Parametric assumptions equate to hidden observations: comparing the efficiency of nonparametric and parametric models for estimating time to AIDS or death in a cohort of HIV-positive women

Table 1 Estimated risks and model efficiency by the number of observations added to the sample

Approach	N	M	Risk^a	95% CLs^a	CL Difference^a	Average CL Difference^b
Nonparametric	1164	0	0.218	0.194, 0.242	0.048	0.047
Generalized Gamma	1164	0	0.210	0.189, 0.231	0.042	0.042
Nonparametric	1164	354	0.218	0.197, 0.239	0.042	0.041
Weibull	1164	0	0.200	0.180, 0.219	0.039	0.040
Nonparametric	1164	593	0.218	0.199, 0.238	0.039	0.038
Exponential	1164	0	0.152	0.141, 0.163	0.023	0.030
Nonparametric	1164	3960	0.218	0.207, 0.230	0.023	0.022

Abbreviations, N sample size, M pseudo-observations added, CL confidence limit
^aAt two years
^bAt all event times
Comparing the estimated two-year risk of AIDS or death, 95% CLs, and CL difference of all four models at the starting sample size (M = 0) and when the nonparametric model became as efficient as the parametric generalized gamma (M = 354), Weibull (M = 593), and exponential (M = 3960) models. The average 95% CL difference at all event times is also given

ISSN: 1471-2288