Heart rate variability (HRV) is a noninvasive marker of cardiac autonomic function that has been extensively studied in a variety of populations. However, HRV analyses require stationarity-thus, limiting the conditions in which these data can be analyzed in physiologic and health research (e.g. post-exercise). To provide evidence and clarity on how non-stationarity affects popular indices of variability and complexity. Simulations within physiologic (restricted to values similar to exercise and recovery RR-intervals) and non-physiologic parameters, with homoscedastic and heteroscedastic variances, across four sample lengths (200, 400, 800, and 2000), and four trends (stationary, positive-linear, quadratic, and cubic) were detrended using 1-3 order polynomials and sequential differencing. Measures of variability [standard deviation of normal intervals (SDNN) and root mean square of successive differences (rMSSD)] as well as complexity [sample entropy (SampEn)] were calculated on each of the raw and detrended time-series. Differential effects of trend, length, and fit were observed between physiologic and non-physiologic parameters. rMSSD was robust against trends within physiologic parameters while both SDNN and SampEn were positively and negatively biased by trend, respectively. Within non-physiologic parameters, the SDNN, rMSSD, and SampEn of the raw time-series were all biased, highlighting the effect of the scale between these two sets of parameters. However, indices of variability and complexity on the original (trended) times-series were furthest from those of the stationary time-series, with indices coming closer to the known values as fit become more optimal. Detrending with polynomial functions provide reliable and accurate methods of assessing the variability and complexity of non-stationary time-series-such as those immediately following exercise.