Effect of nutrition on postprandial glucose control in hospitalized patients with type 2 diabetes receiving fully automated closed‐loop insulin therapy

Fully automated closed‐loop insulin delivery may offer a novel way to manage diabetes in hospital. However, postprandial glycaemic control remains challenging. We aimed to assess the effect of nutritional intake on postprandial glucose control in hospitalized patients with type 2 diabetes receiving fully closed‐loop insulin therapy. The effects of different meal types and macronutrient composition on sensor glucose time‐in‐target (TIT, 3.9‐10.0 mmol/L) and mean sensor glucose were assessed with hierarchical linear models using a Bayesian estimation approach. TIT was lower and the mean sensor glucose slightly higher, after breakfast compared with lunch and dinner, whereas the insulin dose was higher. Across meals, when carbohydrates were replaced by fat, or to a lesser extent by protein, postprandial glucose control improved. For breakfast, a 3.9% improvement in TIT was observed when 10% of the energy from carbohydrates was replaced by fat. Improvements were slightly lower during lunch and dinner (3.2% and 3.4%) or when carbohydrates were replaced by protein (2.2 and 2.7%, respectively). We suggest that reducing carbohydrate at the expense of fat or protein, could further improve glucose control during fully closed‐loop insulin therapy in hospital.


| INTRODUCTION
The prevalence of diabetes among hospitalized patients is growing worldwide, calling for effective, resource-efficient and safe glucose management strategies. 1 Closed-loop insulin therapy, which is an emerging diabetes treatment modality that autonomously modulates insulin pump therapy based on sensor glucose values, was recently shown to achieve better glycaemic control than conventional management in hospitalized patients with type 2 diabetes. 2,3 Yet, reducing glucose excursions after meals using automated subcutaneous insulin delivery remains a challenge. While nutritional intake is known to have an important effect on postprandial glucose control, 4 the impact of specific meal characteristics during fully automated closed-loop insulin delivery is largely unexplored. Thus, the objective of this work was to assess the effect of nutritional intake on postprandial glucose control in patients with type 2 diabetes receiving fully closed-loop insulin therapy while in hospital.

| Study design and participants
This was an exploratory analysis using a subset of data from a twocentre randomized controlled clinical trial assessing the efficacy of closed-loop insulin delivery versus usual care in glucose control in hospitalized patients with type 2 diabetes. 2 The analysis included data from 39 closed-loop participants at a single centre whose nutritional intake was recorded during the study. The protocol was approved by the local ethics committee. All study-related procedures were performed in accordance with the local ethics standards and with the Declaration of Helsinki. All participants gave written informed consent before study enrolment.

| Data collection and pre-processing
Nutritional intake (meal type, energy and macronutrient content) was assessed by a member of the research team using food records and nutritional information from the hospital menu planning system and food database. Continuous glucose monitoring and insulin delivery data were obtained from downloads of study devices. The postprandial period was defined as 3 h after serving the meal. If participants had a snack within the postprandial period, the period was discarded.
Only records containing at least two postprandial data sets for each meal type and participant were eligible. Days with incomplete meal information were discarded for the calculation of daily energy and macronutrient intake. Postprandial glucose control was quantified by the proportion of time with sensor glucose between 3.9 and

| Statistical analysis
The bibliographic reference to the programming language Stan, to the interface rstan and to the package brms is provided in Appendix S1.
This study is based on the retrospective analysis of a clinical trial and no sample size calculation was performed. Nutritional intake was summarized using descriptive statistics. The effects of meal type and composition on postprandial glucose control were assessed with hierarchical linear models accounting for participant-specific effects. A Bayesian estimation approach is described in detail in Appendix S1.
Separate models were implemented for the different glucose metrics.
To assess the difference in postprandial glucose control between breakfast, lunch and dinner, the metric of interest was included as the dependent variable and meal-specific effects for breakfast, lunch and dinner as the independent variable. TIT was modelled with a beta distribution using a logit-link function, MGL was modelled with a normal distribution using an identity-link function, and insulin was modelled with a normal distribution on the log scale using an identity-link function. To assess the effects of macronutrient composition on glucose control, meal type, meal energy content and proportion of energy from fat and protein were included in a isocaloric substitution model whose coefficients indicate changes in outcomes by replacement of carbohydrate (as percentage of energy content) by fat or protein.
Insulin on board and glucose levels at mealtime were included as adjustments. Insulin on board was calculated as described in Toffanin et al. 5 with an insulin peak time of 89 min 6 and a duration of action of 5 h. For the latter, the infused insulin dose per minute was considered a discrete bolus. In the results, we report the posterior mean and the 95% credible interval (CrI) based on the 2.5% and 97.5% quantile of the posterior distribution.

| Nutritional intake
In total, 822 meals from 39 participants (for characteristics see Table S1 in Appendix S1) were analysed: 272 for breakfast, 284 for lunch and 266 for dinner. In total, five meals and postprandial data were excluded from the analysis from participants that contributed less than two observation per meal type. Total daily energy intake was 1445 ± 594 kcal, from which 43 ± 12% (155 ± 66 g) was carbohydrates, 40 ± 10% (65 ± 31 g) fat and 17 ± 6% (61 ± 26 g) protein.
Carbohydrate content was 62 ± 25 g for breakfast, 52 ± 22 g for lunch and 61 ± 31 g for dinner. Further meal characteristics are reported in Appendix S1.

| Glucose control and insulin doses
TIT was 67.5 ± 30.7%, 74.7 ± 35.0% and 75.4 ± 31.6% following breakfast, lunch and dinner, respectively. MGL was 9.1 ± 1.9 mmol/L following breakfast, 8.8 ± 2.1 mmol/L following lunch, and 8.6 ± 1.9 mmol/L following dinner. Insulin doses were 10.4 ± 7.5, 6.6 ± 6.0 and 5.0 ± 4.1 U, respectively. Further metrics of glucose controls are reported in Appendix S1.   To our knowledge, this is the first study evaluating the effects of meal type and macronutrient composition on glucose control during fully closed-loop insulin delivery. The inferior postprandial glucose control and higher insulin doses after breakfast compared with lunch and dinner suggest that breakfast poses greater challenges to automated systems. A similar pattern with higher excursions after breakfast compared with lunch was observed in previous work exploring fully closed-loop insulin delivery in adults with type 2 diabetes in a controlled research setting. 7 Concordantly, the postprandial rise in glucose in response to identical meals was more pronounced in the morning compared with later in the day in insulin-naïve individuals with type 2 diabetes. 8 The underlying mechanisms remain speculative and possible explanations may include the lack of residual insulin from a preceding meal 9 and higher endogenous glucose production 10 in the morning.

| Effect of meal type on postprandial glucose control
The glycaemic benefits of isocaloric substitution of carbohydrates with fat and protein could be simply the consequence of the corresponding reduction in carbohydrates as the major contributor to postprandial glucose excursions. 11 Nonetheless, carbohydrateindependent effects might be considered. Fat intake slows gastric emptying, thereby delaying and attenuating postprandial glycaemic excursions. 12

ACKNOWLEDGMENTS
We are grateful to study participants for their involvement. Michèle interpreted the results and wrote the manuscript. All authors critically reviewed the manuscript and will ensure that questions related to the accuracy and integrity of any part of the work are appropriately investigated and resolved. LB is the guarantor of this work and takes main responsibility for the integrity and accuracy of the data.

PEER REVIEW
The peer review history for this article is available at https://publons. com/publon/10.1111/dom.14187.