statforclinicians

Draft: This page has not been reviewed by a statistician yet. Use it as a starting point and check important choices with a statistician.

Bland-Altman plot and ICC

Shows how closely two methods that measure the same quantity agree, with the average difference and the range in which most differences fall.

Clinical example

Two blood pressure devices used one after the other on the same patients.

When to use it

  • Two methods measure the same continuous quantity in the same patients.
  • You want to know whether the two methods can be used interchangeably.
  • You want the size of the differences, not just whether the values are related.

Assumptions to check

Differences roughly constant across the range

Plot the difference against the average of both methods. If the spread of the differences changes with the size of the measurement, the limits of agreement need another approach.

Acceptable agreement is a clinical judgment

The limits of agreement describe the differences. Decide in advance how large a difference is acceptable.

What to report

  • The Bland-Altman plot
  • Bias (mean difference) with 95% limits of agreement
  • ICC with 95% CI, and which form of ICC you used

Common mistakes

Correlation doesn't show agreement

Two methods can correlate perfectly and still disagree systematically. Use a Bland-Altman plot.

In your software

R
# R
library(irr)
d$diff <- d$device_a - d$device_b
mean(d$diff)
mean(d$diff) + c(-1.96, 1.96) * sd(d$diff)
icc(d[, c("device_a", "device_b")], model = "twoway", type = "agreement", unit = "single")
SPSS
COMPUTE diff = device_a - device_b.
COMPUTE avg = (device_a + device_b) / 2.
EXECUTE.
DESCRIPTIVES VARIABLES=diff /STATISTICS=MEAN STDDEV.
GRAPH /SCATTERPLOT(BIVAR)=avg WITH diff.
RELIABILITY
  /VARIABLES=device_a device_b
  /SCALE('ALL VARIABLES') ALL
  /MODEL=ALPHA
  /ICC=MODEL(RANDOM) TYPE(ABSOLUTE) CIN=95 TESTVAL=0.

Sources