OUP user menu

Diagnosing Lumbar Zygapophysial Joint Pain

Nikolai Bogduk MD, DSc, PhD
DOI: http://dx.doi.org/10.1111/j.1526-4637.2005.05023.x 139-142 First published online: 1 March 2005

In their review of the literature on lumbar radiofrequency neurotomy and zygapophysial joint blocks, Hooten, Martin, and Huntoon [1] make several salient points. They indicate that systematic reviews of lumbar radiofrequency neurotomy have focused on conventional aspects of methodology, such as randomization, sample size, and outcomes, but that these reviews did not address two seminal, clinical matters: diagnosis and operative technique.

On operative technique, Hooten, Martin, and Huntoon [1] correctly point out that for lumbar radiofrequency neurotomy to be credible and effective, electrodes need to be placed parallel to the target nerve. This very point is elaborated in another recent study [2]. None of the controlled studies covered by the systematic reviews used such a technique. Therefore, none of the studies and none of the reviews constitutes an evaluation of the procedure is it should be correctly performed. As Hooten, Martin, and Huntoon [1] state, only Dreyfuss et al. [3] used the correct technique, and their study provides the benchmark for expectable outcomes.

On the matter of diagnosis, Hooten, Martin, and Huntoon [1] reveal that none of the controlled trials properly established a diagnosis of lumbar zygapophysial joint pain before venturing to test a treatment for that condition. The criterion standard for the diagnosis are controlled, diagnostic blocks, but no study used controlled blocks. Therefore, the samples recruited are very likely to have included false-positive cases, which would have confounded the outcomes of the studies, by reducing the apparent success rates. For this reason, none of the controlled studies and none of the systematic reviews constitutes proper evidence of the efficacy of lumbar radiofrequency neurotomy. Again, only Dreyfuss et al. [3] used controlled blocks to select their patients.

Hooten, Martin, and Huntoon [1] proceed to elaborate an algorithm, which they submit could be used to select patients more efficiently, for controlled trials and, by implication, perhaps for treatment in conventional practice. They argue that patients should first be screened using the clinical tests of Revel [4].

What Hooten, Martin, and Huntoon [1] propose is sensible in principle. The argument that they use, however, may be somewhat esoteric to many readers. The argument is based on Bayesian statistics. In essence, this method is based on the relationship that diagnostic confidence is the product of the likelihood ratio of the diagnostic test and the pretest prevalence of the condition in question. Accordingly, a series of tests can be used to increase progressively the pretest prevalence of the condition, and thereby improve diagnostic confidence.

In particular, the argument is that, if the likelihood of zygapophysial joint pain can be increased by clinical examination, the yield of subsequent diagnostic blocks will be greater. Thereby, the diagnostic process can be rendered more efficient, that is, by reducing the number of diagnostic blocks required.

Whereas this argument is correct in principle, certain steps outlined by Hooten, Martin, and Huntoon [1] are either incorrect or ambiguous. By correcting the information and revisiting their argument, the actual efficiency of the algorithm proposed by Hooten, Martin, and Huntoon [1] can be determined in terms that are perhaps of more transparent relevance to practitioners.

For the sake of argument, let us assume that the prevalence of lumbar zygapophysial joint pain in the population is 27%, which is the representative figure selected by Hooten, Martin, and Huntoon [1]. Let us then consider what would happen if this diagnosis were pursued and established using double-blind, controlled diagnostic blocks in 1,000 patients with low back pain.

The data of Dreyfuss et al. [3] show that controlled blocks have a sensitivity of 88% and a specificity of 71%. The false-negative rate, as estimated by Hooten, Martin, and Huntoon [1], is 12%. If the observed prevalence of zygapophysial joint pain is 27%, 1,000 patients would be distributed as shown in Table 1. The actual prevalence of zygapophysial joint pain is 30.6% (306/1,000), but 36 patients fail to be detected because of the false-negative rate, leaving only 270 (27%) to constitute the observed prevalence.

View this table:
Table 1

The outcomes of controlled diagnostic blocks when applied to 1,000 patients in whom the observed prevalence of zygapophysial joint pain is 27%

Second Block Blocks
First blockPositive270202  472
Negative  36492  528

The data in Table 1 also show the properties of single diagnostic blocks. Single blocks have a false-positive rate of 29% (202/694). Consequently, only 270 of 472 positive blocks (57%) are true-positive.

If controlled blocks were used to determine the diagnosis, the cost would be 1,472 blocks. One thousand blocks would be needed to screen the patients for an initial response, and a further 472 control blocks would be needed to check the responses in those who were positive after the first block. The yield of this approach would be 270 true-positive cases, but no false-positive cases, and 36 false-negative cases. The latter would be patients who, indeed, did have zygapophysial joint pain but were missed by the first block and therefore did not proceed to a second block.

If controlled blocks were not used, and the diagnosis was based on relief after only single blocks, the process would appear more efficient. Only 1,000 blocks would be required. The yield, however, would be 270 true-positive responses, but also 202 false-positive responses, with no way of determining which was which. There would also be 36 false-negative cases, as for controlled blocks.

Hooten, Martin, and Huntoon [1] argue that the diagnostic process can be made more efficient if the sample were first subjected to Revel's tests. These tests are said to have a sensitivity of 92% and a specificity of 80%, with a resultant likelihood ratio of 4.6. Although these figures appear impressive, they were not based on controlled blocks; they were based on single blocks. Therefore, the sensitivity and specificity apply to samples of patients who are positive and negative to single blocks. In terms of the example being developed, in a sample of 1,000 patients, 472 would be positive to single blocks, and 528 would be negative (Table 2). With a sensitivity of 92%, Revel's tests would be positive in 434 of those 472 positive patients. With a specificity of 80%, the tests would be false-positive in 106 of the 528 negative patients (Table 2).

View this table:
Table 2

The outcomes of Revel's tests when applied to 1,000 patients in whom the prevalence of zygapophysial joint pain is 27%, but who undergo only single diagnostic blocks

Single Block
Revel's testsPositive434 (248)106  540
Negative 38 (22)422  460
  • The figures in parentheses indicate the number of patients who truly have zygapophysial joint pain according to controlled diagnostic blocks.

The group of 472 patients, who are positive to single blocks, contains the 270 patients who would be true-positive if subjected to controlled blocks. Proportionally, these 270 patients would be distributed as 248 among the 434 patients positive to Revel's tests, and 22 among the 38 patients negative to Revel's tests (Table 2).

If Revel's tests are used as a screening test, to preselect patients for diagnostic blocks, only those who are positive to the tests proceed to blocks. They would number 540, and consist of 434 true-positive and 106 false-positive responders to Revel's tests. The remaining 460 patients do not proceed to blocks, which include the 422 true-negative cases, but also the 38 false-negative cases. The latter include 22 patients with true zygapophysial joint pain who slip through the screening test, because of the false-negative rate of Revel's tests. Of those who proceed to blocks, only 248 truly have zygapophysial joint pain that might be confirmed by controlled blocks.

At this stage in the algorithm, the sample size is 540, and the number of true-positive cases hidden within it is 248. Revel's tests have increased the prevalence of zygapophysial joint pain from 27% to 46% (248/540). When this sample is subjected to blocks the distribution will be as shown in Table 3.

View this table:
Table 3

The outcomes of diagnostic blocks when applied to 540 patients who are positive to Revel's tests

Second Block Blocks
First blockPositive218 85 (31 + 54)303
Negative 30207 (75 + 132)237
  • The figures in parentheses indicate the pedigree of the patients in the respective cells. The 31 and 75 patients were the 106 who were false-positive to Revel's tests. The 54 and 132 were the 186 who were positive to Revel's tests but whom controlled blocks show not to have zygapophysial joint pain.

The figures are based on the properties of diagnostic blocks (Table 1). Although there should be 248 patients with zygapophysial joint pain, only 218 are correctly detected because of the false-negative rate of controlled blocks. Meanwhile, because of the false-positive rate of single diagnostic blocks, 31 of the 106 patients who were negative to Revel's tests appear positive to single blocks, leaving 75 as true-negative; and 54 of the 186 patients who were false-positive to Revel's tests emerge as positive to single blocks, leaving 132 as true-negative. The 31 plus 54 patients have false-positive responses to single blocks, but will prove negative to controlled blocks. From these figures the efficiency of applying Revel's tests as a screening test can be derived.

If single blocks only are used to check the diagnosis, the cost is 540 blocks, which is considerably less than the 1,472 or 1,000 blocks required if Revel's tests are not used, depending on whether controlled blocks or single blocks, respectively, are used. The yield, however, is poor. Of the 540 blocks, 303 are positive but of these, 218 are true-positive and 85 are false-positive, with no way of distinguishing the two. Meanwhile, 30 patients are false-negative, in addition to the 22 genuine cases that were excluded earlier when Revel's tests were applied.

In contrast, if controlled blocks are applied, the cost is 843 blocks, being 540 first blocks and 303 second blocks to check the response in those positive to the first block. The yield is 218 true-positive cases but no false-positive cases. The loss is 30 false-negative cases, in addition to the 22 cases excluded previously.

These various figures can be related to the Bayesian calculations offered by Hooten, Martin, and Huntoon [1]. At the outset, the prevalence of zygapophysial joint pain is 27%. The pretest odds are 0.37 (27:73). Ostensibly, the likelihood ratio of Revel's tests is 4.6 (92/[100 − 80]). In their figure 2, Hooten, Martin, and Huntoon [1] argue that this should increase the prevalence to 63%, that is,

Pretest odds × likelihood ratio = posttest odds Embedded Image

This, however, is not correct. It would be correct only if Revel's tests had been based on controlled blocks. As they are based only on single blocks, the tests increase the prevalence of zygapophysial joint pain to only 46%, that is, 248/540 (Table 2). Hereafter, however, the Bayesian methods are correct.

With a prevalence of 46%, the pretest odds are 0.85 (46:54). The likelihood ratio of single blocks is 3.1. From these figures the post-test prevalence can be calculated.

Embedded Image

This figure constitutes the diagnostic confidence of having found true-positive zygapophysial joint pain after performing both Revel's tests and a single diagnostic block. It corresponds to the figures in Table 3. Of the 303 patients with positive blocks, only 218 (72%) are true-positive.

This figure is less than the 84% implied by Hooten, Martin, and Huntoon [1]. The difference arises because Hooten, Martin, and Huntoon [1] accorded a greater validity to Revel's test, as if they were validated against controlled blocks rather than single blocks. Furthermore, 72% does not constitute strong diagnostic confidence. Although applying Revel's tests does improve efficiency, it does not substitute for eventually performing controlled blocks. Only with controlled blocks is the figure of 72% converted to 100%.

Readers who pursue a diagnosis of lumbar zygapophysial joint pain can use these data to decide which algorithm they wish to follow. Table 4 summarizes the efficiency and yield of various approaches.

View this table:
Table 4

The efficiency and yield of various strategies for the diagnosis of lumbar zygapophysial joint pain. Diagnostic confidence is the proportion of positive responses that are true-positive

StrategyBlocks RequiredYield per 1,000 Patients
True-PositiveFalse-PositiveDiagnostic Confidence
Single blocks only1,000270200 57%
Controlled blocks only1,472270    0100%
Revel's tests plus single blocks  540222113  66%
Revel's tests plus controlled blocks  873222    0100%

Controlled blocks require 1,472 procedures per 1,000 patients but provide 100% diagnostic confidence in the 270 positive diagnoses established. Single blocks require fewer procedures, but the diagnostic confidence is only 57%. Investigators can be only 57% certain that their diagnosis is correct. Coupling Revel's tests with single blocks only detects 303 positive cases but only 218 are true-positive, which amounts to a diagnostic confidence of only 72%. Coupling Revel's tests with controlled blocks preserves 100% diagnostic confidence, but detects only 218 cases. The missing 52 patients are accounted for by the false-negative rates of Revel's tests and of diagnostic blocks.

If a diagnostic confidence of only 57% or 72% is not acceptable, the choice lies between a strategy of controlled blocks only, to detect 270 true-positive cases, and a strategy of Revel's tests plus controlled blocks. The latter uses 629 (43%) fewer blocks, but detects 48 (20%) fewer cases. What might be gained in efficiency is paid for by a decrease in sensitivity; 20% of eligible patients fail to be detected.

These various figures are designed to illustrate, in palpable terms, what the argument of Hooten, Martin, and Huntoon [1] means in practice. The actual figures will differ according to the prevalence of zygapophysial joint pain in the sample studied, and according to the criteria used to define a positive response to a block. Nevertheless, the relative merits of the various diagnostic strategies remain the same.


View Abstract