Authors
Adam Tzur , Richard Nijholt, Vincent Sparagna and Alex Ritson.
Contributions by
Brad Dieter, Brandon Roberts, Zad Chow (Examine), and Michael Hull (Examine).
Reviewed by
Tyler Cartwright, Luis Villasenor, Brianna Stubbs , Stephan Guyenet , Marty Kendall and Kevin Hall.
Updated: December 10, 2018. 1Correction: the meta-analysis p-value was incorrect and was changed from 0.73 to 0.66

Table of Contents
Contents →
TogglePlain Language Summary
What You Will Learn
Understand why some people adhere to the ketogenic diet, and why some people fail.
What We Did and Why
We systematically searched the ketogenic literature for studies on adherence. We gathered data on carb intake, calorie intake, ketone levels and how many people dropped out from their diet.
Our goal is to determine if people can stick to the ketogenic diet in the long-term.
In What Scenarios Do People Adhere to the Ketogenic Diet?
In scientific studies, there are 3 common scenarios where people adhere well to the diet:
- Highly motivated people (e.g., elite athletes)
- Studies where food is provided for the participants.
- People in metabolic wards (e.g., hospitals) where the researchers closely monitor the participants. They can only eat the food they are given. This is forced adherence and does not reflect real-life adherence.
(Read more: At what ketone level are people in ketosis?).
How Long Do People Adhere to the Ketogenic Diet?
People who are left to their own devices generally do not adhere well to the diet. They typically start strong, but have a hard time adhering to the diet after 1-3 months. Carb intake increases and ketone levels drop. In particular, overweight people and diabetics struggle with adherence.
Sign up to the Newsletter for More Nutrition Articles
Is It Easier or Harder to Adhere to Keto vs. Other Diets?
We started by analyzing dropouts: how many people drop out of keto groups versus control diet groups?
The odds of dropping out were similar for all diets. About 24% of people drop out regardless of diet (Odds Ratio = 1.05, P = 0.66). (Jump to our meta-analysis of 30 keto studies.) 2We also wanted to compare self-reported food intake vs. the control diets, but this data is not accurate enough for statistical analysis. The reason is that people often underreport how much they eat
In conclusion we do not yet know whether ketogenic diets are easier or harder to adhere to compared to other diets. We do know that people struggle adhering to all diets over time. Some people do very well on some diets, yet we don’t know why.
The practical takeaway is that you should try different diets, and see which one suits you best. A high protein diet is always a good choice if your goals include fat loss, muscle gain, and/or better satiety.
Hopefully, future research will give us better tools to predict dietary adherence.
“ … no diet has yet been shown to be uniformly easier to stick with than another in the long run.” - Freedhoff and Hall, 2016
Video summary by PictureFit
Summary Infographic
Ketone Levels Drop Over Time
Adherence to a diet is crucial (Johnston et al., 2014; Freedhoff and Hall, 2016; Bray and Siri-Tarino, 2016). An objective way to measure ketogenic adherence is via ketone levels (Nymo et al., 2017; Cipryan et al., 2018). If ketones increase, we know that the person is restricting carbohydrates. If ketone levels fall, then participants are probably not adhering to the diet.
“Because different diets are variably tolerated by individuals, the ideal diet is the one that is best adhered to by individuals so that they can stay on the diet as long as possible.” - Johnston et al., 2014
“ … no diet has yet been shown to be uniformly easier to stick with than another in the long run.” - Freedhoff and Hall, 2016
“[Adherence] is likely a function of psychological issues (e.g., frequency of dietary counseling, coping with emotional eating, group support) rather than macronutrient composition, per se ... Being conscious of one’s behaviors, using social support, confronting problems directly, and using personally developed strategies may enhance long-term success ....” - Freedman et al., 2001
Studies with normal weight, diabetic, or obese participants
After doing a systematic search of the ketogenic literature, we identified ten suitable studies (read more about our inclusion and exclusion criteria). You can see data from these ten studies in the line chart below.
Eight groups never reached ketosis above 0.5 mmol/l. Four were in borderline ketosis (0.3–0.49 mmol/L BHB) by the end of the study. Only two groups were clearly in ketosis at one point during the study.
An interesting trend is that ketone levels decrease over time.
Figure: every line represents a group of people who were aiming to achieve ketosis in various studies. The solid blue lines were in ketosis at least once during the study. Dotted grey lines were in borderline ketosis. Dotted red lines were not in ketosis at the end of the studies. We chose studies that reported BHB levels in the blood. Every line represents one keto group from a study. Note that every study aimed for a carb level that would lead to ketosis (≤70 g per day).
These data suggest that people, primarily diabetics and the obese, struggle to adhere to the ketogenic diet.
Note: some of our reviewers have pointed out that decreasing BHB levels are expected, because they are an adaptation to the diet. They are not an indication of poor adherence. This observation is currently anecdotal and we could not find any published data on this adaptation.
Though, it should also be said that most keto scientists use ketone levels as an objective marker of adherence (see scientific quotes below). Sci-Fit is currently writing an article on this topic.
Study | Ketone level at the end of the fat loss period (BHB mmol/L) |
Krebs | Week 13
2.28 |
Gibas and Gibas | Week 10
0.43 |
Noakes | Week 8
0.2 |
Brehm 2003 | Week 26
0.15 |
Brehm 2005 | Week 17
0.19 |
Fleming | Week 6
0.29 |
Meckling | Week 10
0.38 |
Brinkworth (B) | Week 8
0.3 |
Cipryan | Week 4
0.4 |
Keogh | Week 2
0.41 |
“The data indicated imperfect compliance with the prescribed carbohydrate restriction, but the mean carbohydrate intake of the HPLC group was less than 40 g/day, an intake likely to maintain ketosis. The 13 week serum β-hydroxybutyrate concentrations support this: 2.28 ± 0.34 versus 1.0 ± 0.12, for the HPLC and LF groups, respectively (p = 0.002).” - Krebs 2010
“Compliance to dietary treatment was also confirmed by a change in plasma ketones between diets. VLCARB produced higher plasma levels of ketones (β hydroxybutyrate and acetoacetate) than the VLF or HUF diet treatments (P < 0.01), indicating adherence to a very low carbohydrate intake during the study (Figure 3). Despite continued apparent compliance to the diet plasma ketones declined with time.” - Noakes 2006
“Objective measures of compliance
Diet: Participants met every week for an individual 20 min consultation with a dietician, to review their food records. Urine acetoacetic acid concentration was also measured weekly, using Ketostix reagent strips. Participants who were not ketotic on more than one occasion were considered not compliant and were excluded from the analysis. Concentration of plasma β-hydroxybutyric acid (β-HB) in the fasting state was also measured with a Ketone Body Assay Kit (Mark134, Sigma-Aldrich, St Louis, MO, USA) at baseline, day 3, 5 and 10% WL and weeks 9 and 13.” - Nymo 2017
“Compared to T0 (127.0 ± 57.6 μ mol/L), total serum ketones increased 4-fold during the intervention (561.6 ± 317.9 and 489.5 ± 285.9 μ mol/L for T3 and T6, respectively; P < .001). (…) taken together with the food record, results are evidence that participants were adherent to the dietary recommendations and markedly reduced carbohydrate intake.” - Ballard 2013
“Although compliance with the diets was assessed primarily by dietary records, these data are supported by more objective measures. For example, the average 3-month weight loss in the low fat diet group (∼4 kg) is what would be expected for individuals decreasing their daily caloric consumption by about 400 kcal (28), approximately the restriction these women reported making. In addition, there was a significant correlation between reported changes in caloric intake and weight loss (r = 0.41; P < 0.001). Finally, the presence of measurable ketonemia and ketonuria in the very low carbohydrate group is consistent with severe carbohydrate restriction and was not seen in the low fat dieters.“ - Brehm 2003
“To maintain significant plasma ketosis requires very limited carbohydrate intake, and this measure has been the hallmark of adherence to these diets.” - Brehm 2005
“All subjects in the HFMP group demonstrated β - hydroxybutyrate concentrations above 0.20 mmol·L –1, indicating compliance with the HFMP diet.” - Fleming 2003
“Dietary analysis, compliance, and physical activity
On the basis of the results of the food records, the participants showed good compliance with the prescribed diets.” (...) During the initial stages of the study, plasma β-hydroxybutyrate concentrations increased significantly more in the LC diet group than in the LF diet group, and, although concentrations decreased over time in the LC diet group, they remained higher than those in the LF diet group throughout the intervention (Figure 2), which indicated adherence to a low-carbohydrate intake in the LC group.” - Brinkworth (B) 2009
“Evidence of the adaptation and compliance to VLCHF diet is usually evaluated by the βHB concentration in plasma.” - Cipryan 2018
“Dietary analysis and compliance
Adherence to the dietary interventions was established by concentrations of ketone bodies, the ratio of urinary urea to creatinine, and the dietary analysis data.” - Keogh 2008
Studies in elite athletes, metabolic wards, and with close follow-up
There are also several studies that had tight control over what the participants ate. We didn’t include these in the primary analysis because they do not tell us anything about real-life adherence (external validity).
In other words, we can’t tell if people can adhere to a diet if they are being fed pre-planned meals in a hospital while being monitored. Yet, these tightly controlled studies are very useful if we want to figure out mechanisms related to ketosis (internal validity). For example, if we wanted to test how the ketogenic diet affects energy expenditure, we would prefer to do it in a metabolic ward.
In these types of studies, people have much higher BHB levels:
- Johnstone et al., 2008 (metabolic ward): 1.52 mmol/L urinary BHB at week 4
- Hall et al., 2016 (metabolic ward): 0.76 mmol/L BHB over 4 weeks
- Goday et al., 2016 (feeding study): 1 mmol/L BHB at week 4
- Johnston et al., 2006 (feeding study): 0.72 mmol/L BHB at week 2
- Gomez-Arbelaez et al., 2018 (feeding study): 1 mmol/L BHB at C-2 (week 5.6)
- Burke et al., 2017 (elite athletes): 1.8 mmol/L BHB at week 3
- Wilson et al., 2017 (close follow-up): 0.85–1.4 mmol/L BHB at weeks 4–10
Here are ketone graphs from two of these studies:
Figure: Hall et al., 2016. Ketones remain elevated in a metabolic ward
Figure: Wilson et al., 2017. Ketones remain elevated for two months before refeeding carbohydrates. Note, this is not a metabolic ward study.
The metabolic ward studies show us that ketones can stay elevated when participants are “forced” to adhere to the diet. This is contrasted with the free living studies where participants’ BHB levels typically drop over time. The difference seems to be adherence.
If there is an adaptation that decreases BHB levels over time, it has not shown itself in these studies. At the very least, we must reflect on why BHB levels are consistently higher in metabolic wards, feeding studies, and studies with motivated athletes.
Keto Dieters Eat More Carbohydrates Over Time
We have created two line charts of the studies. The first shows us how many carbohydrates people ate over time. The second looks at short-term studies.
Note that the data is-self reported carbohydrate intake. People often tend to say they ate less than what they really did. This is called underreporting (Baranowsky et al., 1986; Schoeller, 1990; Drougas et al., 1992; Lichtman et al., 1992; Fries et al., 1995; Buhl et al., 1995; Schoeller, 1995; Heitmann and Lissner, 1996; Scagliusi et al., 2003; Okubo et al., 2008; Scagliusi et al., 2008; Bartholome et al., 2013; Capling et al., 2017). Hence, people most likely ate more carbs than what they reported.
Long-term studies
The data in the graph below suggest that very few keto groups are able to eat less than 50 grams of carbohydrates over time. In most studies, they revert to a higher non-ketogenic carbohydrate intake.
Figure: every line represents a group of people who were aiming to eat less than 70 grams of carbs. The solid blue lines report adhering to the diet. Dotted red lines were not adhering to their diet by the end of the study. This graph only includes studies that tracked how carb intake changed over time3Studies with 2+ time point measures after baseline.
Short-term studies
For studies shorter than 13 weeks, people report that they adhere to the diet.
Figure: every line represents a group of people who were aiming to eat less than 70 grams of carbs. The solid blue lines report adhering to the diet. Dotted red lines were not adhering to their diet. We only included short-term studies with single time point measures in this graph.
Are Keto Dieters More Likely to Drop Out?
We did an odds ratio meta-analysis of dropouts from 30 ketogenic and Atkins studies. Neither of our analyses found evidence that the odds favor keto or the control group. In other words, people seem to drop out from the diet groups at about the same rate.
“The odds ratio is the odds of an event occurring in one group divided by the odds of the same event in another group.” - Scott, 2008
In our case, the “event” is someone dropping out of a study. The “groups” are the ketogenic diet groups and control groups. Hence, the odds ratio compares the odds of dropping out from ketogenic diet groups versus control diets. For example, an odds ratio of 2 would mean that people would be more likely to drop out from the keto group. An odds ratio of 1 would mean that the odds of dropping out are identical for both groups.
In our meta-analysis, the odds ratio is 1.05 and the p-value is 0.73. Hence, the effect is tiny and not statistically significant.
Note: the odds ratio should not be interpreted as a risk ratio.
Odds ratio meta-analysis
If we pool the numbers, the included ketogenic diet studies had a total of 1307 participants, of which 319 (24.4%) dropped out from the studies. The control diets had a total of 1294 participants, of which 311 (24%) dropped out.
Meta-analysis OR = 1.05, P = 0.66.
Forest plot: Every line/row represents dropouts from one study. The red dots represent the odds ratio which could favor either the keto or control groups. If the red dot is to the right of the vertical line, then the participants in the keto group had higher odds of dropping out. And vice versa. The horizontal lines represent the 95% confidence interval. Wide lines represent greater uncertainty. Note that more participants increase certainty while few participants decrease certainty.
Odds ratio sensitivity analysis
If we remove the studies with zero dropouts, there is no change to the outcome:
At what Ketone Level Are People in Ketosis?
One of our biggest challenges is to determine when people are in ketosis. It is commonly said to be 0.5 mmol/L. We opt for the following system:
- <0.3 mmol/L BHB: not in ketosis
- 0.3–0.49 mmol/L BHB: borderline ketosis
- ≥0.5 mmol/L BHB: in ketosis
This system is not ideal, because we don’t know precisely when ketosis occurs. There is no data objectively confirming a ketosis cutoff point. However, 0.5 mmol/L of serum BHB is commonly used as a cutoff by keto researchers (McKenzie et al., 2017; Harvey et al., 2018). However, some researchers use 0.3 mmol/L (Wilson et al., 2017), while others use 1 mmol/l (Burstal et al., 2018).
“Ketogenic diets are now commonly applied, for a range of desired outcomes, and with differing definitions of what constitutes a ketogenic diet. Both low-energy diets and VLCKDs with fewer than 50 g of carbohydrate per day typically result in BOHB levels of ≥0.5 mmol L−1 (Gibson et al., 2015). This threshold has been used as a cut-off point for entry into ketosis by Guerci and colleagues (Guerci et al., 2003), and is commonly applied as a marker for entry into NK in the nutrition field … ” - Harvey et al., 2018
“ … nutritional ketosis, defined as a dietary regimen resulting in serum beta-hydroxybutyrate (BOHB) levels between 0.5 and 3.0 mmol·L” - McKenzie et al., 2017
"If deviation from the diet occurred outside of the parameters set forth by the dietitian, or the subjects blood ketones did not register a minimum of 0.3 mmols for ketones then the dietician met with the subject more closely until these parameters fell within the scope of the study." - Wilson et al., 2017
"One hundred non-diabetic adults presenting for elective or emergency surgery were assessed for the presence of hyperketonaemia (beta-hydroxybutyrate levels more than 1 mmol/l)" - Burstal et al., 2018
How We Collected Studies
We:
- Searched PubMed and Google Scholar.
- Extracted studies from other studies/reviews.
- Used studies from our previous articles and our study collection.
- Filtered results through inclusion and exclusion criteria.
- Double checked the data.
PubMed Search terms
(“ketogenic” OR “ketogenic diet” OR “atkin*” OR “very low carb diet” OR “very low carb ketogenic diet” OR “low carb ketogenic diet” OR “ketogenic low-carbohydrate diet” OR “low carbohydrate ketogenic” OR “very low carbohydrate” OR “VLCKD” OR “VLCD” OR “KLCD”) AND (“adhere*” OR “compliance” OR “ketosis” OR “acetoacetate” OR “acetone” OR “urinary ketone*” OR “ketone*” OR “hydroxybutyrate” OR “BHB” OR “ketonemia” OR “ketonaemia”) NOT (“cancer” OR “epilep*” OR “epilepsy” OR “seizure*” OR “rat” OR “rats” OR “mice” OR “cattle” OR “cow” OR “cows” OR “sheep” OR “traumatic brain injury” or “TBI”)
Inclusion criteria
- Ketogenic diet, Atkins diet, or very low-carb diet (less than 70 g carbohydrates): “A daily carbohydrate intake that does not exceed 75 grams on average is needed in order to stay in ketosis; a 50 gram maximum is preferred” (Cavaleri and Bashar, 2018)
- Weight loss study
- The study is in English
- In humans
- The study has data on either carbohydrates, calories, blood ketone levels, or dropouts (all of which are factors which can indicate adherence to the diet)
Exclusion criteria
- Not a ketogenic diet: Studies where participants did not aim to eat less than 70 g of carbohydrates per day.
- Increasing carb intakes: some studies aim to increase the participants’ carb intake on a weekly basis. We only include these studies if they have a period of carb restriction where they do not increase carb intake. We exclude any period where carbohydrates are increased or reintroduced. This is key because eating more carbohydrates over time would indicate a decrease in adherence. However, if the carb increase is planned, it does not indicate a decrease in adherence (i.e., several of the Atkins studies).
- Weight maintenance study: we wanted to see how people adhere to weight loss diets, specifically. Several studies were split into two phases, one with weight loss followed by weight maintenance. We included data from the weight loss phase while excluding data from the weight maintenance phase.
- Cancer patients are a unique population and thus not included. Most keto cancer studies also have small samples, and patients frequently die during the intervention.
- Epilepsy studies because they typically use the classical ketogenic diet, which is very different compared to normal ketogenic diets. Classical keto is so restrictive that it can adversely affect adherence, and thus would be unfair to include in the analysis.
- Very small samples (less than 10 participants). In such studies, individuals have a much stronger influence on the mean. An outlier or two could have powerful effects.
- Lacking data or poor data reporting (i.e., data from only a few participants, or insufficient reporting, numbers not shown, etc.)
- Feeding studies and metabolic wards: several studies control food intake by giving subjects meals which they are allowed to eat. The subjects are either followed up frequently, or they are monitored in training camps or metabolic wards. This study methodology has high internal validity (you know what they are eating), but low ecological validity. We are primarily interested in ecological validity in this review (how people adhere to the diet when others are not watching over their shoulders, or micromanaging their food intake).
- Duration: Very short studies (shorter than two weeks). We generally want to see how adherence changes over time. Two weeks is merely an introduction to the diet.
- Supplement only being given to the keto group (except for general multivitamins or electrolytes).
- Participants who were already participating in a diet or lifestyle program before the study (i.e., Grieb et al., 2008)
- Control groups with no dietary intervention
Notes:
- Studies with type 2 diabetic participants were not excluded because they typically use normal ketogenic diets.
- We excluded feeding studies/metabolic ward studies from the meta-analysis and graphs, but discussed them in “Studies in elite athletes, metabolic wards, and with close follow-up”