Determining the optimal screening interval for type 2 diabetes mellitus using a risk prediction model
Determining the optimal screening interval for type 2 diabetes mellitus using a risk prediction model
Andrei Brateanu 0 1
Thomas Barwacz 1
Lei Kou 1
Sihe Wang 1
Anita D. Misra-Hebert 0 1
Bo Hu 1
Abhishek Deshpande 0 1
Nana Kobaivanova 0 1
Michael B. Rothberg 0 1
0 Medicine Institute, Cleveland Clinic, Cleveland OH, United States of America, 2 Department of Medicine, University Hospitals, Cleveland OH, United States of America, 3 Quantitative Health Sciences, Cleveland Clinic, Cleveland OH, United States of America, 4 Pathology and Laboratory Medicine Institute, Cleveland Clinic , Cleveland OH , United States of America
1 Editor: Massimo Pietropaolo, Baylor College of Medicine , UNITED STATES
Progression to diabetes mellitus (DM) is variable and the screening time interval not well defined. The American Diabetes Association and US Preventive Services Task Force suggest screening every 3 years, but evidence is limited. The objective of the study was to develop a model to predict the probability of developing DM and suggest a risk-based screening interval.
We included non-diabetic adult patients screened for DM in the Cleveland Clinic Health System if they had at least two measurements of glycated hemoglobin (HbA1c), an initial one less than 6.5% (48 mmol/mol) in 2008, and another between January, 2009 and December, 2013. Cox proportional hazards models were created.
The primary outcome was DM defined as HbA1C greater than 6.4% (46 mmol/mol). The
optimal rescreening interval was chosen based on the predicted probability of developing
Of 5084 participants, 100 (4.4%) of the 2281 patients with normal HbA1c and 772 (27.5%)
of the 2803 patients with prediabetes developed DM within 5 years. Factors associated with
developing DM included HbA1c (HR per 0.1 units increase 1.20; 95%CI, 1.13±1.27), family
history (HR 1.31; 95%CI, 1.13±1.51), smoking (HR 1.18; 95%CI, 1.03±1.35), triglycerides
(HR 1.01; 95%CI, 1.00±1.03), alanine aminotransferase (HR 1.07; 95%CI, 1.03±1.11),
body mass index (HR 1.06; 95%CI, 1.01±1.11), age (HR 0.95; 95%CI, 0.91±0.99) and
highdensity lipoproteins (HR 0.93; 95% CI, 0.90±0.95). Five percent of patients in the highest
risk tertile developed DM within 8 months, while it took 35 months for 5% of the middle tertile
to develop DM. Only 2.4% percent of the patients in the lowest tertile developed DM within 5
Funding: The authors received no specific funding
for this work.
Competing interests: The authors have declared
that no competing interests exist.
A risk prediction model employing commonly available data can be used to guide screening intervals. Based on equal intervals for equal risk, patients in the highest risk category could be rescreened after 8 months, while those in the intermediate and lowest risk categories could be rescreened after 3 and 5 years respectively.
Diabetes Mellitus (DM) is a chronic disease associated with micro and macro vascular compli
cations and multiple organ damage.[
] It is estimated that approximately 29 million
Americans have DM, and 8.1 million of these are undiagnosed. In addition, 86 million Americans
aged 20 years or older have prediabetes[
] and 70% of them will ultimately develop DM.[
However, this progression can be stopped with medications and more importantly, with diet and lifestyle changes.
Identifying asymptomatic diabetics and individuals at risk for developing DM may reduce
the burden associated with this disease. While the United States Preventive Services Task
Force (USPSTF) recommends glucose screening in overweight or obese adults aged 40 to 70
] the American Diabetes Association suggests that screening should be done in all
adults beginning at age 45 years regardless of weight, or at any age when obesity or at least one
DM risk factor is present. Glucose abnormalities can be identified measuring the fasting
plasma glucose, glycated hemoglobin (HbA1c), or with the oral glucose tolerance test. HbA1c
has several advantages when compared to the other two tests: it measures chronic glycemic
levels, does not require fasting or two hour testing, and correlates well with the risk of subsequent
diabetic complications. In addition, it can be used to predict the subsequent DM risk.[
Previous models to predict the risk of developing DM in the general population have been
designed in European patient populations,[8±12] or did not use HbA1c.[
In addition, the optimal time for rescreening in patients without DM is not known. A few
studies using mathematical modeling[15±18] or cohorts of Asian populations[
] suggested a
time interval of 3 years for rescreening patients with normal glucose.
The objective of this study was to create a model to predict the risk of progression to DM in patients with normal glucose or pre-diabetes, based upon commonly available risk factors and determine the optimal time for rescreening an individual based on his or her risk.
Research design and methods
Design and setting
We conducted a retrospective cohort study in the Cleveland Clinic Health System (CCHS), a
not-for-profit health system located in Cleveland, Ohio that includes an academic medical
center, 8 community hospitals and 36 community-based practices. CCHS uses a common
electronic health record system (Epic Systems Corporation, Inc.) allowing for the medical data to
be easily extracted. The Cleveland Clinic serves an estimated population of >1.5 million
people, with the majority (75%) of patients coming from the seven counties adjacent to Cleveland,
and the rest from other parts of Unites States or from abroad. To be included, patients had to
be 18 years or older, with at least two measurements of HbA1c, an initial one recorded between
January 1 and December 31, 2008, and a subsequent one between January 1, 2009 and December 31, 2013, in the inpatient or outpatient setting. Exclusion criteria included any of the following prior to 2009: any HbA1c of 6.5% (48 mmol/mmol) or higher, a diagnosis of type 1 or 2
2 / 10
diabetes or gestational diabetes based on ICD-9 codes, and patients taking any glucose
lowering medications or steroids. The Cleveland Clinic Institutional Review Board approved the
The following factors are known to be independently associated with the development of type
2 diabetes: blood glucose level, age, race, family history, increased body mass
] high blood pressure,[
] elevated levels of triglycerides or liver enzymes,[
] and smoking.[
] In addition, an increased level of high-density
lipoproteins (HDL) is known to be associated with a decreased risk of DM. Studies regarding
varying levels of alcohol consumption as a risk factor for developing DM have been
Therefore, the following variables were collected for each patient at baseline: demographic
data (age, sex and race), alcohol consumption and cigarette smoking status, body mass index
(BMI), treatment with statins or antihypertensive medications, family history of DM in a first
degree relative. Laboratory data included: HbA1c, lipid fractions (HDL, low-density
lipoproteins (LDL), triglycerides (TG)), liver enzymes (alanine aminotransferase (ALT)), and
proteinuria, determined within 3 months before or after the first HbA1c measurement.
The primary outcome of the study was progression to DM in patients with prediabetes or no
DM, within five years of the initial HbA1c measurement. DM was diagnosed with at least one reading of HgbA1c greater than or equal to 6.5% (48 mmol/mol).
Cox proportional hazards models were built to predict the risk of developing DM within five
years of the initial HBA1c measurement. Based on the concept of equal treatment for equal
risk, we estimated time intervals for screening among tertiles of risk, so that each tertile would
be screened when 5% of the cohort developed DM. We chose a 5% probability as the threshold,
because this is approximately equal to the proportion of people found to have DM when
patients are initially screened based on the current ADA recommendations.[
Cumulative incidence was plotted to depict the probability of developing DM over time.
Univariate analysis was employed to evaluate the association of all covariates with the
development of DM. For all regression models, candidate variables for multivariable analysis were
evaluated using a stepwise selection procedure, with a P value of 0.05 from the univariate
analysis as the threshold for entry into the model. A multivariate Cox proportional hazard model
was built, to predict the risk of developing DM within 5 years after the initial HBA1c
measurement. Cubic spline models were built to relax model assumption. After the initial model was
created, a stepdown method of backwards elimination was applied to create the reduced
model, which had best prediction power. Then the model was internally validated using 1000
bootstrap iterations. Model discrimination was measured by the C index and calibration
between predicted and observed probabilities was assessed visually via calibration plot.
A total of 5084 participants were included in the study, and 2803 (55.3%) had prediabetes at
study enrollment. Average age was 58.3 ± 13.3 years, 60.3% were men, and the majority were
Caucasians (83.3%). Approximately half of the patients were taking statins (46.3%) and/or
3 / 10
antihypertensive medications (56%). Within 5 years of the initial HbA1c measurement, 872
(17.2%) patients developed DM, of which 100 (11.5%) had normal glucose and 772 (88.5%)
had prediabetes at study enrolment (Table 1).
Compared to patients with no DM within 5 years, patients who developed DM were older
and had a higher BMI. Patients who developed DM were also more likely to be taking statins
and antihypertensive medications, had higher levels of serum TG and lower of HDL.
Based on HbA1c alone, patients had very different probabilities of developing DM at 3 years (Fig 1).
More than one third of patients with an initial HbA1c between 6.1% (43 mmol/mol) and
6.4% (46 mmol/mol) developed DM but only one in ten patients with HbA1c between 5.7%
(39 mmol/mol) and 6.0% (42 mmol/mol), and fewer than 1 in 20 of those with normal glucose.
In the Cox proportional hazards model, the following factors were associated with the risk
of developing DM within 5 years: age, BMI, active smoking, family history of DM, HDL, TG,
ALT, and HbA1c. A family history of DM increased the risk by 30% (HR 1.31; 95%CI, 1.13±
1.51), while a 0.1 unit increase in HbA1c increased the risk by 20% (HR per 0.1 units increase
1.20; 95%CI, 1.13±1.27) and smoking increased the risk by 18% (HR 1.18; 95%CI, 1.03±1.35).
High-density lipoprotein was negatively associated with the risk of DM (HR 0.93; 95%CI, 0.90±0.95) (Table 2).
The final model had a bootstrap bias-corrected c-statistic of 0.809 with a 95% CI (0.795,
0.823). Fig 2 shows a calibration plot of observed vs. predicted risk of DM.
Quintiles of 5-year risk ranged from 0.03±0.49. The model showed good calibration with a ratio of observed to predicted risk of 1.02. An online calculator can be accessed at: http:// riskcalc.org:3838/ab_diabetes_version2/
Based on the model, we divided patients into tertiles of risk for DM. Five percent of patients in the highest risk tertile developed DM within 8 months, while it took 35 months for 5% of the middle tertile to develop DM. Only 2.4% percent of the patients in the lowest tertile developed DM within 5 years (Fig 3).
Discussion and conclusions
In a large cohort of US adults with normal glucose or prediabetes we created a model to predict
the time to develop DM within five years from the initial HgA1c testing, based on the risk of
developing the disease. To our knowledge, this is a novel tool and has the advantage of using
clinical predictors easily obtained when the initial HbA1C is measured. The model can stratify
patients into risk categories to guide future testing intervals. More importantly, our model has
been developed in already screened patients, and not in the general population, like previous
Assuming that patients should be screened when their risk is 5% or higher, patients in the highest risk category should be rescreened after only 8 months, while those in the lowest risk category could go for 5 years or more.
Screening the US population for DM, following the USPSTF or ADA guidelines will
identify a significant number of prediabetics. Most patients with prediabetes will not progress
quickly to DM, and progression can be slowed or even prevented with lifestyle or
] Risk stratification provides 2 potential benefits for these patients.
First, it identifies those most likely to progress to DM, for whom intensive lifestyle therapy or
treatment with metformin would be most indicated. Second, it allows for more efficient
follow-up screening. Determining the appropriate screening interval requires weighing the
benefits of early detection against the costs of frequent monitoring. The USPSTF currently
recommends that individuals with an initial normal glucose should be re-screened every 3
4 / 10
Missing data for
*, 58 patients
², 591 patients
³, 847 patients
§, 213 patients
§§, 73 patients
¶, 1863 patients
#, 1872 patients
**, 1957 patients
²², 1466 patients
BMI, body mass index; DM, diabetes mellitus; HDL, high-density lipoproteins; TG, triglyceride; LDL, low-density lipoproteins; ALT, alanine
aminotransferase; HbA1c, glycated hemoglobin.
5 / 10
Fig 1. Cumulative probability of developing diabetes mellitus. Cumulative probability of 5% would be reached in 19 months for all
patients, in 7 months for patients with HbA1c 6.1±6.4%, and in 26 months for patients with HbA1c 5.7±6.0%. Patients with HbA1c <5.7%
won't reach the 5% threshold within 5 years.
] Applying this interval uniformly may result in over-screening of low-risk patients
and under-screening of high-risk patients. Delaying the diagnosis and treatment of diabetic
patients would be associated with an increase in cardiovascular events and all-cause mortality.
] Alternatively, based on the concept of equal screening for equal risk, the screening interval
for some individuals should be as short as 8 months while for others it might be longer than 5
years. Because of the harms associated with delayed diagnosis, it would probably not be
prudent to wait more than 5 years, even though only a small percentage of low-risk patients would
screen positive at that point.
6 / 10
* Predicted change in the HR for a 5-unit increase in the predictor.
², Predicted change in the HR for a 0.1-unit increase in the predictor.
BMI, body mass index; DM, diabetes mellitus; HDL, high-density lipoproteins; TG, triglyceride; ALT, alanine
aminotransferase; HbA1c, glycated hemoglobin.
Our study has a few limitations. It is based on a retrospective chart review, and the results
may be subjected to bias and errors. We don't know if the population who was screened at
baseline is different from the general population and thus, validation in other medical settings
would be needed. Patients taking steroids were excluded from the study due to the
heterogeneity of doses, duration and individual indication for corticosteroid therapy. In addition, we
Fig 2. The calibration of the model, which measures the relationship between the model's predicted
probability against the actual probability. The final model, had a bootstrap bias-corrected c-statistic of
0.809 with a 95% CI (0.795, 0.823). Quintiles of 5-year risk ranged from 0.03±0.49 and were well calibrated.
7 / 10
Fig 3. Cumulative probability of developing diabetes mellitus. Cumulative probability of 5% would be reached in 8
months for T3 and 35 months for T2. T1 won't reach the 5% threshold.
were not able to identify and include other risk factors that might influence the risk of
developing DM, such as waist circumference, physical activity, and diet.[
In conclusion, we created a tool to risk stratify non-diabetic patients that can be used in
routine clinical practice. Because it is not practical for individual clinicians to seek out an
external website, enter the patient's data, and choose the screening interval, the tool may be
best integrated into the electronic health record (EHR), where it could automatically generate
a personalized screening interval based on risk. Physicians could then receive automatic
reminders to monitor more frequently (e.g. every 8 months) patients with a higher risk of
developing DM. Reminders would be sent less frequently (e.g. every 3, or 5 years) for patients
with intermediate and low risk, respectively. In medical practices where the tool cannot be
practically integrated in the EHR, the HbA1c alone can be used to determine the frequency of
monitoring. Patients with HbA1c between 6.1% (43 mmol/mol) and 6.4% (46 mmol/mol)
could be screened every 7 months, those with HbA1c between 5.7% (39 mmol/mol) and 6%
(42 mmol/mol) every 2 years and those below 5.7% (39 mmol/mol) every 5 years.
Conceptualization: Andrei Brateanu, Thomas Barwacz, Michael B. Rothberg.
8 / 10
Data curation: Andrei Brateanu, Lei Kou, Anita D. Misra-Hebert, Bo Hu.
Formal analysis: Andrei Brateanu, Lei Kou, Bo Hu.
Funding acquisition: Andrei Brateanu, Nana Kobaivanova.
Investigation: Andrei Brateanu, Sihe Wang.
Methodology: Andrei Brateanu, Anita D. Misra-Hebert, Bo Hu, Abhishek Deshpande, Nana
Kobaivanova, Michael B. Rothberg.
Project administration: Andrei Brateanu, Sihe Wang.
Resources: Andrei Brateanu, Sihe Wang, Nana Kobaivanova.
Software: Andrei Brateanu, Lei Kou, Sihe Wang, Bo Hu.
Supervision: Andrei Brateanu, Michael B. Rothberg.
Validation: Andrei Brateanu, Lei Kou, Anita D. Misra-Hebert, Nana Kobaivanova, Michael B.
Visualization: Andrei Brateanu, Lei Kou, Abhishek Deshpande, Nana Kobaivanova.
Writing ± original draft: Andrei Brateanu, Thomas Barwacz.
Writing ± review & editing: Andrei Brateanu, Thomas Barwacz, Lei Kou, Sihe Wang, Anita
D. Misra-Hebert, Bo Hu, Abhishek Deshpande, Nana Kobaivanova, Michael B. Rothberg.
9 / 10
1. American Diabetes A. Diagnosis and classification of diabetes mellitus . Diabetes care . 2014 ; 37 Suppl 1 : S81 ± 90 . https://doi.org/10.2337/dc14-S081 PMID: 24357215 .
2. CDC. http://www.cdc.gov/diabetes/pubs/statsreport14/national-diabetes -report-web.pdf. Access date 01 /01/ 2016 [01/01/2016]. Available from: http://www.cdc. gov/diabetes/pubs/statsreport14/nationaldiabetes-report-web.pdf.
3. Nathan DM , Davidson MB , DeFronzo RA , Heine RJ , Henry RR , Pratley R , et al. Impaired fasting glucose and impaired glucose tolerance: implications for care . Diabetes care . 2007 ; 30 ( 3 ): 753 ±9. https:// doi.org/10.2337/dc07-9920 PMID: 17327355.
4. Khavandi K , Amer H , Ibrahim B , Brownrigg J . Strategies for preventing type 2 diabetes: an update for clinicians . Therapeutic advances in chronic disease . 2013 ; 4 ( 5 ): 242 ± 61 . https://doi.org/10.1177/ 2040622313494986 PMID: 23997928; PubMed Central PMCID : PMC3752182 .
5. Siu AL , Force USPST . Screening for Abnormal Blood Glucose and Type 2 Diabetes Mellitus: U.S. Preventive Services Task Force Recommendation Statement . Annals of internal medicine . 2015 ; 163 ( 11 ): 861 ±8. https://doi.org/10.7326/M15-2345 PMID: 26501513.
6. American Diabetes A. Standards of Medical Care in Diabetes-2016 Abridged for Primary Care Providers. Clinical diabetes: a publication of the American Diabetes Association . 2016 ; 34 ( 1 ):3± 21 . https://doi. org/10.2337/diaclin.34. 1 .3 PMID: 26807004; PubMed Central PMCID : PMCPMC4714725 .
7. Zhang X , Gregg EW , Williamson DF , Barker LE , Thomas W , Bullard KM , et al. A1C level and future risk of diabetes: a systematic review . Diabetes care . 2010 ; 33 ( 7 ): 1665 ± 73 . https://doi.org/10.2337/dc09- 1939 PMID: 20587727; PubMed Central PMCID : PMC2890379 .
8. Lindstrom J , Tuomilehto J. The diabetes risk score: a practical tool to predict type 2 diabetes risk . Diabetes care . 2003 ; 26 ( 3 ): 725 ± 31 . PMID: 12610029 .
9. Schulze MB , Hoffmann K , Boeing H , Linseisen J , Rohrmann S , Mohlig M , et al. An accurate risk score based on anthropometric, dietary, and lifestyle factors to predict the development of type 2 diabetes . Diabetes care . 2007 ; 30 ( 3 ): 510 ±5. https://doi.org/10.2337/dc06-2089 PMID: 17327313.
10. Rahman M , Simmons RK , Harding AH , Wareham NJ , Griffin SJ . A simple risk score identifies individuals at high risk of developing Type 2 diabetes: a prospective cohort study . Family practice . 2008 ; 25 ( 3 ): 191 ±6. https://doi.org/10.1093/fampra/cmn024 PMID: 18515811 .
11. Hippisley-Cox J , Coupland C , Robson J , Sheikh A , Brindle P . Predicting risk of type 2 diabetes in England and Wales: prospective derivation and validation of QDScore . Bmj. 2009 ; 338 :b880. https:// doi.org/10.1136/bmj.b880 PMID: 19297312; PubMed Central PMCID : PMC2659857 .
12. Gray LJ , Taub NA , Khunti K , Gardiner E , Hiles S , Webb DR , et al. The Leicester Risk Assessment score for detecting undiagnosed Type 2 diabetes and impaired glucose regulation for use in a multiethnic UK setting . Diabetic medicine: a journal of the British Diabetic Association . 2010 ; 27 ( 8 ): 887 ± 95 . https://doi.org/10.1111/j.1464- 5491 . 2010 . 03037 . x PMID : 20653746 .
13. Wilson PW , Meigs JB , Sullivan L , Fox CS , Nathan DM , D'Agostino RB Sr . Prediction of incident diabetes mellitus in middle-aged adults: the Framingham Offspring Study . Archives of internal medicine . 2007 ; 167 ( 10 ): 1068 ± 74 . https://doi.org/10.1001/archinte.167.10.1068 PMID: 17533210 .
14. Stern MP , Williams K , Haffner SM . Identification of persons at high risk for type 2 diabetes mellitus: do we need the oral glucose tolerance test? Annals of internal medicine . 2002 ; 136 ( 8 ): 575 ± 81 . PMID: 11955025 .
15. Kahn R , Alperin P , Eddy D , Borch-Johnsen K , Buse J , Feigelman J , et al. Age at initiation and frequency of screening to detect type 2 diabetes: a cost-effectiveness analysis . Lancet . 2010 ; 375 ( 9723 ): 1365 ± 74 . https://doi.org/10.1016/S0140- 6736 ( 09 ) 62162 - 0 PMID: 20356621 .
16. Mortaz S , Wessman C , Duncan R , Gray R , Badawi A . Impact of screening and early detection of impaired fasting glucose tolerance and type 2 diabetes in Canada: a Markov model simulation . Clinicoecon Outcomes Res . 2012 ; 4 : 91 ±7. https://doi.org/10.2147/CEOR.S30547 PMID: 22553425; PubMed Central PMCID : PMCPMC3340109 .
17. Waugh N , Scotland G , McNamee P , Gillett M , Brennan A , Goyder E , et al. Screening for type 2 diabetes: literature review and economic modelling . Health technology assessment . 2007 ; 11 ( 17 ) :iii±iv, ix-xi, 1 ± 125 . PMID: 17462167 .
18. Herman WH , Ye W , Griffin SJ , Simmons RK , Davies MJ , Khunti K , et al. Early Detection and Treatment of Type 2 Diabetes Reduce Cardiovascular Morbidity and Mortality: A Simulation of the Results of the Anglo-Danish-Dutch Study of Intensive Treatment in People With Screen-Detected Diabetes in Primary Care (ADDITION-Europe) . Diabetes care . 2015 ; 38 ( 8 ): 1449 ± 55 . https://doi.org/10.2337/dc14-2459 PMID: 25986661; PubMed Central PMCID : PMCPMC4512138 .
19. Takahashi O , Farmer AJ , Shimbo T , Fukui T , Glasziou PP. A1C to detect diabetes in healthy adults: when should we recheck? Diabetes care . 2010 ; 33 ( 9 ): 2016 ± 7 . https://doi.org/10.2337/dc10-0588 PMID: 20566678; PubMed Central PMCID : PMCPMC2928354 .
20. Rasmussen SS , Glumer C , Sandbaek A , Lauritzen T , Borch-Johnsen K . Determinants of progression from impaired fasting glucose and impaired glucose tolerance to diabetes in a high-risk screened population: 3 year follow-up in the ADDITION study , Denmark. Diabetologia . 2008 ; 51 ( 2 ): 249 ± 57 . https://doi. org/10.1007/s00125-007-0893-8 PMID: 18060659 .
21. Ruchat SM , Vohl MC , Weisnagel SJ , Rankinen T , Bouchard C , Perusse L. Combining genetic markers and clinical risk factors improves the risk assessment of impaired glucose metabolism . Annals of medicine . 2010 ; 42 ( 3 ): 196 ± 206 . https://doi.org/10.3109/07853890903559716 PMID: 20384434 .
22. Link CL , McKinlay JB . Disparities in the prevalence of diabetes: is it race/ethnicity or socioeconomic status? Results from the Boston Area Community Health (BACH) survey . Ethnicity & disease . 2009 ; 19 ( 3 ): 288 ± 92 . PMID: 19769011; PubMed Central PMCID : PMCPMC3706078 .
23. Lyssenko V , Jonsson A , Almgren P , Pulizzi N , Isomaa B , Tuomi T , et al. Clinical risk factors, DNA variants, and the development of type 2 diabetes . The New England journal of medicine . 2008 ; 359 ( 21 ): 2220 ± 32 . https://doi.org/10.1056/NEJMoa0801869 PMID: 19020324 .
24. Brantsma AH , Bakker SJ , Hillege HL , de Zeeuw D , de Jong PE , Gansevoort RT , et al. Urinary albumin excretion and its relation with C-reactive protein and the metabolic syndrome in the prediction of type 2 diabetes . Diabetes care . 2005 ; 28 ( 10 ): 2525 ± 30 . PMID: 16186291 .
25. Yeh HC , Duncan BB , Schmidt MI , Wang NY , Brancati FL . Smoking, smoking cessation, and risk for type 2 diabetes mellitus: a cohort study . Annals of internal medicine . 2010 ; 152 ( 1 ): 10 ±7. https://doi.org/ 10.7326/ 0003 -4819-152-1- 201001050 -00005 PMID: 20048267 .
26. Luo J , Rossouw J , Tong E , Giovino GA , Lee CC , Chen C , et al. Smoking and diabetes: does the increased risk ever go away? American journal of epidemiology . 2013 ; 178 ( 6 ): 937 ± 45 . https://doi.org/ 10.1093/aje/kwt071 PMID: 23817918; PubMed Central PMCID : PMCPMC3816526 .
27. Kao WH , Puddey IB , Boland LL , Watson RL , Brancati FL . Alcohol consumption and the risk of type 2 diabetes mellitus: atherosclerosis risk in communities study . American journal of epidemiology . 2001 ; 154 ( 8 ): 748 ± 57 . PMID: 11590088 .
28. Tuso P. Prediabetes and lifestyle modification: time to prevent a preventable disease . The Permanente journal . 2014 ; 18 ( 3 ): 88 ± 93 . https://doi.org/10.7812/TPP/14-002 PMID: 25102521; PubMed Central PMCID : PMCPMC4116271 .