Diabetes-specific nutritional formulas for tube feeding in patients with hyperglycemia: a systematic review and meta-analysis

Abstract

This systematic review was conducted to evaluate the effects of diabetes-specific nutritional formulas (DSNF) in tube feeding on multiple outcomes in patients with hyperglycemia in real-world clinical settings. Randomized controlled trials (RCTs) comparing outcomes of patients receiving tube feeding with DSNF versus standard formulas were evaluated. RCTs in critical care and long-term care settings were addressed separately. Outcomes reported by two or more RCTs per clinical setting were evaluated, and those addressed by three or more RCTs were assessed using meta-analysis. Five RCTs allocating 483 patients in critical care and three RCTs allocating 137 patients in long-term care met inclusion criteria. DSNF showed advantages over standard formulas for glycemic control in both populations. Meta-analysis of studies enrolling critically ill patients found reductions in mean blood glucose (mean difference (MD) −0.54 mmol/L (95% CI −0.78 to −0.31), p<0.001; I²=9%; moderate certainty), coefficient of variation of glucose (MD −6.56% (95% CI −12.55% to −0.57%), p=0.032; I²=98%; low certainty) and SD of glucose (−0.77 mmol/L (95% CI −1.49 to −0.06), p=0.035; I²=98%; low certainty). Meta-analysis of the studies enrolling patients in long-term care found a reduction in hemoglobin A1c (−0.92% (95% CI −1.72% to −0.13%), p=0.023; I²=85%; low certainty). Findings for other outcomes were uncertain due to a lack of reporting in the included publications, inconsistent findings across studies, small sample sizes or lack of overall statistical significance. Evidence from this review demonstrated that DSNF in tube feeding provides statistically significant improvements in glycemia in both critically ill and long-term care populations with hyperglycemia compared with standard formulas. However, small effect sizes, imprecision and substantial statistical heterogeneity reduced certainty in the strength of evidence for most outcomes. DSNF is an important clinical tool to control hyperglycemia among critically ill and long-term care patients and should continue to be assessed for effectiveness in improving outcomes.

PROSPERO registration number

CRD420251105442.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Diabetes-specific nutritional formulas (DSNFs) have been shown to improve postprandial glycemic responses in various settings. However, their effects on glycemic control and other cardiometabolic parameters particularly among those requiring tube feeding require further evaluation.

WHAT THIS STUDY ADDS

• This review found that DSNFs may improve glycemic control for people with diabetes or hyperglycemia in critical care and long-term care settings compared with standard formulas, although certainty of evidence was generally low.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Additional studies are needed to strengthen the evidence base and improve understanding of outcomes beyond glycemic parameters.

Introduction

Hyperglycemia is a cardinal biochemical feature of type 1 diabetes (T1D) and type 2 diabetes (T2D) and can also be observed during critical illness. Tube feeding may be required for some people with hyperglycemia to meet their nutritional needs. However, continuous feeding may exacerbate hyperglycemia whether or not a person has diabetes,¹ especially when formulas have carbohydrate content and/or high glycemic index carbohydrates.² Hyperglycemia with tube feeding is common, affecting up to 30% of patients.¹ Moreover, it affects 40%³ to 69%⁴ of patients who are critically ill. Research on the treatment and management of hyperglycemia during tube feeding is limited.⁵

Diabetes-specific nutritional formulas (DSNF) are intended to meet the nutritional needs of people with hyperglycemia and are designed to confer potential benefits for glycemic control. Compared with conventional formulas, DSNF have a lower proportion of calories from carbohydrates, typically use lower glycemic-index carbohydrates and contain greater amounts of monounsaturated fatty acids and fiber. Many also include micronutrients and other components intended to support healthy blood sugar levels, such as myoinositol.

A European Society for Clinical Nutrition and Metabolism (ESPEN) expert group has endorsed the use of DSNFs due to improved glycemic control,⁶ while the Society of Critical Care Medicine (SCCM) and American Society for Parenteral and Enteral Nutrition (ASPEN) recommended the use of standard formulas due to insufficient evidence on DSNFs (table 1).⁷ Additional organizations have published relevant statements for hospitalized patients who are not critically ill. The European Federation of Internal Medicine (EFIM) recommended standard formulas as an acceptable selection for hospitalized adults with hyperglycemia.^8–12 Guidance from other organizations either does not address the use of DSNFs for the management of hyperglycemia in non-critical care inpatient settings¹³ or does not specifically recommend for or against DSNFs due to insufficient research.^{14 15}

Table 1. Recommendations on use of DSNF in critically ill or other populations.

Society	Year	Main finding and recommendations
European Society for Clinical Nutrition and Metabolism (ESPEN)6	2017	Endorsed the use of DSNF in enteral nutrition in people with obesity or diabetes to reduce the risk of hyperglycemia, citing short-term and mid-term studies reporting improved glycemic control.
Society of Critical Care Medicine (SCCM) and American Society for Parenteral and Enteral Nutrition (ASPEN)7	2016	Recommend using standard formulas for enteral administration in adult critically ill patients due to insufficient evidence in the clinical literature of clear benefit of disease-specific formulas, including those designed for people with diabetes.
European Federation of Internal Medicine (EFIM)8–12	2025	Recommended standard formulas as an acceptable selection for hospitalized adults with hyperglycemia because ‘there is no evidence that enteral formulas for persons with [diabetes mellitus])have a significant impact on clinical outcomes.’

DSNF, diabetes-specific nutritional formula.

To date, this topic has not been addressed in a clinically relevant peer-reviewed published systematic review. Previous systematic reviews included participants who did not require tube feeding (eg, they were tube fed for experimental purposes only or orally consumed enteral formulas).^16–19 Their applicability to patients who depend on tube feeding in routine care is, therefore, uncertain. In addition, the most recent systematic search conducted across these reviews was in 2018. Using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework, this systematic review evaluates the effects of DSNF on the nutritional needs of people with hyperglycemia who require tube feeding in clinical settings relevant to real-world practice and rates the certainty of evidence for each outcome.

Research design and methods

This systematic review was conducted using standard methods aligned with the Cochrane Collaboration²⁰ and the National Academy of Medicine²¹ and reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) standards.²² This review was registered in PROSPERO (CRD420251105442).

Data sources and searches

We used free-text and controlled vocabulary terms related to T1D and T2D, hyperglycemia, DSNF and tube feeding in PubMed (includes records from MEDLINE) and the Cochrane Central Register of Controlled Trials (CENTRAL; includes records from Embase) on 1 July 2025. Full strategies are provided in the Online-Only Supplementary Materials (online supplemental table 1) (PubMed) and online supplemental table 2) (Cochrane CENTRAL). In addition, we searched relevant bibliographies, performed hand searches, and invited subject matter experts to submit studies.

Study selection

Randomized controlled trials (RCTs) comparing DSNF with standard formulas in patients with hyperglycemia administered by tube feeding and published in full-length text in peer-reviewed literature were eligible for inclusion. To be included as a DSNF in this report, the formula had to be commercially available and have fewer than 50% of calories from (preferably lower glycemic) carbohydrates. A minimum of 3 months of follow-up was required for hemoglobin A1c (HbA1c) and cardiometabolic risk factors, but no limits were applied to other glycemic outcomes or insulin use. No limits on study size or publication date were used, but English language was required. RCTs addressing other clinical contexts or aspects of DSNF not directly pertinent to this use (eg, mechanistic studies, non-clinical studies, studies on healthy volunteers, ‘simulated’ tube feeding, oral consumption) were excluded for lack of clinical relevance.

We screened studies based on titles and abstracts and evaluated full-text articles using Rayyan application (Cambridge, MA) in duplicate with a consensus process.

Data extraction and analysis

Multiple outcomes in the following domains were sought for extraction: glycemia; body composition; lipids; blood pressure; satiety; changes in diabetes medications; quality of life and treatment satisfaction. Patient demographic and health characteristics, details about formulas and protocols, and general study information were collected.

The appropriateness of quantitative synthesis using pair-wise meta-analysis with CMA Version 4 (BioStat, Englewood, NJ) was considered for each outcome if the following conditions were met:

• There were at least three independent studies that met inclusion criteria and addressed an outcome at a similar follow-up period.

• The studies did not have excessive heterogeneity such that they are too dissimilar to reasonably combine.

• The outcomes were statistically compatible or could be converted into compatibility. Medians or median changes with ranges or IQRs were converted into means or mean changes and SD, and SEs of the mean were converted into SD using standard methods and Meta-Analysis Accelerator.²³ Outcomes reported as mg/dL were converted into mmol/L, which was more commonly reported in this evidence base.

Forest plots display the results of individual studies and overall analysis for each meta-analysis. We used random-effects models for meta-analysis due to the anticipated differences across studies in patient populations and clinical protocols and assessed heterogeneity using I².

Exploration of substantial statistical heterogeneity using meta-regression and/or subgroup analysis or robustness testing was planned per protocol but not performed due to an inadequate number of studies per outcome.

For outcomes addressed by two or more studies for which meta-analysis was not appropriate, we used organizational and logic frameworks to evaluate the findings and the consistency of evidence. We focused on assessing the effect sizes and directions, not p values.

Assessment of risk of bias and strength of evidence

We used signaling questions from Cochrane’s Risk of Bias (RoB) 2 instrument to evaluate the risk of bias in individual studies and to guide the critical appraisal.²⁴ Assessment was performed in blinded duplication with a consensus process.

The methods for assessing confidence in the body of evidence were aligned with the GRADE Approach.²⁵ Key factors in determining confidence in the overall evidence base were:

• Quality, considering individual study risk of bias.

• Consistency, in terms of the observed effect sizes and direction of effect across studies.

• Directness, in terms of relevance to the target population and intervention of interest.

• Precision, including consideration of the number of patients represented.

Results

Search results

Bibliographic database searches identified 123 unique records, of which 9 met inclusion criteria (figure 1). No studies meeting inclusion criteria were identified through gray literature or supplemental searches. Five RCTs enrolled critical care patients and evaluated short-term outcomes within approximately a week or less^26–30 and three RCTs reported in four publications enrolled long-term care patients and evaluated long-term outcomes of approximately 3 months.^31–34

Figure 1. Study selection.

Due to the clinical differences between critical care and long-term care patients and the differences in the duration and objectives of their treatment with tube feeding, we evaluated and presented their characteristics and outcomes separately. Data were also presented in the Online-Only Supplementary Materials for patient characteristics and outcomes for each RCT, and for the composition of each DSNF and control formula for long-term care patients (online supplemental table 3) and critically ill patients (online supplemental table 4).

Patient characteristics

Critical care

The five RCTs^26–30 collectively randomly allocated 483 patients with critical illness and were conducted in intensive care units. The studies enrolled patients with hyperglycemia who, where specified, required tube feeding due to acute ischemic stroke,²⁹ acute respiratory insufficiency or pneumonia,²⁷ mechanical ventilation^{26 28} or tube ventilation.³⁰ Mean patient age ranged from 58 years²⁸ to 72 years.³² Patients required short-term enteral feeding (<72 hours) in three studies, with the other two reporting outcomes at 7 days²⁹ and 9–10 days.²⁸ Studies were performed in Australia,²⁶ Spain,^{27 28} China²⁹ and the Netherlands.³⁰ The RCTs enrolled more men than women overall (range 18%²⁷ to 52%)²⁹ and did not report race or ethnicity.

Long-term care

Three RCTs reported in four publications^31–34 collectively randomly allocated 137 patients in residential long-term care, rehabilitation facilities, or nursing facilities who required tube feeding for at least 3 months, with reported reasons including cerebrovascular accident.^32–34 These patients had either T2D or repeated documented hyperglycemia. The populations had mean ages of 69,³³ 76³⁴ and 81 years.³¹ One study was performed in the USA,³¹ another in Germany,^{32 33} and the third was performed in multiple countries (Israel, United Kingdom, Brazil, United States).³⁴ In general, similar proportions of males and females were enrolled (46%³³ and 64%³⁴ females), and one study did not report sex³¹; information about race or ethnicity was not reported.

Study characteristics

Across long-term care and critical care settings, patients typically received DSNF or a control standard formula by nasogastric tube for total or near-total (defined by studies as ≥75%) nutrition. Two long-term care studies included patients receiving nutrition by percutaneous endoscopy gastrostomy^32–34 and one critical care study included some patients with a transpyloric tube.²⁷

Clinical findings

Reported outcomes differed by clinical setting. Long-term care studies focused on HbA1c, lipid profiles and fasting glucose. Critical care studies typically reported hypoglycemic episodes, mean glucose levels, glucose coefficient of variation (CV) and glucose SD. Both settings frequently reported insulin administration data. No studies reported satiety, body composition, changes in diabetes medications beyond insulin, quality of life or treatment satisfaction. Among the few studies that evaluated nutrition status, such as serum albumin, only one provided actual data.²⁹ Outcomes reported by only one or no RCTs were excluded from this review, since such limited evidence is insufficient to inform evidence-based conclusions. A summary of findings is presented in the following text and in table 2.

Table 2. Summary of findings.

Outcome	Number RCTs (citations)	n=analyzed (DSNF, control)	Summary (meta-analysis results where applicable; mean difference with 95% CIs)	Overall benefit with DSNF?	Strength of evidence†
Critical care population (treatment duration up to approximately 1 week)
Mean blood glucose	426 28–30	230, 173*	−0.54 mmol/L (−0.78 to −0.31)‡, p<0.001, I2=9%	Yes	Moderate
Coefficient of variation (CV) of glucose	426 28–30	230, 173*	−6.56% (95% CI −12.55% to −0.57%)‡, p=0.032, I2=98%§	Yes	Low
SD of glucose	328–30	209, 153*	−0.77 mmol/L (95% CI: −1.49 to −0.06)‡, p=0.035, I2=98%§	Yes	Low
Insulin administration	427–30	200, 151*	−8.34 IU/day (95% CI, −19.03 to 2.35)‡, p=0.126, I2=75%§	No	Very low
Peak glucose level	228 29	157, 104	Inconclusive findings, with 1 RCT showing a benefit and 1 showing no difference. Meta-analysis was not performed due to insufficient number of studies.	No	Very low
Hypoglycemic events	426 28–30	230, 173*	All RCTs reporting hypoglycemic events reported infrequent events in both treatment groups, without substantial difference. Meta-analysis was not performed due to zero events in most studies and differences in reporting methods. However, the small total number of patients was likely insufficient to detect unusual events.	Unclear	Very low
Long-term care population (treatment duration approximately 12 weeks)
HbA1c	331 33 34	53, 55	−0.92% (95% CI, −1.72% to −0.13%)‡, p=0.023, I2=85%§	Yes	Low
Fasting glucose	331 33 34	53, 55	Data from 2 of 3 RCTs does not support a conclusion that DSNF confers a benefit. Meta-analysis was not performed due to data reporting issues.	Unclear	Very low
Insulin use	331 33 34	53, 55	Data from 2 of 3 RCTs do not support a conclusion that DSNF confers a benefit, while the third does. Meta-analysis was not performed due to data reporting issues.	No	Very low
High-density lipoprotein	331 33 34	53, 55	−0.11 mmol/L (95% CI, −0.22 to 0.00)‡, p=0.057, I2=46%	No	Moderate

Statistically significant results are presented in bold.

^*Mesejo 2015 had one control group (n=53) and two intervention groups. In this meta-analysis, the control group was split into two groups (n=26 and 27), one for each intervention group in the method of Cochrane, such that both intervention groups are represented and the control group does not receive disproportionate weight/overrepresentation.

^†No downgrade for risk of bias or indirectness; publication bias not quantitatively assessable given too few studies.

^‡Imprecision, serious (−1): 95% CI bounds fall below the clinically meaningful magnitude; very serious (−2) 95% CI spans appreciable benefit to harm.

^§Inconsistency, serious (−1): I²≥50% (substantial heterogeneity).

DSNF, diabetes-specific nutritional formulas; HbA1c, hemoglobin A1c; RCTs, randomized controlled trials.

Critical care

Multiple measures of glycemia were lower with DSNF compared with standard formulas in critical care populations. Mean blood glucose was reported in four RCTs (n=230 DSNF, 173 control) with a duration of follow-up spanning from 1 day to a range of 1–7 days; it was significantly reduced by a mean difference of −0.54 mmol/L (95% CI −0.78 to −0.31), p<0.001, I²=9% (figure 2a).^{26 28–30} The wide CIs with an upper limit of −0.31, an amount of change so small that it could be due to normal fluctuations, indicate that many patients did not experience a meaningful benefit. A visual inspection of figure 2a shows that while all RCTs had findings with the same directionality favoring DSNF, their effect sizes varied considerably, and not all achieved statistical significance.

Figure 2. Meta-analyses for DSNF in tube feeding for critical care patients. Outcomes to the left of zero favor DSNF. (a) Mean Blood Glucose; (b) Coefficient of Variation of Glucose; (c) Standard Deviation of Glucose; (d) Insulin Units per Day. DSNF, diabetes-specific nutritional formulas.

In the same studies, the CV of glucose had a greater reduction with DSNF by a mean difference of −6.56% (95% CI −12.55% to −0.57%), p=0.032, I²=98% (figure 2b).^{26 28–30} These data indicate a benefit for on average decreasing glucose variability with DSNF, but the clinical impact of this effect size is unclear. For people with T1D, a CV of 36% or less is a desirable target and associated with improved clinical outcomes,^{35 36} but a level of clinical importance was not identified for people with stress hyperglycemia, who compose much of this population. In addition, in all these studies, the DSNF group’s CV at follow-up was below that threshold, and it was for all but one of the control groups.²⁸ Pretreatment values were only reported for one of the RCTs,²⁹ which were already well below 36% (median 16.3% in the intervention group and 19.5% in the control group).²⁹ Furthermore, the wide CIs indicate that some patients will experience only a very small benefit.

SD of glucose was reported in three RCTs (n=209 DSNF, 153 control) (figure 2c). It was reduced significantly, favoring DSNF with a mean difference of −0.77 mmol/L (95% CI −1.49 to −0.06), p=0.035, I²=98%. SD of 1.1 mmol/L or higher has been associated with a nearly 10-fold increase in mortality.³⁷ At follow-up, while SD was lower in the DSNF group, the mean levels were all above this threshold so it is unclear whether the difference is enough to impart a clinical impact. The wide CIs indicate some patients will experience little difference.

Insulin administration per day was not reduced with DSNF overall compared with standard formula in a meta-analysis of 4 RCTs^27–30 (n=200 DSNF, 151 control) (mean difference −8.34 IU/day (95% CI −19.03 to 2.35), p=0.126, I²=75%). While the overall effect showed no benefit with DSNF, figure 2d illustrates that some studies found reductions in insulin use with DSNF while others did not. Doola et al²⁶ reported reduced insulin with DSNF but was not included in the meta-analysis due to the incompatible time frame of data collection (rate per hour).

Findings were inconsistent for peak glucose levels, and too few RCTs reported it to enable meta-analysis. Of the two RCTs (n=157 DSNF, 104 control) reporting this outcome, one found a significant benefit with DSNF (Shao et al,²⁹ reporting median 8.8 mmol/L with DSNF and 9.7 mmol/L with control, p<0.05) and one reported similar outcomes among all groups and no significant differences (Mesejo et al,²⁸ reporting a mean of 181.3 mg/dL and 191.3 mg/dL for the two DSNF, respectively, and 193.6 for the control, p=0.68).

All RCTs reporting hypoglycemic events reported infrequent events in both treatment groups, without substantial difference. Meta-analysis was not performed due to zero events in most studies and differences in reporting methods. See online supplemental material tables 3 and 4 for findings from individual studies.

Long-term care

Among people in long-term care tube-fed with DSNF, HbA1c was reduced at 12 weeks by a mean of −0.92% (95% CI −1.72% to −0.13%) in three RCTs (n=53 DSNF, 55 control), which was statistically significant (p=0.023; I² of 85%, figure 3a).^{31 33 34} A reduction of at least 0.5% is generally considered clinically significant.³⁸ As is seen in figure 3a, Vaisman et al³⁴ had a particularly large effect size. In that study, patients in the DSNF group experienced a decrease in HbA1c while patients in the control group experienced an increase. The mean effect in the other two studies was more moderate, but large enough to approximate the threshold of clinical significance.

Figure 3. Meta-analyses for DSNF in long-term tube feeding for long-term care patients Outcomes to the left of zero favor DSNF for HbA1c and outcomes to the right of zero favor DSNF for HDL. (a) HbA1c; (b) HDL. DSNF, diabetes-specific nutritional formulas; HbA1c, hemoglobin A1C; HDL, high-density lipoprotein.

Data from the same three studies did not uniformly report that DSNF conferred benefits for reducing fasting glucose, and due to data reporting limitations, they could not be combined in meta-analysis. Two studies found benefits with DSNF, but only one of those achieved statistical significance. A third found no significant difference but reported no data so whether the direction of the data was consistent in favoring DSNF could not be evaluated. Craig et al³¹ reported the DSNF group’s mean fasting glucose decreased from 7.3 mmol/L to 6.7 mmol/L, while in the control group, it increased from 6.9 mmol/L to 8.3 mmol/L, but this did not achieve statistical significance (p>0.05). Pohl et al³³ reported a median decrease of 1.29 mmol/L in the DSNF group and a decrease of 0.16 mmol/L in the control group (p=0.06). The third RCT, Vaisman et al,³⁴ reported there was no significant difference between the treatment groups but did not report data or an exact p value.

Any potential effect of DSNF on insulin use in long-term care patients was unclear because data were not robustly reported, and findings varied. Craig et al³¹ and Vaisman et al³⁴ reported no significant difference in insulin units administered but did not report data. Pohl et al³³ reported that the DSNF group had a median decrease of 4.0 total daily units while the control group had no change (p=0.014).

Meta-analysis of the three studies did not find a benefit of DSNF for high-density lipoprotein (HDL), and the studies showed trends towards greater reductions in HDL with DSNF compared with standard formulas (p=0.057, see figure 3b). None of the differences in the other lipid outcomes (triglycerides, low-density lipoprotein, total cholesterol) were statistically significant.

Sensitivity analysis

Sensitivity analyses were conducted to remove studies evaluating the Nutrison Diason formula,^{29 34} which was noted as an outlier due to the higher percentage of carbohydrates and lower percentage of fat compared with other formulas. Results from these analyses are reported in online supplemental table 6. The most notable difference in results after removing the outlier was for HbA1c in the long-term care setting.

Study risk of bias assessment

See online supplemental figure 1 for a summary of the risk of bias assessment. The evidence base was rated as low risk of bias overall. All studies were randomized using appropriate methods. Only Mesejo et al²⁷ had ’some concerns’ for the randomization process due to imbalances of baseline characteristics between groups. The long-term care studies were all double-blinded. While not all critical care studies were double-blinded, the very poor health of critically ill patients and administration of the intervention by healthcare professionals collectively provide study participants no opportunity to deviate from the protocol; the risk of non-adherence was, therefore, considered very low. It seems unlikely but not impossible that unmasked providers administered formulas not prescribed for the patient to subvert group allocation. Unmasked providers may have adjusted insulin dosing based on their awareness of group assignment, although the extent to which this occurred is likely limited given the trials had strict protocols to guide insulin administration. However, we note there was limited reporting on whether protocol deviations occurred across the included studies. While some studies did not report follow-up data for all patients, the absence of this data was not expected to depend on the true value of the outcome data and loss to follow-up for any reason was similar between groups in all studies. Outcomes were standard, objective and not anticipated to present risk of measurement bias. Finally, all RCTs had protocols and/or work plans approved by review boards or ethics committees prior to study commencement and data collection.

Strength of evidence assessment

See table 2 for a summary of each evaluated outcome and its strength of evidence rating. Strengths of the evidence include low risk of bias and a high degree of directness to populations of interest as studies enrolled those requiring tube feeding in critical care or long-term care settings. The directionality of findings of outcomes evaluated in meta-analysis consistently favored DSNF across studies and outcomes, as can be observed in the forest plots. The main limiting factor to the strength of evidence is imprecision. Imprecision is observed in the wide CIs of the summary effect sizes even among outcomes with statistically significant findings, encompassing values that are very small and unlikely to be clinically relevant and thus reducing certainty in their importance. Observed imprecision reflects differences in effect sizes among the studies and the small sample sizes, and alongside statistical heterogeneity reduces certainty in the summary effect size. Most studies did not report power analyses, and those that did calculated it based on outcomes other than those of interest in this report.^{29 30} Considered collectively, these factors led to low or very low strength of evidence ratings for most outcomes based on imprecision and heterogeneity for outcomes included in meta-analysis and inconsistent findings among studies with data issues precluding meta-analysis. Publication bias testing and testing to investigate heterogeneity were not performed because there were too few studies per evidence base.

Discussion

This systematic review with meta-analyses closes a gap in the literature and its findings supersede findings of previous systematic reviews addressing this topic, which are outdated and include lower quality studies, studies of limited clinical relevance or both.^16–19 Similarly, identified guidelines on this topic were based on out-of-date literature reviews.

The literature base, while small, consistently indicates benefits for glycemic control with DSNF. Quantification of expected benefit is clouded by statistical heterogeneity and imprecision, but summary statistics suggest most patients experience benefits large enough to improve their clinical outcomes. DSNFs were not consistently associated with improvements in other outcomes such as insulin doses and, in the long-term care population, lipids. Given the small size of the evidence base, with five studies addressing critically ill patients and three studies addressing long-term care populations, assessment of underlying causes of differences among studies could not be conducted. Additional studies are needed to provide clarity about the outcomes with qualitatively inconsistent findings and enable investigation of heterogeneity and to inform the potential impact of DSNF on outcomes for which no data were identified (ie, body composition, blood pressure, satiety, changes in diabetes medications other than insulin, nutritional status, quality of life and treatment satisfaction). The absence of these outcomes is a gap in the literature that precludes understanding of the potential effects of DSNF on cardiometabolic and nutritional outcomes.

Strengths of this review include application of internationally recognized standards and analysis methods. The included evidence base was composed of relevant, well-conducted RCTs with low risk of bias. A key limitation is the small size of the evidence base, in terms of both numbers of patients and numbers of studies. This had multiple impacts, including limited power to detect unusual events (eg, hypoglycemic events), imprecision in computed effect sizes, unsuitability for assessment of heterogeneity and publication bias. Larger enrollments would provide greater assurance of applicability of findings to overall populations. The long-term care studies only followed patients to approximately 12 weeks. Whether benefits for lipids would have been realized with additional treatment time, and to what degree glycemia benefits would be sustained, is unclear. Sparse reporting of patient characteristics such as race and ethnicity limited our ability to explore whether effects vary across different populations. Finally, our review captured RCTs published in English language only, which may have limited the number of studies identified.

From a health economics perspective, evidence regarding the impact of DSNFs on healthcare resource utilization in hospitalized patients with diabetes remains limited. Although economic outcomes were not evaluated in the studies included in this systematic review, findings from retrospective observational studies have suggested potential benefits. For example, a large retrospective analysis over a 10-year period of clinical and cost data from 125 000 hospital inpatient episodes in real-world clinical practice demonstrated that DSNF had a 0.88 day shorter length of hospital stay (LOS) on average compared with those patients provided standard formula, which had only 0.17 day shorter. The shorter LOS contributed to a cost savings of $2586 per admission, largely due to fewer hyperglycemia-related interventions and reduced insulin titration demands.³⁹ Another observational study found that DSNF lowered mean daily glucose by 15–20 mg/dL, reduced insulin requirements by 10%–20%, and cut incidence of hyperglycemic episodes by nearly 40%.⁴⁰ While these findings raise the possibility that improved glycemic management with DSNFs could translate into downstream reductions in resource utilization and healthcare expenditures, economic outcomes were not the primary focus of the studies included in this review. Therefore, additional studies specifically designed to evaluate cost-effectiveness and healthcare utilization are needed to better define the potential economic impact of DSNFs in clinical practice.

Future studies should also evaluate whether DSNFs translate into benefits in clinically important outcomes, including mortality, length of stay and complications. Although these outcomes were outside the scope of the present review, examination of the included RCTs found that approximately half reported mortality and length of stay, while complications were assessed more frequently. Across these studies, there were no significant differences between intervention and control groups, highlighting limited and inconclusive evidence currently available for these clinical endpoints.

Conclusions

Findings from this systematic review and meta-analysis showed that DSNF in tube feeding significantly improves glycemia in patients with hyperglycemia who are critically ill or in long-term care compared with standard formula. DSNFs were associated with reductions in mean blood glucose, CV glucose and SD glucose in patients with critical illness, and reductions in HbA1c in patients in long-term care. However, the strength of evidence for most of these outcomes was rated as low or very low. Findings for other evaluated outcomes were also uncertain due to inconsistent findings among studies or lack of overall statistical significance. In summary, the findings of this review signal clinically relevant improvements in glucose with the use of DSNFs compared with standard formulas in critical care or long-term care settings. Further research is needed to enhance understanding of these outcomes and to evaluate additional outcomes not addressed in this literature base.

Data availability statement

No data are available.