Balkau et al reported models for the risk of a female being diagnosed with diabetes in the future. These can help to identify a patient who should be screened more closely. The authors are from multiple institutions in France participating in the DESIR study.
Patient selection: French adultfemale
Parameters:
(1) diabetes in the family
(2) waist circumference in cm
(3) body mass index in kg per sq meter
(4) hypertension (blood pressure >= 140/90 OR on drugs for hypertension)
(5) fasting glucose in mmol/L
(6) serum triglycerides in mmol/L
(7) number of deleterious alleles in TCF7L2 and IL6
Parameter
|
Finding
|
Clinical Only
|
Clinical + Lab
|
All
|
family history
|
no
|
0
|
0
|
0
|
|
yes
|
1.09
|
0.80
|
0.75
|
waist circum
|
|
0.095 * WC
|
NA
|
NA
|
BMI
|
|
NA
|
4.38*LN(BMI)
|
4.69*LN(BMI)
|
hypertension
|
no
|
0
|
NA
|
NA
|
|
yes
|
0.64
|
NA
|
NA
|
glucose
|
|
NA
|
9.66*LN(Glu)
|
9.35*LN(Glu)
|
serum TG
|
|
NA
|
0.95*LN(TG)
|
0.86*LN(TG)
|
number alleles
|
|
NA
|
NA
|
0.36 * (count)
|
where:
• Glucose and GGT are reported to be log-transformed.
• The coefficient as published for BMI and fasting glucose results in higher-than-expected values. More reasonable values can be obtained if one assumes that the beta-coefficient was published with a misplaced decimal point.
value of X for clinical model =
= SUM(beta-coefficients) - 11.81
value of X for clinical plus laboratory model =
= SUM(beta-coefficients) - 18.91
value of X for clinical plus laboratory plus genetic model =
= SUM(beta-coefficients) - 20.43
probability of diabetes over the next 9 years=
= 1 / (1 + EXP((-1) * X))
Performance:
• The area under the ROC curve is claimed to be 0.84 for the clinical only, 0.92 for the clinical plus laboratory, and 0.91 for all of the parameters.