Risk scores for predicting type 2 diabetes: using the optimal tool

Excerpt

To the Editor: In a recent commentary related to our paper on the performance of a risk score questionnaire for predicting future diabetes [1], Wareham and Griffin make an interesting point about the effect of real life response rates on the true performance of a risk assessment [2]. In our paper we evaluated the performance of the original Finnish diabetes risk questionnaire to predict future screen-detected and clinically diagnosed diabetes. We demonstrated that the performance of the risk score could be improved by adding information on sex, smoking and family history of diabetes [1]. In their commentary, Wareham and Griffin argue that non-response to a questionnaire should be taken into account when evaluating the performance of such a risk questionnaire. They state that when a risk questionnaire is posted out in real life, response rates may be only 50%, and the true sensitivity of the presented score would not be 76% but rather 38%. They go on to suggest that response rates can be improved to nearly 100% by using risk scores that are based on data contained in general practice databases, as has been done in the UK [3, 4], because they do not require collection of new data. They compare this with the 50% response rate for risk score questionnaires reported in the Anglo–Danish–Dutch Study of Intensive Treatment in People with Screen Detected Diabetes in Primary Care (ADDITION)–Denmark Study [5], but do not acknowledge that higher response rates have been achieved—for example 78% in the Hoorn Screening study [6]. …