Identifying cut-off scores with neural networks for interpretation of the incontinence impact questionnaire

We propose to determine cut-off scores for the Incontinence Impact Questionnaire (IIQ) based on the neural network (NN) approach. These cut-off scores should discriminate between patients having poor, moderate, or good quality of life (QoL) secondary to their incontinence problems. Data from two prospectively completed QoL questionnaires, the IIQ (n = 237) and the MOS 36-Item Short-Form Health Survey (SF-36) (n = 237), were analyzed using NN and conventional statistical tools. Kohonen networks identified three distinct clusters of IIQ scores. The three clusters represent the full spectrum of possible scores on the IIQ. We interpreted these clusters as reflecting good, moderate, and poor QoL. We estimated that a score of less than 50 on the IIQ would be representative of good QoL, between 50 and 70 would be moderate QoL, and greater than 70 would be indicative of poor QoL. Validation with the SF-36 data confirmed these categories. The present study demonstrated that the NN approach is opening new areas in the interpretation and clinical usefulness of QoL questionnaires. NN allowed the identification of three levels of QoL and should be useful in clinical decision making.