After reading the case study of 75 HIV infected patients treated with atazanavir-based therapy reported by Gianotti et al. [1], we express concerns regarding the methodology used to measure ECG intervals in this report. In particular, we note that automated measurement by ECG recording machines was used. This is in contrast to the current industry standard for ECG interval measurements to be digitally measured with electronic calipers or confirmed by experienced reviewers. In our view, the absence of digital overreading resulted in inadequate quality control for the QRS interval measurements and a false-positive finding of QRS prolongations in patients treated with atazanavir [1]. The authors report a median increase in QRS duration of 5 ms [interquartile range (IQR) = 0–9 ms; P < 0.0001] on atazanavir after 51 weeks of follow-up (range = 36–89.9 weeks) compared with pretreatment values. In particular, the authors report 14 patients with pretreatment QRS 100 ms or less who experienced an end-of-study QRS interval greater than 100 ms. One of these patients (patient #1) appeared to experience a marked increase in QRS duration (baseline QRS of 88 ms compared with end-of-study QRS of 145 ms). The authors report involvement of a cardiologist in the ECG review. However, because of the possibility of unrecognized ECG machine error, we requested further information from the site. Dr Gianotti et al. graciously supplied photocopies of the actual baseline and end-of-treatment ECGs for the 75 patients in the original report [1]. All of the 150 paper copies (75 patient sets) were electronically scanned using a standard HP Digital Scanner at 300 dpi and then transferred to Scan ECG (AMPS, New York, New York, USA), by Cardiocore (Bethesda, Maryland, USA). Due to poor quality, 16 of the 75 sets could not be interpreted due to unsuccessful digitization. Using electronic calipers, 59 sets of ECGs were measured for QRS width. This technique, often referred to as semiautomatic over-reading, is the industry standard and involves expert review. In the set of 59 digitized samples, we found that only four patients experienced an increase in QRS duration of greater than 10 ms, compared with 15 patients as determined by machine interrogation. Furthermore, no patient experienced QRS duration greater than 116 ms. We found significant discrepancies for two patients reported to have QRS greater than 120 ms in the original report, including the one with the marked increase at end-of-study, patient #1, with corrected values, as shown in Table 1.Table 1: Comparison of QRS values.The machine versus digital-caliper discrepancy is very likely due to random error by the machine algorithm, which can be affected by the variability in rate of change of the tracing as is commonly found with artifact. The end-of-study tracing for patient #1 shows a significant degree of artifact. Taking the 59 digitally measured sets altogether, the median baseline to end-of-study QRS change was 1.0 ms (IQR =–2 to–4; P = 0.246) The corresponding machine-calculated median change was 5 ms (IQR = 1–10; P < 0.001) in the same set of 59 patients. Because the QRS width is difficult to gauge with the human eye using manual calipers (with a likely error of ±15–20 ms) and because the machine QRS calculation is affected by artifact, the conclusions of Gianotti et al.[1] should be regarded with caution. Overreading using digital calipers represents the gold-standard technique. Although we were only able to investigate 80% of the originally reported ECG set, the uncorrected median value of 5 ms for this subset compares closely to the originally reported uncorrected median value of 5 ms for the whole cohort [1]. Furthermore, as the subset includes the extremely anomalous result for patient #1, our reanalysis is likely to be relevant. As such, we note that the corrected median QRS change of 1 ms in these 59 sample sets is likely to have very little clinical relevance. We respectfully disagree with the QRS results and the conclusions presented by Gianotti et al.[1], including their recommendation for routine ECG monitoring with regard to QRS as well as their proposed association between QRS changes and bundle branch block effects. We note that one patient experienced resolution of bundle branch block findings while on atazanavir. As we show here, the study would have been better served by digital collection of ECG tracings with intervals determined by the digital caliper overread technique. We are grateful to the authors for sharing the ECG data.
Read full abstract