---
title: "Half-Life Calculation"
author: "Bill Denney"
output:
  rmarkdown::html_vignette:
    toc: yes
    toc_depth: 6
vignette: >
  %\VignetteIndexEntry{Half-Life Calculation}
  %\VignetteEngine{knitr::rmarkdown}
  %\VignetteEncoding{UTF-8}
---

```{r setup, echo=FALSE, include=FALSE}
library(PKNCA)
library(dplyr)
```

# Introduction

Half-life is calculated by fitting the natural logarithm of concentration by time.  The default calculation method is curve stripping (described in more detail below).  Manual half-life points with no automated half-life selection can be performed, or specific points can be excluded while still performing curve stripping.

# Curve Stripping Method

When automatic point selection is performed for curve stripping, the algorithm described below is used.

## Select the Points

All sets of points that are applicable according to the current options are selected.

* Drop all BLQ values, then
* Choose all sets of points that start from the $T_{last}$ and step back:
    * at least `r PKNCA.options("min.hl.points")` points (customizable with `PKNCA.options("min.hl.points")`)
    * `r c("Not including", "Including")[PKNCA.options("allow.tmax.in.half.life") + 1]` $T_{max}$ (customizable with `PKNCA.options("allow.tmax.in.half.life")`)

As a specific example, if measurements were at 0, 1, 2, 3, 4, 6, 8, 12, and 24 hours; if $T_{last}$ is 12 hours; and if $T_{max}$ is 1 hour then the default point sets that would be fit are:

1. 6, 8, and 12 hours;
2. 4, 6, 8, and 12 hours;
3. 3, 4, 6, 8, and 12 hours; and
4. 2, 3, 4, 6, 8, and 12 hours.

If `PKNCA.options("min.hl.points")` were set to `4`, then the 6, 8, and 12 hour set would not be fit.  If `PKNCA.options("allow.tmax.in.half.life")` were set to `TRUE`, then 1, 2, 3, 4, 6, 8, and 12 hours would be fit.

## Select the Best Fit

After fitting all points, the best fit among the set of possible fit is selected by the following rules:

1. $\lambda_z > 0$ and at the same time the maximum r-squared must be within an adjusted $r^2$ factor of the best.
    1. The adjusted $r^2$ factor is controlled by `PKNCA.options("adj.r.squared.factor")` and it defaults to `r PKNCA.options("adj.r.squared.factor")`.
    2. These rules must be met simultaneously, so if the maximum adjusted $r^2$ is for a line with $\lambda_z \leq 0$, the half-life may end up being unreportable.
2. If fitting the log-linear concentration-time line fails, then it is not the best line.
3. If more than one fit still meets the criteria above, then choose the fit with the most points included.

```{r check-ggplot, include=!requireNamespace("dplyr"), results="asis"}
cat("dplyr is required for this vignette to work correctly.  Please install the dplyr library and retry building the vignette.")
```

## Example

```{r normal-example}
# Perform calculations for subject 1, only
data_conc <- as.data.frame(datasets::Theoph)[datasets::Theoph$Subject == 1, ]

# Keep all points
conc_obj <-
  PKNCAconc(
    data_conc,
    conc~Time|Subject
  )

# Only calculate half-life and parameters required for half-life
current_intervals <- data.frame(start=0, end=Inf, half.life=TRUE)
data_obj <- PKNCAdata(conc_obj, intervals=current_intervals)
result_obj <- pk.nca(data_obj)

# Extract the results for subject 1 
as.data.frame(result_obj)
```


# Manual Point Selection

## Exclusion of Specific Points with Curve Stripping

In some cases, specific points will be known outliers, or there may be another reason to exclude specific points.  And, with those points excluded, the half-life should be calculated using the normal curve stripping methods described above.

To exclude specific points but otherwise use curve stripping, use the `exclude_half.life` option as the column name in the concentration dataset for `PKNCAconc()` as illustrated below.

```{r exclude-points}
data_conc$exclude_hl <- data_conc$Time == 12.12
# Confirm that we will be excluding exactly one point
stopifnot(sum(data_conc$exclude_hl) == 1)

# Drop one point
conc_obj_exclude1 <-
  PKNCAconc(
    data_conc,
    conc~Time|Subject,
    exclude_half.life="exclude_hl"
  )

data_obj_exclude1 <- PKNCAdata(conc_obj_exclude1, intervals=current_intervals)

# Perform the calculations
result_obj_exclude1 <- pk.nca(data_obj_exclude1)

# Results differ when excluding the 12-hour point for subject 1 (compare to
# example in the previous section)
as.data.frame(result_obj_exclude1)
```

## Specification of the Exact Points for Analysis

In other cases, the exact points to use for half-life calculation are known, and automatic point selection with curve stripping should not be performed.

To exclude specific points but otherwise use curve stripping, use the `include_half.life` option as the column name in the concentration dataset for `PKNCAconc()` as illustrated below.

```{r include-points}
data_conc$include_hl <- data_conc$Time > 3
# Confirm that we will be excluding exactly one point
stopifnot(sum(data_conc$include_hl) == 6)

# Drop one point
conc_obj_include6 <-
  PKNCAconc(
    data_conc,
    conc~Time|Subject,
    include_half.life="include_hl"
  )

data_obj_include6 <- PKNCAdata(conc_obj_include6, intervals=current_intervals)

# Perform the calculations
result_obj_include6 <- pk.nca(data_obj_include6)

# Results differ when including 6 points (compare to example in the previous
# section)
as.data.frame(result_obj_include6)
```