The Agree package is developed to calculate the agreement and reliability between the scores of multiple raters or repeated measurments. The (proportion) agreement between multiple raters can be calculated for dichotomous scores as well as for polytomous scores. The confidence interval around the agreement can be calculated, adjusted for multiple raters. Additionally, the Intra Class Correlation (ICC) and Standard Error of Measurement (SEM) can be calculated for continuous scores. These can be obtained from estimated variance components extracted from a multilevel model.

Package installation

The package can be installed directly from GitHub by remotes::install_github(repo = 'iriseekhout/Agree')

Agreement for dichotomous outcomes

The use of the Agree package for the data example that was used in the paper on specific agreement on dichotomous outcomes in the situation of more than two raters. The method was fully explained an described in de Vet, Dikmans & Eekhout (2017).

Dichotomous example data

For the example we used data from a study by Dikmans et al. (2017). This data are based on photographs of breasts of 50 women after breast reconstruction. The photographs are independently scored by 5 surgeons, the patient, and three mothers. They each rated the quality of the reconstruction on a 5 point ordinal scale with the verbal anchors on the left side ‘very dissatisfied’ on the left end and on the right end ‘very satisfied’ on the right end. They specifically rated the volume, shape, symmetry, scars and nipple. In this paper we use the data of 4 surgeons because one surgeon had some missing values and we look at the rates for symmetry. The satisfaction scores were dichotomised into satisfied (scores 4 and 5) and not satisfied (scores 1,2, and 3).

data(breast)
variable <- "symmetry"
raters <- c("PCH1", "PCH2", "PCH3", "PCH4")
ratersvars <- paste(raters, variable, sep="_")
data1 <- data.frame(breast[ratersvars])
 for (r in 1:length(ratersvars)){
    data1[ratersvars[r]] <- ifelse(data1[ratersvars[r]]=="very satisfied"|data1[ratersvars[r]]=="satisfied","satisfied","not satisfied")
 }
data1 <- data.frame(apply(data1[ratersvars], 2, as.factor), stringsAsFactors = TRUE)

Agreement table

First the agreement tables are summed for all rater combinations into one agreement table. Then the off diagonal cells are averaged to obtain symmetry agreement tables. Note that data1 contains a column per rater for the variable of interest.

sumtable(data1,offdiag = FALSE) 
sumtable(data1,offdiag = TRUE) 

Agreement

From the agreement table we can calculate the agreement. And we can calculate the confidence interval around this agreement.

agreement(data1, confint = TRUE)

Specific agreement

The specific agreement for dichotomous data can be evaluated for the satisfied scores and for the not satisfied scores.

agreement(data1, specific="satisfied", confint = TRUE)
agreement(data1, specific="not satisfied", confint = TRUE)

Polytomous outcomes

The use of the Agree package for two data examples that are used in the paper on specific agreement on polytomous outcomes in the situation of more than two raters (de Vet, Mullender, Eekhout, 2018). The first data example is an example of ordinal ratings and the second example of nominal rating.

Ordinal example data

For the ordinal data example we use data from a study by Dikmans et al. (2017). This data is based on photographs of breasts of 50 women after breast reconstruction. The photographs are independently scored by 5 surgeons, the patients, and three mothers. They each rated the quality of the reconstruction on a 5 point ordinal scale with the verbal anchors on the left side ‘very dissatisfied’ on the left end and on the right end ‘very satisfied’ on the right end. They specifically rated the volume, shape, symmetry, scars and nipple. In this paper we use the data of 4 surgeons because one surgeon had some missing values and we look at the rates for symmetry. Data set 1 is used for the example of ordinal categories.

data(breast)

variable <- "symmetry"
raters <- c("PCH1", "PCH2", "PCH3", "PCH4")
ratersvars <- paste(raters, variable, sep="_")
data1 <- data.frame(breast[ratersvars])

Agreement table

First the agreement table are summed for all rater combinations into one agreement table. Then the off diagonal cells are averaged to obtain symmetry agreement tables.


sumtable(data1,offdiag = FALSE)
sumtable(data1,offdiag = TRUE) 

Agreement

From the agreement table we can calculate the agreement. And we can calculate the confidence interval around this agreement.

agreement(data1, confint = TRUE)

Specific agreement

The specific agreement for polytomous data, can be defined in two ways: the agreement of for one category versus not that category (e.g. very satisfied versus all other categories) or the agreement for one category versus any other (e.g. very satistfied versus satisfied). Below the Confidence intervals for the specific agreements are bootstrapped.

agreement(data1, specific="satisfied", confint = TRUE)
agreement(data1, specific=c("satisfied", "very satisfied"), confint = TRUE)
agreement(data1, specific= c("satisfied","neutral"), confint = TRUE)

Conditional probability

We can calulate the probability of any other outcome conditional on an specific outcome.

conditional.agreement(data1) 

Weighted agreement

For ordinal data it might also be useful to look at the agreement when they may be one category off. So the agreement plus or minus one category, that categories is weighted (default weight=1).

weighted.agreement(data1)
weighted.agreement(data1, weight=0.5)

Nominal example data

For the nominal data example we use a data set that was used in a paper by Fleis (1971). In this data patients are diagnosed in 5 categories: Depression, Personality Disorder, Schizophrenia, Neurosis, and Other by 6 raters.


data(diagnoses) 
data2 <- data.frame(lapply(diagnoses,as.factor), stringsAsFactors = TRUE)

 levels(data2$rater1) <- c("Depression", "Pers disord.", "Schizophrenia", "Neurosis", "Other")
 levels(data2$rater2) <- c("Depression", "Pers disord.", "Schizophrenia", "Neurosis", "Other")
 levels(data2$rater3) <- c("Depression", "Pers disord.", "Schizophrenia", "Neurosis", "Other")
 levels(data2$rater4) <- c("Depression", "Pers disord.", "Schizophrenia", "Neurosis", "Other")
 levels(data2$rater5) <- c("Depression", "Pers disord.", "Schizophrenia", "Neurosis", "Other")

Agreement table

First the agreement table are summed for all rater combinations into one agreement table. Then the off diagonal cells are averaged to obtain symmetry agreement tables.

sumtable(data2,offdiag = FALSE) 
sumtable(data2,offdiag = TRUE) 

Agreement

From the agreement table we can calculate the agreement. And we can calculate the confidence interval around this agreement.

agreement(data2, confint = TRUE)

Specific agreement

The specific agreement for polytomous data, can be defined in two ways: the agreement of for one category versus not that category (e.g. Depression versus all other categories) or the agreement for one category versus any other (e.g. Depression versus Schizophrenia). The confidence intervals for specific agreement are bootstrapped.

agreement(data2, specific="Depression", confint = TRUE)
agreement(data2, specific="Pers disord.", confint = TRUE)
agreement(data2, specific="Schizophrenia", confint = TRUE)
agreement(data2, specific="Neurosis", confint = TRUE)
agreement(data2, specific="Other", confint = TRUE)

Conditional agreement

conditional.agreement(data2) %>% kable()

Estimating ICCs and SEMs with multilevel models

The computational background and the use of the icc() function from the Agree package. We developed the icc() functions for this package in connection with a simulation study about sample size requirements for studies on reliability and measurement error (Mokkink et al, tbp) and a methodological paper about how to design and conduct a study on reliability and measurement error (Mokkink et al tbp).

Continuous example data

The intra-class agreement is usually obtained for continuous ratings. As an example we can use data from data study by Dikmans et al. (2017). This data is based on photographs of breasts of 50 women after breast reconstruction. The photographs are independently scored by 5 surgeons, the patients, and three mammography nurses. They each rated the quality of the reconstruction on a 5 point ordinal scale with the verbal anchors on the left side ‘very dissatisfied’ on the left end and on the right end ‘very satisfied’ on the right end. They specifically rated the volume, shape, symmetry, scars and nipple. For the icc examples we can use the sum scores for volume, shape, symmetry, scars and nipple as an overall rating from each rater.

breast_scores <- 
Agree::breast %>%
  dplyr::select(Patient_score, PCH1_score, PCH2_score, PCH3_score, PCH4_score, 
                PCH5_score, Mam1_score, Mam2_score, Mam3_score)

The example data shows missings. The icc function can deal with these missings, because a mixed model is used to estimate the variances to compute the icc with.

For a mixed model, the data needs to be restructured to a long format. we can use the pivot_longer() function from the tidyr package to do that:

breast_long <- breast_scores %>%
 mutate(id = 1:nrow(breast_scores)) %>% #add id column
  pivot_longer(cols = -id, names_to = "rater", values_to = "score")

breast_long

Variance components

The variances that are used to compute the icc are obtained from a linear mixed model. This model is estimated with the lmer() function from the lme4 package. In the breast example we have two levels: patients are level 1 and raters/observers are level 2. The two-level multilevel model is defined as Yijr = β0 + b0j + b0r + εijr, where β0 is the random intercept at the subject level and β0r the random intercept at the rater/observer level. The εijr is the residual error. The r-code for the model in lme4 is: lmer(score ~ (1|id) + (1|observer), data, REML = T)

De exact specification of the multilevel model, depends on the design of the study and the type of ICC that one wants to compute.

Types of ICC

There are three types of icc incorporated in the icc function. The ICC oneway, ICC agreement and the ICC consistency.

ICC oneway

The ICC type oneway is the variance between the subjects (σ2j) divided by the sum of the subject variance (σ2j) and the residual variance (σ2ε). The ICConeway is computed as follows:

ICConeway = σ2j / (σ2j + σ2ε)

The ICC oneway assumes that each subject is rated by a different set of raters, that are randomly selected from a larger population of judges (Shrout & Fleis, 1979). The icc_oneway() uses the varcomp() function to compute the variance components. These variances are estimated from a lmer model with random slope for the subjects. Yij = β0 + b0j + εij

The standard error of measurement (SEM) is the square root of the error variance (i.e. SEM = √σ2ε). The confidence intervals are computed with the exact F method. F = (k σ2j + σ2ε) / σ2ε, with df1 = n - 1 and df2 = n (k - 1) (Shrout & Fleis, 1979).

For the oneway ICC, only the level 1, the patient level, is random. The rater variance is not used.

The r-code to extract the variance component with the varcomp function is:

varcomp(score~(1|id), data = breast_long)

The variance components, can be used to compute the ICC oneway:

vc <- varcomp(score~(1|id), data = breast_long)
vc["id", "vcov"]/sum(vc[,"vcov"])

We have also incorporated a function that computed the ICC oneway directly from the data in the wide format, using the same steps. This is the icc_oneway function.

icc_oneway(breast_scores)

ICC agreement

The icc type agreement is the variance between the subjects (σ2j) divided by the sum of the subject variance (σ2j), rater variance (σ2k) and the residual variance (σ2ε). The ICCagreement is computed as follows:

ICCagreement = σ2j / (σ2j + σ2k + σ2ε)

The ICC for agreement generalizes to other raters within a population (Shrout & Fleis, 1979). All subjects are rated by the same set of raters, and the rater variance is taken into account in the calculation of the ICC. The variance components are computed with the icc_model() function. This is a lmer model with a random slope for the subjects and for the raters. The SEM is the square root of the sum of the rater variance and the error variance (i.e. SEM = √σ2r + σ2ε). The confidence intervals are approximated to account for the three independent variance components, as defined by Satterthwaite (1946) & Shrout & Fleis (1979).

For the ICC for agreement, both the level 1 and level 2 are random.

The r-code to extract the variance component with the varcomp function is:

varcomp(score ~ (1|id) + (1|rater), data = breast_long)

The variance components, can be used to compute the ICC agreement:

vc <- varcomp(score~ (1|id) + (1|rater), data = breast_long)
vc["id", "vcov"]/sum(vc[,"vcov"])

We have also incorporated a function that computed the ICC for agreement directly from the data in the wide format, using the same steps. This is the icc_agreement function.

icc_agreement(breast_scores)

ICC consistency

The ICC type consistency is the variance between the subjects (σ2j) divided by the sum of the subject variance (σ2j) and the residual variance (σ2ε). The rater variance is not used to calculate the ICC and can therefore also be considered as a fixed effect. The ICCconsistency is computed as follows:

ICCconsistency = σ2j  (σ2j + σ2ε)

The ICC for consistency generalizes only to the set of raters in the data (Shrout & Fleis, 1979). The varcomp() function is used to compute the variance components. These variances are computed from a a lmer model with a random slope for the subjects and a fixed effect for the raters. The sem is the square root of the error variance. The confidence are computed with the exact F method. F = (k σ2j + σ2ε) / σ2ε , with df1 = n - 1 and df2 = (n - 1) (k - 1) (Shrout & Fleis, 1979).

For the ICC for consistency, the level 1 is a random effect and the level 2 is fixed.

The r-code to extract the variance component with the varcomp function is:

varcomp(score ~ (1|id) + rater, data = breast_long)

The variance components, can be used to compute the ICC consistency:

vc <- varcomp(score~ (1|id) + rater, data = breast_long)
vc["id", "vcov"]/sum(vc[,"vcov"])

We have also incorporated a function that computed the ICC consistency directly from the data in the wide format, using the same steps. This is the icc_consistency function.

icc_consistency(breast_scores)

Comparing ICC types

We have one general icc function that computes all three ICC types for a data set. The differences in computations between the ICC types can quickly be seen in the variance components returned by the icc function. We can obtain the variances by using var = TRUE in the icc() function, the var_level2 shows the variance between the raters. Only for the ICC for agreement this variance component is estimated.

# ICC for all methods
icc(breast_scores, var = TRUE)

When we estimate the ICC for the surgeons only, we can see that the variance at the rater level is decreased. This effect is directly shown in the ICC.

In the icc we can also use the data in wide format and use the cols option to define the rater columns that we want to use.

# ICC for all methods
icc(breast_scores, 
    cols = c("PCH1_score", "PCH2_score", "PCH3_score", "PCH4_score", "PCH5_score"), 
    var = TRUE)

When we estimate the ICC for the mammography nurses only, we see that the variance at the rater level is increased. This effect is directly shown in the ICC.

# ICC for all methods
icc(breast_scores, 
    cols = c("Mam1_score", "Mam2_score", "Mam3_score"), 
    var = TRUE)

Three-way models

The three-way effects models are an extension of the two-way effects models with an extra random or fixed effect. The additional effect can be defined in the lmer formula in the varcomp function.

# variance components for a three-way model for agreement
varcomp(score ~ (1|id) + (1|rater) + (1|technician), data)

# variance components for a three-way mixed model for consistency with fixed rater and technician
varcomp(score ~ (1|id) + rater + technician, data)

# variance components for a three-way mixed model for consistency with fixed technician
varcomp(score ~ (1|id) + (1|rater) + technician, data)

Overview technical terms

Term Description
β0 Fixed intercept
b0j Random intercept at subject level
b0r Random intercept at rater level
εijr Residual error
σ2j Variance between subjects
σ2r Variance between raters
σ2ε Residual error variance
k Number of raters/observers
n Number of subjects