OBJECTIVES: To determine whether inducing correlation among triglyceride, HDL, and LDL levels in a pharmaceutical treatment Monte Carlo simulation affects parameters' means and variances; proportion with all parameters controlled; and summary statistics of estimated total cholesterol. METHODS: Means, standard deviations, and correlations among the cholesterol parameters were estimated from NHANES data for metabolic syndrome (MS) and diabetic patients with all parameters uncontrolled. For simulation, distributions were fit to the data. Analyses used 1000 replications of populations of 1000. Populations were generated without correlated parameters and with correlation induced in the uncorrelated data. Estimated changes with fenofibrate, statins, and a combination were taken from the literature. Total cholesterol was approximated using HDL, plus LDL, plus 20% of triglycerides. Differences in means and ratios of variances comparing uncorrelated and correlated results were calculated for each replication. Null hypotheses were rejected when the interval the middle 95% of replications spanned did not include zero for differences and one for ratios. RESULTS: Correlations were higher for diabetic than MS patients. Despite the data's and distribution's non-normality, induced correlations were close to NHANES correlations. Correlation did not affect the summary statistics of individual parameters or the proportion with all parameters under control. Correlation affected results for total cholesterol, the sum of other parameters. For MS, variance of total cholesterol was less than 7% lower with uncorrelated data than with correlated data. For diabetic patients, variance of total cholesterol was more than 20% higher with uncorrelated data. Findings held for subpopulations with and without all parameters controlled after taking medication. Variance results were similar across treatments. Total cholesterol means differed primarily for MS subgroups. CONCLUSIONS: Summary statistics (particularly variance) for sums of parameters are affected by correlation in Monte Carlo simulations. Underestimated and overestimated variances increase the risk of Type I and II error respectively.