Discussion
Taken together, the findings of our nationwide case-control study of persons with diabetes do not support the suggestion of lower risk of PD among statin users based on earlier studies on the general population [5,6] or persons with diabetes [8-10]. On the contrary, higher risk of PD in those with highest cumulative statin exposure was observed, regardless of whether nonusers or those with lowest cumulative exposure were used as the reference group. No association was observed when medium or low cumulative exposure tertile was compared to statin no-use, nor when medium tertile was compared to low cumulative exposure tertile. Considering the high DDDs of highest cumulative statin exposure tertiles (mean 2500+ DDDs) in both analyses, statin exposure within these groups can be considered long-term.
Our findings are not in line with previous studies reporting lower risk of PD among statin users with diabetes [8-10]. However, methodological differences may partly explain the differences. In our case-control study, we matched cases and controls according to diabetes duration, whereas only one of the earlier cohort studies propensity score-matched for diabetes duration [10]. In addition, all persons in our study had used diabetes medications during the exposure assessment time, and those 44 (<0.5% of the study population) who had not yet initiated their diabetes medications during the exposure assessment time did so during the lag time [18]. In contrast, two of the prior studies apparently also included those with who managed with lifestyle modifications [8,10]. Furthermore, one study was restricted to metformin users, leaving out persons with diabetes who were treated with other diabetes medications [9]. Further, to account for reverse causality, i.e., impact of prodromal PD symptoms on contact with prescribers increasing the likelihood of changes in drug exposure among cases, we did not consider exposure during the three-year lag before the outcome, while all prior studies considered all statin exposure until the diagnosis of PD. In addition, the exposure levels in our study differ from the earlier studies: The highest cumulative statin exposure tertiles started from 616 [8] and 675 [10] DDDs, which would correspond to less than two years of statin treatment (atorvastatin 20 mg) [17] which is less than the length of lag time used in our study. It should also be noted that two of these earlier studies were based on the same data source, National Health Insurance reimbursement database of Taiwan, although with different population sample, follow-up period and definition of statin use [8,10].
Of the two previous nested case-control studies that were not restricted to persons with diabetes, one found no association between users and nonusers [19] whereas the other reported that statin use of 12 months or more was associated with increased risk of PD compared to statin use of less than six months [20]. These nested case–control studies differed from ours as they included people without diabetes, did not assess cumulative statin exposure, nor did they utilize lag time. Interestingly, a case-control study by Liu et al. reported a higher risk of PD among users of lipophilic statins compared to nonusers, although the study was limited to persons aged less than 65 years which may limit the generalizability of results [21].
We applied a three-year lag period to decrease the effect of possible protopathic bias [22]. The lag duration was based on an earlier FINPARK study that showed an increase in muscle relaxant use already three years before PD diagnosis indicating prodromal motor symptoms [23]. The lag time might be important also due to the observation that cholesterol levels have been reported to begin declining already 4 years before diagnosis [24]. On the other hand, reverse causality may partially explain the inverse association in earlier studies with shorter follow-up time, because initiation of statin therapy or increasing intensity may be less likely during the prodromal phase of PD if there is a decline in cholesterol levels [24].
Prescription register accurately represents statin use in Finnish population as this register includes all reimbursed medication purchases to which all Finnish citizens are eligible. However, data on drugs used in hospitals or public nursing homes was not available. Registry-based study approach effectively controls for selection bias and recall bias. A common limitation of register-based studies is the accuracy of PD diagnosis. In our study, we applied data from multiple sources to ascertain PD cases. Due to reimbursement criteria and additional exclusions performed by us described earlier [16], it is likely that these persons had clinically verified PD. In addition, the proportion of excluded cases in the FINPARK study (25.9%) is in line with the estimated proportion of false diagnoses [15,25,26]. Weaknesses of our study include the possibility of residual confounding explaining our findings.
Duration of diabetes was controlled by matching, but severity of diabetes was unknown. However, if diabetes severity is an intermediate variable between statin use and PD, adjusting for it would not be feasible as it would introduce overadjustment bias [27]. We could not adjust for low-density lipoprotein cholesterol levels as they are not recorded in the registers. However, as higher low-density lipoprotein cholesterol levels have been suggested to be a significant confounder for the association between statin use and lower risk of PD in earlier studies [7], and we observed a slightly higher risk of PD among those with higher statin exposure and no association in the use-nonuse analyses, it seems unlikely that adjustment for low-density lipoprotein cholesterol levels would have led towards an inverse among statin users.
To decrease the influence of confounding by indication we restricted our study to persons with diabetes, a population at increased risk of PD [13,14]. Due to the resulting limitation in sample size, we were unable to assess whether the association of increased risk of PD with high cumulative statin exposure was driven by specific statins, statin therapy intensity or the length of exposure, or any combination thereof. In addition, our study included persons who had used multiple statins and therefore we did not perform more detailed statin -specific analyses. It has been suggested that the association between statins and PD is different for lipophilic and hydrophilic statins [21]. However, all statins regardless of their lipophilicity can cross the blood-brain barrier, although lipophilic statin may impair brain cholesterol synthesis slightly more [28]. In our study the number of hydrophilic statin users was small due to major overlap with lipophilic statin use. Therefore, we did not perform dose-response analyses based on different categories. It has been suggested that initiation of statin therapy may rapidly “unmask” PD in those with preclinical symptoms [11,29]. We did not explore the unmasking theory as any association near PD diagnosis would be indistinguishable from protopathic bias; Incidence of preclinical PD symptoms may increase healthcare contacts which can increase the likelihood of initiation of statin therapy. However, since the prevalence of statin users only during the lag time was indifferent between PD cases and controls, our results do not support this theory. Our nationwide study that controlled for diabetes duration and reverse causality does not provide support for the hypothesis that statin use decreases the risk of PD.