Neurology -- Correspondence for Shy et al., 60 (6) 898-904

Correspondence published:

Reply to Letters to the Editor: Michael E Shy, Elliott M Frohman (20 June 2003)

Quantitative sensory testing: Report of the Therapeutics and Technology Assessment: Peter James Dyck, Peter C O'Brien (20 June 2003)

Quantitative sensory testing: Report of the Therapeutics and Technology Assessment: Neil I Spielholz (20 June 2003)

Reply to Letters to the Editor 20 June 2003

Michael E Shy
Wayne State University Detroit MI,
Elliott M Frohman
Send Correspondence to journal:
Re: Reply to Letters to the Editor

m.shy{at}wayne.edu Michael E Shy, et al.

The letters from Dyck and O'Brien and from Spielholz express concerns with the report on quantitative sensory testing (QST) by the Therapeutics and Technology Subcommittee of the American Academy of Neurology. While we will address these concerns, we think it important to note that neither of these letters challenge the basic conclusions of the report which are (1) that QST should not be the sole criteria used to diagnose structural pathology, (2) that detecting malingering remains a challenge for QST and (3) that the role of QST in longitudinal studies, both in the clinical office and for research studies is not yet known.
Dyck and O'Brien raise a number of points, but most deal with their concern that the subcommittee did not address "the correct question." Our committee was charged with investigating the utility of QST in its present state for neurological practice and research. To fulfill this charge, it was necessary to utilize criteria such as sensitivity, specificity, meaningfulness and usefulness.
We agree with Dyck and O'Brien that a test does not need to be diagnostic to be clinically useful. Moreover, we agree that QST may be clinically useful when it is a component of a complete evaluation including medical history, physical examination, test results, and clinical judgment. In fact, we are in agreement with the results from the Rochester study cited by Dyck and O'Brien, which concludes that QST testing may have value when this assessment strategy is part of a group of tests, including EMG, to evaluate sensory neuropathies. We also have no disagreement with their final conclusion that further research of laboratory systems should be encouraged. The need for further research in QST is a point we emphasized in our conclusions.
With respect to the concern about algorithms, we agree that the issues with algorithms can become quite complex. However, the vast majority of algorithms are based on the concepts of "limits" and "levels" that we discussed in our review. The purpose of our review was not to educate neurologists on the intricacies of QST testing but to give a brief overview of the techniques to those who were not familiar with them.
Dr. Spielholz is concerned that our assessment fell short particularly in its treatment of Current Perception Threshold (CPT) testing, addressing this technology briefly in a single paragraph and focusing briefly on a single manuscript published in 1994. We did not focus on any particular QST technique such as CPT in our review. This was not our charge; again it was to look at the field in its entirety. We reviewed hundreds of manuscripts; those that were chosen were used to demonstrate the types of studies QST had been used in, not to state that a particular technique was good or bad. Given space limitations in our review we were not able to spend as much time on any given manuscript, as we would have liked. While there are manuscripts that have been published more recently than 1994, there are also manuscripts utilizing QST devices that go back more than a quarter of a century. It is based on our review of all of these documents that we reached our conclusions.
Ultimately, QST testing may be useful in the assessment of neuropathy when combined with other analyses of sensory nerve function including the neurological examination, and EMG. QST should not be used as a stand alone technique, particularly in legal matters. Furthermore, malingering continues to constitute a formidable problem in the application of QST techniques in some of the patients that we evaluate in the neurology clinic. Additional research initiatives involving the use of QST in longitudinal studies, and in objectively comparing different systems, should be strongly encouraged.

Quantitative sensory testing: Report of the Therapeutics and Technology Assessment 20 June 2003

Peter James Dyck
Mayo Clinic and Mayo Foundation Rochester MN,
Peter C O'Brien
Send Correspondence to journal:
Re: Quantitative sensory testing: Report of the Therapeutics and Technology Assessment

dyck.peter{at}mayo.edu Peter James Dyck, et al.

The recently published practice parameter on quantitative sensory testing (QST) contains useful information but further tweaking with respect to the primary question asked, the descriptive information provided, and its claim of being evidence based might improve it! Before discussing perceived shortcomings, two disclaimers: 1) we were involved in the development of CASE IV and receive royalty from its sale (see below) and 2) we assume no lack of competence or excessive partiality toward any position of the co-signers or of the oversight committees of the AAN.
The Committee's question was, What is the utility of quantitative sensation testing (QST) in diagnosing specific neurologic diseases? Surely this is not the correct question. The question should be, How sensitive, specific, meaningful and useful is QST for the characterization and quantitation of sensation loss or abnormality (by modality) in health and in disease?
The need to address a somewhat different question to that asked by the Committee is suggested by the observation that essentially no test results by themselves should be considered diagnostic of a medical condition. All abnormal test results must by interpreted taking into account the history, findings, previous treatments and other test results. The neurologist who sees a patient with polyneuropathy and finds a raised fasting plasma glucose result cannot simply diagnose diabetic polyneuropathy. Although the raised fasting plasma glucose may be due to diabetes mellitus, it may also be due to not having fasted, a metabolic disorder other than diabetes mellitus, e.g., liver disease, a medication effect, or a technical mistake in the laboratory. Assuming that the raised plasma glucose is due to diabetes mellitus, the physician still cannot diagnose diabetic neuropathy from the plasma glucose alone, it is necessary to establish (by history, examination, other tests and medical judgement) that the neuropathy is not due to another cause but that it is due to diabetes mellitus. We have reported that approximately five percent of neurologic findings among diabetics are due to another cause. [1] A second example is the patient with typical electrophysiologic findings of median neuropathy at the wrist (carpal tunnel syndrome) still persisting years after having undergone successful section of the carpal ligament with relief of symptoms. The physician makes the judgement that the test results can be set aside as not being diagnostic of CTS since the patient does not have symptoms and findings of CTS and has been successfully treated. Another example is the case of a patient, a plumber whose elevated plasma level of lead (from use of white lead) was not the cause of his neuropathy-it was later established that undiscovered hypothyroidism was its cause and the neuropathy improved rapidly with thyroid replacement. [2] Physicians, not tests, make diagnoses based on medical history, physical examination, test results, and clinical judgment.
There is the further point not adequately expressed in the practice parameter that a test does not need to be diagnostic to be clinically useful. As we have shown previously, many evaluations may be needed to characterize the symptoms and impairments and underlying pathophysiology of a disease such as a specific case of peripheral neuropathy. It may be possible to correctly surmise the correct diagnosis from a very cursory examination with few tests but a more intensive examination and more tests are needed to be certain that the diagnosis is correct. [3]
Many tests, for example, nerve conduction and EMG, are useful for this purpose. The use of visual tests by ophthalmologists and neurologists and of hearing tests by otolaryngologists in diagnosing disease are instructive for how QST may be used in neurologic practice. QST, using psychophysical approaches to assess various thresholds of modalities of sensation and stimulus response characteristics, is the counterpart of assessing vision and hearing by psychophysical approaches. It would be inappropriate to judge the value of tests of vision only by their ability to diagnose specific varieties of eye disease (corneal dystrophy, cataract, vitreous hemorrhage or retinopathy) all of which conditions can cause decreased vision. Likewise, although audiometric tests may be useful in localizing abnormalities to end organ, auditory nerve or brain, it would be wrong to judge its value solely by its ability to diagnose specific diseases.
Therefore, the appropriate questions to ask of QST is its utility to detect, quantitate, or characterize sensation loss or abnormality given that such abnormality is present. The question then is, Does QST identify selective modalities of sensation loss, pan-modality sensation loss or altered stimulus-response characteristics (e.g., hyperalgesia) which information can then be used by neurologists to identify populations of receptors or neurons which are dysfunctional and used for characterization, diagnosis, following course, and recognition of a therapeutic effect?
The evidence that QST is useful for these purposes has probably already been demonstrated. In a series of papers with E. H. Lambert, we showed that a selective loss of a modality of sensation was correlated with a selective decrease in the amplitude of a peak of the compound action potential of a biopsied sural nerve in vitro and also with loss or abnormality of a class of sensory fibers in diameter histograms of a biopsied nerve. [4] Thus in spinocerebellar degeneration, selective loss of touch-pressure and vibration sensation was associated with degeneration of alpha-beta sensory fibers as shown by a decreased amplitude of this alpha-beta peak in the compound action potential in vitro and by a reduced number of large diameter fibers in diameter histograms of transverse sections of nerve. In other studies (certain patients with amyloidosis, varieties of hereditary sensory and autonomic neuropathies, Fabry and Tangier disease) raised thresholds of heat-pain sensation and normal thresholds of mechanoreception had their basis in a selective involvement of unmyelinated and small myelinated sensory fibers as revealed by the characteristic abnormalities of the compound action potential in vitro and by morphometric studies. In still other cases, involvement was broadly distributed among different classes of fibers. The correctness of the forgoing correlations were under girded by the pioneering studies of Gasser and Erlanger [5] and by electrophysiologic studies of sensory units. [6]
The correct question, therefore, should relate to the utility of QST to characterize sensation with development and maturation and with aging in health (among modalities, different sites, and with age and anthropomorphic features) and with alterations due to disease. Since it is sensation, a psychophysical experience, which is to be assessed, it must be judged by how well sensation itself is being assessed, not by how well it correlates with surrogate measures or by detection of disease, albeit also of interest. Other measures, for example, sensory nerve conduction, somatosensory evoked potentials, and even pathologic and morphometric studies of nerve may also be assessed; but they are surrogate measures of sensation, not sensation itself.
We recognize that despite the fact that quantitative sensation tests are used as only a part of the diagnostic process, the statistical entities of sensitivity and specificity are two (among many) useful attributes of a test. Thus, we accept that studies that provide estimates of these types of attributes are important. Is the report correct that there is no class 1 evidence bearing on this question? We suggest that there is such evidence. In the Rochester Diabetic Neuropathy Study (RDNS), we proposed studies that were reviewed by intramural and extramural (NIH) review committees, were approved, and were prospectively performed. A major objective was to determine the sensitivity, specificity, representativeness, reproducibility and meaningfulness of quantitative sensation test results, attributes of nerve conduction, autonomic tests, summated composite scores of selected items of the preceding items and various clinical measures of symptoms and impairment so that the best measures might be used to detect, characterize and follow course of neuropathy and neuropathic outcomes. A large subject cohort of more than 500 persons without diabetes mellitus or neurologic disease agreed to participate and normal percentile values (and normal deviates) were estimated so that results in patients could be expressed as percentiles and normal deviates specific for modality, site, age, sex and applicable anthropomorphic variables. All known diabetic patients in Rochester, MN were contacted and invited to participate in cross-sectional and longitudinal studies employing standard tests. The consenting diabetic persons less than 70 years old were not significantly different from the group who did not agree to participate by the criteria of co-morbidity. All neuropathic evaluations and tests were predetermined and performed using standard baseline and testing conditions. QST was done using predetermined algorithms and independent of other neuropathic evaluations. The results have been extensively published. [7, 8, 9] To illustrate, the intraclass correlation coefficients testing reproducibility of vibration and cooling detection threshold using CASE IV and the 4, 2 and 1 stepping algorithm with null stimuli were better than 0.9 and comparable to what we found assessing the best attributes of nerve conduction. [10] VDT was sufficiently reproducible to show a monotone and significant worsening over long periods of time. [11]
Our published data suggests that some attributes (or preferably summated normal deviate scores of nerve conduction) are more sensitive in diagnosing diabetic polyneuropathy than are QST results. But as we have also suggested, although nerve conductions appear to be more sensitive, QST results tend to be more meaningful - they actually reflect sensation loss or hyperalgesia! There is an additional difference in what should be inferred from QST versus nerve conduction measures. The integrity of the entire sensory apparatus (the receptor, nerve and central nervous system tracts, and cerebral recognition and interpretation) is tested with QST. By contrast nerve conduction provides information only about the cable transmission properties of a class of sensory fibers (essentially large myelinated fibers) from the point of stimulation to a more defined proximal point. Also, nerve conduction measurements may include fibers, which although contributing to the action potential, are already disconnected from their receptors. Importantly also QST allows detection of abnormality of functional classes of sensory fibers, which are not tested by, nerve conduction tests. Small fibers are usually not tested by nerve conduction tests.
Laboratory QST, as compared to clinical bedside testing of sensation, allows standardization of all aspects of testing: the environment, the initial load, use of precisely shaped and defined invariant waveforms and given over a broad range of known stimulus magnitudes, null stimuli and use of sophisticated and standard computer algorithms for testing and finding threshold. Results can then be compared to values obtained from healthy subjects tested by exactly the same approaches so that responses can be expressed as percentiles specific for modality, anatomical site, age, sex and applicable anthropomorphic variables (e.g., height, weight and body mass index). Because clinical evaluation of symptoms and impairments, nerve conduction tests, QST and quantitative autonomic tests provide independent and useful information, which allows for comparison and validation, we advocate use of several of these measurements in composite scores. [12] This makes sense since polyneuropathy is the sum of symptoms, impairments and functional alterations of various classes of fibers.
The comments about algorithms of testing and finding threshold also need tweaking. It is reported that these algorithms can be divided into the method of limits and the method of steps (referred to as forced choice). This summary is too simple and, at least in part, a little askew. In characterizing algorithms of testing and finding threshold, one needs to consider many aspects of the system, the testing procedures and how threshold is to be estimated and validated. True, algorithms may employ stimuli which increase (or decrease) continuously (linear, exponential or other) or in steps (by linear, exponential or other), but this is barely a beginning in describing algorithms. It is quite inaccurate to characterize algorithms by these limited criteria. Algorithms vary depending on the preconditions of testing, the environment, and the instructions given, whether static loads or preconditions are used and what they are, the stimulus waveform magnitude and presentation, whether ramps or steps will be used, whether null stimuli will be given, the specific rules of testing, the specific rules of determining threshold, the number of repetitions and turnarounds, and how results will be presented and compared to normal values. In laboratory testing of QST, we suggest that all of these steps need to be documented, standard, the same at all sites and times, and the same for controls and patients. The specific rules of the algorithm of testing and finding threshold may spell the difference between a valid or an invalid algorithm of testing. To illustrate, we found that too low a number of turnarounds (from sensitivity to insensitivity and vice versa) in the 4, 2 and 1 algorithm with null stimuli might in itself produce spurious results. Difference in results should be due to patients' response differences and not to differences of testing procedures. To clarify, using step testing in an algorithm is not equivalent to forced choice testing! In two alternative forced-choice testing stimulus events are always given in pairs with one being the stimulus and the other the null stimulus. For each pair of stimulus events, indicated, for example, by the display of the numbers 1 and then of 2, a stimulus event is given by chance in the 1 or 2 interval. The null stimulus is given in the other interval. The subject or patient then has to choose whether the stimulus was felt in period 1 or 2. The patient is forced to choose (forced choice) the interval 1 or 2, which is the most likely to have contained the stimulus. Using this method, response criteria should be the same among subjects. Even with forced choice testing the specific rules of testing may be different among algorithms. Therefore it is necessary to use only highly characterized and validated algorithms and not to assume that a standard and validated procedure is being followed simply because the authors state that they used the 'method of limits' or the 'method of forced choice' - without exact specification these terms are almost meaningless.
We also want to comment on the statement about reproducibility among test results. One of the possible reasons for differences in results among different QST systems is use of different expressions of stimulus magnitude, i.e.; absolute measured values versus use of just noticeable difference units (JND). Whereas it is attractive to use absolute units of displacement, for example, �m of displacement superimposed on a standard load or as deltaoC (as we do in CASE 4), care must be taken how these values are interpreted and analyzed.
It has been known for a long time that sensation does not increase linearly but increases as an exponential function. [13] This has been extensively studied for vision and hearing. Therefore, it is better to express results as just noticeable difference. For a variety of reasons, it is even better to express results as percentiles (relative to a healthy population) and do the statistical analyses using normal deviate values.
Secondly, it is well known among statisticians that reproducibility should be concerned with agreement, not correlation. The use of the usual (product-moment) correlation should not be used for this purpose. The intraclass correlation should be used. If one first transforms to ranks, this will facilitate comparisons of correlation with other QST results using other systems and with other nerve test results.
We note that several multicenter trials have already been done or are being done employing centralized calibration of stimuli, standard algorithms of testing and finding threshold, central quality control and QST approaches are proving to be useful in the conduct of controlled trials. Finally we want to reiterate that we think that QST, especially laboratory based QST, is useful for the characterization and quantification of alterations of sensation in health and in disease and increasingly will be used for epidemiologic surveys and controlled clinical trials. Use of QST appears also to be especially useful in detecting thermal hyperalgesia - evidence of sensory receptor and fiber dysfunction. It is now possible to use stimuli, which are defined and accurately calibrated over a broad range of stimulus magnitudes, pre-programmed and validated algorithms of testing using null stimuli, and expression of results as percentiles and normal deviates allowing comparison of results from different systems. Further development of laboratory based systems should be encouraged.
References: 1. 1. Dyck PJ, Kratz KM, Karnes JL, et al. The prevalence by staged severity of various types of diabetic neuropathy, retinopathy, and nephropathy in a population-based cohort: The Rochester Diabetic Neuropathy Study. Neurology 1993;43:817-824.
2. Dyck PJ, Lambert EH. Polyneuropathy associated with hypothyroidism. J.Neuropathol.Exp.Neurol. 1970;29:631-658.
3. Suarez GA, Chalk CH, Russell JW, et al. Diagnostic accuracy and certainty from sequential evaluations in peripheral neuropathy. Neurology 2001;56:1118-1120.
4. Lambert EH, Dyck PJ. Compound action potentials of sural nerve in vitro in peripheral neuropathy. In: Dyck PJ, Thomas PK, Griffin JW, Low PA, Poduslo JF, eds. Peripheral Neuropathy, 3rd ed. Vol. Philadelphia: W. B. Saunders, 1993:672-684.
5. Erlanger J, Gasser HS. Electrical Signs of Nervous Activity. Philadelphia: University of Pennsylvania Press, 1933.
6. Light AR, Perl ER. Peripheral sensory systems. In: Dyck PJ, Thomas PK, Griffin JW,Low PA, Poduslo JF, eds. Peripheral Neuropathy, 3rd ed. Vol. Philadelphia: W. B. Saunders Company, 1993:149-165.
7. Dyck PJ, Bushek W, Spring EM, et al. Vibratory and cooling detection thresholds compared with other tests in diagnosing and staging diabetic neuropathy. Diabetes Care 1987;10:432-440.
8. Dyck PJ. Detection, characterization, and staging of polyneuropathy: assessed in diabetics. Muscle Nerve 1988;11:21-32.
9. Dyck PJ, Dyck PJB, Velosa JA, Larson TS, O'Brien PC. Patterns of quantitative sensation testing of hypoesthesia and hyperalgesia are predictive of diabetic polyneuropathy. A study of three cohorts. Diabetes Care 2000;23:510-517.
10. Dyck PJ, Kratz KM, Lehman KA, et al. The Rochester Diabetic Neuropathy Study: design, criteria for types of neuropathy, selection bias, and reproducibility of neuropathic tests. Neurology 1991;41:799-807.
11. Dyck PJ, Davies JL, Litchy WJ, O'Brien PC. Longitudinal assessment of diabetic polyneuropathy using a composite score in the Rochester Diabetic Neuropathy Study cohort. Neurology 1997;49:229-239.
12. Dyck PJ. Assessment: thermography in neurologic practice. Neurology 1990;40:523-525.
13. Stevens SS. Neural events and the psychophysical law. Science 1970;170(962):1043-1050.

Quantitative sensory testing: Report of the Therapeutics and Technology Assessment 20 June 2003

Neil I Spielholz
University of Miami School of Medicine FL
Send Correspondence to journal:
Re: Quantitative sensory testing: Report of the Therapeutics and Technology Assessment

nspielholz{at}adelphia.net Neil I Spielholz

Technology Assessments should be "fair and balanced" reports based on in-depth, accurate reviews of articles deemed worthy of representing the best in the field. However, the recent Assessment on quantitative sensory test-ing, falls short. [1] Here is an example of how inadequately the Assessment reviewed and reported just one paper.
The single paragraph addressing Current Perception Threshold (CPT) testing reports that one study "detected differences between neuropathic and nonneuropathic groups ... CPT to 2 kHz stimulation correlated best with vibratory thresholds, and CPT to 5Hz stimulation correlated with thermal." [2] But then we are told that a study by Tack et al, [3] found only moderate correlation between clinical evaluation of neuropathy and vibratory thresholds and CPT data. [1]
First, the Assessment designates the paper by Tack et al. as a Class II study (Class I, the highest). [3] However, this contradicts the Assessments classification criteria requiring "a blinded evaluation". [1] Nothing in Tack et al reports blinded evaluations. [3]
Second, the Assessment selectively describes the findings of Tack et al in the following two sentences: "Correlation between VPT and CPT were maximal at 2,000Hz for CPT (r = 0.61). Correlation between CPT at 250Hz or 5Hz with clinical evaluation of neuropathy were less than for 2,000Hz." [1] The Assessment neglects the correlation coefficients between CPT and neurological disability score (NDS) in the subgroup of 22 patients with overt neurologic findings. The coefficient for the 2,000Hz CPT with "large fiber pathology" was 0.88, while between the 5Hz CPT and "small fiber pathology" it was 0.74. [3] So yes, the correlation of 5Hz with clinical examination was less than the correlation of 2,000 Hz with the clinical examination, but they were both higher than the correlation between CPT and VPT. And these correlations were highly substantial (p < 0.002, for the former; p < 0.01, for latter). [3] Why did the Assessment only specify the least important correlation of all?
Third, Tack et al concluded, despite the above correlations of CPT with clinical scores, and also despite mentioning "highly significant differences" in CPT values between patient groups and healthy controls, that "CPT seemed rather insensitive in detecting neuropathy." [3] But Tack et al offered no data to support this apparently contradictory statement. No specifics were given from which either sensitivity or specificity could be calculated. A proper Assessment would have recognized and commented upon such a striking omission. Instead, based on this mediocre 8-year old article, [3] the Assessment implies that CPT has not been shown to be a particularly useful tool. This is misleading, especially in light of many peer-reviewed, blinded clinical studies, which have appeared since 1994.
Therefore, concerning the CPT literature, this Assessment is incomplete, misleading, and 8-years behind.
References:
1) Shy ME, Frohman EM, So Y.T., et al. Quantitative sensory testing: Report of the Therapeutics and Technology Assessment Subcommittee of the American Academy of Neurology. Neurology 2003;60:898-904.
2) Masson, E.A., Veves, A., Fernando, D., Boulton, A.J.M. Current perception thresholds: a new, quick, and reproducible method for the assessment of peripheral neuropathy in diabetes mellitus. Diabetologia, 1989; 32:724-728.
3) Tack CJJ, Netten PM, Scheepers MH, et al. Comparison of clinical examination, current and vibratory perception threshold in diabetic polyneuropathy. Netherlands J Med 1994;44:41-49.