MR Imaging of Hypertrophic Olivary Degeneration: Is There a Role for Metaanalysis?

Metaanalysis is a method used for integrating and combining the results of independent studies. Combining data from a variety of scientific studies can increase the power to detect effects, more precisely estimate the impact of these effects, or address a question not posed by the original investigators (1). Traditionally, metaanalysis has been used to analyze multiple, randomized, controlled trials in which results may be inconsistent or inconclusive. Metaanalysis also can be applied to data from observational studies, if they are of sufficient quality.

Goyal et al, in this issue of the AJNR (page 1073), elegantly used metaanalysis on multiple observational studies to determine the sequential MR imaging findings of inferior olivary nucleus hypertrophy and T2-weighted hyperintensity in hypertrophic olivary degeneration (HOD). The authors studied 45 subjects with this rare disorder by carefully combining 39 patients reported in the literature with six patients from their institution. This pooling of data allowed them to estimate more precisely the temporal evolution of MR imaging changes in HOD.

In order to maximize the validity of the results of a metaanalysis, the following criteria should be fulfilled: 1) All relevant scientific manuscripts should be identified in a comprehensive and exhaustive search of multiple sources. 2) The studies included in the summary should be of high scientific quality, the study populations should be similar, and the outcomes should be measured in the same way. 3) Bias in the studies selected for inclusion should be controlled. 4) Analyses should be done to determine the impact of excluding or including certain studies (1).

In metaanalysis, the collection of all available studies is time consuming and difficult because it involves the identification and assembly of published and unpublished literature found through indices, Medline, registries, and files (2). Furthermore, many studies may appear in several different published formats, such as abstracts, theses, or final scientific articles. Because a sound statistical analysis requires that studies be independent, only one report of any study should be included (2). Goyal et al identified 39 patients in 13 published articles. Inclusion criteria included clear temporal documentation between the onset of HOD and the MR imaging findings.

Goyal et al focused their metaanalysis only on the published literature. There is controversy whether unpublished data should be included in a metaanalysis (1). Negative studies are more likely to be unpublished than positive ones, so metaanalysis is prone to publication bias; relying on published studies could overestimate the presence of HOD patients with positive MR imaging findings. Nonetheless, arguments for exclusion of unpublished data, such as lack of a thorough peer review, are equally valid. Alternatively, sensitivity analysis could be used in metaanalysis to determine the impact of any questionable studies on the results. Sensitivity analysis would consist of the recomputation of estimated effects with the study or studies in question removed, and examination of the influence on the results and final conclusion.

When results of individual studies are inconclusive, or when large samples are required to reveal an effect or correlation, metaanalysis can be a useful technique. Goyal et al pooled 58 MR studies in 45 patients to increase their sample size. A similar strategy has been used in other areas of radiology. For example, recent metaanalysis of the effectiveness of breast cancer screening in women 40–49 years old has contributed to changes in clinical practice recommendations for this controversial group (3).

Metaanalysis is prone to the same pitfalls seen in other study designs. The statistical power of a study decreases if the number of subjects is spread over a long period. Goyal et al combined data collected over more than 10 years from 45 subjects, with a mean number of 1.3 MR imaging studies per patient. The end result of the data being spread over this long time span is the relatively few data points per period (ie, per month and year). Therefore, sound understanding of the data variability for a specific interval is limited because of the low number of subjects and MR imaging studies per period. In the future, the robustness of the results discussed by Goyal et al could be enhanced by adding more data points to the metaanalysis as new articles and reports become available.

Is there an important role for metaanalysis in neuroradiology and the neurosciences? Goyal et al have used this method creatively to elucidate the temporal evolution of MR imaging findings in HOD. Metaanalysis could play a pivotal role in determining the effect of new imaging and interventional techniques on patient outcome in common disorders such as acute cerebrovascular accident, epilepsy, and carotid atherosclerotic disease. Small inconclusive studies could be combined in a scientific manner, so that large trials may be avoided if the accumulating evidence from small trials becomes conclusive. Judicious use of metaanalysis may open the window to new insights about the natural history, as well as the proper use of diagnostic tools and interventions, in multiple neurodisorders.