Patterns of changes in serum lipid prole in prediabetic patients: Results from a 16-year prospective cohort study

Background Lipids abnormality pervasively is associated with the risk of Type 2 diabetes mellitus. To the best of our knowledge, there is no study that examined the longitudinal changes in wide range of serum lipid prole in prediabetic subjects in association with the risk of Type 2 diabetes mellitus in future. This study aimed to identify the patterns of changes in lipids prole over time in prediabetic patients and classify these subjects in order to highlight the high risk people for future diabetes risk. Methods This prospective 16-year (2003–2019) cohort study was conducted among 1228 prediabetic subjects. The study subjects followed over time and changes in their lipid prole include Triglycerides, Cholesterol, high-density lipoprotein cholesterol and low-density lipoprotein cholesterol was evaluated. Latent Markov model was used for data analysis.


Introduction
Type 2 diabetes mellitus (T2DM) is a common chronic disease with major morbidities and mortality rates (1). The World Health Organization (WHO) estimated that the number of diabetic people in the world to reach 522 million by 2030, of whom 439 million will have T2DM (2,3). The prevalence of diabetes between Iranians was 7.9% in 2010, but the distribution of this prevalence in Iran is also diverged widely between 1.3% and 14.5% in various provinces (4).
Prediabetes (PD) is the precursor stage to diabetes mellitus, in which the subject's plasma glucose is higher than normal level, but lower than diabetes mellitus thresholds (5). In recent years, PD prevalence has increased, especially in developing countries. PD prevalence is higher than T2DM (6). It is estimated that 5-10% of subjects with PD, will develop T2DM annually (7). About 30% of diabetic patients in Iran are not aware of their disease; therefore, more attention should be paid to diabetes in Iran (8).
The numerous comorbidities, including obesity and lipid abnormality problems are associated with the risk of developing diabetes (9). Previous evidences suggested that lipid abnormalities are common in people with T2DM and PD (10,11). For instance, a meta-analytic review demonstrated that lipid pro le disorders signi cantly associated with T2DM (12). A community based cross-sectional survey showed a strong association between serum lipid pro le with T2DM and PD (13).
Prediabetic individuals often exhibit an atherogenic pattern of risk factors that includes higher levels of total cholesterol (CHOL), low-density lipoprotein cholesterol (LDL), Triglycerides (TG) and lower level of high-density lipoprotein cholesterol (HDL) than individuals who do not develop diabetes. Lipid abnormalities in diabetics patients, typically characterized by high CHOL, high TG, low HDL and high LDL levels (14).
Although the association between lipid abnormality and T2DM has been investigated in various populations, few studies have been conducted to evaluate such association in prediabetic patients as high risk population.
Due to the increased risk of diabetes progression over the following 5-10 years from the stage of PD (15)(16)(17), it is important to establish appropriate prevention strategies in PD. Lipid pro le is not necessarily stable, especially in prediabetics; accordingly it is necessary to apply an appropriate analytical technique that can provide a comprehensive evaluation of subjects based on changes in lipid pro le over time. Therefore, in current study, an advanced statistical method, i.e. latent Markov model (LMM) was used for addressing the above points.
The previous studies have described the association between lipid pro les with diabetes, without exploring the patterns of changes in lipid pro le over time. LMM, a latent state-switching approach, offer a straightforward approach to classify subjects (latent state) according to the patterns of changes in lipid pro le over time. The application of this method results in the identi cation of subjects within each latent state who are highly similar to each other and uniquely different from those in other states. The model allows us to estimate the probability to move between the states or to retain the same state. The LMM estimates the probability to move between the states or to remain in the same state. Subjects were assigned to the latent states which they had the highest probability for membership. This study aimed to identify the patterns of changes in lipid pro le over time in prediabetic patients and classify these subjects in order to highlight the high risk people for future diabetes risk.

Material And Methods
Page 4/15

Study design and participants
The current study was conducted under the framework of the Isfahan Diabetes Prevention Study (IDPS), which was initiated in 2003 among 3,483 subjects. The IDPS is an ongoing longitudinal study carried out in a cohort of the rst-degree relatives (FDR) of patients with T2DM in Isfahan the largest city in center of Iran to assess the various potential risk factors for diabetes in subjects with a family history of T2DM.
Recruitment methods and examination procedures have been described previously (18). They completed laboratory tests, including a standard 75 g 2-h oral glucose tolerance test (OGTT), and a questionnaire on their health status and on various potential risk factors of diabetes. The subjects were followed up according to standard medical care in diabetes (19). Of the 3,483 subjects who participated at baseline, 1228 had been diagnosed with PD. We used data from 1228 prediabetics. Data include, the baseline, the last measurements and the mean of measurements was recorded among baseline and last, as the second of measurement. Written informed consent was obtained from all subjects in IDPS. The current secondary study has been approved by Bioethics Committee of Isfahan University of Medical Sciences (IR.MUI.MED.REC.1398.532).

Laboratory parameters
Biochemical tests including lipid pro le, fasting plasma glucose (FPG) and OGTT were carried out for all subjects. To determine lipid pro le and FPG, a blood sample was drawn from all subjects after 10-12 h overnight fasting. Postprandial plasma glucose was measured using venous blood sample at 30, 60, and 120 min after oral glucose administration. Plasma glucose and lipid pro le concentrations were determined using enzymatic colorimetric method (ParsAzmoon, Tehran, Iran) adapted to a Selectra-2 auto-analyzer (Vital Scienti c, Spankeren, Netherlands). Serum concentration of LDL was calculated by Friedwald equation in subjects with serum TG levels < 400 mg/dL (20). Serum concentration of HDL, CHOL and TG measured using standardized procedures (20).
De nitions and diagnostic criteria were based on the American Diabetes Association (ADA) guidelines.
Symptomatic subjects with FPG ≥ 11.1 mmol/L were considered diabetic. If FPG was ≥ 7 and < 11.1 mmol/L, a second FPG was measured on another day. If the second FPG was also ≥ 7 mmol/L, subjects were classi ed as diabetic. FPG ≥ 7 mmol/L or 2-hour PG ≥ 11.1 mmol/L also de ned diabetes mellitus. Impaired glucose tolerance (IGT) was interpreted as 7.8 ≤ 2hpost glucose load (75 g glucose) ≤ 11.0 mmol/L(5). If FPG was in the range 5.6 ≤ FPG ≤ 6.9 mmol/L, it was considered as impaired fasting glucose (IFG) (5). In addition, all subjects developing IFG and IGT were pooled in a unique" impaired glucose metabolism "(IGM) group in the analyses.
In the analyses, we considered following categories as abnormal: TG level of more than 150 mg/dL; LDL level of more than 100 mg/dL and CHOL level more than 200 mg/dL; in both men and women (21), HDL level of less than 40 mg/dL in men; less than 50 mg/dL in women (21) and HDL level > 60 mg/dL, optimal condition, considered protective against heart disease (21).

Other variables
The subjects completed a demographic questionnaire including age and gender. Physical activity recorded in an International Physical Activity Questionnaire) IPAQ) (22). Anthropometric and clinical measurements, including body mass index (BMI) (by dividing weight [kg] to the square of height [m 2 ]), FPG and lipid pro le include TG, CHOL, HDL and LDL was recorded. The process of administering and collecting the questionnaires was conducted at the Isfahan Endocrine and Metabolism Research Center, Isfahan University of Medical Sciences.

Statistical analysis
Continuous and categorical basic characteristics of study subjects were presented as mean (standard deviation (SD)) and frequency (percentage), and compared between study groups using analysis of variance (ANOVA) or independent samples T-Test and Chi-square tests, respectively.
To analysis the patterns of changes in serum lipid pro le over time in prediabetic patients, LMM was applied (23). We used three measures of lipid indices obtained from subjects including rst measure at baseline, mean values during follow up period and last measure. LMM identi ed number of latent states in studied subjects based on patterns of changes in study lipid measures, also provided the probability moving between various states. The process of LMM tting was as follows: the following LMM were estimated, 2-State 1-Class, 2-State 2-Class, 2-State 3-Class, 3-State 1-Class, 3-State 2-Class and 3-State 3-Class sequentially, and the model with 2-State 1-Class was selected based on goodness of t criteria.
The balance between t and parsimony (number of parameters) of different models was estimated using Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC). The optimum number of states was determined through comparing the AIC, BIC, classi cation error and entropy indices across models. Lower AIC, BIC and classi cation error and higher entropy, indicate the better model tting and the better states separation (24,25). We extracted two latent states from TG, CHOL, LDL and HDL in order to evaluate their association with progress diabetes over time, and were labelled as "State1" and "State2".
After selecting the number of proper latent states, the LMM without/with covariates including age, gender, physical activity, BMI, and FPG were tted and the patterns of changes in serum lipid pro le over time in prediabetic patients was evaluated. The tted model was adopted separately in gender subgroups. The interpretation of extracted latent states was done based the mean values of lipid pro les calculated by LMM.
Our model also allows estimate the longitudinal change in the metabolic condition of prediabetic subjects. The LMM estimates the initial and transition probabilities according to the mean values of TG, CHOL, LDL and HDL. The initial probability de ne as, probability of current state is that is needed to predict the future. The transition probability is de ned as, probability to move of subjects between different latent states. The subjects in the PD state can remain and/or move to other latent states. The

Results
General characteristics of subjects at baseline are presented in Table 1. Of 1228 study subjects with mean (SD) 44.007 (6.86) years, 73.6% were females. The baseline demographic, age was not different between the males and females. General characteristics of subjects at the end of follow up are presented in Table 2. Over the 16-year follow-up, 339(27.6%) became diabetic, 204(16.6%) was normal, 403 (32.8%) remained PD (IFG and IGT) and the data about nal status of 282 (23%) of subjects was not available.  We estimated a series of LMM to determine the number of latent states (2-3State 1-3Class). All models' t criteria as well as interpretability of extracted states strongly suggested a LMM with two latent states based on the patterns of changes in lipid pro le. According to t criteria, 2-State 1-Class model selected, with lower AIC, BIC, parameter number and classi cation error and as well as higher entropy (Table 3).  Table 4 presents the identi ed latent states of subjects based on lipid pro le resulted from LMM, in total, males and females sample. For total/ males/ and females sample, two latent states were identi ed. Interpretation of states is based on the mean of lipid pro le. The state1 consists of subjects who had lower problem in lipid pro le levels; the subjects in this state had relatively low values of lipid pro le. This state is interpreted as low tendency to progress diabetes in future and consists 74%/ 74%/ 69% of the sample. The state2 consists of subjects who had higher problem in lipid pro le levels; the subjects in this state had relatively high values of lipid pro le. This state is interpreted as high tendency to progress diabetes in future and consist 26%/ 26% / 31% of the sample. Similar ndings were observed when LMM was tted separately in male and female genders (Table 4).  Table 5 presents the initial and transition probabilities observed during the study from one particular state to other states. On the basis of the estimates, at the beginning of the period of investigation, in all groups include total, males, and females more than of half (initial probability was 73%/ 74% / 68%) of subjects were in the latent states1. The probability of being in the state1 (without/with covariates) is higher than the state2.  The state0 is initial state. Probabilities represent the probability of transition from a particular state to other states from row to column According to these results, a subject in the state1 or in the state2 will remain at the same condition (without/with covariates) with the probability of ranging from 77%to 97%. In all groups include total, males and females the transition probability from the state1 to the state2 is lower than the transition probability from the state2 to the state1.

Discussion And Conclusions
In this prospective longitudinal study, we followed 1228 prediabetic patients from 2003 to 2019 and evaluated the changes in serum lipid pro le over time using LMM. Two latent states were identi ed based on the patterns of changes in lipid pro les mean and the states characterized by levels of tendency to progress diabetes (low/ high) with prevalence rates of (74%/ 26%), respectively. We observed that the lipid pro le mean; in subjects assigned to "high tendency to progress diabetes" state was more than "low tendency to progress diabetes" state. The transition probability from the low to high tendency state was lower than the transition probability from high to low tendency state.
We did not nd any study such as current one, which classi ed prediabetic patients into homogeneous states based on lipid pro le mean over time using LMM. However, there are many studies in this regard among general population, and some speci c population with applying simple statistical methods (9,10,28,29).
Previous studies have focused on investigating the association of each lipid pro le; TG, CHOL, HDL and LDL with the risk of diabetes in future or concurrently, separately. It is believed that lipid pro le abnormality is a strong risk factor for T2DM in prediabetic patients (17,32).
In the present study, the subjects in high tendency to progress diabetes state had lipid pro le abnormality.
We observed that the mean of lipid pro le abnormality associated with "high/low tendency to progress diabetes" states. This nding is in line with the results of previous studies have emphasized on the association of lipid pro le disorders with the risk of diabetes (11,12,23,(33)(34)(35)(36).
The Bhowmik et al. study obtained similar results with our study in terms of levels of dyslipidemia.
Results showed a strong association between serum lipid pro le and T2DM and PD. In addition, high levels of TG in combination with low levels of HDL showed the highest association with T2DM and PD. The levels of high CHOL, high TG, and low HDL were more elevated among subjects with T2DM and PD (13).
In the present study, in an irregular pattern, low HDL level was not associated with increased T2DM. In line our study, Hasse et al. reported that genetically reduced HDL was not associated with increased T2DM, suggesting that the corresponding observational association is due to confounding and/or reverse causation (9). In contrast, Hirano in the Hawaii-Los Angeles-Hiroshima study found that HDL is a predictor of T2DM, independent of age and gender in both Japanese-American and native Japanese(37). Janghorbani et al. in a population based longitudinal survey showed that low HDL level was a weak predictor of T2DM independent of age and gender in a cohort of high-risk individuals in Iran (38).
Although numerous researches exist about the risk factors of diabetes, but most research has ignored the complexity of diabetes disease and the reversible of diabetic states. In the current study, the probability for a subject in low tendency to progress diabetic state and to remain in the same condition was more than the probability for a subject in a high tendency to progress diabetic state which to remain in the same condition. Further, the transition probability from the low tendency to progress diabetic state to high tendency to progress diabetic state was lower than the transition probability from the high tendency to progress diabetic state to low tendency to progress diabetic state.
It is important to recognize some strengths and limitations of the present study. A major strength of our study is the applications of latent Markov model for classifying subjects according to the patterns of changes in lipid pro le over time, instead of considering them as a single index. Other strengths of this study are population consisting of a large cohort of prediabetic patients, and the long-lasting followed-up of these subjects (16-year) and adjustment for some potential confounders in the analyses. The current ndings were drawn from a study population of prediabetic patients; therefore, the results may not be applicable to all populations. We found that states identi ed based on lipid pro le by LMM, in particular "low tendency to progress diabetes" and "high tendency to progress diabetes" are associated with the risk of diabetes in future in prediabetic patients.
In conclusion, abnormality of serum lipid pro les remains a signi cant and growing problem in prediabetic subjects as high risk population. The reduction in the problem burden will require changes at the policy level as well as at the personal level. Finally, should draw attention to abnormalities of lipid pro les is as an important step in preventing and managing diabetes.

Declarations
• Ethics approval and consent to participate The current secondary study has been approved by Bioethics Committee of Isfahan University of Medical Sciences (IR.MUI.MED.REC.1398.532). Written informed consent was obtained from all subjects in IDPS.

• Consent for publication
Not applicable

• Availability of data and materials
The data that support the ndings of this study are available from the corresponding author upon reasonable request.

• Competing interests
No potential con ict of interest was reported by the authors. • Authors' contributions AF, MA and AA contributed to the conception and design of the main study, collection and assembly of the data. AF supervised the current secondary study in the framework of a PhD thesis. SS contributed to the statistical analysis, AF and SS contributed to the interpretation of the results. AF and SS contributed to drafting the manuscript. AF, MA and AA revised it critically for important intellectual content in order for the nal approval of the version to be published. All authors read the nal version of manuscript and approved it.