Experimental extension: deterministic heat-injury prediction

This page is a freqTLS-only experimental extension built from synthetic data. It is not a replacement empirical case and does not estimate a repair process.

library(freqTLS)
sim <- simulate_tls(family = "beta_binomial", CTmax = 36, z = 4,
                    phi = 50, seed = 42)
# In this synthetic example, duration and trace time are both hours. No unit
# conversion occurs inside fit_tls() or predict_heat_injury().
fit <- fit_tls(sim, y = survived, n = total, time = duration, temp = temp,
               family = "beta_binomial", tref = 1, quiet = TRUE)
diagnose_tdt_fit(fit)
#> # A tibble: 1 × 9
#>   converged pd_hessian max_abs_gradient gradient_pass optimizer logLik n_params
#>   <lgl>     <lgl>                 <dbl> <lgl>         <chr>      <dbl>    <int>
#> 1 TRUE      TRUE              0.0000184 TRUE          nlminb     -129.        6
#> # ℹ 2 more variables: AIC <dbl>, all_pass <lgl>

trace <- data.frame(
  time = seq(0, 96, by = 0.25)
)
trace$temp <- 27 + 6 * sin(2 * pi * (trace$time - 6) / 24)

Here tref = 1 is one hour, and trace$time is also in hours. The trace time, fitted duration, and reference time must use the same unit because heat-injury damage is accumulated per time unit.

injury <- predict_heat_injury(fit, trace)
head(injury)
#>   time     temp         dose      injury  survival
#> 1 0.00 21.00000 0.000000e+00 0.000000000 0.9892557
#> 2 0.25 21.01285 3.627625e-05 0.003627625 0.9892557
#> 3 0.50 21.05133 7.282956e-05 0.007282956 0.9892557
#> 4 0.75 21.11529 1.102256e-04 0.011022563 0.9892557
#> 5 1.00 21.20445 1.490655e-04 0.014906545 0.9892557
#> 6 1.25 21.31842 1.900114e-04 0.019001136 0.9892557

plot_heat_injury(fit, trace, nboot = 50, seed = 42)

The trajectory is a deterministic transformation of the fitted 4PL and the user-supplied trace. heat_injury_envelope() can add parametric-bootstrap uncertainty from refitted curves. A repair rate supplied by the user remains an illustrative scenario: freqTLS has not fitted that rate from damage/recovery data. Use bayesTLS for its broader Bayesian heat-injury and repair workflow; its posterior intervals are distinct from freqTLS confidence intervals.