Some Statistical Background on SMART

SMART's sample size of 6000 patients was determined by several statistical assumptions, including:

• Primary analysis will be intent-to-treat, using stratified log-rank statistics (strata defined by participating sites).

• There will be a 20% difference in disease progression between the DC group and the VS group.

• Of all primary endpoints, disease progression events will be 70% and deaths will be 30%. This estimate is based on the CPCRA NvR study, which had a study population with more advanced disease at entry. Assuming that half of all deaths will be unrelated to HIV and to the treatment strategies being studied, then the expected treatment difference is reduced from 20% to 17%.
  Advertisement
• Based on the previous assumptions, 910 primary endpoints must be observed. Since the study is event-driven, the study will continue until 910 events occur; this will provide the necessary statistical power.

• Patients with HIV-1 infection and CD4 T cell counts >350 cells/mm³ will be enrolled. The 5-year cumulative event rate is estimated to be 10% and the 7-year cumulative event rate 15%, based on a comparison of early studies before HAART with studies of advanced patients taking protease inhibitors, as well as data from the EuroSIDA project (referenced in study protocol).

• An increasing hazard, with most events occurring late in the follow-up period, is assumed. The event rate per year would increase over 8 years as follows: 1, 1.5, 3, 7, 9, 9, 10, 10. Thus there would be a 10 times greater chance of an event happening in year 8 than in year 1.

• Each year, 2% of patients will be lost to follow-up.

• Patients will be enrolled over 2 years and followed for at least 6 years. Average follow-up will be 7 years and the range will be from 6 to 8 years. If more patients than planned are enrolled in the first 2 years, the average follow-up may be shorter (based on projected event rates). Likewise, if fewer patients are enrolled than planned, the follow-up may be longer.

Back to the RITA! Spring 2001 contents page.