The template catalog

What is here

Every analysis job an HVTI study runs starts from a template: a Quarto file with the study-specific lines marked EDIT:. This page lists them all, what each one does, and how close each is to being delivered. It is built from the package’s own catalog when the page is rendered, so a template added, renamed or shipped shows up here without anyone remembering to update the page.

A template becomes a job in your study with add_job(). The first argument is the template’s prefix, then the subject and type that name the job’s set, and the qualifier when the prefix has one:

add_job("dc", "cohort", "eda", qualifier = "trends")

In a new study that writes 10_descriptive/dc.trends.cohort.eda.qmd; a legacy study that already uses bare folder names gets descriptive/dc.trends.cohort.eda.qmd. The job refuses to overwrite an existing file, because a job accumulates a study’s edits.

Delivery

Each template carries a light.

  • 🟢 Shipped (34): on disk and supported. Scaffold it and use it.
  • 🟡 In progress (27): on disk but being reworked, or scheduled into a delivery batch with nothing blocking it.
  • 🔴 Not yet on the way (11): waiting on a function in another package, or not yet scheduled. The job type is known and counted; for now, write it by hand.

Available now

Every template on disk, grouped by the study folder a job lands in, named without its digits (graphs is 40_graphs in a new study). add_job() places each one; you do not choose the folder.

datasets

Delivery Template Description Scaffold
🟡 In progress bd Cuts a study’s dataset from a master snapshot: cohort list, exclusions with attrition and derived variables, then the working file and its build record, registered as a dated version of the study dataset. add_job("bd", subject, type)

descriptive

Delivery Template Description Scaffold
🟢 Shipped dc-eda The whole data-checking report in one render: every column’s contents, goodness of follow-up, then continuous variables and categorical variables as percentages and counts, each section with its table. add_job("dc", subject, type, qualifier = "eda")
🟢 Shipped dc-general First look at a built cohort: what the data contain, how each categorical variable breaks down, how each continuous variable is distributed, and which variables move together. add_job("dc", subject, type, qualifier = "general")
🟢 Shipped dc-gfup Checks follow-up before any time-related analysis: counts for the cohort, the event and censored subsets, and the missing, negative and zero intervals, then the goodness-of-follow-up figure, each patient’s follow-up against operation year under the diagonal the close date allows. add_job("dc", subject, type, qualifier = "gfup")
🟢 Shipped dc-stddiff The balance table: the standardized difference of each baseline variable between two groups, unadjusted and, when the study has them, in the matched set and under matching weights, with an optional permutation reference. add_job("dc", subject, type, qualifier = "stddiff")
🟢 Shipped dc-tables The formatted descriptive table: every reported variable under its section heading, categorical as n (%) and continuous as the summary you choose, written to Word. add_job("dc", subject, type, qualifier = "tables")
🟢 Shipped dc-trends How the cohort changed over the years of operation: the share with a characteristic, or the level of a measurement, year by year with a smooth. add_job("dc", subject, type, qualifier = "trends")
🟢 Shipped lg Calibrates a continuous or ordinal covariate against an event: it groups the covariate, takes each group’s event rate on a logit or constant-hazard scale, and plots it against six candidate scales of the covariate, so the analyst can pick the straightest and save the term for an lm job’s PREDICTORS. add_job("lg", subject, type)

distributions

Delivery Template Description Scaffold
🟢 Shipped ac Nonparametric life tables by stratum, with estimates at the reporting horizons. Figures over them are an hp job. add_job("ac", subject, type)
🟢 Shipped hz Fits and saves the multiphase parametric hazard model whose shape the later hm, hp and hs jobs read. add_job("hz", subject, type)

analyses

Delivery Template Description Scaffold
🟢 Shipped bc Bootstrap variable-selection screen over a Cox proportional hazards model. add_job("bc", subject, type)
🟢 Shipped bh Bootstrap variable-selection screen for a hazard model: how often each candidate survives, and which to carry into hm. It screens; it does not fit the final model. add_job("bh", subject, type)
🟢 Shipped bl Bootstrap variable-selection screen over a logistic model. add_job("bl", subject, type)
🟢 Shipped br Bootstrap variable-selection screen over a linear model. add_job("br", subject, type)
🟢 Shipped hm Multivariable risk-factor model built on the shape parameters the hz job fitted. add_job("hm", subject, type)
🟢 Shipped lm-balancing_count Poisson or negative-binomial balancing score for a count exposure, with the distribution as a reviewed study choice. add_job("lm", subject, type, qualifier = "balancing_count")
🟢 Shipped lm-binary Binary outcome logistic model, saving the fitted model, predictions and pooled inference across imputations. add_job("lm", subject, type, qualifier = "binary")
🟢 Shipped lm-checkpred Applies a saved lm-binary model to a validation cohort. It never refits the source model. add_job("lm", subject, type, qualifier = "checkpred")
🟢 Shipped lm-nominal Generalized-logit model for a nominal outcome with a declared level set and reference. add_job("lm", subject, type, qualifier = "nominal")
🟢 Shipped lm-ordinal Proportional-odds model for an ordered outcome whose full order is declared, not inferred. add_job("lm", subject, type, qualifier = "ordinal")
🟢 Shipped lm-propensity_binary Propensity model for a binary treatment, scoring the columns that matching and weighting jobs use. add_job("lm", subject, type, qualifier = "propensity_binary")
🟢 Shipped lm-propensity_nominal Generalized-logit propensity model for a nominal treatment with an explicit reference level. add_job("lm", subject, type, qualifier = "propensity_nominal")
🟢 Shipped lm-propensity_ordinal Proportional-odds propensity model for an ordered treatment. add_job("lm", subject, type, qualifier = "propensity_ordinal")
🟢 Shipped nb-boostmtree Fits boosted multivariate trees to a response measured repeatedly over follow-up, and reports its error path, fit over time, importance, partial effects and patient trajectories. add_job("nb", subject, type, qualifier = "boostmtree")
🟢 Shipped rfc-explain Explains a saved classification forest: variable importance and dependence plots. add_job("rfc", subject, type, qualifier = "explain")
🟢 Shipped rfc-fit Grows a classification forest and checks it: out-of-bag error and the ROC curve with its AUC. add_job("rfc", subject, type, qualifier = "fit")
🟢 Shipped rfr-explain Explains a saved regression forest: variable importance and dependence plots. add_job("rfr", subject, type, qualifier = "explain")
🟢 Shipped rfr-fit Grows a regression forest and checks it: out-of-bag error and predicted against observed. add_job("rfr", subject, type, qualifier = "fit")
🟢 Shipped rfs-explain Explains a saved survival forest: variable importance and dependence plots. It never grows a forest. add_job("rfs", subject, type, qualifier = "explain")
🟢 Shipped rfs-fit Grows a random survival forest and checks it learned something: out-of-bag error, predicted survival and the Brier score. add_job("rfs", subject, type, qualifier = "fit")
🟢 Shipped rm-balancing_continuous Least-squares balancing score for a continuous exposure: the averaged fitted value and its strata. add_job("rm", subject, type, qualifier = "balancing_continuous")
🟢 Shipped rm-linear Ordinary least-squares outcome model with Rubin-pooled inference. add_job("rm", subject, type, qualifier = "linear")
🟡 In progress sid-clusters PAM on a sid-fit dissimilarity over a range of K: metrics for choosing K, letters stable across K, a Sankey, and the assignments registered for later jobs to join. add_job("sid", subject, type, qualifier = "clusters")
🟡 In progress sid-fit An unsupervised sidClustering forest on the clustering variables, after mice or forest imputation pooled over M, saving a digest-keyed SID dissimilarity for sid-clusters and vt. add_job("sid", subject, type, qualifier = "fit")
🟡 In progress vp-dependence Unsupervised VarPro: how the variables depend on each other, as a graph whose edges must beat permuted noise, checked across subsamples. add_job("vp", subject, type, qualifier = "dependence")
🟡 In progress vp-explain VarPro partial dependence for the variables a vp-fit screen selected. It never fits varPro. add_job("vp", subject, type, qualifier = "explain")
🟡 In progress vp-fit A VarPro variable screen that says which variables were tested, judges selection against permuted noise of the same kind, and checks the selection across subsamples. add_job("vp", subject, type, qualifier = "fit")
🟡 In progress vp-individual Individual VarPro: each patient’s local gradient for each predictor, from the fit a vp-fit screen saved. It never fits varPro. add_job("vp", subject, type, qualifier = "individual")
🟡 In progress vt-explain The twin contrast a vt-fit job saved: by arm, by eligibility rule and by stratum, with standardized betas and who might benefit. It never grows a forest. add_job("vt", subject, type, qualifier = "explain")
🟡 In progress vt-fit Per-arm survival forests, swapped-arm twin curves and a per-patient RMST difference, with eligibility judged by a choice of methods. add_job("vt", subject, type, qualifier = "fit")

graphs

Delivery Template Description Scaffold
🟡 In progress hp Nomogram and hazard figures, read from the ac life table and the hz fit rather than recomputed. add_job("hp", subject, type)
🟢 Shipped hs-concordance Every patient predicted through every treatment group’s hm model at one horizon, with an optional table of actual against best-predicted treatment. add_job("hs", subject, type, qualifier = "concordance")
🟢 Shipped hs-setup Patient-level predictions from the hm model, set against the survival of a matched general population. add_job("hs", subject, type, qualifier = "setup")
🟢 Shipped lp-mirror The mirrored histogram of the propensity score, treated above the axis and controls below, drawn from a saved lm-propensity_binary bundle without refitting: all patients, under the matching weights, and matched when the study has a match flag. add_job("lp", subject, type, qualifier = "mirror")

Coming

Job types the SAS library and the study corpus carry that have no template yet. SAS jobs counts the distinct studies with a SAS job of that type, so a larger number is a template more studies are waiting for. Unmeasured means nobody has counted it yet, which is not the same as zero.

Delivery Template Description SAS jobs Waiting on
🟡 In progress ar Analysis report 394
🟡 In progress nd Non-linear distributions 244
🟡 In progress np Non-linear plot 241
🟡 In progress dc-variable Distribution of a variable 237
🟡 In progress cd Cumulative distribution 190
🟡 In progress dc-dead Descriptive: mortality 171
🟡 In progress nm Non-linear model 121
🟡 In progress gm Generalized model 73
🟡 In progress rp Regression plot 68
🟡 In progress mm Mixed model 56
🟡 In progress rg Regression trends 45
🟡 In progress mp Mixed model plot 41
🟡 In progress dp-spaghetti Descriptive plot: spaghetti 40
🟡 In progress cm Cox matching 35
🟡 In progress dp-procs Descriptive plot: procedures over time 35
🟡 In progress ls Life table / STS 32
🟡 In progress si Single imputation 1
🔴 Not yet on the way vars Variables 912 hvtiRdatabuild
🔴 Not yet on the way dt Data check 503 hvtiRdatabuild
🔴 Not yet on the way ce Competing events 128 hvtiPlotR#134
🔴 Not yet on the way bn Bootstrap non-linear 108
🔴 Not yet on the way gp Generalized model plot 50 hvtiPlotR#136
🔴 Not yet on the way mi Multiple imputation 18 hvtiRimputation
🔴 Not yet on the way fp Forest plot 11 hvtiPlotR#133
🔴 Not yet on the way dc-boxplot Descriptive plot: boxplot 9
🔴 Not yet on the way cp Cumulative probability plot 4 hvtiPlotR#135
🔴 Not yet on the way bq Bootstrap quantile 2 hvtiRbootstrap#16
🔴 Not yet on the way nb-boostmlr Boosting: BoostMLR unmeasured

Where to read more

  • template_catalog() returns this table with every field the catalog keeps, including the design note behind each row.
  • inst/templates/README.md in the package source explains the naming and the folder layout.
  • The From SAS descriptive jobs to R vignette walks through running the descriptive templates against a study.