The study was an open-label, non-randomized study on participants with type 2 diabetes (T2D) receiving two different treatments, continuous care intervention (CCI) and usual care (UC) (17). The CCI is the treatment arm that consisted of dietary intervention and remote continuous care using telemedicine, while the UC arm received standard of care. The CCI cohort was advised to follow a very low carbohydrate intervention to achieve and sustain nutritional ketosis, i.e., <30 g/day of total carbohydrate in the beginning of the intervention and personalized subsequently based on the individuals’ health goals and carbohydrate tolerance. Protein intake was targeted at a level of 1.5 g/kg of reference body weight, and fat intake targeted to achieve satiety. These participants were monitored telemedically using a web-based software application (app) by their remote care team consisting of a health coach and a healthcare provider. The app was used to upload weight, fingerstick blood glucose and BHB to monitor the patients’ adherence in the diet, their clinical progress, side effects and management of any medication adjustments. BHB values were measured using the Precision Xtra meter (Abbott, Alameda, CA), which has been reported to have a coefficient of variation (CV) of less than 10% across BHB concentrations. BHB values ranging from 0.5 to 5.0 mM are considered within the euketonemia range; however, any value greater than 0.3 mM is still a meaningful indicator that an individual is restricting carbohydrates to a degree that physiologically shifts the body into ketosis (18). The co-primary endpoints of the study were weight and T2D status. Patients were enrolled in the study from August 2015 to May 2016 for a two-year longitudinal assessment. The results from the initial core 2-year study along with the study design, inclusion and exclusion criteria were previously published (17, 19). This post-hoc analysis assesses the intervention effect on 2-year eGFR slope, eGFR stages, and inflammation markers.

There were 262 and 87 participants in the CCI and UC arms, respectively. Kidney-relevant outcomes including serum creatinine, complete blood count (CBC), and hsCRP were collected at baseline, 1 and 2 years. Demographics (age, gender, race, diabetes duration) and other ancillary variables including body mass index (BMI), weight, hemoglobin A1c (HbA1c), systolic blood pressure (SBP), fasting insulin, white blood cell (WBC) and blood urea nitrogen (BUN) that were previously reported in the full CCI and UC cohorts were also included in this post-hoc analysis. eGFR was calculated using the 2021 race-free Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine-based equation (20). Neutrophil-lymphocyte ratio (NLR) was calculated by dividing absolute neutrophil count by the absolute lymphocyte count.

The primary aim of this post-hoc analysis was to assess the relationship between blood BHB and eGFR slope. The secondary aim was to explore changes in inflammatory markers. Other ancillary aims were to assess changes in eGFR and eGFR stage transition from baseline to 2 years among those with baseline eGFR < 60 mL/min/1.73m².

Comprehensive details on the statistical analysis, including the latent class trajectory modeling, linear mixed effect model (LMM), and multiple linear regression analyses, are included in the supplementary statistical method section. The guidelines for reporting on latent trajectory studies (GRoLTS) checklist, as recommended by the Equator Network, are included in the supplementary materials (21).

CCI was associated with a significant increase in eGFR from baseline to 2 years with an eGFR slope 0.91 mL/min/1.73m²/year (p = 0.03). In contrast, UC was associated with a non-significant decrease in eGFR (−0.68 mL/min/1.73m²/year). The between group difference was 1.59 (−3.23, 0.05) mL/min/1.73m²/year and of borderline statistical significance (p = 0.06). Similarly, in a subcohort of participants with baseline eGFR<90 mL/min/1.73m², the difference in the annual rate of change between CCI vs. UC in eGFR was 1.98 (−4.74, 0.79) mL/min/1.73m²/year, where CCI had a positive eGFR slope of 3.04 mL/min/1.73m²/year (p < 0.001).

The results of the multiple regression analyses are presented in Table 2, focusing on predictors previously reported in the literature. Other potential predictors, such as change in lower extremity lean mass, HOMA-IR, and fasting glucose at 365 days, were not significantly associated with change in eGFR at 365 days. After adjusting for baseline predictors (which remained consistent across all models), only mean change in weight (Model 1) and mean BHB (Model 5) at 365 days were significantly associated with mean change in eGFR at 365 days in the models where change-related predictors were included individually. In Model 6, which included both mean delta weight and mean BHB over 1 year, only mean BHB over 1 year (β = 0.1, p = 0.009) was associated with mean change in eGFR at 365 days, with greater mean BHB positively associated with greater mean change in eGFR at 365 days. Additionally, higher baseline eGFR (β = 0.8, p < 0.001) and shorter diabetes duration (β = −0.1, p = 0.01) were associated with a greater mean change in eGFR at 365 days.

As previously published, antiglycemic medications (e.g., SGLT2i, GLP1-RA) (19) were largely deprescribed or remained unchanged at 1 year, with no significant changes in the use of angiotensin-converting enzyme inhibitors (ACEi) and angiotensin II receptor blockers (ARBs) (22), and no participants were on mineralocorticoid receptor antagonists. In the cohort used for this regression analysis, 10.9% of participants were on SGLT2i and 13.7% were on GLP-1 RAs at baseline. GLP-1 RA use remained unchanged at 1 year, with 13.3% of participants continuing therapy. In contrast, SGLT2 inhibitors were deprescribed in most participants as part of the individualized medication management protocol in the CCI group, with only 0.9% remaining on SGLT2i at 1 year. In the final multivariate sensitivity model, mean BHB at 365 days (β = 0.1, p = 0.002) remained significantly associated with change in eGFR at 365 days, even after adjusting for baseline medication use of SGLT2i, GLP1-RA, and ARBs/ACEi (Table 2). Both baseline use of GLP1-RA and ARBs/ACEi were associated with mean change in eGFR at 365 days, with ARBs/ACEi use positively associated and GLP1-RA use negatively associated with a greater mean change in eGFR at 365 days.

To further investigate the possible association between BHB levels and reno-protection, we subclassified the CCI group based on achieved BHB levels. Details on model specification, parameter and selection are included in the Supplementary Appendix A1 Statistical Results and Supplementary Table A1. According to the model specification, the model with 4 latent ketosis classes was deemed to be the best fit (i.e., lower BIC, AIC, and SABIC values) when compared to models with 3 or lower and 5 or more latent classes (Supplementary Tables A2, A3). The entropy of the model was 0.74 indicating good separation between the classes. Supplementary Table A4 includes a completed GRoLTS (Guidelines for Reporting on Latent Trajectory Studies) checklist for the latent class trajectory analysis. The ketosis classes were identified as unsustained nutritional ketosis (UNK, n = 27), low nutritional ketosis (LNK, n = 105), moderate nutritional ketosis (MNK, n = 99) and sustained nutritional ketosis (SNK, n = 17) as illustrated in Figures 1A,B. Table 1 provides details of the mean BHB and percentage of days logging BHB ≥ 0.3 mM in each of the ketosis trajectory classes.

Table 3 presents the estimated eGFR slopes in the different ketosis trajectory classes in comparison to UC. Individuals in the two higher ketosis adherence categories, namely SNK and MNK, had a greater increase in their eGFR slope from baseline to two years relative to those categorized as lower ketosis adherence (LNK and UNK) (Figure 2). Specifically, the eGFR slope within the SNK group was significantly higher compared to the UC group, with an observed between-group difference of 4.06 mL/min/1.73m² annually. Furthermore, analysis revealed a dose–response relationship between the intensity and duration of ketosis, as classified into distinct ketosis classes, and the improvement of eGFR slope (Figure 2A and Table 3). This dose-dependent effect of ketosis adherence on eGFR slope improvement was more pronounced among participants who started off with a baseline eGFR < 90 (Figure 2B and Table 3).

Among participants with baseline eGFR indicating advanced kidney disease (stage 3a or higher), 23 individuals in the CCI group showed significant improvement, with mean eGFR increasing from 49.9 to 64.7 mL/min/1.73m² at 2 years (p < 0.001). Of those with 2-year eGFR data (n = 15), 93% were in stage 3 and 7% in stage 4 at baseline. Notably, none progressed to more severe CKD, and 53% improved to stage 2, 7% improved from stage 4 to stage 3, while 40% remained in stage 3.

In contrast, among 8 individuals with mild to moderate kidney disease in the UC group, 6 had 2-year eGFR data. Only 17% improved to stage 2, 67% remained in stage 3, and 17% worsened to stage 4.

At 2 years, the CCI cohort experienced significant reductions in hsCRP, WBC, and NLR levels, while no significant changes were observed in the UC group (Table 4). Reductions in inflammation markers were observed across all four CCI ketosis trajectory classes, with a clear dose–response relationship. Participants with higher ketosis adherence (SNK and MNK) showed the greatest improvements in hsCRP (−57.5% and −32.0%, respectively) and WBC (−20.6% and −10.7%, respectively) compared to those with lower adherence (LNK: hsCRP -40.7%, WBC -2.9%; UNK: hsCRP -28.0%, WBC -3.7%) (Table 4). Similarly, NLR decreased significantly in the CCI cohort (−8.5%) and in a subcohort with baseline NLR ≥ 2.5 (−15.2%), with no significant changes in UC. Improvements in NLR also followed a dose–response pattern, with greater reductions in higher ketosis adherence categories. These results suggest a robust, dose-dependent anti-inflammatory effect of nutritional ketosis. BUN levels remained stable and within the normal range across all CCI classes and UC (Table 4).

In this study, we demonstrated for the first time that a very low carbohydrate intervention leading to nutritional ketosis is associated with an increase in the rate of eGFR change of 0.91 mL/min/1.73m²/year from baseline to two years in participants with type 2 diabetes. This exceeded the recognized threshold of a clinically meaningful eGFR slope increase of 0.75 mL/min/1.73m² (3, 4). Of note, the increase in the eGFR slope was more evident at higher levels of ketosis and at progressively lower baseline eGFR levels (baseline eGFR <90 mL/min/1.73m²). These findings raise the possibility that a very low carbohydrate intervention could be used to stall or even reverse the progression of diabetic nephropathy to end-stage kidney disease (3, 4).

No other nutritional intervention has demonstrated a similarly positive impact on the eGFR slope. The CORDIOPREV study reported an annual eGFR decline of approximately −1 mL/min/1.73m² with a Mediterranean diet over five years (23), while the MDRD study observed declines of −2.1 and −3.2 mL/min/1.73m² per year in patients with moderate to severe kidney dysfunction following low-protein and very low-protein diets, respectively (24). Similarly, a post-hoc analysis of the Look AHEAD trial found declines of −0.86 and −0.93 mL/min/1.73m² per year over 10 years in the intensive lifestyle intervention (ILI) and diabetes support and education (DSE) groups (25). These studies, however, benefited from longer follow-up periods compared to ours and differed in terms of the study population. Pharmaceutical trials with SGLT2i (8–10), GLP1-RA (26, 27), and finerenone (28) in participants with pre-existing CKD at high risk for progression also reported negative eGFR slopes in both treatment and placebo groups. However, the difference in eGFR decline between the groups ranged from 0.8 to 2.0 mL/min/1.73m²/year, with the placebo arms experiencing greater declines, and the drugs effectively mitigating the rate of decline. In contrast, our nutritional intervention, which targeted nutritional ketosis, uniquely demonstrated a positive eGFR slope of 0.91 mL/min/1.73m²/year in patients with relatively intact kidney function. This improvement was particularly notable in participants with a baseline eGFR <90 mL/min/1.73m², where the difference between the intervention and usual care groups was 1.97 mL/min/1.73m²/year. Moreover, the absence of significant changes in blood urea nitrogen suggests that the positive eGFR slope was not attributable to hyperfiltration from increased protein intake, highlighting the unique renoprotective potential of this approach.

Healthy individuals in the general population usually have an eGFR slope of approximately −1 mL/min/1.73m²/year (29), while patients with T2DM generally have a more rapid decline (30). Our intervention appears to have staved off and even reversed such a decline in the eGFR slope especially among those with lower eGFRs. This trend was consistent among those who started with an eGFR <60 mL/min/1.73m² at baseline, where these patients not only experienced an increase in their eGFR but 53% of them reversed their eGFR staging from stage 3 to stage 2. Nutritional interventions that have the potential to slow the eGFR decline or maintain the eGFR decline to a rate similar to the general population are associated with improved kidney outcomes. Two different post-hoc analyses of the Look AHEAD study data revealed significant benefits for kidney health (31, 32). An earlier study demonstrated that participants in the ILI group had a reduced risk (HR 0.69, 95% CI; 0.55 to 0.87; difference of 0.27 cases per 100 person-years) of developing high-risk CKD, as defined by the 2013 KDIGO classification, compared to the DSE arm (31). A more recent study reported a significant reduction in eGFR < 45 mL/min/1.73m² in the ILI arm compared to the DSE arm during the active intervention phase, with a hazard ratio of 0.80 (95% CI; 0.66 to 0.98) (32).

None of these previous nutritional or pharmacological intervention studies achieved reversals of the eGFR slopes to positive values. Even the best therapies only slow the rate of eGFR decline. The main difference between our nutritional intervention and previous nutritional interventions is that our very low-carbohydrate approach leads to sustained nutritional ketosis in many of the participants. This suggests that the metabolic state of ketosis, characterized by physiologic levels of ketones, may be the key to the observed reno-protective effects. Ketones, particularly BHB, are believed to exert reno-protective benefits through multiple mechanisms, including serving as an efficient alternative energy source for the kidney, reducing oxidative stress, and exhibiting anti-inflammatory and antifibrotic effects, which collectively help preserve kidney function and mitigate damage (33). However, it’s important to consider whether similar benefits would be seen with exogenous BHB supplementation or if sustained endogenous ketosis reflects broader metabolic improvements, particularly reduced hyperinsulinemia, which is known to drive inflammatory pathways (e.g., NLRP3 via PI3K) and other renal stressors (34, 35). Endogenous ketosis likely signals reduced insulin exposure over time, which in turn impacts markers like vitamin D metabolism (36), and inflammatory cytokines (35). Additionally, carbohydrate restriction—which induces endogenous ketosis—may directly ameliorate hyperglycemia and reduce the formation of advanced glycated end products implicated in kidney injury (37). Consequently, the combined effects of carbohydrate restriction, elevated endogenous BHB levels, and lower insulin may contribute to improved eGFR, although further research is required to delineate the independent role of BHB.

The improvement in eGFR in the CCI group was most notable between baseline and one year, coinciding with the period of highest adherence to the carbohydrate restriction. Similar to the observed eGFR improvement trend, a greater increase in mean BHB levels from baseline to 365 days was significantly associated with eGFR improvement, along with shorter diabetes duration, higher baseline eGFR and baseline use of GLP1-RA and ACEi/ARBs after accounting for body weight changes at one year. While GLP-1 RA use at baseline was independently associated with changes in eGFR at 1 year, no such association was observed for SGLT2i, likely due to the limited sample size, substantial deprescription of SGLT2i during the study, and the resulting variability in exposure. It is also important to note that participants on GLP-1 RAs were primarily using older-generation agents, and thus, this observed association should be interpreted cautiously. Future studies with larger cohorts, consistent medication exposure, and newer GLP-1 RA formulations are needed to validate these findings and explore potential additive effects with nutritional interventions. Likewise, our findings were consistent with two other studies that found kidney function improvement from a low carbohydrate nutritional intervention was not correlated with the degree of weight loss (38, 39). Additionally, one of these studies reported an association between eGFR improvement and a decrease in fasting insulin and systolic blood pressure (39), but a similar association was not observed in our study. However, neither of these studies assessed whether changes in mean ketone levels were associated with eGFR improvement (38, 39) or accounted for mean ketones in the model that included fasting insulin (39). It is possible that the association between the decrease in fasting insulin and eGFR improvement was due to carbohydrate restriction, which helps increase insulin sensitivity and reduce the need for greater insulin secretion (40). Moreover, another recent study reported that ketonuria-an indicator of high level of blood ketone levels - was the only significant factor associated with six-month eGFR improvement in those treated with SGLT2i in both univariate and multivariate linear regression (41).

Like the association analysis, we also found a clear dose–response relationship between ketosis classes and eGFR slope. Individuals with closer dietary adherence and maintenance of ketosis categories had a greater increase in their eGFR slope from baseline to two years compared to those demonstrating lower ketosis adherence groups. This relationship was more evident in those with mild baseline kidney dysfunction. These findings align with the reno-protective effect of SGLT2i, which is associated with the presence of ketonuria (41), and mechanistically depends on the production of BHB (13). The improvement in eGFR was approximately seven-fold greater in those on SGLT2i with ketonuria versus those on SGLT2i without ketonuria (41). Likewise, in genetically driven autosomal polycystic kidney disease (ADPKD) a ketogenic diet resulted in increased eGFR in a randomized-controlled trial (42). A post-hoc analysis of the FDA-accepted outcome measure of height-adjusted total kidney volume (htTKV) showed that patients in the ketogenic diet arm who reached a greater ketosis threshold experienced a greater reduction in htTKV (43). The apparent dose-dependent beneficial relationship between BHB levels and reno-protection suggests that nutritional interventions leading to a deeper level of ketosis may be more effective in DN. Alternatively, or in addition, supplementation with exogenous BHB may be of added benefit which is currently being explored for ADPKD (44).

While the mechanistic and biochemical actions of ketones on kidney function could be attributed to their putative anti-inflammatory, anti-fibrotic, and anti-oxidative effects, it is still unknown whether these effects lead to a reduction in adverse kidney outcomes or events. Several clinical and real-world observational studies have reported improvements in kidney function markers among those following a low or very low carbohydrate diet (38, 39, 45–47), including studies that showed improvements in urine albumin creatinine ratio (UACR) (38) and cystatin C levels (47). Even lowering carbohydrate intake to non-ketogenic levels appears beneficial. Thus far, there is only one study that has reported a significant reduction in the risk of doubling of serum creatinine and development of dialysis dependency, or all-cause mortality in a carbohydrate restricted cohort compared to a standard protein restriction diet (48). One large observational study that assessed the association between carbohydrate intake and mortality in those with CKD reported a reduction in mortality in groups classified as having lower carbohydrate intake (HR of 0.76, 95% CI; 0.62 to 0.93) (49).

The close interrelationship between inflammation and CKD is also well established, with numerous studies highlighting the role of various inflammatory markers and the NLRP3 inflammasome in the progression of kidney damage (50, 51). One such marker, the neutrophil-lymphocyte ratio (NLR), has garnered attention for its predictive value in DN (52, 53). Elevated NLR levels indicate an inflammatory state and are associated with worse renal outcomes in patients with T2D, and with increased UACR (54, 55) and lower eGFR (56) in patients with DN. In our study, we observed a significant reduction in NLR, along with hsCRP and WBC, from baseline to two years that followed a dose–response trend related to adherence and maintenance of ketosis level. Despite NLR being a predictive marker for DN and its progression, there are still no studies directly evaluating the impact of NLR reduction on kidney function improvement or the reduction of adverse kidney outcomes. However, several studies indicate that NLR above certain cutoffs, ranging from 2.2 to 3.0, are most likely associated with diabetic nephropathy (56, 57–59). In the subgroup analysis of those starting with NLR ≥ 2.5, there was a 15% average reduction in NLR, with the average falling below the cutoff after two years. Furthermore, both high ketosis adherence classes experienced the greatest decline in NLR, normalizing the NLR levels to <2.5 at two years.

Our study has several strengths, including the inclusion of two distinct groups that were closely followed for at least two years, allowing us to assess differences in eGFR slope between them. Additionally, ketone levels were routinely monitored in the CCI group using blood ketone meters, with values uploaded through the app. This enabled us to examine the association between ketone exposure and improvements in eGFR. However, the study also has several limitations. It represents a post hoc secondary analysis of kidney function markers and was not specifically powered to assess kidney outcomes; thus, the findings should be considered hypothesis-generating. Another key limitation is the lack of detailed information on the underlying causes of CKD and the use of immunosuppressive agents, as the trial was not originally designed to evaluate kidney-specific endpoints. Furthermore, the interpretation of our findings is limited to eGFR improvement; the impact of the intervention on urinary albumin-to-creatinine ratio (UACR) remains unknown. Although urine microalbumin was measured, no significant changes were observed from baseline to two years. The findings from this analysis are limited to patients with type 2 diabetes (T2D). Furthermore, eGFR was estimated using serum creatinine, which can be influenced by weight loss, particularly by changes in lean mass. Rather than using cystatin C or direct measurement via 24-h urinary creatinine clearance. Finally, the moderate sample size may limit the generalizability of the findings.