Skip the LP? CT only for suspected Subarachnoid Haemorrhage

Clinical Question: Does a normal head CT scan within 6 hours of headache onset rule out subarachnoid haemorrhage?

Title: Sensitivity of Early Brain Computed Tomography to Exclude Aneurysmal Subarachnoid Haemorrhage. A Systematic Review and Meta Analysis.

Authors: Nicole M. Dubosh et al.

Published: Stroke, March 2016.

I know not very recent! However it is big on social media and has not made it into established or ad hoc clinical practice locally.

Summary: This was a systematic review and meta-analysis to determine the accuracy of a CT only strategy of investigating suspected SAH with 3rd generation CT performed within 6hrs of headache onset.

3rd generation scanners relate to those >64slices (PRUH has these)

Original articles were selected to answer this question excluding trauma, patients <15years age, older scanners and CT >6hrs of headache onset.

The completed study selection process yielded 5 articles from an original search strategy yield of 882.

One of the studies was prospective, the others retrospective.

Four of the studies were diagnostic accuracy studies, the other a case-control study.

Studies were assessed for quality and bias risk and findings of heterogeneity between studies was noted. The gold standard test for SAH (CT positive or LP positive) was applied unevenly across all the studies.

Estimations were made on 2 of the studies as they only reported on negative CT results.

Several of the studies had different interpretations of what a positive or negative result was so the authors produced pooled results of all the 5 studies based on a worst case scenario.

Pooled figures were Sensitivity was 0.987, Specificity of 0.999 and Negative LR of 0.010.

3 studies with 2 x 2 data were used for a sensitivity analysis to ensure robustness of the main findings.

These gave a pooled incidence of SAH of 19.1%. Sensitivity was 0.986, Specificity of 0.996.  The incidence of missed SAH in this population was 0.87 per 1000.

The Authors conclude that a negative 3rd generation CT brain scan performed within 6 hours of sudden headache onset is sufficient to rule out SAH in adult patients with a normal GCS who are neurologically intact.

Strengths, Weaknesses and Clinical Relevance

This study outlined clearly the aims with a focused question and specific population.

The search for articles appears to have been comprehensive enough (882 articles).  Selection of the articles was robust also with 2 independent investigators screening the article abstracts to select articles for full review (40). Full text review then yielded the final 5 articles. Interobserver agreement was good at 87.5%.

As with all systematic reviews the crucial element is the strength of the individual papers that are in the analysis.  There is considerable heterogeneity of the studies used, and the majority are retrospective.  2 of the papers did not report on positive CT numbers so 2×2 tables to discern accuracy had to be drawn from estimations from the Perry et al paper. The authors did use “worst case scenario” with one of the papers (Marks) where the 11 missed cases and still came up with an excellent sensitivity and Negative LR.

With only 5 papers in this review, I felt it prudent to read through them all.  A brief summary is below –

Perry et al, 2011 –

The only prospective study in the meta-analysis. Multicentre cohort study based in Canadian ED with 3132 pt’s enrolled. Patients were presenting with acute headache maximal <1hr and CT timing was <6hrs from onset of headache. Scans reported by neuro or general radiologist.

Not every patient got the gold standard test of LP (49.4%).  However, follow up was robust – 6 months telephone or if lost (2%) neurosurgical centre/coroners enquiries across whole Ontario state.

CT within 6hrs yielded SAH incidence of 12.7% and 100% for all of Sens, Spec, PPV, NPV.

Backes et al, 2012 –

Retrospective single centre study in Netherlands looking at allcomers with suspected SAH.  137 pt’s had CT scan within 6hrs. All scans read by a neuroradiologist. Of the scans <6hrs, 50% of those were positive although 11% of the positives were perimesencephalic bleeds (venous bleeds with no intervention needed).

All negative CT patients underwent LP.  On those CT negative, 1 patient had positive bleed due to a cervical AVM but this patient had neck pain/stiffness with no headache.  Excluding the 2 patients with no headache yielded 100% again for all of Sens, Spec, PPV, NPV.

Mark et al, 2013 –

USA Retrospective case-control study with 2 goals looking at CT <6hrs as well as a headache decision rule.  Aimed to enrol 70pts but only enrolled 55 cases. Enrollment was based on confirmed SAH, not acute headache.   11 out of 55 patients with a negative CT within 6hrs had a supposed SAH.  However, the LP RBC counts and xanthochromia differ in half of the 11 cases. In addition, significant data was missing and  sudden onset headache made up only 75% of the cases.

Stewart et al, 2014 –

UK based single centre (teaching hospital) retrospective study looking at accuracy of CT within 12hrs of onset in cohort investigated for SAH.  Subset of 65 patients with onset <6hrs looked at. Sens was 100% in this cohort. Authors did not recommend the strategy as cohort small.

Blok et al 2015 –

A multicentre study in non-academic hospitals (i.e. like PRUH). Retrospective review of CT negative patients but also including those CT reports later changed to positive after LP (2nd look and amendment).  760 pts total and 52 had positive LP.  One CT amended.  Of 52 LP positives, 8 had aneurysms on further scans but rupture deemed unlikely.  43 no aneurysm or scan not done (follow up at least 22months).  NPV 99.9% stated.


As you can see the papers have significant differences between them.  Most of the papers to tend to agree on the sensitivity of CT undertaken within 6hrs of onset. Only the Perry paper was prospective and one could argue that the whole meta-analysis is carried by this paper with its design and large data set.

The Mark et al study does stand out as having significant false negatives although this paper had issues with study design and interpretation of the gold standard.

The pooled incidence sounds very high but it is worth noting that those early presenters would be expected to have a higher incidence due the severity of their symptoms.

So, should we expect better data before we come to a definitive conclusion?  A large, multicentre diagnostic study with all patients undergoing the gold standard of LP after negative CT is very unlikely to happen.  An LP is painful with significant complications of post LP headache and false positive results leading to downstream testing and even treatment.

This review is probably as good as the evidence base is going to get I feel.

So what would I do now?

I would definitely consider a rule out after negative CT within 6hrs of thunderclap headache onset.  The patient would have to be well appearing patient with no neuro signs and with SAH being the only likely diagnosis.  Anyone still clutching their head, vomiting, altered GCS etc. I wouldn’t.  Discharging the well patient after a clean CT scan seems reasonable and better for the patient and an already stressed emergency bed base. No rule or pathway has a 0% miss rate.  However <1 in 1000 is arguably better than other EM testing strategies for high risk presentations such as PE, chest pain etc.

However, for those of you that are juniors the following really needs to be stressed –

  • Please do not do this without senior discussion and ideally face to face review.
  • This is only for suspected aneurysmal SAH. There are other sinister causes of headache such as venous thrombosis, meningitis etc. If the differential is wider than SAH alone do not use this guidance.
  • This is not yet widespread practice and is not endorsed by any local or national guideline or any College/Society. As such this is at the fringes of what is deemed acceptable.
  • If you do decide LP is not necessary, make sure you discuss/share the decision with the patient.

So, what to the rest of you think?



Observational pain relief tool for cognitively impaired older people

Clinical Question: Does the use of an observational pain assessment tool reduce the time to analgesia for (non-verbal) older people with advanced dementia presenting to the ED with a suspected long bone fracture?

Title of the Paper: Can an observational pain assessment tool improve time to analgesia for cognitively impaired older persons? A cluster randomised controlled trial

Authors: Fry et al.

Published: EMJ, January 2018

Population: Adults >65 with a clinically suspected long bone fracture, and cognitive impairment (defined as score <4 on Six-Item Screening (SIS)), with exclusions, in 8 EDs in Sydney Australia

Intervention: Pain Assessment in Advanced Dementia (PAINAD) is an observational pain assessment tool, which scores 0-2 across 5 domains, giving a total pain score /10

Comparator: “standard methods of pain assessments”, such as a verbal rating scale or visual analogue scale

Outcome: time from arrival in ED to first dose of analgesia

Summary: The authors note that the most common form of ED pain assessment is patient self-report; however, for patients who are unable to communicate pain intensity verbally, clinicians often rely on subjective assessment, which can be unreliable. Observational pain scoring is widely used in paediatric EM. PAINAD is designed to give a pain score /10 for people with advanced dementia; it is is quick and easy (MDCalc!) but has not been formally trialled in ED.

This is a multi-centre (single city) study, which aimed to investigate whether using an observational pain score would improve time to first (any) analgesia for elderly people with cognitive decline, compared to standard practice. It is cluster randomised by 8 different EDs in the Sydney region. All sites were asked to screen their potentially eligible patients (>65y, suspected long bone fracture) with the SIS scale to identify those with cognitive impairment. At the 4 intervention EDs, those identified to have cognitive impairment would also be pain scored using the observational PAINAD score. The analysis of a cluster randomised trial compares all the people in the intervention clusters to all the people in the control clusters, regardless of whether the intervention was actually applied to all people in the intervention cluster.

The main analysis conducted was intention-to-treat, which means that all the eligible patients in the intervention clusters were compared to all the eligible patients in the control clusters, regardless of whether the nurse or clinician had time to use the SIS score, and the PAINAD score. There were 602 eligible patients, and after adjusting for age, fracture type, arrival mode, and triage category, they were unable to demonstrate a difference in time to analgesia between the two groups (HR 0.97, 95% CI 0.80 – 1.17, p 0.74).

A second sensitivity analysis was conducted including only patients in the study who had a documented SIS score. There were a total of 271 patients with a SIS score <4 across all clusters. 87% of those in the intervention cluster had been PAINAD scored. The sensitivity analysis found that the intervention sites delivered analgesia median 13 mins earlier than control sites, but this was not statistically significant (90 vs 103 minutes, p 0.62).


Strengths, weaknesses, and clinical relevance

This is a recent paper, published online in the middle of last year. It seems to demonstrate a reliable and replicable way of trialling an intervention in a busy and complex environment such as the ED.

There are notable differences between the patients at the intervention clusters compared to the control clusters, and although these are acknowledged and attempts are made to adjust for them, the differences are not fully explained.

It is interesting that time to analgesia has been chosen as an outcome when testing a pain scoring system. As the discussion notes, that there was no difference between clusters suggests that time to analgesia is affected by so much more than just being aware of a patient’s pain score, and suggests other system changes are needed to reach national targets of time to analgesia.

I would like to know wether using an observational pain score provides more appropriate analgesia to patients with advanced dementia. This study looked only at time to first analgesia for suspected long bone fractures, but has not provided details on whether this is, for example, paracetamol or morphine.

Are there any other observational pain scores at use in UK EDs?

Would the >65 years PRUH ED population benefit from an observational pain score?

If so, how could this be implemented?

IV or Oral Paracetamol for Pain

Clinical Question: Used with opioids, is IV paracetamol better than oral paracetamol for managing moderate to severe pain in the Emergency Department?

Title: Intravenous versus oral paracetamol for acute pain in adults in the emergency department setting: a prospective, double-blind, double-dummy, randomised controlled trial

Authors: Furyk, J. et al.

Published: EMJ Dec 2017 (online)

Population: Adults >18 years in the ED, with acute pain >40mm on a Visual Analogue Scale (VAS) (i.e. >4/10) 5 minutes after receiving an IV opioid

Interventions: all patients received an oral and IV substance. Randomised 1:1, half received active oral paracetamol and dummy IV, and half received active IV paracetamol and dummy oral.

Outcome: reduction in pain score on the VAS after 30 minutes

Summary: This is a single centre study, focussing on acute pain in the ED. The authors believed that the assumed benefits of IV paracetamol over oral, such as faster onset of action, greater analgesic efficacy, and reduced opioid use, required evidence of clinical superiority to justify the higher cost, time, and risk of possible complications.

Patients, nurses and doctors were blinded to allocation during treatment and data analysis. Pharmacy staff prepared each patient’s IV and oral preparations as required, using a sealed allocation envelope each time.

The sample size calculation was based on comparing the difference between two means. The authors were looking to detect a difference of 15mm between the VAS means of the two study groups, which was chosen as ‘a compromise between the commonly cited ‘minimum clinically significant’ difference of 13mm and 20mm’.

The final modified intention-to-treat (mITT) analysis included 87 patients, 47 in the IV group and 40 in oral group. They report that after 30 minutes the mean IV group VAS score was 51.5mm, and 54.2mm for the oral group. The difference in pain score was -2.6mm, with 95% CI -13.2 to 7.9 and p=0.62. The authors conclude that IV paracetamol is not superior to oral paracetamol in patients with moderate to severe pain.

Strengths, weaknesses, and clinical relevance

This paper has rapidly generated discussion on blogs (e.g. FOAMShED and EM Lit of Note), podcasts, twitter, and amongst colleagues.

There have been many RCTs comparing oral to IV paracetamol in the past, and 2015 brought two systematic reviews of the evidence. Jibril et al. and Atkinson et al. both concluded there was no efficacy, pharmacokinetic, or safety benefit to IV dosing if the patient can take oral. However, the authors of this month’s paper state this is the first RCT in an ED environment, when managing acute pain is a large part of the workload.

In an ideal setting pain is assessed and managed step-wise up the WHO Analgesic Ladder. There is merit in evaluating fast acting IV preparations of ‘pre-opiate’ drugs for the management of acute pain in ED. It is unfortunate, then, that all patients included in this paracetamol trial have already had opiates, in the previous 5 minutes. The clinical question this study has answered is not the one I have been wondering.

There are a number of places where the analysis might have gone awry. Firstly, this study has been approached statistically as a study of the difference between two means, and the sample size calculation of just 44 per group reflects that. However, it is described as a ‘randomised controlled trial’, and uses RCT methodology. Given that established data finds limited difference in efficacy between oral and IV paracetamol, I believe a RCT sample size calculation would would require an enormous number of patients per group to establish statistical difference between the two groups.

Secondly, the primary outcome is stated to be the reduction in pain score from baseline. Baseline pain scores are reported as 65.0mm for the IV group and 71.3mm for the oral group. After 30 minutes the scores are 51.5mm and 54.2mm respectively. Further analysis compares the 30 minute pain scores of 51.5 and 54.2 rather than the amount of reduction from their baseline. (See Figure 2 for score reduction over time and note the different starting points.)

Unreported, mean pain reduction from baseline after 30 minutes for those receiving IV paracetamol was 13.5mm, and for oral paracetamol was 17.1mm. The mean difference is -3.6 and there remains no statistical significance between them (p=0.50).

Would the results be different in analgesic-naive acute pain?

Does the question of IV versus oral paracetamol need another clinical trial?(!!)

Could or should this study change practice here?

Prednisolone or Dexamethasone in Acute Adult Asthma

Clinical Question: Is a single dose of dexamethasone not inferior to 5 days of prednisolone to treat mild to moderate asthma exacerbation in adults?

Title of the Paper: A Randomized Controlled Noninferiority Trial of Single Dose of Oral Dexamethasone Versus 5 Days of Oral Prednisolone in Acute Adult Asthma

Authors: Rehrer, M.W. et al

Published: Annals of Emergency Medicine, Nov 2016

Population: Adults 18-55yrs requiring nebulisers in ED but able to be discharged home (with exclusions)

Intervention: 12mg dexamethasone in ED and 4 additional placebo capsules

Comparator: 60mg prednisolone OD for 5 days

Outcome: Relapse within 14 days

Summary: This is a single centre study. The authors hypothesised that because there is an association between poor adherence and poor outcomes in asthma, a dose of dexamethasone, known to have a duration of action of 72 hours, could improve outcomes compared to normal standard of care, which requires 5 days of adherence.

The ED staff, patients, and analysers were all blinded to treatment allocation. Anyone who received a nebuliser was screened for possible eligibility. Relapse was detected by telephone follow-up at 2 weeks.

The authors chose a noninferiority margin on 8%. This is because national records and previous studies find a 16% relapse rate for normal standard of care, and a 33% relapse rate for placebo. They state that 50% of the difference between these two relapse rates is 8%. Choosing 8% as a non-inferiority margin also ensures the intervention is >50% superior to placebo.

This study included 376 adults, 173 in dexamethasone group and 203 in prednisolone controls. Relapse occurred for 12.1% taking dexamethasone, and 9.8% taking prednisolone. 95% confidence interval -4.1% to 8.6%. They state that because this is very slightly greater than their pre-planned noninferiority margin of 8%, a single dose of dexamethasone does not demonstrate noninferiority to prednisolone for 5 days.

Strengths, weaknesses, and clinical relevance

I think the paper is very clearly written and methods thoroughly described.

This study relies on noninferiority methodology. They are testing that the new intervention is ‘not unacceptably worse’ than current care, by an arbitrarily chosen margin. By falling outside of their planned margin, the authors can only say that statistically dexamethasone is “not ‘not unacceptably worse'” than prednisolone.

The authors do not describe their local population. It could be presumed that in a context where patients are  less likely to engage with counting out 8 x 5mg prednisolone tablets for 5 days, there would be higher rate of relapse in the control group.

The confidence interval crosses 0 so ignoring the non-inferiority margin it could be considered that there is statistically no difference between the two groups.

The sample size was recalculated when the interim analysis found an overall relapse rate of 11%. The authors do not offer a breakdown of this between the two groups. They did not go on to redefine their noninferiority margin based on a greater difference between this relapse rate and historical placebo relapse rate, because this would have meant changing their outcomes mid-trial; however, should they have done so the noninferiority margin would have been ((33-11)/2=) 11% which would have meant their overall result of 8.6% would have demonstrated noninferiority.

Could the results be different in a UK context prescribing 30-40mg prednisolone instead of 60mg OD?

Is this relevant for the COPD population as well?

Could or should this study change practice here?