The importance of pain measurement in routine cancer patient assessment and in research is advocated by experts and scientific associations [1–5], and several efforts are being made to raise consensus on international recommendations in the choice of standardized measurement tools specific for cancer pain evaluation [3, 6–8] in both clinical practice and research.

Subjective pain intensity is the most often considered among the dimensions of pain that should be assessed [1], both in the clinic and in clinical trials. Among several subjective methods for pain intensity measurement, visual analogue scales (VAS), numerical rating scales (NRS), and verbal rating scales (VRS) proved to be reliable and valid, but no one of them consistently showed to be superior to the others [9–19]. The three scales are significantly different as to number of response categories, patient and clinician preference, likelihood of missing data and administration requirements [1]. Research consistently shows that the use of VAS in elderly patients is associated with higher failure of completion rates than the use of NRS, and also that the elderly prefer to use NRS in respect to VAS [12, 20]. Similar difficulties were observed among patients on high doses of opioids [21]. For these reasons VAS can be considered less suitable for pain evaluation in cancer patients, many of which are old and assume opioids. Yet VAS and NRS have shown a better sensitivity to change with respect to VRS [22] probably due to the usually smaller number of categories in VRS.

For these reasons VAS was not considered in our study, which instead focused on VRS and NRS; both scales are easy to use with most patients and have shown good psychometric properties [22] but no studies have been conducted to compare them for the evaluation of pain exacerbation.

In developing a new questionnaire for breakthrough or intense episodic pain (BP-EP) evaluation, both an 11-point NRS and a 6-level VRS were included in the questionnaire with the aim of comparing their performance in evaluating pain exacerbations in terms of reproducibility and of discriminatory capability to distinguish pain exacerbations over a background of less severe pain.

This study, comparing NRS and VRS psychometric properties in the assessment of pain exacerbations, reveals a significantly higher discriminatory capability of NRS in distinguishing between background and peak pain intensities referred to the pain experienced in the previous 24 hours; patients gave inconsistent evaluations in 23 cases with NRS (14%) versus 40 cases with VRS (25%). NRS also showed higher reproducibility when measuring pain exacerbations (Cohen's K of 0.86 for NRS vs. 0.53 for VRS) while the reproducibility of the two scales was similar in evaluating background pain (Cohen's K of 0.80 vs. 0.77).

In agreement with previous studies [11, 13, 26] NRS and VRS showed high positive correlation (Spearman's rho of 0.86 and 0.84 respectively for background and peak pain intensity measurements) although the comparison of the two scales revealed a rather high individual variability mainly for patients scoring "moderate" on the VRS (FIG 1 and FIG 2). This fact suggests that assuming a direct correspondence between VRS and NRS scores (as for example: 0 corresponding to "None", 1-4 to "Mild pain", 5-6 to "Moderate pain", and 7-10 to severe pain1 [14, 27–29]), should be interpreted cautiously in clinical practice due to relevant individual discrepancies.

Moreover the wider range of NRS scores at any value of VRS suggests that patients benefit from the greater sensitivity offered by the higher number of response levels possible with NRS. The possibility to increase the number of verbal descriptors in VRS scales has also limitations. A study by Rosier et al. [30] showed that among 15 adjectives offered to describe their pain, on average patients used only 6 of them, perhaps also because of a difficulty in distinguishing and ordering such high a number of verbal descriptors.

In this experience no data were missing for both scales. This is probably due to the fact that the pain evaluations were not self-completed by patients but administered by a trained nurse or physician, who could properly help patients in understanding questions. Some patients who did not give consent may have had physical or cognitive impairment and this could have contributed to increase the compliance with pain assessment. Although good compliance with the use of NRS is confirmed also in the clinical use, the two scales applicability should be verified in different conditions such as self-administration and repeated use in time.

One limit of the study could be that the two scales have different upper anchor descriptors: "The worst possible pain" for the NRS and "Very severe" for the VRS. The two scales formats have been chosen because they both have undergone specific validation studies in Italian and other languages [23, 31, 32] and fulfill the requirement of a sufficient number of levels to ensure scale sensitivity [22].

The ability of the patient to report his/her pain assessment over the same 24 hours period 3 to 4 hours after the first administration, could be questionable. This choice is aimed to avoid reproducibility overestimation due to memory effect of the first assessment. Furthermore the potential bias introduced by a 3 to 4 hours interval, should have resulted in an underestimation of reproducibility while the indexes obtained (Cohen's K of 0.80 and 0.77 respectively for baseline NRS and VRS) indicate substantial agreement.

In addition, these results should be considered within the limits of the study methods which required the assessment of previous 24 hours pain in a population of advanced cancer patients with no clinically evident cognitive impairment and in relatively good general conditions (38% of patients were out patients and only 14% were admitted to hospice or home care programs).

Previous studies have already compared various scales for pain measurement and gave different results [13–15, 18, 19, 22, 26, 33–35]. Various factors may have influenced the differences in the results of these studies such as patient's populations (chronic or acute pain, different ages, and different levels of cognitive impairment), types of pain (usual background pain, breakthrough pain), different settings of care (clinical or experimental) and administration methods (self-administration or interview). It s also possible that the lack of agreement on the core properties of the measurement scales and on the analysis methods used to evaluate them, lead to apparently different conclusions depending on the different priority given to various scales properties such as easiness of compilation, validity, sensitivity to change and reliability [11, 15, 26], appropriateness of linearity assumption [18] or stability of intra-individual assessment [19].

The data from the literature favoring the use of NRS for pain measurement are based on its intrinsic measurement properties [36], its cross-cultural validity [29, 37], and its good responsivity properties [38]. Moreover, the high variability of VRS formulations both in the number of response categories and in the labels attached to these categories, support the use of NRS which is applied with more standardized formats (usually 11 levels from 0 to 10) across cultures and languages [3, 30, 39]. The 0-10 NRS has greater sensitivity than the VRS and achieves an adequate level of discrimination [22]. The use of VRS is usually supported by its easy of administration, mainly in some patient's populations [1, 16].

Our results suggest that in the measurement of cancer pain exacerbations, patients use NRS more appropriately than VRS and as such NRS should be preferred to VRS in this patient's population.