No measure is perfect. No one measure assesses all pain domains, nor is any single measure useful in all settings and with all populations. Moreover, because of the imperfection of available instruments, it is theoretically possible to modify any existing measure to improve it further or to develop new and better measures to replace existing ones. As a result, new pain assessment procedures and measures are constantly being developed and published. Thus, the clinician or researcher seeking to find the best measure for his or her needs should not only be aware of the existing pain assessment literature but should also know how to evaluate new measures as they are published.¹ The following section seeks to facilitate this task by briefly summarizing the three key issues that should be considered when evaluating any pain measure: validity, reliability, and utility.

Patients often have more than one pain problem. For example, the majority of individuals with spinal cord injury have chronic pain, and the majority of these report pain at more than one site.³ Clinicians and those researchers who do not limit their sample to the (few) patients with only one pain problem are faced with the difficult task of determining the number of pain problems to assess in any one patient or study participant. If only one “primary” pain problem is assessed at a clinic visit, but on the next visit, a different “primary” pain problem emerges as the most distressing, then it would be very difficult to track the effects of pain treatment from one clinic visit to the next.

Similarly, researchers who limit the number of pain problems assessed to just one primary problem run the risk of underestimating the magnitude of pain and its impact in their research findings. On the other hand, it is not practical to assess every pain problem in every patient seen in the clinic or in every participant of a research study. These considerations suggest that, in many situations, patients should have the opportunity to report on more than one pain problem but not necessarily be required or expected to report on every pain problem that they have at every assessment point.

But how many pain problems should be assessed? Two? Five? More? One approach to deal with this issue is to begin by assessing pain “in general”; for example, asking patients to consider all of their pain problems together when rating the overall average magnitude or intensity of their pain and the impact of pain on their lives. This is a practical solution, especially for assessing pain interference, because it may be very difficult for patients to identify the unique contribution of each different pain problem to interference with different activities. Moreover, assessing global pain intensity and interference allows the clinician or researcher to have a single primary measure of these two key pain domains, making analyses and tracking over time easier.

However, limiting assessment to only pain “in general” may oversimplify assessment and also interfere with determining the true effects of pain treatment. For example, if a pain treatment reduces the pain associated with one pain problem (e.g., headache) but not another (e.g., low back pain or a neuropathic pain condition), the specific effect of the treatment on headache pain might be less noticeable or even lost altogether if a measure of “general” pain intensity is used. So, in many situations, allowing for the assessment of more than one pain problem would be useful.

Unfortunately, however, there is not yet a clear consensus in the field concerning the best number of pain problems to assess. In the clinical setting, it probably makes sense to assess as many of the pain problems that are of concern to the patient. If the patient experiences eight unique pain problems and views each as a significant problem that contributes to dysfunction, then perhaps each of these should be assessed, at least at the initial evaluation, and then tracked at subsequent clinic visits as appropriate.

When determining how many pain problems to assess in a research study, the number of problems that should be assessed would vary as a function of the research question(s) being asked and the specific population being studied. One reasonable option would be to select the number of pain problems to assess that would capture the majority of patients in the population. For example, in persons with spinal cord injury, it has been recommended that investigators should consider assessing basic information (such as pain location and intensity) for up to three presenting pain problems.^4,5 In this instance, three was chosen as a way to balance the need for a thorough assessment against the need to minimize assessment burden, keeping in mind that the majority of persons with spinal cord injury and pain report three or fewer pain problems.³ Although it is unlikely that a single upper limit of pain problems can be identified that should be assessed in every research project and with every patient population, each investigator would do well to consider this issue when developing assessment protocols.

Clinicians and researchers have long recognized that pain is a multidimensional experience that includes a number of measurable qualities such as intensity, affect (global bothersomeness of the pain experience as well as the impact of pain on emotional functioning), sensory quality, spatial quality (location), temporal quality, and impact on or interference with daily activities.^6,7 Although the focus of pain assessment in many clinical and research settings has often been, and continues to be, on pain intensity,⁸ there is now a well-established recognition and interest in the assessment of pain’s other domains.⁹

It is important that clinicians and investigators consider assessing more than just pain intensity for a number of important reasons. First, limiting assessment to only pain intensity leaves clinicians and researchers in the difficult position of having limited information about the presenting pain problem(s). In a clinical situation, changes in a pain domain not assessed might end up being critical for understanding the effects of a pain treatment (e.g., if pain qualities are not assessed, and a treatment reduces the “aching” and “deep” qualities of a pain problem, but perhaps not average pain intensity overall, or if a treatment produces a decrease in the impact of pain on sleep or other areas of functioning, even when there has been a minimal impact on pain intensity). For this reason, clinicians and researchers interested in assessing pain should at least consider all of the pain domains when determining which ones to assess and perhaps only avoid those domains they are certain will not be important to treatment (for clinicians) or understanding (for researchers).

One of the factors that might determine the selection of fewer domains and measures (e.g., perhaps choosing to assess just pain intensity and pain quality) over a more comprehensive assessment (e.g., including measures of pain site, pain interference, and the temporal qualities of pain as well as perhaps more general measures of psychological and physical functioning) is whether the pain problem being assessed is more acute or chronic. Acute pain, which may be defined as pain resulting from current or very recent damage to tissue, includes pain from medical procedures (e.g., injections, lumbar punctures, surgery) as well as both major and minor physical injuries. Because acute pain problems tend to resolve quickly in most individuals, their impact tends to be transitory. In this situation, and if the focus of treatment is on just one or two pain domains (e.g., pain intensity, mood), then it may be appropriate to assess only one or two pain domains.

Chronic pain, on the other hand, tends to be more complex than acute pain. It also tends to influence a large variety of quality of life domains (e.g., employment status, sleep quality, psychological status, social functioning). Patients’ responses to chronic pain can also be quite variable, as can its impact. For chronic pain, then, in both clinical and research settings, and in order to ensure a thorough understanding of the pain problem, more pain domains, and perhaps more measures that assess these domains, are often required.

Often, the clinician or researcher wishes to have a measure of a patient’s usual pain during a specific period of time. A single measure or rating of current pain is not likely going to be an adequate index of usual pain, given that pain can vary from one moment to another. So what is the best way to assess usual pain? The three most common methods to assess pain intensity in clinical trials are (1) ask respondents to provide multiple ratings of current pain on pain diaries during the epoch of interest (e.g., four times per day for 7 days) and then compute the arithmetic mean of the ratings obtained, (2) assess pain once per day on several days, asking respondents to provide a recall rating of their average pain in the previous 24 hours and then computing the arithmetic mean of these 24-hour recalled average pain ratings to create a composite score representing usual pain, or (3) assess pain just once but ask respondents to provide recall ratings of their average pain over the entire epoch of interest. Each approach has strengths and weaknesses. In addition, pain assessment experts have not yet reached a clear consensus on which option should be recommended.

Support for the first approach, using multiple ratings of current pain from daily diary data (most often obtained electronically by asking patients to access a Web site to provide ratings or via automated telephone calls to patients) comes from studies that have demonstrated (1) a biasing impact of recent pain and worst pain (also known as “end” and “peak” effects) on recall ratings^10,11,12 and (2) access to computer-based and electronic diary technology which facilitates data collection.^13,14 When diaries are required, electronic, Web-based, or phone diaries are preferred over paper-and-pencil ones, given the common finding that respondents use paper-and-pencil diaries inappropriately.^15,16 Because of the strengths of diary approaches, there has been an increase over time in the use of this approach to collect data in clinical trials.^17,18,19,20

However, there are also good reasons which make some investigators hesitant to embrace an approach that requires multiple daily diary entries to compute pain intensity scores, especially for use in clinical trials. First, as a practical issue, daily diary data can be expensive to collect. The financial cost of the hardware and software management associated with data collection via electronic diaries may be beyond the means of some investigators. Related to this, there is also a cost in terms of patient assessment burden. Some procedures ask respondents to provide only one assessment per day,^15,21,22,23 but it has been more common to ask patients to provide three²³ or even more (4 to 6 times^17,18,24) ratings per day. This requires a significant effort on the part of patients, which may lower compliance with the task. To the extent that less costly recall ratings (that require just a few or even just one assessment) may be adequately valid, investigators may save substantial resources and significantly decrease the patient assessment burden if recall ratings are used instead of diary ratings.

A second problem with diary data requiring many ratings is that using this approach will result in missing data. The reported percentages of missing data points from electronic diary studies range from 6%¹⁶ to 25%.²⁵ The reported rates of study participants who provide incomplete data (i.e., at least some missing data during the study period) can range from 17%¹⁵ to 46%.²¹ The primary reason reported for missing electronic data is that the patient did not hear the alarm or cue asking for the assessment.²⁶ Other reasons given include the alarm going
off at an inconvenient time, the participant being too busy to respond, technical difficulties with a computer, emotional reasons, and pain being too severe at the time of assessment.²⁶ When data are missing, investigators need to either remove subjects from the analyses (which limits the generalizability of the findings and runs the risk of resulting in findings that overstate the impact of treatment) or use some approach to impute the missing data (i.e., estimate what the missing ratings might have been, had all subjects provided complete data). A variety of data imputation procedures can be used for clinical trial data, such as “last observation carried forward” which involves taking the most recent rating obtained and replacing all missing values with that rating.¹⁷ A more conservative approach is to replace missing values with pretreatment ratings. However, if the treatment being examined is effective, this approach can underestimate the treatment effect size. Regardless of the approach used, however, data imputation adds error; imputed data are estimates only. In fact, the error added by the need to impute missing scores could potentially be greater than that associated with recall bias, so the use of diary data over recall ratings could potentially result in a more costly and effort-intensive assessment procedure that is ultimately also less accurate.

A third issue is that the use of electronic diaries limits the subjects who can participate in a study. For example, in one electronic diary study that approached 52 possible participants, 6 refused participation outright, 1 did not have the motor ability to hold the computer stylus, and 5 had visual problems that interfered with their ability to read the computer display.²¹ By limiting the participants in clinical trials to those who are able and willing to use electronic diaries, their use in these trials limits the generalizability of the study findings.

Perhaps, the strongest argument against the strategy of computing average pain from multiple ratings of current pain over time is that research increasingly supports the conclusion that recall ratings of pain intensity are adequately valid for most research purposes. Although research does indicate that there can be both peak and end effects that bias recall ratings, these effects tend to be small.^10,12 Moreover, research indicates that the correlations between recalled average pain (in the previous 7 days) and actual average pain during that same period (as assessed by diaries) are strong (correlation coefficients range from 0.68 to 0.99^{27,28,29,30,31,32,33})—well within a range that indicate they carry valid variance as measures of average or usual pain. In short, recall ratings reflect actual average pain and are therefore valid indicants of that pain. In addition, and perhaps most critically, the research finding that provides the most support for the validity of recalled pain ratings as outcome measures in pain clinical trials is indisputable: Recall ratings are responsive to the effects of pain treatments known to impact pain. Hundreds, if not thousands, of clinical trials have shown that effective treatments for chronic pain result in reductions in recall ratings of average pain. Thus, there is adequate evidence to support the use of recall ratings as valid measures of pain intensity in clinical and research settings.

That said, two important questions remain unresolved. First, if a researcher chooses to obtain multiple ratings of recalled pain over a relative small recall period (e.g., daily ratings of average pain in the past 24 hours) and compute the arithmetic mean of these ratings to create a composite score representing usual pain, how many of these recall ratings are needed to create an adequately valid measure of a patient’s average pain? Second, might a single rating of recalled pain over a relatively larger recall period (e.g., a single rating of recalled average pain in the past week) be as adequately valid as a composite score made up of multiple 24-hour recall ratings?

Unfortunately, to date, relatively little research has been performed to address these two questions, despite the significant implications of the answers for designing and conducting pain clinical trials. Only three studies were identified which address the first question. In the first of these, investigators compared the reliability and sensitivity (i.e., ability to detect significant treatment effects) of pain intensity scores made up of one to nine 24-hour recall ratings of pain intensity using data from a clinical trial evaluating the efficacy of oxymorphone extended-release for the treatment of low back pain.³⁴ As would be expected based on psychometric theory, the reliability of the outcome measure reflecting average pain intensity increased as the number of ratings used to compute that measure increased from one to nine. However, and unexpectedly, this increase in reliability was not associated with improvements in the ability of the composite score to detect treatment effects. In fact, the first single 24-hour recall rating was about as sensitive for detecting treatments effects as the composite scores made up of 2 to 9 ratings.³⁴ In a commentary on this finding, other researchers noted that these results may have been due to the unusually strong test-retest reliability of the 24-hour recall ratings in the study sample.³⁵ To address this question further, these other investigators performed a similar set of analyses using data from a clinical trial evaluating the efficacy of a behavioral intervention for pain associated with osteoarthritis.³⁵ They found that the individual 24-hour recall ratings of average pain intensity evidenced a great deal of variability in their ability to detect treatment effects, with effect sizes ranging from medium (d = .34, P = NS) to large (d = .88, P < .001). Thus, had only a single 24-hour recall rating been used, the actual treatment effect might have been under- or overestimated and deemed either significant or nonsignificant, depending on the single rating used. On the other hand, the effect sizes for composite scores—which were .51, .55, and .66 for composites made up of two, three, and seven ratings—were more stable. Also, all of the composite scores were statistically significant (Ps = .0005 to .007).³⁵ The third study that addressed this question was a secondary analysis of a pilot study (N = 10 individuals with spinal cord injury and chronic pain) which examined the ability of pain intensity scores to detect changes in pain intensity from before to after 12 sessions of neurofeedback treatment.³⁶ In this study, the single-item ratings (of current pain and 24-hour recalled least, average, and worst pain) also evidenced a great deal of variability in effect sizes (ranging from 0.08 to 1.03). They also became more stable when two ratings were averaged (range, 0.16 to 0.38) and were not that much more stable than this when four ratings were averaged (range, 0.17 to 0.42). These investigators concluded that just two ratings (e.g., two ratings of 24-hour recalled average pain) may be adequate for assessing pain in outcome trials in clinical trials. However, they also noted that it is possible that in some populations—especially populations where pain can vary markedly from 1 day to the next, such as in individuals with chronic headache—more than two ratings might be needed to ensure adequate reliability and validity.

With respect to the second question, regarding the potential of a 7-day recall rating for providing valid ratings of average pain during the previous 7 days, although there are findings showing a tendency for people to overestimate recalled pain over the past week,^25,37 no study was identified which directly compared the validity of a single 7-day pain recall rating with composite scores made up of multiple diary ratings. Thus, this critical question remains unanswered at this point in time.

Given the current state of knowledge, what can or should researchers do with respect to assessing average or usual pain intensity? The answer probably depends on the resources available to the researcher. Certainly, there is evidence that a single 7-day recall rating of pain intensity can be sensitive to changes in pain with treatment (e.g., Hans et al.,³⁸ Branco et al.³⁹). Therefore, 7-day recall ratings can be considered when
resources are limited. However, given that psychometric theory predicts greater reliability (and therefore, ultimately, validity) when more measures are combined into composite scores, and until research is performed to compare single 7-day recall ratings with various combinations of 24-hour recall ratings, it would probably be wise for investigators to consider obtaining more than one rating (e.g., three or four 24-hour recall ratings obtained within a specified number of days) and combining these recall ratings into a single composite score.

The single pain domain assessed most often in clinical and research settings is pain intensity, or the magnitude of felt pain.⁴⁰ The three most commonly used scales to assess pain intensity are (1) the VAS, (2) the Numerical Rating Scale (NRS), and (3) the Verbal Rating Scale (VRS) (Fig. 20.1). The results from research across many different pain populations yield fairly consistent findings concerning the psychometric properties of these measures^6,9,41 and may be summarized as follows:

The affective quality of pain includes both the general unpleasantness and/or bothersomeness of the pain sensation as well as the many varieties of affect (fear, anger, sadness, frustration, feelings of hopelessness) that pain can produce—especially as it becomes chronic. The most common measures of general, global pain unpleasantness are single-item rating scales (VASs, NRSs, and VRSs) that use endpoints that reflect extreme levels of unpleasantness (e.g., for a 0-to-10 NRS or 100 mm VAS, “not bad at all” for the 0 rating or 0-mm mark and “the most unpleasant feeling possible for me” for the 10 rating or 100-mm mark).⁴⁵ In general, these measures have proven useful in highly controlled laboratory studies that seek to differentiate intensity from affective components of pain.^45,46

On the other hand, outside of the laboratory setting, patients appear to treat single-item VAS, NRS, and VRS measures of pain unpleasantness much like measures of pain intensity so that the two are often indistinguishable from one another in clinical populations.^47,48 Moreover, one might question the content validity of single-item measures of affect, given the complex and multidimensional nature of emotional experience.

Pain affect can also be assessed using multiple-item scales, the most common of which are the Affective subscale of the McGill Pain Questionnaire (MPQ)⁴⁹ and its associated short form, the Short-Form McGill Pain Questionnaire (SF-MPQ)⁵⁰ (Fig. 20.2). The original MPQ contains 78 descriptors that are
categorized into 20 subgroups, 5 of which assess the impact of pain on affect. The five affective domains are tension (assessed using “tiring” and “exhausting” descriptors), autonomic (assessed using “sickening” and “suffocating” descriptors), fear (assessed using “fearful,” “frightful,” and “terrifying” descriptors), punishment (assessed using “punishing,” “grueling,” “cruel,” “vicious,” and “killing” descriptors), and affective miscellaneous (assessed using “wretched” and “blinding” descriptors).⁴⁹ Within each domain, when administered the MPQ, respondents are asked to circle or mark the single descriptor within each group that most accurately reflects or describes their pain. Descriptors are then ranked according to their position in the word set. The Pain Rating Index (PRI), which can be computed for each of the four primary MPQ subscales,
including the Affective subscale, is the sum of the rank values of these descriptors.

The short form of the MPQ (SF-MPQ) contains 15 descriptors, four of which come from the MPQ Affective subscale (“tiring-exhausting,” “sickening,” “fearful,” and “punishing-cruel”).⁵⁰ However, unlike the MPQ, which requires respondents to select a single descriptor from each category list that best describes their pain, respondents to the SF-MPQ are allowed to rate the severity of each item individually on a 4-point Likert scale (0 = None to 3 = Severe). A severity or intensity score can then be calculated for the Affective subscale (as well as for Sensory and Total scale scores; see “Measuring Pain Quality” section). Research has shown that the correlations between the corresponding scales on the MPQ and SF-MPQ are high (rs range, 0.68 to 0.92).^50,51,52

There is a substantial amount of data supporting the validity of the MPQ and SF-MPQ Affective subscales. First, like the other MPQ and SF-MPQ scales, the Affective subscale has been shown to be responsive to pain treatment.^53,54,55,56 Additional support for the validity of the MPQ Affective subscale as a measure of the affective component of pain, specifically, was reported by Ahles and colleagues,⁵⁷ who found that this scale was more strongly associated with measures of psychological distress than with measures of pain intensity. Also, Kremer and colleagues⁵⁸ reported that patients with cancer report a greater affective component of their pain on the MPQ Affective subscale than patients with low back pain, consistent with the hypothesis that cancer pain may be associated with higher levels of affect (e.g., be more worrisome and cause more fear) than low back pain.

The experience of pain consists of much more than its magnitude or intensity and affective components. Pain is also often described using a number of different qualities, such as “burning,” “aching,” and “tender,” among many others. Although historically, clinicians and researchers have focused on pain intensity as the single most important pain domain to assess,⁴⁰ there has been an upsurge of interest in the assessment of pain qualities. The two primary purposes of such measures are (1) to help diagnose the pain problem and (2) to more thoroughly describe the pain experience and determine the effects of pain treatments on that experience.