Patients
Patients with neuropathic pain and headache were recruited from the Department of Neurology, University of Würzburg and the Department of Pain Management (Bergmannsheil), Ruhr University, Bochum. Patients with osteoarthritis related pain were recruited from the Department of Orthopedics (König-Ludwig-Haus), University of Würzburg and the Department of Orthopedics (St. Josefs Klinikum), Ruhr University, Bochum. A diagnosis of neuropathic pain was confirmed when associated with a definite neurological cause as determined during routine in- or outpatient management of the patient. In addition, it was required that the pattern of pain distribution was characteristic of the respective diagnosis. Headache was diagnosed using the International Headache Society criteria for tension type headache and migraine [11]. Only patients with a headache present at the time of visit were included. Headache was chosen in order to have a pain condition with a very different distribution from that in most patients with neuropathic pain. Osteoarthritis was diagnosed based on criteria provided by the American College of Rheumatology (http://www.rheumatology.org). Osteoarthritis was chosen to include a condition that (like in most patients with neuropathic pain in our sample) entails pain in the extremities. Exclusion criteria were coexistence of several conditions possibly causing pain, severe depression, chronic alcoholism or substance abuse, and an inability to understand the questionnaire. Patients with neuropathic pain received current symptomatic treatment, including analgesic drugs, when admitted or seen as outpatients.
NPSI Translation into German and adaptation of the NPSI for patients with headache and osteoarthritis
Using the original French NPSI, we used a standard forward-back translation to develop the German equivalent. First, two translators who were native French speakers and fluent in German (one with, one without medical knowledge) translated the questionnaire into German. A reconciled language version was developed using these two forward translations. A third professional translator (a native German speaker fluent in French) produced the back-translation. The final version was evaluated by three independent medical professionals fluent in both French and German usage. Discrepancies in wording were analyzed and when required, resolutions were obtained by consensus. A diagram depicting pain attacks and episodes of continuous pain was inserted to facilitate patient comprehension. For quality control, patients were instructed to indicate areas of pain on an additional diagram (see additional file 2 for a full version of the NPSI-G). Patients with headache or osteoarthritis related pain were additionally required to complete a short questionnaire that addressed aspects of their headache (questions were designed to differentiate between tension headaches and migraines) or, their joint pain (location of pain sites including potentially related sites).
Study design
The study was approved by the Ethics Committees of the Universities of Bochum and Würzburg. After providing informed consent, the patients were required to complete four questionnaires: the NPSI-G, the German version of the NPS (NPS-D), the graded chronic pain scale (GCPS, [12]), and the German version of the Center for Epidemiologic Studies Depression Scale (CES-D) [13]. Patients diagnosed with neuropathic pain were instructed to complete a 2nd NPSI-G as well as the patient global impression of change (PGIC) scale [14] 24 hours later. The entire set of questionnaires was repeated at 4 weeks. Since most of these patients were inpatients, their 24 hour questionnaire was readily distributed and collected. Outpatients with neuropathic pain were given both the 24 hour and 4 week questionnaires with dates for completion clearly marked and prepaid postal envelopes addressed to the investigators. Patients who did not return the questionnaires at the specified time intervals received telephone reminders. During the 4-week period, patients underwent medical treatment as required. Patients with headache (mostly outpatients) or osteoarthritis pain (mostly inpatients) received all of the above questionnaires only once and were instructed to complete their questionnaires while at the hospital.
Depressive symptoms were assessed using the CES-D questionnaire. Scores above 23 points were considered indicative of depressive states. Pain intensities were graded using the 11-point numerical scales (0 - 10) of the GCPS.
Sample description and comparison of the diagnostic groups
Sample description included diagnostic data, age, gender, depression score (CES-D) and general pain characteristics (GCPS scores) for each diagnostic group.
Construction of a score for assessing neuropathic pain
A sum score (NPSI-G score) for assessing the severity of neuropathic pain was calculated using the sum score methodology of our reference study [8]. A possible positive contribution of the categorical items Q4 (frequency of spontaneous pain) and Q7 (number of pain attacks) to the score was examined by transforming them into dichotomous variables and post hoc adding weighted variable scores to the sum score. Then we looked for improvement by comparing the validity measures of the original and the expanded score. Optimized transformation rules and weight-coefficients were generated from an algorithm which tested possible improvements of the correlations in the validity calculations.
Reliability and validity of the NPSI-D
For the group of patients with neuropathic pain, we assessed the test-retest reliability of the NPSI-G and the NPSIG score and analyzed its convergent and divergent validities. The test-retest reliability of the interval-scaled items and the NPSI-G score was assessed by calculating intra-class correlations (ICC, two way random model) for all items included in the initial measurements and the 24-hour follow-up. Since there are no validated instruments available in German comparable to the NPSI, we used two widely employed comparators for convergent validity; the German version of the NPS [7] (NPS-D) and those items from the GCPS that examine intensity of pain (current, average and maximal pain assessments over the 4 week interval). Divergent validity was measured by calculating the Pearson correlations with the CES-D scores. We then compared the reliability of the NPSI-G items and the NPSI-G score with the French reference study.
Sensitivity to change
The Pearson correlations between the 4-week change of the NPSI-G score and the changes of current, average, and maximum pain in the GCPS, the PGIC-score, and the change of the CES-D-score were calculated in patients with neuropathic pain, whereby the CES-D was used as a control measure for divergent characteristics. Correlations were considered low, medium or high as they reached levels of r = 0.1, r = 0.3, or r = 0.5 respectively [15]. To conclude that there was sensitivity to change, we would expect at least medium correlations with the changes of pain intensity in the CGPS (r ≥ 0.3). As the PGIC score is a gross estimate of changes in pain intensity, it was only considered as an additional control instrument. A low to medium correlation (r ≥ 0.2) was judged sufficient to support the hypothesis that the NPSI-G score is sensitive to change. Low correlations with the ASD-score (r < 0.3) would support divergent validity.
Evaluation of the discriminant properties of the NPSI-G and the NPSI-G score
Responses to the NPSI-G items were compared between the diagnostic groups by analyzing the mean values and the frequencies of positive item responses (values > 0), indicating the occurrence of the different qualities of pain for the 10 interval scaled items Q1 - Q3, Q5 - Q6, Q8 - Q12. Mean values of items and sum score were calculated and compared between diagnostic groups. The distribution of the NPSI-G score within the groups was analyzed using box plots. The correlation of the NPSI-G score with age, sex, pain intensity in the CGPS, and the CES-D score was calculated by Pearson correlations to control for a possible bias caused by between-group differences of these parameters. We then examined the diagnostic power of the NPSI-G score using a receiver operating characteristic (ROC) diagram [16]. The ROC diagram shows the possible combinations of sensitivity and specificity that can be achieved for a given score. The non-neuropathic group was formed by combining the osteoarthritis and the headache patients.
Development and analysis of a diagnostic tool based on the NPSI-G
To develop a tool that would separate patients with neuropathic pain from the other groups, we calculated a discriminant score (NPSI-G-dis) using discriminant analysis. This weighted sum score of the variables was constructed using specific coefficients for each variable to allow optimal separation of the different diagnostic groups. We used all variables from the NPSI-G including the two categorical variables Q4 and Q7, which were transformed into dichotomous variables similar to the construction described in the definition of the NPSI-G score, but without weight coefficients which were dispensable in this context. In this case an optimal ROC-diagram was the criterion for an appropriate transformation.
Analysis of different pain profiles in patients with neuropathic pain
Using a cluster analysis (hierarchical Ward analysis [17] with follow up k-means analysis) based on the ten interval items of the NPSI-G, we looked for subgroups with different pain profiles within the neuropathic pain group. Profiles were compared by multivariate variance analysis (MANOVA; Wilks-Lambda used as test statistic) and differences between the item scores within each cluster by variance analysis for dependent measures. An analysis of current and maximal pain levels from the GCPS was done to evaluate whether clusters were based on pain severity only. The distribution of pain intensities was examined by box plots; mean values for different clusters were compared by t-tests.
Test statistics and measurements of coherence
In addition to the already described methods, we used the following statistical tests: Frequencies of item responses were compared using Chi-square tests. Comparison of group means (of the different diagnostic groups or of groups defined by the results of the cluster analysis) was done by t-tests in the case of two groups and by ANOVAs with post hoc tests (Sidak resp. Dunnett T-3 adjustments for multiple testing) when more than two independent groups were involved [18, 19]. To analyze coherence of item responses, we calculated Pearson correlations. Compensation for multiple comparisons was done by Bonferroni adjustments [20].