A total of 1000 subjects were simulated for each scenario. Samples were randomly collected at approximately 0.5, 2, 6, and/or 12 h. The administration of 100 mg doses every 12 h was simulated with 60% of subjects providing three samples and 40% of subjects providing four samples. In a second scenario (scenario 2), a sparse sampling scheme was assumed with dosing at steady state as might be seen in a clinical outpatient study.
In the first scenario (scenario 1), single doses in a rich sampling scheme were assumed with concentrations simulated at 0.5, 1, 1.5, 2, 3, 4, 6, 9, and 12 h after dosing. Sampling strategies were chosen to assure that most simulated concentrations would be neither above nor below LLOQ. Normally distributed body weights with a mean of 70 kg and a standard deviation of 10 kg were used for scaling CL (allometric exponent 3/4) and V (allometric exponent 1).Įach in silico subject received a dose of 100 mg. The residual error model (RUV) used in the simulation was a proportional error model with a coefficient of variation (CV) of 15%. Parameter values (and BSV) used for this simulation were the following: clearance (CL) 8.0 L/h/70 kg (20%), volume of distribution (V) 25.0 L/70 kg (25%), and absorption rate constant 1.5 h −1 (30%). Simulations were performed assuming a one‐compartment PK model with a depot compartment using mrgsolve package version 0.10.1.īetween‐subject variability (BSV) was assumed to be log‐normally distributed. It was not our intent to evaluate bias in decisions made based on plots using weighted residuals. Motivated us to conduct a simulation study to investigate the extent to which weighted residual calculations in subjects having some BLQ data might be biased when using the M3 method together with MDVRES. While using the M3 method, assigning MDVRES = 1 to censored BLQ data excludes the residuals from being calculated for these observations while allowing residuals to be computed for observations above LLOQ. BQL observations influence the fit through the censored likelihood, but these observations are not represented in the residual diagnostic plots.”Īlthough the concern for bias is real, recent NONMEM functionality, MDVRES (missing dependent variable for residual calculation),Īllows one to easily obtain previously suppressed weighted residuals for concentrations above the LLOQ in subjects that have at least one concentration reported as BLQ. A particularly lucid explanation was contributed by Matt Hutmacher, acknowledging that “residuals do not provide great diagnostic value unfortunately for data sets with censored data.
Tom Ludden provided an historical perspective that Stuart Beal intentionally excluded the calculation of weighted residuals for each subject with BLQ data due to a concern that all weighted residuals for that subject might be biased.
#Bug in nonmem software
It was suggested that this was a bug in the NONMEM software (ICON plc Development Solutions).
#Bug in nonmem archive
In a 2010 NONMEM Users Network Archive thread, However, by default, the M3 method suppresses the computation of the entire set of weighted residuals for any subject with at least one BLQ observation. It integrates the likelihood function over the interval and maximizes the likelihood of the concentration being BLQ with respect to model parameters. Although various approaches have been proposed to accommodate BLQ data,īeal's M3 method currently appears to be most common. Limited by the lower limit of quantification (LLOQ) of analytical techniques, it is not uncommon to have concentrations reported as below the LLOQ (BLQ) in PK studies. WRES and CWRES are commonly plotted against TIME and population predictions and are expected to be randomly scattered around zero with the bulk of the data points within two standard deviation units. They represent the difference between the observed concentration and the prediction under the model, which are then weighted to standardize and decorrelate the residuals. Weighted residuals, both traditional weighted residuals (WRES) and conditional weighted residuals (CWRES), are common metrics to graphically evaluate model acceptability in population analyses. Weighted‐residual bias in subjects with BLQ data was found to be small and probably ignorable in both intense and sparse sampling designs. A recent community discussion questioned potential bias in weighted residual plots when M3 is applied, and a simulation study was conducted to evaluate this bias. Concentration data below the limit of quantification (BLQ) are common in population pharmacokinetic (PK) analyses, and one method used to accommodate these during nonlinear mixed effects modeling is the M3 method.