Impact of Meal Insulin Bolus Timing and Bedtime Snacking on Continuous Glucose Monitoring–Derived Glycemic Metrics in Hospitalized Inpatients

Introduction

Hyperglycemia affects 30%–40% of hospitalized patients. ¹ While basal-bolus insulin therapy improves glycemic control and clinical outcomes in hospitalized patients, achieving optimal glycemic control remains a challenge. ² Hospitalized patients experience unique circumstances that impact glucose control, including inconsistent or delayed meals because of illness, procedures, and time off the floor undergoing diagnostic evaluations, as well as medications and comorbidities that may affect appetite. Furthermore, food or snacks may be consumed from sources other than the standardized hospital food service, and prandial insulin administration may not always be ideally matched with meal consumption. For example, prandial insulin may be administered before an anticipated meal, which is subsequently delayed to accommodate an urgent procedure or diagnostic test, resulting in unopposed insulin action and significant hypoglycemic risk. Therefore, appropriate premeal timing of rapid-acting insulin is at risk of being compromised, resulting in insulin: meal mismatch and suboptimal postprandial glucose control. Finally, many inpatient insulin protocols are designed around a standard hospital meal program of three meals per day, providing prandial insulin and basal insulin administration using point-of-care (POC) fingerstick values to direct insulin dosing. ^3,4 Insulin protocols that address nonstandard between meal or evening snacking may be difficult to implement given variable carbohydrate and nutritional content and limited nursing staff time. Anecdotal clinical experience supports the potential negative effects of delayed mealtime insulin administration and evening snacking on inpatient glycemic metrics, although the impact of these two important clinical situations has not formally been evaluated.

Research Design and Methods

We report a post hoc analysis of data from the In-FI trial (ClinicalTrials.gov Identifier: NCT04460326). This trial was a single-center, open-label randomized controlled trial conducted at Boston Medical Center in Boston, MA, which compared the safety and efficacy of Fiasp versus insulin aspart (Novolog) on postprandial glycemic control in hospitalized patients living with type 2 diabetes (T2D). Study participants included hyperglycemic adult subjects (age ≥21 and ≤80 years, body mass index <45 kg/m²), diagnosed with T2D at least 180 days prior to screening. Prior to admission, subjects had to be using one of the following for outpatient diabetes management: insulin, ≥2 oral/injectable agents, or one oral/injectable agent with a hemoglobin A1c of ≥8% within 3 months of enrollment. Exclusion criteria included treatment with glucocorticoids during the index hospitalization, pregnancy or breastfeeding, admission to the intensive care unit, diabetic ketoacidosis, hyperosmolar hyperglycemic state, solid organ transplantation, coronary artery bypass surgery, prior diagnosis of gastroparesis or cirrhosis, acute or chronic kidney disease with a serum creatinine of ≥2 mg/dL at the time of screening, clinically significant nausea and/or vomiting, inability to consume more than 30 g of carbohydrate at each meal, expectation for patient to receive nothing by mouth (NPO) for >24 h, use of continuous or intermittent enteral feeding or parenteral nutrition, use of >1 unit/kg/day of insulin prior to admission, or insulin pump or personal continuous glucose monitoring (CGM) usage within the 2 weeks prior to or during admission.

All enrolled participants (N = 137, 68 Novolog, 69 Fiasp) wore a blinded Dexcom G6 PRO CGM applied by research staff to the upper arm or abdomen for up to 72 h after enrollment. Only those completing at least four meal periods with available CGM data were included in the primary analysis, as well as considered for this post hoc analysis (N = 122). The primary CGM-derived outcome measure was the amount of time spent in the glucose range of 100–180 mg/dL in the 4-h postprandial period, reported in a separate publication. ⁵ Eligible participants were randomized (open label) to receive either Fiasp or Novolog injected with meals multiple times a day, in addition to a once-daily bedtime injection of insulin glargine. The research protocol was approved by the Boston University Medical Center Institutional Review Board.

As part of the In-Fi trial protocol, subjects were offered the standard hospital diet consisting of a menu of food choices with three meals a day, each containing 75 g of carbohydrates per meal. Snacks were not specified or provided as part of the protocol, and snacking in between meals was discouraged. Subjects could, however, request and receive snacks from hospital staff or visitors as per usual hospital practice, and the period of day where snacking was most likely to occur was in the postdinner and overnight period. Rapid-acting insulin was not administered to cover these snacks, which is a common real-world occurrence in hospitalized inpatients living with T2D, even though snacks may be covered in some hospitals, particularly for individuals living with type 1 diabetes. Inpatient nursing staff were directed to administer rapid-acting insulin within 5–15 min before meals, and doses were adjusted per protocol, based on POC blood glucose using the hospital glucometer, Nova Statstrip (Nova Biomedical, Waltham, MA). Our hospital nurse-to-patient ratio varies based on hospital location but is typically one nurse for every five patients in non-intensive care unit wards.

For all analyses, the initial 12 h of CGM data was excluded, and only subjects who completed at least 4 meals, with available CGM data while enrolled in the study, were included in these analyses (n = 122). Food intake (snacks or meals) was inferred using the following heuristic based on meeting all of the following CGM criteria: (1) initial rate of glucose rise ≥1 mg/dL/min, (2) rate of rise ≥0 for at least 15 min, and (3) peak glucose ≥30 mg/dL above initial glucose within 1 h. This heuristic was not formally validated but corresponded well with intuitively identified postprandial glycemic excursions reviewed by clinicians with extensive CGM experience. Exact timing of meal consumption (onset and completion) was not collected, but timing of rapid-acting insulin administration was collected via the electronic medical record (EMR).

For rapid-acting insulin bolus timing analyses, meals were included if they had 4 h of complete CGM data following the start of the meal, without any missing data points (n = 109 participants, 261 meals). The start of a meal was defined either by the administration of prandial insulin (according to EMR timestamps) or by a heuristically identified meal, whichever occurred first. A bolus was classified as late if it was administered more than 5 min after the detected onset of the meal-related glucose excursion.

For bedtime snacking analyses, a single night from day 1 to day 2 (with CGM placement performed on day 0) was assessed for each subject, and subjects were included who had a POC bedtime glucose reported on day 1 and a POC fasting glucose reported on day 2 (n = 100). Subjects who received bedtime rapid-acting insulin to correct for hyperglycemia on the night being studied (day 1) were excluded to reduce confounding (n = 32 excluded). Multivariable linear regression models were used to examine the association of bedtime snacks (9 pm–12 am, identified using the heuristic) with overnight (9 pm–6 am) sensor glucose time in range (TIR_70–180) and fasting POC glucose concentrations the next morning among the remaining 68 subjects, adjusting for bedtime (9 pm) POC glucose concentration. Additionally, the mean difference of overnight glucose standard deviation (SD) and percentage coefficient of variation (% CV) between patients, with and without bedtime snacking, were assessed using multivariable linear regression models adjusting for bedtime sensor glucose concentration.

Results

Demographic and clinical characteristics of this cohort are presented in Table 1, which included a cohort of predominantly non-Hispanic Black participants with markedly elevated A1c (>10% at baseline) and >60% baseline insulin use prior to hospital admission. This population is consistent with the underserved, inner-city, safety-net population cared for at Boston Medical Center.

Insulin bolus timing

We analyzed the TIR_70–180 for 261 qualifying meals identified from 109 subjects. In 112 meals (43%), the insulin bolus was delivered late (mean delay: 58.7 min, SD: 33.8 min). The time course of glucose levels for late versus on-time insulin boluses is shown in Figure 1A. For on-time boluses, the mean TIR_70–180 was 48.1% (n = 149), compared with 24.2% when a meal bolus was administered more than 5 min after the start of the meal (n = 112; P < 10⁻⁶).

OPEN IN VIEWER

FIG. 1.

(A) Summary of CGM data for 4-h postprandial periods from 149 meals with on-time insulin boluses (blue line) and 112 with late boluses (red line). Points indicate mean sensor glucose across postprandial periods at the time from meal start specified by the x-axis. Shaded regions indicate 25th percentile to 75th percentile ranges (red for late boluses, blue for on-time boluses, darker shading where the ranges overlap). Mean TIR_70–180 was 48.1% for on-time boluses, 24.2% for late boluses (P < 10⁻⁶). In the sensitivity analysis, the reduction in 4-h postprandial TIR_70–180 with late boluses persisted (effect size: −30.0%, P < 2 × 10⁻¹⁶). (B) Summary of CGM data for overnight (9 pm–6 am) periods from 26 subjects who consumed a bedtime snack (red line), and 42 subjects who did not consume a bedtime snack (blue line). Points indicate mean sensor glucose across subjects at the time specified by the x-axis. Shaded regions indicate 25th percentile to 75th percentile ranges (red for bedtime snack, blue for no bedtime snack, and darker shading where the ranges overlap). Mean TIR_70–180 was 31.6% lower with bedtime snacks than without in a model adjusting for 9 pm sensor glucose (P < 0.001). CGM, continuous glucose monitoring; TIR, time in range.

As a sensitivity analysis, we applied a mixed-effects multivariable regression that included Subject ID as a random effect, with initial glucose and the interaction between insulin type and bolus timing as fixed effects (Supplementary Table S1). The reduction in 4-h postprandial TIR_70–180 persisted in this model (effect size −30.0%, P < 2 × 10⁻¹⁶). In this model, the 4-h postprandial TIR_70–180 was lower for Novolog than for Fiasp (−10.5%, P = 0.043).

Discussion

Delayed mealtime insulin injection and bedtime snacking without insulin coverage are significant causes of postprandial and overnight hyperglycemia in hospitalized inpatients.

A striking finding of our analysis is that despite enrollment in a controlled clinical trial environment, 43% of mealtime insulin boluses were delivered more than 5 min after the inferred beginning of the meal, and in many cases the delay in bolus administration was significantly longer than what would be expected. The 50% mean reduction in TIR_70–180 when a meal bolus was administered more than 5 min after the start of the meal confirms that even with modern rapid-acting insulin analogs, administration of insulin before meals is important to avoid postprandial hyperglycemia in the hospital. Translating this finding into the real-world clinical environment suggests that inpatient clinical teams should prioritize the importance of careful coordination of meal delivery, patient intent-to-eat, and premeal insulin delivery in the hospital environment. Although the inpatient environment is relatively controlled with widespread care process standardization, our experience suggests that synchronizing insulin administration with meal timing is an aspect of care that is challenging to implement compared with when a patient is ambulatory and has more personal autonomy. Our findings support the real-world anecdotal experience of many inpatient clinicians that delayed mealtime insulin bolus administration is common and that further refinement of inpatient insulin administration protocols is needed. While we are unable to determine the cause of the delayed meal insulin boluses, one possibility may be the valid concern by nursing staff of administering meal insulin to a patient who may not eat sufficiently and develop hypoglycemia. Unfortunately, workloads may cause a significant delay in the nurse’s availability to reassess the patient to assure they have eaten and administer insulin. One potential mitigation strategy that may merit further study is to anticipate this challenge by utilizing an ultra-rapid prandial insulin postmeal based on consumed carbohydrate intake. While postmeal carbohydrate counting was not shown to be superior to fixed-meal dosing in a prior study, that study included a rapid-acting insulin analog, and CGM-derived assessments were not available. ⁷

A finding in this post hoc analysis was that Fiasp outperformed Novolog in our sensitivity analysis of 4-h postprandial sensor glucose TIR_70–180. In the primary outcome analysis, noninferiority was demonstrated, but the analysis was not powered to detect superiority. ⁵ The inclusion of bolus timing and initial sensor glucose as factors increased the variability in the primary outcome analysis, reducing the power to detect differences. By accounting for these factors in the statistical model, we decreased residual variability and effectively increased the power of the analysis. These findings can inform future study designs, potentially allowing for smaller, more definitive studies of postprandial glycemic responses with reduced sample sizes.

We also found that evening snacking that is not covered with presnack insulin administration is associated with overnight and fasting hyperglycemia, with higher overnight glucose SD and CV. This finding highlights the challenges of using fasting glucose values to titrate basal insulin doses without taking into consideration the impact of evening snacking. In this trial, all basal insulin glargine doses were administered at bedtime, and we therefore do not know whether there would be any differences if basal insulin was administered in the morning or twice daily. We hypothesize that the increased overnight glucose variability and unrecognized effects of eating at bedtime may increase the risk of “over-basalization” in response to elevated fasting glucose, where clinicians may be inclined to increase basal insulin doses. ^8–9 The clinical implication is that titration of basal insulin based solely on fasting glucose may need to be reconsidered, and standard inpatient insulin protocols should anticipate the possibility of evening or overnight food consumption. Potential mitigation strategies could include providing very low-carbohydrate or no-carbohydrate snack options and addressing the patient on the impact of food consumption outside of meals. Increased incorporation of real-time CGM in hospitalized inpatients may help to address this concern by revealing the totality of overnight glycemic excursions in relation to insulin administration. ¹⁰

Limitations of our analysis include the consideration that meal intake was not directly recorded—it was estimated using CGM sensor data as described, using a novel, unvalidated heuristic analysis of CGM data to extrapolate meal timing. While we are unable to determine the exact time difference between meal initiation and sensor glucose rise, a 5-min minimum is likely reasonable. Furthermore, our definition of a late meal bolus likely underestimates the actual bolus delay, as the meal detection heuristic is based on the glycemic excursion, which typically occurs with some delay after eating. Even though snacking between meals and nonhospital meals brought in from visitors was discouraged in this trial, it did occur. Therefore, our study examined the impact that delayed mealtime insulin had on glycemic outcomes but also examined whether delayed or complete omission of mealtime insulin influences glycemic outcomes.

In summary, evening snacking and delayed mealtime insulin bolus administration significantly impact inpatient glycemic control. Future research is needed to optimize mealtime insulin protocols to address the challenges of timely meal insulin administration in hospitalized patients. This analysis also highlights the importance of addressing bedtime snacks when optimizing glycemic control and adjusting insulin, as well as the valuable potential role of CGM to support improvements in inpatient glycemic control.