Changes in levels of haemoglobin A_1c during the first 6 years after diagnosis of clinical type 2 diabetes

Abstract

Objective: To assess the variability in levels of glycosylated haemoglobin (HbA_1c) during the first six years after diagnosis of clinical type 2 diabetes in relation to possible predictors. Material and methods: Data were from a population-based sample from general practice of 581 newly diagnosed diabetic patients aged 40 or over. Estimation of HbA_1c was centralized. The changes in levels of HbA_1c were described by HbA_1c at diagnosis and a regression line fitted to the HbA_1c measurements after 1-year follow-up for each patient. The predictive effect of patient characteristics for changes in HbA_1c was investigated in a multivariate mixed model. Results: During the first year after diabetes diagnosis, HbA_1c dropped to near normal average level and then started rising almost linearly. A sharp rise in long-term glycaemic level was observed in approximately a quarter of the patients, especially the relatively young. Of 581 patients, 156 (26.9%) patients, however, experienced a fall in HbA_1c after 1-year follow-up and another quarter showed constant or only slowly rising HbA_1c. The changes in levels of HbA_1c were only predicted by diagnostic HbA_1c and age. Conclusions: During the first 6 years after the diagnosis of clinical type 2 diabetes, changes in levels of HbA_1c show considerable inter-individual variability with age as the only long-term predictor. The results indicate that it is important to monitor changes in HbA_1c more closely and intensify treatment of those often relatively young patients who actually experience the beginning of an apparently relentless deterioration of their glycaemic control.

Key Words::

• Age
• diabetes mellitus
• type 2
• follow-up studies
• glycaemic control
• hemoglobin a
• glycosylated
• hypoglycaemic agents
• sex
• variability

Acknowledgements

The authors are grateful to the patients and doctors who participated in the study and to Lise Bergsøe for her expert technical assistance. Major funding: The Danish Medical Research Council, The Danish Research Foundation for General Practice, The Health Insurance Foundation, The Danish Ministry of Health, Novo Nordisk Farmaka Denmark Ltd., and The Pharmacy Foundation.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

Appendix

Table AI. The estimated changes in levels of HbA_1c during the first 6 years after diabetes diagnosis according to socio-demographic, clinical, biochemical and behavioural patient characteristics at diagnosis.

Table AII. The effect of patient characteristics at diabetes diagnosis on the estimated level of HbA_1c one year after diagnosis and on the estimated slope of the HbA_1c-curve following the first year after diagnosis.

Statistical note to Table AII: The multivariate model for HbA_1c developments includes two effect estimates for each patient characteristic: a main effect and an interaction with time. Hence, (intercept) denotes the constant in the multivariate linear regression and the estimates below it are the main effects of the patient characteristics; (slope) denotes the linear effect of time and the effects listed below it are the interactions with time and thereby denote influences of the characteristics on the time trend.

A model including sex, age and main risk factors (body mass index, HbA_1c, systolic blood pressure, total cholesterols and urinary albumin) measured at diagnosis was contrasted to models augmented with (a) diabetic complications at diagnosis (diabetic retinopathy, coronary heart disease, peripheral vascular disease, peripheral neuropathy, familial disposition to diabetes) and/or (b) lifestyle factors at diagnosis (smoking, physical activity, marital status, residence and GP's acquaintance with the patient) using F-tests. There was no statistically significant reduction of the variation explained by the model with the addition of a (p = 0.99), b (p = 0.23) or both a and b (p = 0.79) to the rest of the model.

Missing values of the patient characteristics were imputed 36 times by a sequential imputation algorithm [1]. Of 2987 observations eligible for the regression model, 344 (11.5 %) were missing. The multiple estimates obtained were combined by Rubin's rule [2], and groups of predictors were assessed by F-tests based on a multivariate extension of Rubin's rule [3].

Appendix References

• Raghunathan TE, Lepkowski JM, Van Hoewyk J, Solenberger P. A multivariate technique for multiply imputing missing values using a sequence of regression models. Survey Methodol 2001;27:85–95.
   Google Scholar
• Rubin DB. Inference and missing data. Biometrika 1976;63:581–90.
   Web of Science ®Google Scholar
• Li KH, Raghunathan TE, Rubin DB. Large-sample significance levels from multiply imputed data using moment-based statistics and an F-reference distribution. J Am Stat Assoc 1991;86:1065–73.
   Web of Science ®Google Scholar