Nonlinear mixed-effects HIV dynamic models with considering left-censored measurements

Journal of Statistical Distributions and Applications

Table 2 Comparison of bias(%) on parameter estimates by three methods: Bayesian, Gaussian quadrature and NLME-SAEM methods

Case	Method	*logc*	*logδ*	*logϕ*	$d_{c}^{2}$	$d_{δ}^{2}$	$d_{ϕ}^{2}$	σ ²
A	Bayesian	16.1	24.2	2.1	49.6	47.7	65.1	543
	Gaussian quadrature	15.6	24	2.2	53.1	44.3	75.5	512
	NLME-SAEM	15.2	23.7	2.4	47.3	46.4	69.5	484
B	Bayesian	78	20	22	78	75	998	657
	Gaussian quadrature	72	25	16	94	61	932	586
	NLME-SAEM	71	22	19	90	68	970	490

In case A, the below detection limit data are hold out before fitting each model; In case B, the below detection limit data are retained but assumed 25 as the true value. The number of simulation run is 100.