Missing things: Addisons

Recently missed an Addisonian crisis.  Here’s a brief summary of the case.

61 y o female attended the ED after seeing her GP with vaginal bleeding.  GP sent the patient to the ED for an inappropriate official reason (vaginal bleeding in a post menopausal lady unless torrential should be managed as a two week wait), but the lady looked unwell, and I think that was why the GP fabricated a reason.

This lady had a PMH of psychotic illness, rectal prolapse, and that was pretty much it.  She was seen with a family member who looked after her.  The history was vague.  She wouldn’t talk much, she said she had felt unwell for a few days, tired, and lethargic.  She had noticed the bleeding intermittently for the last few months.  She had a grey waxy sheen to her skin, and her hair was matted and unkempt.  Was this because of her mental health problem?

Systemic examination was unremarkable apart from a large rectal prolapse which I reduced, it was probably also the source of the blood.  Her results were phoned through:

Na 129, K6.9 Urea 17.2 Creat 152.  She had an elevated neutrophilia.  ECG showed sinus tachy with what were on reflection tented t-waves.  Her VBG showed a normal pH, slightly reduced HCO3- 15, and a BE of -5.

She was dry.  She had a high K+.  I started her on insulin/dex.  Started a septic screen and began rehydrating her with normal saline.  I did not giver her steroids but the medical SpR did about 1 hour after I saw her.

So why didn’t I think of Addisons? (apart from me being a bit thick).

The patient arrived in the ED by ambulance.  The history was confused.  The reason for attendance was really ‘unwell ?cause’ that the GP had recognised but then put more emphasis on the symptom of vaginal bleeding.  I think this led me down a strange path.  Possibly slowing my decision making with an anchoring bias.  I’d found a pathology with the rectal prolapse, and solved it (premature closure).  Did I think that this could explain the GI symptoms, AKI and dehydration?  Possibly, given the mental health problems and poor communication, this lady was likely to present late enough to cause these things if she’d avoided nutrition and fluid because toileting was painful.

Did I also fall prey to confirmation bias?  I think so.  I had a theory, I’d found something to confirm this.  I think I stopped looking for or considering differentials, despite the fact that I was thinking renal failure in someone with a normal venous pH, and only slightly disturbed bicarb.  .  I think there was also an element of fundamental attribution error (stereotyping), I probably placed too much emphasis on this woman’s serious mental health problem, as I tried to fit my theory together.

I may have rejected a new diagnosis of Addisons as a form of availability bias – I’ve never diagnosed primary Addisons, which means it’s not an explanation I often examine or reach for.  This is also called Zebra Retreat.  My normal cue for thinking about Hypoadrenalism is if people are on long term steroids and this lady wasn’t.

So how do we mitigate against cognitive errors?

Well there’s a good article on that here, along with a list of other cognitive errors you can make here.  The biggest step is being aware that they can happen, and what they are.

One way I normally work is I try to write down my differential, weighing the relative merits of each one as best I can.  However the time when I do this is when I’m less busy, and I know that when I am juggling a few Resus patients simultaneously I neglect my documentation, but those are probably exactly the times it needs to be done!  Especially as writing my working out is so important to my ‘cognitive process’.

When I miss something I also like to read up on the thing I missed, as well as trying to understand why I missed it.  Sometimes I write a reflection, sometimes (more often) I chat to my wife about it (she’s an ICU doc).  I just wish more time was allowed for this kind of thing at work.

Some notes on Addisons.

Incidence 0.8 per 100 000 of population.  More common in women than men.

The commonest cause of glucocorticoid deficiency is Iatrogenic, this technically isn’t Addisons.  Consider replacement if patients have been on 7.5mg pred/day (30mg Hydrocortisone/day) for longer than one month.  The mechanism here as I’m sure you know is suppression of the adrenal axis by negative feedback, causing adrenal atrophy.

Classically the causes are usually infection (TB, HIV), trauma, anything causing adrenal haemorrhage, and malignancy (mets like the adrenal glands as they are highly vascular).

Clinical Features:

Vague, frustrating and insidious.  Tiredness is the commonest symptom.  Skin hyperpigmentation (secondary to excess ACTH) looks quite subtle to me.  It’s more pronounced in areas of increased friction like palms, knuckles and scars.  GI symptoms can be severe enough to mimic an acute abdomen.

ED presentations are mostly crises, so you get hypotension, alongside the hyperkalaemia and hyponatraemia.  You can also get hypoglycemia.  On reflection I wonder how many potential mild cases we send back to GPs.

In summary

Mistakes and errors are common, especially with rare conditions and vague presentations. The ED environment sometimes feels like it’s designed to cause and not mitigate cognitive error, with multiple distractions and heavy time pressure.  We all need to develop our own strategies for limiting our own cognitive errors as we develop as clinicians.


Don’t miss an Addisonian crisis. [Low Na, High K –> Give steroids]



Suspicious Histories

Ask any medical student (when they are sober) the clinical features of an Acute Coronary Syndrome they’ll look at you funny for a bit, then talk of  severe, left sided chest pain radiating through to the arm, they might add in diaphoresis, or pulmonary oedema.  If you’ve got a real nugget that might carry on talking about risk factors, relief from GTN and ‘pain like previous MI’ making a cardiac cause more likely.

You’d probably tell them they were right.

I remember hearing the first description of a chest pain history in a big lecture theatre as a first year medical student, and it was pretty much that.

We’ve always been taught that Histories Are Important, we are taught these by rote, by repetition and by  exposure to lots of real world cases.  They are the narrative that gives us the pathology our patients experience. So it’s implicit that the clinical features of those histories are equally important, and give us a sense of the probability of the disease in question.  I doubt I’m alone in labouring to commit ‘red flag symptoms’ to memory for exams, or general day to day work.

Now in my last post I got in a bit of a tiz about the HEART score, and about how the definition of a suspicious history was a little vague.  Now the history component of the HEART score is just worth 0,1, or 2 points out of 10.  So if histories were as important as my old professor said why does the component only make up a maximum of 20% of the score?

Thats because the classical chest pain history is not a very good predictor of ACS.

Let that sink in.  You have been taught that histories are important, they’re the key to finding the pathology.  Sometimes when the diagnosis is essentially a clinical one (eg appendicitis) that still holds true.  However when you have a good quality gold-standard test (eg a Troponin) things become different.  You can test the history, and the chest pain history is not very good at all.

In 2001 Prof Steve Goodacre published a study looking at the clinical features of people with low risk chest pain (normal ECG, no initial trop rise), seeing if any particular features had a predictive effect.  Previous research up to this point suggested left arm pain, diaphoresis, a 3rd heart sound and hypotension predicted acute myocardial infarction, but these studies were done prior to the advent of Troponin.

He conducted a big single centre prospective trial of patients over 25 with chest pain, no known CAD, and a normal initial workup.    He followed his patients up over 6 months looking for AMI and MACE.  He had a ACS rate of 9.1% in his cohort of 893 patients.

If I were to ask you ‘where do most people who have an NSTEMI complain there pain is coming from?’ You’d probably point somewhere over the left side of your chest.

Site Odds Ratio CI P
Both arms 6.0 2.8-12.8 >0.0001
Shoulder 3.4 1.5-7.8 0.004
Right arm 2.5 0.5-11.9 0.24
Throat 2.0 0.4-9.4 0.37
Left arm 1.7 0.9-3.1 0.084
Neck 0.8 0.3-2.5 0.74

Taken from table 4 – Univariate Analysis of Predictors of ACS (Goodacre 2001)

Would you suggest pain in both arms?   What about pain in the right arm being more suggestive of AMI than pain in the left?  That’s what Prof Goodacre found, though his confidence intervals are wide, and some of his findings were not statistically significant.

Now if we look at some his multivariate analysis results we can see that he only really found 3 good predictors.  Pain in both arms, exertional pain and a tender chest wall (this was a good negative predictor).  How many people have you admitted because ‘they had pain at rest’?

Pain in both arms 4.07 4.9 1.3-19.4 0.02
Exertional pain 2.35 3.3 1.3-8.4 0.014
Tender chest wall (as negative predictor) 0.3 0.2 0.05-0.97 0.045

He concludes

“We also found no evidence that the site of pain, the presence of nausea, vomiting, or diaphoresis, and the duration of the pain were at all useful in predicting whether the pain was likely to be cardiac or not”

Despite him having a pretty big sample of low risk, non cardiac chest pain he probably still didn’t have a big enough sample to look at what he wanted to look at.  This dataset was taken from another study primarily looking at the feasibility of a rapid chest pain assessment service.  I guess thats why theres no sample size calculation.

Lets look at my next study.  I love this study, it’s come out of manchester.  Prof Body included everyone over 25 who attended with suspected cardiac chest pain, treating clinicians completed a standard form with tick boxes for clinical features.  All patients had a 12 hour troponin and follow up at 48 hours, 30 days and 6 months.  His primary outcome measure was AMI and secondary outcome was MACE at 6 months.  So we’ve got a slightly different risk profile to the Goodacre study.

They analysed the results from 796 patients, and got 100% follow up.  I feel they had a slightly higher AMI rate of 18.6% (n=148) for initial presentation.  At 6 months they had a MACE rate of 22.9% (n=179).

The Odds ratios for AMI for individual clinical features are below.

bodygraph A few things to point out.  Sweating and abdominal pain seemed to be good predictors of AMI, as did reported N+V (which was not predictive in Prof Goodacres cohort).  Pain location that was in both arms predicted AMI more than pain in the left (and pain in the right was slightly better than left).    My favourite one though is that if the pain was described as ‘like their previous MI’ patients were less likely to have AMI.

Body didn’t find any excellent predictors of AMI.  His best ones were observed sweating, both arm/shoulder pain, and hypotension.

Predictor Sensitivity Specificity PPV NPV LR+ LR-
Sweating Observed 36.5 (28.7-44.8) 94.3 (92.2 – 96.0) 59.3 (48.5 – 69.5) 86.7 (83.9 – 89.1) 6.39 (4.38 – 9.33) 0.67 (0.28 – ).46)
Reported Vomiting 16.2 (10.7 – 23.2) 94.8 (92.7 – 96.3) 41.4 (28.6 – 55.1) 83.2 (80.3 – 85.8) 3.09 (1.89 – 5.05) 0.88 (0.82 – 0.95)
Pain in both shoulders 13.5 (8.5 – 20.1) 94.8 (92.7 – 96.3) 37.0 (24.3 – 51.3) 82.8 (79.8-85.4) 2.58 (1.53-4.34) 0.91 (0.85-0.96)
Same as previous ischemia 22.3 (15.9-29.9) 69.4 (65.7 – 73.0) 14.3 (10-19.5) 79.7 (76.1 – 82.3) 0.73 (0.53-1.01) 1.12 (1.01-1.24)
Right arm/shoulder pain 18.9 (13.0 – 26.2) 91.8 (89.4 – 93.8) 34.6 (24.4 -46.0) 83.2 (80.3 – 85.9) 2.31 (1.52 – 3.53) 0.88 (0.81 – 0.96)

Taken from Table 3 of Body (2010)  Characteristics of each predictive clinical feature as a diagnostic test for AMI

So in two good quality studies of chest pain characteristics we get right arm pain more predictive of AMI than left arm pain.  We can’t with any degree of accuracy state that someones epigastric burning pain isn’t cardiac in origin, and a classical heavy central pain radiating to the jaw or arm isn’t predictive of AMI or 6 month MACE at all.

Caveats.  You’ve got lots of hypothesis tests going in both of these studies.  Maybe too many, given the sample size and the rarity of the condition we are looking for.  A bigger sample would increase the accuracy certainly.

So should we stop talking to people about their chest pain?  Of course not, however we need to be very careful at excluding ACS based on the chest pain history alone.  Using something like the HEART score is probably more accurate, safer for your patient and safer for your registration.