We ascertained stroke events that occurred in the population during 3 year-long study periods: between July 1, 1993, and June 30, 1994, and in calendar years 1999 and 2005. Details of our case ascertainment and how methods have been held constant among the 3 periods have been previously published.4
The GCNK region includes 2 Ohio counties and 3 contiguous Northern Kentucky counties that border the Ohio River with 19 hospitals active in 1993/1994, 18 in 1999, and 17 in 2005. Only residents of the 5 study counties are considered for case ascertainment. This study was approved by the Institutional Review Board at all participating hospitals for each study period.
Study nurses screened the medical records of all inpatients and emergency department visits with primary or secondary stroke-related ICD-9 discharge diagnoses (430–436) from all acute care hospitals in the study region. Additional strokes were ascertained by reviewing 1) cases for which stroke was listed as the primary or secondary cause of death by 1 of the 5 county coroners' offices; 2) all stroke-related visits to local public health clinics and hospital-based outpatient clinics; and 3) records of potential stroke cases in a random sample of primary care physicians' offices and nursing homes in the GCNK region. Details of the exact number of sites by study period may be found in table e-1 on the Neurology
® Web site at www.neurology.org
. Sampling was necessary given the large number of physician offices and nursing homes in the region. Sites were selected randomly, for each study period, by the study statistician from a list generated from a combination of the local yellow pages and the American Medical Association listing of physicians. All events were cross-checked within and between sources to prevent double counting.
To qualify as an incident case, a patient must have met the clinical criteria for one of the stroke categories adapted from the Classification for Cerebrovascular Diseases III11
and used in our previous work1–4
: cerebral ischemia, intracerebral hemorrhage (ICH), subarachnoid hemorrhage (SAH), or stroke of uncertain cause. Imaging results were not considered in this clinical definition except for the presence of hemorrhage on CT or MRI for hemorrhagic events. Only first-ever events were included for this analysis. TIA, defined as symptoms lasting <24 hours regardless of imaging results, were not included. The onset of stroke symptoms must have occurred within the study time periods; discharge lists were screened for 60 days beyond the end of each study period to capture patients with a stroke in the period but not discharged until later.
Study nurses abstracted information on all potential cases regarding stroke symptoms, physical examination findings, past medical/surgical history (including stroke risk factors), prestroke medication use, social history/habits, prehospital evaluation, vital signs and emergency room evaluation, neurologic evaluation, diagnostic test results (testing, EKG/cardiac testing, and all neuroimaging), treatments, outcome, type of insurance, and current address. Classification of race/ethnicity was as self-reported in the medical administrative record; ethnicity was not uniformly collected. The study population of the GCNK region consists of <3% Hispanic and other minorities; all self-identified black or white subjects were included in our analysis. Height and weight information was not collected in 1993/1994, only partially collected in 1999, and collected throughout all of 2005.
Study physicians reviewed every abstract to verify whether a stroke or TIA had occurred, then assigned stroke category and mechanism to each event based on all available information using definitions described previously.
Descriptive statistics included means or medians, as appropriate. In testing for trends over time, we tested for significant changes between periods as well as for overall time trend, with the null hypothesis of no change. We used a mixed-model approach for both categorical and continuous variables to account for the weighting and sampling scheme. We adjusted for multiple comparisons using the Bonferroni correction.
The numerator for incidence rate calculation was the number of first-ever strokes confirmed by physician review, ascertained through inpatient records or emergency departments, plus the number of first-ever strokes ascertained through public health clinics, hospital-based outpatient clinics and family practice centers, and coroners' offices, plus a weighted estimate of the number of strokes ascertained only in the physician's office or nursing home. As an example, events ascertained in physicians' offices and nursing homes for 2005 events were multiplied approximately 16- and 5-fold, respectively, to account for the sampling methodology. Events were considered to be noncases if medical records could not be located to confirm the event.
The denominator for the incidence rate calculation was obtained from the US Census Bureau Web site (www.census.gov
). These estimates are based on extrapolation or interpolation of county population between enumerated census years, accounting for births, deaths, and migration. These denominators were not adjusted to exclude those with prior stroke, as we do not have accurate prevalence estimates in our population for each study year, only from the telephone surveys (see next paragraph). The at-risk population included 218,906 black patients and 1,100,950 white patients for 2005.
The 95% confidence intervals for the incidence rates were calculated assuming a Poisson distribution. Age-, race-, and gender-specific rates were also determined. All adjusted rates were standardized to the 2000 US population. Rates were calculated for adults only (age ≥20).
To examine changes in risk factor prevalence, we utilized data from 3 random-digit dialing telephone surveys performed in our study region during 1995,12
as well as risk factor prevalence data from the National Health and Nutrition Examination Surveys (NHANES).15–17
Survey methods have been previously described and our full population survey results have been reported.12–14