Door-to-needle delays in minor stroke: A causal inference approach

Abstract

Background and Purpose

Thrombolysis rates among minor stroke (MS) patients are increasing due to increased recognition of disability in this group and guideline changes regarding treatment indications. We examined the association of delays in door-to-needle time (DTN) with stroke severity.

Methods

We performed a retrospective analysis of all stroke patients who received tissue plasminogen activator in our emergency department between 7/1/2011 and 2/29/2016. Baseline characteristics and DTN were compared between MS (NIHSS ≤5) and non-minor strokes (NIHSS>5). We applied causal inference methodology to estimate the magnitude and mechanisms of the causal effect of stroke severity on DTN.

Results

Of 315 patients, 133 (42.2%) had NIHSS ≤5. Median DTN was longer in MS than non-minor strokes (58 vs. 53 minutes, P=0.01); fewer MS patients had DTN ≤45 minutes (19.5% vs. 32.4%, P=0.01). MS patients were less likely to use Emergency Medical Services (EMS) (62.6% vs. 89.6%, P<0.01) and to receive EMS pre-notification (43.9% vs. 72.4%, P<0.01). Causal analyses estimated MS increased average DTN by 6 minutes, partly through mode of arrival. EMS pre-notification decreased average DTN by 10 minutes in MS patients.

Conclusion

MS had longer DTN times, an effect partly explained by patterns of EMS pre-notification. Interventions to improve EMS recognition of MS may accelerate care.

Introduction

Rates of thrombolysis for patients with minor strokes (NIHSS≤5) have historically lagged behind those with non-minor deficits. Few minor stroke (MS) patients were included in the efficacy trials of IV-tPA¹ and minor symptoms were previously a relative contraindication to thrombolysis.² Emerging recognition of both disability³ and thrombolysis safety⁴ has led to increased thrombolysis of MS patients.⁵ A continued rise is expected as treatment guidelines change.² Evolving practice patterns make identifying opportunities to expedite MS treatment essential.

In this cohort study, we compare treatment times between MS and non-minor stroke (NMS) patients and use standard methods of causal inference to evaluate mechanisms through which MS cause prolonged door-to-needle (DTN) times.

Methods

We queried our prospective stroke registry for patients who received IV-tPA in our emergency department (ED) from 7/1/2011–2/29/2016. MS was defined as NIHSS≤5.¹ We compared demographics, EMS usage and pre-notification, DTN and component times between MS and NMS (NIHSS >5) patients. We also compared the proportion of MS and NMS patients with the fastest DTN (lowest quartile, ≤45 minutes). All times were abstracted from a standardized stroke consult note in the electronic medical record modeled on Get-With-The-Guidelines requirements. Categorical variables were compared with χ², continuous with Student’s t, and medians with Mann-Whitney U tests.

We applied standard methods of causal inference⁶ to estimate the magnitude and mechanisms of the effect of stroke severity on DTN. In particular, we explored ED mode of arrival (MOA) as a mediator. MOA took three values: walk-in, EMS without pre-notification, and EMS pre-notification. We estimated the (1) average effect of stroke severity on DTN, (2) average effect of stroke severity on MOA, (3) average effect of MOA on DTN, (4) conditional average effects of MOA on DTN in MS and NMS, and (5) controlled direct effects⁷ (CDEs) of stroke severity on DTN with MOA fixed.

‘Conditional average effects’ are causal effects within subpopulations (i.e. MS and NMS patients). CDEs are causal effects had mediators taken a fixed value in the population. For example, the CDE of stroke severity on DTN with MOA fixed at EMS pre-notification is the effect of stroke severity had everybody arrived with EMS pre-notification. This is a ‘direct effect’ because if everybody had the same MOA then any effect of stroke severity must be through pathways other than MOA.

We adjusted for age, sex, and primary language. The conditional expectations in the g-formula were estimated by random forest.⁸ Standard errors were computed using nonparametric bootstrap. Figure 1 describes causal assumptions required to avoid confounding and selection bias.

Figure 1.

A. Causal graph representing assumptions necessary to avoid bias from unobserved confounding in all our analyses. Specifically, it communicates the assumptions that there are no common causes of stroke severity and MOA, stroke severity and DTN, or MOA and DTN other than age, sex, and primary language. Each individual analysis requires only a subset of these assumptions

B. Causal graph which encodes the assumptions necessary to ensure that we did not induce selection bias in any of our causal analyses by restricting the sample only to patients who received IV-tPA. Specifically, we assume that (a) there are no unobserved common causes of MOA and IV-tPA administration; (b) MOA does not influence whether IV-tPA is administered; and (c) the time until it is determined whether to administer IV-tPA does not influence whether IV-tPA is administered.

SPSS v23.0 was used for descriptive analyses; R Statistical Software for causal modeling. P<0.05 was significant. The Columbia University Medical Center IRB approved this study.

Results

Over the study period, 372 patients received IV-tPA. Of these, 9 (2.4%) developed symptoms after ED arrival for a different cause and 48 (12.9%) were stroke mimics. Thus 315 patients with a final diagnosis of ischemic stroke were included; 133 (42.2%) were MS patients.

Compared to NMS, MS patients were younger (64.5 vs. 73.9 years, P<0.01), less likely to use EMS (62.6% vs. 89.6%, P<0.01), and less likely to receive EMS pre-notification (43.9% vs. 72.4%, P<0.01). Median DTN was longer in MS patients (58 vs. 53 min, P=0.01); fewer MS patients had DTN≤45 minutes (19.5% vs. 32.4%, P=0.01). Table 1 includes other DTN component times. In a sensitivity analysis including stroke mimics, DTN and component times remained longer in MS patients.

Table 1.

Cohort characteristics

	NIHSS≤5 (N=133)	NIHSS>5 (N=182)	P-value
Ageˆ	64.5 (14.5)	73.9 (15.8)	<0.01
Male (%)	59 (44.4)	64 (35.2)	0.10
Language (%)			0.90
English	68 (51.1)	94 (51.6)
Spanish	61 (45.9)	84 (46.2)
Other	4 (3.0)	4 (2.2)
EMS-911 (%)	82 (62.6)	163 (89.6)	<0.01
Pre-notification	36 (43.9)	118 (72.4)	<0.01
Arrival “on-hours” (%)•	40 (30.1)	60 (36.8)	0.23
Treatment times (min)*
Onset-to-door	70 [44–112]	65 [43–92.25]	0.18
Door-to-needle	58 [47–79]	53 [42–75]	0.01
Door-to-needle≤45min (%)	26 (19.5)	59 (32.4)	0.01
Door-to-stroke activation	4 [0–14]	1 [0–6]	<0.01
Door-to-imaging	24 [17–35]	20 [15–28]	0.02
Imaging-to-needle	33 [21–45]	30 [19–45]	0.18

^ˆMean(SD)

^•Monday–Friday/7am–5pm

^*Median[IQR]

Table 2 summarizes causal analyses. The estimated effect of MS on DTN through all mechanisms is approximately 6 minutes (95%CI 2–9 minutes). That is, with all other pre-stroke variables equal average DTN would be 6 minutes longer if every stroke were minor than if every stroke were non-minor.

Table 2.

Causal analyses

	EMS pre-notification	EMS without pre-notification	Walk-in
Expected DTN* if all arrive by:	58.5 [55.2, 61.7]	64.6 [60.5, 68.6]	63.6 [59.4, 67.8]
Expected DTN* among MS if all arrive by:	60.5 [55.1, 65.9]	70.9 [63.9, 77.9]	69.7 [63.5, 75.9]
Expected DTN* among NMS if all arrive by:	58.6 [54.1, 63.1]	62.5 [56.4, 68.6]	57.6 [51.4, 63.8]
CDE* of MS on DTN if all arrive by:	+2.6 [−1.2,+6.4]	+5.8 [+1.1,+10.6]	+5.7 [+1.5,+10]
If everyone is MS, probability of MOA being:	28% [17%,39%]	35% [22%,49%]	37% [25%,48%]
If everyone is NMS, probability of MOA being:	78% [69%,85%]	15% [9%,22%]	7% [3%,12%]
Total effect of MS on DTN	+5.7 minutes [+1.9,+9.3]

^*minutes

MS minor stroke; MOA mode-of-arrival; CDE controlled direct effect

We also estimated that stroke severity affects pre-notification, and pre-notification in turn affects DTN. If all strokes were minor, the probability of EMS pre-notification would be 28% versus 78% if all strokes were non-minor. If everyone received EMS pre-notification, average DTN would be approximately 6 minutes shorter than if everyone arrived via walk-in or EMS without pre-notification. Among MS patients, we estimated EMS pre-notification reduces average DTN by approximately 10 minutes (95%CI 3–17 minutes).

To evaluate whether any effect of stroke severity on DTN would remain if everyone received EMS pre-notification, we estimated the CDE of MS with MOA set to EMS pre-notification. This CDE was 2.6 minutes, but the data were consistent with no direct effect. To assess the effect of stroke severity on DTN through other mechanisms in the absence of EMS pre-notification, we considered the CDEs of stroke severity on DTN setting MOA to EMS without pre-notification and walk-in. If everyone had arrived by these MOAs, MS would increase DTN by 6 minutes (95%CI 1–10 minutes).

Discussion

After observing longer DTN in MS patients, we estimated the magnitude and mechanism of the causal effect of stroke severity on DTN. We found that MOA is an important mediator. Specifically, MS decreases EMS pre-notification. In turn, EMS pre-notification decreases DTN, particularly in MS patients. The data were consistent with no direct effect of stroke severity on DTN if EMS pre-notification were universal. Our analyses suggest targeted interventions to increase EMS pre-notification among MS patients could yield large public health benefits.

Decreased pre-notification among MS patients may suggest delayed stroke symptom recognition by EMS personnel.⁹ Alternatively, poor recognition of minor symptoms by patients may affect EMS assessments.¹⁰ While national data confirm increasing odds of EMS pre-notification with greater stroke severity,¹¹ our findings highlight the need for specific interventions among MS patients, a population often missed by current EMS screening tools.⁹ Another factor in longer DTN times in MS patients may be physician uncertainty about utility of IV-tPA treatment in this group. Physician comfort with treating outside guideline recommendations has been demonstrated to play a role in IV-tPA treatment, which may have affected our findings.¹²

This study has several limitations. As an observational study, unobserved confounding bias is likely. Including only tPA-treated patients could introduce selection bias. Assumptions required to avoid these biases are explained in Figure 1 and the online data supplement. We also were unable to account for patient location at the time of stroke, which could affect EMS usage. Lastly, as a single center study, our findings may have limited generalizability.