emergency physicians are being increasingly pushed to make rushed decisions which, on the surface appear to be very reasonable and logical, but for a given patient, may in fact cause more harm than good.
administrators of hospitals are wanting rapid decision making and thus rapid ordering of investigations, so that patients can be either discharged home or admitted to a ward more rapidly to address the increasing problem of ED overcrowding and the new NEAT targets
patients are pushing doctors for ever more investigations and treatments, partly because they have read about it on the internet, and partly because they generally want answers and not uncertainty.
unfortunately, the net outcome of this can be:
more investigations and medications given at greater cost to the community
harm directly from the investigation, procedure or medication
inadvertent commencement on a hazardous clinical pathway that is not in the patient's best interest
more time spent in ED or in hospital addressing “red herrings”
potential harm from subsequent “need” to investigate incidental findings
suspected pulmonary embolism example
a previously well lady is referred to ED with “atypical” chest pain with no obvious risk factors for pulmonary embolism (PE)
a D-Dimer is ordered and it returns a positive result, even though the patient is otherwise “PERC” negative
the ED clinician feels obliged to further rule out PE by ordering a V/Q scan which comes back as indeterminant, and thus he then arranges a CTPA scan which gives a positive PE result.
the patient is then advised to be managed with warfarin for 12 months
now given that 1. a CTPA can be quite nasty for the patient, and 2. warfarin can definitely be nasty for the patient, have we really acted in the patient's best interest and what is the chance she really does NOT have a PE?
let's do some math
the pre-test probability is our gestalt feeling of the probability that this patient has the disease BEFORE we order any tests
clinicians can gain an idea of this by referring to past data
for a patient who is “PERC” negative, the pre-test probability of PE is < 1%
let's assume our D-Dimer test has a sens. 90% and specificity of 70% for low pre-test probability of DVT
using the equation positive likelihood ratio = sens/(1-spec) = 0.90/(1-0.70) = 3
now using the nomogram at bottom of this page, our pre-test probability of 1% becomes a post-test probability of around 3%
now, ignoring the V/Q test, let's just look at the CT-PA:
the PIOPED II study gave CT-PA sens 83% and spec 96% for PE
this gives a positive likelihood ratio = 0.83/(1-0.96) = 21
so given our pre-test probability of 3%, we use the nomogram again, and our post-test probability is 40%
that's right, our positive CT-PA result is more likely to be WRONG than RIGHT just because our pre-test probability was low!
we have just started our patient on 12 months of warfarin with a ~5% chance of major haemorrhage and there is a 60% chance that she has nothing wrong with her!!!
let's look at potential harm
radiation dose from CT-PA
renal toxicity and immune hypersensitivity reaction from CT-PA iv contrast
risk of need for repeat CT-PA study due to inadequate initial CT-PA (eg. contrast timing issues, artefacts)
risk of discovering an incidental finding which may initiate another nasty cycle of clinical investigations
risk of major bleed on warfarin
inconvenience factor of repeated INR measurements
let's look at the potential benefits of Rx
reduced risk of future massive PE
this is only present if she does indeed have a significant DVT or PE
the clinical significance of tiny PE's seen on latest high resolution CT scans is yet to be determined but may be minimal
so how do we quantitate who should be investigated?
the test threshold
this is the pre-test probability of a disease below which, it would be more hazardous to perform the test than not
in other words if the patient's pre-test probability is below the test threshold, we should not do the test and just accept that the patient WILL have a small risk of having the disease and potentially dying from it, but that this risk is LESS than the risk of performing the test with all its consequences.
the test threshold for a CT-PA is probably somewhere around 15% pre-test probability!
how should we decide if we should treat someone
the treatment threshold
this is the probability of the disease above which we feel comfortable that the benefits of treatment outweight the harms of treatment.
ie. probability of disease at the treatment threshold = R/(B + R), where:
R = risk of treating a person without disease
B = benefit of treating a person with disease
to understand this, we need to understand:
the probability the patient has got the disease
the harm the disease is likely to cause if left untreated
the benefits of the treatment
the hazards of the treatment
the possibility of being able to intervene later once the disease better declares itself
the patient's risk aversity and other characteristics
time and resources
for classical untreated non-massive pulmonary embolism, it is said that 30% will die in the next 12 months, although 20% will die from recurrent PE (the other 10% die from cancer, etc. which may have been a risk factor for the PE), while anticoagulation will reduce this mortality to 5%.
the 12-month mortality in the PIOPED study was 24% and these deaths were caused by cardiac disease, recurrent pulmonary embolism, infection, and cancer, not just PE!
let's assume that warfarin Rx does indeed reduce death from recurrent PE from 20% to 5%
if we don't anticoagulate our patient
overall, she has a 8% chance of death from recurrent PE in the next 12 months given the 40% probability of PE, and assuming the above maths is reflective of the current CT scanners, and she fails to return to ED if she develops further symptoms
if she is in the 60% group that did not have a PE then she will do well and not be exposed to harm from PE or warfarin
if she is in the 40% group and does have a PE then she has a 20% 12 month mortality from recurrent PE assuming, she ignores future signs of recurrent PE and does not get it managed at a second presentation
however, if instead of having a massive PE as her 1st recurrent PE, she has a non-massive PE which hopefully will bring her to medical care and appropriate Ix and Mx
if we treat her, she will have the major bleed risk of around 5% plus perhaps 2% mortality from PE's giving an overall nasty outcome of 7%, not much different from not treating her!
thus R = 0.05 (risk of major bleed), B = 0.15 (difference in mortality now 5% not 20%), thus the treatment threshold disease probability should be 5/15 or 33% although this ignores risk of CT-PA and quality of life factors of being on warfarin
one can see that pre-test probability can make an important component of our clinical decision making and yet this is often neglected
95% of cases of recurrence of PE will be non-massive giving the opportunity for re-presentation, diagnosis and commencement of anticoagulation, and secondary prevention
a question today with the earlier diagnosis of PE, and the higher recognition of smaller PE's in the absence of clinically obvious DVT, is how much benefit does anticoagulation really give in this population of patients with almost sub-clinical PE, and does the risks outweigh the benfits?
when we see a patient we have various decisions to make:
do we do nothing “first, do no harm” and adopt a wait and see approach, and patient advised to re-present if further developments
do we order a test - hopefully the pre-test probability is above our test threshold otherwise we may be doing harm
do we just treat the patient without the test - hopefully the pre-test probability is above our treatment threshold otherwise we may be doing harm
if the post-test probability is less than our treatment threshold but more than our testing threshold, then more tests should probably be done
the above is all good in theory, but life for the ED physician is not so simple, as most of the above figures are unknown for their CT scanner, D-Dimer, US service and their patient population not to mention other individual factors that may relate to the patient.
a nomogram for using Likelihood Ratios (LRs) to convert pre-test probabilities into post-test probabilities for diagnostic test results with a known LR
Drag the blue arrows to the values you have for your patient's pre-test probability (%) and the test result's LR, and read off the post test probability from the red arrow on the right.
we need to use the nomogram as clinician's think in terms of probabilities rather than odds which would be much easier:
post-test odds = pre-test odds x likelihood ratio
but odds = probability / (1 - probability) hence the need for the nomogram to avoid worrying about odds
and probability = odds / (1 + odds)
stats/do_no_harm.txt · Last modified: 2013/07/09 17:08 (external edit)