Association of Onset-to-Treatment Time With Discharge Destination, Mortality, and Complications Among Patients With Aneurysmal Subarachnoid Hemorrhage

Key Points Question What is the optimal time between symptom onset and treatment after aneurysmal subarachnoid hemorrhage (SAH) to maximize patient outcomes? Findings In this cohort study, more favorable patient outcomes (discharge home and survival at 12 months) were observed when treatment occurred within 12.5 hours after aneurysmal SAH symptom onset. Treatment delay did not affect neurologic complications after aneurysmal SAH. Meaning The findings of this study provide more information regarding the optimal timelines of surgical treatment for people with aSAH.

This supplementary material has been provided by the authors to give readers additional information about their work. eAppendix 1. Supplemental Methods As reported elsewhere, 31 multiple overlapping sources were used to identify the cases including admission, discharge and ward lists for emergency, neurosurgical and radiology departments across the tertiary centres and referring hospitals. A combination of International Classification of Diseases 10 codes (160.0-160.9, 167.1 and 169.0), as either a primary or secondary diagnosis, and keyword searches were used to ascertain potential cases. A standardised abstraction form using data from radiology, pathology and surgical reports, as well as discharge letters were used to confirm first ever aSAH. Potential cases were coded by one researcher in each site, and a neurosurgeon and a neuro-interventional radiologist resolved any discrepancy in diagnosis. For those who received the two types of procedures (N=4), 'onset-to-treatment'time was defined as the time elapsed between symptoms onset and the first procedure.

Clinical parameters
We used ambulance, emergency, and radiology records to extract demographic information, baseline comorbidities, and clinical details including aSAH complications, severity indices (World Federation of Neurological Surgeons (WFNS) 14 and modified Fisher grade 15 ), and treatment type (endovascular clipping ( EVT) or neurosurgical clipping (NST)). Complications during hospitalisation were extracted from medical records while 12-month survival post-aSAH was determined using data linkage with the Australian Institute of Health and Welfare.
The modified Fisher scale indicates the risk of developing vasospasm (it progressively increases with each grade). The classification is as follows 3 : For the purpose of model adjustment in the present study, WFNS scores were categorised as good (WFNS 1-2) or poor (WFNS≥3) using existing definitions 14 . Since very few patients had grade 0 and grade 1 Fisher scale (i.e. 8 (<1%) and 18 (3.4%), respectively), the modified Fisher scale was categorised (2, 3, 4, lower scores indicating better prognosis. Exploratory single predictor models did not suggest any non-linear effect of these two indices of severity on any of the considered outcomes eAppendix 2. Sensitivity Analyses Multiple overlapping sources of information were used to provide robust information on key exposures, covariates and outcomes for this cohort. There was very minimal missing data, and while time and date stamps for ambulance phone calls, hospital arrivals and procedures times are likely to be highly accurate, the potentially least accurate time may be the patient reported date and time of aSAH symptom onset, which was captured from triage records but also ambulance notes. Our previous analyses within a subset of these cases for the state of Tasmania suggests that the pre-hospital time for most people with aSAH is very short. 1 Therefore, even if there is error in the pre-hospital time, this is unlikely to have affected our results greatly. To examine this, we have repeated our analyses on the time to treatment excluding the pre-hospital time. For these sensitivity analyses, the relevant time variable was defined as 'timeto-first hospitla arrival', defined as the time between the first hospital arrival and the time at which treatment was received (in hours). The distribution of this variable was very similar to the 'onset-to-treatment time' (Fig 1.), with the exception of less ' extreme' observations in the right tail. Bland-Altman plot (Fig 2.), and a Pearson's product moment correlation of 0.74 also suggest very high agreement between these two time measures.
Kernel density plot of the time variable used in the main analyses ('Symptom onset to treatment time') and the time variable used in the sensitivity analyses ('First hospital arrival to treatment time).
Bland-Altman plot of the differences between 'Symptom onset-to-treatment' time (i.e. original time variable) and 'First hospital arrival-to-treatment'time ( i.e. time variable used for sensitivity analyses) Hours Results of sensitivity analyses conducted with 'First Hospital arrival to treatment' time as an alternate exposure for the 'time delay' variable showed essentially the same patterns with all 3 outcomes, suggesting ~12.5 hours post-hospital arrival as the optimal window for aSAH intervention in terms of reducing the odds of being discharged to rehabilitation and maximizing survival at 1 year, with slightly different absolute values due to the shortened overall time to treatment by excluding pre-hospital times. Survival analyses still suggest ~12.3 hours as the treatment time that maximises survival at one year (i.e. similar to the original analysis the nonlinear cubic term is significant in the univariate model (p-val =0.02) but not for the adjusted model (p-val=0.09)) (Fig 3.).
We still were unable to detect an association between this new time variable an any of the considered post-stroke complications. However, the relationship between the odds of being discharged home vs. being discharged to a rehabilitation service and this 'First hospital arrival to treatment' time follows the same shape as the one observed with 'time from onset to treatment', and the non-linear effect is still highly significant even in adjusted models (Fig 4.), suggesting that the odds of being discharged to rehabilitation decrease steeply within the initial 12.5 hours and are the lowers at ~20 hours between time of first hospital presentation and treatment of the aneurysm.

Non-linear effects of 'onset-to-treatment' time and other covariates: Natural cubic splines
For survival models, the optimal degrees of freedom for non-parametric smoothing terms in multivariable Cox models with nonlinear covariate effects was chosen to minimize the corrected Akaike's Information Criterion (AICc) (Meira-Machado et al. 2013). For logistic regression models (i.e. discharge destination and secondary complications models), the nonlinearity of 'treatment delay' was assessed using the estimated degrees of freedom (edf) derived from exploratory univariate generalized additive models (GAM). The edf is a summary statistic of GAM models that reflects the degree of non-linearity of a curve (Wood 2006) (i.e. an edf equal to 1 is equivalent to a linear relationship, 1 < edf ≤ 2 is considered a weakly nonlinear relationship, and edf > 2 implies a highly non-linear relationship). For complications where GAMs with edf> 2, natural cubic splines of degree of freedom equal to rounded-up edf values were introduced in logistic models, to model non-linear effect of time on the probability of developing complications.
Using the approaches highlighted in the paragraph above, in addition to the potential non-linear effect of 'onset-to-treatment time' on each outcome, we investigated the non-linear effect of other chosen continuous covariates ( i.e. age, WFNS, and fisher scale) in single-predictors models. There was no indication of non-linear effect of age, WFNS and Fisher score on survival, discharge destination or complications. Because very few people had modified Fisher scale of grade 0 and 1, we decided to collapse them into 3 categories (2, 3, and 4) in adjusted models. In addition, for the purpose of model adjustment, since the effect of WFNS was not non-linear, we chose to classify WFNS as good grade (WFNS= 1-2) vs. poor grade (WFNS>3) using existing definitions 2 .

Significance of non-linear terms and extraction of 'optimal' onset-to-treatment time from non-linear models
For all models, the significance of the non-linear effects of time on the probability of being discharged home and of developing each complication, and on the Hazard Ratio of death at 12 months was assessed through sequential likelihood ratio testing (LRT tests, where the null hypothesis was that the relationship was linear).
Where the non-linear effects were significant, local extrema in the time-dependant log hazard ratio curves (i.e. survival models) and predicted probability curves (i.e. logistic. Regression models) were extracted using the first derivative method. These points represent the time of greatest survival, greatest discharge home and lowest odds of developing each secondary injury. For 12 month survival, the point in time that was associated with lowest hazard of death was used to derive relative death rate as a function of time-to-treatment, representing the risk of death at 12 month overtime relative to the time where the survival was maximised.

Covariate adjustment
Models were adjusted for age, gender, procedure type, severity, direct admission vs. transfer, ventriculostomy, haematoma evacuation and co-morbidities. To determine if these covariates modified the effect of 'onset-to-treatment' on outcomes we used product terms between the non-linear time to treatment (specified as a natural cubic splines) and covariates. A significant (p0.05) interaction term indicates that the effect of time-to-treatment on a given outcome was different according to levels of that covariate.

eResults. Supplementary Results
69.2% treated participants were female and median age at event was 55.5 (SD=14.5) years, and females were on average 4 years older than male patients. Amongst the 482 treated cases, more patients underwent treatment with endovascular coiling than surgical clipping (61.4% versus 38.6%, respectively, ( Table 1)). Patients who received clipping were on average 4 years younger than those who received endovascular coiling ( Table 1). Time delay in hours between the estimated time at symptom onset and the time at treatment was available for N=474 (98.3%) participants.
Median treatment delay in men was longer by ~3 hours on average than female, but the difference was non-significant ( Table 2).67 patients (13.9%) and 82 (17%) treated patients died within 1-month and 1year post event, respectively, with no sex-differences (Supplementary Table II). The proportion of patients who died of all-cause mortality at 1 month and 1-year post-event were not significantly different between the two treatment modalities (Supplementary Table I). There was no difference in aSAH severity indices between treatment procedure groups or between genders (Supplementary Table I and  Supplementary Table II The 93 untreated participants also experience earlier deaths, with average 'onset-to death' time of 3.3 (sd=4.1) days compared to 21 (sd=26.5) day in the treated group (see kernel density plot below) Kernel density plot of 'onset to death' time in the N=79 untreated patients who died (black line) and the N=82 treated patients who died (Blue line).
The following table presents characteristics of the N=93 untreated participants: Untreated patients were older than the 482 treated patients on average, had a more frequent history of hypertension, and their aSAH was more severe based on fisher scale and WFNS.  The fixed reference (12.25 hours (continuous blue line), was chosen as the "onset to treatment' time where the log Hazard ratio curve was minimized for both groups. (i.e. the risk of death at 12 months = 1 when treatment was received at 12.25 hours post symptom onset). Nonparametric estimates of the dependence of all cause death mortality at 12 months on the delay in receiving aSAH treatment in hours were restricted to the interval between 0 and 100 hours post aSAH symptom onset for each group. In model 2, HR for WFNS3 (vs. WFNS< 3) is 4.16 (95%CI: 2.61-6.52, p-val<0.01).  (Note: covariates were fixed for prediction purposes, so that probabilities shown are for a >55 years old female patient who received endovascular coiling and had a WFNS score <4, a modified fisher scale <3 at hospital admission)