation, broader healthcare system constraints, cultural influences, lifestyle variations, differences in diabetes mellitus management protocols, distinctive clinical characteristics of the participants, the presence of well-trained medical staff, and other methodological considerations. Another possible explanation lies in the contrast in economic development, wherein developed countries boast well-established diabetic care systems. Moreover, the diagnostic criteria for DKA may differ, as developed countries utilize a pH value threshold of < 7.3, while there may be variability in the cutoff points for DKA across different settings.

The current study identified predictors of DKA among type 1 diabetes mellitus adult patients. Age at onset of type 1 diabetes mellitus, sex, family history of type 1 diabetes, T1DM duration, recent six month glycemic control, recent acute illness, comorbidities, medication noncompliance, and follow up frequency were found predictors of the incidence of DKA among type 1 diabetes mellitus.

In the current study sex was significantly associated with the incidence of DKA; it is similar to studies done in USA [36, 42], Scotland [43], and Ethiopia [44, 45]. Males were more likely to develop DKA than females in this study. The possible explanation might be the difference in the health seeking behavior, and sex hormones in responses to treatment and management of T1DM.

According to the current study, the younger age group was significantly associated with the incidence of DKA; it was supported by studies done in USA [36, 42], and Scotland [43]. T1DM adult patients who were in the younger age group were more likely to develop DKA than those in the middle age group. The possible explanation for this difference could be greater insulin resistance results from early adolescent endocrine changes; the level of medication adherence might be poor in young age groups, and in consequence, DKA could be occur. Therefore, it is crucial to strengthen the follow up and care given to the younger age group.

Based on the result of this study, having a family history with T1DM was associated with an increased risk of developing DKA. The reason could be related to an increased genetic risk for T1DM predisposing one to develop DKA. But a study done in Finland revealed that the association was protective [21]. According to their explanation, the possible reason for this could be that a family history of diabetes could lead to better disease recognition and awareness [21]. This might help them to care for themselves to prevent the development of DKA.

In terms of T1DM duration, this study explored that T1DM duration was a strong predictor of the incidence of DKA among T1DM adult patients. Adult patients with T1DM who had less than 3 years’ duration were 13.30 times more hazardous to develop DKA as compared to those with longer (≥ 3 years) duration. This finding was supported by studies conducted in Ethiopia on hyperglycemic emergencies [23], and Iraq [46]. This reason for this could be T1DM patients with shorter (< 3 years) durations of diabetes might be less knowledgeable and experienced with T1DM management and, correspondingly, less adherent to insulin adjustment and diabetes mellitus management [47]. In addition, studies showed that diabetic mellitus patients during the early years of diagnosis had stress and anxiety about the disease [23]. These could increase the risk of developing DKA. Therefore, it is better to follow and give special attention to those T1DM patients with a shorter duration (< 3 years). However, a study done in Nigeria contradicts this significant relationship between T1DM duration and incidence of DKA [48]. This contradiction could happen due to the difference in methodological issues, clinical characteristics of the participants, and study population differences.

Regarding the preceding six-month glycemic control, this study noted that poor glycemic control was a predictor of DKA. This was comparable with studies done in Ethiopia on hyperglycemic emergencies [23], USA [49], and Australia [40]. Patients with poor six-month glycemic control had 2.29 times more hazard of developing DKA as compared to patients with controlled glycemic control. The reason might be that both lower and higher blood sugar levels from the normal range can lead to a point spectrum of acute complications of diabetes by increasing counter-regulatory hormones, altered glucose production, and lipolysis [50, 51]. Therefore, focusing on follow-up frequency and assessing patient adherence to medication would reduce the likelihood of having poor glycemic control.

Patients with a history of a present acute recent illness were 1.41 times more hazardous to develop DKA as compared to those without a history of a present acute recent illness. This was supported by studies in Ethiopia [23], china [52], USA [21], Nigeria [48], and Nature Review Primers 2020 [53]. The reason could be that recent acute illnesses are associated with a hypermetabolic state due to the release of excessive levels of stress hormones, including glucocorticoids, leptin, counter-regulatory hormones, cortisol, adrenaline, and growth hormone, which works against the effects of insulin [50]. This could induce insulin resistance and increase hepatic glucose production, contributing to the development of DKA [54]. Another possible clinical explanation could be patients with recent acute illness complained of nausea, vomiting, and stomach pain, which might be interpreted as a requirement to lower or discontinue their insulin regimen. The problem might be exacerbated when such patients consume large amounts of sugar-rich diets to treat apparent hypoglycemia. Therefore, to prevent such consequences, early prevention and management of recent acute illness are very important.

Based on the current study, T1DM adult patients with the presence of comorbidities were 1.78 times more hazardous to develop DKA as compared with patients without comorbidities. This finding is supported by studies done in Ethiopia [23], Nigeria [55], and USA [56]. This might be because T1DM patients who have comorbidities, such as hypertension, chronic heart failure, gastrointestinal disturbance, renal disease, and others, can either directly or indirectly interfere with insulin’s and other counter-regulatory hormones’ regular functions, and also comorbidities can impact adherence to therapy [57]. Additionally, co-morbidities, which are typically the source of polypharmacy, might cause poor drug compliance, and this could result in the development of DKA [57]. Therefore, extra attention should be paid to individuals with comorbidities to halt the occurrence of DKA.

According to the result of this study, a history of medication noncompliance was a significant predictor of DKA. Other studies conducted in China, Nigeria, china, and Ethiopia [52, 57–59] supported this evidence. The risk of developing DKA among T1DM adult patients who had a history of medication noncompliance was 2.35 times more hazardous compared to those patients without a history of medication noncompliance. This might be because not taking or not following the prescribed course of anti-diabetic medication treatment could increase the blood glucose level, which in turn contributes to the development of DKA [50]. The study’s findings have a substantial influence on preventing DKA in people with T1DM. High-risk periods should be the focus of targeted therapy, particularly in the early years following diagnosis. Structured diabetes education programs should be implemented to improve patients’ comprehension of insulin adherence, early symptom recognition, and sick day management. Additionally, employing mobile health (mHealth) strategies like SMS reminders may enhance drug adherence and follow-up. Healthcare facilities should also develop and implement standardized sick-day management protocols and strengthen routine follow-up methods to ensure early detection and management of problems. Additionally, increasing access to insulin and glucose monitoring equipment is still necessary to reduce the burden of DKA in places with limited resources.

Strength and limitation of the study

The study appears to be the first of its kind in Ethiopia to investigate the incidence and predictors of DKA among adult patients with established T1DM. The study design, being a cohort study, provides better evidence. Furthermore, the study was conducted in a large setting involving multiple centers, with no issue of generalizability. However, it is important to note that the study data were obtained from secondary sources, which resulted in the omission of certain important predictors. These include factors such as nutritional history, economic considerations, and behavioral traits (such as smoking, tobacco and alcohol use, and chat consumption), as well as sociodemographic characteristics (like educational status, religion, and marital status) that could potentially have a significant impact on the incidence of diabetic ketoacidosis. Furthermore, since the data were extracted from patients’ medical cards, there is a possibility of missing records. The findings were primarily based on the information documented in the patients’ medical charts, relying on physician diagnoses (the DKA diagnosis was based on clinical judgment without consistent laboratory confirmation, which may lead to misclassification). This approach carries the risk of potential misdiagnosis of DKA or its predictors. The presence of wide confidence intervals for certain predictors (e.g., chronic complications) indicates reduced precision of the estimates, likely due to limited sample size or sparse events in some categories, and warrants cautious interpretation of these associations. Additionally, it is worth noting that the conclusions drawn from this study may be subject to limitations due to potential documentation errors in the patients’ medical records.

Conclusions, and recommendations

This study found that 39.8% (95% CI: 35.96–43.77) of adults with T1DM developed DKA, with an incidence rate of 15.96 per 100 person-years (95% CI: 14.04–18.12). Key predictors included age at diagnosis, sex, family history of T1DM, glycemic control, duration of diabetes, recent acute illness, comorbidities, and medication noncompliance. Improving glycemic control, strengthening medication adherence, and ensuring close follow-up particularly during the first three years after diagnosis are essential to reduce DKA risk. Early detection and management of acute illnesses and comorbid conditions should also be prioritized. At the health system level, targeted patient education and strengthened monitoring of high-risk patients are recommended. The Amhara Regional Health Bureau should prioritize patient education programs targeting DKA risk awareness, medication adherence, and self-management skills for T1DM patients. Hospitals in the Amhara region could enhance monitoring of T1DM patients with high-risk factors for DKA, such as recent acute illness or comorbidities, to reduce DKA incidence. Future research could explore longitudinal studies to better understand the progression of DKA risk factors over time and evaluate interventions aimed at improving diabetes management in resource-limited settings.

Abbreviations

AHR

Adjusted hazard ratio

AIC

Akaike information criterion

BMI

Body mass index

COPD

Chronic obstructive lung disease

CI

Confidence interval

CHR

Crude hazard ratio

DMCSH

Debre Markos comprehensive specialized hospital

DTCSH

Debre Tabor comprehensive specialized hospital

DKA

Diabetic ketoacidosis

DM

Diabetes mellitus

EPI DATA

Statistical software

FMOH

Federal ministry of health

FHCSH

Felegehiwot comprehensive specialized hospital

HA1C

Hemoglobin A1C

IR

Incidence rate

NPH

Neutral protamine hagedorn

PH

Proportional hazard

SD

Standard deviation

STATA

Statistical software

T1DM

Type 1 diabetes mellitus

UoGCSH

University of Gondar comprehensive specialized hospital

USA

United States of America

WHO

World health organization

Authors' contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or all these areas; took part in drafting, revising, or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Funding

This study received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Data availability

All the data are included within the manuscript. Anyone in need of further details may contact the corresponding author.

Declarations

Ethics approval and consent to participate

Ethical clearance was obtained from the ethical review committee of the University of Gondar College of Medicine and Health Sciences School of Nursing with S/N/167/2015 and from THE Amhara Public Health Institute with a Ref.No አሕጤኢ/ም/ል/ዳ/-ት-03/1920. A formal letter of cooperation was received from UoGCSH, FHCSH, DMCSH, and DTCSH for collecting data. We had taken permission from hospital medical directors, and data was kept confidential. Informed consent was not applied due to the data being taken from chart review only. Moreover, names and medical card numbers were not included in the data-collecting format, and the data was not shared with anybody except the principal investigator and utilized exclusively for research. Patient consent was waived as data were anonymized and extracted from medical records without identifying information. The study adhered to the Declaration of Helsinki and ethical guidelines of the University of Gondar.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.