Potential for bias in case-crossover studies with shared exposures analyzed using SAS.