Package 'dcurves'

Convert DCA Object to tibble

Description

Convert DCA Object to tibble

Usage

## S3 method for class 'dca'
as_tibble(x, ...)

Arguments

x	dca object created with dca()
...	not used

Value

a tibble

Author(s)

Daniel D Sjoberg

See Also

dca(), net_intervention_avoided(), standardized_net_benefit(), plot.dca()

Examples

dca(cancer ~ cancerpredmarker, data = df_binary) %>%
  as_tibble()

---

Perform Decision Curve Analysis

Description

Diagnostic and prognostic models are typically evaluated with measures of accuracy that do not address clinical consequences. Decision-analytic techniques allow assessment of clinical outcomes but often require collection of additional information may be cumbersome to apply to models that yield a continuous result. Decision curve analysis is a method for evaluating and comparing prediction models that incorporates clinical consequences, requires only the data set on which the models are tested, and can be applied to models that have either continuous or dichotomous results. The dca function performs decision curve analysis for binary outcomes. Review the DCA Vignette for a detailed walk-through of various applications. Also, see www.decisioncurveanalysis.org for more information.

Usage

dca(
  formula,
  data,
  thresholds = seq(0, 0.99, by = 0.01),
  label = NULL,
  harm = NULL,
  as_probability = character(),
  time = NULL,
  prevalence = NULL
)

Arguments

formula	a formula with the outcome on the LHS and a sum of markers/covariates to test on the RHS
data	a data frame containing the variables in ⁠formula=⁠.
thresholds	vector of threshold probabilities between 0 and 1. Default is seq(0, 0.99, by = 0.01). Thresholds at zero are replaced with 10e-10.
label	named list of variable labels, e.g. list(age = "Age, years")
harm	named list of harms associated with a test. Default is NULL
as_probability	character vector including names of variables that will be converted to a probability. Details below.
time	if outcome is survival, ⁠time=⁠ specifies the time the assessment is made
prevalence	When NULL, the prevalence is estimated from ⁠data=⁠. If the data passed is a case-control set, the population prevalence may be set with this argument.

Value

List including net benefit of each variable

as_probability argument

While the ⁠as_probability=⁠ argument can be used to convert a marker to the probability scale, use the argument only when the consequences are fully understood. For example, when the outcome is binary, logistic regression is used to convert the marker to a probability. The logistic regression model assumes linearity on the log-odds scale and can induce miscalibration when this assumption is not true. Miscalibration in a model will adversely affect performance on decision curve analysis. Similarly, when the outcome is time-to-event, Cox Proportional Hazards regression is used to convert the marker to a probability. The Cox model also has a linearity assumption and additionally assumes proportional hazards over the follow-up period. When these assumptions are violated, important miscalibration may occur.

Instead of using the ⁠as_probability=⁠ argument, it is suggested to perform the regression modeling outside of the dca() function utilizing methods, such as non-linear modeling, as appropriate.

Author(s)

Daniel D Sjoberg

See Also

net_intervention_avoided(), standardized_net_benefit(), plot.dca(), as_tibble.dca()

Examples

# calculate DCA with binary endpoint
dca(cancer ~ cancerpredmarker + marker,
    data = df_binary,
    as_probability = "marker",
    label = list(cancerpredmarker = "Prediction Model", marker = "Biomarker")) %>%
  # plot DCA curves with ggplot
  plot(smooth = TRUE) +
  # add ggplot formatting
  ggplot2::labs(x = "Treatment Threshold Probability")

# calculate DCA with time to event endpoint
dca(Surv(ttcancer, cancer) ~ cancerpredmarker, data = df_surv, time = 1)

---

Simulated data with a binary outcome

Description

Simulated data with a binary outcome

Usage

df_binary

Format

A data frame with 750 rows:

patientid

Identification Number

cancer

Cancer Diagnosis: 0=No, 1=Yes

dead

Dead (1=yes; 0=no)

risk_group

Patient Risk Group (Low, Intermediate, High)

age

Patient Age, years

famhistory

Family History of Cancer: 0=No, 1=Yes

marker

Marker

cancerpredmarker

Prob. of Cancer based on Age, Family History, and Marker

---

Simulated data with a case-control outcome

Description

Simulated data with a case-control outcome

Usage

df_case_control

Format

A data frame with 750 rows:

patientid

Identification Number

casecontrol

Case-control Status: 1=Case, 0=Control

risk_group

Patient Risk Group (Low, Intermediate, High)

age

Patient Age, years

famhistory

Family History of Cancer: 0=No, 1=Yes

marker

Marker

cancerpredmarker

Prob. of Cancer based on Age, Family History, and Marker

---

Simulated data with a survival outcome

Description

Simulated data with a survival outcome

Usage

df_surv

Format

A data frame with 750 rows:

patientid

Identification Number

cancer

Cancer Diagnosis: 0=No, 1=Yes

cancer_cr

Cancer Diagnosis, competing event: "censor", "dead other causes", "diagnosed with cancer"

ttcancer

Years to Cancer Dx/Censor

risk_group

Patient Risk Group (Low, Intermediate, High)

age

Patient Age, years

famhistory

Family History of Cancer: 0=No, 1=Yes

marker

Marker

cancerpredmarker

Prob. of Cancer based on Age, Family History, and Marker

---

Add Net Interventions Avoided

Description

Add the number of net interventions avoided to dca() object.

Usage

net_intervention_avoided(x, nper = 1)

Arguments

x	object of class 'dca' calculated with dca()
nper	Number to report net interventions per. Default is 1

Value

'dca' object

Author(s)

Daniel D Sjoberg

See Also

dca(), standardized_net_benefit(), plot.dca(), as_tibble.dca()

Examples

dca(
  cancer ~ cancerpredmarker,
  data = df_binary
) %>%
  net_intervention_avoided()

dca(
  Surv(ttcancer, cancer) ~ cancerpredmarker,
  data = df_surv,
  time = 1
) %>%
  net_intervention_avoided(nper = 100)

---

Plot DCA Object with ggplot

Description

Plot DCA Object with ggplot

Usage

## S3 method for class 'dca'
plot(
  x,
  type = NULL,
  smooth = FALSE,
  span = 0.2,
  style = c("color", "bw"),
  show_ggplot_code = FALSE,
  ...
)

Arguments

x	dca object created with dca()
type	indicates type of plot to produce. Must be one of c("net_benefit", "net_intervention_avoided", "standardized_net_benefit"). The default is "net_benefit", unless the net intervention has been calculated when "net_intervention_avoided" is used, or if "standardized_net_benefit" has been calculated.
smooth	Logical indicator whether plot will be smooth with ggplot2::stat_smooth(). Default is FALSE
span	when smooth = TRUE, Controls the amount of smoothing for loess smoother. Smaller numbers produce wigglier lines, larger numbers produce smoother lines. Default is 0.2.
style	Must be one of c("color", "bw"). Default is "color", and "bw" will print a black and white figure
show_ggplot_code	Logical indicating whether to print ggplot2 code used to create figure. Default is FALSE. Set to TRUE to perform advanced figure customization
...	not used

Value

a ggplot2 object

Author(s)

Daniel D Sjoberg

See Also

dca(), net_intervention_avoided(), standardized_net_benefit(), as_tibble.dca()

Examples

p <-
  dca(cancer ~ cancerpredmarker, data = df_binary) %>%
  plot(smooth = TRUE, show_ggplot_code = TRUE)
p

# change the line colors
p + ggplot2::scale_color_manual(values = c('black', 'grey', 'purple'))

---

Add Standardized Net Benefit

Description

Add the standardized net benefit to dca() object.

Usage

standardized_net_benefit(x)

Arguments

x	object of class 'dca' calculated with dca()

Value

'dca' object

Author(s)

Daniel D Sjoberg

See Also

dca(), net_intervention_avoided(), plot.dca(), as_tibble.dca()

Examples

dca(Surv(ttcancer, cancer) ~ cancerpredmarker, data = df_surv, time = 1) %>%
  standardized_net_benefit()

---

Test Consequences

Description

Test Consequences

Usage

test_consequences(
  formula,
  data,
  statistics = c("pos_rate", "neg_rate", "test_pos_rate", "test_neg_rate", "tp_rate",
    "fp_rate", "fn_rate", "tn_rate", "ppv", "npv", "sens", "spec", "lr_pos", "lr_neg"),
  thresholds = seq(0, 1, by = 0.25),
  label = NULL,
  time = NULL,
  prevalence = NULL
)

Arguments

formula	a formula with the outcome on the LHS and a sum of markers/covariates to test on the RHS
data	a data frame containing the variables in ⁠formula=⁠.
statistics	Character vector with statistics to return. See below for details
thresholds	vector of threshold probabilities between 0 and 1. Default is seq(0, 0.99, by = 0.01). Thresholds at zero are replaced with 10e-10.
label	named list of variable labels, e.g. list(age = "Age, years")
time	if outcome is survival, ⁠time=⁠ specifies the time the assessment is made
prevalence	When NULL, the prevalence is estimated from ⁠data=⁠. If the data passed is a case-control set, the population prevalence may be set with this argument.

Value

a tibble with test consequences

statistics

The following diagnostic statistics are available to return.

Statistic	Abbreviation	Definition
Outcome Positive Rate	"pos_rate"	(a + c) / (a + b + c + d)
Outcome Negative Rate	"neg_rate"	(b + d) / (a + b + c + d)
Test Positive Rate	"test_pos_rate"	(a + b) / (a + b + c + d)
Test Negative Rate	"test_neg_rate"	(c + d) / (a + b + c + d)
True Positive Rate	"tp_rate"	a / (a + b + c + d)
False Positive Rate	"fp_rate"	b / (a + b + c + d)
False Negative Rate	"fn_rate"	c / (a + b + c + d)
True Negative Rate	"tn_rate"	d / (a + b + c + d)
Positive Predictive Value	"ppv"	a / (a + b)
Negative Predictive Value	"npv"	d / (c + d)
Sensitivity	"sens"	a / (a + c)
Specificity	"spec"	d / (b + d)
Positive Likelihood Ratio	"lr_pos"	sens / (1 - spec)
Negative Likelihood Ratio	"lr_neg"	(1 - sens) / spec

Examples

test_consequences(cancer ~ cancerpredmarker, data = df_binary)

---