Open annaschultze opened 4 years ago
The QOF codes were an early draft and have not been checked. They were mostly used for the MVP. In the end, the risk factors study uses actual systolic blood pressure and diastolic blood pressure (latest values) for both to indicate raised bp. @laurietomlinson was involved in decisions around this.
I feel reasonable confident about the hypertension QOF codes but I just wanted to flag up that this has not gone through the rigorous sign off procedure used for the other variates as we moved early on to values.
For NSAID and ICS we originally thought to adjust for diagnosed hypertension. For most recent BP, I realise this could be high but not genuine hypertension or normal in someone with hypertension who's well controlled. What was the thinking begind going with most recent BP instead (and apologies for going over old ground!!)?
I am not sure about the NSAID or ICS studies, certainly for the risk factor studies, we are using blood pressure value rather than any hypertension codes. @alexwalkercebm may know more about the drug studies and hypertension.
I think we'd want to be consistent between them all. For the RF study, this would tell you about people who've had a recent(ish) high BP reading, but not about people with diagnosed hypertension, though there would presumably be overlap. Was the thinking that current BP might be more relevant biologically than being in a group wih diagnosed hypertension?
@laurietomlinson suggests deriving a variable coded 0 or 1 with
1= coded hypertension regardless of most recent bp or most recent bp is high in the absence of a hypertension code
0=all other possibilities e.g. no hypertension code or recent bp, no hypertension code and recent bp is norma
l
LSHTM Hypertension Readcodes Hypertension.txt
It makes sense to me that what's of interest for the risk factor study is whether or not someone has high BP close to the time of their admission/death, rather than whether they had previously diagnosed (but managed) hypertension. I think the theory is that it's high BP that's implicated - but not a clinician.
For the drug studies I assume we're interested in things that may determine whether you are currently on a certain drug (NSAID/ICS) and your risk of the outcome, and for that I guess the broader category might be useful if those with diagnosed HT were considered for different treatments, for example...
So personally think it's fine if we're not consistent between studies - for one we're considering a potential risk factor, the others a confounder?
Here are SNOMED and QOF code lists for hypertension Hypertension_CodeList.xlsx
My two cents on the BP/HTN debate that you are most welcome to ignore:
Whether for risk factor or cov adjustment, a combo variable is probably warranted since they both capture information that make up the overall story. If we find elevated BP alone or diagnosed HTN alone increases risk for [insert outcome here], the next question will be which of the two are driving the association. So something like this: 0 – no HTN, normal BP 1 – HTN, controlled 2 – HTN, not controlled
Of course could also bring in meds to help define control as well.
I like @DarthCTR 's combination variable? Here are my comments on the SNOMED/QOF codes: Hypertension_CodeList (1)_LT.xlsx
Need to have a quick chat with @CarolineMorton when she's free :-)
I like it too - would need to decide what to do with no HTN code + no BP (put them in category 0?) and HTN code but no BP record. Not sure what we do with them?
Here are my thoughts on the full QOF code list that @CarolineMorton sent me: some repition remains but much reduced. Caroline do you have enough with this and the one I did above to move this forward? Thanks QoFClusteres_CTV3Codes - Sheet1_BP codes LT.xlsx
Summary for TPP:
CTV3 codes from @chris-tpp (many thanks!)
Hi all, I've checked the above list for hypertension. Quite a few exclusions, and I've tried to be clear about reasons.
One main point to note: I've excluded transient hypertensions (pre-eclampsia, drug side-effects etc), on the assumption they resolve and don't translate into long-term cardiovascular risk.
Happy to discuss or be corrected!
Thanks @HenryDrysdale
I have reviewed as well, and changed a couple. For note we have not include primary pulmonary hypertension and pre-eclampsia.
Reviewed: htn_reviewed.xlsx
Thanks and sorry for the delay Actually I have removed few codes in the end. This has raised some concerns about process. In general I don't use 'process codes' like monitoring to define disease but as they are widely used in other codelists I have kept them in. I would also normally exclude codes related to renal failure due to hypertension as this is widely miscoded but have retained them here. If they are in this list via 'Hypertension_CTV3_Raw_HD.xlsx' but weren't included in 'QoFClusteres_CTV3Codes - Sheet1_BP codes LT.xlsx' is that a problem?
No i don't think so because the CT3 codes included codes mapped from Read 2 and SNOWMED
I have removed the code you indiciated. this is the final codelist: final-htn.xlsx
Posting for sign-off in a minute below, I did an extra check against LT's review and;
Can @laurietomlinson or @HenryDrysdale do clinical sign-off; @ianjdouglas epidemiological?
This covariate will capture only diagnosed hypertension. Blood pressure measurements are described under ebmdatalab/opensafely-risk-factors-research#48
DEFINITION: Any record of hypertension ever in existing patient history. Absence of a code is taken as absence of hypertension.
Example Output:
patient_id | code | hypertension | date |
---|---|---|---|
123 | H/O: hypertension | 1 | 1/2/2009 |
332 | Hypertension resolved | 1 | 2/4/2016 |
CODE LISTS:
This was created through the following process:
The following were excluded:
FLEXIBILITY NEEDED BETWEEN STUDIES: There will likely significant variation required for different studies - with some, depending on their objective, relying on BP alone, some on BP and diagnostic codes and some on either of these in combination with treatments. This code list and variable represents ever diagnosed hypertension, and can be used in conjunction with BP measurements. and separate drug lists as required in different studies. It was considered preferable to have a clean definition for diagnosed hypertension alone.
EFFECTS ON COHORT SELECTION: None.
POTENTIAL BIASES: Hypertension may be underdiagnosed. Care required when using this variable; diagnosed hypertension may not equal current hypertension.
CLINICAL SIGN OFF & DATE: Laurie Tomlinson @laurietomlinson 28/4/2020 EPIDEMIOLOGY SIGN OFF & DATE: @ianjdouglas 04/05/2020
SHARED WITH WIDER TEAM:
FINAL SIGN OFF DATE (and apply label):
For the NSAID and ICS studies, it is of interest to capture hypertension using both BP and diagnostic codes. The rationale is that we're more interested in capturing potential confounding by hypertension, as opposed to describing the precise relationship between hypertension and the outcome (as in the risk factor analysis). A broad definition may therefore be more suitable for these studies.
Attaching codes pulled from QoF by @CarolineMorton during early discussions that we can use as a basis.
https://github.com/ebmdatalab/tpp-sql-notebook/blob/htn/data/htn.csv
Would be good to discuss next steps @laurietomlinson @ianjdouglas - do we need an LSHTM Read2 list and SNOMED codes in addition?