tablearn: Learning-curve analysis for sequential clinical or procedural...

In R4VN: Health Data Analysis and Publication-Ready Reporting

Learning-curve analysis for sequential clinical or procedural performance

Description

tablearn() analyzes performance across consecutive cases, procedures, or observations. It keeps individual case-level data for statistical inference while allowing visually stable rolling or block summaries for learning-curve display. A right-aligned rolling window such as cases 1-5, 2-6, 3-7, ... is particularly useful when the question is: "What is my current performance based on the most recent N cases?"

Usage

tablearn(
  outcome,
  data = NULL,
  order = NULL,
  operator = NULL,
  type = c("auto", "continuous", "binary", "count"),
  event = NULL,
  better = c("auto", "lower", "higher"),
  window = 5L,
  window_type = c("rolling", "block", "none", "cumulative"),
  window_align = c("right", "center", "left"),
  window_step = NULL,
  window_complete = TRUE,
  window_fun = c("auto", "mean", "median", "trimmed_mean", "proportion", "rate"),
  window_trim = 0.1,
  window_weight = c("equal", "linear", "exponential"),
  window_decay = 0.85,
  window_ci = TRUE,
  ci_level = 0.95,
  window_ci_method = c("auto", "t", "normal", "wilson", "exact"),
  interval = c("ci", "sd", "iqr", "range", "none"),
  ewma = FALSE,
  ewma_lambda = 0.2,
  ewma_init = c("first", "mean", "target"),
  smooth = TRUE,
  smooth_method = c("loess", "spline", "lm", "gam", "none"),
  smooth_on = c("window", "raw", "ewma"),
  smooth_span = 0.6,
  smooth_df = NULL,
  smooth_ci = TRUE,
  breakpoint = TRUE,
  break_on = c("raw", "window"),
  break_min_n = 8L,
  break_grid = NULL,
  break_boot = 0L,
  phase = TRUE,
  phase_breaks = NULL,
  phase_names = c("Learning", "Consolidation", "Proficiency"),
  phase_method = c("auto", "combined", "manual", "breakpoint", "proficiency"),
  proficiency = TRUE,
  proficiency_method = c("auto", "combined", "target", "lccusum", "breakpoint", "manual"),
  proficiency_case = NULL,
  proficiency_case_rule = c("confirmed", "first_stable"),
  target_hold = 3L,
  target_tolerance = 0,
  plateau = TRUE,
  plateau_ratio = 0.25,
  plateau_slope = NULL,
  stability = TRUE,
  stability_metric = c("auto", "sd", "iqr", "cv", "none"),
  stability_window = NULL,
  stability_ratio = 0.75,
  cusum = FALSE,
  target = NULL,
  cusum_method = c("deviation", "llr"),
  cusum_alt = NULL,
  cusum_reset = FALSE,
  cusum_limit = NULL,
  lccusum = FALSE,
  p_acceptable = NULL,
  p_unacceptable = NULL,
  alpha = 0.05,
  beta = 0.1,
  racusum = FALSE,
  expected = NULL,
  adjust = NULL,
  racusum_or = 2,
  racusum_limit = NULL,
  sensitivity = FALSE,
  window_sensitivity = NULL,
  plot = TRUE,
  plot_type = c("auto", "learning", "cusum", "both"),
  x_axis = c("case", "order"),
  operator_display = c("facet", "overlay"),
  show_raw = TRUE,
  show_window = TRUE,
  show_smooth = TRUE,
  show_ewma = TRUE,
  show_interval = TRUE,
  show_break = TRUE,
  show_proficiency = TRUE,
  show_phase = TRUE,
  show_target = TRUE,
  show_current = TRUE,
  title = NULL,
  subtitle = NULL,
  caption = NULL,
  xlab = NULL,
  ylab = NULL,
  theme = c("publication", "minimal", "classic", "bw", "gray"),
  legend = "bottom",
  plot_opts = NULL,
  lang = c("en", "vi"),
  digit = 2,
  p_digit = 3,
  interpretation = FALSE,
  viewer_plot_format = c("png", "svg"),
  console = FALSE,
  show = TRUE
)

Arguments

outcome	Outcome variable. May be a bare variable name, character name, or vars(...) containing multiple outcomes.
data	Data frame. If NULL, tablearn() attempts to use the R4VN active data frame.
order	Optional ordering variable such as case number or procedure date. If omitted, current row order is used. Data are sorted by this variable within each operator.
operator	Optional operator/surgeon/trainee variable. Curves and analyses are then calculated separately within operator.
type	Outcome type: "auto", "continuous", "binary", or "count". In auto mode, logical/factor/character variables with exactly two non-missing levels and numeric 0/1 variables are treated as binary. A numeric variable with two other observed values (for example, 50 and 100) is treated as continuous unless event is supplied or type = "binary" is requested. This conservative rule prevents a two-valued continuous measurement from being silently converted to a proportion.
event	Event level for a binary outcome. If omitted, the second factor level, TRUE, or the larger numeric level is used.
better	Direction of better performance: "lower", "higher", or "auto". Auto uses "lower" for numeric outcomes and for an event coded as failure/complication it should normally be set explicitly by the analyst.
window	Number of cases in a rolling or block window. Default 5.
window_type	"rolling" (1-5, 2-6, 3-7, ...), "block" (1-5, 6-10, ...), "cumulative" (1, 1-2, 1-3, ...), or "none".
window_align	Alignment for rolling windows: "right" (recommended and default for current-performance monitoring), "center", or "left".
window_step	Number of cases to advance each window. Default is 1 for rolling/cumulative and window for block summaries.
window_complete	If TRUE, only complete windows are retained. If FALSE, partial windows at the beginning/end are allowed.
window_fun	"auto", "mean", "median", "trimmed_mean", "proportion", or "rate". Auto uses a proportion for binary outcomes and mean otherwise.
window_trim	Trim proportion used by window_fun = "trimmed_mean".
window_weight	"equal", "linear", or "exponential". Equal weighting gives the ordinary rolling mean/proportion; alternatives give more weight to recent cases within each window.
window_decay	Decay in ⁠(0,1]⁠ for exponentially weighted windows.
window_ci	Show point-wise interval for each window.
ci_level	Confidence level, default 0.95.
window_ci_method	"auto", "t", "normal", "wilson", or "exact". Auto uses Wilson for binary proportions, t intervals for small continuous windows, and normal intervals for larger continuous windows.
interval	Window interval displayed/calculated: "ci", "sd", "iqr", "range", or "none".
ewma	Logical; calculate an exponentially weighted moving average.
ewma_lambda	EWMA smoothing parameter in ⁠(0,1]⁠; larger values react more strongly to the latest case.
ewma_init	EWMA starting value: "first", "mean", or "target".
smooth	Logical; add a fitted smooth trend.
smooth_method	"loess", "spline", "lm", "gam", or "none". GAM requires package mgcv; otherwise loess is used as fallback.
smooth_on	Data used only for the descriptive smooth: "window", "raw", or "ewma".
smooth_span	LOESS span.
smooth_df	Optional degrees of freedom for smoothing spline.
smooth_ci	Show a point-wise confidence band where the smoothing method supports one.
breakpoint	Logical; estimate a one-change-point piecewise regression.
break_on	"raw" (recommended default for inference) or "window". Overlapping rolling windows are correlated, so fitting the inferential model to raw cases avoids treating overlapping windows as independent observations.
break_min_n	Minimum observations required on each side of a candidate breakpoint.
break_grid	Optional numeric vector of candidate case positions.
break_boot	Number of bootstrap replications for an exploratory percentile CI for the breakpoint. 0 disables bootstrap.
phase	Logical; create a phase summary table using the estimated or user supplied phase breaks.
phase_breaks	Optional numeric vector of manual phase boundaries. This is useful for 3+ named phases; automatic estimation currently provides one main change point.
phase_names	Names of phases. Defaults include Learning, Consolidation, and Proficiency. Manual names are used when phase_breaks is supplied.
phase_method	Phase classification rule: "auto"/"combined" uses the detected learning change point together with the estimated proficiency case; "manual" uses only phase_breaks; "breakpoint" uses the change point; and "proficiency" divides the series at the proficiency case. Automatic classification never labels a phase Proficiency unless proficiency is actually estimated.
proficiency	Logical; assess whether proficiency has been reached.
proficiency_method	"auto", "combined", "target", "lccusum", "breakpoint", or "manual". The recommended "auto" resolves to a combined rule. A supplied target and an enabled LC-CUSUM are treated as required criteria; plateau/stability strengthen the evidence.
proficiency_case	Manual case number used only with proficiency_method = "manual".
proficiency_case_rule	For sustained-target proficiency, report the first qualifying window ("first_stable") or the window in which the required run is confirmed ("confirmed", default).
target_hold	Number of consecutive window estimates that must satisfy target before target-based proficiency is confirmed. Default 3.
target_tolerance	Non-negative tolerance around target. For better = "lower", values <= target + tolerance qualify; for better = "higher", values >= target - tolerance qualify.
plateau	Logical; assess whether the post-breakpoint slope is sufficiently small to be interpreted as an operational plateau. A breakpoint alone is not automatically called proficiency.
plateau_ratio	Relative plateau threshold when plateau_slope is NULL. The post-breakpoint absolute slope must be <= plateau_ratio times the initial absolute slope. Default 0.25.
plateau_slope	Optional absolute slope threshold overriding plateau_ratio. This is outcome-scale specific and can be useful when a clinically meaningful slope threshold is known.
stability	Logical; assess whether performance variability has fallen.
stability_metric	"auto", "sd", "iqr", "cv", or "none". Auto uses SD for continuous/count outcomes and does not use variability as proficiency evidence for binary outcomes.
stability_window	Number of early and late raw cases used to compare variability. Default is at least the selected learning-curve window size.
stability_ratio	Late/early variability ratio required for stability. Default 0.75 means late variability must be at most 75% of early variability.
cusum	Logical; calculate a conventional CUSUM from individual cases.
target	Clinical/quality target for CUSUM, target/reference line, and optional EWMA initialization.
cusum_method	"deviation" or binary likelihood-ratio "llr".
cusum_alt	Alternative binary failure probability for LLR-CUSUM.
cusum_reset	If TRUE, use a one-sided tabular CUSUM reset at zero.
cusum_limit	Optional decision limit. If omitted, CUSUM is descriptive.
lccusum	Logical; calculate a binary LC-CUSUM designed to signal evidence that an acceptable failure rate has been reached.
p_acceptable	Acceptable failure probability for LC-CUSUM.
p_unacceptable	Unacceptable failure probability for LC-CUSUM; must be larger than p_acceptable.
alpha, beta	Type-I and Type-II error probabilities used in the LC-CUSUM decision boundary formula.
racusum	Logical; calculate a binary risk-adjusted CUSUM using likelihood scores and individual expected risks.
expected	Optional expected-risk variable or numeric vector for RA-CUSUM. Supplying externally validated or pre-operative expected risks is preferable.
adjust	Optional covariates used to fit a logistic expected-risk model if expected is not supplied. May be character names or vars(...).
racusum_or	Odds ratio representing deterioration to be detected. Must be greater than 1.
racusum_limit	Optional RA-CUSUM decision limit.
sensitivity	Logical; calculate window-size sensitivity summaries.
window_sensitivity	Numeric vector of window sizes. If omitted and sensitivity = TRUE, sensible values are chosen from the sample size.
plot	Logical; create figures. Standard figures and Viewer figures use base R and therefore require no add-on package. When ggplot2 is installed, an advanced ggplot object is also retained in ⁠$plots⁠.
plot_type	"auto", "learning", "cusum", or "both". Auto shows the learning curve and also the CUSUM panel whenever conventional CUSUM, LC-CUSUM, or RA-CUSUM has been requested.
x_axis	"case" (default) or "order". Case number is usually preferable for learning curves; date/time can be shown using "order".
operator_display	"facet" or "overlay" when operator is supplied.
show_raw, show_window, show_smooth, show_ewma, show_interval, show_break, show_proficiency, show_phase, show_target, show_current	High-level plot layer switches.
title, subtitle, caption, xlab, ylab	Plot labels. Defaults are generated from the outcome and selected language.
theme	Plot theme: "publication", "minimal", "classic", "bw", or "gray".
legend	Legend position: "bottom", "top", "left", "right", or "none".
plot_opts	Nested list for advanced plot customization. See the dedicated Plot options section below. Values supplied here override defaults.
lang	"en" or "vi" for generated labels/messages.
digit	Number of decimals for ordinary estimates in Viewer tables.
p_digit	Number of decimals for p-values in Viewer tables.
interpretation	Logical; include a short interpretation section in the Viewer/console report. Default is FALSE; the interpretation text is still retained in ⁠$interpretation⁠ for programmatic use.
viewer_plot_format	Self-contained Viewer image format: "png" or "svg". Both are produced with base R graphics and need no extra package.
console	Print a concise analysis summary to the console. Default FALSE.
show	Open the complete HTML report in the RStudio Viewer (or browser) and display requested figures in the Plot pane. Default TRUE.

Value

An object of class r4vn_tablearn. For one outcome it contains at least raw, window, ewma, smooth, breakpoint, proficiency, proficiency_evidence, phases, phase_classification, current, current_status, cusum, lccusum, racusum, sensitivity, plots, settings, plot_options, tables, and interpretation. With show = TRUE, the returned object also carries the generated self-contained Viewer HTML/file path. Multiple outcomes return class r4vn_tablearn_multi containing one analysis per outcome.

Rolling-window interpretation

With window = 5, window_type = "rolling", window_align = "right", and window_step = 1, the first displayed point summarizes cases 1-5, the next summarizes 2-6, then 3-7, and so on. Therefore the point at case 100 represents performance in the most recent five cases (96-100). A newly observed case 101 updates the curve to cases 97-101. This is different from non-overlapping block summaries and from a cumulative mean.

Statistical inference versus visual smoothing

Overlapping rolling windows share observations and are therefore correlated. tablearn() can display rolling windows for a stable curve while fitting the change-point model and CUSUM on the original case sequence. The recommended default is break_on = "raw"; CUSUM, LC-CUSUM and RA-CUSUM always operate on individual sequential cases in this implementation.

Advanced plot options

plot_opts is a nested list. Every field is optional. Main groups are:

• raw: show, color, fill, shape, size, alpha, stroke.

• window: show, geom ("line", "point", "point_line"), color, fill, shape, size, alpha, line_color, line_width, line_type.

• window_interval: show, geom ("ribbon" or "errorbar"), color, fill, alpha, line_width, width.

• smooth: show, color, fill, line_width, line_type, alpha, ci_show, ci_alpha.

• ewma: show, color, line_width, line_type, alpha.

• breakpoint: show, color, line_width, line_type, alpha, label, label_text, label_size, label_angle, label_hjust, label_vjust.

• proficiency: independent proficiency-line controls: show, color, line_width, line_type, alpha, label, label_text, label_size, label_angle, label_hjust, label_vjust.

• phase: show, fills, alpha, border_color, border_width, label, label_size, label_position.

• target: show, color, line_width, line_type, alpha, label, label_text, label_size.

• current: show, color, fill, shape, size, alpha, stroke, label, label_text, label_size, hjust, vjust.

• axes: x_limits, y_limits, x_breaks, y_breaks, x_expand, y_expand, y_percent, percent_accuracy, x_reverse, y_reverse, x_trans, y_trans, x_date_format, x_date_breaks, clip.

• legend: show, position, title, direction.

• facet: ncol, nrow, scales.

• operator: colors, shapes, line_types; layout is selected by the high-level operator_display argument.

• cusum: CUSUM-figure controls including colors, line_types, line_width, alpha, point controls, zero-line controls, decision-limit controls, signal marker controls, title/subtitle/caption/axis labels, and CUSUM-specific axis limits/breaks.

• theme: name, base_size, base_family, grid_major, grid_minor, panel_border, axis_line, plot_title_face, legend_key_size.

• text: optional title/subtitle/caption/axis/legend text sizes.

• margins: top, right, bottom, left in points.

• panel: optional background/border customization.

• annotation: show_n, show_window_label.

This layered design lets the raw observations remain visible while the rolling curve, uncertainty, fitted smooth, target, breakpoint, estimated proficiency, phases, and current performance are styled independently.

Proficiency and phase classification

tablearn() deliberately distinguishes a statistical/descriptive change point from proficiency. A change point indicates a change in the learning trajectory; proficiency is estimated from one or more operational criteria. With the default combined rule, a supplied clinical target must be sustained for target_hold consecutive displayed windows, and an enabled LC-CUSUM must cross its competency decision boundary. A post-change plateau and reduced variability strengthen the evidence. If no target or LC-CUSUM is supplied, a clear plateau can provide a limited, data-driven proficiency estimate. Manual proficiency is also supported.

Automatic phase classification uses these results rather than forcing every dataset into three phases. When both an earlier learning change point and a later proficiency case are found, phases are Learning -> Consolidation -> Proficiency. If proficiency is not established, a post-change segment is labelled Consolidation rather than Proficiency. phase_breaks always allows complete manual control for study protocols with pre-specified phases.

The evidence label (Strong, Moderate, Limited) is an R4VN rule-based summary of concordant criteria, not a confidence probability and not a substitute for a clinically defined competency standard.

Viewer and dependency policy

With show = TRUE (default), tablearn() opens one self-contained HTML report containing the key publication-ready tables and every requested figure. The same figures are also sent to the Plot pane. Standard analysis, HTML rendering, learning curves, CUSUM, LC-CUSUM, and RA-CUSUM use only base/recommended R packages. ggplot2 is optional: when already installed, an advanced ggplot object is retained in ⁠$plots⁠; when it is absent, plotting still works through the base-R fallback. mgcv is needed only when the user explicitly selects smooth_method = "gam"; otherwise the default smoothing methods use base R.

Examples

# --------------------------------------------------------------------------
# Reproducible demonstration data used by the examples below
# --------------------------------------------------------------------------
set.seed(2026)
n <- 150
d <- data.frame(
  case = 1:n,
  date = as.Date("2025-01-01") + 0:(n - 1),
  surgeon = rep(c("A", "B", "C"), each = n / 3),
  complexity = rbinom(n, 1, 0.35),
  age = round(rnorm(n, 58, 12), 1)
)
d$time <- 115 - 48 * (1 - exp(-d$case / 30)) +
  9 * d$complexity + rnorm(n, 0, 9)
d$score <- 55 + 28 * (1 - exp(-d$case / 35)) + rnorm(n, 0, 5)
d$expected_risk <- plogis(-1.8 + 0.9 * d$complexity + 0.012 * (d$age - 58))
actual_risk <- plogis(qlogis(d$expected_risk) - 0.010 * d$case)
d$complication <- rbinom(n, 1, actual_risk)
d$errors <- rpois(n, pmax(0.15, 3.2 * exp(-d$case / 45)))

# The full catalogue is interactive so R CMD check stays fast.
if (interactive()) {

# 1. Simplest end-user command. In an interactive session this opens the
# complete Viewer report and sends the learning curve to the Plot pane.
if (interactive()) {
  m1 <- tablearn(time, data = d, order = case)
}

# 2. Right-aligned rolling window: 1-5, 2-6, 3-7, ...
m2 <- tablearn(time, data = d, order = case, window = 5,
               show = FALSE, plot = FALSE)
head(m2$tables$Window_performance)
m2$tables$Current_performance

# 3. Non-overlapping blocks: 1-10, 11-20, 21-30, ...
m3 <- tablearn(time, data = d, order = case, window = 10,
               window_type = "block", show = FALSE, plot = FALSE)
head(m3$window[, c(".start", ".end", ".value")])

# 4. Cumulative learning curve: 1, 1-2, 1-3, ...
m4 <- tablearn(time, data = d, order = case,
               window_type = "cumulative", show = FALSE, plot = FALSE)

# 5. Individual-case series without aggregation.
m5 <- tablearn(time, data = d, order = case,
               window_type = "none", show = FALSE, plot = FALSE)

# 6. Median and IQR for a skewed continuous outcome.
m6 <- tablearn(time, data = d, order = case, window = 7,
               window_fun = "median", interval = "iqr",
               show = FALSE, plot = FALSE)

# 7. Give recent cases greater weight within the rolling window.
m7 <- tablearn(time, data = d, order = case, window = 10,
               window_weight = "exponential", window_decay = 0.85,
               show = FALSE, plot = FALSE)

# 8. Add EWMA to the ordinary rolling curve.
m8 <- tablearn(time, data = d, order = case, window = 10,
               ewma = TRUE, ewma_lambda = 0.20,
               show = FALSE, plot = FALSE)
tail(m8$ewma)

# 9. Automatic piecewise change point. Inference uses raw cases by default.
m9 <- tablearn(time, data = d, order = case, window = 5,
               breakpoint = TRUE, break_on = "raw",
               show = FALSE, plot = FALSE)
m9$tables$Change_point

# 10. Sustained clinical target: <= 70 for 3 consecutive windows.
m10 <- tablearn(time, data = d, order = case, window = 5,
                better = "lower", target = 70, target_hold = 3,
                proficiency_method = "target",
                show = FALSE, plot = FALSE)
m10$tables$Proficiency
m10$tables$Proficiency_evidence

# 11. Report the first qualifying window rather than the confirmation window.
m11 <- tablearn(time, data = d, order = case, window = 5,
                better = "lower", target = 70, target_hold = 3,
                proficiency_method = "target",
                proficiency_case_rule = "first_stable",
                show = FALSE, plot = FALSE)

# 12. Manual proficiency and prespecified study phases.
m12 <- tablearn(time, data = d, order = case, window = 5,
                proficiency_method = "manual", proficiency_case = 60,
                phase_method = "manual", phase_breaks = c(25, 59),
                phase_names = c("Learning", "Consolidation", "Proficiency"),
                show = FALSE, plot = FALSE)
m12$tables$Phase_classification

# 13. Binary outcome. Numeric 0/1 is recognized automatically; event = 1 is
# explicit and makes the scientific meaning clear.
m13 <- tablearn(complication, data = d, order = case, event = 1,
                better = "lower", window = 20,
                window_ci_method = "wilson",
                show = FALSE, plot = FALSE)
m13$tables$Current_performance

# 14. Count outcome.
m14 <- tablearn(errors, data = d, order = case, type = "count",
                better = "lower", window = 10,
                show = FALSE, plot = FALSE)

# 15. Higher values can represent better performance.
m15 <- tablearn(score, data = d, order = case, better = "higher",
                target = 80, target_hold = 3,
                show = FALSE, plot = FALSE)

# 16. Conventional deviation CUSUM for a continuous outcome.
m16 <- tablearn(time, data = d, order = case, target = 75,
                better = "lower", cusum = TRUE,
                plot_type = "auto", show = FALSE, plot = FALSE)
m16$tables$Sequential_monitoring

# 17. Binary likelihood-ratio CUSUM.
m17 <- tablearn(complication, data = d, order = case, event = 1,
                better = "lower", target = 0.10,
                cusum = TRUE, cusum_method = "llr", cusum_alt = 0.20,
                show = FALSE, plot = FALSE)

# 18. LC-CUSUM: evidence that an acceptable failure rate has been reached.
m18 <- tablearn(complication, data = d, order = case, event = 1,
                better = "lower", window = 20,
                lccusum = TRUE, p_acceptable = 0.10,
                p_unacceptable = 0.25,
                proficiency_method = "lccusum",
                show = FALSE, plot = FALSE)
m18$tables$Proficiency

# 19. RA-CUSUM with externally supplied case-specific expected risks.
m19 <- tablearn(complication, data = d, order = case, event = 1,
                better = "lower", window = 20,
                racusum = TRUE, expected = expected_risk,
                racusum_or = 2,
                show = FALSE, plot = FALSE)

# 20. RA-CUSUM can estimate expected risk from covariates using base glm().
# For prospective monitoring, an external/pre-specified risk model is preferred.
m20 <- suppressWarnings(tablearn(
  complication, data = d, order = case, event = 1,
  racusum = TRUE, adjust = vars(complexity, c.age), racusum_or = 2,
  show = FALSE, plot = FALSE
))

# 21. Separate learning curves by operator/surgeon.
m21 <- tablearn(time, data = d, order = case, operator = surgeon,
                window = 8, operator_display = "facet",
                show = FALSE, plot = FALSE)
m21$tables$Current_performance

# 22. Overlay operators in the same graph.
m22 <- tablearn(time, data = d, order = case, operator = surgeon,
                window = 8, operator_display = "overlay",
                show = FALSE, plot = FALSE)

# 23. Window-size sensitivity analysis.
m23 <- tablearn(time, data = d, order = case, window = 5,
                sensitivity = TRUE,
                window_sensitivity = c(3, 5, 10, 20),
                show = FALSE, plot = FALSE)
m23$tables$Window_sensitivity

# 24. Multiple outcomes in one command.
m24 <- tablearn(vars(time, complication), data = d, order = case,
                event = 1, window = 10,
                show = FALSE, plot = FALSE)
names(m24$outcomes)

# 25. Use procedure date on the x-axis instead of consecutive case number.
m25 <- tablearn(time, data = d, order = date, x_axis = "order",
                window = 7, show = FALSE, plot = FALSE)

# 26. Interpretive prose is opt-in; default is FALSE.
m26 <- tablearn(time, data = d, order = case,
                interpretation = TRUE, show = FALSE, plot = FALSE)
m26$interpretation

# 27. Vietnamese generated interpretation/labels.
m27 <- tablearn(time, data = d, order = case, lang = "vi",
                interpretation = TRUE, show = FALSE, plot = FALSE)

# 28. Conservative auto-detection: a numeric variable with two values other
# than 0/1 remains continuous unless event/type explicitly says binary.
d2 <- data.frame(case = 1:20, value = c(rep(100, 10), rep(50, 10)))
m28 <- tablearn(value, data = d2, order = case,
                show = FALSE, plot = FALSE)
m28$settings$type

# 29. All publication-ready tables are directly accessible.
names(m10$tables)
summary(m10)$tables

# 30. Plot methods work even when ggplot2 is not installed because tablearn()
# has a base-R plotting fallback. These also appear inside the Viewer report.
if (interactive()) {
  plot(m10, type = "learning")
  m30 <- tablearn(complication, data = d, order = case, event = 1,
                  lccusum = TRUE, p_acceptable = .10,
                  p_unacceptable = .25, plot_type = "both")
  plot(m30, type = "cusum")
}

# 31. High-level publication styling. Advanced ggplot styling is used when
# ggplot2 is installed; the Viewer/base-R figure remains available otherwise.
if (interactive()) {
  m31 <- tablearn(
    time, data = d, order = case, window = 5, target = 70,
    plot_opts = list(
      raw = list(alpha = .15, size = 1.0),
      window = list(color = "#1F5A94", line_width = 1.2),
      smooth = list(color = "#B23A48", line_width = 1.4),
      proficiency = list(color = "#00796B"),
      phase = list(alpha = .08),
      current = list(fill = "#FFD166"),
      theme = list(base_size = 12, grid_minor = FALSE)
    )
  )
}

} # end full interactive example catalogue

R4VN documentation built on Sept. 30, 2026, 5:13 p.m.