A Novel Approach to Synthesize the Evidence on Analgesic Adjuvants for Postoperative Pain
We should expect the results of RCTs to vary, even if they investigate the same intervention in a similar population. Each RCT recruits patients by a convenience sample from a local subpopulation; it is not a random sample of the entire population who might receive the adjuvant. Furthermore, by pure chance alone, each RCT has a chance to overestimate or to underestimate the effect of an intervention. Also, larger and smaller studies will lead to more or less precise estimates of the intervention effect. Meta-analysis pools available RCTs to synthesize the evidence for a more precise and robust effect estimate. This may reduce uncertainty in the face of seemingly contradictory results. However, if the results and studies are too heterogeneous, evidence synthesis may be inappropriate. Such an approach is frequently critiqued as mixing apples and oranges. Excessive between-study heterogeneity in meta-analysis raises concerns that the included studies are clinically and methodologically too different, making pooling all identified RCTs unreasonable. In the face of substantial between-study differences in results, we should explore its possible causes.2
One approach to explain why RCTs yield contradictory results is to group studies by surgical intervention. Stratification by surgical intervention makes clinical sense because between-study heterogeneity may be smaller within each stratum. This led to the current paradigm of procedure-specific pain control3: it is expected that different surgical procedures cause different amounts of pain. Populations undergoing different interventions for different diseases may vary in how they respond to pain, in their comorbidities, pharmacokinetics, etc. Clearly, a population of elderly men after thoracotomy for lung cancer will differ from a population of young women after cesarean delivery. It follows that postoperative pain control should be tailored to the specific surgical intervention and the particular population. While this seems intuitive, it drastically reduces the number of available studies for clinical decision-making for a particular patient population undergoing a particular procedure. For example on thyroidectomy, Doleman et al1 found only 1 single RCT investigating the effect of intravenous acetaminophen. Pooling only studies on a particular procedure limits the number of studies available and hence poses a significant challenge to any evidence-based approach to procedure-specific postoperative pain management.
A second approach to explain differences in results between RCTs is to control for baseline risk.4 For studies investigating adjuvants for improved postoperative pain control, investigators frequently use the mean difference in morphine dose between those receiving and those not receiving the adjuvant to estimate the effect.