Age-related selection bias in Parkinson's disease research: are we recruiting the right participants?

To describe, and explore heterogeneity in, age at onset/diagnosis in Parkinson's disease (PD) and compare mean age at onset/diagnosis in incidence studies with that in general PD research studies. We systematically reviewed studies of PD incidence. We meta-analysed mean age at onset/diagnosis and age-stratum-specific incidence rates. We compared age-specific incidence rates in screening studies in the elderly with whole-population studies. We collated mean ages at onset/diagnosis in clinical studies of PD in five journals July–December 2016. In 17 studies reporting sufficient data to pool, mean age at onset/diagnosis was 69.6 years (95% CI 68.2–71.1), but heterogeneity was high (I2 = 96%). In ten of these studies reporting age at diagnosis specifically, the pooled mean age at diagnosis was slightly higher (71.6 [95% CI 70.6–72.6]) with lower, but still high, heterogeneity (I2 = 84%). In twelve whole-population studies reporting age-specific incidence rates, these peaked in age 70–79 (pooled incidence rate per 100,000 = 93.8 [95% CI 80.3–107.4]). Heterogeneity increased with each increase in age stratum (0% in youngest to 88% in oldest age stratum). Pooled age-specific incidence rates in five population-based screening studies of older age groups were several-fold higher than in whole-population studies. The mean of the reported mean ages at onset/diagnosis in recently published research studies was 60.8 (SD 5.6). The mean age of onset/diagnosis PD is about 70, although this may be an underestimate due to under-diagnosis in the elderly. Many published studies use age-unrepresentative subjects: the effect of this selection bias deserves further study. •The mean age at onset/diagnosis in PD in incidence studies is about 70.•This may be an under-estimate because of under-representation of the elderly.•Patients involved in general research studies in PD have a much lower average age at onset, about 60.•Further studies should investigate the impact of this selection bias.