(Slightly extended) Bland-Altman plots BlandAltmanLeh

Description

Bland-Altman Plots for assessing agreement between two methods of clinical measurement and returning associated statistics. Plots are optionally extended by confidence intervals as described in "J. Martin Bland, Douglas G. Altman (1986): Statistical Methods For Assessing Agreement Between Two Methods Of Clinical Measurement" but not included in the graphics of that publication. Either base graphics or ggplot2 can be used.

Details

Bland Altman plots are a diagnostic tool for assessing the agreement between two methods of measurement or assessing retest reliability from two measurements. This package offers plots in base and ggplot2 graphics as well as detailed descriptive statistics, thus supporting the construction of individual plots based on Bland Altman plots. Bland and Altman describe a way for constructing confidence intervals. This package computes these confidence intervals and includes them into the plots. It also invents the Sunflower-Bland-Altman plot for data with ties.

Author(s)

Bernhard Lehnert

Maintainer: Bernhard K. Lehnert <bernhard.lehnert@uni-greifswald.de>

References

Bland JM, Altman DG, Statistical Methods For Assessing Agreement Between Two Methods Of Clinical Measurement, Lancet, 1986; 307-310.

Altman DG, Bland JM, Measurement in medicine: the analysis of method comparison studies, The Statistician 1983; 32, 307-317.

Vaz S et al., The Case for Using the Repeatability Coeffcient When Calculating Test-Retest Reliability, PLOS ONE, Sept. 2013, Vol 8, Issue 9.

See Also

bland.altman.plot,bland.altman.stats

Examples

# simple basic Bland Altman plot
a <- rnorm(40,30,10)
b <- 1.01*a + rnorm(40)
bland.altman.plot(a,b, xlab="mean", ylab="difference")

# to get all the data for further analysis
bland.altman.plot(a,b, xlab="mean", ylab="difference", silent=FALSE)

# to include confidence intervals into the plot
bland.altman.plot(a,b, xlab="mean", ylab="difference", conf.int=.95)

# to plot in ggplot2
bland.altman.plot(a,b, graph.sys="ggplot2")

# to mark ties in a Sunflower-Bland-Altman plot
a <- sample(1:5, 40, replace=TRUE)
b <- rep(c(1,2,3,3,5,5,5,5),5)
bland.altman.plot(a, b, sunflower=TRUE)
 

PEFR Data from Bland JM and Altman DG 1986

Description

Peak expiratory flow data from 17 members of Bland's family, taken with two different instruments, each twice. This data is for explanatory use only. Columns 1 and 2 were measured with the "Wright" peak flow meter, columns 3 and 4 with the "Mini Wright" peak flow meter. These are the data behind fig. 1, fig. 2 and fig. 6 of the original paper and these can be easily reconstructed

Usage

bland.altman.PEFR

Format

An object of class data.frame with 17 rows and 4 columns.

Examples

# this is what fig. 1. would have looked like in R:
x <- bland.altman.PEFR[["bigger.first"]]
y <- bland.altman.PEFR[["smaller.first"]]
plot(x,y, xlab="PEFR by large meter",ylab="PEFR by mini meter", 
     xlim=c(0,800), ylim=c(0,800))
abline(0,1)

Produce Bland-Altman Plot Bland-AltmanPlots for assessing agreement between two measuring methods or repeatability (test-retest agreement) of measurements. Using either base graphics or ggplot2.

Description

Produce Bland-Altman Plot

Bland-AltmanPlots for assessing agreement between two measuring methods or repeatability (test-retest agreement) of measurements. Using either base graphics or ggplot2.

Usage

bland.altman.plot(group1, group2, two = 1.96, mode = 1,
  graph.sys = "base", conf.int = 0, silent = TRUE, sunflower = FALSE,
  geom_count = FALSE, ...)

Arguments

Value

Depends on graphic system chosen. In case of "base" depending on whether silent==TRUE. If silent==TRUE then no returns. If silent==FALSE than returns list of statistics as returned by bland.altman.stats(). In case the graphics system is "ggplot2" than the graphic object is returned so that it can be printed or altered.

Author(s)

Bernhard Lehnert <bernhard.lehnert@uni-greifswald.de>

See Also

bland.altman.stats

Examples

bland.altman.plot(rnorm(20), rnorm(20), xlab="mean measurement", 
                  ylab="differences", main="Example plot")
                  
bland.altman.plot(rnorm(20), 2+.8*rnorm(20), xlab="mean measurement", 
                  ylab="differences", conf.int=.95)
                  
bland.altman.plot(rnorm(200), 2+.8*rnorm(200), xlab="mean measurement", 
                  ylab="differences", conf.int=.95)
                  
# this is what fig.2 in Bland&Altman1986 would have looked like
PEFR1 <- bland.altman.PEFR[,1]
PEFR2 <- bland.altman.PEFR[,3]
bland.altman.plot(PEFR1, PEFR2, silent=TRUE, xlim=c(0,800),
                  xlab="Average PEFR by two meters",
                  ylab="Difference in PEFR (large-mini)")

# and this is the same but with additional 95 percent CIs
data(bland.altman.PEFR)
bland.altman.plot(PEFR1, PEFR2, silent=TRUE, conf.int=.95, xlim=c(0,800))
                  
# an example with many ties and the 'sunflower'-option
a <- rep(c(1,1,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,4,4,4,5,6,6),2)
b <- rep(c(1,1,1,2,2,2,3,1,4,2,5,3,3,3,3,3),3)
bland.altman.plot(a,b,sunflower=TRUE, xlab="Mean",ylab="Difference",
                  main="discrete values lead to ties")
                  
library(ggplot2)
a <- bland.altman.plot(rnorm(20), rnorm(20), graph.sys="ggplot2", conf.int=.9)
print(a + xlab("you can change this later") + ggtitle("Title goes here"))                  

Calculate statistics for Bland-Altman-Plot

Description

Does the computation for Bland Altman plots. This will usually be called from graphic functions like bland.altman.plot but will be usefull for customized plot (see examples for color coded BA plot). Offers symmetric confidence intervalls for bias and upper and lower limits.

Usage

bland.altman.stats(group1, group2, two = 1.96, mode = 1, conf.int = 0.95)

Arguments

Value

means vector of means, i. e. data for the x axis

diffs vector of differences, i. e. data for the y axis

groups data.frame containing pairwise complete cases of group1 and group2. NAs are removed.

based.on count of pairwise complete cases in groups

lower.limit lower limit for BA plot

mean.diffs mean of differences, also called 'bias'

upper.limit upper limit for BA plot

lines vector containing y values where to draw horizontal lines, i. e. mean of differences minus "two" standard deviations, mean of differences and mean of differences plus "two" standard deviations (i. e. c(lower.limit, mean.diffs, upper.limit). This is convenient for printing.

CI.lines vector of confidence intervalls for the values of lines (based on the assumption of normal distribution of differences diffs).

two the argument 'two'

critical.diff critical difference, i. e. 'two' times standard deviation of differences, equals half the difference of lower.limit and upper.limit

Author(s)

Bernhard Lehnert <bernhard.lehnert@uni-greifswald.de>

See Also

bland.altman.plot

Examples

# simple calculation of stats:
a <- rnorm(20)
b <- jitter(a)
print(bland.altman.stats(a, b))
print(bland.altman.stats(a, b)$critical.diff)

# drawing Bland-Altman-Plot with color coding sex:
example.data <- data.frame(sex = gl(2,6,labels=c("f","m")),
                 m1 = c(16,10,14,18,16,15,18,19,14,11,11,17),
                 m2 = c(18, 9,15,19,19,13,19,20,14,11,13,17))
ba <- bland.altman.stats(example.data$m1, example.data$m2)
plot(ba$means, ba$diffs, col=example.data$sex, ylim=c(-4,4))
abline(h=ba$lines, lty=2)
              
# compute 95%-CIs for the bias and upper and lower limits of PEFR data as 
# in Bland&Altman 1986
bland.altman.stats(bland.altman.PEFR[,1],bland.altman.PEFR[,3])$CI.lines
# apparently wrong results? CAVE: Bland&Altman are using two=2, thus
bland.altman.stats(bland.altman.PEFR[,1],bland.altman.PEFR[,3], two=2)$CI.lines