Evaluation of Computer-Based Insulin Infusion Algorithm Compared With a Paper-Based Protocol in the Treatment of Diabetic Ketoacidosis

Abstract

Background: Development of computer-based software, termed electronic glucose management system (eGMS), offers an alternative strategy to manage diabetic ketoacidosis (DKA) compared with institution-specific paper protocols by integrating glucose and insulin titration into the electronic medical record. Objective: To evaluate the safety and efficacy of eGMS versus a paper-based DKA protocol in an urban academic medical center. Methods: Single-center, retrospective analysis of patients admitted for DKA. The primary objective of this study was the time to transition from intravenous to subcutaneous insulin after resolution of DKA pre- and post-eGMS implementation. Secondary outcomes included incidence of hypoglycemia while on an insulin infusion, intensive care unit (ICU) length of stay, and total hospital length of stay. Results: Time to DKA resolution was similar in both groups with a median time of 8.6 versus 8.8 hours in the paper-based (n = 133) and eGMS groups (n = 84), respectively (P = 0.43). Hypoglycemia occurred more frequently in the paper-based group compared with eGMS during insulin infusion (14 vs 3 patients, P = 0.06). The median ICU (36.5 vs 41.4 hours; P = 0.05) and hospital length of stay (67.9 vs 77.8 hours; P = 0.05) were shorter in the paper-based group compared with the eGMS group. Conclusion and Relevance: Similar rates of DKA resolution were seen for patients managed with a paper-based protocol compared with eGMS. Patients in the paper-based protocol had a shorter ICU and hospital length of stay; however, eGMS had improved clinically relevant safety outcomes.

Background

Treatment of diabetic ketoacidosis (DKA) involves fluid resuscitation, electrolyte replacement, and correction of hyperglycemia with intravenous (IV) insulin.^1-3 As current guidelines may only offer general guidance on insulin dosing and titration, many institutions have developed institution-specific protocols to manage DKA. The development of computer-based software, termed electronic glucose management system (eGMS), offers an alternative management strategy by integrating glucose and insulin titration, based on proprietary treatment algorithms, into the electronic medical record (EMR). ⁴ With an eGMS system, initial insulin dosing is determined by integrating multiple patient-specific factors, including height, weight, hemoglobin A1C, a history of insulin resistance, or a history of renal disease. Insulin doses are then adjusted based on a patient’s glycemic trends after receiving insulin, integrating factors such as carbohydrate consumption to determine the need for changes to both mealtime and correctional insulin doses moving forward. Insulin doses in eGMS are not provider-titrated, as eGMS uses Food and Drug Administration (FDA)-cleared proprietary algorithms to determine both starting insulin doses and dose adjustments based on glycemic control trends.

Various studies evaluating the efficacy and safety of FDA-cleared eGMS have demonstrated improved glycemic control regarding reduced hypoglycemic events and faster resolution of DKA compared with paper-based protocols in DKA.^4-7 Despite these findings, there is potential that both institution-specific and user-specific factors may impact the safety and efficacy of eGMS. The objective of this study was to evaluate the safety and efficacy of eGMS compared with a paper-based DKA protocol in an urban academic medical center.

Methods

This study was a single-center, retrospective analysis of patients admitted for management of DKA at a large university-based teaching hospital with >600 beds located in the Southwestern United States. The time frame was between April 1, 2017, through April 30, 2018 (paper-based protocol) and April 1, 2019, through April 30, 2020 (eGMS). Implementation of eGMS made by Glytec occurred on October 22, 2018. We allowed for a 6-month “washout period” following implementation of eGMS to account for staff training, eGMS troubleshooting, and eGMS implementation hospital-wide.

For patients with DKA treated with the paper-based protocol at our medical center, they would first start with fluid resuscitation with normal saline, followed by initiation of an insulin infusion. Insulin used for the insulin infusion was regular insulin. The insulin infusion would start at a rate of 0.1 units/kg/h which was titrated by 0.05 units/kg/h until blood glucose (BG) was between 150 and 250 mg/dL, and at the same time, the patient’s fluid was changed to dextrose in water (D5W) with half-normal saline until DKA resolution and transition to subcutaneous (SUBQ) insulin.

For patients with DKA treated with eGMS at our medical center, the default treatment parameters are a goal BG between 140 and 180 mg/dL with an initial multiplier of 0.01 units/kg/h. The multiplier determines the initial IV insulin dose. The multiplier then adapts to a patient-specific value based on hourly BG data points reported in the eGMS, which is how insulin dose adjustments are calculated and IV insulin rate changes are made. The software used at our center allows for manual entry adjustments to insulin doses recommended by the eGMS. The need for these adjustments may arise in situations where there may be staff concerns regarding the insulin rate adjustments in the interest of patient safety. These situations can include, but not limited to, clinical changes such as evolving sepsis, acute kidney injury, and worsening hepatic function or need for custom dosing to resume IV insulin after surgery.

Patients were included in this study if they were 18 years of age and older admitted with a primary diagnosis of DKA, defined as a BG >250 mg/dL, positive serum/urine ketones, and an elevated anion gap (AG) >15. ¹ Patients could have either type 1 or type 2 diabetes. All patients with DKA during the selected time period who met inclusion criteria were looked at. Exclusion criteria for this study were diagnosis of hyperosmolar hyperglycemic state (HHS), no administration of an insulin infusion, DKA management outside of the intensive care unit (ICU), continued IV insulin after resolution of DKA for other medical reasons, inability to tolerate oral intake at the time of insulin transition, transfer from an outside hospital, pregnancy, or incarceration.

Data parameters collected included gender; age; weight; body mass index (BMI); race; ethnicity; hemoglobin A1c; baseline pH; serum bicarbonate and potassium; baseline and first AG <12; BG at baseline, throughout infusion, and 24 hours after insulin infusion; duration of insulin infusion; APACHE (Acute Physiology and Chronic Health Evaluation) II score; duration of ICU stay; and total hospital length of stay. Data were collected from the EMR Cerner Corporation—Millennium/Power Chart Version 2018.01 (Kansas City, Missouri). Data surrounding eGMS use, including manual eGMS adjustments, were collected from Glytec (Waltham, Massachusetts). This study was approved by our institution’s Research Review Committee (Study ID 21-048).

The primary outcome of this study was the time to transition from IV to SUBQ insulin (in hours) after resolution of DKA pre- and post-eGMS implementation. This time frame was measured from when the DKA protocol was initiated to the time the insulin infusion was transitioned to SUBQ insulin. Diabetic ketoacidosis resolution for this study was defined as an AG <12. An AG of 12 allows for better comparisons between the cohorts as transition to SUBQ in the paper-based protocol could not occur until that defined marker was met. Secondary outcomes included the incidence of hypoglycemia, defined as BG <70 mg/dL, and severe hypoglycemia, defined as BG <40 mg/dL, both during the insulin infusion and in the first 24 hours following the transition to SUBQ insulin, duration of ICU stay (in hours), and total hospital length of stay (in hours).

Statistical analysis was performed using IBM SPSS Statistics, Version 19. Differences in transition times, ICU, and hospital length of stay between the groups were compared using a t test or Mann-Whitney U. Data were reported as means and standard deviations for normally distributed continuous variables or medians with interquartile ranges for non-normally distributed data. Incidence of hypoglycemia was compared through χ² or Fisher exact test to evaluate categorical variables and reported as proportions. A P value of ≤0.05 was considered statistically significant.

Results

A total of 456 patients were screened for inclusion during the defined study periods. Of the screened hospital encounters, 239 of those admissions were excluded (Figure 1). The most common reasons for study exclusion were no administration of an insulin infusion (n = 71), continued IV insulin after DKA resolution for other reasons (n = 54), and transfer from an outside hospital (n = 53). Of the 217 patients included, 133 were in the paper-based protocol cohort and 84 were in the eGMS cohort. Baseline characteristics are detailed in Table 1. Groups were similar regarding age, weight, BMI, race, ethnicity, hemoglobin A1c, baseline labs, and APACHE II scores.

Figure 1.

Patient flow chart.

Abbreviations: DKA, diabetic ketoacidosis; eGMS, electronic glucose management system.

Table 1.

Baseline Characteristics.

	Paper (n=133)	eGMS (n=84)	P value
Sex, male, No. (%)	59 (44.3)	31 (36.9)	0.28
Age, y (mean ± SD)	41.9 ± 16.7	45.9 ± 16.1	0.08
Weight, kg (mean ± SD)	73.2 ± 19.9	74.8 ± 22.8	0.60
BMI, mean ± SD	25.7 ± 6	27.5 ± 9.3	0.19
Race, No. (%)			0.97
White/Anglo	95 (71.4)	62 (73.8)
American Indian/Alaska Native	22 (16.5)	11 (13)
Black/African American	7 (5.2)	4 (4.7)
Asian	0	1 (1.1)
Unavailable	9 (6.7)	6 (7.1)
Ethnicity, No. (%)			0.51
Hispanic/Latino	70 (52.6)	47 (55.9)
Not Hispanic/Latino	60 (45.1)	36 (42.8)
Unavailable/decline to answer	3 (2.2)	1 (1.1)
Hemoglobin A1c (mean ± SD)	12.3 ± 2.5	12.4 ± 2.8	0.82
Baseline laboratory values (mean ± SD)
BG, mg/dL	557 ± 248	614 ± 248	0.10
Anion gap	21.8 ± 5.3	22.8 ± 4.9	0.18
pH	7.18 ± 0.39	7.13 ± 0.16	0.29
Serum sodium bicarbonate, mEq/L	10.0 ± 4.1	9.4 ± 4.1	0.26
Serum potassium, mmol/L	5.1 ± 3.2	4.9 ± 1.1	0.50
APACHE II score (mean ± SD)	13.2 ± 5.9	13.9 ± 5.4	0.40

Abbreviations: APACHE, Acute Physiology and Chronic Health Evaluation; BG, blood glucose; BMI, body mass index; eGMS, electronic glucose management system.

Time to DKA resolution (Table 2) was similar in both groups with a median time of 8.6 versus 8.8 hours in the paper-based and eGMS groups, respectively (P = 0.43). Normalization of BG to <250 mg/dL was faster with the paper-based protocol compared with in the eGMS group (median 3.2 vs 5.6 hours, P < 0.001). Furthermore, the time from insulin infusion initiation to the first insulin glargine dose was faster with a median time of 17.3 hours in the paper-based group and 20.4 hours in the eGMS group (P = 0.002). Evidence of DKA recurrence was not seen in either group in the 24-hour period after transitioning to SUBQ insulin.

Table 2.

Time to Transition From IV to SUBQ Insulin.

	Paper (n = 133)	eGMS (n = 84)	P value
Time from insulin infusion start to BG <250 mg/dL, h, median (IQR)	3.2 (1.9–5.2)	5.6 (3.5–8.8)	<0.001
Time from insulin infusion start to first anion gap <12, h, median (IQR)	8.6 (4.9–12.5)	8.8 (6.4–13.0)	0.43
Time from insulin infusion start to first glargine dose, h, median (IQR)	17.3 (11.9–24.7)	20.4 (16.0–27.5)	0.002
Time on insulin infusion, h (mean ± SD)	21.1 ± 11.8	25.3 ± 10.7	0.42

Abbreviations: BG, blood glucose; eGMS, electronic glucose management system; IQR, interquartile range; SUBQ, subcutaneous.

Although not statistically significant (Table 3), hypoglycemia occurred more frequently in the paper-based group compared with eGMS during the insulin infusion (14 vs 3 patients, P = 0.06). There were 3 patients with severe hypoglycemia during the insulin infusion in the paper-based group, with 1 experiencing both a hypoglycemic (BG between 41 and 70 mg/dL) and severe hypoglycemic (BG ≤40 mg/dL) event. In the first 24 hours following transition to SUBQ insulin, 20 patients experienced hypoglycemia in the paper-based group compared with 6 patients in the eGMS group (P = 0.08). Blood glucose between 41 and 70 mg/dL was also more frequent in the paper-based group (19 vs 5 patients, P = 0.06). The rates of severe hypoglycemia following transition were similar between groups.

Table 3.

Hypoglycemia Episodes, ICU LOS, and Hospital LOS.

	Paper (n = 133)	eGMS (n = 84)	P value
Hypoglycemia during infusion a
All BG <70mg/dL, patients	14 (10.5)	3 (3.6)	0.06
BG 41-70 mg/dL, patients	12	3	0.12
BG ≤40 mg/dL, patients	3	0	0.29
Hypoglycemia following transition to SUBQ insulin for the first 24 h a
All BG <70mg/dL, patients	20 (15.0)	6 (7.1)	0.08
BG 41–70 mg/dL, patients	19	5	0.06
BG ≤40 mg/dL, patients	2	1	1.00
ICU LOS (hours), median (IQR)	36.5 (23.4-53.2)	41.4 (29.2-54.8)	0.05
Total hospital LOS (hours), median (IQR)	67.9 (42.8-108.2)	77.8 (53.2-113.2)	0.05

Abbreviations: BG, blood glucose; eGMS, electronic glucose management system; IQR, interquartile range; LOS, length of stay; SUBQ, subcutaneous.

^aOne patient had a BG both between 41 and 70 mg/dL and ≤40 mg/dL.

The median ICU (36.5 vs 41.4 hours; P = 0.05) and median hospital length of stay (67.9 vs 77.8 hours; P = 0.05) were shorter in the paper-based group compared with the eGMS group.

Discussion

This study demonstrated comparable timing of DKA resolution among patients managed with either a paper-based protocol or eGMS at our center. While it did not reach statistical significance, this study demonstrated a trend favoring fewer hypoglycemic events when eGMS was used over a paper-based DKA protocol. There were also modest reductions in ICU length of stay with a difference of 4.9 hours between the groups and a difference of 9.9 hours for total hospital length of stay favoring paper-based treatment over eGMS, which is in contrast to previous literature.

A retrospective, multicenter study identified that patients with DKA managed with eGMS had a faster time to normalization of BG (9.1 ± 8.9 vs 10.97 ± 10.2 hours, P = 0.001), resolution of metabolic acidosis (13.6 ±11.8 vs 17.6±19.6 hours, P = 0.001), and decrease in hospital length of stay (3.2 ± 2.9 vs 4.5 ± 4.9 days, P = 0.01) when compared with a paper based protocol. ⁵ These findings should be interpreted with caution, as this study evaluated outcomes from 34 centers, each using an institution-specific paper-based DKA protocol. A second retrospective review of patients with DKA found a shorter time to DKA resolution (6.6 ± 3.7 vs 7.3 ± 4 hours, P =0.02) and a decrease in both mild hypoglycemia (glucose <80 mg/dL, 1% vs 8%) and severe hypoglycemia (glucose <50 mg/dL, 12% vs 54%) in those treated with eGMS compared with a paper-based protocol. ⁶ Although these 2 retrospective studies demonstrate statistically significant findings for faster time to normalization of BG and shorter time to DKA resolution when eGMS was used, clinical significance is questionable given the differences of only 1 to 2 hours between groups. Furthermore, the potential for confounding variables in both studies remains high in the setting of inconsistent and varied institution-specific paper-based DKA protocols.

In contrast to the current body of published evidence evaluating eGMS use for patients with DKA, we evaluated the severity of illness between groups via the APACHE II scores, as a patient’s acuity of illness may impact the duration of both ICU and hospital length of stay in the scope of DKA treatment. This is the first study to our knowledge that included APACHE II scores as a component of baseline characteristics for their respective cohort.

In the retrospective study by Ullal et al, patients were differentiated into mild, moderate, and severe DKA based on published diagnostic criteria; however, the groups appeared to be unbalanced, with more conventional protocol patients classified with mild DKA compared with the eGMS cohort, which were predominately classified as severe DKA. This imbalance makes comparison difficult to interpret as one would expect a longer length of stay in the more severe population which was not reflected in that study. ⁵ When using the same published diagnostic criteria, the severity of DKA in the present study was similar between paper-based and eGMS-treated patients, with the minority having mild DKA and the majority being diagnosed with severe DKA. ¹ Another consideration to take into account is the possible effect of a closed versus open ICU system. Most eGMS DKA studies evaluated patient populations managed in an open ICU. ⁶ Our center is a closed-system ICU that potentially has more standardized practice patterns given patient admittance is solely under an intensivist compared with an open-system ICU where intensivists are solely available for consultation. The impact of a closed versus open ICU practice model on length of stay remains unclear and may be an area for future study.

Given the ongoing concerns of institution- and user-specific factors that may affect the efficacy of eGMS, the continuous adjustments of insulin by eGMS minimize human-related errors that may occur with interpretation of a paper-based protocol. This is particularly important in the treatment of DKA, given that frequent insulin rate titrations are needed to treat hyperglycemia and resolve ketoacidosis. ³ The initial BG ranges and eGMS multiplier at our center for DKA were set compared with the prior DKA study by Ullal et al that used BG targets between 120 and 180 mg/dL and multipliers ranging from 0.01 to 0.03. The aforementioned study used ranges of BG targets and multipliers to try to determine which range was most efficacious while preventing hypoglycemic events. Of note, 2 of the 84 patients on eGMS had a manual entry adjustment of their IV insulin multiplier performed in the middle of their DKA treatment although they were already established on IV insulin. Justifications for the reductions in multipliers were not readily available, but presumed due to safety concerns. None of the patients in the eGMS cohort had eGMS discontinued and then re-entered to start from scratch; this process has been known to occur at our center as an attempt to reset the insulin dosing due to perceived safety concerns from staff. Manual entry adjustments of eGMS multipliers have not been discussed or evaluated in prior published studies.

Hypoglycemic events in the present evaluation are similar to those found in other DKA studies that compared institution-specific, paper-based protocols with eGMS. The difference in hypoglycemic events seen in our patient cohorts, although not statistically significant, is hypothesis-generating for an area of future study given the overall trend favoring eGMS for fewer hypoglycemic events. Episodes of hypoglycemia with glucose ≤40mg/dL occurred in 5 patients in the paper-based protocol and 1 patient in the eGMS group. Upon closer investigation, the paper-based protocol used communication orders in a section of the EMR not associated with medication or medication administration to direct staff to adjust insulin infusions per the DKA protocol without clear and specific guidance located within the actual order as seen in eGMS. Communication orders do not repeatedly flag or alert the staff to actionable items; the increase in hypoglycemic events in the paper-based cohort may attributable to this EMR limitation. Although it is difficult to determine the adherence to eGMS, given that eGMS provides specific actionable items like timing of BG checks and titration adjustments, deviations from protocol are likely much less frequent. Of note, adherence to insulin titrations and hypoglycemic treatment for the paper-based DKA protocol was not assessed in this evaluation. Manual entry adjustments of eGMS multipliers occurred in a number of instances as previously described; however, the impact of this is unclear. When looking closer at the BG both before and after the manual adjustments, the rate changes did not appear to have a noticeable impact on the BG trend in either direction, making it hard to assess and infer the rationale behind the adjustments.

There are a number of limitations to this study to note. First, this was a retrospective, single-center chart review using an institution-specific paper-based protocol, which may limit generalizability to broader populations of patients with DKA managed with IV insulin due to the small sample size. Second, as the paper-based protocol was not fully documented in the EMR, data regarding protocol adherence, provider and/or personnel manipulation, or deviation were unable to be obtained. Interruptions in insulin infusions due to electrolyte derangements that may have artificially prolonged the duration of insulin infusions were not evaluated and may have influenced findings. Although the frequency of hypoglycemia was assessed in this study, appropriate and timely treatment of hypoglycemic events was not reviewed or evaluated. Inpatient insulin glargine dosing compared with home insulin dosing was not collected or compared, so the extent of insulin resistance cannot be inferred. The presence of unequal sample sizes may have also affected statistical power and validity. Lastly, cost savings evaluations around implementation of eGMS, including factors such as hypoglycemic events and length of stay, were not evaluated.

Conclusion

Patients had similar rates of DKA resolution when managed with a paper-based protocol compared with eGMS. Although ICU and total hospital length of stay were shorter in the paper-based group, the higher incidence of hypoglycemia both during the insulin infusion and after the transition in this group is worth investigating in further studies as hypoglycemia may be associated with worse outcomes including, but not limited to, seizure, coma, myocardial infarction, residual neurological impairment, or death. ⁸ The paper-based protocol overall decreased both ICU and hospital length of stay; however, eGMS had improved clinically relevant safety outcomes. Future prospective studies that also evaluate cost savings would be beneficial to both validate findings and determine safe and efficient management of DKA.