We have identified several limitations that may affect the results, interpretation, and generalizability of the study. We aimed to exclude patients with a clear indication for IV fluid administration with diagnosis- and narrative-based exclusion criteria; however, it is possible that patients may have inadvertently been included or excluded if, for example, a diagnosis was not documented or went undetected, or an assessment of volume status was not documented. This is inherent to the retrospective design of the study. Similarly, the administration of prehospital IV fluids and sedating medications was not accounted for, nor was the infusion rate, volume, or type of IV fluids, although in most cases it consisted of 1–2 liter boluses of either normal saline or lactated Ringers solution.

Notably patients were not necessarily excluded from the non-IV fluid group if they had a contraindication to IV fluid administration (eg, current or high risk of volume overload). This increased study generalizability, however, may also have led to increased variance, although on subjective narrative review it was not a regular occurrence. Recall bias may also have influenced the results because the data point for alcohol involvement was at times determined based on the narrative documentation of a patient interview while they were intoxicated and could have resulted in misclassification bias, although in many cases blood alcohol levels were available.

Vital signs were collected and shock index calculated to be considered as surrogates for hypovolemia, but further exclusion accuracy could have been achieved through laboratory value interpretation such as kidney function, hemoconcentration, and urine studies. In many cases, no laboratory studies were ordered, and when significant abnormalities were present, they often were identified in the exclusion criteria through supporting diagnoses, such as acute kidney injury. The requirement of a heart rate ≥ 90 bpm was used as another surrogate means of increasing the likelihood that the patient in fact was intoxicated with a stimulant as is commonly the case; however, especially in the context of polysubstance use or even withdrawal, this surrogate marker may be misleading. For example, a patient who used stimulants and opioids and required naloxone for respiratory depression may be diagnosed, among other things, with stimulant intoxication. The addition of vital signs at discharge to the analysis could have provided more clarity between groups. Additionally, an attempt to elucidate a clear indication for IV fluid administration could have allowed for subgroup analyses with less variability; however, this would not have been reliable with the retrospective nature of the study, which instead addressed this by previously described exclusion criteria. Although the administration of IV fluids may expose patients to multiple adverse effects beyond an increase in discharge LOS,^4–6 these events were not quantified in this study and often resulted in exclusion because of a requirement of admission; they represent an important area of future research.

Similarly, given the wide variety of stimulant intoxicants and their associated durations of effect in addition to dose and route of administration variability, the details of stimulant use were not quantified among the study groups and would likely be difficult to generalize to other centers, if even detectable, in the context of regional variability in drug supply. Based on urine drug screens and patient history, this study involved patients intoxicated with stimulants that included cocaine, methamphetamine, and synthetic marijuana; however, the exact intoxicant(s) from the local drug supply could not be confirmed with certainty in the clinical setting. Another potential contributing factor to variability was the lack of a standardized approach to the management of the stimulant-intoxicated patient, where the clinician’s practice pattern could have influenced discharge LOS significantly, although this study involved patients managed by dozens of different physicians over multiple years at a single center.

Finally, this study did not account for time of presentation or season of the year in addition to other commonly considered throughput modifiers such as peak volumes, variation of average acuity in the department, and general hospital throughput including severity of boarding of admitted patients in the ED having a ripple effect on resource demand. These limitations highlight the complexity of studying acutely intoxicated patients in the ED.