A statistical decision-making system has been developed which will predict the clinical status of a patient with cystic fibrosis based on daily self measurements obtained at home. The data for the study were collected from CF patients within 7-12 years of age. Thirty-two participants recorded four daily measurements (weight, vital capacity, breathing rate, and resting pulse) and one weekly measurement (height). In addition to the 4 daily measured values, the clinical status of each patient at his,/her most recent previous clinic visit was used as a predictor variable. The measured values were used as the basis for the development of a discriminant rule. The goal of the rule was to determine whether each patient's clinical status was deteriorating, stable, or improving at the time of the most recent set of weekly measurements. Three types of analysis were performed: linear discriminant analysis, quadratic discriminant analysis, and nearest neighbor. Quadratic discriminant analysis provided the best discrimination due to the differences in the covariance matrices among the populations. The rule was able to correctly classify 77% of the 103 cases in the learning set. To further evaluate the rule, both a weighted classification percentage and weighted kappa statistic were calculated for the rule. Bootstrapping was used to predict the performance of the rule on the population with results of 77'/ correctly classified overall.
- Home monitoringMultivariate discriminant analysisComputer-assisted discrimination Cystic fibrosisChronic diseaseBootstrapping