Participant Characteristics
Table 1 presents baseline characteristics of the 262 CCI and 87 UC participants. At baseline, 88% of CCI participants were prescribed diabetes medication (57% were prescribed a diabetes medication other than metformin, 30% prescribed insulin) and 93% were obese. Eighty-seven percent of participants in UC at baseline were prescribed diabetes medication (46% prescribed insulin), and 82% were obese. Forty-four participants (16.8%) withdrew from the CCI, 22 from each education delivery mode. Baseline characteristics of CCI dropouts did not differ significantly from the 218 completers except none of the five thiazolidinedione users were dropouts (Table 1). At baseline, characteristics of CCI participants who self-selected web-based versus on-site education were not significantly different after accounting for multiple comparisons (see Table S1 in the electronic supplementary material). Compared to the 78 UC participants who completed the study, the nine that withdrew tended to be older (58 versus 52 years old), had lower TSH, and fewer were prescribed insulin, SGLT-2, DPP-4, GLP-1, or blood pressure medications (Table 1).
Effectiveness
Table 2 presents mean 1-year changes in biomarkers. In the CCI, HbA1c was significantly reduced 17%, from 60 ± 1.0 mmol mol−1 (7.6 ± 0.09%) at baseline to 45 ± 0.8 mmol mol−1 (6.3 ± 0.07%) after 1 year (nominal significance P < 1.0 × 10−16; Fig. 1). Eighty-five percent (174/204) of CCI participants completing 1-year HbA1c testing observed a decline greater than 2.2 mmol mol−1 (> 0.2%) in the measure. When adjusted for multiple comparisons, significant within-CCI reductions were observed in fasting glucose (− 22%, P < 1.0 × 10−16), fasting insulin (− 43%, P = 6.7 × 10−16), C-peptide (− 23%, P = 2.2 × 10−16), HOMA-IR derived from fasting insulin excluding exogenous users (− 55%, P = 3.2 × 10−5), HOMA-IR derived from C-peptide (− 29%, P = 1.0 × 10−13), weight from clinic measurements (− 12%, P < 1.0 × 10−16), weight from home scales (− 13%, P < 1.0 × 10−16, Fig. 2), triglycerides (− 24%, P < 1.0 × 10−16), high-sensitivity C-reactive protein (− 39%, P < 1.0 × 10−16), ALT (− 30%, P = 2.4 × 10−10), AST (− 21%, P = 5.1 × 10−7), and alkaline phosphatase (− 13%, P < 1.0 × 10−16). HDL-cholesterol increased 18% (P < 1.0 × 10−16) and calculated LDL-cholesterol increased 10% (P = 5.1 × 10−5) while apolipoprotein B (ApoB) concentration was unchanged (P = 0.37) for participants in the CCI. There were no significant differences in mean biomarker changes between CCI-web and CCI-onsite (see Table S2 in the electronic supplementary material). In contrast to the CCI, patients enrolled in UC for 1 year showed no Bonferroni-adjusted significant change for any of the biomarkers measured (Table 2).
Following 1 year of CCI, usage of all diabetes medications combined (excluding metformin) was reduced significantly (56.9 ± 3.1% to 29.7 ± 3.0%, P < 1.0 × 10−16) through decreased prescriptions for DPP-4 (9.9–6.3%, P = 0.11), insulin (29.8–16.7%, P = 4.3 × 10−9), SGLT-2 inhibitors (10.3–0.9%, P = 9 × 10−7), sulfonylureas (23.7–0%, P < 1.0 × 10−16), and thiazolidinediones (1.5–0.4%, P = 0.23) (Fig. 3). GLP-1 prescriptions were statistically unchanged (13.4% at baseline to 14.4% at 1 year, P = 0.67), and metformin decreased slightly (71.4–65.0%, P = 0.04) for CCI participants. Forty percent (31/78) of CCI participants who began the study with insulin prescriptions (average dose of 64.2 units) eliminated the medication, while the remaining 60% (47/78) of insulin users reduced daily dosage from 105.2 to 53.8 units (P < 0.0001). Patients enrolled in UC for 1 year showed no Bonferroni-adjusted significant change for prescription of medication. For the 34 UC participants that continued using insulin, the average daily dose increased from 96.0 to 111.9 units.
The proportion of participants in the total imputed CCI group with HbA1cbelow 48 mmol mol−1 (< 6.5%) increased from 19.5 ± 2.4% to 69.8 ± 3.1%. Of those in the CCI with HbA1c reported at 1 year, 72% (147/204) achieved HbA1c below 48 mmol mol−1 (6.5%) and 60.3% (123/204) of participants achieved HbA1c below 48 mmol mol−1 (< 6.5%) while taking no diabetes medication or only metformin. Of those in the CCI with HbA1c below 48 mmol mol−1 (< 6.5%) at 1 year, 42.3% (52/123) were prescribed no diabetes medication and 57.7% (71/123) were prescribed metformin only. The proportion of the total imputed CCI group with fasting glucose below 6.99 mmol L−1 at 1 year increased from 34.9 ± 3.3% to 58.4 ± 3.9%, and the proportion with class III obesity decreased from 45.5 ± 3.1% to 19.6 ± 2.8%.
Compared to UC, the CCI showed significant Bonferroni-adjusted (P < 0.0017) net reductions in HbA1c (nominal significance for the two-group comparison, P < 10−16; Fig. 1), fasting glucose (P = 2.1 × 10−6), fasting insulin excluding exogenous users (P = 4.6 × 10−5), C-peptide (P = 5.3 × 10−5), HOMA-IR derived from insulin excluding exogenous users (P = 6.0 × 10−5)
or derived from C-peptide (P = 3.0 × 10−5), weight (P < 10−16), triglycerides (P = 1.0 × 10−6), hsCRP (P = 9.3 × 10−7), ALT (P = 4.6 × 10−5), and alkaline phosphatase (P = 3.1 × 10−8). All of these group differences remained significant when adjusted for the baseline age, sex, insulin medication use, and body mass index (Table 2). The CCI decrease in diabetes medication use was significantly greater than the changes in the UC group for all diabetes medications (P < 10−16) and all diabetes medications excluding metformin (P = 9.0 × 10−9), including sulfonylurea (P = 3.3 × 10−7) and insulin (P = 0.0002) (Fig. 3).
The CCI-web and CC-onsite sub-cohorts provide replication of the above results. Specifically, Table S2 (see electronic supplementary material) shows that within-group Bonferroni significance was achieved separately for the mean 1-year reductions in HbA1c, fasting glucose, fasting insulin, C-peptide, HOMA-IR, triglycerides, and hsCRP, and the significant increases in HDL-cholesterol and LDL-cholesterol. The Bonferroni-adjusted significant differences from the UC cohort were also replicated by the two educational sub-cohorts for HbA1c, fasting glucose, insulin-derived HOMA-IR, weight, HDL-cholesterol, LDL-cholesterol, triglycerides, hsCRP, and alkaline phosphatase, with or without adjustment for baseline covariates.
Time Course of Biomarker Change in CCI
Over the course of the intervention at baseline, 70 days [23], and 1 year, the proportion of participants in the total imputed CCI with HbA1c below 48 mmol mol−1 (< 6.5%) increased from 19.5 ± 2.4 to 60.7 ± 3.1 to 69.8 ± 3.1%; the proportion with fasting glucose below 6.99 mmol L−1 (< 126 mg dL−1) increased from 34.9 ± 3.3 to 55.5 ± 3.3 to 58.4 ± 3.9%, and the proportion with class III obesity decreased from 45.5 ± 3.1, to 30.2 ± 3.1, to 19.6 ± 2.8%.
The time course of biomarker changes also differed by variable (see Table S3 in the electronic supplementary material). Most of the 1-year improvements in diabetes risk factors were achieved during the first 70 days of the intervention including 84% of the HbA1c decrease, 90% of the fasting glucose decrease, 73% of the fasting insulin decrease, 64% of the C-peptide decrease, and 87% and 74% of the decreases in HOMA-IR as estimated from fasting insulin and C-peptide concentrations, respectively. Improvements in blood pressure also mostly occurred in the initial 70 days, as did reductions in alkaline phosphatase, serum creatinine, and eGFR. Most of the plasma triglyceride decrease occurred during the first 70 days (87%), whereas essentially all the substantial increase in HDL-cholesterol occurred between the initial 70 days of the intervention and 1 year (99%). About 60% of weight loss occurred in the first 70 days.
Retention and Adherence in CCI
Eighty-three percent of participants remained enrolled in the CCI at 1 year. Nearly all CCI participants (96%) reported at least one BHB reading of 0.5 mmol L−1 or more by handheld measure, and among completers, the group mean at 70 days by laboratory measure was over threefold the baseline (0.54 ± 0.04 versus 0.17 ± 0.01 mmol L−1). Laboratory-measured BHB at 1 year (0.31 ± 0.03 mmol L−1) was nearly double the baseline value (Fig. 4). The intention-to-treat analysis yielded similar results, with an increase in average from baseline (0.17 ± 0.01 mmol L−1) to 70 days (0.54 ± 0.04 mmol L−1), followed by a decrease at 1 year (0.30 ± 0.02 mmol L−1), though still nearly twofold the baseline concentrations.
Safety and Adverse Events
For CCI participants, acid–base physiology was normal; no cases of metabolic acidosis were observed. One CCI patient (0.38% of starters) had a clinically significant rise in serum creatinine, but group mean declined at 1 year. Mean blood urea nitrogen increased significantly in the CCI group, possibly indicating increased dietary protein consumption although high protein intake was not recommended. Mean uric acid in the CCI rose transiently at 70 days, but was unchanged at 1 year; no new cases of gout were diagnosed. Mean free T4 level was unchanged, and TSH was significantly lower at 1 year; two new cases of subclinical hypothyroidism were observed (0.76% of starters) in the CCI [24].
Adverse events occurred in 6/262 CCI participants including one non-ST-segment myocardial infarction, one inferior myocardial ischemia by electrocardiogram, one metastatic neuroendocrine carcinoma, one malignant cancer with multiple brain lesions and lung tumor, and death from renal hemorrhage and failure and hyperkalemia. Also, one episode of hypoglycemia occurred following a motor vehicle accident and medical records indicated the patient was not taking insulin as prescribed; no other episodes of symptomatic hypoglycemia requiring assistance were reported. None of the adverse events were attributed to the intervention.
Adverse events were reported in 6/87 UC participants, including one percutaneous coronary intervention (PCI) to left anterior descending stenosis, one PCI to right coronary artery, two carotid endarterectomies (one of which was successful), multifactorial encephalopathy, and diabetic ketoacidosis with pulmonary emboli.