Analyze Reported Results With Sources

 Long Analysis Plan Copied

No data was supplied, so I cannot execute the deconvolution itself. To run Weighted Least Squares (WLS) deconvolution, I need your concentration–time (or equivalent response) data, the unit impulse response function, and per-point variance estimates or weighting assumptions. If you upload these, the analysis can proceed.

What WLS deconvolution does (per the primary methodological source)

The canonical method models measured concentrations as the convolution of a drug input function with a unit impulse response (e.g., a two-exponential two-compartment disposition), represents the input as a polynomial, and solves for input coefficients via linear regression with Best Linear Unbiased Estimators. Weights derive from the error structure, and input rate/extent differences between products are tested within-subject via t-tests and F-tests (with discussion of unequal variances). Residual diagnostics include Durbin–Watson and runs tests; sampling should follow a Chebyshev design 1980 J. Pharm. Sci. WLS deconvolution method paper.

Reported strengths and limitations

• Recovering input rate and cumulative input without AUC extrapolation; it can reveal input differences masked by Cmax/tmax/AUC Avoids AUC extrapolation; detects input differences.
• Validation was simulation-only (three simulated subjects, two-compartment IV bolus plus first-order absorption with lag); real-data validation was not demonstrated in the source.
• Key assumptions: linear, time-invariant response; noninteracting primary input; correct weighting model; sensitivity to noise, normality, and sample size; the Gauss–Markov framework is only approximate when regressors are stochastic.
• Results depend on the unit impulse response; parameter non-uniqueness can exist without necessarily biasing the overall input estimate.

What to upload to run it

1. Time and concentration columns for each subject/occasion (test and reference products).
2. Impulse response data or parameters (e.g., IV bolus biexponential coefficients), or the data to fit it.
3. Variance model for weights (replicates, observed SD, or a variance function).

Confidence note: the methodological description above rests on a single, simulation-based 1980 source; treat real-data performance as unverified until your data are analyzed.