npj Schizophrenia (Feb 2022)

Development and validation of a web-based prediction tool on minor physical anomalies for schizophrenia

  • Xin-Yu Wang,
  • Jin-Jia Lin,
  • Ming-Kun Lu,
  • Fong-Lin Jang,
  • Huai-Hsuan Tseng,
  • Po-See Chen,
  • Po-Fan Chen,
  • Wei-Hung Chang,
  • Chih-Chun Huang,
  • Ke-Ming Lu,
  • Hung-Pin Tan,
  • Sheng-Hsiang Lin

DOI
https://doi.org/10.1038/s41537-021-00198-5
Journal volume & issue
Vol. 8, no. 1
pp. 1 – 7

Abstract

Read online

Abstract In support of the neurodevelopmental model of schizophrenia, minor physical anomalies (MPAs) have been suggested as biomarkers and potential pathophysiological significance for schizophrenia. However, an integrated, clinically useful tool that used qualitative and quantitative MPAs to visualize and predict schizophrenia risk while characterizing the degree of importance of MPA items was lacking. We recruited a training set and a validation set, including 463 schizophrenia patients and 281 healthy controls to conduct logistic regression and the least absolute shrinkage and selection operator (Lasso) regression to select the best parameters of MPAs and constructed nomograms. Two nomograms were built to show the weights of these predictors. In the logistic regression model, 11 out of a total of 68 parameters were identified as the best MPA items for distinguishing between patients with schizophrenia and controls, including hair whorls, epicanthus, adherent ear lobes, high palate, furrowed tongue, hyperconvex fingernails, a large gap between first and second toes, skull height, nasal width, mouth width, and palate width. The Lasso regression model included the same variables of the logistic regression model, except for nasal width, and further included two items (interpupillary distance and soft ears) to assess the risk of schizophrenia. The results of the validation dataset verified the efficacy of the nomograms with the area under the curve 0.84 and 0.85 in the logistic regression model and lasso regression model, respectively. This study provides an easy-to-use tool based on validated risk models of schizophrenia and reflects a divergence in development between schizophrenia patients and healthy controls ( https://www.szprediction.net/ ).