Application of a Prediction Model for Identification of Individuals at Diabetic Risk

Summary

As a follow-up to our preceding paper, we attempted to extract features of health risk progression for diabetes in Sequential Multi Layered Perceptron (SMLP) via inverse processing of the learned structure. The time-varying risk progress was assessed with risk trajectory and conditional mixture model.

Overall risk cut along with the prediction was stable over time and high body mass index (BMI) tops the health behavioral risks predicting the onset of diabetes. For the initial prediction, high BMI (obesity), high blood pressure (BP), high cholesterol, and diet in fatty food were significant. Over time, variations in trajectory were due to changes in BMI, stress, BP, cholesterol, and fatty food intake.

We tested the effectiveness of identifying prediabetics by the SMLP by applying the implemented SMLP to a test population of employees from a large manufacturing company, where an early worksite health promotion was initiated (1984). This resulted in a potential sensitivity (71.4%) although there were issues like mapping corresponding risks and large time lags.

A secondary test on the similar population as in the previous paper showed a promising sensitivity (86.5%) over 3 years.

When combining with targeted screening such as impaired glucose tolerance test only for those predicted to be diabetics, the presented prediction model and extracted features can be used in implementing an effective disease prevention and management program.

Keywords

Precursors - diabetes - time-varying - SMLP - screening - health risk appraisal

• References

• 1 National Institutes of Health, National Heart, Lung, and Blood Institute. Clinical guidelines on the identification, evaluation, and treatment of overweight and obesity in adults – the evidence report. Obes Res 1998; 6 suppl (Suppl. 02) 51S-209S.
• 2 Hitman GA. Type 2 diabetes: prediction and prevention. Wiley Europe. 1999
• 3 Park J, Edington DW. A sequential neural model for prediction of diabetes (NIDDM). Artif Intell Med 2001; 23: 277-93.
• 4 Bazan J, Skowron A, Synak P. Dynamic Reducts as a Tool for Extracting Laws from Decision Tables, Proc Symp on Methodologies for Intelligent Systems Charlotte, NC, USA, Lecture Notes in Artificial Intelligence. Springer Verlag 1994; 869: 346-55.
• 5 Tafeit E. et al. Artificial Neural Networks compare to factor analysis for low dimensional classification of high-dimensional body fat topography topography data of healthy and diabetic subjects. Comput Biomed Res 2000; 33: 365-74.
• 6 Tickle AB, Andrews R. The truth will come to light: directions and challenges in extracting the knowledge embedded within mined artificial neural networks. IEEE Trans Neural Networks 1998; 9 (06) 1057-68.
• 7 Edington DW, Yen L, Braunstein A. The Reliability and Validity of HRAs. SPM Handbook of Health Assessment Tools. 1999
• 8 Verbeke G, Lesaffre E. A linear mixed-effects model with heterogeneity in the randomeffects population. J Amer Stat Assoc 1996; 91: 217-21.
• 9 Bryk AA, Raudenbush SW. Hierarchical Linear Models: applications and data analysis methods, Newbury Park, Sage Publications. 1992
• 10 American Diabetes Association (ADA). The prevention or delay of type 2 diabetes. Diabetes Care 2002; 25: 742-9.
• 11 ADA. Screening for type 2 diabetes. vol 21, supplement 1: clinical practice recommendations. 1998
• 12 McCulloch D. et al. Improvement in diabetes care using an integrated population-based approach in a primary care setting. Disease management 2000; 3 (02) 75-82.
• 13 Sergio S. et al. Diet and risk of type 2 diabetes. N Engl J Med 2002; 346 (04) 297-8.
• 14 Diabetes prevention program research group. Reduction in the incidence on type 2 diabetes with lifestyle intervention or metformin. N Engl J Med 2002; 346 (06) 393-403.
• 15 ADA. Type 2 diabetes: incremental medical care costs using the first 8 years after diagnosis. Diabetes Care 1999; 22 (07) 1116-24.