89 number of overlapping abnormal classifications found by the two methods. Only 8% of CDT measurements and 6% of WDT measurements were abnormal with both methods. Fourthly, the results presented in Tables 4 and 5, show that leaving MLe out of the calculations results in improved diagnostic accuracy. Moreover, comparing WDT MLi, CDT MLi, TSL, PHS, CPT and HPT measurements at the test side with the contralateral side shows that a 5% lower sensitivity was found at the test side (see Table 5). Since MLe was only performed at the test side, this could indicate that MLe had a negative impact on the sensitivity of the combined measurements at the test side. To summarize, a high degree of variance in the number of levels required to determine a detection threshold, a low specificity, not improving MLi, a risk of desensitization and decreased accuracy are reasons not to measure with MLe. A minor limitation of our study was that patients were not age- and sex-matched during the inclusion process. Since the study was performed in a population with a combination of rare diseases and a challenging protocol for SFN diagnosis, the priority was to include a population with a pure diagnosis of sarcoidosis as well as SFN, and to minimize comorbidities. Additionally, results from TTT measurements were corrected for age, sex and measuring site using z-transformation, which should limit the bias caused by the lack of age- and sex-matching. A major limitation of our study was the lack of structural assessment with skin biopsy. The reason for this was that its complex and labor-intensive protocol would result in high costs and limited clinical feasibility. Since multiple Dutch hospitals are struggling with the implementation of skin biopsies for SFN diagnosis, this highlights the challenges of implementing 50 µm IENFD as a diagnostic standard for SFN. Diagnostic accuracy measures the level of agreement between index test results and the outcome of a reference standard. Not using any diagnostic technique to classify SFN and therefore inverting the standard clinical approach, impairs validity and reproducibility of the ROC-analysis. Alternatively, we investigated methods to test for diagnostic accuracy when no gold standard is available,22 one of them being known as “validate index test results”. This method abandons the test accuracy paradigm and test results were related to other relevant clinical characteristics. Hence, we defined “probable SFN” as the reference standard to test the diagnostic accuracy of TTT measurements. In order to be able to use this method, we emphasize that we applied strict inclusion criteria, resulting in a pure study population with symptoms and at least two clinical signs of SFN. The high SFNSL scores of sarcoidosis patients with probable SFN compared with sarcoidosis patients without SFN confirms the strict patient selection method. Although “probable SFN” remains suboptimal for diagnostic accuracy calculations, it could still be used as an informative method to show the effects of different parameters and measuring site selections, rather than showing actual diagnostic accuracy. Moreover, those analysis nicely shows how to determine a cutoff value for TTT NOAs and indicates which parameters show added values. In contrast with papers proposing the assessment of detection thresholds with MLe, our study shows the limitations of MLe. Moreover, current diagnostic criteria suggest the use of decreased IENFD and/or abnormal QST for the definitive SFN diagnosis. However, inconsistent use of the term “abnormal QST” is seen between studies and the definition does not correct for the number of TTT measurements. The large number of parameters measured with TTT and the heterogenic results of individual parameters make it quite challenging to interpret QST, while TTT NOAs clearly relates to the severity of SFN, which is underlined by the good correlation between TTT NOAs and SFNSL. 93 5
RkJQdWJsaXNoZXIy MjY0ODMw