 |
 |
Dear Guylene Theriault (Ben, James and Neil)
I woke up this morning to read this fascinating discussion! To my mind, screening, diagnostic investigation, diagnosis confirmation and treatment decisions involve combinations of thresholds, individual risk assessments and personal preferences.
A ‘sensitive’ threshold can be set for a diagnostic screening test using a ROC curve, on one side of which no reasonable person would ever accept a treatment for that diagnosis because the probability of benefit would be so low. This screening test result would be deemed ‘negative’ and the patent reassured. However, for a ‘positive result’, one has to consider its differential diagnosis (the ‘specificity’ is of little help for this purpose). The numerical value of the ‘positive’ test result for an individual patient would determine the probabilities of the differential diagnoses for that patient (the probabilities would differ greatly for borderline and extreme test results for example).
The criterion for labeling a patient with a diagnosis should depend on the probability of some benefit at the cut-off (even in the absence of any possible harm) that should lead to a suggestion that the patient should consider its treatments. Thus if the probability of benefit were to be so low that no reasonable patient would accept a treatment, then all results at that level and below could be regarded as ‘negative’ and those above the cut-off as positive. (See my recent paper in JECM on arriving at diagnostic definitions in an evidence-based way: https://onlinelibrary.wiley.com/doi/abs/10.1111/jep.12981).
|
onlinelibrary.wiley.com
In order that evidence‐based medicine can prevent “too much medicine”, it has to provide evidence in support of “gold standard” findings for use as diagnostic criteria, on which the assessment of other diagnostic tests and the outcomes of randomized controlled
trials depend.
|
However, if the ‘diagnostic test result’ (or score based on a number of test results) were positive it would be important to estimate the probability of benefit in the light of that patient’s particular test result(s). The probabilities of various harms should also be estimated based on that patient’s various characteristics. The probable effects on that particular patient of the benefits and harms would also have to be considered. From this information the patient would be expected to assess the probability of personal overall well being with and without the treatment. (Perhaps this might be done by using Decision Analysis.) On this basis, the patient might be able to make a decision, perhaps by first specifying a personal threshold of a probable resultant benefit above which a decision would be made to go ahead.
The diagnosis appears to be important. For example, I would not have thought giving a lipid lowering drug to high CVS risk patient with a BP of 200/130 and a cholesterol/HDL ratio of 3.5 would have the same effect as giving a lipid lowering agent to someone with cholesterol/HDL ratio of 7.7 and a BP of 120/80. In the former case, the diagnosis would be ‘hypertension’ causing a high CVS risk whereas in the latter it would be ‘dyslipidaemia’ causing a high CVS risk. Many patients would have both or many more diagnoses of course. So the process involves diagnoses, test result thresholds, probability estimates based on a patient’s individual tests results, personal value judgements and personal thresholds for risk.
Added to this is the positive cost-benefit to the health service budget of reducing the number of CVS events. Treating all those with a slightly elevated risk would have a substantial effect because there are many more patients with slightly
elevated risks (as they are nearer the mean of the distribution) than those with a high risk (of which there are numerically fewer as they are further out in the tail). Perhaps we should ask patients to accept treatment for slightly elevated risks for the
common good as an act of altruism! They could stop the treatment if they actually suffered some adverse effect.
Small wonder that many patients ask “What would you do if you were in my position?”
Huw Llewelyn
Hi Neal - thanks for coming on board - I would agree cost effectiveness is obviously important from a societal perspective but I’m unaware that guidelines typically consider that when it comes to recommendations and certainly not typically when it comes to developing treatment thresholds. If they did they would have to have different thresholds for every different treatment (statins, PCSK9 inhibitors, metformin, thiazides, bisphosphonates, ACEI) because each one has different costs.
For instance, the PCSK9 inhibitors - monoclonal antibodies for lipids - have an effect on CVD but I’m not sure if they are cost effective given that in my mind they should be priced the same as statins - which will likely never happen.
My point is a more generic one - assuming a treatment can be afforded either by society or by an individual. Let’s assume the patient is Bill Gates and he could afford a PCSK9 inhibitor if he felt it would provide therapeutic value - although given him he might put the money into more global issues.
So cost aside, I see no reason to have treatment thresholds in guidelines. Why debate about the treatment thresholds for BP, glucose or cholesterol. Who cares what a guideline committee thinks is the proper threshold for treatment - they aren’t the ones who are going to take the treatment.
Why don’t we just have guidelines that provide tools that help clinicians say based on the best evidence we think you risk of a CVD or a fracture is ~ X% over the next 10 years - if we treat we can decrease it to ~Y% - the harms are as follows - if you want to take it great, if not great.
Now a patient might say to a clinician “Well what would you do” and if you know the evidence you could answer that truthfully.
My point is no one on the guideline committee can apriori ever know the specific values and preferences of the next patient you see.
However, I agree with you that if a country as a whole needs to figure out if they can afford a treatment a cost effective analysis could be useful - except that in my experience cost effective analyses can be subtly tweaked to come up with almost any answer you want.
Hope this makes sense.
James
On Dec 3, 2018, at 11:25 PM, Neal Maskrey <[log in to unmask]> wrote:
Most interesting again, Ben and James!
Don’t we initially need an assessment at the population level - not least in many countries to determine whether proposed interventions are cost effective (affordable) by the health system?
Only then can the options be presented in the guideline in the way James suggests? So a 2 stage process?
NealNeal Maskrey
But, again, how would you help people make decisions ...information on risk alone is not sufficient... ( however, what you are describe is a sort of the threshold exercise..)
Sent from my iPhone(Please excuse typos & brevity)Ben: Again - not sure why guidelines need to define a treatment threshold. Why not have the guideline provide tools to estimate risks and show how treatment changes that risk. In other words they give tools for shared-decision making not treatment thresholds (regardless of how these could be developed)
James
On Dec 3, 2018, at 7:52 PM, Djulbegovic, Benjamin <[log in to unmask]> wrote:
James, indeed the guidelines panel may define the threshold but that does not mean that they defined it correctly (as I said in my first reply to your message , the METHOD for determining the threshold is crucial but this is completely different issue )Ben
Sent from my iPhone(Please excuse typos & brevity)Thanks Ben - one totally needs info on harm - my example was strictly limited to the discussion of the risks and benefits. I think maybe we are talking at cross purposes??
When I say treatment thresholds I mean guidelines that say if your risk is >10% then you should treat - is that what you mean by thresholds or something else?
James
On Dec 3, 2018, at 7:40 PM, Djulbegovic, Benjamin <[log in to unmask]> wrote:
Yes, you do - you need not only benefit information ( risk goes down to 15%) but information on harms...Also, even in your example you implicitly defined the threshold at 15% ( but, as I just said, determination of more precise threshold will require information on harms and if harms are higher than benefits, the threshold will exceed 100% ie normatively no one should be treated despite benefits )Ben
Sent from my iPhone(Please excuse typos & brevity)Hope people find this interesting - an example of what could be said
Your risk of a clinical fracture in the next 10 years is ~20% (based on clinical features etc) - if you take treatment your risk goes down to ~15%
What further information would you need around risk and benefits? Still don't need a treatment threshold.
James
On Dec 3, 2018, at 6:27 PM, Djulbegovic, Benjamin <[log in to unmask]> wrote:
But, how do you anchor/frame discussion without (first) determining the thresholds? Theoretically, risk can range from 0 to 1 - where do you start?This is another important discussion that you have started James, and I would love to hear what other folks think. I should also add this is linked to the fundamental insights that all our policy decisions ( including guidelines) have to work within framework of “irreducible uncertainty, inevitable errors, unavoidable injustice”( Paul Ash and I had a piece along these lines a number of years ago:Ben
Sent from my iPhone(Please excuse typos & brevity)HI Ben - then what is the point of the threshold. Just present the risks and then potential benefits - do shared decision-making and leave it at that.
James
On Dec 3, 2018, at 6:07 PM, Djulbegovic, Benjamin <[log in to unmask]> wrote:
Hi James,My own view is that once you determine the thresholds , say by using decision -analytic methods, you elicit patient’s V&P to adjust the threshold , up or down ...ThanksBen
Sent from my iPhone(Please excuse typos & brevity)Hi Ben - if you have treatment thresholds (instead of discussion thresholds) how do you incorporate patient values and preferences.
James
On Dec 3, 2018, at 5:57 PM, Djulbegovic, Benjamin <[log in to unmask]> wrote:
I think it is a mistake to get away from action (treatment) thresholds; (decision analytic) threshold helps link the quality evidence (which exist on the continuum of credibility) with the decision-making (which is a categorical exercise).
Without threshold, there is no way to help people make decisions ( NB a completely separate issue is how and which model we should use to determine threshold, but determination of the threshold is of fundamental importance for any decision-making under risk/uncertainty)
Ben
[Attention: This email came from an external source. Do not open attachments or click on links from unknown senders or unexpected emails.]
Hi Guylene - really love your approach - we did this with our lipid guideline http://www.cfp.ca/content/61/10/857 where we had absolute risks (determined by using the cvdcalculator.org) that defined when we would have treatment discussions - in other words there were NO treatment thresholds just discussion thresholds. This ideology, if you will, changes the entire focus of a guideline.
You are correct that there is no such thing as overdiagnosis or false positive/false negative when it comes to risk reduction.
The only concept of overdiagnosis would be if you have a person on a treatment for risk (BP, statin, bisphosphonate) yet if they knew or had been given the absolute benefit they wouldn’t have chosen to take the treatment given the benefits and harms.
Hope that makes sense.
James
HI all,
I need your help
I am on the Canadian Task Force and we are looking at doing a guideline on screening to prevent fragility fractureWe wish to get away from a treatment threshold in our recommandations. If we use a threshold if would be for shared decision making not for treatment.We also wish to get away from a disease label (osteoporosis) and use the risk level (% risk) as the outcome to determine actions following screening.We encounter diverging views on certain definitions.
What would these be in that context? Our real question is the last one about overdiagnosis.
False positive and false negative: For us there cannot be a false positive or false positive.If you are labeled at higher risk (example 30% risk of a fracture) there cannot be a false positive because not having a fracture is a possible issue included in the risk.
If you are labeled at lower risk (example 5% risk of fracture) there cannot be a false negative because having a fracture is a possible issue included in the risk.
But what is overdiagnosis in that context? Can we even talk about overdiagnosis?Considering that osteoporosis even if it has a name (diagnostic name) is essentially a risk assessment.
I am eager for your thoughts
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1
*SECURITY/CONFIDENTIALITY WARNING:
This message and any attachments are intended solely for the individual or entity to which they are addressed. This communication may contain information that is privileged, confidential, or exempt from disclosure under applicable law (e.g., personal health information, research data, financial information). Because this e-mail has been sent without encryption, individuals other than the intended recipient may be able to view the information, forward it to others or tamper with the information without the knowledge or consent of the sender. If you are not the intended recipient, or the employee or person responsible for delivering the message to the intended recipient, any dissemination, distribution or copying of the communication is strictly prohibited. If you received the communication in error, please notify the sender immediately by replying to this message and deleting the message and any accompanying files from your system. If, due to the security risks, you do not wish to receive further communications via e-mail, please reply to this message and inform the sender that you do not wish to receive further e-mail from the sender. (LCP301)
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1
To unsubscribe from the EVIDENCE-BASED-HEALTH list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=EVIDENCE-BASED-HEALTH&A=1