Prediction of extended period of vasopressor infusion requiring central venous catheterisation: A burning issue in critical care

Vasopressor infusion is one of the most important interventions in the critical care setting, especially for patients with circulatory failure. ¹ While vasopressors can help to elevate arterial blood pressure and maintain adequate tissue perfusion, prolonged use of vasopressors is associated with non-trivial complications, including cardiac arrhythmia, peripheral organ ischaemia, and a requirement for a central venous catheter (CVC). It is well established that prolonged central venous catheterisation carries a high risk of catheter-related bloodstream infection (CRBSI) and deep vein thrombosis, in addition to mechanical complications related to the cannulation procedure itself. ² Thus, minimising the placement and use of CVCs is an important quality improvement target in the critical care setting.

There has been accumulating evidence showing that infusing vasopressors through a peripheral venous catheter (PVC), at least in a diluted form, is both feasible and safe.^3–6 Infusing vasopressors via a PVC for a short period of time is helpful to avoid CVC placement. Therefore, it is important to identify risk factors for extended vasopressor infusion, and to predict who will need prolonged use of vasopressors as early as possible. Early identification of patients who will need extended vasopressor administration can help to develop tailored treatment strategies. For example, patients with a low risk of requiring extended vasopressor infusion can have a PVC placed instead of a CVC. In this issue of Anaesthesia and Intensive Care, Haimovich and colleagues conducted an interesting study to investigate risk factors for the prolonged use of vasopressors by using a large critical care database. ⁷ The authors first described the epidemiology of vasopressor use in a United States critical care setting. Interestingly, the majority of patients with a short duration of vasopressor use (<24 hours) had CVCs placed. Although CVCs may be inserted for other indications such as central venous pressure and/or oxygen saturation monitoring, the data would suggest that they might be overused in the current critical care practice. In our clinical experience, physicians are likely to insert a CVC when patients show signs of circulatory shock, irrespective of the anticipated duration of the shock. We posit that patients who require a single vasopressor for less than six hours may not need a CVC. As there is accumulating evidence supporting the safety of using a PVC for (diluted) vasopressor infusion, ⁸ the liberal use of CVCs for short-term vasopressor therapy, as in current clinical practice, needs to be re-evaluated.

In some patients who require vasopressor therapy, their shock state can be corrected after adequate fluid resuscitation, with the vasopressor weaned off quickly. The difficulty is how we can identify those who would do so at the time when the vasopressor is initiated. The lack of an accurate tool to predict how long a vasopressor is needed for critically ill patients invariably leads to the overuse of CVCs in our current clinical practice. The study by Haimovich et al. is thus highly relevant in this context. Some interesting risk factors of prolonged vasopressor use were identified in this study, including cardiac problems as the main diagnosis, renal impairment, endotracheal intubation, and older age. These risk factors are common and would agree with our clinical intuition regarding factors that were previously reported to be associated with prolonged vasopressor use, such as cirrhosis, ⁹ race-ethnicity, ¹⁰ the use of antihypertensives and a history of chemotherapy.^11,12

Translating research findings into clinical practice is an important step. Reliable and user-friendly decision-support tools are pivotal in bridging the gap between research and clinical practice. Predictive analytics have been widely investigated in the critical care setting,^13–15 primarily in the area of risk stratification. Interest in predictive analytics has recently been boosted by the surge in popularity of machine learning (ML). Although in most cases the use of complex ML algorithms does not necessarily improve the prediction accuracy,^16,17 researchers remain convinced that ML has a huge potential to improve patient care in the future either directly or indirectly. A major drawback of ML for clinicians is the ‘black-box’ nature of the ML-derived prediction models, ¹⁸ resulting in clinicians not trusting the models if underlying biological or pathophysiological rationale of the models is unclear or if the variables chosen by the models are not clinically relevant or applicable in practice. In this regard, Haimovich et al. ⁷ should be congratulated for developing a clinically relevant multivariable regression model for the prediction of prolonged vasopressor use by using only variables that are available at the time of intensive care unit (ICU) admission.

One caveat that warrants further discussion prior to adoption of the tool is the fact that the prolonged use of vasopressors does not necessarily mean the need for a CVC. Furthermore, the current prediction model has not been validated in external cohorts. In particular, patients who were already deemed to require a CVC prior to ICU admission were excluded in the current study. Given there could be substantial difference across institutions in the practice of vasopressor infusion and CVC insertion both in the operating theatre and emergency department, further evaluation of this new prediction model is needed by different institutions before the model is used routinely as a clinical decision-support tool at the ‘coalface’. The accuracy of any prediction models is bounded by space and time. Clinical algorithms require evaluation using local data prior to implementation, as well as continuous monitoring and recalibration. ¹⁹