Abstract
Purpose
Artificial intelligence and real-time continuous glucose monitoring (rtCGM) are important inventions in the history of diabetes. The aim of this study was to compare the effectiveness of two methods of glycemic control (smart insulin dosage software and traditional physician’s experience).
Methods
86 patients with type 2 diabetes (T2DM) who received multiple daily insulin injections were randomized 1:1 into the smart insulin group (App group) and the traditional physician group (Control group). The former calculates the insulin doses based on an application (which is equipped with trend arrows that reflect the magnitude and direction of glucose changes), while the latter calculates the insulin doses based on physician’s experience. During the study, all participants underwent rtCGM and capillary blood glucose (CBG) testing. The study duration was 5 to 7 days. We compared glucose profile of patients in both group after the intervention.
Results
Patients in the App group had lower mean CBG (mmol/L) before dinner (10.07 (8.70, 11.60) vs 11.42 (9.89, 14.86), P = 0.006) and at bedtime (9.93 (8.83, 11.33) vs 10.71 (9.62, 12.72), P = 0.039) compared to the control group. The results of rtCGM showed that patients in the App group had lower mean glucose (mmol/L) (9.01(8.04, 9.70) vs10.45 (8.78, 12.27), P = 0.003), higher glucose management index (7.81% (7.09%, 8.59%) vs 7.19%(6.77%, 7.49%), P = 0.003), higher time in range (64.0% (51.0%, 73.0%) vs 47.0% (29.5%, 67.0%), P = 0.001) and less time above of range (35.0% (23.0%, 47.0%) vs 52.0% (29.5.0%, 70.0%), P = 0.003). No significant difference was observed in time below of range (1.0% (0%, 2.0) % vs 0% (0%, 2.5%), P = 0.469). There was no difference in the incidence of hypoglycemia between the two groups of patients (14.0% vs 20.9%, P = 0.394).
Conclusion
Among T2DM receiving insulin treatment, the smart insulin dosage software with trend arrows has more advantage on controlling hyperglycemia than the physician’s experience. Further research with larger samples and longer durations is needed to support our findings.
The clinical trial registration number was NCT05389839.
Keywords: Insulin dosage adjustment, Trend arrows, Continuous glucose monitoring, Artificial intelligence, Time in range
Introduction
The latest data show that the number of diabetes patients worldwide has exceeded 500 million [1], with a prevalence of more than 10% in the Chinese population [2]. Insulin is an important tool for controlling hyperglycemia, and the method of adjusting insulin dosage is indeed purely empirical, which is particularly difficult in non-endocrinology departments. Studies have shown that non-endocrinologists generally lack knowledge of insulin type, dosage, and glycemic management, and there are significant barriers to in-hospital glycemic management [3]. Disease stress, therapeutic agents, and disease severity are important causes of hyperglycemia during hospitalization [4]. Hyperglycemia is an important factor affecting the length of hospital stay and total cost of hospitalization in diabetic patients. Artificial Intelligence (AI) is a great boon for diabetic patients and plays an important role in insulin dosage adjustment. The insulin dosage system is a software package that incorporates patient-specific dose-response data into a data model and then calculates the dose of new medication needed to reach the next desired goal. In 2003, Gross first suggested that smart insulin dose calculators could play an important role in controlling postprandial glucose in patients with type 1 diabetes mellitus (T1DM) [5]. Since then, smart dosage systems have begun to spread among diabetic patients and have shown some advantages over clinicians adjusting glucose empirically. Results of a randomized controlled trial suggest that smart software to calculate insulin doses improves the quality of glycemic control and does not increase hypoglycemia compared to nurse-adjusted insulin doses [6]. Meanwhile, researchers in Germany have found that hemoglobin A1c (HbA1c) can be better controlled through the use of a smartphone glucose management app [7]. Thus, we proposed to explore the glucose lowering effect of smart insulin dosage system in non-endocrinology hospitalized diabetic patients. Therefore, smart insulin dosage systems offer many benefits to diabetic patients. We propose to investigate the glucose controlling effect of smart insulin dosage system on hospitalized diabetic patients in non-endocrine departments.
Continuous glucose monitoring (CGM) technology is an important and effective tool for glucose management. Factors such as accuracy and convenience have led to the widespread use of CGM for diabetic patients and its great clinical benefits [8, 9]. Given the unique benefits of AI and CGM, several researchers have combined the two for glycemic management [10], Breton et al. found that CGM combined with an intelligent insulin dose titration system reduced glycemic fluctuations [11], both of which were conducted in patients with T1DM. CGM data is a continuous nontemporal series that can provide rich information about glucose, which promotes the use of data mining to further characterize glucose dynamics. Time in range (TIR) is a new metric for assessing glycemic management. Several observational studies have shown that TIR is strongly associated with pathophysiologic characteristics of diabetes, indicators of glycemic variability, and risk of complications and mortality [12–14]. The trend arrow is an important feature of the CGM that shows the magnitude and direction of blood glucose changes. Users can utilize the trend arrows to predict future blood glucose levels so that they can take the next step in treatment, which can help alleviate impending hypoglycemia or hyperglycemia. Most studies on smart insulin dosing software have not included trend arrow information from CGM. Therefore, this study proposes to make full use of trend arrow information to combine smart insulin dosing software and CGM as a method of glycemic control. The glucose control effect of this method was compared with the traditional empirical glucose adjustment by physicians in hospitalized in non-endocrine department.
Materials and methods
Patients
This study included 86 patients with type 2 diabetes (T2DM) who were admitted to Shanghai Public Health Clinical Center from October 2023 to September 2024. All patients were non-endocrinology inpatients, including 57 internal medicine patients and 29 surgery patients, among whom 22 underwent surgical treatment. The detailed distribution of patients across departments is shown in the Supplementary Table S1. All patients received multiple daily insulin injections (MDI) to control hyperglycemia, maintained a regular three-meal diet during hospitalization, and were provided with standardized diabetic meals by the hospital. This study was approved by the Ethical Research Committee of the Shanghai Public Health Clinical Center, with a clinical trial registration number of NCT05389839. All participants signed written informed consent forms.
Inclusion criteria: ①18 to 75 years old; ②Using MDI to control glycose without oral hypoglycemic agents; ③Patients with T2DM; ④Non-endocrinology inpatient; ⑤Having regular three meals every day.
Exclusion criteria: ①Patients who had a myocardial infarction or cardiovascular event within in the past three months; ②Patients with renal insufficiency (estimated glomerular filtration rate <45 ml/min/1.73m2); ③Patients with hepatic insufficiency (alanine aminotransferase or aspartate aminotransferase exceeding more than twice the upper limit of normal);④Pregnant or lactating patients; ⑤Patients with asymptomatic hypoglycemia; ⑥Patients who are unwilling to participate or unable to cooperate.
Withdrawal criteria:①Patients unable to complete capillary blood glucose (CBG) testing during the trial; ②Patients unable to maintain regular meals.
Methods
Subjects meeting the inclusion criteria were divided into two groups by randomization: the smart insulin dosage software-adjusted glucose group (App group) and the traditional physician experience-adjusted glucose group (Control group) (Fig. 1). All patients performed CBG tests four times daily—one before each meal and one before bedtime. Meanwhile, they wore a real-time CGM (rtCGM) device, which was a Silicon Kinetics Continuous Glucose Monitoring System (SiJoy GS1 CGM, Shenzhen Silicon Bionics Technology Co., Ltd.), with the sensor lifespan having a maximum validity of 14 days. The rtCGM silicon sensor used in this study has the following key performance characteristics: it requires 1-hour for initialization, and it’s interstitial fluid glucose readings synchronize with fingerstick glucose measurements within 5–8 minutes. The sensor collects glucose data at 5-minute intervals. The mean absolute relative difference (MARD) of this sensor is 8.83%. It adopts factory batch calibration technology, requiring no fingerstick blood glucose calibration during the entire usage period. Subjects in both groups had no prior experience with rtCGM. The study duration was 5 to 7 days. At the end of the study, a comparative analysis of CBG and CGM data between the two groups was conducted.
Fig. 1.
Research route
Principles of blood glucose adjustment
Basal insulin was calculated in both groups as 0.25 u per kg of body weight in both groups. Mealtime insulin was calculated differently, with the App group calculating the insulin dose based on the Smart App. In this study, the glucose regulation method adopted by physicians in the control group was the sliding-scale insulin method recommended by the American Diabetes Association.
The smart insulin dosage system is a software that integrates patient-specific dose-response data into a data model and then calculates the required insulin dose—the dose needed to reach the next target [15]. The calculation method for the smart insulin dosage system adopted in this study is detailed in the Supplementary. In this study, the rtCGM used is adjunctive; therefore, we take the patients’ CBG as the basis for the App to calculate insulin doses.
Hypoglycemia is defined as the presence of clear hypoglycemic symptoms, or CGM data showing glucose levels below 3.9 mmol/L for 15 minutes or longer. Severe hypoglycemia is defined as a CBG level below 2.8 mmol/L, or CGM data showing glucose levels below 2.8 mmol/L for 15 minutes or longer. The glucose range for TIR was defined as 3.9 to 10 mmol/L, time above range (TAR) is greater than 10 mmol/L, time below range (TBR) is less than 3.9 mmol/L. The formula for GMI (%) is 3.31 + 0.02392 x (mean glucose (mg/dl) form CGM). The GRI is calculated as described in previous literature [16].
Statistical analysis
Statistical analyses were performed using SPSS software (Version 26.0). Normally distributed data were presented as mean + standard deviation and non-normally distributed data were expressed as median and interquartile intervals (IQR 25-75%). Categorical variables were described by frequencies or percentages. For comparisons between the two groups, t-tests were used for continuous variables, and chi-square tests were applied for categorical variables. Line charts, scatter plots, and bar graphs were generated using GraphPad software (Version 8.0). Gplus system (Version 3.58.0) (Shanghai iMedpower Tech. LTD, China) was applied to process CGM data. Two-sided tests p < 0.05 were considered statistically significant.
Results
Study population
A total of 92 patients with T2DM who received MDI at baseline and did not use oral hypoglycemic agents were included in this study (Fig. 2). After randomization, 47 patients were assigned to the App group and 45 to the control group. Four patients in the App group withdrew from the study (2 had severe conditions that prevented them from eating, and 2 refused capillary blood glucose monitoring) and 2 patients in the control group refused CBG testing and withdrew from the study. Eventually, 43 patients in each of group completed the study. Among the total study population, 65.1% were male, with a median age of 58 years, and the median duration of intervention was 7 days. All participants were inpatients from non-endocrinology departments, and the distribution across departments is presented in Supplementary Table S1. Shanghai Public Health Clinical Center is a grade A tertiary hospital in Shanghai with specialized expertise in the field of viral hepatitis. Consequently, 36% (31/86) of the subjects in this study had diabetes with liver cirrhosis.
Fig. 2.
Flow chart of this study
Comparison of baseline data between the two groups of patients
As shown in Table 1. Participants in both groups were well-matched in terms of gender, age, body mass index (BMI), systolic blood pressure (SBP) and diastolic blood pressure (DBP). There were no significant differences between the two groups in fasting blood glucose, fasting C-peptide, fasting insulin, HbA1c, or duration of diabetes mellitus. Baseline CBG data were missing for some subjects and we present baseline CBG and baseline insulin usage for 33 subjects. Patients in the App group had higher pre-breakfast CBG levels than the control group (10.00 (9.00, 14.30) vs 8.50 (7.43, 9.83)mmol/L, P = 0.027). No differences were observed between groups for pre-lunch, pre-dinner, or bedtime CBG levels. There was no difference in total daily insulin dosage. 36% (31/86) of the subjects had cirrhosis, and the percentage of patients with cirrhosis did not differ between the two groups (32.6% vs 39.5%, P = 0.500). A total of 25.6% (22/86) of subjects underwent surgery during the study period, and the percentage of operated patients did not differ between the two groups (20.9% vs. 30.2%, P = 0.323).
Table 1.
Baseline information for both groups of subjects
| Characteristics | App group (n = 43) | Control group (n = 43) | P |
|---|---|---|---|
| Sex (Male) | 28 (65.1%) | 28 (65.1%) | 0.665 |
| Age (year) | 58.0 (43.0, 62.0) | 58.0 (50.0, 65.0) | 0.282 |
| BMI (kg/m2) | 24.3 (21.5, 26.4) | 24.16 (21.15, 25.71) | 0.675 |
| SBP (mmHg) | 125.0 (117.0, 136.0) | 125.0 (117.0, 136.0) | 0.248 |
| DBP (mmHg) | 79.0 (70.0, 89.0) | 79.0 (70.0, 89.0) | 0.247 |
| FPG (mmol/L) | 8.49 (7.07, 10.74) | 8.99 (7.9, 13.16) | 0.090 |
| FCP (pg/dL) | 1.20 0.60, 2.56) | 1.76 1.08, 2.72) | 0.172 |
| FIN (ug/dL) | 8.90 (6.00, 14.14) | 13.08 (5.91, 25.04) | 0.132 |
| HbA1c (%) | 9.80 (8.00, 11.1) | 9.40 (8.25, 10.5) | 0.675 |
| CBG before breakfast (mmol/L) (n = 33) | 10.00 (9.00, 14.30) | 8.50 (7.43, 9.83) | 0.027 |
| CBG before lunch (mmol/L) (n = 33) | 16.95 (9.43, 18.98) | 14.50 (8.90, 17.63) | 0.296 |
| CBG before bedtime (mmol/L) (n = 33) | 11.80 (10.00, 14.30) | 11.55 (8.48, 13.65) | 0.395 |
| CBG before bedtime (mmol/L) (n = 33) | 14.80 (10.50, 17.40) | 11.90 (9.45, 14.05) | 0.108 |
| Total insulin dose (U) (n = 33) | 30.00 (27.00, 40.00) | 31.00 (29.00, 38.25) | 0.610 |
| Creatinine (mg/dL) | 61.9 (50.6, 81.2) | 69.9 (56.0, 82.6) | 0.195 |
| eGFR (ml/min1.73m2) | 108.4 (88.4, 145.0) | 107.1 (79.7, 127.9) | 0.210 |
| Duration of DM (year) | 5.0 (0.1, 12.0) | 7.0 (0.5, 15.0) | 0.384 |
| Cirrhosis (N, %) | 14 (32.6%) | 17 (39.5%) | 0.500 |
| Surgeries (N, %) | 9 (20.9%) | 13 (30.2%) | 0.323 |
BMI body mass index, SBP systolic blood pressure, DBP diastolic blood pressure, FPG fasting blood glucose, FCP Fasting C-peptide, FIN fasting insulin, HbA1c hemoglobin A1c, CBG capillary blood glucose, GFR estimated glomerular filtration rate, DM diabetes mellitus
Comparison of capillary blood glucose and insulin dosage between the two groups of patients
we compared the mean values of CBG during the study period, with results presented in Table 2 and Fig. 3. There was no significant difference in pre-breakfast CBG between the App group and the control group (7.56 (6.94, 9.53) vs 8.26 (7.28, 9.65) mmol/L, P = 0.148), nor in pre-lunch CBG (10.21 (8.58, 12.16) vs 11.17 (9.89, 14.86) mmol/L, P = 0.096). Patients in the App group had a lower pre-dinner glucose (10.07 (8.70, 11.60) vs 11.42 (9.94, 14.46) mmol/L, P = 0.006) and a lower bedtime glucose (9.93 (8.83, 11.33) vs 10.71 (9.62, 12.72) mmol/L, P = 0.039). There were no significant differences between the two groups in total insulin dosage (230.0 (200.0, 293.0) vs 244.0 (172.0, 297.0) U, P = 0.972), daily average insulin dosage (37.0 (28.7, 46.4) vs 35.7 (31.9, 45.0) U, P = 0.675), or insulin dosage per kilogram of body weight (0.59 (0.47, 0.73) vs 0.57 (0.47, 0.74) U/kg, P = 0.789). 17.4% (15/86) of patients experienced hypoglycemia, with no difference in the number of hypoglycemic patients between the two groups (14.0% vs 20.9%, P = 0.394). According to CGM data, there was no difference in the frequency of hypoglycemia between the two groups (2.0 (0, 3.0) vs (0, 3.5), P = 0.152), and no severe hypoglycemia occurred in either group.
Table 2.
Comparison of mean capillary blood glucose and dose of insulin used in the two groups at the end of the study
| Parameters | App group (n = 43) | Control group (n = 43) | P |
|---|---|---|---|
| Days | 7.0 (6.0, 7.0) | 7.0 (6.0, 7.0) | 0.271 |
| Pre-breakfast CBG (mmol/L) (mmol/L) (mmol/L) | 7.56 (6.94, 9.53) | 8.26 (7.28, 9.65) | 0.148 |
| Pre-lunch CBG (mmol/L) | 10.21 (8.58, 12.16) | 11.17 (9.89, 14.86) | 0.096 |
| Pre-dinner CBG (mmol/L) | 10.07 (8.70, 11.60) | 11.42 (9.94, 14.46) | 0.006 |
| Pre-bedtime CBG (mmol/L) | 9.93 (8.83, 11.33) | 10.71 (9.62, 12.72) | 0.039 |
| Total insulin dose (U)(U) | 230.0 (200.0, 293.0) | 244.0 (172.0, 297.0) | 0.972 |
| Total insulin dose /weight (U/kg) | 0.59 (0.47, 0.73) | 0.57 (0.47, 0.74) | 0.789 |
| Average daily insulin dose (U/day) | 37.0 (28.7, 46.4) | 35.7 (31.9, 45.0) | 0.675 |
| Patients with CBG <3.9 mmol/L (N, %) | 6 (14.0%) | 9 (20.9%) | 0.394 |
| CGM-detected hypoglycemia(<3.9 mmol/L) (N) | 2.0 (0, 3.0) | 0 (0, 3.5) | 0.152 |
CBG capillary blood glucose
Fig. 3.
capillary blood glucose in both groups after intervention. * represents P <0.05, ** represents P <0.01
Comparison of CGM parameters between the two groups of patients
As show in Table 3, both the mean glucose (MG) (9.01 (8.04, 9.70) vs 10.45 (8.78, 12.27)mmol/L, P = 0.003) and standard deviation (SD) of glucose (3.00 (2.29, 3.46) vs 3.48 (2.84, 4.04)mmol/L, P = 0.011) of the App group were lower than those in the control group. The GMI of the App group was lower than that of the control group (7.81 (7.09, 8.59) vs 7.19 (6.77, 7.49)%, P = 0.003) (Fig. 4A). Patients in the App group had a higher TIR (64.0 (51.0, 73.0) vs 47.0 (29.5, 67.0)%, P = 0.001) and a lower TAR (35.0 (23.0, 47.0) vs 52.0 (29.5, 70.0)%, P = 0.003), while there was no significant difference in TBR between the two groups (1.0 (0, 2.0) vs 0 (0, 2.5)%, P = 0.469) (Fig. 4B). Indicators reflecting glycemic fluctuations also differed between the two groups. Patients in the App group had a lower mean amplitude of glycemic excursions (MAGE) (7.28 (5.80, 8.39) vs 7.98 (6.82, 9.44)mmol/L, P = 0.018) (Fig. 4C), a lower mean of daily difference (MODD) (2.42 (1.91, 2.76) vs 2.85 (2.13, 3.47) mmol/L, P = 0.018) (Fig. 4D), and a lower largest amplitude of glycemic excursion (LAGE) (14.20 (12.10, 17.80) vs 16.10 (13.15, 18.85)mmol/L, P = 0.029) (Fig. 4E). There was no significant difference in the coefficient of variation (CV) between the two groups (0.34(0.28, 0.37) vs 0.33 (0.27, 0.38), P = 0.757). Patients in the App group had a lower GRI (35.84 (28.44, 52.60) vs 62.98 (39.50, 83.07), P = 0.000) (Fig. 4F). Figure 5 shows the ambulatory glucose profiles (AGP), and the overall glucose of patients in the App group was lower than that of the control group.
Table 3.
Comparison of CGM data between the two groups of patients after the intervention
| Parameters | App group (n = 43) | Control group (n = 43) | P |
|---|---|---|---|
| MG (mmol/L) | 9.01 (8.04, 9.70) | 10.45 (8.78, 12.27) | 0.003 |
| SD (mmol/L) | 3.00 (2.29, 3.46) | 3.48 (2.84, 4.04) | 0.011 |
| CV (%) | 0.34 (0.28, 0.37) | 0.33 (0.27, 0.38) | 0.757 |
| MAGE (mmol/L) | 7.28 (5.80, 8.39) | 7.98 (6.82, 9.44) | 0.018 |
| MODD (mmol/L) | 2.42 (1.91, 2.76) | 2.85 (2.13, 3.47) | 0.018 |
| LAGE (mmol/L) | 14.20 (12.10, 17.80) | 16.10 (13.15, 18.85) | 0.029 |
| TIR (%) | 64.0 (51.0, 73.0) | 47.0 (29.5, 67.0) | 0.001 |
| TAR (%) | 35.0 (23.0, 47.0) | 52.0 (29.5, 70.0) | 0.003 |
| TBR (%) | 1.0 (0, 2.0) | 0 (0, 2.5) | 0.469 |
| GMI (%) | 7.81 (7.09, 8.59) | 7.19 (6.77, 7.49) | 0.003 |
| GRI | 35.84 (28.44, 52.60) | 62.98 (39.50, 83.07) | 0.000 |
MG Mean glucose, CV coefficient of variation, SD standard deviation, MAGE mean amplitude of glucose excursions, MODD mean of daily difference, LAGE largest amplitude of glycemic excursion, Min-GM minimum glucose, Max-G maximum glucose, TIR time in range, TAR time above range, TBR time below range, GMI glucose management index, GRI glycemic risk index
Fig. 4.
Comparison of CGM parameters, GMI (A), TIR/TAR/TBR (B), MAGE (C), MODD (D), LAGE (E) and GRI (F) in two groups of patients. * represents P <0.05, ** represents P <0.01
Fig. 5.
Ambulatory glucose profiles of the two groups of patients. Data were expressed as median and interquartile intervals (IQR 25-75%)
Comparison of patients with TIRå 70% between the two groups
At the end of the study, 17 patients in the App group and 8 patients in the control group had a TIR > 70%, and the comparison results of these patients are shown in Supplementary Table S2. There were no significant differences in age, gender, BMI, baseline FPG, baseline FCP, or baseline HbA1c between the two groups. There were no significant differences in the proportion of patients with liver cirrhosis or the proportion of patients who underwent surgery between the two groups. During the study period, there were no significant differences in the total insulin dosage or daily insulin dosage between the two groups. Additionally, there was no significant difference in the incidence of hypoglycemia between the two groups during the study period.
Discussion
Hyperglycemia is common in hospitalized patients and is a detrimental factor in the severity of illness and cost of hospitalization. Insulin is an important tool for glycemic control, and insulin dose adjustments are often based on physician experience. Smart insulin dosage system helps doctors and patients get more accurate insulin doses. In this study, we combined the mature AI technology with the glycemic profile characteristics of T2DM, based on the individualized characteristics of patients, fully utilized the advantages of CGM and referred to its trend arrow information to provide timely and effective pre-meal insulin dosage adjustment for diabetic patients. We conducted this randomized controlled trial aiming to compare the difference in effectiveness between the smart insulin dosage software and the empirical glucose adjustment by non-endocrine physicians. In this study, patients whose glucose was adjusted according to App had lower CBG levels, especially before dinner and at bedtime, compared with those whose glucose was adjusted empirically by non-endocrine physicians. According to the CGM data, the average glucose of the App group was lower, which also indicated the consistency between CGM and CBG data. Previous studies have shown that insulin dose calculators are better for postprandial glucose control in insulin treated patients, and this study only measured CBG [5]. Similar studies have also shown the importance of smart insulin dosing software for postprandial glucose [17]. These studies are consistent with the results of this study, showing that smart insulin dosage software has better hypoglycemic effects in diabetic patients. In addition, there was no difference in total insulin use between the two groups, and patients in the traditional physician group had a higher incidence of hypoglycemia. Previous research findings indicate that smart insulin dosing software can reduce total insulin dosage in both T1DM [18] and T2DM [19], while in this study, patients in the App group experienced an increase in total insulin dosage compared to baseline. The incidence of hypoglycemia in the App group was numerically lower than that in the control group, the reasons may be as follows. First, the insulin algorithm of the App incorporates comprehensive information such as residual insulin and trend arrows, resulting in higher accuracy of insulin doses. Second, it may be related to the clinical experience of physicians in the control group, such as using insulin more conservatively.
CGM recording of dynamic glucose data and trend arrows is another feature of this study. CGM is one of the great advances in diabetes technology, which not only records glucose data, but also provides more meaningful indicators of diabetes management than HbA1c, such as TIR, CV, GMI, MAGE [20, 21]. Studies have shown that TIR is directly associated with the risk of diabetes complications and death [22, 23]. In hospitalized patients, hyperglycemia, hypoglycemia, and glucose variability were all associated with adverse outcomes. In our study, patients in the App group had higher TIR and lower MAGE, suggesting that patients in the App group had less glucose fluctuation [24]. This is slightly different from the results of Secher’s study, which found that smart insulin dosing software lowered CV but did not significantly change TIR compared to usual care [25].
Unlike previous studies, this study took into account the information from the trend arrows and fully considered the magnitude and direction of the patients’ glucose changes to avoid too high or too low glucose. The trend arrows might be one of the reasons why the incidence of hypoglycemia in the App group was smaller than that in the control group.
In this study, 29% of the patients had a TIR of more than 70%, which may be related to the subjects having very high glucose and undergoing surgery during the study period. In addition, 36% of diabetic patients had cirrhosis, which has a higher risk phase for hyperglycemia [26]. Moreover, studies have shown that diabetic patients with cirrhosis have worse glycemic control compared to diabetic patients without cirrhosis [27]. This study has several limitations. First, the sample size of this study was small, and due to the nature of hospitalization, the study period was limited to one week, necessitating support from prospective studies with larger samples and longer follow-up periods. Second, all subjects were non-endocrinology inpatients, resulting in partial loss of baseline CBG data and the absence of a washout period prior to intervention. Third, the study population exhibited selection bias, such as an excessively high proportion of patients with liver cirrhosis. In summary, among non-endocrinology hospitalized diabetic patients, smart insulin dosage software with trending arrows was shown to be more effective in controlling hyperglycemia without increasing the risk of hypoglycemia compared with traditional physician experience. Additional clinical studies are needed to further support our findings.
Supplementary information
Acknowledgements
We are grateful to Mr. Yunyang Sun and Shanghai Imedpower Tech Ltd. Provided the CGMs data analysis.
Author contributions
Author contributions contributed to the study conception and design. Material preparation, data collection, and analysis were performed by L.H.Z, C.W.W, C. H. W, and D.D.Q. The first draft of the manuscript was written by L.H.Z. and X.L.Z. All authors commented on previous versions of the manuscript. Xia Yu provided technical support. All authors read and approved the final manuscript.
Data Availability
No datasets were generated or analysed during the current study.
Declarations
Conflict of interest
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Chenwei Wu and Chunhong Wang contributed equally to this work.
Supplementary information
The online version contains supplementary material available at 10.1007/s12020-025-04509-z.
References
- 1.K.L. Ong, L.K. Stafford, S.A. McLaughlin et al. Global, regional, and national burden of diabetes from 1990 to 2021, with projections of prevalence to 2050: a systematic analysis for the Global Burden of Disease Study 2021. The Lancet 402(10397), 203–234 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.W. Jia, J. Weng, D. Zhu et al. Standards of medical care for type 2 diabetes in China 2019. Diabetes Metab Res Rev 35(6), e3158 (2019) [DOI] [PubMed] [Google Scholar]
- 3.S. Latta, M.N. Alhosaini, Y. Al-Solaiman et al. Management of inpatient hyperglycemia: assessing knowledge and barriers to better care among residents. Am J Ther 18(5), 355–365 (2011) [DOI] [PubMed] [Google Scholar]
- 4.C. Jin, Y. Lai, Y. Li et al. Changes in the prevalence of diabetes and control of risk factors for diabetes among Chinese adults from 2007 to 2017: An analysis of repeated national cross‐sectional surveys. Journal of Diabetes 16(2), e13492 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.T.M. Gross, D. Kayne, A. King, C. Rother, S. Juth, A bolus calculator is an effective means of controlling postprandial glycemia in patients on insulin pump therapy. Diabetes Technol Ther 5(3), 365–369 (2003) [DOI] [PubMed] [Google Scholar]
- 6.J. Dubois, T. Van Herpe, R.T. van Hooijdonk et al. Software-guided versus nurse-directed blood glucose control in critically ill patients: the LOGIC-2 multicenter randomized controlled clinical trial. Critical Care 21(1), 212 (2017) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.N. Hermanns, D. Ehrmann, K. Finke-Groene et al. Use of smartphone application versus written titration charts for basal insulin titration in adults with type 2 diabetes and suboptimal glycaemic control (My Dose Coach): multicentre, open-label, parallel, randomised controlled trial. The Lancet Regional Health - Europe 33, 100702 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.E.E. Wright, S. Subramanian, Evolving use of continuous glucose monitoring beyond intensive insulin treatment. Diabetes Technology & Therapeutics 23(S3), S-12–S-18 (2021) [DOI] [PubMed] [Google Scholar]
- 9.M.A. Jackson, A. Ahmann, V.N. Shah, Type 2 diabetes and the use of real-time continuous glucose monitoring. Diabetes Technology & Therapeutics 23(S1), S-27–S-34 (2021) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.M. Vettoretti, G. Cappon, A. Facchinetti, G. Sparacino, Advanced diabetes management using artificial intelligence and continuous glucose monitoring sensors. Sensors 20(14), 3870 (2020) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.M.D. Breton, S.D. Patek, D. Lv et al. Continuous glucose monitoring and insulin informed advisory system with automated titration and dosing of insulin reduces glucose variability in type 1 diabetes mellitus. Diabetes Technology & Therapeutics 20(8), 531–540 (2018) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.J. Ye, J. Deng, W. Liang et al. Time in range assessed by capillary blood glucose in relation to insulin sensitivity and β-cell function in patients with type 2 diabetes mellitus: A cross-sectional study in China. Journal of Diabetes Investigation 13(11), 1825–1833 (2022) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.R.M. Goldenberg, V.R. Aroda, L.K. Billings et al. Correlation between time in range and HbA1c in people with type 2 diabetes on basal insulin: post Hoc analysis of the SWITCH PRO study. Diabetes Therapy 14(5), 915–924 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.R.M. Bergenstal, E. Hachmann-Nielsen, K. Kvist, A.L. Peters, J.M. Tarp, J.B. Buse, Increased Derived time in range is associated with reduced risk of major adverse cardiovascular events, severe hypoglycemia, and microvascular events in type 2 diabetes: a post Hoc analysis of DEVOTE. Diabetes Technology & Therapeutics 25(6), 378–383 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.C.B. Cook, L.J. Mann, E.C. King et al. Management of insulin therapy in urban diabetes patients is facilitated by use of an intelligent dosing system. Diabetes Technol Ther 6(3), 326–335 (2004) [DOI] [PubMed] [Google Scholar]
- 16.D.C. Klonoff, J. Wang, D. Rodbard et al. A glycemia risk index (GRI) of hypoglycemia and hyperglycemia for continuous glucose monitoring validated by clinician ratings. Journal of Diabetes Science and Technology 17(5), 1226–1242 (2022) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.P. Rossetti, F.J. Ampudia-Blasco, A. Laguna et al. Evaluation of a novel continuous glucose monitoring-based method for mealtime insulin dosing—the iBolus—in subjects with type 1 diabetes using continuous subcutaneous insulin infusion therapy: a randomized controlled trial. Diabetes Technology & Therapeutics 14(11), 1043–1052 (2012) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.S. Ruddock-Walker, S. Shaaban, J.L. Jacobson, L. Meltzer, C.Z. Minutti, S.W. Hovey, Improving timeliness of insulin administration by using an insulin dose calculator. Hospital Pediatrics 11(10), 1163–1173 (2021) [DOI] [PubMed] [Google Scholar]
- 19.M. Warren, R.M. Bergenstal, M. Hager, E. Bashan, I. Hodish, A scalable application of artificial intelligence-driven insulin titration program to transform type 2 diabetes management. Diabetes Technol Ther 26(8), 556–565 (2024) [DOI] [PubMed] [Google Scholar]
- 20.American Diabetes Association, 6. Glycemic Targets: Standards of Medical Care in Diabetes-2021. Diabetes Care 44(Suppl 1), S73–s84 (2021) [DOI] [PubMed] [Google Scholar]
- 21.T. Battelino, T. Danne, R.M. Bergenstal et al. Clinical targets for continuous glucose monitoring data interpretation: recommendations from the international consensus on time in range. Diabetes Care 42(8), 1593–1603 (2019) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.J. Wan, J. Lu, C. Li, X. Ma, J. Zhou, Research progress in the application of time in range: more than a percentage. Chinese Medical Journal 136, 522–527 (2023) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.R.W. Beck, R.M. Bergenstal, T.D. Riddlesworth et al. Validation of Time in Range as an Outcome Measure for Diabetes Clinical Trials. Diabetes Care 42(3), 400–405 (2019) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.M.O. Seisa, S. Saadi, T. Nayfeh et al. A systematic review supporting the endocrine society clinical practice guideline for the management of hyperglycemia in adults hospitalized for noncritical illness or undergoing elective surgical procedures. The Journal of Clinical Endocrinology & Metabolism 107(8), 2139–2147 (2022) [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.A.L. Secher, U. Pedersen-Bjergaard, O.L. Svendsen et al. Flash glucose monitoring and automated bolus calculation in type 1 diabetes treated with multiple daily insulin injections: a 26 week randomised, controlled, multicentre trial. Diabetologia 64(12), 2713–2724 (2021) [DOI] [PubMed] [Google Scholar]
- 26.W.G. Lee, C.I. Wells, J.L. McCall, R. Murphy, L.D. Plank, Prevalence of diabetes in liver cirrhosis: A systematic review and meta-analysis. Diabetes Metab Res Rev 35(6), e3157 (2019) [DOI] [PubMed] [Google Scholar]
- 27.L. Castera, K. Cusi, Diabetes and cirrhosis: current concepts on diagnosis and management. Hepatology 77(6), 2128–2146 (2023) [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 Availability Statement
No datasets were generated or analysed during the current study.