Provides tools for assessing the shape of a dose-response curve by testing linearity and non-linearity at user-defined cut-offs. It also provides two methods of estimating a threshold dose, or the dose at which the dose-response function transitions to significantly increasing: bi-linear (based on pkg 'segmented') and smoothed with splines (based on pkg 'mgcv').
NEWS drsmooth 1.9.0
Version 1.9.0 introduces major additional functionality, changes several functions' parameters, and implements fixes as follows:
Supports data with dichotomous outcome using new
required parameter data_type = "continuous" or "dichotomous".
Outcome proportion is modeled in dichotomous data. Note: depending on data format, new function drsmooth::expand may be required to format summarized dichotomous data before executing drsmooth, see example.
Adds a Fisher's Exact Test to the available tests from the noel function, for use with dichotomous data.
Adds an optional output file of predicted values, which may be returned to the user inreal time or written to a .csv file.
Allows user specification of the basis of the spline dimensions from the default of 4 to 3.
MINOR FUNCTION CHANGES & BUG FIXES
Function 'smooth' is changed to 'drsmooth' to avoid conflict with stats::smooth. 'smooth' is deprecated. Please see drsmooth::drsmooth help for details.
Only the following functions are now exported, to come into closer alignment with package development best practices: 'prelimstats', 'noel', 'segment', 'nlbcd', nlaad', 'lbcd', 'expand', 'spline.plot', 'drsmooth'.
Functions 'noel', 'drsmooth', 'spline.plot', parameters have changed to include data_type. Please see drsmooth:: help for details.
Functions 'prelimstats', 'lbcd', 'nlaad', 'nlbcd', and 'segment' have not been modified to accommodate dichotmous data.
STD bias calculation in drsmooth may be turned off, yielding considerable savings in processing time.
Step intervals used in predicted data is now relative to dose range.
CURRENT ISSUES & FUTURE FUNCTIONALITY (v.2.0.0)
'drsmooth' calculations are based on the response to the lowest dose, even if the dose-response is negative. This is maximally conservative and should be interpreted carefully, as there is no reason to prefer any dose in a non-significant postive (or negative!) initial dose range. A more comprehensive solution, including some parameterization of absolute change in slope, is underway.
Include relevant preliminary statistics, additional noel tests (e.g. a test for trend in proportions) and cut-off dose functionality for dichtomous data.