The current review suggests that the following quantitative sensory measures can predict analgesic effects:
In addition, mechanical allodynia (chronic pain patients) and preoperative hyperalgesia (surgical patients) correlated with the effect and cumulative consumption of analgesics. Conflicting findings on the predictive effect of electrical pain threshold were obtained with three studies in surgical patients producing insignificant results.
4.1 Interpretation and applicability of results
We primarily base our knowledge about human pain on clinical studies and the term QST may give the impression that it is a reliable and easily reproducible test that quantifies sensory function (Siao and Cros, 2003). To the best of our knowledge, this is the first review that combines data across experimental, postoperative and clinical studies in order to assess whether QST can predict analgesic effects. Two previously published systematic reviews focusing on the correlation between preoperative pain sensitivity and postoperative pain intensity, and a review on the clinical relevance of experimental pain assessment suggested that heat stimuli of suprathreshold intensity produced a better predictive ability than heat pain threshold, particularly in women (Edwards et al., 2005; Werner et al., 2010; Abrishami et al., 2011). This was not evident in our review. Although suprathreshold heat pain was found to be a significant predictor of analgesic effect in studies of postoperative pain, heat pain threshold was identified as the principal predictor in studies of experimental and chronic pain. Furthermore, Werner et al. (2010) suggested that electrical pain thresholds have a much greater predictive potential than the mechanical or thermal pain thresholds. Werner et al. (2010) in part based their observation on the study by Nielsen et al. (2007), also included in our review, in which it was demonstrated that preoperative electrical pain threshold significantly correlated with post-caesarean section pain scores at rest and upon movement, but not with postoperative use of supplemental morphine. These apparently conflicting results regarding electrical stimulation may be related to several factors. The advantage of electrical stimulation is that it allows the administration of standardized stimuli of predictable intensity. Timing of the stimuli can also be accurately programmed. Despite these advantages and easy clinical application, electrical stimulation bypasses the peripheral receptors and activates all nerve fibre types (i.e., Aβ-, Aδ- and C-fibres) directly and therefore the method is not a specific activation of the nociceptors (Staahl and Drewes, 2004). Because of the non-specific nature of the stimulus, subjects may find it difficult to reliably assess its intensity (Drewes et al., 2003). Type of surgery, age and psychological distress have, in a recent systematic review, been identified as the three most important predictive factors for postoperative analgesic consumption (Ip et al., 2009). Surgery with low-intensity noxious stimuli combined with an outcome such as consumption of (supplemental) opioid analgesics may thus not generate enough difference between study patients. Hence, the choice of analgesic outcome measure (e.g., cumulative paracetamol vs. morphine consumption) may add to the conflicting results. Unfortunately, heterogeneity in the studies included in our review does not allow for comparisons of predictive values, and age and gender effects across studies.
4.2 Limitations of the original studies
A shortcoming of this review is that the included studies had significant heterogeneity and were generally of poor or moderate quality according to the criteria proposed by Hayden et al. (2006). Accordingly, several factors may limit firm conclusions. There are two obvious explanations why data were inconclusive: (1) the QST measure in focus is not predictive, and an inadequate study design produces false positive results or (2) the QST measure in focus is predictive, and an inadequate study design permits false negative results. In brief, two types of factors can affect predictive performance: factors that influence the measurement of the predictor and those that affect the outcome. The rationale for the use of QST in analgesic treatment prediction is that QST (when tested in a standardized and reproducible fashion) can detect and quantify excitatory or inhibitory alterations of peripheral and central nervous system function (i.e., nociceptive neuroplasticity) (Coderre et al., 1993). The neurophysiological mechanisms that underlie individual pain perceptions are also important in determining the degree of response to analgesic treatment. Nevertheless, QST assessment is still dependent on the participant and not ‘objective’ in the sense of being independent of perception (Backonja et al., 2009). In mechanism-based pain diagnosis and treatment, the assessment tools should be sufficiently sensitive and advanced to deliver valid and credible information (Woolf et al., 1998; Woolf and Max, 2001). The term surrogate pain model has been defined as an indirect measure that predicts the efficacy of a drug in a clinical patient population (Arendt-Nielsen and Hoeck, 2011). Obviously, an acute experimental pain model does not fully simulate the drug action in a complex clinical condition involving multiple mechanisms (Arendt-Nielsen and Yarnitsky, 2009; Staahl et al., 2009a,b; Olesen et al., 2012a). However, the pain biomarker approach has several advantages in human pain research and QST may be used to link symptoms and clinical signs with mechanisms and thus improve treatment strategies for neuropathic pain (Jensen and Baron, 2003; Cruccu et al., 2010). Hence, the individual pattern of sensory symptoms in neuropathic pain most likely closely reflects the underlying pain-generating mechanisms and might also determine the reason for differential and individual treatment responses (Baron et al., 2010). Of note, using drugs themselves as a probe of mechanism have potential advantages over sensory tests. Unfortunately, a recent systematic review demonstrated that intravenous analgesic tests have limited overall clinical utility in selecting patients for long-term treatment with specific oral analgesic agents (Cohen et al., 2009). On the other hand, there are probably more ways to stimulate or block sites of pain processing with selective drugs than with bedside examining tools (Woolf and Max, 2001) and better studies with more appropriate designs are highly warranted. Importantly, the reliability of the individual stimulation measures and/or paradigms was not evaluated in any of the included studies. In addition, QST and outcome measurements were evaluated at very different time-points, particularly in surgical populations, limiting firm conclusions concerning optimal timing of the QST assessments. However, the included stimulation parameters are all widely accepted methods in experimental pain assessment and most studies managed to adequately measure the predictor of interest sufficiently to limit potential bias. Additionally, the methodology, reliability, reproducibility, limitations and potential clinical applications of QST have previously been addressed in a number of reviews (Yarnitsky, 1997; Shy et al., 2003; Siao and Cros, 2003; Chong and Cros, 2004; Hansson et al., 2007; Arendt-Nielsen and Yarnitsky, 2009; Backonja et al., 2009; Oono et al., 2011a,b; Olesen et al., 2012b; Nahman-Averbuch et al., 2013). The German Research Network on Neuropathic Pain has also published extensive validation data for all somatosensory submodalities (Rolke et al., 2006a,b). There are several other important considerations to make when interpreting the type of information obtained from QST, including patient characteristics, instructions, examiner, the tissue and organs tested and the type and method of application of the stimulus (Backonja et al., 2009; Staahl et al., 2009a).
Regarding the validity of the dependent variable, response to analgesic treatment was inconsistently defined and most frequently reported as a secondary outcome. Analgesic consumption is influenced by both pharmacokinetics and health convictions, whereas pain intensity and relief are subjective measures. There is large between-patient variability in analgesic consumption and this measure may become problematic when used as a surrogate marker for pain in populations or groups, particularly in studies with small sample sizes (Moore et al., 2011). The validity of analgesic consumption as an outcome measure remains unclear; however, several factors have been proposed as potential confounding factors in analgesic consumption (McQuay et al., 2008). A basic assumption must be that patients will consume analgesic until pain intensity is reduced and they reach a state of comfort. Obviously, this entails similar pain scores across patients. Thus, titration must have failed if pain scores are different. Indeed, we can only test the null hypothesis (i.e., no difference in analgesic consumption between groups) when pain scores and consumption are similar. Furthermore, analgesic consumption may also be affected by the pharmacology of the drug(s) in focus. Opioid-induced side effects may thus interfere with analgesic consumption. For example, sedation could lead to reduced consumption, and subsequently be mistaken for analgesia. Similarly, if increasing intake of an opioid analgesic corresponds to nausea and vomiting, patients will likely limit their analgesic intake accordingly. Additionally, fear of addiction and acceptance of pain may also reduce and thereby affect analgesic consumption. Unfortunately, none of the included studies were sufficiently powered to detect differences in adverse events. Studies using consumption of non-opioid analgesics (e.g., paracetamol and nonsteroidal anti-inflammatory drugs) as an outcome measure may additionally be hampered by a dose-limiting factor and an analgesic ceiling effect associated with these drugs. Consequently, patients experiencing moderate to severe pain may be unable to reduce their pain scores as much as preferred, whereas patients with no or mild pain would experience superior pain relief receiving the same amount of drug.
As a result of the abovementioned shortcomings, a number of alternative clinical endpoints have been suggested including simple multiplication of raw pain scores and hourly analgesic consumption (Lehmann et al., 1986), dichotomous classification of analgesic consumption at 750 μg fentanyl (equivalent to 75 mg morphine) over 48 h (Moore et al., 2011), and integrated assessment of pain scores and analgesic consumption based on a percentage difference from the mean rank of the specific variable (Silverman et al., 1993). Notably, none of these measures have gained widespread acceptance and therefore their validity remains unknown. In addition, the use of a combined endpoint may cause problems if the predictor effect is in opposite directions for different outcomes included in the composite endpoint (Bouwmeester et al., 2012).
Measurement of pain intensity may be reliable for multiple measurements within a single individual. However, there is also high variability in pain, particularly among chronic pain patients, and adding to the complexity, pain scales are also typically influenced by clinical co-morbidities (Price et al., 2008; Farrar, 2010). Furthermore, the improvement in pain within an individual patient, expressed as a percent change, is highly correlated with his or her overall condition (Farrar, 2010). Likewise, there is no linear relationship between nociception and clinical pain and particularly chronic pain can be influenced by psychosocial factors, such as catastrophizing, depression and anxiety in up to 50% of patients (Vase et al., 2011; Mao, 2012). Interestingly, the study by Eisenberg et al. (2010) in healthy volunteers was the only study that used changes in QST measures during a cold pressor test (120 s hand immersion in 1 °C ice water). Unfortunately, almost half of their study population withdrew their hand within 15 s of hand immersion precluding firm conclusions on the prediction of the magnitude of opioid analgesia. Finally, analgesic treatment responses may be driven by variables not analysed in the included studies. Thus, recent research indicates that individual responses to analgesic treatment are partly genetically controlled (Zwisler et al., 2010).
4.3 Limitations of the review
Despite widespread and increasing use of QST in pain research, only a few studies of the predictive effects on analgesic treatment response were identified. As discussed earlier, the risk of bias was high or unclear for many of the included studies, leading to uncertainty regarding the internal validity of the findings. A variety of predictors and outcomes was used at different points in time across studies making it difficult to arrive at any firm conclusion about the best strategy to use QST in predicting analgesic response. In many cases, the reporting of study details was inadequate and therefore, the size of the predictive effect remains uncertain. The focus on a limited number of QST parameters in the current review provides no information on other tests that may be used in clinical or experimental practice. Likewise, we have focused on selected outcomes ignoring other potentially relevant endpoints such as physical and emotional functioning, sleep, safety, tolerability and patient adherence to treatment (Turk et al., 2003; Dworkin et al., 2009).
4.4 Implications for current practice and future research
Experimental models using acute stimuli may activate the nervous system in different ways compared with tonic pain generated from ongoing inflammation (Olesen et al., 2012a). Other symptoms also often influence patients' pain perception confounding a clear evaluation (Olesen et al., 2012a). Likewise, drug response can be modulated by a number of non-genetic factors, especially by co-medications and the presence of concurrent diseases (Stamer and Stuber, 2007). Existing analgesic treatments are inefficient or poorly tolerated in approximately half of patients with chronic pain and, in those who do respond, it is relatively rare for pain to be completely relieved or even reduced to mild severity (Dworkin et al., 2009). In addition, treatment of chronic pain is challenging because of much more influential co-morbidities, as discussed earlier. The goal remains to identify a true clinical profile and to use such a profile to formulate individualized medicine (Mao, 2012).
The current findings, particularly the most recent, are indeed interesting and suggest that QST may have some predictive value regarding analgesic treatment response. However, our review does not yet suggest that it is reasonable that clinicians use any specific QST measure to (1) discriminate between patients who are likely to respond to analgesic treatment or (2) to identify patients at higher risk for severe postoperative pain and high analgesic requirement outside the context of experimental or clinical studies. On the other hand, this does not imply that clinicians should discontinue clinically driven sensory testing, as individual differences in pain sensitivity that can be assessed by QST may still be relevant for treatment decisions. As previously suggested, we need to recognize that clinical pain cannot be addressed by only understanding limited dimensions of the pain process (Mao, 2012). Studies including deeper, tonic pain stimulations as well as methods evoking hyperalgesia are therefore warranted in future prediction studies (Olesen et al., 2012a). In our opinion, experimental pain models should aim for translation into clinical practice. Importantly, the use of QST to follow patients over time requires that the test is reproducible in test-retest conditions. It is important to keep in mind that the sensitivity of a given experimental model for detecting analgesia is also affected by the method used to measure specific pain conditions (Staahl et al., 2009a). In addition to clinical outcomes, such as pain intensity and consumption of analgesics, a composite score that takes both these measures into account and perhaps also includes other relevant outcomes may also be beneficial to examine the effectiveness of any given QST parameter. However, large prospective studies are needed to establish the clinical value of such approaches.