Objective: Short-term cardiovascular compensatory responses to perturbations in the circulatory system caused by haemodialysis can be investigated by spectral analysis of heart rate variability. This could provide an important variable for categorising individual patients response to haemodialysis leading to a more personalised treatment. However, data obtained over a four-hour haemodialysis treatment is significant in volume and subject to artefacts that can compromise its analysis. Methods: The Lomb-Scargle Periodogram can provide a robust method of generating power spectral density estimates for large, irregularly sampled and noisy data sets obtained in clinical settings, provided that careful attention is given to frequency limits. The effect of different pre-processing methods on the resulting power spectrum is explored with simulated and real heart rate variability data. Results: Common pre-processing methods for correcting individual artefacts in heart rate records, such as interpolation, are unreliable as they act as non-linear low-pass filters and distort the resulting spectral analysis. These distortions are present, but less apparent within patient data and can mislead clinical interpretations. Conclusion: It is more appropriate to exclude suspect data points than to edit them prior to spectral analysis via the Lomb- Scargle periodogram, and where required, de-noise the entire heart rate signal by empirical mode decomposition. The use of a False Alarm Probability metric can help establish whether spectral estimates are valid Significance: Methods established to pre-process time-invariant data prior to power spectral density estimation fail when used in conjunction with the Lomb-Scargle method.