Preventing Overdiagnosis 2016 in Barcelona ( 20-22 September, 2016 ) Measurement error of HbA1c for screening for type 2 diabetes among healthy Japanese adults Kuniyoshi Hayashi 1, 2, Sachiko Ohde 1, Gautam Deshpande 1, Osamu Takahashi 1 1 Center for Clinical Epidemiology / Graduate School of Public Health Planning Office, St. Luke s International University 2 CREST, Japan Science and Technology Agency
Contents 1 Background and objectives 2 A statistical method for screening for diabetes 3 Analysis results 4 Concluding remarks
1 Background and objectives 2 A statistical method for screening for diabetes 3 Analysis results 4 Concluding remarks
Report of Ministry of Health, Labour and Welfare The number of diabetic patients in Japan (unit: 1000) http://www.dm-net.co.jp/calendar/chousa/population.php Year / Month Male Female Total 2014 / 10 1768 1401 3166 2011 / 10 1487 1215 2700 2008 / 10 1312 1061 2371 2005 / 10 1323 1147 2469 2002 / 10 1208 1076 2284
Result of 2012 National healthy nourishment investigation" A survey, National healthy nourishment investigation, carried out in November 2012 by the Ministry of Health, Labour and Welfare The number of persons for whom diabetes is strongly suspected is approximately 9.5 million. The number of persons for whom there is a possibility of diabetes is approximately 11 million. www.mhlw.go.jp/stf/houdou/0000032074.html
Criteria for judgment of hyperglycemia 1. FPG 126 mg dl. Fasting is defined as no caloric intake for at least 8 h. 2. 2-h plasma glucose 200 mg dl during an oral glucose tolerance test (OGTT). 3. In a patient with classic symptoms of hyperglycemia or hyperglycemic crisis, a random plasma glucose 200mg dl. 4. A1C 6.5%. The test should be performed in a laboratory using a method that is based on NGSP. Standards of medical care in diabetes-2013. Diabetes Care, 2013, Suppl 1:S11-S66.
Diagnosis and classification of diabetes mellitus Previous reports 1), 2) recommended that HbA1c can be used to diagnose diabetes and that the diagnosis can be made if the HbA1c level is more than 6.5%. The American Diabetes Association has suggested 5.7%-6.4% as the high-risk range 1). The International Expert Committee recommended that persons with a HbA1c level of between 6.0% and 6.5% were at particularly high risk and may be considered for diabetes prevention interventions 1).
Purpose of our study We used the statistical evaluation method for measurement errors used in our previous study. We first estimated the common standard deviation of measurement errors for each patient at each time point. With the estimated common standard deviation, we developed the 95% confidence interval for the expected HbA1c value for each patient at each time point. We investigated the magnitude of the estimated common standard deviation. Finally, based on the above approach, we detected signal HbA1c values to be re-examined on the longitudinal observed data in each patient.
Global report on diabetes by WHO Managing diabetes The starting point for living well with diabetes is an early diagnosis the longer a person lives with undiagnosed and untreated diabetes, the worse their health outcomes are likely to be (Global report on diabetes, executive summary. World Health Organization, 2016). On the other hand, we also have to consider overdiagnosis in each patient. Therefore, it is important for us to accurately distinguish a true signal increment in HbA1c from noise and perform an optimal diagnosis.
1 Background and objectives 2 A statistical method for screening for diabetes 3 Design and analysis results 4 Concluding remarks
A statistical method for screening for diabetes A representative technique for the resolution of noise and signal for the data has been already proposed (Glasziou et al., 2008). Recently, we have examined an approach to distinguish signals from noise based on the concept of statistical confidence intervals.
A statistical method for screening for diabetes
n i t j y j i, 1, ;,, 1 i t n j n n t i j i i t j y y y n y y y i y y y t j 1 1 1 1 1 1 1 1 Subject Point Time j : nt, N t j j i y i t 1 1 ˆ : measurements in the -th time point of the -th patient
Constitution of the 95% confidence interval [ ˆ 2 ˆ, ˆ 2 ˆ i common i common] 1 ˆ ˆ 2 n t common y i 1 j 1 i j N n 2, where the degree of freedom of is. ˆcommon i 2 N n
Purpose of our study We used a statistical evaluation method used in our previous study for measurement errors. We first estimated the common standard deviation of measurement errors for each patient at each time point. With the estimated common standard deviation, we developed the 95% confidence interval for the expected HbA1c value for each patient at each time point. We investigated the magnitude of the estimated common standard deviation. Finally, based on the above approach, we detected signal HbA1c values to be re-examined on the longitudinal observed data for each patient.
About design and target patients Study design: retrospective open cohort study Patients: Japanese individuals aged 30-74 years who visited a private preventive health center at least twice between 2005 and 2014 ( healthy adults: 66,160; males: 31,760 ) Exclusion criteria: those on diabetes medication, with baseline HbA1c 6.5, or CVD at the first visit
Japanese individuals aged 30-74 years who visited a private preventive health center at least twice between 2005 and 2014 ( healthy adults: 66,160; males: 31,760 )
1 Background and objectives 2 A statistical method for screening for diabetes 3 Analysis results 4 Concluding remarks
Japanese individuals aged 30-74 years who visited a private preventive health center at least twice between 2005 and 2014 ( healthy adults: 66,160; males: 31,760 ) Mean (SD) age: 46.5 years (10.4 years) Mean (SD) HbA1c: 5.4% (0.3%) Mean (SD) BMI: 22.3 (3.2) The overall HbA1c measurement error ( ): 0.17% ˆcommon
Japanese individuals aged 30-74 years who visited a private preventive health center at least twice between 2005 and 2014 ( healthy adults: 66,160; males: 31,760 ) The HbA1c measurement error for men in each age group: 30-44: 0.16 %, 45-59: 0.17%, 60-74: 0.17 % The HbA1c measurement error for women in each age group: 30-44: 0.18 %, 45-59: 0.18%, 60-74: 0.17 %
Result of analysis 95% upper limit 95% lower limit
Result of analysis 95% upper limit 95% lower limit
1 Background and objectives 2 A statistical method for screening for diabetes 3 Analysis results 4 Concluding remarks
Summary We found that HbA1c includes 0.17% measurement error, overall. There was a higher measurement error among younger women while there were no gender differences among elderly groups. Clinicians should consider that HbA1c includes a substantially large measurement error, which also varies by gender. By using the estimated common standard deviation at each age group, we can decide whether values exceed CI thresholds at past and current HbA1c measurements. Therefore, clinicians can better identify which patients warrant ongoing examination. Moreover, when we estimate the predicted values of HbA1c at the next time point based on the AR and mixed effect models, if all the predicted HbA1c values exceed the upper 95% confidence threshold, intervention may be warranted.
Concluding remarks We are estimating the common standard deviations with the patients data stratified by gender, age group, and current BMI. Using the estimated standard deviations, we can precisely detect the movement of the observed measurements of each patient. Based on our approach, these findings can be useful in avoiding overdiagnosis not only in diabetes but also in other diseases.
References Glasziou PP, Irwig L, Heritier S, Simes RJ, and Tonkin A, for the LIPID Study Investigators, Monitoring cholesterol levels: Measurement error or true change? Annals of Internal Medicine, 2008;148:656 661. Global report on diabetes, executive summary. World Health Organization, 2016. International Expert Committee report on the role of the A1C assay in the diagnosis of diabetes, Diabetes Care, 2009, 32:1327-1334. Standards of medical care in diabetes-2013. Diabetes Care, 2013, Suppl 1:S11-S66. Use of Glycated Haemoglobin (HbA1c) in the Diagnosis of Diabetes Mellitus: Abbreviated Report of a WHO Consultation, World Health Organization, 2011. http://www.dmnet.co.jp/calendar/chousa/population.php www.mhlw.go.jp/stf/houdou/0000032074.html
Preventing Overdiagnosis 2016 in Barcelona ( 20-22 September, 2016 ) Thank you for your attention!