DASH4D diet for glycemic control and glucose variability in type 2 diabetes: a randomized crossover trial.

Results

Patient disposition

The parent trial screened 1,284 individuals and enrolled 105 participants (Figure 1, Panel A). The first participant was enrolled June 2, 2021, and the last participant was enrolled December 19, 2023. Of these 105 participants, 3 were excluded from the analysis in the parent study (2 left the study before starting the intervention diets, and 1 left after one week of controlled feeding).

Among 102 participants randomized and analyzed in the parent DASH4D trial, 89 (87%) consented to participate in the DASH4D-CGM Study and completed at least one CGM assessment (Figure 1, Panel A). The first participant for the CGM ancillary study was enrolled June 2, 2021, and the last participant was enrolled December 19, 2023.

The mean age in the DASH4D-CGM study was 67 years, 67% were female, and 88% were Non-Hispanic Black ethnicity (Table 1). At baseline, 54% of participants were using 2 or more glucose-lowering medications, mean HbA1c was 7%, mean CGM sensor glucose was 130.7 mg/dL, mean time-in-range was 84.1%, and mean coefficient of variation was 26.3%. Baseline rates of CGM-detected hypoglycemia were low (mean percentage of time below 70 mg/dL: 3.3%).

Study participants were fed four diets (DASH4D diet with higher sodium, DASH4D diet with lower sodium, comparison diet with higher sodium, and comparison diet with lower sodium) in a random order (Figure 1, Panel B). Each feeding period lasted five weeks and was separated by a break of at least one week (median length of breaks: 14 days). Participants consumed their own food and beverages during breaks. Figure 1 (Panel C) and Extended Data Table 1 provide examples of meals given to participants during the study intervention, and a full description of the nutritional composition and food content of all four diets is included in Extended Data Table 2.

Self-reported adherence to the intervention diets (ate all study food and ate no outside food) was high and did not differ across diets (~95% of study days during the DASH4D diet and ~96% of study days for the comparison diet) (Extended Data Table 3). End-of-period body weight was similar in both diets. Urinary sodium excretion was higher in the higher (versus lower) sodium diet, and urinary potassium excretion was higher in the DASH4D (versus comparison) diet.

Participants completed a total of 302 CGM assessments (146 during the DASH4D diet and 156 during the comparison diet) (Extended Data Table 4). Sensors were worn for a median of 13.6 (out of a maximum of 14) days in both diets. During CGM assessments, the median number of glucose measurements recorded was 1,339 (out of a maximum of 1,344) in the DASH4D diet and 1,338 in the comparison diet. A total of 9 sensors were dislodged and replaced (1 during the DASH4D diet and 8 during the comparison diet).

Primary outcomes

During the DASH4D diet (versus comparison diet), participants had significantly lower mean glucose (adjusted difference: −11.1 mg/dL; 95% CI: −15.8 to −6.3; P<0.001) and higher time-in-range (adjusted difference: +5.2 percentage points; 95% CI: 2.8 to 7.7; P<0.001). Glucose coefficient of variation did not differ between diet (P=0.52) (Figure 2, Table 2).

The effect of the DASH4D diet (versus comparison diet) on mean glucose and time-in-range was observed by the first day of the CGM assessment period (which occurred during weeks 3 to 4 on the diet) and was consistent to the end of the feeding period (Figure 3, Panel A-B). Differences in mean glucose and time-in-range across the diets were largest during daytime hours, particularly from ~12pm to 9pm (Figure 3, Panel C-D).

The effects of the DASH4D diet on mean glucose and time-in-range were observed across all subgroups (Figure 4, Panel A-B). There was some evidence that the benefits of the DASH4D diet were stronger for persons with higher HbA1c levels at baseline. For instance, mean glucose was 12.9 to 17.9 mg/dL lower when consuming the DASH4D diet (versus comparison) in persons with baseline HbA1c ≥7% compared to 7.6 mg/dL lower in participants with baseline HbA1c <7%.

Secondary outcomes

Participants consuming the DASH4D (versus comparison) diet had a lower glucose standard deviation (adjusted difference: −3.3 mg/dL; 95% CI: −4.9 to −1.6) and spent less time with glucose >180 mg/dL (adjusted difference: −5.9 percentage points; 95% CI: −8.3 to −3.5) and >250 mg/dL (adjusted difference: −2.7 percentage points; 95% CI: −4.5 to −0.9) (Table 2). The percentage of time spent with hypoglycemia (<70 mg/dL and <54 mg/dL) was low and did not differ between diets.

Safety

As reported in the parent trial24, there were 2 serious adverse events (1 hospitalized myocardial infraction during near the start of the DASH4D lower sodium diet and 1 hospitalized heart failure during the break between feeding periods). Three participants had elevated (≥5.5 mmol/L) potassium levels (2 during the DASH4D lower sodium diet and 1 during the comparison lower sodium diet). No participants experienced severe hypoglycemia during the study.

Within the CGM ancillary study, one participant reported pain at the CGM sensor insertion site but completed the full 14-day assessment. No other CGM-related adverse events were reported.

Exploratory outcomes

Participants spent more time with glucose between 70 to 140 mg/dL and less time with glucose above 140 mg/dL when consuming the DASH4D diet (Table 2). The proportion of participants meeting recommended targets for time-in-range and time spent with hyperglycemia was higher during the DASH4D diet. The proportion of participants meeting recommended targets for time spent with hypoglycemia did not differ between diets.

Sensitivity analyses

The effects of sodium reduction on the primary outcomes (CGM mean glucose, time-in-range, and coefficient of variation) were small and similar within the DASH4D and comparison diets (Extended Data Table 5). The DASH4D diet (versus comparison diet) reduced mean glucose and increased time-in-range at the higher and lower sodium levels. There was no significant diet-by-sodium interaction for all three primary outcomes (P>0.10 for tests for interaction).

The effect of the DASH4D diet on the primary outcomes was consistent with the intention-to-treat results when using a per-protocol approach and when restricting the analysis to those who wore CGM sensors for at least 10 days during each feeding period (Supplemental Table 1).

Post-hoc analyses

During the DASH4D (versus comparison) diet, participants had lower mean values for the Glucose Management Indicator and Glycemia Risk Index (Supplemental Table 2).

Discussion

In this controlled feeding study of adults with type 2 diabetes, participants had lower mean glucose and higher time-in-range when following the DASH4D (versus typical American) diet. The effects of the DASH4D diet were evident by the first day of CGM assessment (which occurred after 3 weeks on the diet), persisted throughout the entire two-week CGM assessment, and were driven by reductions in hyperglycemia. Benefits of the DASH4D diets were observed for all subgroups and were most pronounced for persons with HbA1c >7% at baseline. Overall, these results suggest that the DASH4D diet is a promising nutritional approach to improve glycemic control in adults with type 2 diabetes.

Our results build on the evidence from prior clinical trials. A crossover study of 21 adults with type 2 diabetes found a 1.3 percentage-point reduction in HbA1c when receiving counseling to consume the DASH diet (versus a control diet representative of a typical Iranian diet) for 8 weeks 16. A second study randomized 40 adults with type 2 diabetes to 4 weeks of dietary counseling to follow a DASH (versus control) diet, finding no difference in HbA1c 15. However, both analyses tested the effects of nutritional counseling (rather than the DASH diet itself) and had high attrition, low adherence to intervention diets, and no assessment of short-term glucose patterns. By pairing CGM assessments with a controlled feeding design, our study maximized dietary adherence and directly tested the effect of a DASH-style diet on short-term glycemic control and variability.

The effects of the DASH4D eating pattern on glycemic control may partly reflect the types of carbohydrates included in the diet. The DASH4D diet was higher in fruits, vegetables, whole grains, and nuts and lower in refined grains and sweets than the comparison diet. High-quality carbohydrates such as vegetables are nutrient-dense and rich in fiber, which can slow the digestion and absorption of glucose and improve insulin sensitivity 25. In trials of persons with type 2 diabetes, increased consumption of fiber and whole grains has been linked with reductions in HbA1c, fasting glucose, and postprandial hypergylemia 26–30.

The effects of the DASH4D diet on glycemic control are clinically relevant. An international consensus statement recently defined interventions that improve average time-in-range by at least 3 percentage points as clinically significant 31. In our analyses, the DASH4D diet exceeded this threshold; time-in-range increased by an average of 5.2 percentage points when participants consumed the DASH4D (versus comparison) diet. Emerging studies have linked higher time-in-range with a lower risk of major complications and mortality 22,23. For example, a 5-percentage point increase in estimated time-in-range lowered the risk of retinopathy and kidney disease by ~20–30% in the Diabetes Control and Complications Trial 32. These results suggest that the DASH4D diet may offer long-term benefits in adults with type 2 diabetes.

The benefits of the DASH4D diet may differ by baseline HbA1c. The diet increased time-in-range by ~7 to 12 percentage points (equivalent to ~2 to 3 additional hours each day) and reduced mean glucose by ~13 to 18 mg/dL in participants with HbA1c greater than 7%. Prior studies suggest a 10 percentage-point increase in time-in-range roughly translates to a 0.8% reduction in HbA1c 33. Thus, the DASH4D diet may have clinically meaningful benefits for individuals with HbA1c above 7%, who represent half of all adults with diabetes in the US 4. In contrast, the DASH4D diet had a smaller effect on mean glucose and time-in-range in participants with HbA1c less than 7%. Nonetheless, our results suggest that the diet can be a useful strategy for maintaining glycemic control across the spectrum of HbA1c levels.

The DASH4D diet improved time-in-range by reducing hyperglycemia. This included a ~3 percentage-point decrease (~1 hour less per day) in the time spent with clinically significant hyperglycemia (glucose >250 mg/dL). In contrast, there was no difference in CGM-detected hypoglycemia (percentage of time glucose <70 mg/dL and <54 mg/dL) between the DASH4D and comparison diet. However, persons with a history of severe hypoglycemia were excluded and baseline rates of CGM-detected hypoglycemia were low (~3% of time spent with glucose <70 mg/dL). Our study was not powered to detect differences in these rare outcomes. Further studies in populations with higher baseline risk (e.g., patients treated with intensive insulin therapy) are needed to examine the effects of a DASH-style diet on hypoglycemia.

The DASH4D diet did not influence the glucose coefficient of variation, the pre-specified primary outcome for glycemic variability. However, the coefficient of variation is a ratio calculated by dividing the standard deviation of glucose by mean glucose. In our study, the DASH4D diet significantly reduced both the standard deviation and mean glucose. As a result, the ratio of the two measures (i.e., the coefficient of variation) was unchanged. These results suggest that the DASH4D diet may reduce absolute glycemic variability (standard deviation) but may have a limited effect on relative variability.

Our study had several limitations. First, as a tightly controlled feeding experiment, our findings may not be generalizable to the general population with type 2 diabetes. Nonetheless, the DASH4D diet was designed using commonly available foods to facilitate broad adoption, and prior studies have shown the efficacy and acceptability of a DASH-style diet across diverse populations in real-world settings 34–39. Second, our study may underestimate the effect of the DASH4D diet on glycemic control and variability because the trial did not include persons with HbA1c >9%. Our analyses show that participants with the highest baseline HbA1c benefitted the most from the DASH4D diet. Third, we used an older system (Abbott Libre Pro) for our CGM assessments. However, the Libre Pro is FDA approved and remains used in clinical practice, and the accuracy of the Libre Pro is comparable to the newest generation of devices 40,41. Fourth, we could not assess the effects of the DASH4D diet on long-term glycemic control because each feeding period was 5 weeks long. Nonetheless, the length of the intervention periods in the DASH4D trial is similar to other intensive feeding trials and the follow-up period was sufficient to observe clinically meaningful changes in glucose control 7–9. Fifth, because CGM sensors were placed during the third or fourth week of each feeding period, we could not examine the early effects of the DASH4D diet on glycemic outcomes.

The DASH4D trial was a tightly controlled feeding study in persons with type 2 diabetes. Trained staff prepared over 40,000 individualized meals for participants in a metabolic research kitchen. The randomized crossover study design enabled each participant served as their own control. There was high adherence to study diets (~96% of study days). The use of CGM technology allowed us to capture detailed changes in glucose patterns. The study population included a large number of Non-Hispanic Black participants, who have traditionally been under-enrolled in diabetes trials 42.

In conclusion, the DASH4D diet improved short-term glycemic control compared to a typical US diet in adults with type 2 diabetes. Incorporating the DASH4D eating pattern into nutritional policies and clinical guidelines may be an effective strategy to promote glycemic control in adults with type 2 diabetes.

Methods

Study design and participants

The DASH4D Study (parent study) was a single center, four-period, crossover, controlled feeding trial. 24 This study aimed to assess the effects of the DASH4D diet and sodium reduction on blood pressure in persons with type 2 diabetes. From June 2021 to December 2023, the parent trial enrolled 105 adults (aged 18 years or older) from the Baltimore metropolitan area with type 2 diabetes, systolic blood pressure between 120–159 mmHg, and diastolic blood pressure <100 mmHg. Type 2 diabetes was defined as an HbA1c (point-of-care) ≥6.5% or current use of glucose-lowering medications. Further details about the parent trial have been published elsewhere. 43,44 Enrolled participants without contraindications for CGM were invited to the DASH4D-CGM ancillary study (Figure 1, Panel A).

Ethics

The CGM study protocol is provided in the Supplementary Information. All participants provided written informed consent for both the parent and CGM study. The parent and CGM study were registered on ClinicalTrials.gov with the identifying number NCT04286555. The study protocol for both the parent trial and the DASH4D-CGM study were approved by the Johns Hopkins University School of Medicine Institutional Review Board (reference number IRB00000025).

Eligibility criteria for the DASH4D parent trial and DASH4D-CGM ancillary study

Inclusion Criteria for DASH4D Parent Trial

Age 18 or older

Diabetes Mellitus Type 2 defined by: HbA1c ≥6.5%, orTreatment of diabetes with diabetes medication(s)

HbA1c ≥6.5%, or

Treatment of diabetes with diabetes medication(s)

Baseline systolic BP of 120–159 mmHg (based on average across 3 screening visits)

Baseline diastolic BP <100 mmHg (based on average across 3 screening visits)

Willing and able to eat on site for one meal per day, 3 days per week, and eat only and all food provided as part of the study diets during the controlled feeding periods (run-in and four 5-week feeding periods). Note that actual frequency of on-site dining may be fewer than 3 days per week due to COVID-related restrictions, but participants will still need to be on site to pick up food and be weighed 3 days per week, and will still be expected to have meals monitored (in-person or remotely) for one meal per day, 3 days per week.

Willing and able to complete required measurement procedures

Have access to a mobile device or computer with video conferencing capabilities, or be willing to use a device for video conferencing provided by the study

Exclusion Criteria for DASH4D Parent Trial

Laboratory Exclusions Serum potassium ≥5.2 mmol/L or <3.5 mmol/L Estimated GFR <30 mL/min by commercial lab result (note that prior to 7/12/22, the lab was using the race-based CKD Epi equation, and on/after 7/12/22, the lab switched to using the CKD-Epi 2021 equation, which does not provide different estimated GFR by race)

Serum potassium ≥5.2 mmol/L or <3.5 mmol/L Estimated GFR <30 mL/min by commercial lab result (note that prior to 7/12/22, the lab was using the race-based CKD Epi equation, and on/after 7/12/22, the lab switched to using the CKD-Epi 2021 equation, which does not provide different estimated GFR by race)

Estimated GFR <30 mL/min by commercial lab result (note that prior to 7/12/22, the lab was using the race-based CKD Epi equation, and on/after 7/12/22, the lab switched to using the CKD-Epi 2021 equation, which does not provide different estimated GFR by race)

HbA1c >9.0%

Medication Exclusions Unstable dose (i.e., change in the 2 months prior to screening or prior to randomization) of any of the following: Anti-hypertensive medicationsSodium-glucose co-transporter 2 (SGLT2) inhibitors or glucagon-like peptide-1 (GLP-1) receptor agonistsStimulants, including oral medications for asthma or chronic obstructive pulmonary disease (COPD)Hormone replacement therapy or thyroid hormoneWeight-increasing psychotropic agents Antipsychotic agentsLithiumMirtazapineUse of any of the following medications: Potassium supplementation in any form, including a multivitamin or electrolyte drink mix, with a dose >99 mg/day, which is the allowable amount in over-the-counter productsPrandial or short-acting insulinGLP-1 receptor agonist if on weight loss doseWarfarin (Coumadin)Chronic oral corticosteroid (intermittent use is okay)Weight loss medicationsTirzepatide (Mounjaro™)Unwillingness to keep same dose of vitamin, mineral, and botanical supplements

Unstable dose (i.e., change in the 2 months prior to screening or prior to randomization) of any of the following: Anti-hypertensive medicationsSodium-glucose co-transporter 2 (SGLT2) inhibitors or glucagon-like peptide-1 (GLP-1) receptor agonistsStimulants, including oral medications for asthma or chronic obstructive pulmonary disease (COPD)Hormone replacement therapy or thyroid hormoneWeight-increasing psychotropic agents Antipsychotic agentsLithiumMirtazapine

Anti-hypertensive medications

Sodium-glucose co-transporter 2 (SGLT2) inhibitors or glucagon-like peptide-1 (GLP-1) receptor agonists

Stimulants, including oral medications for asthma or chronic obstructive pulmonary disease (COPD)

Hormone replacement therapy or thyroid hormone

Weight-increasing psychotropic agents Antipsychotic agentsLithiumMirtazapine

Antipsychotic agents

Lithium

Mirtazapine

Use of any of the following medications: Potassium supplementation in any form, including a multivitamin or electrolyte drink mix, with a dose >99 mg/day, which is the allowable amount in over-the-counter productsPrandial or short-acting insulinGLP-1 receptor agonist if on weight loss doseWarfarin (Coumadin)Chronic oral corticosteroid (intermittent use is okay)Weight loss medicationsTirzepatide (Mounjaro™)

Potassium supplementation in any form, including a multivitamin or electrolyte drink mix, with a dose >99 mg/day, which is the allowable amount in over-the-counter products

Prandial or short-acting insulin

GLP-1 receptor agonist if on weight loss dose

Warfarin (Coumadin)

Chronic oral corticosteroid (intermittent use is okay)

Weight loss medications

Tirzepatide (Mounjaro™)

Unwillingness to keep same dose of vitamin, mineral, and botanical supplements

Any medication not compatible with participation as determined by the investigators

Medical History Exclusions Type 1 diabetesHypoglycemia requiring hospitalization or the assistance of another person in the last 12 monthsActive CVD or any event in the prior 6 months, including coronary artery bypass grafting (CABG), percutaneous transluminal coronary angioplasty (PTCA), myocardial infarction (MI), cerebrovascular accident (CVA), or congestive heart failure (CHF) exacerbation requiring hospital admissionCancer diagnosis or treatment in the last 2 years (benign tumors or non-melanoma skin cancer or localized breast or prostate cancer not requiring chemotherapy is acceptable)Active inflammatory bowel disease, bowel resection, malabsorptive syndrome, pancreatitis (episode within past year), history of Roux-en-Y gastric bypass, or history of other bariatric surgery that limits food intake volume or that requires a specific diet planPregnancy or lactation or planned pregnancyAny emergency department (ED) visit for asthma or COPD in the last 6 months

Type 1 diabetes

Hypoglycemia requiring hospitalization or the assistance of another person in the last 12 months

Active CVD or any event in the prior 6 months, including coronary artery bypass grafting (CABG), percutaneous transluminal coronary angioplasty (PTCA), myocardial infarction (MI), cerebrovascular accident (CVA), or congestive heart failure (CHF) exacerbation requiring hospital admission

Cancer diagnosis or treatment in the last 2 years (benign tumors or non-melanoma skin cancer or localized breast or prostate cancer not requiring chemotherapy is acceptable)

Active inflammatory bowel disease, bowel resection, malabsorptive syndrome, pancreatitis (episode within past year), history of Roux-en-Y gastric bypass, or history of other bariatric surgery that limits food intake volume or that requires a specific diet plan

Pregnancy or lactation or planned pregnancy

Any emergency department (ED) visit for asthma or COPD in the last 6 months

Any other serious illness or condition not compatible with participation as determined by the investigators

Physical Exclusions Body weight >420 poundsArm circumference >50cmWeight loss or gain of >5.0% of body weight during 2 months prior to screening, or large weight change during screening prior to randomization

Body weight >420 pounds

Arm circumference >50cm

Weight loss or gain of >5.0% of body weight during 2 months prior to screening, or large weight change during screening prior to randomization

Lifestyle and Other Exclusions Significant food allergies, preferences, intolerances, or dietary requirements that would interfere with diet adherenceNot able to self-monitor glucose if neededConsumption of more than 14 alcoholic drinks per week or consumption of more than 6 drinks on one or more occasion per weekActive substance use disorder that would interfere with participationParticipation in or planning to start weight loss programCurrent participation in another clinical trial that might affect blood pressure or ability to comply with study proceduresPlanning to leave area prior to end of studyInvestigator discretion

Significant food allergies, preferences, intolerances, or dietary requirements that would interfere with diet adherence

Not able to self-monitor glucose if needed

Consumption of more than 14 alcoholic drinks per week or consumption of more than 6 drinks on one or more occasion per week

Active substance use disorder that would interfere with participation

Participation in or planning to start weight loss program

Current participation in another clinical trial that might affect blood pressure or ability to comply with study procedures

Planning to leave area prior to end of study

Investigator discretion

Continuous Glucose Monitoring (CGM) Ancillary Study Exclusion

History of allergic skin reaction to adhesive

Implantable pacemaker

Dietary intervention

Study participants were fed four diets (DASH4D diet with higher sodium, DASH4D diet with lower sodium, comparison diet with higher sodium, and comparison diet with lower sodium) in a random order. Each feeding period lasted five weeks and was separated by a break of at least one week (median length of breaks: 14 days). Participants consumed their own food and beverages during breaks (Figure 1, Panel B).

Similar to the original DASH diet, 7 the DASH4D diet is rich in fruits, vegetables, whole grains, and low-fat dairy. The DASH4D diet emphasizes lean protein sources (e.g., lean meats and beans) and is low in saturated fat and sugar-sweetened foods and beverages. The carbohydrates target for the DASH4D diet (45% of total calories) was lower than the original DASH diet (55% of total calories) to facilitate diabetes management. The DASH4D diet also had higher targets for potassium and unsaturated fat than the original DASH diet to better align with dietary recommendations for persons with type 2 diabetes. 7–9,45,46

The comparison diet was developed following a similar process used in the original DASH trial. 7 This diet was based on national survey data from the National Health and Nutrition Examination Survey (NHANES) and reflects consumption patterns of the general population of US adults, 47,48 as well as individuals with diabetes. 49 The comparison diet was low in fruits, vegetables, whole grains and plant-based sources of protein (e.g., nuts). Compared to the DASH4D diet, the comparison diet included more red meat, solid fats (e.g., butter), and sugar-sweetened foods. The macronutrient targets in the comparison diet matched the average distribution of calories from proteins, fats, and carbohydrates among US adults with diabetes. For micronutrients, we selected targets near the 25th percentile among US adults to create a contrast with the DASH4D diet.

The lower sodium intake target (1,500 mg/day at 2,000 kcal) was selected based on its efficacy in a prior sodium reduction trial. 8 This lower target also aligns with dietary recommendation from the American Heart Association. 50 The higher sodium (3,700 mg/day at 2,000 kcal) was based on average consumption in the NHANES and aligns with the 75th percentile among US adults. 49

All meals, snacks and most beverages were prepared in a metabolic kitchen and provided to participants during the feeding periods. Participants consumed three meals per week at the study center, and all other food was eaten offsite. Participant caloric targets were set to minimize weight change, and caloric levels were adjusted as needed to maintain a stable weight.

The study team developed a 7-day menu cycle consisting of 21 meals (7 breakfasts, 7 lunches, 7 dinners) and 7 sets of snacks for all four study diets. Diets were prepared at five calorie levels (1500, 2000, 2500, 3000, 3500 kcal). A full description of the nutritional composition and food content of all four diets is included in Extended Data Table 2.

Adherence to study diets was self-reported daily during each feeding period. Adherence was defined as self-reporting eating all study foods and abstaining from eating outside food or unallowed beverages each day. Dietary adherence was also objectively assessed by 24-hour urinary excretion of sodium, potassium, and creatinine. These objective measures were collected during week 4 or 5 of each feeding period.

Randomization

Using a computer algorithm, participants were randomized with equal probability to one of 24 four-diets sequences, with a block size of one. Investigators were masked diet assignment and all study team members were blinded to CGM measurements.

Continuous glucose monitoring

The Abbott FreeStyle Libre Pro was used for CGM assessments. The Libre Pro is a masked CGM system (participants cannot see glucose values), comes factory calibrated (does not require calibration with fingerstick testing), and measures glucose every 15 minutes for up to 14 days. 51 Trained research staff placed the Libre Pro on the back of the upper arm during a screening visit (baseline) and the third or fourth week of each feeding period. CGM sensors that were dislodged or accidently removed before the completion of the 14-day wear period were replaced.

Outcomes

All CGM outcomes were calculated over the entire 14-day CGM assessment period and reflect glycemic control and variability at the end of each feeding period. Outcomes were selected based on clinical guidelines and recommendations from a consensus statement on CGM endpoints for clinical trials. 31,52

The primary outcomes were CGM mean glucose, percentage of time spent with glucose between 70 to 180 mg/dL (time-in-range), and glucose coefficient of variation (glucose standard deviation divided by mean glucose). Secondary outcomes were glucose standard deviation, the percentage of time with hyperglycemia (glucose above 180 mg/dL and 250 mg/dL, respectively); and the percentage of time with hypoglycemia (glucose below 70 mg/dL and 54 mg/dL, respectively). 31,53

Exploratory continuous outcomes were percentage of time between 70 and 140 mg/dL and percentage of time above 140 mg/dL. Exploratory binary outcomes included meeting targets for time-in-range (>70% time), coefficient of variation (<36%), time above 180 mg/dL (<25% time), time above 250 mg/dL (<5% time), time below 70 mg/dL (<4% time), time below 54 mg/dL (<1% time).

Post-hoc outcomes were the Glucose Management Indicator and Glycemia Risk Index. Both post-hoc outcomes were calculated used previously published equations. 54,55

Other measures

Sociodemographic characteristics (age, sex, race, and ethnicity) were self-reported during in-person interviews. HbA1c was measured at a screening visit using the Afinion AS100 analyzer (Abbott Laboratories), a point-of-care device. Medication use was assessed through pill bottle review and categorized into classes by research physicians.

Safety monitoring and adverse events

The parent trial ascertained serious adverse events and four prespecified adverse events: (1) systolic blood pressure >180 mmHg or diastolic blood pressure >110 mmHg; (2) systolic blood pressure 171–180 mmHg or diastolic blood pressure 106–110 mmHg; (3) serum potassium ≥5.5 mmol/L; (4) blood glucose <70 mg/dL.

Potassium was measured at the end of each study period for all participants. Individuals with chronic kidney disease and those using potassium-sparing diuretics also had potassium assessed during the second week of each feeding period. Participants with hyperkalemia (confirmed potassium ≥5.5 mmol/L) were withdrawn from the study.

Participants using agents associated with a high risk of hypoglycemia (insulin, sulfonylureas, or meglitinides) recorded self-monitored blood glucose. A study clinician reviewed these data and followed a standardized medication adjustment protocol to prevent hypoglycemia as needed.

Participant also reported adverse events directly to study staff or in study diaries. Diaries included the following question related to adverse events: “Is there anything you would like us to know regarding your participation in this study, including any new medications/pills, vitamins/supplements, or any health problem or issue?” The diary also included an open-ended section for participants to provide details about any adverse events. Participants completed study dairies each day of controlled feeding.

Sample size and power calculation

The sample size for the DASH4D-CGM study was fixed per the design of the parent trial. Because sodium is unlikely to affect short-term glucose control, 56 we pre-specified analyses combining the lower and higher sodium feeding periods for each diet to increase statistical power (see Supplemental Information for statistical analysis plan). Using this two-arm study design, a sample size of 100 participants could detect a difference of 8.8 mg/dL in mean glucose, 4.2 percentage-point difference in time-in-range, and 1 percentage-point difference in coefficient of variation with 90% power (see Supplemental Table 3 for details about power calculation). 44

Statistical analyses

We conducted all analyses on an intention-to-treat basis. We estimated the effect of the DASH4D (versus comparison) diet on all outcomes using linear mixed-effects models (for continuous outcomes) or logistic mixed-effects models (for binary outcomes). In secondary analyses, we assessed the effect of the DASH4D diet on the primary outcomes (mean glucose, time-in-range, coefficient of variation) across pre-specified subgroups (age, sex, categories of baseline HbA1c, number of glucose-lowering medications, type of glucose-lowering medications). All models were adjusted for study feeding period, included a random intercept for participants to account for within-person correlation, and used an unstructured variance-covariance matrix. Mixed-effects models account for missing data using maximum likelihood estimation and assume that data were missing at random.

We conducted three sensitivity analyses. First, we tested our pre-specified assumption that sodium would not affect glucose measures. For this analysis, we assessed the effect of higher (versus lower) sodium on the primary outcomes within each study diet. We also examined the effect of the DASH4D (versus comparison) diet on the primary outcomes within lower and higher sodium levels. We tested for interactions between sodium and diet. Second, we re-estimated all primary analyses using a per-protocol approach. For this analysis, we restricted our sample to participants who were adherent to intervention diets >90% of study days for a given feeding period. Third, for each feeding period, we only included data from participants who wore CGM sensors ≥70% of the time (10 days). 31

In exploratory analyses, we visually compared daily and hourly mean glucose, time-in-range, and coefficient of variation across the DASH4D (versus comparison) diet. Daily averages were calculated using each 24-hour of CGM data. Consistent with prior CGM trials, 57–59 these analyses were based on raw CGM data and not model-based.

All analyses were conducted using Stata version 18. All P-values were two sided. Because CGM outcomes are not independent, we did not adjust for multiple comparisons.

Extended Data

Example 2000 kcal Daily Menus of DASH4D and Comparison Diet

Note: Household measurements were estimated from the weight of each food item.

Varying amounts of salt added to meet the lower and higher sodium target

Food content and daily nutrient targets for the four study diets (at 2,000 kcal level)

Note: Seven-day average nutrient and food group calculations of the four study diets are shown at the 2000 kilocalorie (kcal) level. Prespecified tolerances for each seven-day menu cycle were ±2% kcal for total fat, saturated fat, carbohydrates, and protein; ±5% mg for sodium, potassium, and calcium; and ±10% mg for magnesium and fiber. Diets were prepared at five kcal levels: 1500, 2000, 2500, 3000 and 3500 kcal. Sodium, potassium, calcium and magnesium were adjusted in increments per kcal, maintaining an approximately 50% difference between the higher vs. lower sodium diets for sodium, or between the DASH4D vs. comparison diet for other nutrients. ‘n/a’ indicates that there was no target for this nutrient or food group

Self-reported and objectively measured diet adherence and measured body weight, by study diet

Adherence for each participant was calculated as the number of study days in which they consumed all study food and ate no outside food. Adherence was calculated from day 1 of each feeding period to the day the blood pressure outcome was assessed. 24-hour urine was collected during week 4 or 5 of each feeding period to assess urinary metabolite excretion. Baseline weight was the average of all weight measurements collected during the run-in period. End-of-period weight was the average of all weight measurements collected during feeding period week 4 or 5.

Continuous glucose monitoring (CGM) adherence (N=89)

Adjusted differences in primary outcomes within different sodium levels and diets

Abbreviations: CI = confidence interval

Note: Time-in-range was defined as the percentage of time spent with CGM glucose between 70 and 180 mg/dL. Estimates in each row were based on separate linear mixed-effects models that included a random intercept for participants and adjusted for study feeding period. The 95% confidence intervals for all estimates are indicated in paratheses.

Supplementary Material