STUDY LIMITATIONS
Our study has a number of limitations which need to be addressed. Firstly, there are constraints with using administrative claims data such as the NIS36. It is a de-identified database; making it impossible to validate individual ICD-9 codes. Therefore, inaccuracies in coding may occur. Furthermore, electrograms and telemetry strips were not available to validate the type of arrhythmia in our sample. We were unable to subcategorize the type of arrhythmia is our sample, such as identifying persistent versus paroxysmal AF. In addition, the NIS does not provide long term follow-up data. Therefore, it is probable that outcomes were underestimated in our study. The NIS regards each hospitalization as a separate entity, so it is possible that readmissions were regarded as distinct hospitalizations, leading to an overestimation in the number of hospitalizations. The cause of in-hospital mortality could not be ascertained since hospitalization notes are unavailable through the NIS. It is uncertain whether changes in coding practice patterns contribute to the changes in reported arrhythmia frequency, using NIS methodology.
Apart from the aforementioned limitations, it is also possible that patients with arrhythmias had comorbid conditions that we have not assessed. Therefore, we cannot completely exclude the possibility that additional unknown confounding variables, not included in our multivariate analysis, may explain some of the associations found. We could not, for instance, ascertain the extent at which concurrent medication use such as anti-retroviral therapy (ART) impacted the incidence of arrhythmias, nor could we determine the relationship between disease severity (CD4 count or viral load, for instance) and arrhythmia risk. Future studies should be aimed at investigating the various drug regimens for HIV and their associated risk of developing arrhythmias. Data on CD4+ cell count and HIV viral load was not collected in our sample as they were not available in the NIS registry. Further studies are warranted to elucidate the specific mechanisms by which HIV can lead to the development of AF. Lastly, our sample did not include uninfected controls to allow a direct comparison of the burden of arrhythmias between HIV and non-HIV patients. Such a study is warranted to determine if HIV infection is an independent risk factor for the development of arrhythmias. Unfortunately, it was not possible to perform this analysis using the NIS registry, given the very large population of uninfected controls that would have to be included in the analysis (approximately 80 million patients from the years 2005-2014).
These limitations are, however, counterbalanced by the presence of a large unrestricted study sample which lacks the selection bias found in studies reported by individual specialized centers and skilled operators.