Prevalence and Incidence of Huntington's Disease

We read with interest the updated review of the epidemiology of Huntington's disease (HD) by Medina et al.¹ In their article, the authors present results from a series of meta-analyses of prevalence and incidence studies conducted in populations in Africa, Asia, Europe, and the Americas between 2011 and 2022. Worldwide pooled estimates are reported for prevalence and incidence, along with separate pooled incidence estimates for each continent where there was more than a single study. In each case, estimates were derived from a random-effects meta-analysis.

As is common in systematic reviews of prevalence and incidence, the included studies are heterogeneous in terms of their methodology, data source, and population. For example, whereas the majority of studies in the review reported prevalence and incidence for all ages, Gavrielov-Yusim et al² provided results only for those ≥18 years and Evans et al³ only for those ≥21 years. These two studies derived their estimates from administrative and research databases, whereas Kounidas et al⁴ used genetic laboratory, clinic, and hospital records. These differences in population and data source matter; the epidemiology of HD in children and adolescents is not the same as in adults, and different data sources are derived from populations with different disease risk.

Significant heterogeneity in a meta-analysis results in pooled estimates that are difficult to interpret, and this is very much the case here. The pooled prevalence and incidence estimates reported by Medina et al¹ do not in any meaningful sense represent the prevalence or incidence in a defined population. However, this is exactly how the pooled estimates reported in a previous meta-analysis study⁵ have been used.^6-9

This misinterpretation is made even more likely due to an error in the reporting of the key measures of study heterogeneity, Q (which follows a χ² distribution and therefore allows us to test the significance of the heterogeneity) and I². The values of Q and I² reported in tables 1 and 2 of Medina et al¹ suggest that heterogeneity is very small or absent. However, this is not actually the case. Unfortunately, the software that the authors used, Comprehensive Meta-Analysis Software, rather confusingly reports a “Q* statistic” (along with an I ² value calculated from this value of Q*), which should be used “only for the analysis of variance, to partition Q* into its various components,” and it is these values that appear in the article. The software authors note that these statistics are not measures of heterogeneity and state that “[r]ather, the Q statistic computed using fixed-effect weights [our emphasis] is the one that reflects the between-studies dispersion.”¹⁰

We have calculated the correct values of Q and I², and in contrast with the values reported in the article, the results suggest a very high degree of heterogeneity (Table 1).

The high degree of heterogeneity can also be seen clearly in forest plots generated from the data presented in the article. See Figure 1 for an example (European prevalence studies).

In conclusion, we caution against interpreting the pooled estimates of prevalence and incidence reported in Medina et al¹ as meaningful for any population. We would also encourage authors of meta-analysis studies to publish forest plots, either in the body of the paper or as a supplementary file, so that readers can visually assess the degree of heterogeneity in the study estimates.