Abstract
Background:
Inpatient hyperglycemia is common and associated with adverse outcomes. Subcutaneous insulin remains the mainstay of treatment for most non-critically ill patients, yet appropriate dosing is complex and prone to variation. Clinical decision support (CDS) tools may improve insulin prescribing, but few have been evaluated in real-world settings.
Methods:
We conducted a retrospective cohort study of adult inpatients with diabetes and hyperglycemia admitted to 2 academic medical centers between December 1, 2015, and July 30, 2018. We assessed uptake, effectiveness, and adherence to an optional subcutaneous insulin CDS tool embedded in the Epic electronic medical record (EMR). Propensity score matching with Crump trimming and 1:3 nearest-neighbor matching was used to compare glycemic outcomes between patients who used the tool within 24 hours of admission and matched controls.
Results:
Of 5485 eligible admissions, 662 (12.1%) involved tool use on day 1. Tool use was more common among patients with type 1 diabetes, chronic kidney disease, or higher baseline glucose and varied by diagnosis and provider type. Among matched patients (n = 1750), tool use was associated with greater odds of any euglycemia (70-180 mg/dL) on day 2 (75.1% vs 69.8%; odds ratio [OR] = 1.30, 95% CI = 1.04-1.63; P = .02). Mean glucose levels, hypoglycemia, and hyperglycemia did not differ significantly. Only 43.7% of patients for whom the tool was used received a total daily insulin dose within 20% of the user-selected dose.
Conclusion:
Use of an EMR-integrated insulin CDS tool was associated with modest glycemic improvements. Low adherence may have attenuated its clinical impact.
Keywords
Introduction
Hyperglycemia affects up to 40% of hospitalized patients and is associated with increased in-hospital mortality, longer length of stay, health care-associated infections, acute kidney injury, stroke, and post-discharge disability.1-4 Insulin remains the primary therapy but carries risk due to its narrow therapeutic index. Inpatient dosing must account for renal function, glucocorticoid use, nutritional intake, and prior insulin requirements, and limited clinician expertise may increase dosing errors. 5 Inpatient glycemic control is also increasingly relevant to hospital quality reporting, as Centers for Medicare & Medicaid Services (CMS) inpatient dysglycemia measures include severe hyperglycemia (>300 mg/dL) and severe hypoglycemia (<40 mg/dL). 6
Clinical decision support (CDS) tools such as inpatient insulin dosing calculators may standardize insulin prescribing and promote guideline-concordant care. 7 Most prior studies have focused on intravenous insulin algorithms in critical care, 8 although the majority of hospitalized patients receive subcutaneous insulin outside intensive care settings. Commercial platforms have demonstrated effectiveness but typically operate outside the electronic medical record (EMR), potentially limiting workflow integration.9-11
Our institution developed an EpicCare-embedded subcutaneous insulin CDS tool to guide dosing in non-critically ill adults with diabetes. 12 The tool generates evidence-based recommendations based on patient characteristics and is integrated into admission workflows to facilitate individualized basal-bolus prescribing. However, its real-world uptake and impact on glycemic outcomes have not been fully evaluated. We therefore conducted a retrospective cohort study to assess the utilization, effectiveness, and adherence of the CDS tool. 12
Methods
Study Design and Population
We conducted a retrospective cohort study of adults admitted to Johns Hopkins Hospital or Johns Hopkins Bayview Medical Center from December 1, 2015, through July 30, 2018. The Institutional Review Board approved the study with a waiver of informed consent.
Eligible admissions met the following criteria: (1) diagnosis of diabetes mellitus, (2) hospital length of stay ≥2 and ≤10 days, (3) ≥4 blood glucose measurements during admission, (4) receipt of ≥1 unit of subcutaneous insulin, (5) available weight data, and (6) hyperglycemic blood glucose (≥180 mg/dL) value recorded within the first 24 hours of admission. This analytic cohort was therefore not intended to represent all hospitalized patients with hyperglycemia, but rather patients for whom the subcutaneous insulin CDS tool was clinically relevant and evaluable. Exclusion criteria included pregnancy, diabetic ketoacidosis or hyperosmolar hyperglycemic state, cystic fibrosis, intensive care unit (ICU) admission, intravenous insulin or insulin pump use, Nil Per Os (NPO) status during admission, or missing blood glucose data.
Clinical Decision Support Tool
The CDS tool is an Epic SmartForm linked to the subcutaneous insulin order set. 12 A detailed description and sample workflow are provided in the Supplement.
In this analysis, the exposure of interest was use of the CDS tool on hospital day 1 (within 24 hours of admission); patients without tool use in this window served as the comparison group for effectiveness analyses. We selected a 24-hour window to capture when initial basal-bolus regimens are typically initiated. In addition, key variables from the tool were examined in the adherence analysis. The recommended total daily dose (TDD) refers to the body mass index (BMI)-specific dose highlighted in green within the displayed range of weight-based estimates; the user-selected TDD is the provider-entered value based on these options; and the administered TDD is the total insulin delivered over 24 hours.
Utilization-Related Outcomes
We compared baseline demographic and clinical characteristics between encounters with and without tool use. Variables included age, sex, race, diabetes type, BMI, renal function, glycated hemoglobin (A1C), home insulin use, early blood glucose values, admitting service, and provider type. In addition, among encounters where the tool was used within 24 hours, we described patterns of use. Outcomes included provider type completing the tool, admitting service, and primary admission diagnosis.
Clinical Effectiveness-Related Outcomes
Effectiveness outcomes included any euglycemia on day 2, defined as having at least 1 blood glucose value in the target range (70-180 mg/dL), geometric mean blood glucose on day 2, and the percent change in geometric mean blood glucose from day 1 to day 2. Daily geometric mean blood glucose was calculated as the exponentiated mean of the natural log-transformed blood glucose values for each hospital day. We also evaluated the presence of hyperglycemia (blood glucose >180 mg/dL), severe hyperglycemia (blood glucose >300 mg/dL), hypoglycemia (blood glucose <70 mg/dL), and severe hypoglycemia (blood glucose <40 mg/dL) on day 2. Among patients with a hospital length of stay of at least 4 days, we calculated the average of daily geometric mean blood glucose values from days 2 through 4 as an additional secondary outcome. We did not assess all glucose values in the target range as an outcome because few patients met this criterion.
All blood glucose values were derived from point-of-care and serum measurements obtained through routine clinical care and were aggregated into daily geometric means. Point-of-care glucose values were measured using the Nova StatStrip glucose meter (Nova Biomedical, Waltham, Massachusetts) at both hospitals.
Adherence-Related Outcomes
The CDS tool does not auto-populate insulin orders; therefore, clinicians must manually re-enter the insulin doses that they selected within the CDS, when placing the actual orders. This design introduced the potential for discordance between the dose recommended by the tool and the dose ultimately prescribed. Across all encounters, we assessed execution fidelity by calculating the proportion of cases in which the administered TDDs was within ±20% of the user-selected TDD in the CDS. We also computed the ratio of administered to selected TDDs to evaluate systematic deviations in either direction.
Because the CDS tool supports multiple TDD estimation strategies, the approach to assessing adherence differed by method (Figure 1). For weight-based dosing, the tool displays a calculated dosing range based on weight multipliers (ie, unit/kg/day) and highlights a BMI-based recommendation in green. This allowed us to define a “recommended TDD” against which both provider-selected and administered TDD could be compared (Supplemental Figure 1). In contrast, for home-based or other non-weight-based strategies, no standardized recommendation was provided by the tool, and adherence could only be assessed based on concordance between selected and administered doses.
Classification schema for provider adherence to subcutaneous insulin-dosing recommendations.
For encounters using weight-based dosing, we compared the recommended, selected, and administered TDDs and classified adherence into 5 categories: (1) Full concordance – the user selected a dose within the recommended range and a similar TDD was administered to the patient; (2) Deviation After Selecting Recommended Dose – the user selected a dose within the recommended range but a different TDD (not within ±20%) was administered to the patient; (3) Full Discordance – the user selected a dose outside the recommended range and a different TDD (not within ±20% of the selected TDD) was administered to the patient; (4) Execution of Non-Recommended Dose – the user selected a dose outside the recommended range and a similar TDD was administered to the patient; and (5) Reversion to Recommended Dose – the user selected a dose outside the recommended range, but a TDD within the recommended range was administered to the patient.
To further quantify adherence, we calculated the ratio of both the selected and administered TDDs to the BMI-based recommended TDD and recorded whether these values fell above or below the recommended range.
Statistical Analysis
Baseline characteristics were compared between patients who received the insulin dosing tool within 24 hours of admission and those who did not. Categorical variables were compared using chi-square tests. For continuous variables, 2-sample t-tests were applied to normally distributed variables, while Wilcoxon rank-sum tests were used for non-normally distributed variables. Variation in tool uptake across subgroups was described using stratified cross-tabulations for variables such as provider type, admission service, and primary diagnosis. Adherence to the tool’s recommended insulin dose was summarized using frequencies and proportions.
To reduce confounding, we performed propensity score matching. The propensity score, defined as the conditional probability of CDS tool use on hospital day 1 given observed covariates, was estimated using multivariable logistic regression including demographics (age, sex, race/ethnicity), diabetes type, renal function (chronic kidney disease [CKD] and end-stage renal disease [ESRD]), baseline glycemic status, outpatient insulin use, day 1 diet type, admitting and discharge service, provider type, and primary admission diagnosis. We applied 2-tailed Crump trimming (0.1-0.9) to ensure common support. 13 Patients were matched 1:3 using nearest-neighbor matching with replacement and a caliper of 0.2 standard deviations of the logit of the propensity score.14,15 Covariate balance was assessed using standardized differences, with <10% indicating adequate balance.16,17 Outcomes were analyzed using linear regression for continuous variables and logistic regression for binary variables and are reported as mean differences or odds ratios with 95% confidence intervals (CIs).
Adherence patterns among recommended, selected, and administered TDDs were summarized using descriptive statistics. Categorical adherence classifications were reported as frequencies and proportions. Continuous variables, including TDD values and ratios to the recommended TDD, were summarized using means, standard deviations, medians, and interquartile ranges.
As a subgroup analysis, we evaluated adherence and effectiveness outcomes among patients with CKD or ESRD.
All reported P-values are 2-sided. All analyses were conducted using Stata/MP version 17.0 (StataCorp, College Station, Texas).
Results
Study Population
Of 74 743 admitted patients, 5485 met eligibility criteria (Figure 2). The median age was 63 years, the median BMI was 30.3 kg/m², 51% were female, and approximately half of the patients identified as white (51%). Most patients had type 2 diabetes (98%), 38% had CKD, and 11% had ESRD. Median first blood glucose was 202 mg/dL, day 1 geometric mean glucose was 194 mg/dL, and day 1 maximum glucose was 264 mg/dL (Table 1).
Study flow diagram.
Baseline Characteristics of Total Study Population and Matched Population.
Values are presented as number (%), where otherwise not specified. A1C data were available for 258/662 (39.0%) tool users and 2046/4823 (42.4%) non-users in the unmatched cohort, and for 230/606 (38.0%) tool users and 475/1144 (41.5%) non-users in the matched cohort.
A1C data were available for 258/662 (39.0%) patients for whom the tool was used and 2,046/4,823 (42.4%) patients for whom the tool was not used in the unmatched cohort, and for 230/606 (38.0%) patients for whom the tool was used and 473/1,144 (41.3%) patients for whom the tool was not used in the matched cohort.
Abbreviations: BMI, body mass index; IQR, interquartile range; CKD, chronic kidney disease; ESRD, end-stage renal disease; Ob/Gyn, obstetrics and gynecology; ICD-10, International Classification of Diseases, 10th Revision.
Utilization Outcomes
Patients in whom the insulin dosing tool was used were slightly younger (median 62 vs 63 years, P =.033) and more likely to have type 1 diabetes (6.3% vs 0.7%, P < .001), CKD (47.1% vs 37.9%, P < .001), and ESRD (17.7% vs 11.2%, P < .001). There were no significant differences in sex, race, BMI, or home insulin use. Patients in whom the tool was used had higher day 1 maximum glucose (281.5 vs 264.0 mg/dL, P < .001) but similar initial and day 1 mean glucose values. Microvascular complications were more common among patients for whom the tool was used (77.3% vs 67.8%, P < .001); other baseline characteristics, including provider service, were similar.
Among encounters where the insulin dosing tool was used, the majority were completed by attendings (41%) and residents (40%), followed by physician assistants (7%) and fellows (10%). Uptake of the tool varied modestly by admitting service, with similar use rates observed across medicine (12.9%), surgery (12.9%), and other services (11.8%). Tool use was most frequent among patients admitted for endocrine/metabolic conditions (15.9%) and genitourinary diseases (16.5%), compared to lower use among those with respiratory (6.6%) or eye/ear (7.7%) diagnoses. Monthly uptake remained consistently modest over the study period (Supplemental Figure 2).
The most common method for estimating the TDD of insulin was the patient’s home regimen (77.2%), followed by weight-based estimation (19.8%) and prior 24-hour subcutaneous insulin use (3.0%).
Clinical Effectiveness Outcomes
After Crump trimming and propensity score matching, 56 were excluded due to a lack of common support in the propensity score distribution, and 606 patients for whom the tool was used were matched to 1144 controls with adequate covariate balance (all standardized differences <10%; Table 1).
In the matched cohort, tool use was associated with higher odds of achieving any euglycemia on hospital day 2 (75.1% vs 69.8%; odds ratio [OR] = 1.30, 95% CI = 1.04-1.63; P = .021). There were no significant differences in rates of hyperglycemia (>180 mg/dL), severe hyperglycemia (>300 mg/dL), hypoglycemia (<70 mg/dL), or severe hypoglycemia (<40 mg/dL) on day 2. Mean geometric glucose on day 2 and percent change from day 1 did not differ significantly between groups. Among patients with the length of stay ≥4 days, mean glucose across days 2 to 4 was also similar (Table 2).
Clinical Decision Support Tool Effectiveness Outcomes.
At least 1 blood glucose value between 70 and 180 mg/dL on hospital day 2.
Blood glucose value >180 mg/dL on hospital day 2.
Blood glucose value >300 mg/dL on hospital day 2.
Blood glucose value <70 mg/dL on hospital day 2.
Blood glucose value <40 mg/dL on hospital day 2.
Among patients with length of stay ≥4 days (n = 313 for patients for whom the tool was used; n = 610 for controls).
Adherence Outcomes
Among encounters with CDS tool use within 24 hours (n = 662), concordance between user-selected and administered TDD (±20%) occurred in 43.7% of cases. Adherence was slightly higher with home regimen-based dosing (45.7%) than with weight-based dosing (38.2%) (Figure 3). The median administered-to-selected TDD ratio was 0.67 and 0.90 on days 1 and 2, respectively, indicating underdosing relative to selected doses.
Adherence distributions by (A) Home-based Insulin Dosing and (B) Weight-Based Insulin Dosing.
In weight-based encounters (n = 131), adherence categories were distributed as follows: Full Concordance, 26.0%; Deviation After Selecting Recommended Dose, 47.3%; Reversion to Recommended Dose, 13.7%; Execution of Non-Recommended Dose, 6.9%; and Full Discordance, 6.1%.
The median administered-to-recommended ratios were 0.56 and 0.78 on days 1 and 2, respectively.
Subgroup Analysis in Patients With Chronic Kidney Disease and End-Stage Renal Disease
Among CKD/ESRD encounters in which the tool was used, overall TDD concordance occurred in 129 of 312 encounters (41.3%), including 109 of 257 home-based dosing encounters (42.4%) and 17 of 45 weight-based dosing encounters (37.8%) (Supplemental Tables 1 and 2). In the matched CKD/ESRD subgroup (n = 840; 289 encounters in which the tool was used and 551 controls), tool use was associated with higher day 2 euglycemia (82.7% vs 75.1%), lower day 2 hyperglycemia >180 mg/dL (85.1% vs 90.4%), lower severe hyperglycemia >300 mg/dL (26.6% vs 33.8%), and lower day 2 geometric mean blood glucose (188.9 vs 201.0 mg/dL), without significant differences in hypoglycemia <70 mg/dL or severe hypoglycemia <40 mg/dL (Supplemental Table 3).
Discussion
In this retrospective cohort study of hospitalized adults with diabetes and hyperglycemia, use of a subcutaneous insulin CDS tool within 24 hours of admission was associated with limited improvements in glycemic outcomes. Although patients for whom the tool was used had higher odds of achieving any euglycemia on hospital day 2, no significant differences were observed in mean blood glucose, rates of hyperglycemia or hypoglycemia, percent change in glucose from day 1 to day 2, or mean glucose across hospital days 2 through 4 compared with matched controls. These findings suggest that the clinical impact of this CDS tool, as implemented in routine practice, was modest. In subgroup analyses, findings appeared more favorable among patients with CKD/ESRD, with improvements across multiple glycemic outcomes suggesting that patients with renal insufficiency may derive greater benefit from structured insulin dosing support.
The adherence findings may help explain these limited clinical effects. Concordance between user-selected and administered TDD occurred in fewer than half of tool encounters, suggesting that implementation and execution challenges, rather than dosing logic alone, constrained effectiveness. In weight-based encounters, providers commonly selected a TDD within the recommended range, yet dosing frequently diverged at order execution, and administered-to-recommended TDD ratios were consistently below 1.0, indicating systematic underdosing. This pattern is consistent with conservative dose adjustment at order entry, potentially to mitigate hypoglycemia risk.
Implementation design may have contributed to these execution gaps. Although embedded within the EMR, the CDS tool did not populate insulin orders, requiring manual transcription of the selected dose into the order set. This was due to limitations in Epic SmartForm functionality, which does not allow automatic population of medication doses from SmartForm outputs, and to institutional CDS governance requirements mandating manual dose entry as an additional safety safeguard. This added step increases friction and creates opportunities for both transcription error and intentional modification. This manual translation step reflects the specific Epic SmartForm infrastructure used in our implementation and differs from some commercial glycemic management systems, which may use standalone, browser-based, HL7-integrated, or point-of-administration workflows. Unlike CDS systems that automate order population, 8 a display-only approach may be less likely to change prescribing behavior, consistent with the “last mile” problem in CDS implementation. 18
Additional factors may have contributed to the tool’s limited success. Clinician inertia, time constraints, and competing priorities during busy hospital workflows may have reduced uptake of the tool. Moreover, because the tool was optional, clinicians could bypass it entirely, rely on alternative insulin ordering pathways within the EMR, or use parallel supports for inpatient glycemic management, including consultation with the Inpatient Diabetes Management Service (IDMS). Inexperience or discomfort with insulin titration may have led to conservative dosing despite tool guidance. Institutional culture, lack of reinforcement through feedback mechanisms, and absence of real-time support from diabetes specialists may have further undermined adoption and adherence. Furthermore, the tool did not integrate continuous glucose monitoring data, which was not routinely implemented at our institution and may represent future directions in inpatient insulin management. 19 Artificial intelligence (AI)-enabled decision support tools that automate data extraction and generate personalized insulin recommendations with minimal user input may improve adoption but require prospective evaluation. 20
Most users in this study selected the home insulin dosing approach, which required manual entry and provided minimal decision support. Weight-based estimation, which the tool was designed to support through automated calculations and a highlighted BMI-based recommendation, was used in only 20% of encounters. This is notable given guideline recommendations supporting weight-based dosing in hospitalized patients3,21 and trial evidence demonstrating improved glycemic control compared with home dose-based approaches. 22 Observational data further suggest that, particularly in patients with higher outpatient insulin requirements, weight-based approaches may be a preferred strategy. 23 Thus, broader adoption of this structured functionality in the tool may be necessary to realize its full clinical impact on insulin prescribing.
The CDS tool was intentionally designed to preserve provider autonomy. Its non-mandatory structure and absence of forced logic reflected early feedback from residents and hospitalists, who emphasized the importance of maintaining workflow flexibility and avoiding perceived intrusions into clinical judgment. 12 However, these same features may have limited the tool’s impact. Passive CDS designs, such as those that rely solely on information display, may be insufficient to change prescribing behavior. 24 Incorporating soft stops or inline alerts when users override guideline-based recommendations may help preserve autonomy while promoting safer, more consistent insulin dosing.
Previously studied CDS tools for subcutaneous insulin dosing in the hospital setting vary widely in their integration and dosing approach. 25 Our EpicCare SmartForm uses transparent rule-based dosing options such as weight/BMI-based estimates, prior insulin requirements, home insulin regimen, and nutrition status; commercial glycemic management systems vary in both workflow integration and algorithmic inputs. Publicly available descriptions of commercial systems describe features, although exact weighting and proprietary calculation logic are not always fully described. GlucoTab is a computerized decision support system for subcutaneous basal-bolus insulin dosing that incorporates glucose values, nutrition status, body weight, renal function, and prior insulin exposure to generate dose recommendations. Prospective studies have demonstrated mean glucose levels of 154 to 159 mg/dL, time-in-range up to 68.8% (70-180 mg/dL), low hypoglycemia rates, and dose-suggestion adherence exceeding 90%11,26,27 Glucommander supports both intravenous and subcutaneous insulin management using patient-specific clinical data, including glucose values, body weight, nutrition status, and prior insulin response. Published studies have reported improved glycemic control and time-in-range compared with provider-managed insulin therapy.10,28-30 EndoTool is an insulin dosing platform that uses a proprietary patient-specific mathematical model incorporating glucose values and other clinical variables. Published outcome data are primarily from intravenous insulin studies and demonstrate improvements in hypoglycemia, hyperglycemia, and time to target.9,25,31
Our observed TDD concordance of 44% is lower than the adherence reported for tightly integrated glycemic management systems such as GlucoTab, in which dose-suggestion acceptance has exceeded 95%. 27 However, these studies generally measure whether clinicians accepted or modified dose recommendations within the software workflow, rather than whether a provider-facing recommendation was ultimately translated into administered insulin over the subsequent hospital day. In contrast, our EMR-embedded tool required manual translation of selected TDD into medication orders, creating additional opportunities for discordance between recommendation, order entry, and administration. More comparable computerized insulin order templates and protocols have shown variable and sometimes limited uptake, with 1 review noting protocol uptake as low as 27%. 32 Thus, the adherence gap observed in our study may reflect a broader implementation challenge for inpatient insulin CDS, particularly when recommendations are not directly coupled to order entry and administration workflows.
This study adds to the limited literature on subcutaneous insulin CDS tools. Study strengths include the large sample size, evaluating over 5000 admissions across 2 academic centers. We used propensity score matching to minimize confounding, yielding a more rigorous estimate of effectiveness than prior pre-post or uncontrolled designs. Finally, we conducted a detailed analysis of adherence – an often underreported aspect of CDS implementation. 33
This study has several limitations. First, unmeasured confounding remains possible because tool use may have reflected unmeasured factors such as clinical complexity or provider comfort with insulin dosing. Second, matching with replacement may have increased the complexity of variance estimation. Third, we lacked encounter-level IDMS consultation data, which may have attenuated observed differences between groups. Fourth, glucose measurements were obtained during routine care and may have varied in timing and frequency. Fifth, we did not assess provider satisfaction, perceived usefulness, or usability. Sixth, incomplete payor and pre-admission A1C data precluded inclusion of these variables in the propensity score model. Seventh, adherence analyses did not capture dose-specific ordering or insulin timing relative to meals, nutrition changes, or glucose measurements. Finally, the study period ended in 2018, which may limit generalizability to current inpatient CDS workflows.
Conclusion
In this retrospective study, use of an EMR-integrated subcutaneous insulin CDS tool was associated with modest improvements in early glycemic control among hospitalized adults with diabetes. Future efforts should focus on improving adherence, integrating recommendations more tightly with order entry, and providing more active CDS. Future studies should evaluate the impact of glucose monitoring practices, integration with specialist diabetes management services, and comparative effectiveness, adherence, provider satisfaction, and implementation costs of CDS tools.
Supplemental Material
sj-docx-1-dst-10.1177_19322968261468379 – Supplemental material for Adherence and Clinical Effectiveness of a Subcutaneous Insulin Clinical Decision Support Tool in Hospitalized Patients
Supplemental material, sj-docx-1-dst-10.1177_19322968261468379 for Adherence and Clinical Effectiveness of a Subcutaneous Insulin Clinical Decision Support Tool in Hospitalized Patients by Benjamin Lalani, Mohammed S. Abusamaan, Andrew P. Demidowich, Mihail Zilbermint, Sudipa Sarkar and Nestoras Mathioudakis in Journal of Diabetes Science and Technology
Footnotes
Abbreviations
A1C, glycated hemoglobin; BG, blood glucose; BMI, body mass index; CI, confidence interval; CDS, clinical decision support; CKD, chronic kidney disease; CMS, Centers for Medicare & Medicaid Services; DKA, diabetic ketoacidosis; EMR, electronic medical record; ESRD, end-stage renal disease; HHS, hyperosmolar hyperglycemic state; HL7, health level 7; ICU, intensive care unit; ICD-10, International Classification of Diseases, 10th Revision; IDMS, inpatient diabetes management service; IQR, interquartile range; IV, intravenous; NPO, Nil Per Os; OR, odds ratio; SD, standard deviation; TDD, total daily dose.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by a grant from the National Institutes of Diabetes and Digestive and Kidney Diseases (K23DK111986).
Declaration of Conflicting Interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: B.L. is a co-founder of Endocrine Technologies LLC, a company not involved in the intervention or tools evaluated in this study. M.Z. reports consulting for DexCom, Inc. All other authors have no declarations of interest.
Supplemental Material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
