So, a challenge we have as it relates to observable data is that we can define a cohort definition with a exit strategy that ‘right-censors’ after we see two successive BP without treatment, but if a data source doesn’t have BP measures, we may not observe when a person leaves the disease state (I don’t think that not seeing drugs or conditions would be sufficient to declare that the hypertension as ended). This could represent a form of ‘cohort end date misspecification’ error.
It may be wrong for other more acute phenotypes, but essential hypertension we may not need/be able to get precision in start date. Most clinicians will not start treatment on the date of diagnosis (unless there is hypertensive urgency or crisis).
Sure. But if we are observing ppl on anti hypertensive medications they are not new essential hypertension.
That’s the debate. Do we need explicit indicators of termination of a state to define cohort end date, or do we use implicit indicators.
In observational data, absence of information can be because of missing information.
We don’t know
Great thread with implications for many chronic conditions with intermittent acute exacerbations. All comes back to what is the question. I know we are all striving to have one clinical definition for each entity. However, I still think defining a particular set of conditions that define the onset of a clinical condition is different from use case to adjust for a clinical condition as a potential confounder, also may differ from use case for exclusion of an individual as the ultimate form of potential confounder adjustment. Once I have HTN I forever have a history of hypertension. The database available may or may not know the day I first experienced hypertension.
I regularly check in with my Juxtaglomerular Apparatus on my blood pressure and push a non-standard OMOP code to the CDM.
If we observe a censoring event, then we do know (seeing SBP <130 is a nice hard marker). so, i would favor including that in an HTN definition, at least for purposes of evaluating its impact. but censoring requires observing an event, and absence of an event (such as ‘no hypertension code’) isn’t necessarily evidence of anything.
That’s interesting. How about right censoring an acute event.
I would disagree here, no news maybe good news in health care.
I have surgery, I have surgical scar, I get discharged, I go home and don’t have any subsequent surgical care.
Then is my surgical scar healed? There is no evidence in the data that my scar has healed
Right now, I don’t think we know how to define end of essential hypertension. I skipped over that part in when i wrote clinical description - because i couldn’t get consensus within myself of what signals the end of hypertension. What marks the end of this disease - is there consensus among experts on this? If not - we cant write the clinical description describing the end of hypertension - and absence of clinical description we dont have a guide to define end of hypertension (i.e. cohort end date) when we build cohort definition
- is it repeated BP measurements that are within the normal range (repeated how often) in a person who is not receiving medical therapy.
- is it documentation using a code (that doesnt exist) that the person is normotensive
- is it absence of health care visits for management of hypertension and absence of treatment for hypertension.
Malignant hypertension and hypertensive crisis - we can define the end of. If BP goes below the crisis range - that marks the end of the hypertensive crisis episode. We can also mark the end of secondary hypertension – if there was a definitive treatment for hyperthyroidism or pheochromocytoma and after that the BP was found to be in normal range - we can use that to mark the end of secondary hypertension.
But how do we mark the end of essential hypertension
i dont think we are - we are striving to match the cohort definition to a clinical description. The clinical description is driven by the use case i.e. the study.
atleast that is my current position - ofcourse may change tomorrow as our understanding evolves.
- as long as we have clinical descriptions that are different - there may be different cohort definitions for the same clinical idea (e.g. in this case hypertension)
What do you think?
Thank you for this detailed and very helpful walk through for phenotyping. I had a very quick question about obtaining the clinical definition. I was wondering whether there are any examples other than Harrison’s manual (preferably with open access) that you would recommend in order to get to these definitions that would be considered as an authoritative consensus document. I would highly appreciate your input about this, especially about diseases for which the etiology is not very well established and in disease areas where categorization is more complex.
For diseases that are “established” we an expect to have abundant of authoritative resources. Many guidelines from professional societies come to mind such as JNJ Hypertension guidelines
For diseases that are not yet established, but are considered of public health importance - there may be Government/regulatory authority sponsored initiatives to define them e.g. the various covid related task forces come to mind for Covid.
I dont know of any single authoritative resources that compiles everything and is available world over.
Maybe others have found other resources?
This is it, @Gowtham_Rao. These should be two phenotypes: One for hypertensive crisis, and another one for the chronic condition of hypertension that leads to such crises or to a chronically high blood pressure.
Same thing: For the crisis, it’s the measurement of a low BP, or the end of the visit, or the administration of a slew of hypertensive drugs, or just a couple days as a means of last resort. It doesn’t really matter exactly when it ends, you just need to have some reasonable end date so you can have distinct crises for calculating their rates. For the ongoing hypertension as a disease, there is no end needed. We won’t calculate incidence rates of those conditions beyond the first onset (which is why those incidence rates are really proportions of patients in the population).
My problem with all this, as declared elsewhere, that we bury this in some text. We don’t make categories explicit:
- Acute onset conditions
- Chronic condition with flares
- Chronic conditions with a cure
- Chronic conditions without a cure
The criteria would be very different for these.
We also don’t seem to classify our criteria set:
- Improvement over low specificity (e.g. through repetition)
- Improvement over low sensitivity (e.g. through combination with diagnostic procedure)
- Improvement of index date (e.g. through combination with diagnostic procedure)
- Re-diagnosis because we trust neither of the above (e.g. using BP measurements for hypertension)
- Mixture thereof
And then there is the avoidance of some confounder, as @Kevin_Haynes mentioned, which should be absent for any of this. That’s for the studies.
Perfectly right - thats why we right the clinical description upfront! in this case - i explicitly excluded hypertensive crisis in my clinical description for essential hypertension.
And i agree with you here!
are you referring to “clinical description”
Yes. Nothing wrong with the clinical description. But there are categories of conditions requiring different approaches, and we could use them.
i agree there is scope to improvement on the clinical description - but the purpose of this whole thread is to emphasize this point and to ground ourselves to this best practice.
I agree with both @Gowtham_Rao and @Christian_Reich. I understand Christian’s concern to be that phenotypes live on but the clinical description and phenotype development decisions behind them do not. It seems like it would be easy to develop a convention whereby the phenotype documentation is always stored alongside the phenotype, for example in the git repository with a standard filename.
I see - so the is point that cohort definition and the clinical description are not tightly coupled together. If they are coupled together, then we will know why a certain cohort definition was built.
I would add that if there are evaluations on the performance of the cohort definition - that should also be coupled to the cohort definition. But they can be slightly different by data source…
ie. the unit to attach all this intelligence (i.e. clinical description, evaluation of estimates of sensitivity and specificity) is the cohort definition.
Thanks Gowtham. This is the clearest description I have seen for a repeatable process that can develop reliable phenotypes. In teaching new investigators in OHDSI, this something everyone struggles with.
To keep this thread alive so that we converge on a phenotype that the community can use, I propose the following phenotype (ATLAS)
This uses measurement values of systolic blood pressure as a qualifying entry event, but requires observing either diagnosis of treatment afterwards to avoid a one-time errant measure. It also proposes an exit criteria for the cohort: observing at least 2 SBP < 120 without current of future HTN treatments and no future SBP > 140.