Variability in experimental pain studies: nuisance or opportunity?

Editor—Pain is a variable experience, even in studies that use controlled nociceptive stimuli in healthy humans.¹ This variability is unsurprising considering that nociception (the neural process of encoding noxious stimuli) and pain (a conscious experience) do not share an isomorphic relationship.² Pain is influenced by a broad range of biological, cognitive, contextual, and mood-related factors that may vary from moment to moment.³ Even a well-controlled experimental laboratory environment presents limited scope to control all these factors, and it is rare for pain to be elicited with high consistency.

The intraindividual variability in pain reporting may reflect important personal features that are relevant to our understanding of pain and the impact of analgesic strategies. With this in mind, several clinical studies have used high frequency pain assessment in longitudinal designs to examine the possibility that intraindividual variability may be relevant to clinical outcomes. They report that intraindividual variability in pain ratings may be related to depression,^4,5 self-efficacy,⁴ emotional and physical functioning,⁶ and may predict benefits from sham medication or active treatments.7, 8, 9

Nevertheless, experimental pain research in humans has largely neglected to acknowledge explicitly the importance of intraindividual variability in pain reporting. To our knowledge, only a handful of studies have attended to intraindividual variability.^1,10, 11, 12, 13, 14 Instead, the common practice is to analyse averages gained from repeated measurements and thus smooth out variability. Raw data are seldom presented. Thus, intraindividual variability is considered a nuisance, rather than a feature worthy of attention.

Obscuring variability has practical disadvantages. Many experimental studies rely on calibrating stimulus intensities to each individual participant and then assuming that subsequent stimuli are experienced at an intensity that reflects the data from that initial calibration. Intraindividual variability in trial-by-trial pain reports undermines this assumption, and points to contemporary experimental designs that allow for drift in stimulus-response relationships and detect shifts over time in, for example, the effect of one stimulus-response on a subsequent stimulus-response. Accounting for variability in research design or statistical analyses will allow greater confidence in interpreting the effect of analgesic interventions on stimulus-response relationships.

Obscuring variability has external validity disadvantages. The common approach to experimental pain studies is sequential averaging, where the averaged intraindividual measurements are then averaged across individuals to realise a group average. It reduces variance but makes two flawed assumptions: (1) that the appropriate measure of central location (e.g. arithmetic mean, median) was used to capture the centre of location for the intraindividual variation and (2) that only interindividual variation is important. The cost of sequential averaging is that both intraindividual variation and interindividual variation are lost (Fig. 1). This removes potentially fertile data as to the effect of any analgesic strategy on variability itself – a treatment that decreases variability of a stimulus-response without reducing the average response within individuals will not be differentiated from a treatment that has no effect on either. It may be that decreased variability, or removal of peaks in pain, is a clinically worthwhile outcome. Moreover, sequential averaging has less relevance to clinical practice. Patients with pain typically seek relief of their own pain, not of the pain of a group and there may well be no individuals who have this ‘average response’.

Fig 1.

Selected data (from Madden and colleagues¹⁵) used to illustrate the consequences of averaging. (a) Rating given by six participants at each stimulus intensity. (b) Average rating from the same six participants at each stimulus intensity. (c) Average rating across all six participants at each stimulus intensity. A sequential loss of information is observed across the three panels as variability is obscured by averaging. Moreover, that the two stimuli marked with arrows (a) are associated with identical ratings illustrates the risk that, if only one reading were taken per stimulus intensity, an erroneous result is highly possible.

Obscuring variability has important implications for power calculation simulations that are used when designing complex studies. This has ethical and resource implications because concealing the extent of variance in reporting will lead to underpowered studies. Such studies are known to be associated with spurious findings as to the direction and magnitude of an intervention's effect.^16,17

We suggest three remedies for experimental studies:

• (1)Report variability data at the individual level, whether in the main body of a manuscript or in supplementary data. Ideally, raw data should be made available, and the data displayed using appropriate graphical methods.
• (2)Use statistical techniques that allow for individual-level variability in pain reporting. Modern analytical techniques allow for hierarchical nesting of repeated measures within individuals and of individuals within groups. Such an approach would seem warranted when analysing experimental pain reports. Considering that experimental studies on pain lay the groundwork for clinical studies that inform treatments, acknowledging individual fluctuations in pain report in the analysis of experimental data stands to improve translation between experimental and clinical research studies.
• (3)Ask participants to report the percept on every stimulation trial, so as to verify the stimulus–response relationship on each trial and verify or modify calibrations, or evaluate drift and fluctuation in stimulus–response over repeated trials.

Experimental examination of intraindividual variability in pain responses should improve clinical phenotyping. There have been calls to consider matching treatment to patients defined by their responses to psychophysical phenotyping procedures.18, 19, 20, 21 In the context of the growing clinical application of these phenotyping procedures to inform treatment decisions, clarifying the true variability in individual responses is of some urgency. For example, it is unclear how the three-replicate method used for some modalities in the DFNS Quantitative Sensory Testing paradigm²² was derived, and whether it is sufficient. Further, the substantial overlap of ratings that can exist across a wide range of stimulus intensities (as visible in Fig. 1a) indicates that, in some cases, three replicates may be too few for confidence.

The relevance to clinically meaningful outcomes of intraindividual variability when reporting pain could be clarified with appropriate designs. For example, it would be interesting to understand whether intraindividual variability differs systematically between people who recover from a painful episode and those who do not. That is, does variability have utility as a risk factor for chronicity? We recently speculated that the variability with which an individual rates different, intensity-matched stimulation trials could reflect the flexibility of his/her perceptual processing¹⁵ and could represent a beneficial feature of a sensory processing system. Indeed, preliminary data infer the possibility that pain thresholds are more variable in people who do not have persistent pain than in people who do.²³ The idea that variability could be beneficial also loosely reflects the recent recognition that cognitive flexibility may be beneficial with regard to persistent pain.²⁴ The variability with which an individual reports on painful experimental stimuli could represent a useful biomarker of risk or vulnerability, or actual or predicted response to treatment. We speculate that clinical variables relevant to perioperative care (e.g. anxiety) may also influence variability in clinical pain report, in which case addressing those influential variables could yield beneficial outcomes. Experimental designs which study variability in pain reports and how that variability relates to other, clinically relevant individual features, may be the method of choice for these and other lines of inquiry.

Fully acknowledging and reporting intraindividual variability in reports of experimentally induced pain, and handling such variability appropriately, stand to move the field closer to understanding pain. Individual variability in stimulus-response may reflect important phenotypic features to inform clinical subgrouping and predict treatment outcomes. If the role of experimental work in humans is to clarify the contributions of various variables to pain outcomes, acknowledging intraindividual variability will be key to optimising the translation of experimental research to the clinical context.

Declarations of interest

VJM receives speaker's fees for lectures on pain and rehabilitation. PK is on retainer for Partners in Research, and receives speaker's fees for lectures and professional development courses on pain. GLM has received support from Pfizer, Kaiser Permanente, Workers’ Compensation Boards in Australia, Europe, and North America, the International Olympic Committee, the Port Adelaide Football Club, and the Arsenal Football Club. He receives royalties for books on pain and speaker's fees for talks and professional development courses on pain and rehabilitation. The authors declare no other competing interests related to this work.