Multivariate random effects metaanalysis of diagnostic tests with multiple thresholds
 Taye H Hamza^{1, 2}Email author,
 Lidia R Arends^{1, 3},
 Hans C van Houwelingen^{4} and
 Theo Stijnen^{4}
DOI: 10.1186/14712288973
© Hamza et al; licensee BioMed Central Ltd. 2009
Received: 25 March 2009
Accepted: 10 November 2009
Published: 10 November 2009
Abstract
Background
Bivariate random effects metaanalysis of diagnostic tests is becoming a well established approach when studies present one twobytwo table or one pair of sensitivity and specificity. When studies present multiple thresholds for test positivity, usually metaanalysts reduce the data to a twobytwo table or take one threshold value at a time and apply the well developed metaanalytic approaches. However, this approach does not fully exploit the data.
Methods
In this paper we generalize the bivariate random effects approach to the situation where test results are presented with k thresholds for test positivity, resulting in a 2 by (k+1) table per study. The model can be fitted with standard likelihood procedures in statistical packages such as SAS (Proc NLMIXED). We follow a multivariate random effects approach; i.e., we assume that each study estimates a study specific ROC curve that can be viewed as randomly sampled from the population of all ROC curves of such studies. In contrast to the bivariate case, where nothing can be said about the shape of study specific ROC curves without additional untestable assumptions, the multivariate model can be used to describe study specific ROC curves. The models are easily extended with study level covariates.
Results
The method is illustrated using published metaanalysis data. The SAS NLMIXED syntax is given in the appendix.
Conclusion
We conclude that the multivariate random effects metaanalysis approach is an appropriate and convenient framework to metaanalyse studies with multiple threshold without losing any information by dichotomizing the test results.
Background
Metaanalysis of diagnostic accuracy studies depends on the type of data that is available from different studies. The most frequently reported measures of diagnostic test accuracy are sensitivity and specificity or a two by two table, i.e. with a single threshold value. Metaanalytic methodologies for such kind of data have been developed to summarize sensitivity and specificity separately or jointly in a fixed or random effects context, for example [1–6]. In recent years the bivariate random effects metaanalysis of diagnostic tests has become a well established approach, which can be fitted in many statistical packages [1, 2]. The bivariate approach has many advantages over separate random effects metaanalysis of sensitivity and specificity and the traditional summary receiver operating characteristics (SROC) method of Littenberg and Moses [1, 2, 4]. Besides it is flexible to derive different outcome measures, such as overall sensitivity and/or specificity, diagnostic odds ratio and SROC curves, from the estimated parameters.
In this article we consider the situation where diagnostic results for the test under evaluation are reported in three or more categories; for example, disease severity classified as malignant, suspect or benign. One straightforward approach often followed in practice is to dichotomize the test results into two categories and apply the well developed bivariate methods separately for each of the thresholds. When data is presented for many thresholds, a ROC can be calculated per study, and metaanalytic methods have been developed to derive a SROC from them [7–9]. Poon [10] discusses a latent normal distribution model for analysing ordinal responses with applications in metaanalysis. This model can be applied to multiple threshold diagnostic metaanalysis data. However, this paper only considered fixed effect modeling. Bipat et al [11] discussed a multivariate randomeffects approach for metaanalysis of cancer staging. They assumed the correlations between different cancer stages to be independent, i.e. SROC curves can not be derived from their model. Specifically for diagnostic accuracy studies, Dukic et al [12] discussed both ordinal regression and hierarchical approaches based on latent variable modeling.
The above approaches are no direct extensions of the nowadays popular bivariate metaanalysis approach for the one threshold case. The aim of this article is to generalize this approach to the situation where test results are presented using more than one threshold or in more than two categories. Not necessarily in all studies the same number of categories is presented, however, in this article we focus on the case where the number of categories is equal across studies and we discuss how our approach can accommodate studies reported with unequal number of thresholds. Our approach can be implemented in standard statistical packages. In the methods section we briefly review the bivariate random effects approach, and we introduce the multivariate approach to metaanalyse studies that report test results with more than one threshold. We illustrate the methods using published metaanalysis data and end with a discussion.
Methods
The bivariate random effects (BREM) approach
For the situation where each study presents one pair of sensitivity and specificity with corresponding standard errors, the bivariate metaanalysis approach [13] has become a well established method [1, 2, 14]. The approach preserves the twodimensional nature of the original data taking into account the betweenstudies correlation of sensitivity and specificity. It can be seen as an improvement on the method of Littenberg and Moses [4], which has been the standard method to construct a SROC for more than a decade.
In this section first we will introduce the bivariate random effects model (BREM) in its standard form. Subsequently we will derive another form of the model, which starts from a model for study specific ROCs and has a different parametrization. This formulation of the model is the natural one to generalize to the case where we have two or more pairs of specificity and sensitivity per study. This formulation also sheds more light on the interpretation of SROCs, which is problematic in the case where only one pair of sensitivity and specificity is available.
For study i, denote ξ_{ i }= logit(1  specificity_{ i }) and η_{ i }= logit(sensitivity_{ i }). Let x_{1i}be the number of true positives, n_{1i}the total number of diseased subjects, x_{0i}the number of false positives and n_{0i}the total number of nondiseased subjects. Then the observed sensitivity and specificity for a given study i are x_{1i}/n_{1i}and (n_{0i}x_{0i})/n_{0i}respectively. Note that the underlying sensitivity and specificity tend to be negatively correlated across studies because of explicit or implicit differences in the thresholds. Therefore ξ_{ i }and η_{ i }will tend to be positively correlated.
Betweenstudies variability
Different choices of summary lines as resulting from the BREM:y = α + βx where y = logit(sensitivity) and x = logit(1specificity). []
Parameter  Type of regression line  

ηon ξ  ξon η  Don S  R & G  Major Axis  
β 





α  βξ 
The different SROCs can be vastly different in applications, see for instance Arends et al [2] and the data examples. The BREM approach as introduced by Reitsma et al [1] and discussed by Arends et al [2] does not assume anything about study specific curves. The method simply leads to an estimated underlying bivariate distribution of the true sensitivities and specificities as reported by the different studies included in the metaanalysis. This means that the chosen SROC does not necessarily correspond with the true curves of the studies. The true study specific curves might have a substantially different shape, and the SROC cannot be interpreted as a kind of average or overall ROC representative for the ROCs of the different studies. There might even be no study specific curves at all, in case the diagnostic test cannot be thought of as a continuous test. However, this does not mean that the analysis does not make sense in this case, since the existence of study specific ROC curves is not assumed by the method. In the remainder of this section we introduce a new formulation of the BREM, which starts with the study specific ROCs. This will make clear under which extra assumption the BREM describes the distribution of study specific ROCs and the calculated SROC can be considered to be a real overall SROC.
This model is just the same as (1), only with a different parametrization. However, the number of parameters is one more, which means that one of them is unidentifiable. To make the model identifiable, we need a further assumption on how the ξ_{ i }'s in the different studies are selected. For instance we could assume that σ_{ αξ }is zero. This means that the individual investigators, in selecting their ξ_{ i }value, are not lead by the level of their line. However it is perfectly conceivable that an investigator who happens to have a ROC that is relatively low, tends to choose a relatively high value for his ξ_{ i }if a high sensitivity is preferred, or just a relatively low value of ξ_{ i }if high specificity is preferred.
If we assume that the correlation between α_{ i }and ξ_{ i }is zero, it can be seen that β is given by the slope of the regression line of η on ξ. In this case the η on ξ type SROC is the true SROC in the sense that it really can be interpreted as such. In the (ξ, η) space it is just the average line over the population of studies, in the ROC space it can be interpreted as a kind of median ROC.
Another assumption could be that the correlation between η and α is zero. This means that we assume and therefore the correlation between α and ξ_{ i }is given by ρ_{ αξ }= σ_{ α }/(βσ_{ ξ }). The slope, then, is rewritten as , which is the slope of the regression of ξ on η. Thus under this assumption the ξ on η type SROC is the real one.
More general, we could assume that some linear combination aξ + bη of ξ and η is not correlated to α, for some value of a and b. We have already seen that if a = 1 and b = 0, the η on ξ type SROC is the correct one. If a = 0 and b = 1, then the ξ on η type is the correct one. If we assume a = b = 1, then one can check that β is equal to the slope of the regression of D = η ξ on S = η + ξ, and the Littenberg & Moses type SROC is the correct one. One can also check that the assumption a = β and b = 1 leads to the Rutter & Gatsonis (R&G) type SROC. The slope of their method is the geometric average of the η on ξ and ξ on η regression line, that is, their regression line always lies in between the two curves. For a detailed discussion we refer to Arends et al [2]
 1.
The η on ξ, ξ on η and major axis methods give an improper SROC curve if and only if sensitivity and specificity are positively correlated.
 2.
The D on S method give an improper SROC curve if and only if σ_{ ξη }<  and σ_{ ξη }>  or σ_{ ξη }>  and σ_{ ξη }<  . In either cases sensitivity and specificity are positively correlated.
Notice that the R&G method always yields a proper SROC because it implies that the slope parameter is always positive, being equal to the ratio of the between studies standard deviations of sensitivity and specificity. It is remarkable that the R&G SROC can be calculated from two simple univariate metaanalyses on the sensitivities and specificities separately. This is an advantage since no bivariate modeling is needed at all, but it might also question its value because apparently it does not use the possible correlation between sensitivity and specificity. We conclude that in the situation where we have only one pair of sensitivity and specificity per study a calculated SROC can only be interpreted as a real overall ROC under an untestable assumption. The assumption is especially sensitive when the differences among the estimated betweenstudies variances and covariance of sensitivity and specificity are large. This issue seems to have been overlooked in the literature. In a recent letter in Biostatistics Chu and Guo [15] claim that the SROC given by Harbord et al [14] and Rutter and Gatsonis [6] are "are incorrect and potentially misleading". Chu and Guo assume that the η on ξ type is the "real" SROC, but they forget that this one is based on an untestable assumption too. However, the situation changes as soon as more pairs of sensitivity and specificity are available per study.
Withinstudy variability
The withinstudy variability can be modeled using an approximate normal distribution [1, 2] or a binomial distribution [2, 14, 16]. Hamza et al [17, 18] compared in extensive simulation experiments the binomial and approximate normal withinstudy models, and showed that in general the performance of the binomial withinstudy model is much better. Chu and Cole [16] also showed similar results using a selected number of simulations. Therefore in this paper we restrict to the binomial withinstudy model. For the approximate approach we refer to [1, 2].
The ξ_{ i }and η_{ i }= α_{ i }+ βξ_{ i }are the true logit transformed (1specificity) and sensitivity from (2). The bivariate model given by (24) can be fitted using generalized linear mixed model procedures in standard statistical packages, such as the SAS procedure NLMIXED, STATA gllamm or the R/SPlus program nlme.
Multivariate random effects metaanalysis (MREM)
Twobythree contingency table for study i for relating the FNAC outcome to the final diagnosis of breast lesion (The FNAC data is given in the additional file 1).
FNAC outcome  Malignant  Suspect  Benign  Total 

Final diagnosis  
Malignant  x _{1i1}  x _{1i2}  x _{1i3}  n _{1i} 
Benign  x _{0i1}  x _{0i2}  x _{0i3}  n _{0i} 
The model
Let the true logit transformed 1  specificity and sensitivity for a given threshold j be denoted by ξ_{ ij }and η_{ ij }respectively, where ξ_{ ij }'s and η_{ ij }'s are ordered in the j index. We assume a hierarchical model that is a direct generalization of model (23). In contrast to the one threshold case (bivariate approach), when we have more than one threshold (multiple points per study), the SROC curve is identifiable. The between and withinstudy models are given as follows:
Betweenstudies model
 1.
Model for the relation between ξ_{ ij }and η_{ ij }:
 2.Model for the ξ_{ ij }'s:(6)
Note that the covariance structure for the ξ_{ ij }'s is of compound symmetry, i.e. the betweenstudies variances of ξ_{ ij }'s are assumed to be equal and the covariances between any of the ξ_{ ij }'s are assumed to be the same. We have chosen this structure since it is popular in repeated measures modeling as a simple but often realistic covariance structure. Moreover the parameters have a nice interpretation. However, one can choose any structure. The assumption of compound symmetry might be strong; for example, the covariance (correlation) between consecutive thresholds may be larger compared to between nonconsecutive ones. If so, a richer structure is needed. In general, the betweenstudies variances and covariances of ξ_{ ij }'s could be allowed to be all different (unstructured covariance matrix) and, in the spirit of the general guidelines for mixed model building given by Verbeeke and Molenberghs [23](chapter 9), it could then be simplified into another covariance structure, such as Toeplitz, autoregressive or compound symmetry.
Withinstudy model
Note that ξ_{ ij }is logit(1specificity) and η_{ ij }is logit(sensitivity) for threshold j, and the cell probabilities for diseased (or nondiseased) subjects for category j is the difference of sensitivities (or 1specificities) between category j and j  1.
As seen from 7, we assumed a multivariate normal distribution for the random effects. However, the density g(B) can also be assumed to belong to some other parametric family of distributions [24]. In summary the MREM model allows to calculate different relevant measures of diagnostic test accuracies:

The mean logit sensitivity and logit specificity along with their standard errors for any known threshold, j, are estimated by the model. One can then derive sensitivity and specificity as follows: sensitivty_{ j }= and specificity_{ j }= , and the corresponding standard errors can be calculated using delta method [25]. SAS NLMIXED users avoid hand calculation by using the 'Estimate' statement (see the SAS syntax for example).

The diagnostic odds ratio (DOR), that can be derived from sensitivity and specificity, for a given threshold, j, is given by DOR_{ j }=

The estimated parameters from model (5  7) are used to derive the overall median SROC curve that is given by sensitivity(specificity) = . Besides, study specific ROC curves can be generated from the empirical Bayes estimates of the random effects model. In SAS NLMIXED, the empirical Bayes estimates are generated automatically and collected from the output file ('out') specified in the 'random' statement (see SAS syntax).

Uncertainty around the SROC can be characterized by calculating the confidence interval at each point along the curve.

A prediction band for the true ROC curve of a new study can be calculated by adding and subtracting 1.96 times the estimated standard deviation of α_{ i }in the equation above.

The model accommodates study level covariates, for example corresponding to two different diagnostic tests, that enables to test the hypothesis for the differences between groups, e.g. diagnostic tests. A thorough discussion of comparison between groups or test results using the bivariate model is given in Hamza et al[26]. In the second data example of this paper we show how to test for the significant difference between groups.
Fitting the model
 1.
We fitted the model using Proc NLMIXED of SAS. This procedure does not support directly the multinomial distribution. However, the procedure allows a user specified loglikelihood function. This is easily done for the multinomial distribution. In the appendix the syntax for an example is given. The NLMIXED procedure calculates the likelihood function by numerical integration, using adaptive Gaussian quadrature. The number of quadrature points is specified by the user or automatically by SAS. The larger that number is chosen the better the approximation, but at the cost of more computational time. For a detailed discussion of different choices of quadrature options, optimization methods and convergence criteria we refer to the SAS manual [27]. In practice, when a final model is reached, one increases the number of quadrature points until the parameter estimates do not change anymore.
 2.NLMIXED allows user specified likelihoods, but many other programs do not. Usually the binomial distribution is supported, therefore we also mention another possibility to fit the model in Generalized Linear Mixed Model programs. The trick is to write the multinomial pdf as a sequence of conditional univariate pdf's, i.e. the pdf of (x_{Di1}, ⋯, x_{ DiJ }) is expressed as:
where π_{0ij}and π_{1ij}are calculated as in (8 & 9) with j = 1, ..., J  1.
Results
To illustrate the methods discussed in this article, we apply them to two published metaanalysis datasets. One is relatively large (29 studies) with three test result categories (2 thresholds). The second data set is small (10 studies) with five test result categories (4 thresholds). Here our objective is to fit the models discussed in the methods section, and to derive the SROC curves.
Example 1: Fineneedle aspiration cytologic examination
Giard and Hermans [19] present 29 studies evaluating the accuracy of fineneedle aspiration cytologic examination (FNAC) of the breast to assess the presence of breast cancer (see additional file 1. FNAC provides a nonoperative way of obtaining cells for the establishment of the nature of a breast lump and therefore plays a pivotal role in the preoperative diagnostic process [19, 20]. The selected FNAC results were classified in the following four cytologic categories: definitely malignant, suspect for malignancy, benign, and unsatisfactory specimen for diagnosis (acellular aspiration). The last category is for those who do not have satisfactory specimen for diagnosis. Following the authors, we merged this group with the benign group, which resulted in a two by three table (Table 2).
Parameter estimates (standard errors) and AUC of the SROC curves from the BREM and MREM approaches for the FNAC data
Type of SROC  α  β  AUC 

BREM  
η on ξ  2.110(0.321)  0.107(0.118)  0.882 
ξ on η  7.636(6.307)  2.276(2.463)  0.955 
D on S  2.643(0.371)  0.316(0.137)  0.918 
Rutter and Gatsonis  3.094(0.319)  0.493(0.112)  0.935 
Major axis  2.191(0.406)  0.138(0.153)  0.889 
MREM  2.368(0.135)  0.224(0.016)  0.902 
Next we fitted the MREM method. The estimated median specificities (standard error) at the first and second thresholds respectively are 0.999(0.0003) and 0.928(0.018), and the corresponding sensitivities can be estimated by and are 0.687(0.028) and 0.858(0.016). Note that the estimated median specificity and sensitivity at the second threshold are approximately equal to the BREM estimates. The estimated variances and covariances were = 0.363(0.117), σ_{αΔ }= 0.045(0.143), σ_{Δ} = 0.042(0.443) and = 1.841(0.615). The test for the significance of the covariance between α_{ i }and ξ_{ i }in the MREM approach is not significant (likelihood ratio = 0.999, pvalue = 0.317), and hence there is no indication for the choice of specificities to depend on the level of individual curves. The estimates for the intercept and slope, and AUC are given in table 3. The SROC from the MREM is depicted in figure 1. Comparing the estimated SROC curves from the two approaches, one can see that the BREM approach underestimates or overestimates the SROC curves depending on the choice of the type of SROC. This can be seen clearly from the AUC of the SROC curves in table 3. Of course, if we had chosen another cut point for the BREM, we would have ended up with a different SROC estimate for each of the choices. By the delta method it is possible to construct a confidence band for the SROC curve, as explained above. We did not do here, to avoid a too busy picture.
Example 2: CAGE in screening for alcoholism
Parameter estimates (standard errors) and AUC (for nonprimary care patients (nonPC), and for primary care patients (PC)) of the SROC curves from the BREM and MREM approaches for the CAGE dataset
Type of SROC  α  β  γ  AUC _{ nonPC }  AUC _{ PC } 

BREM  
η on ξ  2.775(0.712)  0.840(0.316)  0.050(0.389)  0.886  0.890 
ξ on η  3.618(0.827)  1.235(0.364)  0.113(0.479)  0.902  0.908 
D on S  3.160(0.710)  1.020(0.314)  0.079(0.419)  0.895  0.900 
R & G  3.156(0.657)  1.019(0.286)  0.079(0.417)  0.895  0.900 
Major Axis  3.165(0.776)  1.023(0.347)  0.079(0.420)  0.895  0.900 
MREM  2.537(0.312)  0.795(0.047)  0.207(0.382)  0.849  0.888 
The twobyfive tables from the CAGE metaanalysis were also analyzed using the MREM approach. Here the betweenstudies model in (5) can be rewritten as η_{ ij }= α_{ i }+ β_{ i }ξ_{ ij }+ γz_{1} to adjust for the covariate z_{1}. Note that the slope parameter is random and assumed to be independent of α_{ i }and ξ_{ ij }; i.e. β_{ i }~N ( , ). In fact the ξ_{ ij }can also be adjusted for study level covariates or the interaction between ξ_{ ij }and z's could be added, but we did not do that in this example. First we tested the random slope assumption ( ≠ 0) with the likelihood ratio test using mixture of chisquare test statistics. It turned out the null hypothesis that the slope parameter is not random was not rejected (p = 0.19), and therefore we fitted the MREM model assuming fixed slope parameter. The estimated parameter for primary versus nonprimary population is 0.207(0.382), P = 0.589. Therefore there is no significant difference between the two group of population. The estimated median specificities (standard error) at thresholds 14 respectively are 0.994(0.002), 0.976(0.006), 0.902(0.021) and 0.742(0.045). And the estimated median sensitivities (standard error) are 0.172(0.041), 0.403(0.068), 0.684(0.061) and 0.846(0.038) for nonprimary care patients; and 0.203(0.046), 0.453(0.069), 0.727(0.056) and 0.871(0.033) for primary care patients. The estimated variances and covariances are = 0.392(0.211), σ_{αΔ }= 0.217(0.178), σ_{Δ} = 0.463(0.226) and = 0.036(0.022). The test for the correlation between the random intercept, α_{ i }and ξ_{ i }is not significant (χ^{2} = 2.1, pvalue = 0.147). Therefore there is no indication for the choice of the ξ_{ i }'s to depend on the level of individual curves. The estimated SROC parameters are given in table 4.
As shown in figure 3 and the AUCs from table 4 the bivariate approach seems to over estimate the SROC curve, for any of the 5 choices of the type of the SROC. Again this would possibly be changed if we choose another cutoff point for positivity on the screening test for alcoholism.
Discussion
The summary ROC curve has been introduced as a way to assess the diagnostic accuracy of a diagnostic test in a metaanalysis [4, 5, 7, 29, 30]. For the most frequent situation, when one point per study is presented, the medical (and statistical) articles seem to have overlooked the problems inherent to SROCs based on studies with only one point. Although recent developments in the area have shown that the bivariate random effects metaanalysis approach has important advantages over the standard SROC approach of Littenberg and Moses [1, 2, 4], the problem of identifiability and therefore interpretability of the resulting SROC remains. When studies present more than one point per study, commonly the test results are reduced to two categories and metaanalysed using a well established approach such as the BREM, which is a suboptimal approach. In our data examples we illustrated this by considering a single cutoff value and applying the BREM approach. The results from the two data examples showed that differences between the estimated SROC curves based on the BREM approach can be large, as in the first example, or relatively small, as in the second example. The sizes of the differences depend on the values of the three covariance parameters. The η on ξ and ξ on η curves are always most extreme in the sense that the other three lie between them. Therefore a necessary and sufficient condition for the 5 different curves to be equal is the correlation being one, which is not very probable in practical situations. Equality of the variances of ξ_{ i }and η_{ i }is a sufficient condition for equality of the SROCs from the three intermediate approaches (D on S, Rutter and Gatsonis, and major axis). In the first example the variances differ a factor 4 and the correlation is relatively small (0.22). In the second example the variances are almost equal and the correlation is relatively large (0.82). This explains why the differences are large in the first and small in the second example.
In this article we generalized the BREM approach for one threshold to the situation where more than one point per study is available and the number of thresholds is equal across studies. In our opinion, the MREM approach is relatively easy to understand and has several advantages. First, data of the full 2 by k table is used without losing any information by dichotomizing the test results. Second, different outcome measures can be derived from the fitted model, such as SROC curves and overall sensitivity and/or specificity for any choice of the threshold. Third, in contrast to the BREM approach, the summary ROC and the study specific curves are identifiable. Fourth, the model is symmetric in the ξ_{ ij }'s and η_{ ij }'s. Interchanging their role leads to the same model. Fifth, it is straightforward to include study level covariates. They can be added directly to the intercept and slope of the SROC, and also to the threshold values. Sixth, the MREM can be fitted in standard statistical packages without extra programming. In equation (7), we specified compound symmetry for the covariance structure of the ξ_{ ij }'s. However, one can also choose another, possibly richer structure and simplify it using the likelihood ratio test. However, more covariance parameters gives more risk of nonconvergence problems. For the withinstudy model we used the multinomial distribution instead of the summary statistic approach usually followed in metaanalysis. This avoids problems with small numbers and zero cells are allowed.
We used NLMIXED from SAS to fit our models and noticed that convergence of the program is sensitive for starting values. It turned out that good starting values are obtained by fitting the BREMs according to all possible cutoff values.
Related work was done by Dukic and Gatsonis [12]. They used ordinal regression and a hierarchical approach based on latent variable modeling and fitted their model by Bayesian methods. To our knowledge their approach is rarely used in practice, probably due to the inherent complexity of the model and fitting methods. The difference between Dukic and Gatsonis model and the MREM is mainly in the modeling of the ξ_{ ij }'s. They treated them all as fixed parameters, leading to as many ξ parameters as there are data points, while we modeled them using the standard multivariate metaanalysis model [13]. The motivation for this is to reduce the number of parameters and to correct for the measurement errors in the 's. In our opinion, Dukic's method does not correct for measurement error in the 's and leads to an inconsistent estimate of the summary ROC curve, for reasons set out in Van Houwelingen and Senn [31]. Another difference is that Dukic's method assumes independence between the choice of the specificities as represented by the ξ_{ ij }parameters and the level of the study specific ROC curve as determined by α_{ i }; a not necessarily realistic assumption.
Model Extension and Limitation
In this paper we focused on the situation where test results are presented with equal numbers of thresholds, but the method is more general. In practice test results can possibly be presented with different numbers of thresholds. Often the different numbers of categories arises while all studies in principle use the same categorization, but in different studies different categories are lumped together. Then our model can still be applied without any modification, since our approach allows missing data points. Many other situations can be covered by allowing the parameters of the model for the ξ_{ ij }'s (6) to be different for different groups of studies. For instance, suppose the goal of the metaanalysis is to compare two tests A and B, one with 3 and the other with 5 categories, and each study reporting either A or B. Then the ξ_{ ij }model for A can be specified to be completely different from that of B. Such a model can still be estimated in for instance using Proc NLMIXED. The total number of different thresholds across all studies is the limiting factor in our approach. If it is too large, the number of parameters might be too large to estimate and the likelihood method may not work properly.
Conclusion
The multivariate random effects metaanalysis approach is an appropriate and convenient framework to metaanalyse studies with multiple threshold without losing any information by dichotomizing the test results. The model can be implemented on standard statistical packages.
Appendix
SAS syntax to fit the MREM model by writing the likelihood in the SAS procedure NLMIXED
*PROC NLMIXED = call SAS procedure NLMIXED, and the options included represents:
df = degree of freedom, we specified 1000 to get Wald test in stead of ttest
qpoints = number of quadrature points, if not specified SAS automatically does
miniter = is the minimum number of iteration;
PROC NLMIXED DATA = cage DF = 1000 MINITER = 30 QPOINTS = 5;
*Specifies initial values for the maximum likelihood estimates
if not specified, all the initial values automatically assigned to be one;
PARMS ma = 2.6 b = 0.8 b2 = 0 mxi1 = 5.2 mxi2 = 3.7 mxi3 = 2.2 mxi4 = 1.0
va = 0.4 cavdi = 0.2 vdi = 0.5 vdij = 0.02;
*Constraint to make sure that xi's are in the right order;
BOUNDS mxi1mxi2< = 0, mxi2mxi3< = 0, mxi3mxi4< = 0;
*Betweenstudies model (equation ( 5 ));
eta1 = a + b*xi1 + b2*z1;
eta2 = a + b*xi2 + b2*z1;
eta3 = a + b*xi3 + b2*z1;
eta4 = a + b*xi4 + b2*z1;
*Withinstudy model (equation ( 8 ));
p01 = 1/(1+exp((xi1)));
p02 = 1/(1+exp((xi2)))  1/(1+exp((xi1)));
p03 = 1/(1+exp((xi3)))  1/(1+exp((xi2)));
p04 = 1/(1+exp((xi4)))  1/(1+exp((xi3)));
p05 = 1  1/(1+exp((xi4)));
*Withinstudy model (equation ( 9 ));
p11 = 1/(1+exp((eta1)));
p12 = 1/(1+exp((eta2)))  1/(1+exp((eta1)));
p13 = 1/(1+exp((eta3)))  1/(1+exp((eta2)));
p14 = 1/(1+exp((eta4)))  1/(1+exp((eta3)));
p15 = 1  1/(1+exp((eta4)));
*Loglikelihood (equation ( 10 ));
if (p01^ = 0 and p02^ = 0 and p03^ = 0 and p04^ = 0 and p05^ = 0 and
p11^ = 0 and p12^ = 0 and p13^ = 0 and p14^ = 0 and p15^ = 0) then
ll = n01*log(p01)+n02*log(p02)+n03*log(p03)+n04*log(p04)+n05*log(p05)+
n11*log(p11)+n12*log(p12)+n13*log(p13)+n14*log(p14)+n15*log(p15);
else ll = 1**100;
*SAS maximized the likelihood using the general distribution;
MODEL n11 ~general(ll);
*Random effects parameters alpha_i and xi_ij's are normally distributed around their mean and betweenstudies variance covariance matrix (equation 7);
RANDOM a xi1 xi2 xi3 xi4 ~normal([ma, mxi1, mxi2, mxi3, mxi4],
[va,
cavdi, vdi+vdij,
cavdi, vdi, vdi+vdij,
cavdi, vdi, vdi, vdi+vdij,
cavdi, vdi, vdi, vdi, vdi+vdij])
SUBJECT = study OUT = study_specific; *Empirical Bayes estimate;
*Estimate sensitivities and specificities at each of known thresholds (1,2,3,4);
ESTIMATE 'sensitivity_1' exp(ma + mxi1)/(1 + exp(ma + mxi1));
ESTIMATE 'sensitivity_2' exp(ma + mxi2)/(1 + exp(ma + mxi2));
ESTIMATE 'sensitivity_3' exp(ma + mxi3)/(1 + exp(ma + mxi3));
ESTIMATE 'sensitivity_4' exp(ma + mxi4)/(1 + exp(ma + mxi4));
ESTIMATE 'specificity_1' exp(1)/(1 + exp(mxi1));
ESTIMATE 'specificity_2' exp(1)/(1 + exp(mxi2));
ESTIMATE 'specificity_3' exp(1)/(1 + exp(mxi3));
ESTIMATE 'specificity_4' exp(1)/(1 + exp(mxi4));
run;
Abbreviations
 ROC:

Receiver operating characteristics
 SROC:

Summary receiver operating characteristics
 BREM:

Bivariate random effects model
 MREM:

Multivariate random effects metaanalysis
 FNAC:

Fineneedle aspiration cytologic examination
 AUC:

Area under the curve.
Declarations
Acknowledgements
Taye H. Hamza was financially supported by a program grant from The Netherlands Organization for Health Research and Development (ZonMW 91204016, Advanced Metaanalytical methods for medical data)
Authors’ Affiliations
References
 Reitsma JB, Glas AS, Rutjes AWS, Scholten RJPM, Bossuyt PM, Zwinderman AH: Bivariate analysis of sensitivity and specificity produces informative summary measures in diagnostic reviews. Journal of Clinical Epidemiology. 2005, 58: 982990. 10.1016/j.jclinepi.2005.02.022.View ArticlePubMed
 Arends LR, Hamza TH, Van Houwelingen JC, Heijenbrokkal , Hunink MGM, Stijnen Th: Multivariate random effects metaanalysis of ROC curves. Medical Decision Making. 2008, 28 (5): 621638. 10.1177/0272989X08319957.View ArticlePubMed
 DerSimonian R, Laird N: Metaanalysis in clinical trials. Controlled Clinical Trials. 1986, 7 (3): 177188. 10.1016/01972456(86)900462.View ArticlePubMed
 Littenberg B, Moses LE: Estimating diagnosticaccuracy from multiple conicting reports  a new metaanalytic method. Medical Decision Making. 1993, 13: 313321. 10.1177/0272989X9301300408.View ArticlePubMed
 Moses LE, Shapiro D, Littenberg B: Combining independent studies of a diagnostic test into a summary ROC curve: Dataanalytic approaches and some additional considerations. Statistics in Medicine. 1993, 12: 12931316.View ArticlePubMed
 Rutter CM, Gatsonis CA: A hierarchical regression approach to metaanalysis of diagnostic test accuracy evaluations. Statistics in Medicine. 2001, 20: 28652884. 10.1002/sim.942.View ArticlePubMed
 Irwig L, Tosteson ANA, Gatsonis C, Lau J, Colditz G, Chalmers TC, Mosteller F: Guidlines for Metaanalyses Evaluating Diagnostic Tests. Annals of Internal Medicine. 1994, 120: 667676.View ArticlePubMed
 Tosteson AN, Begg CB: A general regression methodology for ROC curve estimation. Medical Decision Making. 1988, 8: 20415. 10.1177/0272989X8800800309.View ArticlePubMed
 Kester ADM, Buntinx F: Metaanalysis of ROC curves. Medical Decision Making. 2000, 20: 430439. 10.1177/0272989X0002000407.View ArticlePubMed
 Poon WY: A latent normal distribution model for analysing ordinal responses with applications in metaanalysis. Statistics in Medicine. 2004, 23: 21552172. 10.1002/sim.1814.View ArticlePubMed
 Bipat S, Zwinderman AH, Bossuyt PMM, Stoker J: Multivariate RandomEffects Approach: For MetaAnalysis of Cancer Staging Studies. Acad Radiol. 2007, 14: 974984. 10.1016/j.acra.2007.05.007.View ArticlePubMed
 Dukic V, Gatsonis C: Metaanalyis of Diagnostic test accuracy assessment studies with varying number of thresholds. Biometrics. 2003, 59: 936946. 10.1111/j.0006341X.2003.00108.x.View ArticlePubMed
 van Houwelingen HC, Arends LR, Stijnen T: Advanced methods in metaanalysis: multivariate approach and metaregression. Statistics in Medicine. 2002, 21 (4): 589624. 10.1002/sim.1040.View ArticlePubMed
 Harbord RM, Deeks JJ, Egger M, Whiting P, Sterne JA: A unification of models for metaanalysis of diagnostic accuracy studies. Biostatistics. 2006, 1 (1): 121.
 Chu H, Guo H: Letter to the Editor. Biostatistics. 2009, 10 (1): 201203. 10.1093/biostatistics/kxn040.PubMed CentralView ArticlePubMed
 Chu H, Cole SR: Bivariate metaanalysis for sensitivity and specificity with sparse data: a generalized linear mixed model approach (letter to the Editor). Journal of Clinical Epidemiology. 2006, 59: 13311331. 10.1016/j.jclinepi.2006.06.011.View ArticlePubMed
 Hamza TH, van Houwelingen HC, Stijnen T: Random effects metaanalysis of proportions: The binomial distribution should be used to model the withinstudy variability. Journal of Clinical Epidemiology. 2008, 61 (1): 4151. 10.1016/j.jclinepi.2007.03.016.View ArticlePubMed
 Hamza TH, Reitsma JB, Stijnen T: Metaanalysis of diagnostic studies: a comparison of random intercept, normalnormal and binomialnormal bivariate Summary ROC approaches. Medical Decision Making. 2008, 28 (5): 639649. 10.1177/0272989X08323917.View ArticlePubMed
 Giard RWM, Hermans J: The value of Aspiration Cytologic Examination of the Breast. A Statistical Review of the Medical Literature. Cancer. 1992, 62: 21042110. 10.1002/10970142(19920415)69:8<2104::AIDCNCR2820690816>3.0.CO;2O.View Article
 Mushlin AI: Diagnostic tests in breast cancer: Clinical strategies based on diagnostic probabilities. Annals Internal Medicine. 1985, 103: 7985.View Article
 Riley RD, Abrams KR, Sutton AJ, Lambert PC, Thompson JR: Bivariate randomeffects metaanalysis and the estimation of betweenstudy correlation. BMC Medical Research Methodology. 2007, 7 (3): 115.
 Aergeerts B, Buntinx F, Kester A: The value of the CAGE in screening for alcohol abuse and alcohol dependence in general clinical populations: a diagnostic metaanalysis. Journal of Clinical Epidemiology. 2004, 57: 3039. 10.1016/S08954356(03)002543.View Article
 Verbeeke G, Molenberghs G: Linear Mixed Models for Longitudinal Data. 2000, NewYork: SpringerVerlag
 Molenberghs G, Verbeke G: Models for Discrete Longitudinal Data. 2005, NewYork: SpringerVerlag
 Cox C: "Delta Method," Encyclopedia of Biostatistics. Edited by: Peter Armitage, Theodore Colton. 1998, New York: John Wiley, 11251127.
 Hamza TH, van Houwelingen HC, Heijenbrokkal MH, Stijnen T: Associating explanatory variables with summary receiver operating characterstics curves in diagnostic metaanalysis. Journal of Clinical Epidemiology. 10.1016/j.jclinepi.2009.02.002.
 SAS Institute Inc: SAS/STAT(r) 9.1 User's Guide. 2004, Cary, NC: SAS Institute Inc
 Agresti A: Categorical Data Analysis. 2002, Wiley & Sons, Inc., New JerseyView Article
 Irwig L, Macaskill P, Glasziou P, Fahey M: Metaanalytic methods for diagnostic test accuracy. Journal of Clinical Epidemiology. 1993, 48 (1): 119130. 10.1016/08954356(94)00099C.View Article
 Midgette AS, Stukel TA, Littenberg B: A metaanalytic method for summarizing diagnostic test performances: Receiveroperatingcharacteristicsummary point estimates. Medical Decision Making. 1993, 13: 253256. 10.1177/0272989X9301300313.View ArticlePubMed
 van Houwelingen H, Senn S: Investigating underlying risk as a source of heterogeneity in metaanalysis (by S. G. Thompson, T. C. Smith and S. J. Sharp, Statistics in Medicine, 16, 27412758 (1997)). Statistics in Medicine. 1999, 18 (1): 110115. 10.1002/(SICI)10970258(19990115)18:1<110::AIDSIM14>3.0.CO;2C.View ArticlePubMed
 The prepublication history for this paper can be accessed here:http://www.biomedcentral.com/14712288/9/73/prepub
Prepublication history
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.