Accuracy of neuropathic pain measurements in patients with symptoms of polyneuropathy: validation of painDETECT, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs, and Douleur Neuropathique 4

Abstract

Pain is a common symptom in patients referred to polyneuropathy assessment. Diagnostic evaluation and choice of treatment may depend on whether the pain is likely to be neuropathic or not. This study aimed to investigate the diagnostic accuracy of 3 tools commonly used to differentiate between neuropathic and nonneuropathic pain. To accomplish this, we included patients with bilateral distal lower extremity pain, referred to neurological outpatient clinics at 5 Norwegian University hospitals for polyneuropathy assessment. The patients filled in Norwegian versions of painDETECT, the Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs (S-LANSS), and the clinician-rated Douleur Neuropathique 4 (DN4). All patients underwent a clinical examination and nerve conduction measurements and were classified according to the NeuPSIG neuropathic pain criteria (reference standard). In total, 729 patients were included, of which 63% had neuropathic pain by the reference standard. Only DN4 demonstrated high sensitivity (0.87), whereas all 3 tools had low specificity (≤0.65). Importantly, the tools' predictive ability was unsatisfactory; The probability of getting a correct test result was 3 quarters at best, and at worst, no better than two fifths. Consequently, we show that neither DN4, painDETECT, nor S-LANSS can be confidently used to assess neuropathic pain in a neurological outpatient population with symptoms of polyneuropathy.

In Brief

Among patients referred to neurological outpatient clinics with symptoms of polyneuropathy, painDETECT, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs, and Douleur Neuropathique 4 demonstrated unsatisfactory predictive diagnostic accuracy.

1. Introduction

Neuropathic pain represents a substantial part of the worldwide burden of pain, with an estimated population prevalence of 7% to 10%.⁶⁶ The prevalence of neuropathic pain is expected to increase, as an ageing population, improved cancer survival rates and rise in diabetes mellitus cases are likely to cause an upswing in painful polyneuropathies.¹³ This increase imposes a need to identify patients with neuropathic pain in patient populations with likely polyneuropathy, to offer optimized care.

Patients with polyneuropathy often present with distal pain and a mixture of neurological signs and symptoms. This presentation is common in the neurological outpatient clinic, where clinicians need to determine whether the pain is likely to have a neurological cause or component. Discriminating between predominantly neuropathic and nonneuropathic pain is important because it can have direct consequences for further examination and choice of treatment.

Many clinical measurement tools have been developed to help clinicians distinguish between pain that is predominantly neuropathic and nonneuropathic. Among the most commonly used for both clinical and research purposes are 2 questionnaires for self-assessment: painDETECT¹⁸ and the Self-Completed Leeds Assessment of Neuropathic Pain (S-LANSS),⁵ and one structured clinical evaluation tool, the “Douleur Neuropathique 4” (DN4).⁸ All 3 are frequently used and have been cross-culturally adapted or translated to many languages. However, it is not clear whether they are adequately valid or if the questionnaire items are sufficiently consistent and suitable for use in patients with polyneuropathy symptoms in an outpatient neurological setting.³⁷

The diagnostic accuracy of neuropathic pain tools depend on the patient population in question.^{2,20,24,26,28,53} Most of the knowledge about the diagnostic accuracy of painDETECT, S-LANSS, and DN4 in patients with polyneuropathy is derived from studies of patients in pain clinics with mixed etiology, including, eg, arthritis, phantom pain, complex regional pain syndrome, or back pain.^{5,14,32,41,49} This is problematic because different patient groups display diverse clinical presentations and degrees of co-morbidity that may affect the score of the clinical tools. For example, the score increases with pain severity,^12,18,35,43 anxiety,^18,54 depression,^18,54,68 and reduced quality of life.^11,68 To confidently apply painDETECT, S-LANSS, and DN4 in patients with possible polyneuropathy, the validity of these tools should be evaluated in the target population.

Therefore, the general aim of this study was to assess the diagnostic accuracy of painDETECT, S-LANSS, and DN4 in distinguishing between neuropathic and nonneuropathic pain in patients referred to polyneuropathy assessment. Specific objectives were to estimate the tools' discriminative and predictive abilities. Our secondary aim was to evaluate the tools' internal consistency and explore which items contributed the most to false positives and false negatives. This study is part of a large Norwegian multicenter study including 5 clinical neurophysiology departments at our University hospitals.

2. Methods

2.1. Overview

This study was performed in 2 steps. First, the original version of painDETECT, S-LANSS, and DN4 were translated and cross-culturally adapted into Norwegian. Next, diagnostic accuracy and internal consistency of the 3 tools were tested using a cross-sectional design. The NeuPSIG criteria¹⁷ were treated as the reference standard for diagnosing neuropathic pain. We included patients with pain in the distal lower extremities who were referred to neurological outpatient clinics for polyneuropathy assessment. The study was approved by The Regional Committee for Medical Research Ethics, South-East Norway (ref no. 2017/1593), and all subjects gave informed consent before inclusion.

2.2. Translation and cross-cultural adaptation to Norwegian

The process of translation and cross-cultural adaptation of painDETECT, S-LANSS, and DN4 was performed according to international guidelines^3,23 by Grotle, Nilsen, and Killingmo (2015, Appendix 5-7, available as supplemental digital content at https://links.lww.com/PAIN/B729). Two native Norwegian speakers (1 philologist and 1 clinician) independently translated the original tools from English into Norwegian. The 2 Norwegian versions were synthesized into one version before being back translated into English. Two native English speakers (1 philologist and 1 clinician), both blinded to the original tools, independently performed the back translation and synthesized the 2 English versions into one. An expert committee consisting of the translators and 2 researchers from our research group reviewed all translations. In a formal meeting, the committee discussed deviations until consensus on a prefinal version was reached. The goal was that the prefinal Norwegian tool should be as concise and easy to understand as possible. The prefinal version was tested on 10 patients from the neurological outpatient clinic at Oslo University Hospital. None of the patients had difficulties understanding the meaning of items or responses, and they found it easy to comprehend. No further changes were made, and the final versions of the Norwegian tools evaluated in this study are the same as the prefinal versions.

2.3. Study sample and recruitment procedure

Patients aged 18 to 70 years, referred to neurological hospital outpatient clinics for polyneuropathy assessment, were consecutively asked to participate in the study. Five hospitals participated in the multicenter data collection between May 2017 and July 2021: Oslo University Hospital; Haukeland University Hospital, Bergen; Stavanger University Hospital; Trondheim University Hospital (St. Olavs Hospital); and The University Hospital of North Norway, Tromsø. The exclusion criteria were acute polyneuropathies (eg, acute inflammatory demyelinating polyradiculopathy and acute motor axonal neuropathy), nerve entrapment without polyneuropathy, limited capacity to give informed consent (eg, language barrier, dementia, and psychiatric illness), and patients being too sick to participate (eg, bed-ridden and high fever).

2.4. Measurement tools for neuropathic pain

The painDETECT questionnaire was originally developed by Freynhagen et al.¹⁸ as a screening tool for neuropathic pain components in patients suffering from low back pain. The tool is self-completed and is made up of a pain drawing, 3 questions regarding current, maximum, and average pain intensity (Numeric Rating Scale and Visual Analogue Scale [0-10], not scored), as well as 3 distinct main parts: gradation of pain, pain course pattern, and answering whether the pain radiates. The gradation of pain consists of 7 items for characterizing and grading pain and other sensations, from never experiencing them (0) to experiencing them very strongly (5). The questions cover pain sensation (eg, burning and shooting pain), paresthesias (eg, numbness and tingling), and allodynia (to light touch, pressure, heat, or cold). Four different pain course patterns are illustrated and described, and the patient chooses the one that best describes their course of pain. Two patterns describe persistent pain with slight fluctuations or pain attacks—these are graded 0 or −1, ie, principally persistent pain patterns do not contribute positively to the painDETECT score. The remaining 2 pain patterns describe pain attacks with or without baseline pain and are both graded as +1. Finally, the patient answers whether the pain radiates. A positive answer gives +2, whereas a negative answer is neutral (0). Maximum sum score is 38. The original painDETECT suggests 2 cutoffs: a sum score of <12 means it is unlikely that the pain has a neuropathic component (<15% chance), whereas a cut-off of >18 indicates that it is likely that the pain has a neuropathic component (>90% chance). In this study, only the cut-off at >18 was used.

Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs⁵ is a revised version of the Leeds Assessment of Neuropathic Symptoms and Signs,⁴ intended to make self-completion possible. It is a 7-item tool for identifying pain that is predominantly neuropathic in origin. It was originally validated for a broad mix of patients with chronic pain in pain clinics, day care wards, and inpatient wards. The first 5 items are weighted descriptors of the patients' pain and other symptoms, to which the patient answers yes (graded as 1, 2, 3, or 5 depending on the item) or no (0). The questions cover paresthesias (eg, pins and needles and numbness), autonomic response (skin color change), hyperesthesia to touch, paroxysmal or shooting pain (eg, electric and bursting), and burning sensations. For the last 2 items, the patients examine themselves with gentle pressing and rubbing of the painful area. If gentle rubbing leads to discomfort, pain, or paresthesia, it is scored as +5, whereas tenderness or numbness following gentle pressing scores +3. Maximum sum score is 24. The original S-LANSS operates with a cut-off at ≥ 12, suggesting pain of predominantly neuropathic origin, and this cutoff was used in this study.

DN4 was developed by Bouhassira et al.⁸ and originally validated in patients with chronic pain of different etiologies and at least moderate pain severity (>40 mm on a 100 mm Visual Analogue Scale). It is intended to be a clinician-administered diagnostic tool for neuropathic pain. Douleur Neuropathique 4 consists of 2 main subgroups of items: the 7-item interview (Q1 about pain characteristics and Q2 associated symptoms) and the 3-item clinical examination (Q3 about sensation loss and Q4 about allodynia). All items within questions are scored as yes (+1) or no (0). The first items cover whether the pain is characterized by burning, painful cold, or electric shock-like sensations, and whether the pain is associated with tingling, “pins and needles,” numbness, or itching. Following this, the patient is examined for hypoesthesia to touch and pin prick and for mechanical brush allodynia. Maximum sum score is 10. The final 10-item tool was developed from an original 17-item questionnaire, and a cut-off score at ≥4 for the diagnosis of neuropathic pain was determined by maximal Youden index. The same cut-off score was used in this study.

2.5. Reference standard

The reference standard for the diagnosis of neuropathic pain was the NeuPSIG criteria.¹⁷ Published by the International Association for the Study of Pain's special interest group, these criteria are increasingly recognized as the gold standard for assessing neuropathic pain in clinical practice and for research purposes. The NeuPSIG criterion classifies patients by the level of confidence that neuropathic pain is present: unlikely, possible, probable, and definite neuropathic pain. Possible neuropathic pain requires a history of relevant neurological lesion or disease, as well as a pain distribution that is neuroanatomically plausible. Failure to meet these first criteria classifies the patient's pain as unlikely to be neuropathic. Probable neuropathic pain requires the former, in addition to an examination revealing that the pain is associated with sensory signs in the same distribution. Definite neuropathic pain requires all of the above, as well as a confirmatory diagnostic test for a lesion or disease of the somatosensory nervous system that can explain the pain.

In this study, patients were dichotomized into 2 groups: neuropathic pain (definite, probable) and no neuropathic pain (unlikely, possible). Since all patients in the study were subject to sensitive confirmatory tests for a lesion or disease of the somatosensory nervous system, patients defined as having probable neuropathic pain all have negative electrodiagnostic tests. To ensure that any false positives in the reference standard did not change our results, a sensitivity analysis was performed where patients with probable neuropathic pain were also included as true negatives (no neuropathic pain) and excluded from the analysis. Furthermore, to reduce the chance of error in NeuPSIG classification, all data were reviewed and validated by a third party team from the research group. Any inconsistencies were flagged, and agreement was reached through discussion and assessment of the patient journal with a representative physician at the hospital in question.

2.6. Assessment procedure

Both painDETECT and S-LANSS were sent by traditional mail and filled out on paper by the patients preferably 0 to 14 days before their appointment. Douleur Neuropathique 4 was administered by face-to-face interview by a physician clinical neurophysiologist as part of the routine assessment of polyneuropathy. History-taking, clinical examination, and nerve conduction studies (NCS) were performed as part of routine assessments for polyneuropathy. All NCS were performed in concordance with the Norwegian national guidelines.³⁴ A minimum of 2 sensory nerves in the feet were tested (sural nerve and medial plantar nerve) as well as 2 motor nerves (tibial nerve and peroneal nerve), including F-responses. If the neurographic findings were not clear, one extra sensory nerve was tested (superficial peroneal nerve).

Of particular note, quantitative thermal testing was considered a confirmatory test for small-fiber lesions. This is not completely in line with the NeuPSIG classification but consistent with clinical practice and recent guidelines for diagnosing small-fiber neuropathy.^57,69 Detection and pain thresholds for cold and heat were obtained in the lower extremities. Only detection thresholds (cold or warm) were used to define small-fiber abnormality—pain thresholds were used as supportive information. The method of limits was used, with a baseline temperature of 32°C, ramp-rate of 1°C/s, and thermode size of 9 to 12 cm², as per the national guidelines³⁴ and the hospitals' own protocol.

For both NCS and quantitative thermal testing, local reference values were applied when possible. Alternatively, references values from PowerPack^TM (Stefan Stålberg Software AB, Helsingborg, Sweden)⁵² or the national guidelines (quantitative thermal testing) were used.

2.7. Polyneuropathy assessment

Polyneuropathy was defined in accordance with the criteria for diabetic polyneuropathy suggested by Tesfaye et al.,⁵⁸ as these also fit general distal, symmetrical polyneuropathies. Patients were classified as having polyneuropathy if they fulfilled the criteria for confirmed diabetic peripheral sensorimotor polyneuropathy (DPSN). The criteria for confirmed DPSN in myelinated nerve fibers require the presence of an abnormal NCS or small-fiber test and symptom(s) or sign(s) of neuropathy. In accordance with the Norwegian national guidelines for clinical neurophysiology,³⁴ at least 2 nerves of different roots had to be abnormal to constitute a positive finding on NCS (3 if abnormal medial plantar nerve). For small-fiber neuropathy, the Tesfaye criteria require the presence of length-dependent symptoms, clinical signs of small-fiber damage, normal (sural) NCS, and either altered intraepidermal nerve fiber density or abnormal thermal detection thresholds in the feet.⁵⁸ In this study, the latter was most commonly used.

2.8. Statistical analysis

2.8.1. Sample size

The current study is part of a large, multicenter study. Minimum sample size was conservatively calculated based on Buderer recommendations.⁹ With expected sensitivity and specificity of maximum 0.8, sample prevalence for neuropathic pain of 50%, and 95% confidence intervals with 5% precision, the study required a minimum of 492 patients included in the analyses.

2.8.2. Analyses of data quality

Descriptive statistics were conducted to explore distribution of the included tools; continuous variables with normal distribution are presented by mean values with SDs, whereas skewed variables are presented by their median with interquartile ranges. Categorical variables are presented as absolute and relative (percentage) frequencies.

Each participating hospital coded their own patients independently before data pooling. Item missingness was analyzed visually and by Little's Missing Completely at Random (MCAR) test and was found to be random. The missing percentage of clinical tool items ranged from 0% to 4.1%, with the exception of painDETECT's “pain pattern” (8%) and “radiating pain” (17.4%). Missing items were imputed by single imputation. The method of single imputation was based on fully conditional specification (chained equations) (SPSS v25). Twenty imputations (10 iterations) were run, and the rounded average value (discrete items) or mode (categorical items) of these were used for the final calculations.

Floor and ceiling effects (ie, how many scored “never experiencing [item]” (0) and “experiencing [item] very strongly” (5)) for the pain gradation items in painDETECT were assessed through frequency tables, with a cut-off of >15%.

2.8.3. Main analyses of diagnostic accuracy and internal consistency

We assessed diagnostic accuracy in several ways. First, agreement between the neuropathic pain tools and the NeuPSIG criteria was determined by calculating Cohen kappa. Kappa scores indicating agreement between the tool and the NeuPSIG classification were interpreted as none to slight (0.01-0.20), fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1.00).³⁹

Second, potential discriminative ability (to distinguish between neuropathic and nonneuropathic pain) was analyzed by receiver operating curves (ROC) analysis and presented by the area under the ROC curves (AUC). Area under the ROC curve values were considered nondiscriminative (0.5-0.6), poor (0.6-0.7), acceptable (0.7-0.8), excellent (0.8-0.9), or outstanding (>0.9).²⁹ However, since AUC values do incorporate cut-offs that are clinically nonsensical,³⁶ the absolute discriminative ability was calculated for 3 different cut-offs: the original cut-off for each tool, a “best cut-off” (based on the highest Youden Index, ie, sensitivity + specificity − 1), as well as a cut-off weighted towards the highest possible sensitivity with specificity above 0.5. In the event of substantially better sensitivity or specificity at the additional cut-offs, predictive values would also be calculated.

Third, the tools' predictive ability (how well the tools can predict the presence of neuropathic pain in a real-life clinical setting) was assessed by positive predictive values (PPV), negative predictive values (NPV), and positive/negative likelihood ratios. Predictive values are clinically intuitive and important for the clinician interpreting actual results, ie, “how likely is it that my patient has/doesn't have neuropathic pain, given the test result?”^36,62 With a sample prevalence of neuropathic pain close to 50%, likelihood ratios were understood as how much the result of each tool increased the probability of neuropathic pain (from pretest to posttest). Specifically, the likelihood ratios were interpreted as small, rarely important (1-2 and 0.5-1), small but sometimes important (2-5 and 0.5-0.2), moderate (5-10 and 0.1-0.2), and large, often conclusive (>10 and <0.01).³⁰

Furthermore, we calculated Cronbach alpha (α) to determine the internal consistency of the 3 tools in our particular study sample. Item total correlations (correlation between each item and the total score) were calculated as a complementary measure to interpret Cronbach α values and to assess each item's discriminative ability. Cronbach alpha values between 70 and 95 were considered to imply good internal consistency, and item total correlations between 0.3 and 0.5 were considered discriminative.^15,42

2.8.4. Additional analyses

To examine which items contributed the most to false positives and false negatives, we explored item response frequency tables for each possible outcome.

Analyses were performed in SPSS v25 (IBM, Armonk, NY) and R v4.1.1⁵⁵ with the packages tidyverse v1.3.1⁷⁰, table1 v1.4.2⁴⁵, pROC v1.18,⁴⁶ and epiR v2.0.33.⁵¹

3. Results

In total, 1498 patients were eligible for inclusion. We recruited 1163 patients, of which 729 were included in the main analysis (Fig. 1). Patient demographics and clinical variables are presented for the entire sample, as well as stratified by the presence of neuropathic pain (Table 1). The group with neuropathic pain was, on average, older, consisted of more males and diabetes patients, had a much higher prevalence of polyneuropathy and higher use of pain medication. The prevalence of neuropathic pain and polyneuropathy was 63% and 53%. The mean (SD) scores for painDETECT, S-LANSS, and DN4 were 17.85 (7.23), 13.03 (6.40), and 4.83 (2.04), respectively.

Figure 1.: Flowchart of the recruitment process. DN4, Douleur Neuropathique 4; S-LANSS, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs.

Table 1 - Patient demographics and clinical variables.

	All patients (n = 729)	Nonneuropathic pain* (n = 264)	Neuropathic pain* (n = 465)
Age, y, mean (SD)	55 (11)	52 (11)	57 (10)†
Sex, female, n (%)	405 (56)	169 (64)	236 (51)†
Pain medication, yes, n (%)	336 (46)	99 (38)	237 (51)†
Pain duration, n (%)
1-3 mo	12 (2)	6 (2)	6 (1)
3-12 mo	124 (17)	48 (18)	76 (16)
1-5 y	289 (40)	102 (39)	187 (40)
>5 y	0 (0)	0 (0)	0 (0)
Mean pain last 3 mo, mean (SD)‡	5.5 (2.1)	5.4 (2.2)	5.6 (2.1)
Diabetes, yes, n (%)	197 (27)	36 (14)	161 (35)†
Type 1	40 (6)	13 (5)	27 (6)†
Type 2	145 (20)	21 (8)	124 (27)†
Polyneuropathy etiology, n (%)
Diabetes	153 (21)	11 (4)	142 (31)†
Idiopathic	154 (21)	13 (5)	141 (30)†
Chemotherapy-induced	26 (4)	1 (0)	25 (6)†
Small-fiber	112 (15)	14 (5)	98 (21)†
Hereditary	9 (1)	1 (0)	8 (2)
Posttraumatic nerve injury	2 (0.3)	1 (0)	1 (0)
Postoperative (iatrogenic)	1 (0.2)	0 (0)	1 (0)
Other§	103 (14)	65 (25)	38 (8)†
None or missing	169 (23)	158 (60)	11 (2)†
NeuPSIG grading, n (%)
Unlikely	157 (21)	157 (59)	0 (0)
Possible	107 (15)	107 (41)	0 (0)
Probable	103 (14)	0 (0)	103 (22)
Definite	362 (50)	0 (0)	362 (78)

^*Patients classified as having neuropathic pain (definite, probable) and no neuropathic pain (unlikely, possible) according to the NeuPSIG criteria for diagnosing neuropathic pain.

^†Significant between the 2 NeuPSIG groups (neuropathic and nonneuropathic) at P < 0.05 after Bonferroni adjustment.

^‡Numeric Rating Scale (0-10), where 0 = no pain and 10 = worst imaginable pain.

^§Chronic inflammatory neuropathies and neuropathies related to vitamin deficiency, Lyme disease, and toxins (eg, alcohol).

3.1. Diagnostic accuracy for distinguishing between neuropathic and nonneuropathic pain

There was none or slight to fair agreement between the 3 tools and the reference standard. Cohen kappa values for the NeuPSIG classification and painDETECT were κ = 0.12 (95% confidence interval [CI] 0.05-0.19, P = 0.002), κ = 0.13 (95% CI 0.05-0.22, P = 0.001) for S-LANSS, and κ = 0.38 (95% CI 0.31-0.45, P < 0.001) for DN4.

Discriminative ability by AUC was acceptable for DN4 with an AUC of 0.7 7, poor for S-LANSS, and nondiscriminative for painDETECT (Fig. 2). Youden index for the original cut-offs were 0.14 for painDETECT, 0.14 for S-LANSS, and 0.37 for DN4, the latter largely driven by a sensitivity of 0.87. Changing the cut-offs to maximize Youden's index or improve sensitivity did not improve discriminative ability to a clinically relevant degree (Fig. 2).

Figure 2.: Area under the receiver-operating curves (with 95% confidence intervals in the frame) for painDETECT, S-LANSS, and DN4. Sensitivity and specificity for the following 3 cut-offs are included for each tool: (1) the original cut-off, (2) the cut-off maximizing the Youden index (sensitivity + specificity − 1), and (3) the cut-off with the highest possible sensitivity while keeping specificity above 0.5. DN4, Douleur Neuropathique 4; S-LANSS, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs.

Among patients with a positive result on either tool, the probability of having neuropathic pain was 71% to 75% (Table 2). On the flipside, among patients with a negative result on painDETECT or S-LANSS, the probability of not having neuropathic pain was 39% to 42%, rising to 68% for DN4.

Table 2 - Discriminative and predictive ability of painDETECT, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs, and Douleur Neuropathique 4.

		Neuropathic pain (NeuPSIG)				Point estimate (95% CI)
painDETECT	Positive Negative Total	Positive 194 195 389	Negative 73 127 200	Total 267 322 589	Sensitivity Specificity Positive predictive value Negative predictive value Positive likelihood ratio Negative likelihood ratio	0.50 (0.45, 0.55) 0.64 (0.56, 0.70) 0.73 (0.67, 0.78) 0.39 (0.34, 0.45) 1.37 (1.11, 1.68) 0.79 (0.68, 0.91)
S-LANSS	Positive Negative Total	Positive 232 152 384	Negative 95 110 205	Total 327 262 589	Sensitivity Specificity Positive predictive value Negative predictive value Positive likelihood ratio Negative likelihood ratio	0.60 (0.55, 0.65) 0.54 (0.47, 0.61) 0.71 (0.66, 0.76) 0.42 (0.36, 0.48) 1.30 (1.10, 1.54) 0.74 (0.62, 0.88)
DN4	Positive Negative Total	Positive 399 62 461	Negative 132 130 262	Total 531 192 723	Sensitivity Specificity Positive predictive value Negative predictive value Positive likelihood ratio Negative likelihood ratio	0.87 (0.83, 0.90) 0.50 (0.43, 0.56) 0.75 (0.71, 0.79) 0.68 (0.61, 0.74) 1.72 (1.52, 1.95) 0.27 (0.21, 0.35)

Cut-off value for neuropathic pains: painDETECT ≥ 19, S-LANSS ≥ 12, and DN4 ≥ 4.

95% CI, 95% confidence interval; DN4, Douleur Neuropathique 4; S-LANSS, Self-Administered Leeds Assessment of Neuropathic Symptoms and Signs; NeuPSIG, the NeuPSIG criteria for diagnosing neuropathic pain were used as the reference standard, with patients classified as having neuropathic pain (definite, probable) and no neuropathic pain (unlikely, possible).

None of the tools reached a positive likelihood ratio above 2, which means that a positive test result will rarely give the clinician important information, as it will not appreciably increase the posttest probability of neuropathic pain. Only DN4 had a somewhat promising negative likelihood of 0.27, meaning a negative test implies a small, but sometimes important decrease in the likelihood of the patient having neuropathic pain. Neither painDETECT nor S-LANSS reached a negative likelihood ratio of <0.5, showing poor ability for ruling out disease.

In the sensitivity analysis, excluding the “NeuPSIG probable” group had little effect on overall diagnostic accuracy, with the exception of an increase in DN4's NPV (0.68-0.75). Including the probable group as true negatives lead to a decrease in specificity for painDETECT (0.65-0.58) and DN4 (0.50-0.41). This also decreased PPV by 0.15 for all tools and increased NPV by approximately 0.12 (Appendix 1, available as supplemental digital content at https://links.lww.com/PAIN/B729).

3.2. Data quality and internal consistency

painDETECT had good internal consistency (Table 3), with a Cronbach alpha score of α = 0.79, and all items except radiating pain showed good item total correlation (the item pain pattern was excluded from the analysis because of its −1 to 1 scoring). Neither S-LANSS nor DN4 reached good internal consistency. S-LANSS had a Cronbach α of 0.61, with 2 items' item total correlation below 0.3. Douleur Neuropathique 4 had a Cronbach α of 0.58, with 6 items below item total correlation <0.3. For painDETECT, a floor effect was observed for the items burning pain, light touch, electric shocks, painful temperature, and light pressure, whereas no ceiling effects were present.

Table 3 - Item quality and internal consistency of painDETECT, Self-Completed Leeds Assessment of Neuropathic Symptoms and Signs, and Douleur Neuropathique 4.

	Missing, %*	No, n (%)	Yes, n (%)	Item total correlation†	Cronbach α (if deleted‡)
DN4 (n = 723)					0.58
Burning pain	0.5	255 (35)	468 (65)	0.19	0.57
Cold pain	1.0	546 (76)	177 (25)	0.20	0.56
Electric shocks	0.7	377 (52)	346 (48)	0.22	0.56
Tingling	0.8	253 (35)	470 (65)	0.25	0.55
Pins and needles	0.4	149 (21)	574 (79)	0.33	0.53
Numbness	0.5	136 (19)	587 (81)	0.33	0.53
Itching	1.4	554 (77)	169 (23)	0.23	0.56
Hypoesthesia touch	1.4	453 (63)	270 (37)	0.35	0.52
Hypoesthesia prick	1.8	407 (56)	316 (44)	0.31	0.53
Painful brush	1.9	623 (86)	100 (14)	0.17	0.57
S-LANSS (n = 589)					0.61
Tingling and pins and needles (0/5)	1.2	100 (17)	490 (83)	0.21	0.62
Color (0/5)	2.2	465 (79)	125 (21)	0.24	0.61
Sensitive to touch (0/3)	1.9	323 (55)	267 (45)	0.49	0.52
Electric shocks (0/2)	1.2	233 (40)	357 (61)	0.30	0.59
Burning pain (0/1)	2.2	253 (43)	337 (57)	0.36	0.60
Rubbing discomfort (0/5)	3.6	304 (52)	286 (49)	0.52	0.49
Numbness (0/3)	3.2	200 (34)	390 (66)	0.42	0.55
painDETECT (n = 589)		Min score, n (%)	Max score, n (%)		0.79
Pain pattern (−1 to 1)§	8.0	151 (25)	153 (26)
Radiating pain (0/2)	17.4	316 (53)	280 (47)	0.23	0.80
Burning pain‖	3.9	101 (17)	58 (10)	0.48	0.77
Tingling‖	2.9	59 (10)	77 (13)	0.56	0.76
Light touch‖	2.7	206 (34)	24 (4)	0.60	0.75
Electric shocks‖	4.1	151 (25)	52 (9)	0.55	0.76
Painful temperature‖	3.2	224 (38)	26 (4)	0.56	0.76
Numbness‖	3.0	47 (8)	83 (14)	0.42	0.78
Light pressure‖	3.4	160 (27)	39 (7)	0.57	0.76

^*Before imputation, results otherwise based on imputed data set.

^†Correlation between each item and the total score.

^‡Cronbach α calculation without the item in question.

^§Pain pattern excluded from calculation of item total correlations.

^‖Items are scored on a 0 to 5 point scale (5 represents more severe symptoms).

3.3. Item response frequency

The item response frequency tables for each tool showed that true positive and true negative patients had different item score distributions as compared with their wrongly classified counterparts (Appendix 2-4, available as supplemental digital content at https://links.lww.com/PAIN/B729). However, several items may have contributed to the false-positive and false-negative results. All patients frequently reported pins and needles, tingling, numbness, and sometimes burning or electric/shooting pain, in particular. Patients classified as false negatives rarely reported items related to itching, hypoesthesia to touch, temperature allodynia, or mechanical allodynia. False-positive patients commonly experienced all items, except for changing color (S-LANSS), cold pain (DN4), itching (DN4), hypoesthesia to touch (DN4 examination), hypoesthesia to pinprick (DN4 examination), and brushing pain (DN4 examination). Overall, items related to physical examination, small-fiber lesions, and mechanical allodynia contributed the least to false positives and negatives.

4. Discussion

The results from our large multicenter study on patients with suspected polyneuropathy showed acceptable discriminative ability for DN4. The discriminatory ability of S-LANSS was poor and nondiscriminative for painDETECT. Douleur Neuropathique 4 showed a promising negative likelihood value and the best predictive values of the 3, although neither tool demonstrated good overall predictive ability. painDETECT demonstrated good internal consistency in our study sample, whereas S-LANSS and DN4 did not. Regardless of the neuropathic pain classification, patients frequently reported pins and needles, tingling, and numbness, which are common symptoms of polyneuropathy.

Owing to the lack of published papers on polyneuropathy populations, it is difficult to compare our results directly with previous studies. We identified one study of DN4 in patients with diabetic polyneuropathy,⁵⁰ which used an earlier version of the NeuPSIG criteria⁶¹ as the reference standard, with both probable and definite reflecting neuropathic pain. The authors reported moderately good discriminative and predictive ability, concluding that the trade-off in diagnostic accuracy may be worthwhile because of its simplicity and user friendliness. Aside from this, many studies report on the diagnostic accuracy of the 3 tools in other patient populations and across several languages but with widely different results. For example, Youden index for S-LANSS ranges from 0.06 in a study on cancer patients²⁸ to a near perfect 0.95 in patients with mixed etiology.⁶³ Likewise, studies on painDETECT and DN4 report Youden indices from below 0.2^16,44 to 0.71²² and 0.92,²⁵ respectively. All 3 tools usually perform better in other patient populations than in this study, with DN4 tending to outperform the other 2.

It is not surprising that DN4 has shown acceptable diagnostic accuracy in a previous study and performs best in this study, when one considers that it is the only tool administered by healthcare professionals (eg, physicians). Such a healthcare professional could likely also use the NeuPSIG guidelines to determine whether the patient is likely to have polyneuropathy. However, when compared with DN4, the NeuPSIG guideline requires a higher level of clinical judgment to establish probable (or definite) polyneuropathy. Since the time needed for clinical examination is only marginally longer for the NeuPSIG approach, the choice between the 2 may depend more on the clinical skills of the available personnel.

The nature of our sample likely contributes to the suboptimal predictive abilities in this study. The tools were originally developed for use in patients with chronic low back pain (painDETECT), or in mixed groups of patients (S-LANSS, DN4), with painDETECT and DN4 primarily intended as screening tools. Singling out homogeneous groups, for which the tools were not originally intended, can be expected to affect their diagnostic accuracy. As screening tools, the results could also plausibly be expected to reflect a greater ability to pick up on neuropathic pain (eg, higher sensitivity and positive likelihood ratios), although this was only true for DN4 in our sample. We found that numbness, tingling, and pins and needles were among the most frequently reported items (cf. “easy” items in item response theory¹⁵). This is in concordance with earlier studies on all 3 tools, across an array of different patient populations.^{1,7,8,11,12,16,21,25,27,32,40,48,53,56,59,60,64,65,67,68} By themselves, “easy items” can be expected to be sensitive, with discriminative power in patients with few symptoms.¹⁵ However, numbness, tingling, and pins and needles are also cardinal symptoms of polyneuropathy, that is, hypoesthesia and paraesthesia in the feet. This entails that for patients with symptoms of polyneuropathy, these items can be expected to form a “baseline score” which artificially inflates sensitivity, reduces specificity, and consequently leads to an increase in false positives and worse predictive ability. Although DN4 had the best predictive ability in our sample, it is still mediocre with likelihood ratios implying a small effect on posttest probability and results only being correct 68% to 75% of the time. Thus, neither tool can be used confidently in patients with symptoms of polyneuropathy.

The multidimensional nature of pain makes it particularly difficult to create clinical tools that successfully isolates and measures certain elements or aspects. Earlier studies have shown that both painDETECT and DN4 scores may be associated with pain severity, pain catastrophizing, health-related quality of life, depression, anxiety, stress, and disability.^18,54,68 The effect of these factors on scoring may help explain the low specificity and the low PPV observed despite the high true prevalence in our sample. First, our study sample experienced high pain severity, which may inflate the scores of the clinical tools, leading to more false positives.^{6,10,12,18,35,38,43} Second, a substantial portion of the patients included suffered from diabetic polyneuropathy—a group that often has a high burden of illness related to anxiety, depression, insomnia, and disability.^33,47 As such, instead of picking up on only neuropathic pain, the tools could be particularly sensitive to patients with comorbidities and low health-related quality of life, making them incapable of ruling out neuropathic pain in patients with a high burden of disease.

The internal consistency of the 3 tools varied. painDETECT (pain pattern excluded) had good internal consistency, largely comparable with previous reports.^{6,10,14,18,22,38} By contrast, we found poorer internal consistency for S-LANSS and DN4, for which earlier studies have been more conflicting (eg, S-LANSS^2,19,48,63 and DN4^31,49,59,65). Although Cronbach α can be expected to increase with the number of items, the tools are comparable (7, 8, and 10 items) and the observed differences do not follow such a pattern in our data, nor in previous reports. Furthermore, neither Cronbach α nor item total correlation can discriminate between different constructs. This means that if the clinical tools do in fact also pick up on, eg, polyneuropathy in itself, health-related quality of life, or psychosocial factors, Cronbach α values make little sense, and we cannot deduce which constructs the items are actually correlating with or even if they correlate with the same ones. This could help explain some of the lower item total correlation values found in S-LANSS and DN4 but does not resolve whether painDETECT is actually more consistent in measuring neuropathic pain or if the items measure different constructs that correlate well with each other. Going forward, future studies may look to explore the construct validity of painDETECT, S-LANSS, and DN4 in patients with polyneuropathy, to better understand which items have adequate discriminative ability for which constructs.

This study is well-powered and has a number of strengths. First, it is a large multicenter study including 5 different Norwegian University hospitals. Second, we did not use inclusion or exclusion criteria that are likely to impact scoring, eg, pain severity. Third, the use of the recommended NeuPSIG criteria for diagnosis of neuropathic pain should improve the internal validity of our study. We included patients with probable neuropathic pain as true positives in the reference standard. However, we are aware that dichotomization practices varies between studies,^{44,56,59,65,68} and since patients with probable neuropathic pain could potentially increase the rate of false positives in the reference standard, we performed a sensitivity analysis that largely confirmed our findings.

Some limitations should be mentioned. Since an assessment of neuropathic pain must include descriptors of signs and symptoms to properly capture its subjective nature, it is unavoidable that there is some overlap between the tools, the clinical examination in DN4 (and to an extent, self-examination in S-LANSS), and the reference standard. In particular, both the clinical examination and lack of physician blinding may cause some bias and could theoretically contribute specifically to why DN4 performs better. Although the NeuPSIG criteria makes it easier to standardize the diagnostic process in multicenter studies, it is possible that a different reference standard would be a better fit for patients with polyneuropathy.

5. Conclusion

The discriminative ability of DN4 was acceptable, whereas poorer results were observed for painDETECT and S-LANSS. The predictive ability of the 3 tools were poor to mediocre, with DN4 performing best. Applying any of the 3 tools had only a small effect on posttest probability of neuropathic pain. The probability of getting a correct test result was 3 quarters at the very best and only two fifths at worst. Hence, neither tool is appropriate when trying to distinguish between neuropathic and nonneuropathic pain in patients referred to polyneuropathy assessment at neurological outpatient clinics. The internal consistency of painDETECT was good, whereas that of S-LANSS and DN4 did not reach conventional limits for confirmatory studies. Crucially, common polyneuropathy symptoms likely form a baseline score for each clinical tool, reducing their diagnostic accuracy. This study identifies a need for better supportive tools for differentiating between predominantly neuropathic and nonneuropathic pain in patients with symptoms of polyneuropathy.

Conflict of interest statement

The authors have no conflicts of interest to declare.

Appendix A. Supplemental digital content

Supplemental digital content associated with this article can be found online at https://links.lww.com/PAIN/B729.

Supplemental video content

A video abstract associated with this article can be found at https://links.lww.com/PAIN/B730.