Answers are grounded in this paper and cite their sources.
Studies on postoperative pain inherently involve statistics and include some outcome domains. In this issue of the Journal, Andersen et al1 showed how variations in the choice of statistical methods and outcome variables for studies assessing the treatment effect on pain management can lead to different results and conclusions. In their article, they reanalyzed their data from their previous published study, wherein they compared transversus abdominis plane block with either ropivacaine or saline after laparoscopic colon resection.2 Pain intensity measurements were measured at 0, 2, 4, 6, 12, 18, and 24 hours postoperatively. They compared the patients’ visual analog scale (VAS) pain scores, morphine consumption, and side effects using standard statistical tests such as the Mann-Whitney U and Fisher exact tests and used P < .05 as significance criterion. They found no difference between groups in the pain scores while coughing at 6 hours (P = .20), their primary outcome, or in the area under the curve (AUC) 2 to 24 hours (P = .32). There was a significant reduction in the 24-hour morphine consumption in the local anesthetic group (P = .008).2 In their present article, the investigators reanalyzed their VAS data and patient controlled analgesia-opioid consumption and calculated the same summary measures of pain intensity ratings (AUC-VAS, mean VAS across the time points), but for each 0 to 6 and 0 to 24 hours1 instead of 2 to 24 hours as in their original work. In addition, they introduced a novel version of an integrated composite end point that includes longitudinally measured pain intensity and opioid consumption (PIOC). Authors reported (in Table 2) statistical improvements in active versus placebo both at rest and during coughing in mean VAS, opioid consumption, and PIOC, at each of 0 to 6 and 0 to 24 hours, and for AUC-VAS only at 0 to 6 hours (not 0–24). The authors discussed the merits and limitations of simple measurement of PIOC. Their use of a composite end point, integrating PIOC, is welcome. Such composite end points, combining with other outcome measures, need to be employed and refined in future studies. Andersen et al1 note that conclusions can change depending on the choice of statistical methods and outcome variables for pain studies. This is certainly true, but the particular details of a study often help in narrowing the choice of methods. Using AUC-VAS implies that researchers are comfortable interpolating pain values between measurement points, which is most reliable when the time points are not too far apart. Also important is planning a similar temporal pattern of measurements for patients in each group and across groups, as was the case here. However, if patients did not have similar follow-up lengths, time-weighted average VAS (AUC-VAS divided by total measurement time) is an attractive option. Repeated-measures analysis of variance may be the best choice if the group-by-time interaction is of interest; the method also allows appropriate modeling of the correlation between measurements. Morphine consumption is often log-normal in shape (skewed to the right), and so can be analyzed on the log scale and the treatment effect given as the ratio of geometric means. The main contribution of Andersen et al1 is the application of the integrated pain/opioid consumption score of Dai et al3 to longitudinal data. The integrated score is intuitive because pain and opioid consumption are vital elements of pain management, and an integrated score cleanly allows a single treatment effect and conclusion. However, the integrated score may be best used in conjunction with other methods. For example, and importantly, the integrated score does not distinguish the direction or magnitude of either effect, eg, a significant overall effect might occur if pain is increased while opioid consumption is reduced or vice versa. Andersen et al1 state that the integrated score above zero indicates increased pain and opioid consumption, but this could easily not be the case—a positive score could result if one outcome was increased and the other decreased. Therefore, before comparing groups on the integrated score, it would be important to ensure noninferiority on each outcome using a joint hypothesis-testing framework.4 Such a framework efficiently controls type I error when assessing effects on multiple components and assures that, when a global test is applied across components, the inference is clear. For some areas of research, it may be acceptable if effects on certain prespecified components of a composite are negative as long as the majority of components have a positive effect. This is not the case with pain and opioids; care must be taken to assure that neither is “worse” with the intervention than control and the relevant testing planned a priori. Otherwise, interpreting claims of a positive effect for the integrated score or other global test will be difficult. This issue is even more important as additional components such as functional status are added to the assessment of treatment effects on pain management, as we suggest below. Another aspect of the article by Andersen et al1 is their choice of outcome measures that included both pain and adverse events. Since pain is multidimensional, studies on pain should assess several outcome domains. The Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials (IMMPACT) group on the research design for single-dose analgesic clinical trials in acute pain recommended pain, adverse events, and global assessment of the responders5 and additional domains such as physical function, ability to ambulate without assistance, and other “quality of recovery” (QoR) measurement tools for longer duration studies. For pediatric acute pain trials, the PedIMMPACT group recommended pain intensity, global judgment of improvement and of satisfaction with treatment, adverse events and symptoms, physical recovery, emotional response, and economic factors.6 Currently, published studies on postoperative pain usually assess only pain and side effects; physical and emotional functioning are rarely noted. There are several sophisticated outcome measurement tools7–12 that assess several outcome domains (Table). All instruments have very good psychometric properties, and the difference is in the domain that they address. Although these instruments are multidimensional and have been employed in postoperative pain studies, they were not designed or have been validated for use in the immediate postoperative period.Table.: Outcome Domains Covered by the Sophisticated Measurement ToolsThere are instruments validated for use in the perioperative period. Myles et al13 developed a series of QoR scores for use in postoperative pain studies (Table). The longer form, QoR-40, exhibits excellent psychometric properties, high patient recruitment, and excellent completion and return rate.13 The shorter form, QoR-15, was developed to shorten the administration burden.14 A comparison of the QoR scores showed the minimal change in score that would indicate a meaningful change in a patient’s health status, or minimal clinically important difference, to be 8.0 for the QoR-15 and 6.3 for the QoR-40.15 Another measurement tool is the International Pain Outcomes Questionnaire (IPO), which was based on the American Pain Society Outcome Questionnaire-Revised (APS-POQ-R).16 In summary, the article by Andersen et al1 highlights the need to improve and agree on a common methodology for assessing treatment effects on pain outcomes. Integrating several measures into a composite end point is attractive and promising but needs to be done carefully. Their use of pain and adverse events as the only outcome measures needs to be addressed. Pain is multidimensional; physical and emotional functioning and patient satisfaction should also be investigated. Utilization of the QoR40 or the IPO, especially the QoR instrument as it has been extensively used, should include most of the outcome domains recommended by the IMMPACT group. The addition of patient satisfaction to the QoR completely addresses the multidimensional nature of acute pain. DISCLOSURES Name: Honorio T. Benzon, MD. Contribution: This author helped write the manuscript. Name: Edward J. Mascha, PhD. Contribution: This author helped write the manuscript. Name: Christopher L. Wu, MD. Contribution: This author helped write the manuscript. This manuscript was handled by: Richard Brull, MD, FRCPC.
No takes yet. Share an insight, caveat, or question.
Benzon et al. (2017) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: