Meta-analysis and Network Meta-analysis: Do They Simplify Clinical Care?

At the turn of this century, the number of drugs available to treat psychiatric disorders was a handful. It witnessed a paradigm shift from the use of first-generation to second-generation antipsychotics (SGAs). The first quarter of this century has seen the introduction of several new and emerging drugs for the treatment of mental health conditions. The amount of research data on each of these drugs is enormous, often with conflicting results, which can be overwhelming for clinicians in terms of choosing particular treatments and educating well-informed patients. The problem is that while multiple primary studies aim to address related problems, the varying populations and settings as well as conflicting results make it challenging to apply the overall understanding of the findings to clinical practice. The statistical method employed in systematic reviews, known as meta-analyses, offers an answer to this conundrum. To “make clear what is known and not known about the potential benefits and harms of alternative drugs, devices, and other health-care services, and combines the results from many primary studies, usually randomized controlled trials (RCTs),” a systematic review synthesizes pertinent research.[1] Its explicit objective is to provide the answer(s) for a specific research question through an exhaustive and well-planned protocol, identifying similar but separate studies, and filtering and assessing them systematically to come to a conclusion. This makes it possible to assess the overall efficacy of a particular health-care intervention. The advantage of a meta-analysis is that it increases the power of the results by incorporating all of the research that is currently available on a certain subject into a bigger, multiparticipant study. A meta-analysis yields a precise assessment of the intervention’s outcome by combining the findings of several smaller studies into one bigger investigation. By statistically combining study results, a common treatment effect of one intervention compared with another (usually a control intervention) and the magnitude of that treatment effect can be calculated. For example, a question may be asked, “Do second-generation antipsychotics work for the acute management of mania?” A meta-analysis helps answer this question vis-à-vis a comparator treatment and provides the magnitude of the effect. Meta-analytic concepts have been used for about 100 years, with Glass coining the term in 1976.[2] Its explicit aim is to provide a comprehensive qualitative synthesis of the studies in the systematic review. The steps to follow while performing a meta-analysis[3,4] are formulation of the research question, identification of the relevant literature, extraction of relevant study-level data, data quality appraisal, synthesis of summary measure from study-level data, further exploratory analyses, and knowledge synthesis.[5] These steps, according to the critical aspects of any meta-analysis, have remained unchanged from 1976: extensive collection of studies, rigorous analysis of analyses, and integration of evidence. Meta-analyses can remove biases that creep within narrative reviews, answer specific questions by updating the literature systematically, and provide a clear roadmap on how to replicate the review through transparent measures such as preregistration and following reporting guidelines,[6] combine the data from studies and provide a clear picture of the benefits or risks of interventions and are helpful in the evaluation of rare events and secondary outcomes. Its weaknesses lie within its robust methodological process.[6] A meta-analysis can only be as good as the original studies. Biases and dropout rates in controlled trials, publication bias, which meta-analysis may detect but cannot usually overcome, and the inherent retrospective nature of a meta-analysis are further shortcomings. Rare but serious adverse drug reactions may be difficult to answer through meta-analyses. Disparity may come from the differential numbers of studies conducted region wise and may require a region-based sensitivity analysis. Further, meta-analyses may require quality appraisal as poorly done meta-analyses, if included, may further compromise the quality of network meta-analyses (NMAs). While in the above example, a meta-analysis may tell us that a SGA works better than a placebo in treating acute mania, the clinician would always be interested in knowing which drug works best compared to the others and which drug is most tolerable. These important questions drive drug choices and well as adherence and, in turn, could influence treatment outcomes. The use of NMAs, also referred to as multiple treatment meta-analyses or mixed treatment comparisons, can fill this need. Typically, NMA is a single analysis that compares three or more interventions simultaneously. RCTs contrast newly developed medications with a standard treatment or a placebo, but they do not compare the new medication to all other treatments that are offered in the same study. Consequently, it is typically not possible to compare any of the competing interventions head to head. For example, consider two different studies evaluating acute mania-one study comparing olanzapine versus placebo, and a second study with two-arms comparing haloperdol versus placebo. For each of the treatment comparisons, estimated effect sizes are obtained directly. In the absence of a two-arm study of the treatment comparison of olanzapine versus haloperidol, which is of interest to a clinician, a procedure for indirect comparison of effect size is adopted in such a manner that within each trial, the benefit of randomization is well maintained.[7] An indirect comparison can contribute as useful evidence of the difference in outcome measures among competing interventions (all SGA for acute mania) and help select the best available treatment (s) for a health outcome. Moreover, combining the results of direct evidence with the results of indirect estimates may yield a more precise estimate of the intervention or treatment effects.[8] This concept is extended to NMA, which allows multiple pairwise comparisons across a range of interventions and simultaneously combines direct and indirect comparisons. As a result, an NMA offers concise estimates of how relative treatment affects comparisons between several treatments. In addition, conducting an NMA is a cost-effective clinical decision support method when there is insufficient direct evidence to support treatment comparisons, and starting a new RCT research encompassing all competing treatments is not practical. Therefore, methods of NMA are preferred to synthesize the treatment effects if multiple RCTs are well connected. If the multiple RCTs are not well connected, it may potentially lead to unreliable relative treatment effect estimates. The perception of systematic reviews and meta-analyses by stakeholders – including patients, clinicians, and other parties – has evolved with the global shift toward evidence-based medicine (EBM). In clinical practice, meta-analyses and NMAs help collate, evaluate, and understand available evidence.[9] NMAs can assess the relative efficacy of two or more competing interventions, even when not studied head to head, aiding clinical decision-making.[10] This critical appraisal of evidence is viewed as the final step in translating research into practice but also serves as the proximal step in clinical decision-making. These methods also address feasibility and risk–benefit analysis for an index patient in specific geographical locations, which are essential for EBM when planned along those lines. Consequently, they contribute to clinical practice guidelines and policy-making, addressing the health-care needs of larger populations. It is possible that systematic reviews, meta-analyses, and NMA will not be sufficient in the quickly changing field of EBM. The current paradigms of living systematic reviews and umbrella reviews (systematic reviews of systematic reviews) emphasize expanding the scope of these reviews and updating the sources regularly.[11] Meta-analysis and NMA do simplify clinical care. Clinicians must invest some time in understanding them.