drugDemand-package: Drug Demand Forecasting

In drugDemand: Drug Demand Forecasting

Drug Demand Forecasting

Description

Performs drug demand forecasting by modeling drug dispensing data while taking into account predicted enrollment and treatment discontinuation dates. The gap time between randomization and the first drug dispensing visit is modeled using interval-censored exponential, Weibull, log-logistic, or log-normal distributions (Anderson-Bergman (2017) \Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v081.i12")}). The number of skipped visits is modeled using Poisson, zero-inflated Poisson, or negative binomial distributions (Zeileis, Kleiber & Jackman (2008) \Sexpr[results=rd]{tools:::Rd_expr_doi("10.18637/jss.v027.i08")}). The gap time between two consecutive drug dispensing visits given the number of skipped visits is modeled using linear regression based on least squares or least absolute deviations (Birkes & Dodge (1993, ISBN:0-471-56881-3)). The number of dispensed doses is modeled using linear or linear mixed-effects models (McCulloch & Searle (2001, ISBN:0-471-19364-X)).

Details

In clinical trials, patients do not always follow protocol-specified visit and drug dispensing schedules. Patients may encounter delays in their drug dispensing appointments, skip visits altogether, or receive doses different from the protocol-specified target. Relying solely on protocol-based predictions tends to result in an overestimation of drug demand. Consequently, we propose a method that models observed drug dispensing data, thereby accounting for these deviations.

• k0: The number of skipped visits between randomization and the first drug dispensing visit.

• t0: The gap time between randomization and the first drug dispensing visit when there is no visit skipping.

• t1: The gap time between randomization and the first drug dispensing visit when there is visit skipping.

• ki: The number of skipped visits between two consecutive drug dispensing visits.

• ti: The gap time between two consecutive drug dispensing visits.

• di: The dispensed doses at drug dispensing visits.

For k0 and ki, we explore several modeling options, including constant, Poisson, zero-inflated Poisson (ZIP), and negative binomial distributions.

For t0, we consider various models such as constant, exponential, Weibull, log-logistic, and log-normal.

For t1 (given k0) and ti (given ki), we apply linear regression models using least squares or least absolute deviations.

For di, we evaluate constant, linear, and linear mixed-effects models with subject random effects.

Once the dosing models are fitted to the observed drug dispensing data, we draw model parameters from their approximate posterior distributions. Subsequently, we simulate drug dispensing data after cutoff for both ongoing and new patients.

Finally, we estimate the dose to dispense based on the simulated data.

Author(s)

Kaifeng Lu, [e-mail]

References

Clifford Anderson-Bergman. icenReg: Regression Models for Interval Censored Data in R. J Stat Softw. 2017, Volume 81, Issue 12.

Achim Zeileis, Christian Kleiber, and Simon Jackman. Regression models for count data in R. J Stat Softw. 2008, Volume 27, Issue 8.

David Birkes and Yadolah Dodge. Alternative Methods of Regression. John Wiley & Sons: New York, 1993.

Charles E. McCulloch and Shayler R. Searle. Generalized, Linear, and Mixed Models. John Wiley & Sons: New York, 2001.

drugDemand documentation built on May 29, 2024, 8:43 a.m.