Predicting Alzheimer's Disease

My boyfriend explained his data science project today. He is trying to predict alzheimer's disease based on a huge set of data from USC--medical data (MRI results, etc), demographic data, questionnaire answers. Using all of the data, his accuracy during test was around 85 percent. Surprisingly, just by predicting with the questionnaire answers, his accuracy was 75%, which is not too different from when you use all of the data. He also found out that the accuracy was 60% when he just used one question (recalling words). It was interesting for me how just one simple question could predict whether someone had the alzheimer's disease or not.