Methods
Approval was granted by the Richmond VA Medical Center (VAMC) Institutional Review Board in Virginia, and by the VA Informatics and Computing Infrastructure (VINCI) data access request tracker system for extraction of data from the VA Corporate Data Warehouse. Microsoft Structured Query Language was used for data programming and query design. Statistical analysis was conducted using Stata version 15.1 with assistance from professional biostatisticians.
Only male veterans with a nervous system disorder affecting the spinal cord or with myelopathy were included, based on International Classification of Diseases (ICD) version 9 and 10 codes, corresponding to traumatic and nontraumatic myelopathy. Veterans diagnosed with myelopathy based on ICD codes corresponding to progressive or degenerative myelopathies, such as multiple sclerosis or amyotrophic lateral sclerosis, were excluded.
For each veteran, extracted data included the unique identification number, date of birth, ICD code, date ICD code first appeared, race, gender, death status (yes/no), date of death (when applicable), date of each PSA test, PSA test values, and the VAMC where each test was performed. Only tests for total PSA were included. The date that the ICD code first appeared served as an approximation for the date of SCI. The time frame for the study included all PSA tests in the VINCI database for 2000 through 2017. However, only post-SCI PSA tests were included in the analysis. Duplicate tests (same date/time) were eliminated.
Race is considered a risk factor for prostate cancer only for African American patients, likely due to racial health disparities.17 Given this, we chose to categorize race as either African American or other, with a third category for missing/inconsistent reporting. Age at time of the PSA test was categorized into 4 groups (≤ 39, 40-54, 55-69, and ≥ 70 years) based on AUA guidelines.4 The annual PSA testing rate was calculated for each veteran with SCI as the number of PSA tests per year. A mean annual PSA test rate was then calculated as the weighted (by exposure time) mean value for all annual PSA testing rates from 2000 through 2017 for each age group and race. Annual exposure was calculated for each veteran and defined as the number of days a veteran was eligible to have a PSA test. This started with the date of SCI diagnosis and ended with either the date of death or the date of last PSA. If a veteran moved from one age group to another in 1 year, the first part of this year’s exposure was included in the calculation of the annual PSA testing rate for the younger group and the second part was included for the calculation of the older group. For deceased veterans, the death date was excluded from the exposure period, and their exposure period ended on the day before death.
Statistical Analysis
To compare PSA testing rates between African American race and other races, Poisson regression was used with exposure treated as an offset (exposures were summed across years for each veteran). An indicator (dummy) variable for African American race vs other races was coded, and statistical significance was set at P < .05. To check sensitivity for the Poisson assumption that the mean was equal to the variance, negative binomial regression was used. To assess for geographic PSA testing rate variability, the data were further analyzed based on the locations where PSA tests were performed. This subanalysis was limited to veterans who had all PSA tests in a single station. For each station, the average PSA testing rate was calculated for each veteran, and the mean for all annual PSA testing rates was used to determine station-specific PSA testing rates.