Multivariable Survival Analysis — multisurvival • ClinicoPath

Performs multivariable survival analysis using Cox proportional hazards regression. In multivariable survival analysis, person-time follow-up is crucial for properly adjusting for covariates while accounting for varying observation periods. The Cox proportional hazards model incorporates person-time by modeling the hazard function, which represents the instantaneous event rate per unit of person-time. When stratifying analyses or examining multiple predictors, the model accounts for how these factors influence event rates relative to the person-time at risk in each subgroup.

Usage

multisurvival(
  data,
  elapsedtime,
  tint = FALSE,
  dxdate,
  fudate,
  timetypedata = "ymd",
  timetypeoutput = "months",
  uselandmark = FALSE,
  landmark = 3,
  outcome,
  outcomeLevel,
  dod,
  dooc,
  awd,
  awod,
  analysistype = "overall",
  explanatory,
  contexpl,
  multievent = FALSE,
  hr = FALSE,
  sty = "t1",
  ph_cox = FALSE,
  km = FALSE,
  endplot = 60,
  byplot = 12,
  ci95 = FALSE,
  risktable = FALSE,
  censored = FALSE,
  medianline = "none",
  pplot = TRUE,
  cutp = "12, 36, 60",
  calculateRiskScore = FALSE,
  numRiskGroups = "four",
  plotRiskGroups = FALSE,
  ac = FALSE,
  adjexplanatory,
  ac_method = "average",
  showNomogram = FALSE,
  use_modelSelection = FALSE,
  modelSelection = "enter",
  selectionCriteria = "aic",
  pEntry = 0.05,
  pRemoval = 0.1,
  use_stratify = FALSE,
  stratvar,
  person_time = FALSE,
  time_intervals = "12, 36, 60",
  rate_multiplier = 100,
  use_tree = FALSE,
  min_node = 20,
  complexity = 0.01,
  max_depth = 5,
  show_terminal_nodes = TRUE,
  use_time_dependent = FALSE,
  td_format = "wide",
  time_dep_vars,
  change_times = "6, 12, 18",
  td_suffix_pattern = "_t{time}",
  start_time_var,
  stop_time_var,
  use_frailty = FALSE,
  frailty_var,
  frailty_distribution = "gamma",
  use_splines = FALSE,
  spline_vars,
  spline_df = 3,
  spline_type = "pspline"
)

Arguments

data: The dataset to be analyzed, provided as a data frame. Must contain the variables specified in the options below.
elapsedtime: The numeric variable representing follow-up time until the event or last observation. If tint = false, this should be a pre-calculated numeric time variable. If tint = true, dxdate and fudate will be used to calculate this time.
tint: If true, survival time will be calculated from dxdate and fudate. If false, elapsedtime should be provided as a pre-calculated numeric variable.
dxdate: Date of diagnosis. Required if tint = true. Must match the format specified in timetypedata.
fudate: Follow-up date or date of last observation. Required if tint = true. Must match the format specified in timetypedata.
timetypedata: Specifies the format of the date variables in the input data. This is critical if tint = true, as dxdate and fudate will be parsed according to this format to calculate survival time. For example, if your data files record dates as "YYYY-MM-DD", select ymd.
timetypeoutput: The units in which survival time is reported in the output. Choose from days, weeks, months, or years.
uselandmark: If true, applies a landmark analysis starting at a specified time point.
landmark: The time point (in the units defined by timetypeoutput) at which to start landmark analyses. Only used if uselandmark = true.
outcome: The outcome variable. Typically indicates event status (e.g., death, recurrence). For survival analysis, this may be a factor or numeric event indicator.
outcomeLevel: The level of outcome considered as the event. For example, if outcome is a factor, specify which level indicates the event occurrence.
dod: The level of outcome corresponding to death due to disease, if applicable.
dooc: The level of outcome corresponding to death due to other causes, if applicable.
awd: The level of outcome corresponding to alive with disease, if applicable.
awod: The level of outcome corresponding to alive without disease, if applicable.
analysistype: Type of survival analysis: - overall: All-cause survival - cause: Cause-specific survival - compete: Competing risks analysis
explanatory: Categorical explanatory (predictor) variables included in the Cox model.
contexpl: Continuous explanatory (predictor) variables included in the Cox model.
multievent: If true, multiple event levels will be considered for competing risks analysis. Requires specifying dod, dooc, etc.
hr: If true, generates a plot of hazard ratios for each explanatory variable in the Cox model.
sty: The style of the hazard ratio (forest) plot. "finalfit" or "survminer forestplot".
ph_cox: If true, tests the proportional hazards assumption for the Cox model. Use if you suspect violations of the PH assumption.
km: If true, produces a Kaplan-Meier survival plot. Useful for visualization of survival functions without covariate adjustment.
endplot: The maximum follow-up time (in units defined by timetypeoutput) to display on survival plots.
byplot: The interval (in units defined by timetypeoutput) at which time points or labels are shown on plots.
ci95: If true, displays 95\ estimates on plots.
risktable: If true, displays the number of subjects at risk at each time point below the survival plot.
censored: If true, marks censored observations (e.g., using tick marks) on the survival plot.
medianline: If true, displays a line indicating the median survival time on the survival plot.
pplot: If true, displays the p-value from the survival comparison test on the survival plot.
cutp: .
calculateRiskScore: If true, calculates a risk score from the Cox model coefficients for each individual.
numRiskGroups: Select the number of risk groups to create from the risk scores. The data will be divided into equal quantiles based on this selection.
plotRiskGroups: If true, stratifies individuals into risk groups based on their calculated risk scores and plots their survival curves.
ac: .
adjexplanatory: .
ac_method: Method for computing adjusted survival curves
showNomogram: .
use_modelSelection: If true, applies a variable selection procedure to find the best-fitting model based on criteria like AIC or likelihood ratio tests.
modelSelection: The method used to select variables: - enter: Includes all variables (no selection) - forward: Adds variables one at a time if they improve the model - backward: Removes variables that do not significantly contribute - both: Combination of forward and backward steps
selectionCriteria: The criterion used for adding or removing variables in model selection: - aic: Balances model fit and complexity - lrt: Uses likelihood ratio tests to decide inclusion/removal
pEntry: Significance level at which a variable enters the model during forward or stepwise selection.
pRemoval: Significance level at which a variable is removed from the model during backward or stepwise selection.
use_stratify: If true, uses stratification to handle variables that violate the proportional hazards assumption. Stratification creates separate baseline hazard functions for different groups.
stratvar: Variables used for stratification. When proportional hazards are not met, stratification can adjust the model to better fit the data by allowing different baseline hazards.
person_time: Enable this option to calculate and display person-time metrics, including total follow-up time and incidence rates. These metrics help quantify the rate of events per unit of time in your study population.
time_intervals: Specify time intervals for stratified person-time analysis. Enter a comma-separated list of time points to create intervals. For example, "12, 36, 60" will create intervals 0-12, 12-36, 36-60, and 60+.
rate_multiplier: Specify the multiplier for incidence rates (e.g., 100 for rates per 100 person-years, 1000 for rates per 1000 person-years).
use_tree: If true, fits a survival decision tree to identify subgroups with different survival outcomes. Decision trees provide an intuitive alternative to Cox regression for identifying risk factors.
min_node: The minimum number of observations required in a terminal node. Larger values create simpler trees that may be more generalizable but potentially miss important subgroups.
complexity: The complexity parameter for tree pruning. Higher values result in smaller trees. This parameter controls the trade-off between tree size and goodness of fit.
max_depth: The maximum depth of the decision tree. Limits the complexity of the tree to avoid overfitting.
show_terminal_nodes: If true, displays Kaplan-Meier survival curves for each terminal node of the decision tree.
use_time_dependent: Enable time-dependent covariates for Cox regression. This allows modeling variables that change values at specific time points during follow-up (e.g., treatment changes, biomarker measurements, disease progression).
td_format: Specify whether your data is in wide format (one row per subject with time points as separate variables) or long format (multiple rows per subject with time intervals).
time_dep_vars: Variables that change values over time. In wide format, these are baseline variables that will be updated at change points. In long format, these are the time-varying variables.
change_times: Time points (in same units as survival time) when time-dependent variables change. For wide format data, specify comma-separated time points (e.g., "6, 12, 18"). The function will create intervals and update covariate values at these times.
td_suffix_pattern: For wide format: Pattern for time-specific variable names. Use time as placeholder. Example: if baseline variable is 'treatment' and pattern is '_ttime', the function looks for 'treatment_t6', 'treatment_t12', etc.
start_time_var: For long format only: Variable indicating the start time of each interval. Leave empty for wide format data.
stop_time_var: For long format only: Variable indicating the stop time of each interval. Leave empty for wide format data.
use_frailty: Add a frailty term to account for unobserved heterogeneity or clustering in the data. Frailty models add random effects to the Cox model.
frailty_var: Clustering variable for the frailty term (e.g., hospital, family, or study center). Each level represents a cluster with shared frailty.
frailty_distribution: Distribution of the frailty term. Gamma is most commonly used and assumes multiplicative effect on the hazard. Gaussian assumes additive effect on log-hazard.
use_splines: Use penalized splines to model time-varying effects (non-proportional hazards). This is an alternative to stratification for handling PH violations.
spline_vars: Variables to model with time-varying coefficients using splines. These are variables that violate the proportional hazards assumption.
spline_df: Degrees of freedom for the spline functions. Higher values allow more flexible time-varying effects but may lead to overfitting.
spline_type: Type of spline basis to use. Penalized splines provide smooth functions with automatic smoothness selection. Natural splines are constrained to be linear at the boundaries.

Value

A results object containing:

`results$todo`					a html
`results$text`					a html
`results$text2`					a html
`results$personTimeTable`					a table
`results$personTimeSummary`					a html
`results$plot`					an image
`results$plot3`					an image
`results$cox_ph`					a preformatted
`results$plot8`					an image
`results$plotKM`					an image
`results$risk_score_analysis`					a preformatted
`results$risk_score_analysis2`					a html
`results$riskScoreTable`					a table
`results$riskScoreMetrics`					a html
`results$riskGroupPlot`					an image
`results$stratificationExplanation`					a html
`results$calculatedtime`					an output
`results$outcomeredefined`					an output
`results$addRiskScore`					an output
`results$addRiskGroup`					an output
`results$plot_adj`					an image
`results$plot_nomogram`					an image
`results$nomogram_display`					a html
`results$tree_summary`					a html
`results$tree_plot`					an image
`results$node_survival_plots`					an image
`results$mydataview_modelselection`					a preformatted
`results$text_model_selection`					a html
`results$selection_method`					a html
`results$text2_model_selection`					a html

Tables can be converted to data frames with asDF or as.data.frame. For example:

results$personTimeTable$asDF

as.data.frame(results$personTimeTable)

Examples

# Example 1: Basic multivariable Cox regression
library(survival)
data(colon)

multisurvival(
    data = colon,
    elapsedtime = "time",
    outcome = "status",
    outcomeLevel = "1",
    explanatory = c("sex", "obstruct", "perfor"),
    contexpl = c("age", "nodes"),
    timetypeoutput = "days",
    hr = TRUE  # Show hazard ratio plot
)
#> Error in multisurvival(data = colon, elapsedtime = "time", outcome = "status",     outcomeLevel = "1", explanatory = c("sex", "obstruct", "perfor"),     contexpl = c("age", "nodes"), timetypeoutput = "days", hr = TRUE): argument "adjexplanatory" is missing, with no default

# Example 2: Using dates to calculate survival time
# Assuming you have diagnosis and follow-up dates
multisurvival(
    data = mydata,
    tint = TRUE,
    dxdate = "diagnosis_date",
    fudate = "last_followup_date",
    timetypedata = "ymd",
    timetypeoutput = "months",
    outcome = "vital_status",
    outcomeLevel = "Dead",
    explanatory = c("stage", "grade"),
    contexpl = "age"
)
#> Error: object 'mydata' not found

# Example 3: Risk stratification analysis
multisurvival(
    data = colon,
    elapsedtime = "time",
    outcome = "status",
    outcomeLevel = "1",
    explanatory = c("sex", "obstruct"),
    contexpl = c("age", "nodes"),
    calculateRiskScore = TRUE,
    numRiskGroups = "three",
    plotRiskGroups = TRUE,
    addRiskScore = TRUE,  # Add risk score to data
    addRiskGroup = TRUE   # Add risk group to data
)
#> Error in multisurvival(data = colon, elapsedtime = "time", outcome = "status",     outcomeLevel = "1", explanatory = c("sex", "obstruct"), contexpl = c("age",         "nodes"), calculateRiskScore = TRUE, numRiskGroups = "three",     plotRiskGroups = TRUE, addRiskScore = TRUE, addRiskGroup = TRUE): unused arguments (addRiskScore = TRUE, addRiskGroup = TRUE)

# Example 4: Model with stratification for non-proportional hazards
multisurvival(
    data = colon,
    elapsedtime = "time",
    outcome = "status",
    outcomeLevel = "1",
    explanatory = c("obstruct", "perfor"),
    contexpl = c("age", "nodes"),
    use_stratify = TRUE,
    stratvar = "sex",  # Stratify by sex if PH assumption violated
    ph_cox = TRUE      # Test proportional hazards assumption
)
#> Error in multisurvival(data = colon, elapsedtime = "time", outcome = "status",     outcomeLevel = "1", explanatory = c("obstruct", "perfor"),     contexpl = c("age", "nodes"), use_stratify = TRUE, stratvar = "sex",     ph_cox = TRUE): argument "adjexplanatory" is missing, with no default

# Example 5: Stepwise model selection
multisurvival(
    data = colon,
    elapsedtime = "time",
    outcome = "status",
    outcomeLevel = "1",
    explanatory = c("sex", "obstruct", "perfor", "adhere"),
    contexpl = c("age", "nodes"),
    use_modelSelection = TRUE,
    modelSelection = "both",  # Stepwise selection
    selectionCriteria = "aic",
    pEntry = 0.05,
    pRemoval = 0.10
)
#> Error in multisurvival(data = colon, elapsedtime = "time", outcome = "status",     outcomeLevel = "1", explanatory = c("sex", "obstruct", "perfor",         "adhere"), contexpl = c("age", "nodes"), use_modelSelection = TRUE,     modelSelection = "both", selectionCriteria = "aic", pEntry = 0.05,     pRemoval = 0.1): argument "adjexplanatory" is missing, with no default

# Example 6: Person-time analysis
multisurvival(
    data = colon,
    elapsedtime = "time",
    outcome = "status",
    outcomeLevel = "1",
    explanatory = "sex",
    contexpl = "age",
    person_time = TRUE,
    time_intervals = "180, 365, 730",  # 6mo, 1yr, 2yr
    rate_multiplier = 1000  # Rate per 1000 person-days
)
#> Error in multisurvival(data = colon, elapsedtime = "time", outcome = "status",     outcomeLevel = "1", explanatory = "sex", contexpl = "age",     person_time = TRUE, time_intervals = "180, 365, 730", rate_multiplier = 1000): argument "adjexplanatory" is missing, with no default