Reply to: Diabetes and risk of Parkinson's disease

Abstract

We thank Cereda et al. for including our study1 in their updated meta-analysis of diabetes and Parkinson disease (PD) published in Movement Disorders. There were only 4 prospective studies available at the time of publication of the original meta-analysis and inclusion of our investigation substantially increased the number of PD cases from 3284 in the original meta-analysis to 3940 in the updated version. Overall, this addition changed the pooled relative risk estimate for diabetes from 1.37 (95% confidence interval [CI], 1.21–1.55) to 1.25 (95% CI, 1.03–1.55), but with significant heterogeneity between studies. A key question is: what are the sources of this heterogeneity? An important difference among the studies included in this meta analysis are the methods of PD ascertainment. Those studies that relied on register data (Hu et al.)2 or unconfirmed self-report (Driver et al.),3 may have included as incident PD cases individuals with vascular Parkinsonism. The potential bias from this source was reduced in the study by Xu et al.,4 who were able to obtain confirmation from the treating neurologist of about two-thirds of the incident PD cases, and minimized in the studies by Simon et al.,5 and Palacios et al.,1 which included only neurologist-confirmed incident PD cases. Vascular parkinsonism may contribute to explain the fact that the strongest association between diabetes and PD was found in Finland (Hu et al.),2 which has relatively high rates of cardiovascular disease. Another important difference is residual confounding. For example, adjustment for changes in smoking behavior was only implemented in Simon et al.,5 and Palacios et al.;1 in the remaining studies, a spurious positive association between diabetes and PD could have resulted if individuals with diabetes were more likely to quit smoking (a negative risk factor for PD)6–8 during the follow-up than those without diabetes. Finally, losses to follow-up may contribute to bias—because only about 63% of the participants in the cohort studied by Xu et al. responded to the PD survey, a higher response rate among diabetics could induce a spurious positive association between diabetes and PD. As Cereda et al. acknowledge, it is not possible to take these differences into account in the meta-analysis. The results from the meta-analysis should therefore be considered with caution. A close analysis of the results of the studies that report a positive association between diabetes and PD raises further doubt on whether such association truly reflects an increased PD risk among diabetics The 2 U.S. studies that reported a positive association are in conflict as to the timing of diabetes prior to onset of PD. Xu et al. found an overall increased risk of PD associated with diabetes within the National Institutes of Health and AARP (formerly the American Association of Retired Persons) (NIH-AARP) cohort, but most of the risk was attributed to patients with 10 or more years of diabetes at baseline. At the same time, in the other positive study, by Driver et al., the increase in PD risk was restricted to patients with <10 years of diabetes at baseline, and a 7-fold increase in risk was reported among individuals with a recent diagnosis of diabetes. The authors of the latter study acknowledged that their results were most likely due to ascertainment bias or reverse causation. Thus, pooling the relative risk estimates from these 2 studies does not have a biological rationale. Overall, in spite of the interesting work by Cereda et al, we believe that there is little evidence that PD risk is increased among individuals with diabetes in the United States. The possibility of an association in Finland remains open. Obesity has also been found to be associated with an increased PD risk in Finland,9 but not in the United States,1,10 and it remains possible that genetic or lifestyle differences between the Finnish and the U.S. population may explain this apparent discrepancy.