The heterogeneity of reversion to normoglycemia according to prediabetes type is not explained by lifestyle factors

Healthy lifestyle interventions and drug therapies are proven to have a positive preventative influence on normal glucose regulation in prediabetes. However, little is known on the specific role that these factors play on reversion to normal glycemia according to type of prediabetes. We used data from the Observational prospective cohort study, The Cohort study in Primary Health Care on the Evolution of Patients with Prediabetes from 2012 to 2015. A total of 1184 individuals aged 30–74 years old were included and classified based on the ADA in three mutually exclusive groups using either fasting plasma glucose (FPG) levels (from 100 to 125 mg/dl, FPG group), HbA1c (5.7–6.4%, HbA1c group) or both impaired parameters. Information on lifestyle factors and biochemical parameters were collected at baseline. Reversion to normal glucose regulation was calculated at third year of follow-up. Relationship of lifestyle factor and type of prediabetes with reversion were estimated using odds ratios (ORs) with 95% confidence intervals (95% CIs) adjusting by different groups of confounders. Proportion of reversion rates were 31% for FPG group, 31% for HbA1c group and 7.9% for both altered parameters group, respectively. Optimal life style factors such as BMI < 25 kg/m2[OR (95% CI): 1.90 (1.20–3.01)], high adherence to Mediterranean diet 1.78 (1.21–2.63) and absence of abdominal obesity 1.70 (1.19–2.43) were the strongest predictors for reversion to normal glucose. However, those did not modify the ORs of reversion to normal glucose. Taking as reference those with both impaired parameters, subjects with FPG impairment (FPG group) had an OR of 4.87 (3.10–7.65) and 3.72 (2.39–5.78) for HbA1c group. These estimates remained almost the same after further adjustment for biochemical parameters and lifestyle factors (4.55(2.84–7.28) and 3.09 (1.92–4.97), respectively). Optimal lifestyle factors showed to be a positive predictor for reversion to normal glucose regulation however, the differences of reversion risk according type of prediabetes are not explained by lifestyle factors.

The global prevalence of diabetes among adults aged 18 and beyond, has risen from 4.7% in 1980 to 8.5% in 2014, representing almost 422 million people by 2014 1 . In Spain the prevalence of diabetes in 2010 we 13.6% 2 . Moreover, the mortality of diabetes has scaled up to be positioned in the eight causes of dead rank worldwide 3 . Risk factors for developing diabetes type 2 have been well established and characterized [4][5][6] , which has allowed to implement interventive measures to reduce the burden of this disease across different healthcare systems. As an example, adherence to a healthy lifestyle [including healthy diet, smoking cessation, increase in physical activity, reduction in alcohol consumption, and reduction in body mass index (BMI)] at age 50-75, has been associated with six to ten years increase in life years and a significant improvement of quality of life. These implementations also resulted in a reduction in prevalence of major chronic diseases (including cancer, cardiovascular diseases and diabetes) compare to individuals with no vast improvement in lifestyle 7,8 . Prediabetes status has been associated with a higher risk of developing diabetes type 2. This stage is characterized by either an impaired fasting plasma glucose (IFG) or impaired glucose tolerance (IGT), or elevated glycated hemoglobin A1c (HbA1c) depending on the established diagnostic criteria 9 . Prior research on this topic highlights how optimal lifestyle factors and drug therapies (majority oral antidiabetic medications) implementations are effective predictive positive factors to conversion to normal glucose regulation in subjects with prediabetes 5,10,11 . However, those interventions do not seem to play the same role depending on the type of prediabetes. There is sparsity of data on reversion to normal glucose according to IFG criteria, an interventional study of physical activity found higher reversion rates among individuals with IFG compared with those with IGT or both altered parameters 12 . In addition, a recent study observed how individuals with HbA1c levels in range of prediabetes were less likely to revert to normal glycaemia after physical activity recommendations 12 . Although these interventions could be effective, some authors have evaluated the role of phenotypic and genetic variables to determine the effectiveness of a lifestyle intervention to prevent diabetes 13 .
This apparent heterogeneous association, between optimal lifestyles and reversion to normal glycaemia regulation according prediabetes type, could be the cornerstone for individualized prevention strategies in subjects with prediabetes. However, there is little evidence on how healthy lifestyle factors can explain the variation in the proportion of patients reverting to normal glucose regulation according to type of prediabetes. In order to develop a better understanding of the following matter, the current study aimed to study the heterogeneity in the reversal rate to normal glucose regulation among the three groups and to study the key role of lifestyle factors in predicting the reversal using a prospective cohort of individuals with prediabetes followed up by primary care physicians in Spain.

Material and methods
Study design. The Cohort study in Primary Health Care on the Evolution of Patients with Prediabetes (PREDAPS Study) is a prospective study encompassing two cohorts of patients: those with prediabetes status and those with normal glycemia (i.e. non-prediabetes neither diabetes) with the attempt to study the progression, prognosis and behavior of prediabetes. Details on information and design published previously by the same authors 14 . Briefly, this prospective study conducted at the primary care setting, started in 2012. To be member of the prediabetes cohort individuals aged 30-74 years old were included when meeting the following prediabetes criteria based on the definition established by American Diabetes Association 15 using only FPG and HbA1c parameters as there were not data available to identify impaired glucose tolerance (IGT). First group, namely (1) group 1 (isolated IFG), included all individuals with FPG 100-125 mg/dl, (2) group 2 (isolated A1c group), included all individuals with HbA1c 39-47 mmol/mol (5.7%-6.4%) and (3) group 3 (both altered parameters group), included all individuals with HbA1c 39-47 mmol/mol (5.7-6.4%) and FPG 100-125 mg/dl. Participants aged 30-74 years old with HbA1c < 39 mmol/mol (< 5.7%) and FPG < 100 mg/dl were assigned to the normoglycemia cohort. Individuals with the following criteria were excluded (to be members) from the study cohort: diabetes, terminal disease, pregnancy, surgery, or hospital admission in the previous 3 months at study entry, or any hematologic disease, which could alter HbA1c values. A total of 2022 individuals gave their written informed consent for participation: 1184 subjects with prediabetes and 838 without impaired glucose metabolism.
The present study analyzed the relationship between lifestyle and other variables (i.e. including lifestyle factors and metabolic conditions) measured at baseline and the situation of reversion to normal glycaemia in the third year of follow-up among the cohort of subjects with prediabetes. Thus, out of 1184 subjects with prediabetes, a total of 948 (80.1%) attended their third follow up visit and were retained to be members of the final cohort, therefore remaining patients were excluded. The main causes of exclusion were the administrative assignment of the participants to another general practitioner, change of their place of residence and refusal to continue in the study. Reversion to normal glucose regulation, was considered if FPG and HbA1c values were FPG < 100 mg /dl and HbA1c < 39 mmol/mol (< 5.7%), respectively, at third year of follow up (Fig. 1).

Data collected.
The study protocol has been published 14 . Basically, the questionnaire included more than 200 items and all physicians were trained to carry out the interview, collect information and complete the questionnaire. During the medical visit blood and urine analyses were requested to determine FPG, HbA1c, lipid profile, transaminases, blood count, iron levels and renal function. Of note, all variables were treated as categorical data. Similarly, visit a physical examination was performed, which included anthropometry and determina- www.nature.com/scientificreports/ tion of blood pressure. Three readings of height, weight, waist circumference, systolic and diastolic blood pressure were taken. The mean of the three readings was used for the analysis. The questionnaire of health surveys carried out at national level was used to measure smoking and physical activity. Smoking habit was classified into three mutually exclusive categories: current smoker, former smoker, and non-smoker. Individuals were asked to state which of the following alternatives best reflected their alcohol consumption frequency: never drinker, former drinker, occasional drinker, or daily drinker. For the present analysis individuals were classified into non-drinkers, occasional drinkers and daily drinkers. Physical activity data were collected based on the frequency-number of times in the last two weeks-and amount-mean time in minutes for each session as well as different types of physical activity, and, on the basis of the data collected, the minutes per week of physical activity performed by each participant were estimated. Subjects were classified into two categories to their compliance with the World Health Organization (WHO) physical activity recommendations-accumulate at least 150 min per week of moderate aerobic activity or 75 min per week of vigorous aerobic activity-, or an equivalent combination of moderate and vigorous physical activity 16 Diet information was obtained by a simplified 20-item food frequency questionnaire, based on validated instrument which included standard portions of foods 17,18 . And several response categories: daily consumption, ≥ 3 times / week, 1-2 times / week, < 1 time / week, never or almost never. The foods were: dairy products, meats, cold meats and sausages, fish, eggs, legumes, potatoes, vegetables, fruit, rice and pasta, bread, cakes or sweets, olive oil, other oils, animal fat, fried foods, ready meals, preserved food, nuts, bag snacks. Adherence to the Mediterranean diet was estimated through an adaptation of the score used by Panagiotakos in the ATTICA study 19 . A score of 0 was considered minimum adherence, compared to 80, which would be maximum adherence. Adherence to the Mediterranean diet was grouped into three categories low (0-53 points), medium (54-59 points) and high (60-80 points).
Overweight and general obesity was defined as a Body Mass Index (BMI) ≥ 25 g/m 2 , and abdominal obesity as a waist circumference ≥ 102 cm in men and ≥ 88 cm in women. Hypertension was defined as systolic blood pressure ≥ 140 mmHg, or diastolic blood pressure ≥ 90 mmHg, or current use of antihypertensive treatment or having a personal history of hypertension. Hypercholesterolemia was defined as total serum cholesterol ≥ 250 mg/ dl, high-density cholesterol level (HDL-C) of as < 40 mg/dl in men and < 50 mg/dl in women, and hypertriglyceridemia as serum level of triglycerides ≥ 200 mg/dl.

Statistical analysis.
A descriptive analysis of the distribution of demographic characteristics, lifestyle variables, obesity, and hypertension and biochemical parameters, according to type of prediabetes was performed using the chi-square test, Pearson's chi square (categorical variables), p values < 0.05 were considered statistically significant. Then it was calculated the percentage of subjects who reverted to normal glycaemia according to these variables and types of prediabetes. Multivariate logistic regression was used to estimate odds ratios (OR) with 95% confidence intervals (CIs) to quantify the association between demographic characteristics, lifestyle www.nature.com/scientificreports/ variables, obesity, hypertension and biochemical parameters with reversion to normal glycaemia after adjustment for age and sex. ORs of reversion to normal glycaemia associated with type of prediabetes was estimated using three sequential models of adjustment: adjusting for age and sex (Model A); hypertension, hypercholesterolemia, HDL levels and triglycerides (Model B), and alcohol consumption, smoking, BMI, abdominal obesity, physical activity, adherence to Mediterranean diet (Model C). Confounders were included in each model based on its significant association (p < 0.05) or based on established knowledge of acting a risk factor biological mechanism of action (i.e. BMI). Each successive model included the factors from the previous model. Finally, prediabetes was further subdivided using as cut-off levels FPG < 110 and ≥ 110 mg/dl and HbA1c < 42 and ≥ 42 mmol/mol (< 6 and ≥ 6%), respectively, and the relationship between subtype of prediabetes and the reversion to normal glycaemia was also estimated by models A, B y C. Statistical analyses were performed using the STATA package version 12.0 (StataCorp LP, College Station, TX, USA).
Ethics approval. The study was classified by the Spanish Drug and Health Product Agency as a Non-Interventional (Observational) Post-Authorization Study, and the protocol was approved by the Parc de Salut Mar Clinical Research Ethics Committee in Barcelona. Informed consent was obtained from all participants and/or their legal guardians. Authors confirm that all methods were performed in accordance with the relevant guidelines and regulations.

Results
Baseline characteristics. Among our cohort of prediabetes, mean age was 58.7 years (median: 60 years).
Amongst them, 21% of patients were classified as having isolated impaired FPG (group 1), 27.6% had isolated elevated HbA1c levels (group 2) and 50.9% had both altered parameters (group 3). Table1 shows the baseline characteristics of study cohort according to type of prediabetes. There was an inverse proportion of men and women according to each prediabetes criteria; a total of 61% of group 1 population were men, corresponding percentages were 38.9% for group 2 and 50.3% for group 3, respectively (p < 0.001). Distribution of age was similar within groups, although group 1 tended to be younger. In terms of alcohol consumption, the proportion of daily drinkers were 29% for group 3 compared to 19.5% for group 2 and 34% for group 1. Besides group 3 presented a higher proportion of BMI > 25 kg/m 2 and abdominal obesity. There were no differences in distribution of remaining lifestyle factors as smoking, physical activity and adherence to diet. Those in group 3 had a higher frequency of hypertension and triglycerides levels and there were no differences in distribution of hypercholesterolemia or HDL levels.
Factors associated with the reversion to normoglycemia. Comparison with existing literature. Previous studies evaluating the proportion of reversion to normoglycaemia were heterogeneous in design, duration of follow-up and criteria definition, yielding a broad range of reversion rates [20][21][22][23][24] . Only few did it according to prediabetes criteria 13,25 . A Japanese study similar to ours observed a greater proportion of reversion rates among those with elevated HbA1c levels 25 . Our results showed an opposite trend resulting in lower rates for individuals with HbA1c levels ≥ 6.0% regardless FPG levels, similar to the results provided by a British study 13 . Using the same prediabetes criteria than ours, other study observed that individuals who reverted to normoglycaemia had a low insulin resistance and optimal beta-cell function at baseline 26 . Therefore, it would be reasonable to think that individuals with HbA1c ≥ 6% at baseline, had an increased insulin resistance and/or a decreased beta cell function, which might explain the lowest reversion rates found. Although we did not capture fasting plasma insulin, we used other parameters that might serve as a proxy for recognizing insulin resistance. We calculated the triglyceride (TG) glucose (TyG) index and several lipid ratios: TG/HDL ratio, the total TC/HDL ratio, and the LDL/HDL ratio and analysed them in tertile strata. Individuals with HbA1c ≥ 6% at baseline presented higher levels (Tertile 1) compared with individuals with only impaired FPG. For example, almost 50% of subjects with only impaired FPG were located in tertile 3 while 38.9% of only impaired HbA1c were in tertile 1 and 33% for those with both altered parameters (data not shown). Baseline characteristics such as age less than 50 years, normal weight, absence of abdominal obesity, physical activity, adherence to Mediterranean diet and absence of hypertension have been associated with a higher likelihood to normal glucose regulation. There are studies in subjects with prediabetes that evaluate reversion to normoglycemia through intervention trials focusing on optimal lifestyle actions. The vast majority, although not all 13,27 , draw similar conclusions than ours 4,20,21,23,24,28 . Several trials that compared diet and physical activity promotion programs versus usual care in person with prediabetes reported reversion to normoglycemia between 20 and 52% 29 . Those findings have been confirmed in an intervention trial among persons with high cardiovascular risk 30 .
Both, obesity and body fat distribution are critical factors to decrease insulin sensitivity and B cells function 31 . Physical activity causes increased glucose uptake into active muscles balanced by hepatic glucose production and it improves insulin action 32 . There is also evidence that Mediterranean diet improves insulin sensitivity and prevents from diabetes 33 . In our study, at baseline, only BMI and waist circumference showed an association with FPG, HbA1c, triglycerides and cholesterol, but not the rest of lifestyles (Supplementary Table 1). However, baseline characteristics such as normal weight, absence of abdominal obesity, physical activity, adherence to  www.nature.com/scientificreports/ Mediterranean diet shown association with a higher likelihood to normal glucose regulation. It is probable that individuals reverting to normoglycemia followed an optimal lifestyle behaviour long time before baseline state.

Strengths and limitations.
This is the first study evaluating the role of optimal lifestyle factors in the reversion to normoglycemia according type of prediabetes. Our study highlights the feasibility of conducting a prospective study, with data collected nation-wide by primary care physicians during routine clinical practice. Analytical determinations were performed at different laboratories. This fact could result in some source of misclassification. Since each subject was assigned to the same laboratory during the follow-up, this limitation should be minor and non-differential in relation to the outcome, as it is unlikely the relation between the methods employed by specific laboratories and reversion. A potential source of misclassification when classifying subjects according to FPG levels cannot be ruled out. HbA1c reflects average plasma glucose over the previous eight to 12 weeks 34 , while FPG is subjected to daily variation levels 35 . If any substantial impact of misclassification, we would not be able to find the important Table 2. Percentage of reversion to normal glucose regulation and odds ratio (OR) according to the characteristics of the subjects. *Sex and age adjusted odds ratio, except the odds ratios according sex and age. All variables were considered and treated as categorical variables.

Characteristics, N = 948
Percentage of reversion www.nature.com/scientificreports/ observed differences to reversion according isolated impaired FPG levels. In addition, this study did not consider participants with prediabetes diagnosed based on oral glucose overload 14 , therefore it was not possible to estimate the prognosis for different categories according to this criterion. Although this is a limitation, in any case its impact would be minor as this measure is rarely used in routine practice. Although 24-h dietary recall and dietary records have been used to measure usual dietary intake, both instruments are expensive and unrepresentative of usual intake and therefore, inadequate for the assessment of past dietary intake 36,37 . The food frequency questionnaire is the most commonly used instrument to assess past dietary intake in epidemiological studies. While the questionnaire of other Spanish study (PREDIMED study) used 14 item food, the PREDAPS Study used 20 in order to include some foods that are eaten as substitutes for the foods that are part of the Mediterranean diet. Therefore, the subjects with high adherence to the Mediterranean diet in the present analysis are strict adherents to this diet, since in the calculation of the score the absence of consumption of these substitute foods has been weighted more. Perhaps this explains that in these subjects the reversion to normoglycemia is 78% higher than in the rest of the subjects. On the other hand, an overestimation of physical activity cannot be excluded due to a possible recall bias of the activities carried out in the last two weeks. However, it is unlikely that this overestimation is differential with respect to the type of prediabetes of the subjects.
Researchers were unable to determine a time-dependent variable. However, the vast majority of the factors considered in the present study are chronic conditions or long-term lifestyle factors not susceptible to a fast variation within the follow-up during the study period. Likewise, we did not include in the analyses the existence or not of pharmacological treatment, as antihypertensive agents and lipid lowering drugs. In any case, there no was association between pharmacological treatment and reversion to normal glycemia. The odds ratio of association was 0.90 (0.62-1.29) in the case of use of antihypertensive agents and 1.06 (0.72-1.55) in the case of lipid lowering drugs.
Finally, since reversion to normoglycemia might not be a permanent condition (i.e. some subjects might change from first visit to third) we decided to use as a cutoff point the third visit of follow up to ensure a minimum time-lapse to measure the reversion. The subjects who did not reach the third visit were excluded. When evaluating the baseline characteristics of the dropped-out subjects, there were no significant substantial differences with respect to the subjects who remained in the study (Supplement Table 2). Therefore, a possible selection bias should not be ruled out, it should be minor.
Because the group sizes are small, the confidence intervals of the estimates in the different prediabetes categories overlap. However, in the analyzed prediabetes categories, the heterogeneous distribution of the estimates conforms to what is known about the FPG and/or HbA1c values used to consider prediabetes. Therefore, the magnitude of the association found could be very useful for decision-making in clinical practice.
Implications for research and/or practice. The criteria of prediabetes are a controversial topic 38,39 . On the other hand, our findings suggest that beyond optimal lifestyles, FPG and HbA1c could be in themselves key markers to revert to normal glycaemia in subjects with prediabetes. Since 41% and 33% of the subjects with isolated FPG < 110 mg/dl and isolated HbA1c < 6% reverted to normoglycemia, respectively, this might lead into an overdiagnosis of prediabetes. Perhaps those subjects would not be the specific target for intensification of optimizing lifestyle factors and other actions such as initiating antidiabetic therapy. Further studies evaluating the role of optimal lifestyles in prognostic results, among subjects with prediabetes classified with different criteria, www.nature.com/scientificreports/ will be very useful to harmonize definitions on prediabetes and to better identify specific subjects with a low and high probability of normalizing glycaemia levels.

Conclusions
In conclusion, optimal lifestyle factors showed to be a positive factor to reversion to normoglycaemia after three years of follow up in our prediabetes cohort however, they do not seem to explain differences in the reversion to normal glucose regulation according type of prediabetes.

Data availability
Data will be available upon request.