Demographic differences in referral rates to neurologists of patients with suspected Parkinson's disease:...

Although patients with Parkinson's disease (PD) are often treated by neurologists, most are first diagnosed by a primary care physician. Epidemiologic studies which use PD cases from only neurology clinics may be subject to referral bias due to differential case selection in the referral process. To study this methodological issue, a historical cohort study of all outpatients with PD diagnosed by a non-neurologist from January 1, 1988 to June 1, 1992 at the Henry Ford Health System was conducted. This health system has specialty and primary care outpatient clinics located at the main facility in Detroit, Mich. and throughout the metropolitan area. The analysis included those who had at least one outpatient visit during the follow-up period from January 1, 1988 to June 30, 1993 (n = 588), of whom 183 (31.1%) were referred to a neurologist. The univariate analysis resulted in crude risk ratios for referral of 2.14 [95% confidence interval (CI) = 1.56-2.93] for age less than 70, 2.11 (95% CI = 1.57-2.84) for male sex, 1.64 (95% CI = 1.17-2.32) for nonwhite race, 2.48 (95% CI = 1.77-3.47) for private health insurance, 1.79 (95% CI = 1.32-2.43) for being married, 1.44 (95% CI = 1.08-1.92) for 10 or more health care visits per year and 2.27 (95% CI = 1.66-3.11) for having the initial visit for PD at the main Detroit clinic. In a Cox proportional hazards model which included all study variables, race and marital status were no longer statistically significant (p < 0.05), and no risk estimates were greater than 2.0. In summary, patients referred to neurologists in this study were more likely younger, male, to have private health insurance, frequent users of health care and to use hospital-based clinics for their primary health care. We suggest that future case-control studies of PD avoid potential referral bias by not using a study population comprised solely of patients referred to a neurology clinic or design studies which can estimate the potential effect of this bias.