Database

After consulting with the director of the Institutional Review Board (IRB) of Taipei Medical University, this retrospective cohort study was exempt from full review by the IRB of Taipei Medical University because the LHID2000 consists of de-identified secondary data released to the public for research purposes.

We retrieved data of sampled subjects for this study from the Taiwan Longitudinal Health Insurance Database (LHID2000). The LHID2000, which was created by the Taiwan National Health Research Institute (NHRI), includes all medical claims and registration files for 1 000 000 enrollees under the Taiwan National Health Insurance (NHI) program. These 1 000 000 enrollees of the LHID2000 were randomly selected from all enrollees listed in the 2000 Registry of Beneficiaries (n = 23.72 million) under the NHI program. The Taiwan NHRI validated the representativeness of the LHID2000 relative to the entire population of NHI enrollees in terms of sex distribution. Some prior studies also demonstrated the high validity of the LHID2000.

Study Sample

This retrospective cohort study included a study cohort and a comparison cohort. We first identified 1083 subjects who had received a first-time diagnosis of neovascular AMD (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes 362.42, 362.43, 362.52, or 362.53) between January 1, 2001 and December 31, 2008. We selected the date of receiving their first-time neovascular AMD diagnosis as the index date. Since incidences of neovascular AMD and Parkinson disease are very low in subjects ≤40 years old, we excluded this age group of subjects (n = 44). In addition, we further excluded all subjects who had ever received a diagnosis of Parkinson disease (ICD-9-CM code 332), dementia (ICD-9-CM code 290, 294.1, or 331), or stroke (ICD-9-CM code 430-438) prior to the index date (n = 162). As a result, 877 subjects with neovascular AMD were included in the study cohort.

For the comparison cohort, we likewise retrieved subjects from the remaining enrollees of the LHID2000. We first identified all subjects ≥40 years who had never received a diagnosis of neovascular AMD since initiation of the NHI program in 1995. We then randomly selected 8770 comparison subjects (10 comparison subjects per study patient) matched to the study subjects in terms of sex, age group (40-44, 45-49, 50-54, 55-59, 60-64, 65-69, 70-74, 75-79, and >79 years), urbanization level, and index year using the SAS program proc surveyselect (SAS System for Windows, ver. 8.2; SAS Institute, Cary, North Carolina, USA). In this study, we chose 10 comparison subjects per study subject on account of the very low prevalence of Parkinson disease. Furthermore, we matched the distribution of urbanization level between comparison subjects and study subjects in order to help assure that they were reasonably similar in terms of unmeasured neighborhood socioeconomic characteristics.

For study subjects, the year of the index date was the year in which they had received their first diagnosis of neovascular AMD, while for comparison subjects, the year of the index date was simply a matched year in which comparison subjects had visited a physician. We assigned the date of their first visit to a physician that occurred during that matched year as the index date for comparison subjects. We also ensured that no selected comparison subjects had ever received a diagnosis of Parkinson disease, dementia, or stroke prior to the index date.

Ultimately, there were 9647 subjects in this study. We then individually tracked each subject for 3 years following the subject’s index date to identify those subjects who subsequently received a diagnosis of Parkinson disease during that 3-year follow-up period (all Parkinson disease cases received a first-time Parkinson disease diagnosis between 2001 and 2011). This study only included Parkinson disease cases if their diagnosis consistently appeared in at least 2 medical claims on different dates in order to increase the diagnostic validity. We did not include Parkinson disease cases in which patients had received a diagnosis of dementia or stroke, or took antipsychotics during the study period, which may lead to Parkinson disease. Furthermore, subjects with Parkinson plus syndrome (ICD-9-CM code 333.0) were not included in our study.

We also censored subjects who died during the 3-year follow-up period; 535 subjects had died, including 53 from the study cohort (6.0% of the study cohort) and 482 from the comparison cohort (5.5% of the comparison cohort) ( P = .357).

Statistical Analysis

We used the SAS statistical package (SAS System for Windows, ver. 8.2, Cary, North Carolina, USA) to conduct all statistical analyses performed in this study. We used the Kaplan-Meier method and log-rank test to examine differences in 3-year Parkinson disease–free survival rates between the study and comparison cohorts (time zero = index date). We also used Cox proportional hazards regression (stratified by sex, age group, urbanization level, and the index year) to calculate the 3-year risk of subsequent Parkinson disease following a diagnosis of neovascular AMD. Before conducting the Cox proportional hazards regression analysis, we tested the proportional hazards assumption and found that this assumption was satisfied since the survival curves for both strata (subjects in the study cohort and comparison cohort) had hazard functions that were proportional over time by observing the plots. In this study, we also adjusted for the selected comorbidities including hypertension, diabetes, hyperlipidemia, and coronary heart disease in the regression model. Hazard ratios (HRs) along with 95% confidence intervals (CIs) were used to present the risk of Parkinson disease, with a 2-sided P value of <.05 considered statistically significant.