Clustering of HIV-1 infected haemophiliacs by time-series data of CD4+and CD8+cell numbers and comparison of blood viral load in each cluster

We applied three-way data clustering to the numbers of CD4+ and CD8+ cells in the interval from 1986 to 1992 observed in Japanese haemophiliacs infect ed with HIV-1. A total of 131 haemophiliacs were classified into four duste rs. On the basis of clinical status at the end of October 1997, the cumulat ive fraction of patients with AIDS was the highest among the 36 haemophilia cs in Cluster 1. On the other hand, there was a smaller fraction in Cluster 2, consisting of 49 haemophiliacs, and a much smaller fraction in Clusters 3 and 4, which included 33 and 13 haemophiliacs respectively. However, a s ignificantly lower blood viral copy number was found in Cluster 3 than in C luster 4. There was no significant difference in age at the first term of 1 986 among the four clusters. Thus, further investigation is necessary to de tect background factors, other than age, that resulted in the present class ification. As demonstrated in this analysis, three-way data clustering will result in efficient data mining in the handling of various clinical data. Med Sci Res 27:295-298 (C) 1999 Lippincott Williams & Wilkins.