Diabetes Mellitus (DM) is a chronic metabolic disease characterized by high serum glucose levels that lead to multi-organ disorders that affect the eyes, kidneys, nerves, heart, and blood vessels [1]. In particular, amylin, which is a hormone secreted from pancreatic beta-cells, forms a neurotoxic polymer in the brains of patients with DM and is associated with neurodegenerative diseases [2]. Parkinson Disease (PD) is a progressive and incurable neurodegenerative disorder caused by a deficiency in the active neurotransmitter dopamine. It shows the typical symptoms of tremors at rest, stiffness, a mildly bent forward posture, slower conscious movement, and a lack of facial expression known as the masked face, which affects approximately ten million individuals globally [3]. In Korea, the number of PD patients under 65 years old has more than doubled over the past ten years to 4,193 in 2019 [4]. However, the causes of PD are not yet clear; therefore, they are difficult to prevent. In addition, if DM is present when PD is diagnosed, the burden of the disease increases. Evidence from a literature review shows that the symptoms of PD are worse and develop faster in patients with DM [5].

Pagano et al. [6] found that DM is associated with Parkinson's-like pathology and affects the markers of Parkinson's pathology. In individuals with DM but without PD, dopamine transporter binding was significantly reduced in the corpora striata of their brains, which implies associations with PD [6]. Santiago and Potashkin [7] also asserted associations between DM and PD since the two neurodegenerative chronic diseases share "remarkably similar dysregulated pathways". Exposure to environmental risk factors is combined with inherited susceptibility, "mitochondrial dysfunction, endoplasmic reticulum stress, inflammation, and alterations in metabolism may lead to insulin resistance and, ultimately, DM and/or neurodegeneration [7]".

The link between PD and DM has been investigated in several meta-analyses. Cereda et al. [8] conducted a metaanalysis on the risk of PD associated with DM and reported discrepancies between cohort and case-control studies. While cohort studies showed a significant increase in the risk of developing PD in DM, case-control studies showed the opposite results, although they were not statistically significant. Lu et al. [9] also conducted a meta-analysis and reported that the probability of PD in DM patients in case-control studies was significantly reduced by 25%. However, Yue et al. [10] argued in their meta-analysis of cohort studies that the risk of PD in patients with DM was 38% higher than that in patients without DM. As such, the link between DM and PD has been continuously studied but without a clear conclusion. Cohort studies can provide a powerful basis for causality better than case-control studies can, because exposure to risk factors is identified before outcomes [11]. Therefore, in this study, we included only cohort studies to further clarify the temporal sequence between DM and PD. One prior meta-analysis [10] only analyzed cohort studies, but more than half of the included studies assessed exposure or outcome using self-reported data. All but one of the cohort studies included in the previous meta-analysis studies [8, 10] were conducted in the United States or Europe. However, three of the four cohort studies that were published after 2016 and added to the current analysis had been conducted in Asia. In addition, all four studies utilized objective data, such as medical records, when ascertaining exposures and outcomes. Thus, we attempted to estimate the robust pooled effect of DM on the risk of PD by only using cohort studies, geographically expanded samples, and objectively evaluated data. These may also provide better internal and external validities [12].

The purpose of this meta-analysis with large population-based cohort studies is to investigate whether DM at baseline is a risk factor for PD. In addition, this study updates the results of existing meta-analyses [8, 10] by including cohort studies published thereafter. This may help indirectly manage PD, an incurable neurodegenerative disease, by managing relatively well-known DM on the prevention and management of diseases in terms of disease prevention. The results of this study could benefit nurses in clinical and community settings during their daily practice with patients with DM. The results could inform nurse researchers and educators to develop better management strategies for DM patients to prevent and detect PD early.

This meta-analysis was performed to analyze the effect of DM on the risk of PD in ten large cohort studies. The overall effect size of the ten cohort studies was estimated to be 1.30. In other words, DM increased the risk of developing PD by 30%. The overall effect size was similar to those reported in the two previous meta-analyses: 1.37 (95% CI=1.21~1.55) in a study by Cereda et al. [8] with four cohort studies and 1.38 (95% CI=1.18~1.62) in a study by Yue and colleagues [10] with seven cohort studies. The results of our study supported the concluding remarks of the two previous studies [8, 10], which indicated that DM was a risk factor for PD development. In addition, Santiago and Potashkin [30] showed that the two diseases are highly interconnected at the molecular level by a network biological analysis. Previous studies have suggested that insulin resistance causes neurodegeneration [7]. This implies that DM and PD are biologically and functionally related, supporting the findings of previous and current studies. However, inconsistent results have been reported in these studies. As also noted in the Santiago and Potashkin study [30], this is thought to have resulted from sometimes self-reported DM diagnoses and the impact of reactions among drugs to treat DM and PD.

All ten primary studies analyzed the data from credible sources. Among the ten studies, six used national databases: English National Hospital Episode statistics and mortality data in England [19], the Nationwide Social Insurance Institution's drug register database in Finland [21], and National Health Insurance databases in Korea and Taiwan [22, 23, 26, 28]. The remaining studies used data from the American Association of Retired Persons database with over 40 million members in the USA [27], a nutrition cohort with over 140,000 adults followed for 14 years in the USA [24], a cohort of over 20,000 male physicians followed for more than 23 years in the USA [20], and data from the Nurse's Health Study (121,046 women) and the Health Professional Follow-up study (50,833 men) [25]. These data sources had large sample sizes and considerably long monitoring periods, which were highly likely to confer representativeness to the data and validity of the results [31].

The levels of heterogeneity in the current and two previous meta-analyses were inconsistent. Although the heterogeneity was low (I²=34.3%) in the first meta-analysis [8] and relatively high (I²=71.2%) in the second meta-analysis [10], it was considerably high (I²=98.2%) in the current study. One reason for this inconsistency is the large number of primary studies and their sample sizes [15]. Only four and seven cohort studies were included in the first and second meta-analyses, respectively [8, 10]. However, ten were included in our study. The sample size was over eight million in the study by De Pablo-Fernandez et al. [19] and over one million in Sun et al.'s study [26]. Another source of heterogeneity is the differences between the sample characteristics and study design of the primary studies [32]. Including studies conducted in Asia may have contributed to the high heterogeneity. One Korean study [22] and three Taiwanese studies [23, 26, 28] were added to the first meta-analysis. The data sources of all four studies conducted in the USA were self-reports of participants, while the data sources were government-controlled national databases in the remaining studies. Self-reporting might have led to diagnostic and reporting errors, particularly underreporting. As such, subgroups based on geographical regions of studies were found to be significantly heterogeneous. In summary, the high heterogeneity in this study was understandable from a design-related practical standpoint. In addition, from a statistical standpoint, high heterogeneity did not mean that the pooled effect was not significant from a statistical standpoint, since the 95% CI of the pooled RR did not stretch below one and all the studies turned out to be good [33]. We are confident that the positive effect we found was robust.

Several subgroup analyses revealed interesting results. First, the subgroup analysis based on geographical regions revealed the source of heterogeneity in the current study. In the four studies conducted in the USA, the RR was not significant, while the RRs from studies conducted in Asia and Europe showed an association between DM and PD. Among the four studies from the USA, two showed association [20, 27], while the others did not [24, 25]. Data extracted from three [20, 25, 27] out of four studies [20, 24, 25, 27] in the USA were reported incorrectly in the meta-analysis by Cereda et al. [8]. Yue et al. [10] also reported incorrectly extracted data [24, 27]. Second, the results of the subgroup analysis according to sex were similar to those reported in the study by Yue et al. [10]. RRs for men and women were significant, and the RR for women was slightly higher than that for men. Third, a DM duration of less than ten years was found to be a risk factor for PD in the current study and the study by Yue et al. [10]. The reason for a DM duration of less than ten years, rather than one greater than ten years, is a significant risk factor for PD is unclear. The reason for this might be ascertainment bias, as many of the authors of the primary studies indicated. Patients with DM are more prone to other diseases and are, therefore, likely to be in contact with their healthcare providers more often than patients without DM. Hence, increased surveillance might have led to the detection of PD more often in this group. However, the long-term effect of increased surveillance might have decreased because DM and PD are age-related chronic diseases, and diabetes-associated comorbidities exacerbate PD symptoms. Additionally, this may have resulted from the longterm use of anti-diabetic drugs, such as metformin, eliciting neuroprotection [30]. Further prospective investigations are required to clarify the etiological role of DM. Fourth, a subgroup analysis based on comorbid status was performed only in the current study. When patients with DM had comorbid hypertension, stroke, and hyperlipidemia combined, the risk of PD was higher. However, when each comorbid status was considered separately, only stroke was a risk factor for PD. Perhaps this finding, in which stroke acted as a risk factor for PD, resulted from a similar path on which the two are based. If ischemic stroke-induced cerebrovascular changes affect subcortical white matter, basal ganglia, thalamus, and upper brain stem related to PD, it could result in vascular parkinsonism [34]. In summary, the risk of PD in patients with DM was similar in men and women and significantly higher in patients with DM for less than ten years and those with stroke. However, the results need to be interpreted with caution because the number of studies in each subgroup analysis was small, ranging only from two to seven. In addition, the number of cohort studies included in the analysis to estimate the pooled effect was small. The possibility of bias in the publication bias test was less likely, but it could not be ruled out completely. Details of individual patients, such as history of using anti-diabetic drugs and DM severity, were not identified. Therefore, this was not considered in estimating the pooled effect on causality between DM and PD. Subsequently, a systematic literature review is required to cover all related papers for a broad understanding of the effect of DM on the risk of PD. Moreover, an additional meta-analysis is needed considering the details affecting causality between DM and PD to increase clinical applicability in patients. However, the current study has complemented power compared to previous meta-analyses with the advantage of larger populations, longer follow-ups, and a larger number of primary studies. These findings can serve as a basis for nursing practice. Based on these findings, it is necessary to educate patients and their families about the association between DM and PD, especially for patients with DM durations of less than ten years and those who have suffered from strokes, and to develop and apply nursing intervention programs to reduce PD incidence risk. In addition, it is necessary to identify PD outbreaks as early as possible to prevent the worsening of symptoms. Based on the results of this study, the education and application of interventions for DM patients could increase disease insights and self-care activities. This consequently reduces the risk of PD outbreaks, suggesting that DM patients continue to live independently.

In conclusion, although the mechanism of the association between DM and PD has not yet been clearly demonstrated, the association was found globally in ten population-based cohort studies and consistently confirmed with three meta-analyses, including the current one. In addition, further studies should focus on causality or mechanisms that underlie DM and PD to elucidate the clarity of the association.