摘要
目的：本文提出了一种制定多元参考值范围的新方法——全息主成分叠加复元法(简称全息元法)。用以降低多个单变量参考值范围联合判断所产生的高假阳性率，改善传统多元方法对变量正态性和（或）指标独立性的严格要求，为多元参考值范围的临床运用拓展更大的空间。方法：采用该法将n个m维多指标数据转换为n个全息主成分值TZ，再针对n个一维TZ值应用正态近似法或百分位数法制定相应的参考值范围。并将单指标法、4种其它多元方法及该法对儿童发育资料应用结果作比较。结果：本法识别阳性率最接近5％的理论水平，为5.02％，检出假阳性率为适中，为0.85%，其假阴性率水平符合应用实际要求。结论：本法理论依据充分，并具有（1）不丢失任何信息，（2）既适用于正态分布资料，亦适用于偏态分布资料，（3）允许指标间存在任何程度的相关性等多个优点。因此值得推广使用。
Objective: A new method named full information principal component analysis (simplified as full information analysis) for constructing multivariate reference ranges is presented, by which the high frequency of false positive is reduced when several laboratory test results are interpreted by comparing each single test result with corresponding reference intervals. The requests of multivariate Gaussian distribution and /or independence of tests by general multivariate analysis is lowered. Methods: Reference range of m one-dimensional TZ, which is transformed from n m-dimensional data by the new method, is constructed using parametric or non-parametric method, and comparing between the results of single variable, 4 other multivariable approaches and the new method when applying to the children' s general physical examination data is also presented. Results: The new method has the highest efficiency in reducing false frequency(13.2%), the lowest difference between the actual and the expected positive probability(0.02%), a suitable false positive frequency by which the hidden abnormal data can be accurately found, and the discriminating frequency is in accordance with the actual rule. Conclusion: No demands on the distribution and the maximum tolerance to the correlation of variables，no information missed and easily applied are characterized by this method and the comparing also showed that it is superior to the others. So the new method is worth applying.
出处
《重庆医科大学学报》
CAS
CSCD
2001年第2期171-174,共4页
Journal of Chongqing Medical University
