Gender differences in time from arrival to stroke team activation in patients presenting with acute stroke symptoms

2.1 Study design and setting

This was an observational cross-sectional study in which we retrospectively queried the prospective stroke registry of our urban, regional referral stroke center hospital, which included all consecutive patients who presented to the adult ED between June 15, 2014 and June 14, 2015 for whom the stroke team was activated. To investigate gender differences in care, we employed a previously developed methodology to investigate a novel subset of the stroke registry [6]. During the study period, there were 67,795 total adult ED visits and approximately 27,000 adult inpatient admissions. The center is a primary teaching site for multiple residencies, including emergency medicine and neurology, and there is a stroke team available in-house 24 hours per day, 7 days a week. The stroke team may be activated prior to hospital arrival by emergency medical services (EMS) or at any time during triage or ED care by ED nurses or physicians. The ED is staffed by board-certified/board-eligible attending emergency physicians who supervise emergency medicine and off-service rotating residents. All American Heart Association/American Stroke Association Get With The Guidelines recommendations have been implemented [23, 24], and ED nursing and physician staff undergo periodic acute stroke continuing education. The study was approved by our Institutional Review Board.

2.2 Selection of participants

The institution maintains a prospective registry of all patients for whom the stroke team is activated, which includes patient demographics and time stamps for key care events including ED arrival time, stroke team activation time, imaging initiation and completion times, and time to thrombolytic administration. A stroke nurse coordinator maintains the registry and verifies its accuracy based upon established institutional guidelines. Automated and manual processes exist to ensure 100% registry capture of all patients for whom stroke resources are activated.

By protocol, the stroke team was activated when any patient presented to the ED with symptoms or findings consistent with an acute stroke within 12 hours of symptom onset. Our multidisciplinary stroke committee previously established the 12-hour window accounting for 3 key considerations: prioritizing sensitivity over specificity, availability of resources enabling treatment beyond 4.5 hours of symptoms in select cases, and institutional research protocols.

We excluded patients who presented via EMS with pre-hospital stroke activation because no data existed with which to determine timeliness of stroke recognition before hospital arrival. All other patients in the stroke registry were activated by a member of the ED staff, after arriving either via EMS or by other means and were included in our investigation. Patient’s presenting with symptoms concerning for stroke with symptom onset less than 12 hours were initially evaluated in the ED resuscitation rooms to facilitate rapid evaluation and care. Use of these rooms for potential stroke patients was determined by the evaluation of the triage nurse.

2.3 Methods and measurements

Trained research assistants, who were blinded to the study hypothesis, retrospectively reviewed the electronic health record (EHR) (ED PulseCheck, Optum Clinical Solutions, Inc., Eden Prairie, MN; Soarian, Cerner Corporation, North Kansas City, MO; and OnBase, Hyland Software, Inc., Westlake, OH) for each patient in the registry to validate the registry data and abstract the following fields (determined a priori) using standardized abstraction forms: age, gender, triage emergency severity index (ESI) score [25], mode of arrival (EMS versus non-EMS), whether the patient was initially seen in an ED resuscitation/critical care room, initial vital signs (heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure, and oxygen saturation), supplemental oxygen use and delivery method (none, nasal cannula, face mask, bag-mask ventilation, or intubated), Glascow coma scale (GCS) score, level of orientation (person, place, and time-range of 0-3), National Institute Health Stroke Scale (NIHSS) score, initial blood glucose value, elapsed time since the patient was last known to be at their baseline neurologic condition, previous history of stroke or transient ischemic attack, previous history of diabetes mellitus, and previous history of hypertension.

The EHR was also reviewed to determine whether the ED team documented treatment of another emergent, life-threatening condition that delayed care, categorized as airway/breathing intervention required, hypertension, hypotension, hypoglycemia, emergent electrolyte abnormality, or more than one of the above conditions. However, all documented delays pertained to the interval between stroke team activation and neuroimaging, and there were no documented delays in stroke recognition or activation. The registry and our EHR considered gender to be binary and self-reported by the patient. Rare missing values in the registry were obtained from the EHR by the abstractor. A second investigator independently searched the EHR for missing values after the initial abstraction and also independently validated all abstracted data for a subset of cases primarily abstracted by each research assistant. Missing values not available in either the registry or the EHR (vital signs, n = 1; glucose, n = 4; and GCS, n = 75) were replaced with the corresponding median value for the remaining data set. One entry in the registry was an exact duplicate, so the affected patient was analyzed only once. We have previously validated and employed a similar abstraction and data verification methodology for another registry-based study [6].

2.4 Outcomes

Our primary outcome was DTA time, defined as the interval between the EHR timestamp for ED arrival and the time of institutional stroke team activation by the ED staff, as recorded in the registry. We chose DTA given its prevalence in the literature as a key step in ED stoke care [5, 8, 11, 23, 24, 26] and its inherent threshold effect on other metrics, such as door-to-imaging time and door-to-needle time for thrombolytics. In our primary analysis, we considered DTA as a continuous outcome. In a secondary analysis, we considered the binary outcome of DTA ≤ 15 minutes (the national goal). Patients for whom the documented duration of symptoms was shorter than the DTA time (one possible explanation being that symptoms may have begun while already in the ED) were reviewed for potential exclusion by full-text review of the EHR documentation by a senior investigator, but no cases of documented symptom onset while in the ED were identified in the included population.

Our secondary outcome was the binary variable of DIT within the national goal (at the time of the study) of ≤ 25 minutes [4], defined as the interval between ED arrival and initial neuroimaging completion (non-contrast computerized tomography at our institution) with images available for interpretation [1]. Additional secondary clinical endpoints were admission to a neurology service, final ED diagnosis of stroke or intracranial hemorrhage, and administration of intravenous thrombolytics or neurointerventional procedure, which were recorded directly in the registry and verified in the EHR.

2.5 Analysis

Our primary predictor of interest was patient gender. We identified 18 additional candidate covariates by investigator consensus, which are listed in Table 1 and Table 2, based upon their plausibility as confounders and/or inclusion in prior studies. We also considered arrival hour of day and day of week as temporal effects but did not find an effect of month/year. We calculated mean arterial pressure as 1/3 times systolic blood pressure plus 2/3 times diastolic blood pressure and used it in lieu of including both systolic and diastolic values to reduce dimensionality. Because NIHSS and age had nonlinear associations with DTA, we performed a priori recursive partitioning analyses to identify potential cut-points to further reduce dimensionality and collapse these nonlinear continuous effects to discrete predictors.

Our primary analysis of continuous DTA was a multivariable Cox Proportional Hazards regression to account for the highly skewed distribution typical of time-to-event data, which typically violate many assumptions of ordinary least squares regression. The Cox model was based upon a non-parametric hazard function, defined as the conditional probability of a given patient achieving DTA in the next instant (i.e. their stroke symptoms being recognized), provided their stroke had not been recognized by that point. (“Hazard” is a misnomer in this context, where it is desirable to have a shorter time-to-event, and the term used in its statistical definition, not it’s colloquial meaning.) For each predictor, we calculated HR, the ratio of hazard rates per unit change in a covariate. We verified the Cox proportionality assumption by generating time-dependent covariates using interactions between each predictor and log-transformed DTA. With the exception of high flow oxygen, none of the time dependent covariates were significant. For high flow oxygen, we graphically verified that Kaplan-Meier curves were parallel across both levels of the covariate and concluded that the very small number of patients with high flow oxygen (causing wide confidence intervals for this covariate) were not likely to meaningfully jeopardize our overall analysis.

Given the prevalence of logistic regression in the existing literature on this topic, we also fit a multivariable logistic regression model of DTA ≤ 15 minutes with maximum likelihood estimation as a secondary analysis. We used an identical approach for our secondary analysis of DIT, except we also added ED occupancy rate [27] at the time of patient arrival, based on prior literature [6]. Given our desire to control for a number of plausible confounders, based mostly on prior research, we included all candidate variables in our model, regardless of univariate statistical significance. We assessed for baseline univariate differences between group using Fisher’s Exact Test for categorical variables and Wilcoxon Rank Sums Test for continuous variables. The logistic regression model exhibited good overall fit (AUC 0.86 and Hosmer-Lemeshow lack-of-fit test P = 0.31). Analyses were conducted using JMP Pro 13.2 (SAS Institute Inc., Cary, NC).

3.1 Characteristics of study subjects and baseline gender differences

Among 490 consecutive stroke activation patients in the registry during the study period, 211 did not have pre-hospital stroke activation and were eligible for inclusion. Table 1 and Table 2 report the baseline characteristics of included patients, stratified by gender. We observed no statistically or clinically significant differences in baseline characteristics by gender, except that male patients tended to be slightly younger (Table 2).

Among all patients, DTA ranged from less than 1 minute to 3 hours, 37 minutes (median 19 minutes, interquartile range [IQR] 23 minutes, 42.7% ≤ 15 minutes). For women, the range was less than 1 minute to 3 hours, 37 minutes (median 22, IQR 26 minutes). For males, less than 1 minute to 1 hour, 32 minutes (median 14, IQR 22 minutes). 53% of males achieved DTA within the goal of 15 minutes, compared to 34% of women (P = 0.008 for two-tail univariate difference).

DIT ranged from 10 minutes to 3 hours, 41 minutes (median 36, IQR 25 minutes) among all patients and was 10 minutes to 3 hours, 41 minutes (median 36, IQR 27 minutes) for women and 13 minutes to 1 hour, 56 minutes (median 33, IQR 25 minutes) for males. DIT was within the 25-minute goal for 26% of patients overall, 35% for males and 18% for women (P = 0.007 for univariate difference).

3.2 Main results

Table 3 reports the HRs for each variable in the multivariable Cox model of DTA. The interpretation of HRs in this context is, “controlling for covariates, what is the probability of a patient with a specific characteristic achieving DTA time faster than a patient without that characteristic?” Being a woman was associated with an adjusted HR of 0.60 (95% confidence interval 0.43 to 0.83) for timely recognition of stroke symptoms and activating stroke resources, compared to male gender. The HR for women in an unadjusted model was 0.73 (95% confidence interval 0.55 to 0.96).

3.3 Secondary outcomes

Logistic regression models of DTA ≤ 15 minutes and DIT ≤ 25 minutes had good overall fit (area under the curve 0.86 and 0.88, respectively). Odds ratios for each variable are given in Table 4. Being a woman was a significant predictor of delays in both DTA and DIT.

There were no gender differences in the secondary clinical outcomes. The final ED diagnosis was stroke or intracranial hemorrhage in 66 (31%) of patients; 128 (61%) were admitted to a neurology service, and 7 (3.3%) received thrombolytics or a neurointerventional procedure. Non-stroke diagnoses were varied and consistent with prior reports [21].