A randomized phase II trial of gemcitabine, nab‐paclitaxel, cisplatin with or without a medically supervised ketogenic diet for patients with metastatic pancreatic cancer

ABSTRACT

Background

A randomized phase II screening trial of gemcitabine, nab‐paclitaxel, and cisplatin with a medically supervised ketogenic diet (MSKD) versus usual diet (non‐MSKD) was conducted in patients with treatment‐naive metastatic pancreatic ductal adenocarcinoma (PDAC).

Methods

Patients with untreated metastatic PDAC were randomized 1:1 to MSKD or non‐MSKD while receiving gemcitabine, nab‐paclitaxel, and cisplatin on days 1 and 8 of a 21‐day cycle. The MSKD was guided by a remote health care team and daily ketone (beta‐hydroxybutyrate) levels, with goal beta‐hydroxybutyrate of 0.5 to 3.0 mM. The primary endpoint was progression‐free survival (PFS) using a one‐sided alpha level of 0.20. Secondary endpoints included overall survival (OS), safety, and quality of life (QOL). Changes in microbiome were an exploratory endpoint.

Findings

Overall, there were 32 evaluable patients. In the MSKD arm, 15 of 16 patients achieved nutritional ketosis; the median proportion of days in ketosis was 39.4%. The median PFS was 8.5 months in MSKD patients and 6.2 months in non‐MSKD patients: hazard ratio, 0.53 (95% CI, 0.21–1.37); one‐sided p = .096. The median OS was 13.7 months with MSKD and 10.2 months in the non‐MSKD arm: hazard ratio, 0.58 (95% CI, 0.25–1.37); one‐sided p = .107). All MSKD‐related adverse events were grade 1‐2. There were no significant differences in grade ≥3 chemotherapy‐related adverse events between the arms. MSKD patients had no decline in QOL and had significant enrichment of beneficial taxa in the microbiome (p < .05, log‐fold change ≥2).

Conclusions

The MSKD is feasible in patients with PDAC and, although not powered for definitive outcomes, shows trends in improved PFS and OS when combined with gemcitabine, nab‐paclitaxel, and cisplatin, without added toxicity or detriment to QOL. Larger studies are required to confirm these findings and establish the value of the MSKD in pancreatic cancer treatment.

Short abstract

The authors conducted a randomized phase II screening trial of gemcitabine, nab‐paclitaxel, and cisplatin with a medically supervised ketogenic diet (MSKD) versus usual diet (non‐MSKD) in patients with treatment‐naive metastatic pancreatic ductal adenocarcinoma. The MSKD is feasible and safe and showed trends toward improved progression‐free survival and overall survival in patients with treatment‐naïve metastatic pancreatic ductal adenocarcinoma receiving gemcitabine, nab‐paclitaxel, and cisplatin, without added toxicity or detriment to quality of life.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is the third leading cause of cancer‐related death, and more than 80% of patients present with locally advanced or metastatic disease. ¹ A phase Ib/II study of gemcitabine, nab‐paclitaxel, and cisplatin and a large phase II trial in patients with untreated metastatic PDAC showed promising results, with objective response rates of 40.5% to 71% and median overall survival (OS) of 9.8 to 16.4 months. ² , ³ Further improvements to first‐line therapy are urgently needed.

The intersection of diet and cancer therapy has long intrigued clinicians and researchers, with dietary interventions being explored as adjunctive strategies to conventional cancer treatments. ⁴ The ketogenic diet (KD)—a higher‐fat, lower‐carbohydrate, moderate‐protein regimen—exploits the metabolic vulnerabilities of cancer cells ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ by reducing glucose and insulin levels and increasing levels of the ketone bodies, β‐hydroxybutyrate (BHB), acetoacetate, and acetone. BHB is synthesized from circulating fats during carbohydrate restriction and serves as an energy carrier, carbon source, reducing agent, and signaling metabolite affecting gene expression, lipid metabolism, and metabolic rate. ⁵ In murine KPC pancreatic cancer models, adding the KD to gemcitabine, nab‐paclitaxel, and cisplatin synergistically suppressed tumor growth, tripling the survival benefit of chemotherapy alone. ⁶

We conducted an open‐label, randomized phase II screening trial with a relaxed alpha level ⁷ comparing a medically supervised KD (MSKD) to a usual diet (non‐MSKD) in combination with gemcitabine, nab‐paclitaxel, and cisplatin in patients with treatment‐naïve metastatic PDAC. The purpose of such a screening trial design was to decide whether the MSKD warrants a larger, definitive phase III trial. The MSKD was monitored by telemedicine using a continuous care intervention (CCI) developed by Virta Health to actively coach patients to maintain nutritional ketosis. ⁸ , ⁹ To our knowledge, this is the first randomized study to evaluate an MSKD by a CCI in patients with PDAC receiving first‐line chemotherapy.

METHODS

Ethics and compliance

The study protocol and all amendments were approved by the institutional review boards or independent ethics committee at each participating institution. All patients provided written informed consent in accordance with the Declaration of Helsinki ¹⁰ before participation in the study. The study was registered under NCT04631445.

Patient Selection

Key inclusion criteria were: age 18 years or older, histologically or cytologically confirmed metastatic PDAC, no prior treatment for metastatic disease (prior treatment in the adjuvant setting was allowed, provided at least 6 months had elapsed since completion of the last therapy), and measurable disease per Response Evaluation Criteria in Solid Tumors (RECIST) v1.1, ¹¹ Karnofsky Performance Status score of ≥70%, life expectancy ≥12 weeks, and adequate hematologic, hepatic, and renal function. Key exclusion criteria were: evidence of central nervous system metastasis, unwillingness or inability to comply with the MSKD, severe malnutrition or body mass index <18, albumin <3.0 g/dL, type 1 diabetes, and history of diabetic ketoacidosis.

Study design and treatment

Patients randomized to the MSKD arm received remote support, health coaching, daily biometric feedback, and peer support via the Virta Clinic. Nutritional and behavior change educational content was provided before starting chemotherapy and for the duration of study treatment. Access to a web‐based software application was provided for biomarker reporting and monitoring, education, and communication with the remote care team (via telemedicine) consisting of a health coach and medical provider. Dietary recommendations were to restrict total carbohydrates to ≤30 g/day (typically ≤10% of total energy intake). The daily protein intake target was 1.5 g/kg of reference body weight, and participants were coached to incorporate dietary fats to satiety. Other aspects of the diet were individually prescribed to ensure safety, effectiveness, and satisfaction, including consumption of three to five servings of nonstarchy vegetables and adequate mineral and fluid intake for the ketogenic state. A preprepared meal shipment to cover the first 24 to 48 hours of the MSKD was provided.

Patients in the MSKD arm monitored BHB levels twice daily via a home blood ketone monitor during the first month, then daily (pm) thereafter, and daily (am) glucose monitoring via a home glucometer. These values were uploaded to the Virta‐based app for close monitoring and nutritional adjustments as needed by the Virta health coach and/or medical provider. Real‐time monitoring by the Virta team allowed individualized nutrition recommendations to achieve and sustain nutritional ketosis with a goal of 0.5 to 3.0 mmol/L blood BHB. Patients were encouraged to report daily hunger, cravings, energy, and mood. These ratings and BHB concentrations were used to adjust nutritional guidance.

Patients randomized to the non‐MSKD arm did not have any nutritional intervention provided as part of their participation in the trial and served as the control arm. Patients in this arm met with a local dietitian and were instructed to follow their regular diet and not a ketogenic diet. Follow‐up with the site‐based dietician was per standard of care.

Initial consultation with the Virta Health registered dietitian for patients in the MSKD arm and the on‐site registered dietitian for patients in the non‐MSKD arm occurred within 21 days before the first dose of chemotherapy.

All patients received intravenous (IV) nab‐paclitaxel 125 mg/m², gemcitabine 1000 mg/m², and cisplatin 25 mg/m² on days 1 and 8 every 21 days until development of unacceptable toxicity, disease progression, or clinical deterioration. IV premedications consisted of dexamethasone 12 mg, palonosetron 0.25 mg, and fosaprepitant 150 mg, followed by oral dexamethasone 4 mg and ondansetron 8 mg twice daily for 2 days after each chemotherapy. The sequence of drug administration was IV hydration followed by nab‐paclitaxel, then cisplatin, and then gemcitabine. Patients received additional IV hydration on days 2 and 9 and received pegfilgrastim, 6 mg, subcutaneously on day 9 of each cycle as primary prophylaxis.

Study assessments

Contrast‐enhanced computed tomography evaluations were performed once every three cycles (prior to cycles 4, 7, 10, etc). Adverse events (AEs) were assessed according to the Common Terminology Criteria for Adverse Events v5.0. Radiological response assessments were not blinded and were conducted at each institution.

Stool samples were collected from four patients in the MSKD group (13 samples) and eight patients in the non‐MSKD group (23 samples). Deep shotgun metagenomic sequencing was conducted using NovaSeg (Illuminia). Taxonomical and functional microbiome characterization was performed using MetaPhlAn 4.0 and HUMAnN 3.6 (The Huttenhower Lab, Harvard). Bioinformatics analyses using QIIME2 were performed to analyze microbiome composition and function over time.

Study endpoints

The primary endpoint was progression‐free survival (PFS) per RECIST 1.1. PFS was defined as the time from randomization to first documentation of objective tumor progression or to death from any cause while on study.

Secondary endpoints included

• Overall survival

• The number of complete responses/partial responses as defined by computed tomography scan using RECIST 1.1

• Changes in CA 19‐9 (or CA 125, or CEA if nonexpressers of CA 19‐9) and return to normal limits (from at least >2× the upper limit of normal)

• Disease control rate using RECIST 1.1 ‐ partial response + complete response + stable disease at 9 weeks

• Change in weight

• Change in insulin levels

• Change in BHB levels

• Change in HbA1c levels

• Change in quality of life via the European Organization for Research and Treatment of Cancer Quality of Life Questionnaire QLC‐C30 (EORTC QLQ‐C30) assessment: global health status and summary score.

All randomized patients were evaluated for efficacy based on their randomization group if they were evaluable for response (modified intent‐to‐treat). Patients were defined as not evaluable for response if they did not receive at least one dose of cisplatin, nab‐paclitaxel, and gemcitabine and/or if they did not have at least one post‐baseline tumor assessment without death due to any cause. Patients with death due to any cause were considered evaluable for progression‐free survival and overall survival if they received at least one dose of cisplatin, nab‐paclitaxel, and gemcitabine, irrespective of whether they had at least one post‐baseline tumor assessment.

Statistical analysis

This pilot study was designed as a randomized phase II screening trial with a sample size of 32 (16 evaluable patients in each arm), to detect a doubling of PFS from 10 months to 20 months with a power of 80%, assuming a one‐sided alpha level of 0.20. We used this screening phase II model with an alpha level of 0.20 and moderate sample size which is supported by Rubinstein et al., ⁷ to determine whether the MSKD intervention warrants exploration in a larger phase III study. The sample size was re‐evaluated in July 2023 based on median PFS of 6.2 months observed in the larger patient number study of the triple chemotherapy regimen – the AXCLPANC study. ³ The sample size of 16 evaluable patients in each arm provided > 80% power to detect a doubling of PFS from 6.2 months to 12.4 months (assuming a one‐sided alpha level of 0.20). Kaplan‐Meier curves were created for each survival outcome, by arm. Differences in survival by arm were tested using log‐rank tests and Cox proportional hazards regression. Disease control rate at 9 weeks, best overall response (CR/PR versus other), and CA 19‐9 normalization rates were compared using Fisher’s exact tests. Differences in each continuous outcome (CA 19‐9, weight, insulin, BHB, glycated hemoglobin, EORTC QLQ‐C30 global health status, EORTC QLQ‐C30 summary score) were tested in multiple ways: change from screening to C4D1 (Wilcoxon rank‐sum test), maximum change from C1D1 (Wilcoxon rank‐sum test), normalized AUC (Wilcoxon rank‐sum test), and a linear mixed effects model that included an arm‐by‐time interaction term. Only PFS and OS were assessed using a one‐sided alpha level of 0.20; all other endpoints were assessed using a two‐sided alpha level of 0.05, and no adjustments were made for multiple comparisons. All statistical tests were conducted using Stata 18 (StataCorp, College Station, TX).

RESULTS

Patient characteristics

Fifty‐six patients with untreated metastatic PDAC were consented, of which 41 were eligible and 36 were enrolled and randomized (Figure 1). Among 32 evaluable patients (median age 65.9 years; 53% male), 16 were randomized to each arm. Baseline characteristics were generally well balanced between the arms. Both arms had similar baseline weight, HgbA1c, insulin, and BHB levels. CA 19‐9 levels and neutrophil‐to‐lymphocyte ratio were non‐significantly higher in the non‐MSKD arm at baseline (Table 1).

FIGURE 1.

CONSORT diagram.

TABLE 1.

Baseline patient characteristics.

Characteristic a	Non‐MSKD n = 16	MSKD n = 16	Total n = 32	p value b
Age (year)
Median	65.9	66.1	65.9	.985
Range	46.0–75.6	40.7–75.8	40.7–75.8
Weight (kg)
Median	76.3	73.2	74.2	.731
Range	52.5–120.5	45.9–127.0	45.9–127.0
Glucose c (mg/dL)
Median	112.5	105.5	110.5	.485
Range	84.0–171.0	85.0–208.0	84.0–208.0
HbA1c (%)
Median	6	5.9	6	.545
Range	5.4–7.6	4.6–9.0	4.6–9.0
Insulin (pmol/L)
Median	90.3	76.4	78.5	.744
Range	27.8–659.8	29.2–138.9	27.8–659.8
BHB (mmol/L)
Median	.29	.26	.26	.940
Range	0.08–0.97	0.07–1.10	0.07–1.10
Neutrophil‐to‐lymphocyte ratio
Median	4.27	2.55	3.35	.119
Range	1.90–6.88	1.65–7.13	1.65–7.13
Sex, n (%)				.723
Male	9 (56.2)	8 (50.0)	17 (53.1)
Female	7 (43.8)	8 (50.0)	15 (46.9)
Race/ethnicity, n (%)				.451
Non‐Hispanic White	10 (66.7)	14 (87.5)	24 (77.4)
Hispanic	3 (20.0)	1 (6.2)	4 (12.9)
Black	1 (6.7)	1 (6.2)	2 (6.5)
American Indian or Alaska Native	1 (6.7)	0 (0.0)	1 (3.2)
Karnofsky performance status, n (%)				.319
80	8 (50.0)	4 (25.0)	12 (37.5)
90	6 (37.5)	8 (50.0)	14 (43.8)
100	2 (12.5)	4 (25.0)	6 (18.8)
History of diabetes, n (%)	5 (31.2)	4 (25.0)	9 (28.1)	.694
Insulin use, n (%)	3 (18.8)	3 (18.8)	6 (18.8)	1.000
CA 19‐9 expressor (>35), n (%)	16 (100.0)	13 (81.2)	29 (90.6)	.069
CA 19‐9 among expressors (U/mL)
Median	8954	3479	5484	.329
Range	512–336,300	102–200,000	102–336,300

Abbreviations: BHB, β‐hydroxybutyrate; HbA1c, glycated hemoglobin; MSKD, medically supervised ketogenic diet.

^a Missing data: insulin (n = 10), BHB (n = 3), race/ethnicity (n = 1).

^b Rank‐sum test (continuous variables) or chi‐squared test (categorical variables).

^c Glucose levels were obtained in the clinic and were nonfasting.

Patients in the MSKD arm received a median (range) duration of 8.0 (2.0–19.0) cycles of therapy, while the non‐MSKD arm had a median (range) treatment duration of 6.5 (1.0–16.5) cycles. There was no significant difference in treatment duration between the arms (p = .249). Most patients in both arms completed at least 3 cycles of therapy (88% non‐MSKD, 94% MSKD).

Feasibility and safety of the medically supervised ketogenic diet (MSKD)

In the MSKD arm, BHB levels were monitored by daily fingerstick tracking which were submitted remotely to the Virta Health team. On days with multiple measurements, the maximum value was used for determining whether the participant was in ketosis.

In both arms, BHB levels were also measured at each site through the institutional laboratory: on the first day of every three cycles in the MSKD arm and on the first day of every cycle in the non‐MSKD arm.

Using daily ketone tracking, 15 of 16 patients in the MSKD arm achieved nutritional ketosis (0.5–3.0 mM) at any point during the study, with mean BHB levels of 0.49 mM (95% CI 0.36–0.63) and median proportion of days in ketosis of 39.4% (range 0–95.8%). The mean percentage of treatment days that a participant submitted a BHB measurement was 58.5% (range 3.5–100%). In an exploratory analysis, using a BHB cutoff of ≥ 0.3 mM, all 16 MSKD patients achieved ketosis at any point in the study.

Based on laboratory data obtained on the first day of every three cycles, the MSKD arm had a significantly higher change in median BHB levels from C1D1 to C4D1 (median, IQR) (0.04, −0.03 to 0.78) versus the non‐MSKD arm (−0.05, −0.41 to −0.02) arm (p = .021).

All MSKD‐related adverse events (AEs) were Grade 1‐2 and included fatigue (n = 3), constipation (n = 3), weight loss (n = 3), decreased appetite (n = 1), dehydration (n = 1), dizziness (n = 1), nausea (n = 1) and body ache (n = 1). The incidence of these Grade 1‐2 MSKD‐related AEs was not different than those reported by MSKD participants that were unrelated to diet, or the incidence in non‐MSKD participants. None of the patients stopped the MSKD due to AEs.

There were no unexpected chemotherapy‐related AEs and no treatment‐related deaths on study. When examining Grade ≥ 3 chemotherapy‐related AEs between the groups, there were no significant differences between the two arms for anemia, diarrhea, or platelet count decrease (Fisher's exact test, all p > .2) (Table 2, Supplementary Table S1, and Table S2).

TABLE 2.

Summary of grade ≥ 3 chemotherapy‐related adverse events.

Adverse event	Non‐MSKD (n = 16)	MSKD (n = 16)
Anemia	9 (56.3%)	6 (37.5%)
Neutrophil count decreased	6 (37.5%)	8 (50%)
Febrile neutropenia	1 (6.2%)	0
Platelet count decreased	13 (81.2%)	9 (56.2%)
WBC decreased	2 (12.5%)	2 (12.5%)
Dehydration	0	1 (6.2%)
Hypokalemia	2 (12.5%)	1 (6.2%)
Hyponatremia	1 (6.2%)	0
Hypophosphatemia	1 (6.2%)	0
Peripheral sensory neuropathy	0	1 (6.2%)
Hematuria	1 (6.2%)	0
Pulmonary embolism	0	1 (6.2%)
Sepsis	1 (6.2%)	0
Colitis	1 (6.2%)	0
Diarrhea	3 (18.8%)	0
GI hemorrhage	1 (6.2%)	0
Intra‐abdominal hemorrhage	1 (6.2%)	0
Nausea	0	1 (6.2%)
Vomiting	0	1 (6.2%)
Aspartate transaminase increased	0	1 (6.2%)
Fatigue	0	1 (6.2%)
Mucosal inflammation	1 (6.2%)	0

Abbreviations: GI, gastrointestinal; MSKD, medically supervised ketogenic diet; WBC, white blood cell.

Efficacy and survival outcomes

The endpoints of progression‐free survival (PFS) and overall survival (OS) were examined using a one‐sided alpha level of 0.20. All other endpoints were assessed using a two‐sided alpha level of 0.05.

Using RECIST criteria, patients on the MSKD arm had a median PFS of 8.5 months, as compared to 6.2 months on the non‐MSKD arm, HR 0.53 (95% CI, 0.21–1.37, one‐sided p = .096) (Table 3 and Figure 2). Furthermore, patients in the MSKD arm had an improved median OS compared to those on the non‐MSKD arm, 13.7 versus 10.2 months, HR 0.58 (95% CI 0.25–1.37, one‐sided p = .107) (Figure 3).

TABLE 3.

Progression‐free survival (PFS), overall survival (OS), disease control rate (DCR).

Outcome	Non‐MSKD (n = 16)	MSKD (n = 16)	p two‐sided	p one‐sided
PFS (RECIST)
Failure, n (%)	10 (62.5%)	8 (50.0%)
Median survival (95% CI), months	6.2 (3.2– a )	8.5 (4.7–9.6)
Log‐rank test			.182	.091
Hazard ratio (95% CI) b	1.0 (ref.)	0.53 (0.21–1.37)	.192	.096
OS
Failure, n (%)	13 (81.2%)	9 (56.2%)
Median survival (95% CI), months	10.2 (4.7–16.0)	13.7 (9.1– a )
Log‐rank test			.209	.104
Hazard ratio (95% CI) b	1.0 (ref.)	0.58 (0.25–1.37)	.214	.107
Best overall response
Complete response	0 (0.0%)	0 (0.0%)
Partial response	5 (31.2%)	11 (68.8%)
Stable disease	9 (56.2%)	4 (25.0%)
Progressive disease	2 (12.5%)	1 (6.2%)
DCR at 9 weeks c	14 (87.5%)	15 (93.8%)

Abbreviation: MSKD, medically supervised ketogenic diet.

^a Upper bound of the 95% CI cannot be estimated because of the small sample size.

^b Cox proportional hazards regression model.

^c Includes two patients whose end‐of‐study scans occurred at 6.0 and 8.7 weeks, respectively.

FIGURE 2.

Progression‐free survival.

FIGURE 3.

Overall survival.

The disease control rate (DCR) at 9 weeks did not significantly differ between the non‐MSKD (87.5%) and MSKD arms (93.8%) (p = 1.000). More patients on the MSKD arm achieved a best overall response of partial response (PR) compared to the non‐MSKD arm, but this did not reach statistical significance (68.8% versus 31.2%, p = .110) (Table 3). The maximum change in target lesion tumor burden (an exploratory endpoint) did not significantly differ between the two arms (p = .090). Figure 4 is the swimmers plot illustrating time on treatment, RECIST progression, death, and time at censoring for alive patients.

FIGURE 4.

Efficacy outcomes. (A) A waterfall plot of the maximum percent change from baseline in target lesion tumor burden (sum of the longest diameters) per investigator assessment by treatment arm. (B) A swimmer plot of outcomes, including time on treatment, PD by RECIST, and patient status by treatment arm. Arrows indicate that the patient is alive at last contact.

Patients in the MSKD arm had a greater CA 19‐9 decline than those in the non‐MSKD arm (linear mixed effect model, p‐interaction = .076). There was no significant difference in the rate of CA 19‐9 normalization between the two arms, MSKD 23% versus non‐MSKD 6% (p = .299).

Changes in metabolic parameters, microbiome and quality of life measures

Blood metabolic parameters were similar between the arms at baseline (Table 1). Over the course of the study, patients in the non‐MSKD arm gained more weight than in the MSKD arm (linear mixed effects model, p‐interaction <.001). Importantly, there was no significant weight loss and weight remained stable between C1D1 and C4D1 in the MSKD arm (data not shown).

During study treatment, glucose levels were higher in the non‐MSKD arm than in the MSKD arm (AUC, p = .019). Similarly, glucose levels increased more across time in the non‐MSKD arm than in the MSKD arm (linear mixed effects model, p‐interaction = .072). Furthermore, glycated hemoglobin levels decreased more over time in the MSKD arm compared to the non‐MSKD arm (linear mixed effects model, p for interaction = .076). No significant differences between arms were seen in insulin levels for the study duration.

Microbiome profiles between the two groups remained distinct throughout therapy. Though not statistically significant due to sample size, MSKD patients showed lower alpha diversity (Shannon diversity; p = .13), likely reflecting reduced intake of fermentable fiber. Differential abundance analysis using the ANCOM‐BC method revealed a significant enrichment of beneficial taxa in the MSKD group compared to the non‐MSKD group (p < .05, log‐fold change ≥2), including Akkermansia muciniphila (linked to improved response to immune checkpoint inhibitors; p = .018, q = 1), Intestinimonas butyriciproducens (a short‐chain fatty acid and butyrate producer; p < .0001, q = .017), and Streptococcus thermophilus (a supporter of gut mucosal health; p < .0001, q = .006).

During the study, only MSKD patients showed sustained increases in Roseburia hominis (short‐chain fatty acid/butyrate producer; p < .0001, q = .021) and Lacticaseibacillus rhamnosus (with anti‐inflammatory and immune‐modulating properties; p < .0001, q < .0001), along with a marked reduction in potentially pro‐inflammatory Actinomyces spp. (p < .0001, q = .024).

Functional profiling showed increased abundance of cancer‐relevant microbial pathways in MSKD patients, including phosphatidate metabolism (PWY‐7039; involved in lipid signaling and immune regulation; p = .0004, q = .20) and allantoin degradation (PWY‐5705; linked to redox balance; p < .0001, q = .004). Conversely, L‐ascorbate biosynthesis V (PWY‐6415) was reduced (p = .0004, q = .17), likely because of decreased D‐galacturonate substrate availability from fiber restriction.

With respect to quality of life (QOL) measures, EORTC QLQ‐C30 Global Health Status and Summary scores did not significantly differ at baseline between arms. Over the course of therapy, summary scores increased more in the non‐MSKD arm compared to the MSKD arm (linear mixed effects model, p for interaction = .005). In the MSKD arm, summary scores remained unchanged from C1D1 to C4D1. Similarly, Global Health Status scores increased significantly more in the non‐MSKD arm between screening and C4D1, compared to the MSKD arm (Wilcoxon rank‐sum test, p = .045). Importantly, there was no significant decline in either QOL score in the MSKD arm (Table 4).

TABLE 4.

EORTC QLQ‐C30 global health and summary scores during study treatment.

Score	Non‐MSKD (n = 15)	MSKD (n = 16)	p
Global health score
Median (IQR)
Screening	58.3 (41.7, 75.0)	70.8 (58.3, 91.7)
C4D1 a	75.0 (66.7, 83.3)	66.7 (50.0, 83.3)
Change, screening to C4D1 a	16.7 (0.0, 25.0)	0.0 (−8.3, 0.0)	.045 b
Maximum change from C1D1	−12.5 (−33.3, 33.3)	−16.7 (−33.3, 25.0)	.553 b
Normalized AUC	70.8 (52.0, 80.3)	77.1 (61.0, 79.4)	.498 b
Linear mixed effects model	0.208 (0.120) c	−0.013 (0.093) c	.145 d
Summary score
Median (IQR)
Screening	72.6 (60.7, 82.1)	80.0 (60.2, 93.1)
C4D1 a	84.1 (75.8, 89.7)	80.0 (66.7, 83.7)
Change, screening to C4D1 a	7.6 (−4.8, 17.1)	0.1 (−1.4, 3.9)	.133 b
Maximum change from C1D1	14.0 (−20.4, 17.1)	−10.5 (−19.7, 13.6)	.313 b
Normalized AUC	84.3 (73.6, 89.7)	85.5 (70.3, 89.2)	.918 b
Linear mixed effects model	0.261 (0.073) c	0.001 (0.057) c	.005 d

Abbreviations: AUC, area under the curve; IQR, interquartile range; MSKD, medically supervised ketogenic diet.

^a Sample size for C4D1: n = 11 (non‐keto diet), n = 11 (keto diet).

^b Wilcoxon rank‐sum test.

^c Beta coefficient (standard error) for effect of time on summary score.

^d Test for arm‐by‐time interaction.

DISCUSSION

Our randomized phase II screening trial is the first to demonstrate that the MSKD is feasible with an acceptable safety profile and is associated with trends in improved PFS and OS in patients with metastatic PDAC receiving chemotherapy.

The “ketogenic diet” was first formally described more than 100 years ago as an effective therapy for epilepsy. ¹² Most evidence assessing the KD as an adjunctive cancer therapy lies in glioblastoma, ¹³ where a systematic analysis suggested a survival benefit, but the studies were of limited size. ¹⁴ Similarly, KD trials in breast and other solid tumors have been small, heterogeneous, and shown mixed findings. ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹

Accumulating evidence supports BHB as an antitumor effector alongside cytotoxic and immunotherapy in PDAC. Mechanistically, BHB has far‐reaching effects on cancer cell metabolism, epigenetic and transcriptional regulation, and the gut microbiome. ²⁰ Importantly, BHB has been shown to serve as a histone deacetylation inhibitor. ²⁰ Findings from microbiome analysis in our study demonstrated enrichment of anti‐inflammatory and immune‐mediating bacteria in the MSKD group, which may have partly contributed to the improved outcomes and warrant further exploration. Furthermore, KD‐mediated pancreatic tumor regression has been linked to reduction in circulating insulin levels and decrease in the preferred fuel source of glucose, leading to ERK and AKT inhibition and upregulation of JAK/STAT3 and interferon signaling. ⁶ , ²⁰ , ²¹

In the preclinical impetus for this trial, adding the KD to triplet chemotherapy in murine KPC pancreatic tumor models led to decreased tumor glucose use, enhanced 3‐hydroxybutyrate use, and boosted reactive oxygen species, leading to more durable benefits from chemotherapy. ⁶ Furthermore, glutamine metabolism inhibitors combined with the KD had robust effects in these murine models, revealing an additional metabolic vulnerability to be explored. ²² , ²³ In a recent study of immunodeficient murine xenograft PDAC models undergoing anti‐mesothelin CAR‐T therapy, BHB supplementation resulted in higher peripheral CAR‐T cell expansion, elevated serum effector cytokines, and more frequent tumor responses compared to controls. ²⁴

Consistently following a prescribed diet is difficult, and patients with PDAC have compounding challenges, including taste alterations, abdominal symptoms, anorexia‐cachexia, asthenia, and depression. Although certain patients will use the opportunity of an MSKD as needed guidance and part of their disease that they can control, others may find the routine monitoring burdensome and the allowed diet restrictive. Importantly, in our study, no patients assigned to the MSKD discontinued the diet because of adverse events, and there were no significant differences in treatment‐related AEs, including cytopenias, gastrointestinal effects, or peripheral neuropathy between the arms.

With respect to trial design, most nutritional cancer research to date has been in the preventive or adjuvant settings and reductionist in nature, focusing on one nutrient or food but not addressing the complexity of the whole diet or guiding patients with active cancer on therapy. ²⁵ , ²⁶ Conversely, even specific diets retain a degree of heterogeneity, which makes discerning the anticancer element complex. Furthermore, dietary interventional studies often rely on patient recollection and assume accurate documentation of intake into diaries which makes capturing compliance cumbersome and less reliable.

The MSKD circumvents these challenges by using a CCI and relying on goal BHB levels as an objective measure and biologic correlate of compliance. Using a benchmark BHB level of 0.5 to 3.0 mM and with a median 39.4% of days in ketosis, the MSKD led to observed trends toward improvements in PFS and OS. It is critical to note that ketogenic diets examined in the literature widely vary with respect to how ketosis is defined and achieved, including different caloric allowances and proportions of fats, proteins, and carbohydrates. The MSKD in this study allows eating to satiety within the prescribed macronutrient ratios, with the aim of balancing adherence with efficacy.

Notably, patients on the non‐MSKD arm experienced improved QLQ‐C30 Global Health Status scores during the first three cycles of chemotherapy, consistent with previous studies of the triplet regimen, ³ whereas those on the MSKD did not demonstrate improvement. Importantly, there was no decline in QOL in the MSKD group.

Merging mechanistic rationale with insights from this initial clinical experience, future investigations should aim to: (1) optimize the level, duration, and timing of ketosis alongside systemic therapy; (2) explore the synergy between the MSKD and ketone supplementation, sodium‐glucose cotransporter 2 inhibitors, and/or glutamine metabolism inhibitors; (3) determine the utility of noninvasive markers and metabolites (e.g., insulin, glucose, gut microbiome) in guiding the use of the MSKD; (4) integrate the MSKD into the perioperative setting and with other treatment modalities (e.g., radiation) and novel therapeutics; and (5) examine the benefit of the MSKD in clinically relevant subgroups including those defined by demographic characteristics, tumor gene expression profiles, and radiographic correlates (i.e., sarcopenia indices, site of tumor metastases).

Limitations of the study

Our study has its strengths and limitations. The primary strength is the randomized design, but the small sample size not powered for definitive outcomes is a key limitation and is a result of having restricted resources. This was a phase 2 screening trial that used a relaxed alpha level of 0.20 for the outcomes of PFS and OS. As such, the results of our study are ultimately hypothesis‐generating and intended to guide the decision to proceed with a phase III trial. Nonetheless, although the study did not meet its primary endpoint by conventional statistical standards, the observed trends warrant further investigation in a larger, adequately powered trial. Our study used gemcitabine, nab‐paclitaxel, and cisplatin; whether the MSKD yields clinical benefit with more commonly used first‐line regimens (gemcitabine plus nab‐paclitaxel, FOLFIRINOX) is unknown. The frequency and method of BHB measurements varied between the arms, which limits the strength of this comparison. Another important limitation is that the radiological review for response assessment was not blinded. We acknowledge that broad application of the MSKD model may not be practical in certain socioeconomic contexts, including underserved populations or patients without access to smartphones.

Conclusions

The MSKD is feasible and safe, and trends were observed toward improved PFS and OS in patients with treatment‐naïve metastatic PDAC receiving gemcitabine, nab‐paclitaxel, and cisplatin. Our trial demonstrates the feasibility of conducting a multi‐institutional randomized dietary intervention in a medically complex patient population and hopefully paves the way for larger studies, which are warranted.

Author contributions

Gayle S. Jameson: Conceptualization; investigation; writing — original draft; writing — review and editing; resources; data curation. Denise J. Roe: Conceptualization; writing — original draft; writing — review and editing; methodology; formal analysis. Erkut Borazanci: Conceptualization; investigation; writing — original draft; writing — review and editing; resources; data curation. Diana L. Hanna: Conceptualization; investigation; writing — original draft; writing — review and editing; data curation; resources. Caroline G.P. Roberts: Conceptualization; writing — review and editing; writing — original draft; resources; investigation; data curation. Meredith S. Pelster: Investigation; writing — review and editing; resources; data curation. Richard C. Frank: Investigation; writing — review and editing; resources; data curation. Angela T. Alistar: Investigation; writing — review and editing; resources; data curation. Alan M. Miller: Writing — original draft; writing — review and editing; project administration. J. Erin Wiedmeier‐Nutor: Conceptualization; writing — review and editing. Sandra D. Algaze: Investigation; writing — review and editing; data curation; resources. Alison R. Zoller: Conceptualization; writing — original draft; writing — review and editing; resources; data curation; investigation. Sarah J. Hallberg: Conceptualization; resources. Betsy C. Wertheim: Conceptualization; methodology; formal analysis; writing — review and editing; writing — original draft. Keehoon Lee: Writing — review and editing; formal analysis; resources; data curation; investigation. Derek Cridebring: Conceptualization; writing — review and editing. Joshua D. Rabinowitz: Conceptualization; writing — review and editing. Stephen Gately: Conceptualization; funding acquisition; writing — original draft; writing — review and editing; project administration; supervision; resources; validation. Jennifer Keppler: Writing — review and editing; project administration. Sunil Sharma: Investigation; writing — review and editing; resources; data curation. Daniel D. Von Hoff: Conceptualization; funding acquisition; writing — original draft; writing — review and editing; supervision; validation; resources; data curation. Drew W. Rasco: Investigation; writing — review and editing; resources; data curation.

CONFLICT OF INTEREST STATEMENT

G.J.: Bristol Myers Squibb. E.B.: Merus, VCN, Corcept, Atheneum, Taiho, and Arcus Biosciences (consulting). C.R.: Virta Health (employee, stock options). M.P.: Abbvie, Actuate Therapeutics Affini‐T Therapeutics, Agenus, Arcus Biosciences, Artios, Astellas, BeiGene, BioNTech, Bristol‐Myers Squibb, Codiak, Compass, CytomX, Eisai, Elevation Oncology, Elicio, Exelixis, Fate Therapeutics, Fog Pharmaceuticals, Gilead, GlaxoSmithKline, HiberCell, Immune‐Onc Therapeutics, Impact Therapeutics, Jazz Pharmaceuticals, Kura Oncology, Leap Therapeutics, Neogene, Novartis, OncXerna Therapeutics, Panbela Therapeutics, Revolution Medicines, Roche, SeaGen, SQZ Biotechnologies, Surface oncology, Tachyon Therapeutics, Takeda, TD2, Translational Genomics, TransThera Sciences, ZielBio, and 1200 Pharma (Research Funding); and Arcus Biosciences, AstraZeneca, Curio Science, CytomX, Elevation Oncology, EMD Serono, Ipsen Biopharmaceuticals, Jazz Pharmaceuticals, Kura Oncology, Pfizer, and Takeda (consulting). A.Z.: Virta Health (employee, stock options). S.H.: Virta Health (employee, stock options). J.R.: Colorado Research Partners, Bantam Pharmaceuticals, Rafael Pharmaceuticals, and Empress Therapeutics (advisor and stockholder); Farber Partners and Raze Therapeutics (founder, director, and stockholder); Marea Therapeutics and Fargo Biotechnologies (founder, advisor, and stockholder); and Princeton University (inventor of patents). D.V.H.: Medtronic, CerRx, SynDevRx, United Healthcare, Anthem, Inc. Stromatis Pharma, Systems Oncology, StingRay, Orpheus Bioscience, AADI, Origin Commercial Advisors, Halia Therapeutics, Lycia Therapeutics, (3 + 2) Pharma, and AcuViz (stockholder/ownership interests); Imaging Endpoints, CanBas, Lixte Biotechnology, TD2, Phosplatin Therapeutics, SOTIO, Immunophotonics, Oncology Venture, Novita Pharmaceuticals, Vicus Therapeutics, Sirnaomics, AiMed Bio, Erimos Pharma, Pfizer, ImmuneOncia, Viracta Therapeutics, AlaMab, Xerient, Lycia Therapeutics, EXACT Therapeutics, ImaginAb, SignaBlok, Compass Therapeutics, Sellas Life Sciences, Catamaran Bio, Remix Therapeutics, SMP Oncology fka SDP/Tolero, Coordination Pharmaceuticals, Orphagen Pharmaceuticals, Red Arrow Therapeutics, Soley Therapeutics, Invios GmbH, Mekkanistic Therapeutics, POINT Biopharma, Peptomyc, Remunity, SIWA Therapeutics, Xenthera, Indaptus fka Decoy, Panavance Therapeutics fka Geistlich, CyMon Bio, Bryologyx, Moleculin Biotech, EnGeneIC, Race Oncology, Autonomix, Econic Biosciences, Crinetics Pharmaceuticals, Diakonos Research and Improve Bio (consulting); Lillly, Genentech, Celgene, Incyte, Merrimack, Plexxikon, Minneamrita Therapeutics, Abvie, Aduro, Cleave Biosciences, CytRx, Daiichi Sankyo, Deciphera, Endocyte, Exelixix, Five Prime Therapeutics, Gilead Science, Merck, Pfizer, Pharmacyclics, Phoenix Biotech, Samumed, Strategia, and Halozyme (research funding); and Intramedullary Catheter, Methods of Human Prostate Cancer, Use of 5,6‐Dihydro‐5‐Azacytidine in the Treatment of Prostate Cancer, Targeting Site‐2 Protease (S2P) for the Treatment of Pancreatic Cancer (pending), Targeting Ecto‐5‐Nucleotidase (Cd73) for the Treatment of Pancreatic Cancer, Targeting a Protein Tyrosine Phosphotase‐PRL‐1 for the Treatment of Pancreatic Cancer (pending), Targeting a Protein PRC1 for the Treatment of Pancreatic Cancer (pending), Targeting Ecto‐5‐Nucleotidase (CD73) for the Treatment of Pancreatic Cancer (pending), Protein Kinase Inhibitors (pending), Methods, Compounds and Compositions with Genotype Selective Anticancer Activity (pending), Methods and Kits to Predict Therapeutic Outcome of BTK Inhibitors (pending), Muscle Fatigue Substance Cytokines and Methods of Inhibiting Tumor Growth Therewith (pending), and 2‐aryl‐pyridylazoles for the Treatment of Solid Tumors such as Pancreatic Cancer (pending) (patents, royalites, and other intellectual property).

Supporting information

Supplementary Material

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.