Monotonicity of Nerve Tests in Diabetes: Subclinical nerve dysfunction precedes diagnosis of polyneuropathy

For these studies we used cross-sectional and longitudinal data obtained from the Rochester Diabetic Neuropathy Study (RDNS) cohort, an epidemiologic study of 504 diabetic patients from Olmsted County (prevalence date 1 July 1986), who were Caucasian and mainly of northern European extraction (13,14,28). For the present study, a data audit was completed to the end of 2003.

The study population for the present study are the 238 (of 327 evaluated at least twice) patients with diabetes who at first evaluation did not fulfill minimal nerve conduction test criteria for DSPN (i.e., their composite nerve conduction scores [Σ 5 NC nds], the derivation of which is given in the appendix, were <97.5th percentile) (group 1) and a subgroup of 90 of these 238 patients that was evaluated at least six times (group 2). For the latter group, six or more visits was chosen as the smallest number of evaluations needed to test (with a reasonable degree of certainty) for worsening, improvement, or no change in linear regressions of nerve conduction tests (e.g., of Σ 5 NC nds) on time (years).

Individual and sum scores of attributes of nerve conduction tests and modalities of quantitative sensation test (QST) thresholds (determined with CASE IV; WR Medical Electronics, Stillwater, MN) were expressed as measured units, percentiles, and normal deviates after correction for applicable biographic and demographic variables based on studies of a healthy subject cohort (RDNS of healthy subjects [RDNS-HS]) using precisely the same approaches and standards of testing and quality control as used for patients with diabetes (29,30).

The Σ 5 NC nd was chosen as the minimal criteria for DSPN because nerve conduction abnormality is objective (results cannot be willed by the patient), provides a quantitative measure over a wide range, has been shown to be as (or more) sensitive and specific for sensorimotor polyneuropathy than other nerve conduction criteria, is more reproducible than most individual attributes of nerve conduction, and generally correlates well with neuropathic impairment, and its components include several nerves of the legs and representative attributes (conduction velocities, latencies, and amplitudes) (16). The denervation of the ∑ 5 NC nds is provided in the appendix. To set percentile (and normal deviate values) of sum scores such as Σ 5 NC nd, we plotted values of patients from the RDNS-HS cohort on age, fitted a common regression line using least-squares regression, and provided corresponding percentile lines empirically by increasing or decreasing the intercept. Considering its derivation, it follows that if every one of the five nerve conduction values making up the composite score were at the 50th percentile, the composite score would add up to 0; if all were at the 1st percentile (−2.33 normal deviate), the composite score would be −11.65; and if all were at the 99th percentile (2.33 normal deviate), the composite score would be 11.65.

All clinical and nerve tests were prescheduled, standard, and independent (of information from previous or other examination test results) as part of a prospective cross-sectional and longitudinal study of diabetes and its complications and risk covariates (NS36797). Also, examination times were not linked to intercurrent health problems. The neurologic examinations were done by one of us (P.J.D.) without reviewing previous examination information or test results and without knowledge of symptoms (the examination done before inquiry regarding symptoms). All items of the standard Neuropathy Impairment Score (NIS) were graded by defined criteria. Then a 37-item symptom score (Neuropathy Symptoms and Change [NSC]) was completed. All of the clinical information was entered into a paper and electronic form (Clinical Neuropathy Assessment [CNA]). The CNA is a paper and electronic record of the study, biographic and demographic data, nerve test (NIS), neuropathy symptoms (NSC), and disability (Dyck modification of the Rankin score). The electronic record (CNA) provides a high level of quality control and accuracy and transfer of data, which is described in our previous article (31). Nerve conduction tests were performed by technicians following defined protocols of study under the supervision of one of us (C.M.H.), and QSTs were performed by technicians using verbal (from printed) instructions and programmed algorithms of testing and finding threshold. The stimulus-response pattern was quality controlled by visual inspection (by the technologists and P.J.D.), looking for inadequate patterns due to inadequate testing or inattention or drowsiness. When the record was inadequate, the test was repeated. For vibration and cooling thresholds, 4, 2, and 1 stepping with the null stimuli algorithm was used; for heat pain a nonrepeating ascending stepping algorithm with null stimuli was used (rev. in 16). As for clinical evaluations review of earlier test results was not allowed for nerve tests. Nerve conductions were obtained using standard conditions and stimulating and recording electrode placements.

In group 1 patients, we assessed whether the mean Σ 5 NC nd score of patients with diabetes but with values <97.5th percentile was higher at onset and at last examination than that of healthy subjects (also with values <97.5th percentile). If mean values are higher in diabetic subjects, and especially at the last examination, it would suggest that nerve conductions, even within normal limits, are gradually worsening in diabetes.

In the analysis of the 90 group 2 subjects (of 238) without DSPN at baseline, we regressed individual and sum scores of nerve conduction and other test results on time. This analysis was performed separately for each subject, resulting in a slope (b) and a correlation coefficient (r) for each subject. The mean slope (b̄) and correlation (r̄) were calculated. One sample t test was used to test that the slopes and correlation coefficients differed from 0. The square of the mean slope was computed for each variable, indicating that the percent variance explained by the regression, which provided a measure of monotonic worsening ranging from 0.0 to 1.0, which can be compared among end points. Assessed as risk covariates against the slope of Σ 5 NC nds on time of group 2 patients were the risk covariates assessed in our previous article (14). Associations between the slope of worsening and risk factors were calculated quantitatively with Spearman rank correlation for continuous variables and rank sum tests for dichotomous variables. For multivariate analysis, stepwise regression was used, stepping up with P < 0.05 as the criterion.

Of 327 patients in the RDNS who were evaluated two or more times, there were 238 subjects at the first evaluation and 196 at the last evaluation who were without DSPN by the criterion of Σ 5 NC nds <97.5th percentile. Their clinical features are summarized in Table 1. At baseline, the 238 patients had a mean Σ 5 NC nd score of 1.08, which was considerably higher than that in healthy subjects whose Σ 5 NC nds <97.5th percentile was −0.12. At the last examination the 196 patients had a normal deviate mean score of 3.63. These results suggest that in community diabetic subjects a shift of nerve conduction values toward abnormality has occurred even before they can be declared abnormal by the normal limits criterion of ≥97.5th percentile. It cannot be attributed to a change in age or weight during the follow-up period because these factors had already been corrected for.

To provide the differences in measured units of nerve conduction that a typical patient (a 45-year-old man, 1.78 m tall, and weighing 82 kg) would undergo if his Σ 5 NC nd value were to change from −0.12 nds (mean value of healthy subjects) to 1.08 nds (value at the first evaluation of RDNS patients) and 3.63 (value at the last evaluation of RDNS subjects), see Table 2. For these estimates, we make the assumption that each of the five attributes of nerve conduction in Σ 5 NC nds changed by an equal amount.

The same phenomenon (a shift toward abnormality while nerve conduction still remained within normal limits) can be seen by looking at the distribution of Σ 5 NC nds plotted against age among RDNS (diabetic patients) compared with healthy subjects (Fig. 1). In the RDNS-HS, there were 156 values above and 155 values falling below the 50th percentile line. For diabetic patients with Σ 5 NC nds <97.5th percentile at baseline, there were 147 values above the 50th percentile line and 90 below. At last examination, there were 154 above and 42 below the line.

Table 3 shows information on which nerve tests provide the best measures of monotone worsening by the criteria of the number of individuals whose values significantly worsened, improved, or remained unchanged and on the magnitude of the mean change with time and the percentage of the variance explained (r̄²). Composite nerve conduction tests were markedly superior to individual attributes. Especially favorable indications of monotone worsening were Σ 5 NC nd and Σ Per. Tib. Ul. MNDL nds. By contrast, neurologic signs (e.g., NIS [lower limb]) barely showed a statistically significant change (P = 0.032), whereas symptoms (NSC) and QST did not.

The second question addressed was “Did patients show significant worsening over time even before minimal nerve conduction test criteria for DSPN had been met?” Statistically significant worsening of Σ 5 NC nds occurred in 42 of 90 patients (46.7% of patients); 22 of 90 (24.4%) had shown significant worsening but at the last examination had nerve conduction tests <97.5th percentile. It was of particular interest that of patients who at baseline had nerve conduction NC values <1st percentile, 13 (14.4%) had last values <50th percentile after 13.9 ± 2.25 years of follow-up.

Many univariate risk covariates were associated with rate of worsening of Σ 5 NC nds on time; but in multivariate analysis, only 24-h urine microalbuminuria (a measure of microvessel disease) remained as a significant risk covariate (P < 0.001) (Table 4).

To answer the question “Which nerve tests show the largest monotone worsening in a diabetic cohort expected to develop DSPN?”, we compared the ability of nerve tests to demonstrate a statistically significant monotone worsening over time, assuming that such a change is taking place. Is the assumption of worsening correct? The studies by Pirart (32,33), the Diabetes Control and Complications Trial studies (34,35), our cross-sectional and longitudinal studies (36), and our present studies of group 1 and 2 patients unequivocally show that over time an increasing number of individuals with diabetes develop DSPN. It is therefore reasonable to compare different end point measures for their ability to recognize this worsening. Direct comparison among different nerve tests was optimized by expressing the abnormality of each test as a normal deviate value (from percentiles) with values corrected for applicable biographic and anthropomorphic characteristics (as obtained from the study of a healthy subject cohort). We did this for all our tests of individual and summed attributes of nerve conduction tests and for QSTs. For clinical measures (i.e., NIS and NSC), corrections for the influence of age, sex, physical fitness, height, and weight were also attempted, but such corrections are based only on judgment, which presumably is at best only approximate. Although we have previously demonstrated that there can be a high level of agreement in NIS scores among investigators who are trained and agree on criteria for making judgments (28), it is unknown what degree of agreement would be achieved without this training and agreement.

The present studies did not address the closely related, but different, topic of the lowest level of cumulative glycemic exposure inducing polyneuropathy or the mechanisms that might be involved. Patients classified as having impaired glucose tolerance were not included in the present analyses.

By the criteria of monotone worsening in this cohort of subjects with diabetes but without DSPN at onset, nerve conduction tests were unequivocally superior to clinical impairment, symptoms, or QSTs. The best composite scores of nerve conduction test abnormality (Σ 5 NC nds) were a reliable indicator of functional worsening, as 42 subjects showed significant worsening and only 1 showed significant improvement, with the remainder showing no change. Neuropathic impairment (NIS and NIS [lower limb]) also showed a statistically significant mean slope of worsening, but it was not reliable for identification of worsening of individual patients, as five patients significantly worsened and three significantly improved. Symptoms (NSC severity) also worsened, but statistical significance was not quite achieved (P = 0.053). Here also symptoms were not reliable indicators of worsening because significant worsening was demonstrated in three subjects only, and in one there was significant improvement.

Next, which were the best indicators of functional worsening of nerve conduction tests? Composite nerve conduction test scores performed better than individual nerve conduction test attributes. The two that performed best were Σ 5 NC nds and Σ Per. Tib. Ul. MNDL nds, but some individual attributes (e.g., Per. MNDL nds and Tib. MNDL) also performed well. We emphasize that this worsening can be attributed to actual worsening of nerve function due to diabetes, because corrections already had been made for the nondisease variables of change in age and weight.

Do the present results support the conclusion of the recently published evidence-based case definition of polyneuropathy (37)? These authors concluded that “The combination of neuropathic symptoms, signs, and electrodiagnostic findings provide the most accurate diagnosis of distal symmetric polyneuropathy.” In addition to the evidence listed in the article, the conclusion these authors came to seems intuitively correct. The present study, however, extends the evidence on this question by showing that low levels of functional abnormalities of nerve conduction are earlier and more certain indicators of nerve dysfunction than are neurologic symptoms or signs. Alternatively or additionally, nerve conduction tests may provide reliable evidence of lesser involvement. This was recognized by the San Antonio consensus group who suggested a categorization of severity of DSPN, taking nerve conduction, QST, and neuropathic signs and symptoms into account (38). Our approach to staging severity of DSPN also recognizes that lesser degrees of functional abnormality are recognized by nerve conduction tests compared with clinical assessment. In our staging approach, abnormality of nerve conduction is the least degree of abnormality (stage 1A), followed by nerve conduction test abnormality and neurologic signs (stage 1B), and then by symptoms (stages 2A and 2B and 3) (12). In the present studies, direct comparison of nerve conduction test abnormality compared with neuropathic signs or symptoms by the criterion of monotone worsening over time showed that composite scores of nerve conduction tests are clearly superior. Although we argue here that nerve conduction tests at defined percentile levels are reliable minimal criteria for DSPN and are useful for epidemiologic surveys and therapeutic trials, there are compelling reasons (cost, complexity, expertise needed, and others) that for many clinical purposes they may not be needed.

The second question addressed here “Can latent dysfunction of nerves be demonstrated in patients with diabetes before nerve conduction test criteria for DSPN have been met?”, has been unequivocally answered in the affirmative. In perhaps one-third of subjects who had not met minimal nerve conduction criteria for DSPN (i.e., their Σ 5 NC nds were <97.5th percentile), we were able to demonstrate subtle functional alterations by a shift of composite nerve conduction values within the range of normal values toward abnormality and more definitively by statistically significant worsening with time, even when all or most serial nerve conduction test values fall within the normal range. This subtle functional alteration began anywhere within the range of normal test values (e.g., anywhere between the 1st and 97.5th percentile). Because these subtle alterations begin anywhere within the range of normal values, changing the minimal criteria to a lower level (e.g., to <95th or <90th percentile) would not have prevented the phenomenon from occurring. In fact, there were examples of nerve conduction test worsening within the ranges of the 1st to 50th percentile of the composite scores. We make the further observation that this worsening usually developed over years. In none of the subjects could this worsening be recognized by altered symptoms or impairments. Because serial measurements over time are needed to show significant change, it would not be a useful minimal criterion for DSPN in clinical practice.

Is there an explanation for why the use of normal limits does not detect the latent functional abnormalities reported here? Perhaps the major reason is that normal NC values vary widely over a broad range (up to 100%) so that a change in nerve function is not readily recognized when values change within these normal limits. Following the course of nerve function, therefore, has clearly been shown to be more sensitive in recognizing such change.

The third question that our studies address is “Is there evidence that the functional worsening, mainly within the range of normal nerve conduction test values, which we have demonstrated, is meaningful and relates to diabetes?” (39) The fact that the composite measure of nerve conductions (Σ 5 NC nds) significantly worsened in 42 of 90 patients and improved in only 1 patient and that this worsening was not explained by a change in age, height, or weight means that it is attributable to the diabetic condition. Furthermore, many univariate risk factors were related to this worsening. In multivariate analysis, 24-h microalbuminuria remained as the significant risk factor. In a previous study of the entire RDNS cohort, we showed that several markers of microvessel disease (retinopathy and 24-h proteinuria) and mean HbA_1c (A1C) assessed serially (and many times) over time were the major risk covariates for severity of diabetic sensory polyneuropathy at last evaluation (14). (See also ref. 39 for a discussion of the role of vascular risk factors in DSPN.)

The present evidence of early asymptomatic worsening may be useful for conducting epidemiological and prospective controlled clinical trials. We advocate use of composite nerve conduction test scores for epidemiological and prospective controlled clinical trials because they have been shown to be sensitive and reproducible, to indicate monotone change with time, and to correlate with neuropathic impairment. The magnitude of change considered to be meaningful could be estimated by correlative studies of the degree of change of sum scores of nerve conduction tests equivalent to a clinical meaningful change (40).

CDT is the cooling detection threshold. In CASE IV testing, it is expressed as the just noticeable difference test level or step (from 1 to 25), Δ°C from 30°C, percentile or normal deviate considering modality, site, age, sex, and applicable physical variables.

DSPN indicates diabetic sensorimotor polyneuropathy; other causes of this pattern of polyneuropathy have been considered and excluded. The contribution of other diabetic neuropathies should also have been excluded (cranial, entrapment, or radiculoplexus neuropathies [cervical, thoracic or lumbosacral]).

The NIS consists of a summation of 37 items of the neuromuscular examination with muscle weakness graded from 0 = normal to 4 = paralyzed and tendon reflex and sensation loss graded 0 = normal, 1 = decreased, and 2 = absent. The NIS score may vary from 0 to 244 points. Abnormality is scored taking age, sex, height, weight, and physical fitness into account. Scores for the NIS of the lower limbs may vary from 0 to 88 points.

The NSC is a 38-item score of the number, severity, and change of neuropathic symptoms.

Σ 5 NC nds indicates summated normal deviates (nds) of peroneal motor nerve amplitude, velocity, and distal latency, tibial motor nerve distal latency, and sural nerve amplitude (point A). For each attribute of nerve conduction, the percentile value is estimated for age and applicable variables. Abnormality is expressed in the upper tail of the normal distribution. To illustrate, for amplitude and velocity a percentile of 5 is changed to 95 and so on. Each percentile is converted to a normal deviate value (e.g., the 50th percentile is 0 nd, the 5th is ∼−1.6, and the 95th is ∼ +1.6. The nds of the five attributes are summed, divided by the number of measurable variables (when amplitude is 0 velocity and latency cannot be measured), and multiplied by the number of tests in the composite score (in this case five). ∑ per, tib, ul, and MNDL are the peroneal, tibial, and ulner motor nerve distal latencies, respectively, normal deviate values from percentiles expressed in the upper tail of the normal distribution and then summated by the number of measureable values and multiplied by three.

This work was supported in part by a Grant NINDS 36797 from the National Institute of Neurological Disease and Stroke.

The authors thank Vicki Clark for activities related to patient recruitment and evaluation, Jenny Davies for data analysis, and Mary Lou Hunziker for manuscript preparation.