Approval of the study was obtained from the regional Medical Ethics Committee and the trial was registered at ClinicalTrials.gov (Identifier: NCT00435240). The study was a randomized, cross-over, dietary intervention study in 13 patients with type 2 diabetes without insulin treatment, 3 women and 10 men, recruited from three primary health care units in the Lund area in Sweden. We included adult patients with type 2 diabetes and a C-peptide value above zero, unaltered medical diabetes treatment and stable weight since three months before start of study, HbA1c above 5.5% by Mono-S standard, creatinine below130 μmol/L, liver enzymes below four times their respective upper reference value, no chronic oral or injection steroid treatment and no acute coronary event or change in medication of beta blockers or thyroxin since six months before start of study. Exclusion criteria during ongoing study were change in beta blocker or thyroxin medication, chronic oral or injection steroid treatment, warfarin treatment, creatinine above 130 μmol/L or liver enzymes above four times their respective upper reference value, acute coronary event, and physical or psychological illness or changes in personal circumstances which would make further study participation impossible.

Recruitment for the study during routine clinical work was performed by TJ, UCB, GP, AH and MS. In addition, a letter containing written study information was sent by TJ to subjects at two of the health stations who from journal data seemed to match the inclusion criteria. All recruited subjects were given oral and written study information prior to signing a consent form to participate in the study and were then further assessed for eligibility.

All eligible subjects were informed of the intention to compare two healthy diets in the treatment of type 2 diabetes and that it was unknown if any of them would be superior to the other. At study start all eligible subjects were randomized to start with either a Diabetes diet in accordance with current guidelines 17 or a Paleolithic diet. Randomization was performed by UCB, GP and AH by opening opaque, sealed envelopes (prepared by TJ) containing a note of the initial diet with equal proportions of envelopes for both diets. After randomization, there was no blinding of dietary assignment to study participants, nor to those administering the interventions or assessing the outcomes. Immediately after randomization, all subjects received oral and written information individually (by UCB, GP or AH) in the morning about their respective initial diet. After three months all subjects switched diets and received new oral and written information individually (by UCB, GP or AH) about the diet of the following three months. Written information with dietary advice and food recipes were similarly formulated for both diets. For increased conformity, the dietary advice and data collection procedure were discussed by all authors except YG at several meetings prior to start of study. Advice about regular physical activity was given equally to all subjects.

The information on the Diabetes diet stated that it should aim at evenly distributed meals with increased intake of vegetables, root vegetables, dietary fiber, whole-grain bread and other whole-grain cereal products, fruits and berries, and decreased intake of total fat with more unsaturated fat. The majority of dietary energy should come from carbohydrates from foods naturally rich in carbohydrate and dietary fiber. The concepts of glycemic index and varied meals through meal planning by the Plate Model were explained 18. Salt intake was recommended to be kept below 6 g per day.

The information on the Paleolithic diet stated that it should be based on lean meat, fish, fruit, leafy and cruciferous vegetables, root vegetables, eggs and nuts, while excluding dairy products, cereal grains, beans, refined fats, sugar, candy, soft drinks, beer and extra addition of salt. The following items were recommended in limited amounts for the Paleolithic diet: eggs (≤2 per day), nuts (preferentially walnuts), dried fruit, potatoes (≤1 medium-sized per day), rapeseed or olive oil (≤1 tablespoon per day), wine (≤1 glass per day). The intake of other foods was not restricted and no advice was given with regard to proportions of food categories (e.g. animal versus plant foods). The evolutionary rationale for a Paleolithic diet and potential benefits were explained 19.

An oral glucose tolerance test (OGTT) was performed in the morning after obtaining venous blood samples and measurements of blood pressure, weight and waist circumference in the primary care unit (by UCB, GP or AH) at study start, after 3 months (when switching to a new diet) and at the end of the study (after 6 months). 75 g glucose was ingested. Blood samples for plasma glucose and insulin during OGTT were obtained at 0, 15, 30, 60, 90 and 120 minutes. Changes in the area under the curve (AUC) between 0 and 120 min during OGTT for plasma glucose (AUC Glucose_0–120) and plasma insulin (AUC Insulin_0–120) were predefined primary endpoints, along with changes in body weight, waist circumference, serum lipids, CRP, blood pressure and glycated haemoglobin A1c (HbA1c) by Mono-S standard. The base of the AUC was set at 0 mmol/L for glucose and 0 pmol/L for insulin. The stimulated secretion was represented by the areas under the glucose and insulin curves using levels at 0 min as the base of the area. The Homeostatic model assessment (HOMA) was used for assessing beta-cell function (%B) and insulin sensitivity (%S), as percentages of a normal reference population, and insulin resistance (IR, the reciprocal of %S (100/%S)) 20. Values for %B, %S, and IR were derived from fasting plasma glucose and insulin using the HOMA2 computer model v2.2 20. Insulin sensitivity index (ISI_0,120) was calculated from fasting (0 min) and 120 min (post-OGTT) insulin and glucose concentrations 21. A 4-day weighed food record on four consecutive days, including one weekend day, with weighing of each food item on a digital weighing scale (that could be set to zero), was completed by the participants, starting 6 weeks after initiating each diet. Nutrient compositions were calculated by YG using data from The Swedish Food Database of the National Food Administration in Sweden. GL and GI for the two diets were calculated. Underlying concept of dietary GL and dietary GI is food GI, introduced by Jenkins et al 22, reflecting the postprandial glucose response after a specific food rich in carbohydrate, expressing the quality of the carbohydrates. Wolever and Jenkins also suggested the possibility of ranking diets based on dietary GI calculated from the proportional GI contribution of the included foods containing carbohydrate 23. To include also the quantity of carbohydrates consumed GL was introduced by Salmerón et al expressing the glycemic effect of the diet 24. While dietary GI is expressing the quality of the carbohydrates consumed GL represent both the quantity and the quality of the carbohydrates consumed. Thus, dietary GL in this study was calculated as the result from multiplying available carbohydrate (g) for the food reported by the subjects during the 4-day weighed food record with the specific food's GI divided by 100. Available carbohydrate was based on total carbohydrate minus dietary fibre. Food's GI values (glucose as reference) were taken from the compilation by Foster-Powel et al 25. Dietary GI was calculated as 100 multiplied with dietary GL divided by the amount of available carbohydrate (g) in the diet.

A pre-study power calculation showed that 15 subjects would be required to detect, with 80% power and at a significance level of 5%, a 15% reduction in AUC Glucose_0–120. Two-way paired t-test was used to analyze within-subject changes in absolute values, while two-way unpaired t-test was used to analyze between-subject changes in absolute values. All outcome variables showed reasonable normal distribution in normal plots. Within-subject changes in outcome variables after first and second diet and within-subject changes in reported dietary intake during first and second diet were used to check for period effects 26. Mean values of outcome variables and reported dietary intakes for the group starting with Paleolithic diet was compared with the group starting with Diabetes diet in order to check for carry-over effects 26. Exploratory analyses were performed on outcome variables with significant effects from the Paleolithic diet as compared to the Diabetes diet. Exploratory analyses consisted of bivariate correlations between within-subject differences (Δ) in outcome and dietary variables. Significantly correlating variables were entered into a stepwise forward linear regression analyses.

The study started in January 2005 and the last participant was followed up in September 2007 after which the study was stopped. Out of 26 subjects assessed for eligibility, nine were not eligible since they did not meet the inclusion criteria or refused to participate. Out of the remaining 17 eligible subjects, who were all randomized and started on the study, four subjects were excluded for the following reasons: one starting with Paleolithic diet was wrongly included with ongoing warfarin treatment, one starting with Paleolithic diet was unwilling to continue due to abdominal pains and bloating, one starting with Diabetes diet was excluded after developing leukemia, and one starting with Diabetes diet was excluded after developing heart failure. All reported analyses are "per protocol" analyses on the 13 participants who completed the trial.

Study participants were on average treated with just above four drugs per day, which usually included metformin alone (3 subjects) or metformin in combination with a sulfonylurea (3 subjects) or a thiazolidinedione (3 subjects)(Table 1, 2). Medication usually also included a lipid lowering drug (8 of 13 study participants and always statin treatment) and more than one anti-hypertensive drug per day (Table 1, 2). All medication remained unchanged during the whole study with the following exceptions:

One participant stopped taking sulfonylurea (glibencklamide 87.5 mg daily) the day after starting the study with the Paleolithic diet, and was thus on a low dose sulfonylurea at baseline, but without sulfonylurea during both the Paleolithic and Diabetes diet. Exclusion of this participant would not negate any significant effects from the Paleolithic diet compared to the Diabetes diet, but would negate the effect from the Paleolithic diet compared to baseline on systolic blood pressure, and the effect from the Diabetes diet compared to baseline on BMI. Due to concerns for rising blood sugar levels one participant switched hypertensive treatment from a thiazide diuretic to a beta blocker for seven weeks during the Paleolithic diet. Exclusion of this participant would not negate the significant effect on HbA1c, but would negate the effect on diastolic blood pressure. Due to concerns about muscle ache one participant was without lipid-lowering drug treatment for four weeks during the Paleolithic diet. Exclusion of this participant would not negate the significant effects on TG and HDL, but would instead cause also total cholesterol to be significantly lower following the Paleolithic diet compared to the Diabetes diet (p = 0.03). One participant was put on finasteride (5 mg daily, a drug versus benign prostate hyperplasia) during the Paleolithic diet and continued this medication during the following Diabetes diet.

The group starting with the Paleolithic diet differed at baseline only with regard to fasting plasma glucose and AUC glucose being lower and HOMA2 %B being higher compared to the group starting with a Diabetes diet (Table 2). There was no difference between starting groups before or at the end of the study in inclusion/exclusion variables.

Compared to the Diabetes diet, the Paleolithic diet resulted in lower mean values of HbA1c, TG, DBP, weight, BMI and waist circumference, while mean values for HDL were higher (Table 3, Figure 1). The larger decrease of fasting plasma glucose following the Paleolithic diet nearly reached significance, and SBP also tended to decrease more following the Paleolithic diet. Compared to baseline, the Paleolithic diet lowered mean values of HbA1c, TG, SBP, weight, BMI, waist circumference, fasting plasma glucose, fasting plasma insulin, AUC glucose, ISI_0,120, HOMA2 %S and HOMA2 %IR (Table 3). Compared to baseline, the Diabetes diet lowered mean values of BMI, waist circumference and HOMA2 %S (Table 3). Period effects were seen in AUC insulin_0–120, AUC insulin_0–120 stimulated secretion and HOMA2 %B (Table 4). Carry over effects were seen in HbA1c (Table 2, Figure 1).

There were no period or carry-over effects in reported dietary intakes (data not shown). Reported daily food intake differed between diets mainly in that the Paleolithic diet was markedly lower in cereals and dairy products, and lower in potatoes, beans and bakery, and much higher in fruits, vegetables, meat and eggs (Table 5). Further, the Paleolithic diet was somewhat lower in total energy, energy density, carbohydrate, fiber, saturated fatty acids and calcium, and higher in monosaccharides, dietary cholesterol, some vitamins (vitamin B6, vitamin C, niacin) and minerals (potassium, selenium) (Table 5). During the Paleolithic diet, there was a lower relative intake (as a percentage of total macronutrient energy intake [E%]) of carbohydrate and a higher relative intake of protein and fat (Table 5). Both dietary GL and dietary GI were determined to be lower for the Paleolithic diet than for the Diabetic diet (Table 5).

In exploratory analyses of primary endpoints, within-subject differences (Δ) in HbA1c (ΔHbA1c) correlated with Δwaist circumference, which correlated with Δweight, which correlated with ΔCRP (Table 6). Furthermore, ΔHDL correlated with Δcholesterol and ΔDBP with ΔHOMA2 IR (Table 6). In exploratory analyses of estimated intake of nutrients, ΔHbA1c correlated with Δpotassium, ΔHDL with Δfatty acid C20:5 n-3, ΔTG with Δthiamin, ΔDBP with Δdietary cholesterol, Δweight with Δenergy density per meal, and Δwaist circumference with Δbakery, Δenergy density per meal, Δsauce and Δvitamin E (Table 6).

The advice for patients with type 2 diabetes to follow a Paleolithic diet resulted in lower HbA1c, TG, DBP, weight and waist circumference, and higher HDL, as compared to a Diabetes diet according to current guidelines. In addition, fasting glucose and SBP tended to decrease more after the Paleolithic diet. Changes in glucose tolerance were not significantly different between diets. The two diets differed mainly in that the Paleolithic diet was lower in cereals and dairy products, and higher in fruits, vegetables, meat and eggs. Further, the Paleolithic diet was lower in total energy, energy density, carbohydrate, dietary GL, saturated fatty acids and calcium, and higher in unsaturated fatty acids, dietary cholesterol and several vitamins. Dietary GI was lower in the Paleolithic diet (GI = 50) than in the Diabetic diet (GI = 55).

No advice was given to restrict food intake. Therefore, the lower reported energy intake during the Paleolithic diet despite no difference in weight of reported food intake agrees with the notion that such a diet is satiating and facilitates a reduced caloric intake 427. Accordingly, energy density was lower in the Paleolithic diet and also correlated with alterations of both weight and waist circumference. The higher amount of fruit and vegetables during the Paleolithic period may have promoted weight loss due to its high content of water, which is thought to be satiating 28. Interestingly, the Paleolithic diet appeared to be satiating despite a lower content of fiber in this study. The slightly higher relative protein intake, as percentage of total calorie intake, may also have added to a satiating effect 2930. Alternative explanations on satiation, such as dietary effects on leptin resistance, could also be considered 31.

A reduced energy intake would evidently be a major explanation for the beneficial effects of the Paleolithic diet on weight and waist circumference. Meta-analyses and large trials with various lifestyle interventions indicate that reduced caloric intake is more important for long-term weight loss than other known dietary factors, including macronutrient composition 323334353637383940. In studies shorter than 6 months, such as this one, differences in GI and/or GL may also have played a role for weight change. A Cochrane review found that overweight or obese people lost slightly more weight during 5–12 weeks of low GI diets 41, and short-term carbohydrate restriction possibly results in greater weight loss than low-fat diets 29. However, dietary GI and dietary GL did not correlate with alterations of weight, waist circumference or metabolic variables in our study. It should also be noted that, in the present study, reported mean absolute carbohydrate intake in the Paleolithic diet (g per day) was only slightly below the 130 g per day recommended by the American Diabetes Association, and clearly above 50 g per day, which has been proposed as the level below which a diet should be termed a low carbohydrate diet 42.

Paleolithic diet improved the glycemic control in the subjects, as evident by the reduction of HbA1c levels by -0.4 percentage points lower as compared to the diabetes diet. Since both glucose and insulin levels declined during Paleolithic diet, a main mechanism behind the improved glycemic control is probably improved insulin sensitivity, which may have allowed the released insulin to work more efficiently. The difference in reduction in HbA1c of 0.4% units between the Paleolithic and Diabetes diet is close to the average 0.5% units in a recent Cochrane review of diets with a low glycemic index or glycemic load 8. However, the differences in GI between diets in that meta-analysis were considerably larger than in our trial. Glucose tolerance, which also determines the glucose response and thereby HbA1c, did not improve more during the Paleolithic diet. This result agrees with findings from Frassetto et al 10, but differs from our previous parallel-group trial which compared a Paleolithic diet with a Mediterranean-like diet in subjects with diabetes or impaired glucose tolerance 4. Glucose tolerance has not been shown to improve after reduced carbohydrate intake in earlier dietary studies 43444546.

The much higher fruit intake of the Paleolithic diet probably resulted in a slightly higher intake of fructose which may have aided in the reduction of HbA1c. Fructose in exchange for starch, sucrose or glucose decreases postprandial glycemia 47, while the effect on glucose tolerance and insulin sensitivity is more uncertain 48. The effect of fruit on TG and other risk factors is expected to have been neutral in this study 4849. Total intake of monosaccharides was 46 g per day, including approximately equal amounts of glucose and fructose, which was well below the suggested safety limit of 50 g fructose per day 48. Our study lends further support to the notion that fruit intake should not be restricted in patients with type 2 diabetes.

The lower DBP after the Paleolithic diet compared to the Diabetes diet did not correlate with sodium intake, which did not differ significantly and was rather low in both diets (2.5 g and 3.0 g per day respectively for the Paleolithic and Diabetes diet).

The reduction of TG after the Paleolithic diet was possibly due to greater loss of abdominal fat 50 or lower GL compared to the Diabetes diet 6, although no correlation of TG with waist loss or GL was seen in exploratory analyses. A small additional effect on TG may be attributable to a trend for higher content of long-chain omega-3 fatty acids in the Paleolithic diet, while the higher dietary cholesterol content of the Paleolithic diet is probably of minor significance 51.

All improvements in markers of the metabolic syndrome on the Paleolithic diet are in line with findings from epidemiological studies in non-Western populations 141516. Improvements in HbA1c 4, weight 41252, BMI 52, waist circumference 452, DBP 10, and TG 10, compared to baseline, on a Paleolithic diet have been observed before in intervention studies, while improvements in HDL have not. Similar differences in weight and DBP on a Paleolithic diet, compared to a cereal based diet, have been observed before in an intervention study on domestic pigs 11. A lower reported energy intake and energy density of food despite food intake ad libitum agrees with our earlier findings that a Paleolithic diet facilitates a reduction of caloric intake 41127.

Also, lower intake of cereals, dairy products, carbohydrates, dietary GL and saturated fat, and higher intake of fruit and potassium have been observed before 410. Lower intake of potatoes, bakery, fiber, phosphorous and calcium, and higher intake in vegetables, meat, eggs, monosaccharides, dietary cholesterol, vitamin B6, vitamin C, niacin and selenium have not been observed before in intervention studies with a Paleolithic diet. Dietary GI for a Paleolithic diet has not been determined before.

A limitation of this study, as with most other dietary trials, is the lack of blinding after randomization. To minimize this problem, all study participants were informed of the intention to compare two healthy diets in the treatment of type 2 diabetes and that it was unknown if any of them would be superior to the other. Also, written information with dietary advice and food recipes were similarly formulated for both diets. Furthermore, for increased conformity, the dietary advice and data collection procedure were discussed by all those administering the interventions at several meetings prior to start of study.

Another limitation of this study is its small size which did not reach the number of participants needed as calculated in the pre-study power calculation. The decision to end the study was taken when recruitment for the study had not yielded new participants for more than six months. The population of patients with type 2 diabetes is much larger and therapy continues for substantially longer than in this study. Moreover, many patients with type 2 diabetes have illnesses and treatments that excluded them from the current study. Consequently, the results of this study do not address the occurrence of rare adverse events, nor can they be extrapolated to all patients seen in general clinical practice.

The carry-over effects on HbA1c were not due to carry-over or period effects in reported food intake. Instead, they could be true carry-over effects of the first diet. This is particularly likely for HbA1c, since HbA1c represents a weighted average of the blood glucose concentration over the previous two to three months (53. If results from the second period were discarded (owing to carryover 54), the reduction of HbA1c from the Paleolithic diet compared to the Diabetes diet was still significant (p = 0.01) and even larger (-1.3% units) than when results from the second period were included. However, this approach could lead to biased answers to our hypothesis and results from both periods are therefore used in this study 54.

The lack of carry-over or period effects in reported food intake indicates fairly good adherence to intervention diets. Reported food intake in this study seemed reasonable both in distribution and quantity, as subjectively assessed by a nutrition engineer skilled in analyzing reported food intake (YG). Furthermore, the reported lower energy intake of 1.3 MJ per day on a Paleolithic diet equals about 3.2 kg fat during three months, which almost exactly accounts for the observed 3.3 kg difference in weight loss between diets. This indicates both good reporting by the participants and good adherence to reported food intake during the study.

The favourable results in this study are in line with previous findings and increase the generalizability of the Paleolithic diet by testing it in both men and women in a primary care setting. A limitation of the study is the small size of the study population. This prevents the conclusions from resulting in nutritional recommendations for patients with type 2 diabetes. A long-term study in a larger population is therefore required. In parallel, further research into possible mechanisms for the beneficial effects of a Paleolithic diet should be done.

Total protein intake in g per day did not differ between the diets, but, as a result of the difference in total energy intake, the energy percentage (E%) from dietary protein on the Paleolithic diet (24 E%) slightly exceeded US and European recommendations for people with diabetes (<20 E%) 1755. The debatable disadvantage for long-term kidney function 5657 should be weighed against the benefits of attenuated postprandial glycemia when protein replaces starch or glucose 58.

Calcium intake did not meet recommendations for any of the diets, and it was particularly low in the Paleolithic diet. Recent calcium balance studies indicate that human calcium requirements are lower than previously thought 59, and meta-analyses of randomized controlled trials suggest that the effect of calcium supplementation for bone strength is limited 6061. It has been suggested that absorption and excretion of calcium are more important than calcium intake for whole-body calcium balance 62. In this context, the lower content of calcium-binding phytate and the lower dietary acid load from a Paleolithic diet may hypothetically compensate for the low amount of calcium 63. Supporting this view are the findings of Frassetto et al, where calcium intake remained unchanged and urine calcium decreased after a Paleolithic diet compared to baseline 10.

As has been discussed, there may be a challenge to implement and adopt the Paleolithic diet on a worldwide scale in subjects with type 2 diabetes. However, this aspect is beyond the objective of this paper and requires more research.