Predictive Equation of Stern et al to Screen for Type 2 Diabetes Mellitus

Parameters:

(1) age of the patient in years

(2) body mass index in kilograms per square meter

(3) fasting glucose in mg/dL (not stated as plasma or whole blood)

(4) systolic blood pressure in mm Hg

(5) gender

(6) HDL cholesterol in mg/dL

(7) race

(8) family history of diabetes in parent or sibling

where:

• Blacks were not included in the study population.

X =

= (-13.415) + (0.028 * (age in years)) + (0.079 * (fasting glucose)) + (0.661 * (points for gender)) + (0.070 * (BMI)) - (0.039 * (HDL cholesterol)) + (0.481 * (family history)) + (0.018 * (systolic blood pressure)) + (0.412 * (points for race))

probability of diabetes =

= 1 / (1 - EXP((-1) * X))

NOTE: Almost all of the logistic regression equations I have seen have had the form P = (1 / (1 + EXP((-1) * X))). When the equation was implemented as described, unreasonable values are returned. More reasonable results are returned when the typical equation is used. It is interesting that this is the same apparent error as shown in Predictive Equation of Tabaei and Herman to Screen for Diabetes Mellitus.

alternative equation for probability of diabetes giving more reasonable results =

= 1 / (1 - EXP((-1) * X))