Draw an interactive nomogram for a regression model

Draws the nomogram of a fitted regression model: one axis per predictor, scaled in points, a total points axis and one or more axes that turn the total into a prediction. In the widget every predictor has a handle. Dragging it, clicking a category or using the arrow keys sets a patient's values, and the points, the total and the prediction with its 95% confidence interval follow, computed in the page from the model's coefficients and their covariance.

Usage

ggnomogram(
  fit,
  data = NULL,
  values = NULL,
  labels = NULL,
  ranges = NULL,
  outcome = NULL,
  times = NULL,
  level = NULL,
  title = NULL,
  caption = NULL,
  family = "Lato"
)

Arguments

fit

A fitted regression model. See Details for the classes read.

data

The data the model was fitted to. Defaults to the data named in the model's call, which is enough when that data is still around.

values

Starting values for the predictors, as a named list. The others start at the median, or at the most common category.

labels

Axis labels for the predictors, as a named character vector. The others use the column's label attribute or its name.

ranges

Ranges for numeric axes, as a named list of pairs.

outcome

Label of the prediction axis. For a model with several, a character vector with one label each.

times

For a survival model, the times to predict survival at, named to label the axes, such as c("1 year" = 365, "5 years" = 1826). Defaults to the median follow-up time.

level

For a multinomial model, the category whose nomogram the static copy draws. Defaults to the first after the reference.

title, caption

Title above the nomogram and note below it.

family

Font family. The package ships Lato and registers it on load.

Value

An object of class ggnomogram, which prints as an interactive widget. Use graph_widget(), graph_plot() or graph_save() for the widget, a static ggplot or a file. nomogram_predict() gives the prediction for any set of values, as the widget computes it.

Details

Contributions are computed through the model's design matrix, so transformed and nonlinear terms, such as log(x), poly(), ns(), rcs() or a smooth from 'mgcv', are drawn as they were fitted. An axis whose effect rises and then falls is folded onto more than one line, as rms does. When predictors interact, the axis of a later predictor in the interaction is drawn for the current values of the earlier ones and redraws as they change. Numeric axes run from the 2.5th to the 97.5th percentile of the data unless ranges says otherwise.

The prediction depends on the model:

Linear models

lm(), nlme::gls(), quantreg::rq(), rms::ols() and Gaussian models with an identity link: the predicted mean, with a prediction interval for lm().

Generalized linear models

glm(), MASS::glm.nb(), geepack::geeglm(), logistf::logistf(), rms::lrm() and mgcv::gam(): the mean on the response scale, such as a probability or a rate, through the model's link.

Mixed models

lme4::lmer(), lme4::glmer(), nlme::lme() and glmmTMB::glmmTMB(): the prediction for a typical cluster, with the random effects at zero. On a nonlinear link this is the prediction for that cluster, not the average over the population. The interval reflects the fixed effects only. For glmmTMB, zero inflation and dispersion models are left out.

Cox models

survival::coxph() and rms::cph(): survival at each of times, with the interval survival::survfit() gives for the same patient. A stratified model gets a choice of stratum.

Parametric survival models

survival::survreg() and rms::psm(): the median survival time and survival at times.

Ordinal models

MASS::polr(), ordinal::clm(), rms::orm() and ordinal rms::lrm(): the probability of each category or above, with the probability of every category beside the plot.

Multinomial models

nnet::multinom(): one nomogram per category, scaled on its odds against the reference category, with the probability of every category beside the plot.

Examples

bw <- MASS::birthwt
bw$race <- factor(bw$race, labels = c("White", "Black", "Other"))
bw$smoke <- factor(bw$smoke, labels = c("No", "Yes"))
fit <- glm(low ~ age + lwt + race + smoke, family = binomial, data = bw)
n <- ggnomogram(fit, outcome = "Risk of low birth weight",
                labels = c(age = "Age (years)", lwt = "Weight (lb)"))
n
nomogram_predict(n, list(age = 30, lwt = 110, race = "Black", smoke = "Yes"))
#> $points
#> [1] 247.8336
#> 
#> $lp
#> [1] 0.5663902
#> 
#> $se
#> [1] 0.6063791
#> 
#> $outputs
#>                     output  estimate    lower     upper
#> 1 Risk of low birth weight 0.6379298 0.349306 0.8525662
#>