For diabetes data, the partially varying-coefficient single-index model with distorted measurement error was used for fitting. Due to the large dimension of experimental data and compared with the traditional parametric model and nonparametric model, the application of semiparametric model can not only fit the data better, but also avoid the problem of “curse of dimensionality”. In addition, if the influence of error is ignored during fitting, it may lead to deviation in model estimation. Therefore, body mass index (BMI) was further selected as a potential confounding factor, and it was assumed that both the response variable and the singleindex parameter were contaminated by the BMI. The observation of the experimental results showed that the coefficient of the measurement data of the six serum indicators and sex would vary with BMI, and comparing the results in two different situations, it can be found that the quantitative measurement value, age and average blood pressure of diabetic patients were all polluted by BMI. These results indicate that it is reasonable to select the partially varyingcoefficient single-index model with measurement error for the fitting of this data set, and compared with the semi-parameter model without measurement error, this semi-parametric model can better mine the information in the data.