CLASSIF1 Patient Classification: PROLOGUE Trial (Sitagliptin versus Diet/Training)

Günter Valet

Summary: The analyis of the PROLOGUE clinical trial data (157 parameters, 463 patients, 47 hospitals) for the Diet/Training and Sitagliptin (50-100mg/d) therapies at 0, 12, 24 months by the CLASSIF1 percentile classifier showed similarities for both therapies. The algorithm searches unsupervised for hidden data patterns to identify individual patients. In particular, no perceptible changes in the 42 paraneter pattern of the arterial intima/media thickness (IMT) measurements were detected, indicating no apparent progress of atherosclerosis during the observation period of 2 years for both therapies.

1. Introduction:
The intima/media thickness (IMT) in the Diabetes mellitus type 2 patients of the PROLOGUE ( 1) multicenter PROBE (prospective, randomized, open label, blinded endpoint) design clinical trial, comprising 463 patients in 47 participating hospitals did not increase within the 2 years (0, 12, 24 months) of the Diet/Muscular Training or Sitagliptin (50-100mg/d) therapy. After earlier statistical data reanalyis ( 2) the purpose of this study concerned individual patient classiication by the CLASSIF1 percentile classifier (3, 4).

2. Material and Methods:
The downloaded experimental data (157 data columns: allocation, IMT, other data 1/3, 2/3, 3/3) were analyzed by the CLASSIF1 percentile classifier in search for differences between both therapies at the individual patient level.

3. Results and Discussion:
The classification accuracy for sensitivity and specifcicity was limited (<90%) (tab.1) with sensed differences preferentially at the 0 months time point of the matched patient groups. Prior to the training phase every 5th patient record was assigned to the embedded unknown validation set until database record nr.300, supplemented by the remaining patient records 301-463 as additional unknown validation patients. This results in 239 training set and 224 validation set patients. No high accuracy (>90% correct) resolution between the Diet/Training and the Sitagliptin therapies was obtained during training (tab.1 left) and similarly during validation (tab.1 right) with significant numbers of double classifications (ntsd=38, nvsd=43, see ---> patients 1,2,6,9,42..,, and 8,57,113,120...) occuring in case of closely related patient groups. Some patient records were unclassifiable due to unavailable experimental values (ntsu=11, nvsu=8) in the classification masks.

4. Conclusion:
Both therapies Diet/Training and Sitagliptin decrease HbA1c blood levels to a similae extent without favoring atherosclerosis as determined by multiparameter IMT measurements (n=42) and CLASSIF1 data pattern classification (tab.2).