Phenotype Phebruary 2023 P8 Parkinson's disease

[allanwu]

Update on P8 Parkinson’s Disease.
@Gowtham_Rao what are next steps for this Phenotype?
Here’s what I think seem to be current status and next steps on this roadmap:

1. Two proposed flavors of phenotypes: unanimity and tiered consensus.

• last week, we created two initial unanimity phenotypes (without and with meds)
• more about tiered consensus below

2. Unaminity - proof of concept run results:

• inital results posted above by @Gowtham_Rao
  https://data.ohdsi.org/PhenotypePhebruary2023_P8_ParkinsonsDisease/
• I have not seen the Atlas attrition tables for 309 and 310
• review of Cohort Diagnostics suggests the following:
• need to add concept Progressive Supranuclear Ophthalmoplegia to the “parkinsonism (neurodegenerative, nonPD) conditions” concept sets. I have updated in: Atlas-demo concept set 1872601 (2/13/23); not yet updated in atlas-phenotype.ohdsi.org yet
• more details about how phenotype behaves in two datasets require attention: jmdc and truven_mdcd in that phenotype does not result in expected male, older predominance demographics; it seems there is more secondary parkinsonism in both datasets overall, but still have to figure out why. From phenotype point of view, would like to identify if I am missing possible secondary parkinsonisms concepts that should be excluded in last 3 years; one way to approach this is to partially address is to create a unanimity cohort that will exclude ALL secondary parkinsonisms for all time (not just 3 years) and see if that results in expected demographics in these datasets. Another option worth exploring is if we should include a limitation in visit types to office visits (as the original phenotype paper did).

3. Would benefit from guidance on next steps to consider how to address above issues and to consider additional analysis or cohorts to help refine this phenotype.

For tiered consensus criteria

1. I agree that specialty seems to be challenging for P8 for now.

• Movement disorder specialty is not practical (once we have our own OMOP-CDM instance, we can post how much difference this might make)
• there remains value in including neurologist visits if possible. Again, in published algorithms for detecting PD with specificity, the tired consensus perfoms with higher sensitivity, specificity and PPV compared to unanimity algorthm.

2. I believe there is still value in assessing a tiered consensus criteria

• even without a visit specialty criterion, a tiered criteria will allow a comparison between a full 3 year exclusion (unanimity) vs a tiered non-specialty consensus criteria.
  There are likely 3 approaches to take to tiered consensus criteria:
• (1) validate/review the 6 atlas-demo cohort defintions as proposed by @allanwu on 2/8/23 and run those
• (2) consider adaptation of those 6 atlas-demo cohort defintions and remove the specialty criteria and just focus on effect of the tiers year-by-year. suggested options for these 6 cohorts: “tiered consensus wo meds” (current), “tiered consensus w meds” “tiered non-specialty consensus wo meds” “tiered non-specialty consensus w meds” (only the first is available in atlas-demo currently).
• (3) I remain unsure how scalable the Phea approach is for networks tudy

@fabkury I reviewed the Phea package and I think I see what you are trying to do. At each time point (essentially visit), you identify if that particular visit is included in the cohort or not. I would add that if a particular visit is NOT included in the cohrt, then at that point, the person would be consided to drop out of the cohort. This raises very interesting possiblities in tracking the evolution of patients.
However, for the basic P8 implementation of the tiered consensus criteria (understanding you did not implement specialty yet), one would merely need to apply the SQL/Phea approach to the entry point (most recent) visit, look back 3 years, and either include or not include that person into the cohort or not.

And I remain interested in any of above cohorts, in the future, how we would be able to do an inference of incidence (earliest entry point). I didn’t quite follow @Patrick_Ryan post on this.

[Gowtham_Rao]

Thank you @allanwu

@allanwu

what are next steps for this Phenotype?

atlas-demo.ohdsi.org has the following definitions - i have taken a snapshot of it.

Regarding Unaminity definitions:
Regarding output and shiny: I will update the shiny app output to reference the updated definitions in atlas-demo.ohdsi.org. this should have the attrition output. i.e. lets continue the iteration on atlas-demo.ohdsi.org and only move the mature definitions to atlas-phenotype.ohdsi.org at a future stage. This strategy will allow us to iterate and update quickly.

Regarding observations in the tiered definitions and how to address them: I suggest making a list of observations and frame them as sensitivity, specificity errors. You may also consider saying why you think they may be errors. This would help some of us non Parkinson’s experts provide ideas and collaborate on solving it. Examples are:

• variability in male and older preponderance by data source (e.g… possible sensitivity errors, younger individuals less likely to have Parkinson’s disease and mean age lower than expected)
• higher secondary parkinsonism (source of specificity errors?)

Regarding Tiered definitions: To pursue the tiered definitions, we need some time and it may well be atlas only, atlas + Phea or Phea only. But there are many outstanding issues in the atlas based definitions. Of note:

• Blanks for provider specialty: @Chris_Knoll shared with me that it is still a valid cohort definition and it will execute - but will make it more explicit
• There may be some logical errors in some of the definitions.

For next steps - for tiered definitions specifically - I think we should do a targeted re-review of the implemented cohort definition to make sure it is complete and has transformed the logic.

I am attempting to get an output on https://data.ohdsi.org/PhenotypePhebruary2023_P8_ParkinsonsDisease/ later today with updated output for at least the Unaminity definitions (it will have output of tiered definitions, but I think we should ignore them for now because of possible errors). I will update this thread when ready.

[Chris_Knoll]

@Gowtham_Rao

Blanks for provider specialty: @Chris_Knoll shared with me that it is still a valid cohort definition and it will execute - but will make it more explicit

‘valid’ has mixed meanings here. For purposes of SQL generation, it’s ‘valid’ to leave it blank and we’ll ignore it and generate a query. But is a blank provider ‘valid’ for this phenotype?

[Gowtham_Rao]

@fabkury

Generate a new study package using the modified SQL file. (did I do this correctly? I am not sure)

This design is very promising @fabkury . There maybe some potential errors in the current tiered Atlas cohort definition (not limited to the missing provider specialty information). Once those are addressed, I can attempt to use Phea SQL.

[allanwu]

@Gowtham_Rao and group,
I have made no changes in the unanimity phenotypes in Atlas-demo.
Since last post, I have updated Atlas-demo definitions for tiered consensus (2/13/23 10-11pm CST).
I was able to import desired neurology specialties into the specialty criteria, so I am comfortable with the content of the specialty neurology criteria (within CDM standard vocabularies). I realize not all OMOP-CDM instances may not all have similar implementations of specialty, so I have created two flavors of the tiered consensus criteria below
I believe these are ready for logic/technical review.

I have updated 6 phenotypes of the tiered consensus algorithm that include the neurology criteria these are tagged with [Pheb2023][ucepd]
Persons with PD tiered consensus w specialty [no]neuro#year (6 definitions)

• 3 definitions are for persons who have neurologist visits in last 3 years (for each of the 3 years)
• 3 definitions are for persons who do not have neurologist visits in last 3 years

I have created 3 further tiered consensus defintions - ignoring specialty criteria
Persons with PD tiered consensus wo specialty #year (3 definitions)

• 3 definitions that ignore the specialty criteria but do evaluate PD defintions vs exclusions for each of the last 3 years.

The rationale is to evaluate the increase in PPV, sensitivity and specificity for detecting PD (as opposed to similar parkinsonisms and not-PD) in 3 sets of phenotypes:

• unanimity definition w and wo meds - good PPV/specificity; lowest for all
• tiered consensus wo specialty - better PPV, higher sensitivity, risk for specificity better or worse given allowance for some exclusions 2 or 3 years prior to current year
• tiered consensus w/specialty - anticipate better PPV, sensitivity and specificity - testing assumption that weighting neuro specialty diagnoses provide added value to these EHR cohorts. But this likely works only for OMOP-CDM instances that have reliable specialty coding.

atlas-demo phenotypes:
1781748 unanimity wo meds
1781760 unanimity w meds
1781814 tiered consensus wo specialty last 1 year (1 of 3)
1781815 tiered consensus wo specialty 2 years ago (2 of 3)
1781816 tiered consensus wo specialty 3 years ago (3 of 3)
1781732 tiered consensus w specialty PD neuro last 1 year (1 of 6)
1781809 tiered consensus w specialty PD neuro last 2 years ago (2 of 6)
1781810 tiered consensus w specialty PD neuro last 3 years ago (3 of 6)
1781811 tiered consensus w specialty PD no-neuro last 1 year (4 of 6)
1781812 tiered consensus w specialty PD no-neuro last 2 years ago (5 of 6)
1781813 tiered consensus w specialty PD no-neuro last 3 years ago (6 of 6)

[Gowtham_Rao]

@allanwu

OMOP does not support movement disorder specialty

@aostropolets for vocabulary landscape assessment. I have struggled with provider specialty vocabulary.

@allanwu do you know how ‘movement specialist’ are documented in your data sources e.g. is it a designated with a specialist code in your health care system? I do not think it is a American Board of Medical Specialties certified specialty (or is it?). Is this a physician specialty or a non physician specialty?

As FYI OMOP vocabulary requests are handled here i believe Issues · OHDSI/Vocabulary-v5.0 · GitHub . If you show them there is an authoritative vocabulary that is generally accepted and if licensure allows them to import it into OMOP - they usually do it.

Current physician specialty list is here and more general provider specialty is here

As @Patrick_Ryan said here, the community does not have the experience with using Provider Specialty mostly because most data sources do not have this information in a reliable way - but if and when available, both the CDM and tools can support it.

@Patrick_Ryan

As a community, we are still harmonizing on provider specialty, and I know many databases either do not provide this information or if its particularly noisy (including not mapped into visit occurence table). Many of the NUCC concepts aren’t specific to Neurology (they combine Psychiatry and Neurology), so I’m just out of my depth to implement this reliably.

Also discussed here

[Gowtham_Rao]

Hi @allanwu - the shiny application should have the most recent definitions here https://data.ohdsi.org/PhenotypePhebruary2023_P8_ParkinsonsDisease/

Some suggested topics to discuss on todays call at 12pm est (meeting invite - everyone welcome to attend)

• status of local omop instance
• unanimity definition to completion (is it complete e.g. concepts complete, rules accurate, orphan included, do we need unanimity specific subgroups e.g. jmdc and truven_mdcd issue, review attrition, evaluation using the framing of sensitivity, specificificity, index date misspecification errors)
• tiered consensus definition using atlas only (review the 9 definitions output. note - i extracted all definitions in atlas-demo, so it has the ones not part of these 9)

[Azza_Shoaibi]

Here are some notes from the evaluation we did today to the unanimity definition (Persons with Parkinson’s disease unanimity- c1781748 ):

A. Attrition/ Effect of inclusion criteria:

• “has at least 1 PD specific code criterion” has no effect- suggesting that all the data is loaded on the “specific code” and the broader codes are not utilized.
• " has 2 encounters with PD code that are at least 30 days apart" has no effect - that was due to a bug/error in the cohort definition which is now corrected-
• “has no secondary parkinsonism conditions” and “has no (non-PD) neurodegenerative parkinsonism conditions” criteria lead to a loss of 0.4-7% each, which is inline with expectations- it is estimated that 5% of Parkinson patient have secondary Parkinson or (non-PD) neurodegenerative parkinsonism conditions.

B. Index event breakdown:

• In US data sources “Parkinson’s disease-G20” accounts for 34% (in MDCD) -83 % (in optum EHR) of persons on index and Paralysis agitans accounts for 19-70%. In JMD, Germany France, 100% has
  “Parkinson disease” on index and no other code.

C. Time distribution:

• on average data sources had a median time of 3 years before index. This is assuring, since all inclusion criteria is based on 3 years prior index.

D. cohort characterization:

• In most data sources, patients were of older age and more likely to be males-which is inline with expectation to the known trends of Parkinson . However, JMDC and MDCD had higher proportions of younger age groups and higher females when compared to other data sources and when compared to expectations.
• The age and gender distribution in JMDC and MDCR may be an indication of a specificity error. In specific, Parkinson disease among younger /female group is unlikely to be idiopathic Parkinson’s Disease but may be related to “drug induced Parkinson”.
• In JMDC around 50% of the cohort have Schizophrenia compared to only 2.1% in optum DOD. Also a much higher proportion of the cohort in JMDC are on Olanzapine (15%) and Risperidone (20%) when compared to optum DOD (1.2% and 4.0% respectively). Antipsychotic drugs such as Risperidone and lanzapine are known to be associated with Parkinson disease (drug-induced Parkinson). The high prevalence of Schizophrenia and it’s treatments in JMDC may suggest that a proportion of the cases are drug-induced Parkinson, representing false positive cases.

[allanwu]

Thank you for great notes @Azza_Shoaibi

Current goal is to confirm the technical and clinical logic for the unanimity PD cohort defintions.
Once initial evaluation of logic is done (12 databases were available in discussion above), we can run as network study and fully evaluate the definition with potential sources of acceptable or understood error.

Revision of logic of unanimity PD cohort definitions based on discussion.
All work is in atlas-demo site.

1. New “Parkinson’s Disease [ucepd][Pheb2023]” concept set that includes PD condition and children with appropriate exclusions to make sure all standard conditions that would include PD are captured. ID 1869664.

2. Created “Medications associated with parkinsonism [ucepd][Pheb2023]” concept set. It includes both Classification and Ingredient standard concepts. It also includes several exclusions that are acceptable in a cohort defintion of PD (as described in clinical description). ID 1872667.

3. Updated/created 4 versions of unanimity PD cohort definition.

• fixed the 2 PD conditions separated by 30 days logic using nested criteria and “30 days before and 1 day before index start date” (not 0 days)
• used the new PD concept set for defining PD conditions
• all tagged wtih [ucepd][Pheb2023] to find
  1781748 Persons with PD unanimity (no med criteria)
  1781760 Persons with PD unanimity w PD med criteria (include PD meds)
  1781843 Persons with PD unanimity wo confounding meds (meds that cause parkinsonism)
  1781844 Persons with PD unanimity w PD med and wo confounding meds

The 1781843 is designed to address limitations in JMDC and MDCD datasets; we include an exclusion if the person is exposed, in the last 3 years, to a drug that causes drug-induced parkinsonism. This would exclude those that are coded with PD (but are suspected are more likely to have drug-induced parkinsonism) if they are treated with meds that treat schizophrenia (and can cause parkinsonism). This should increase specificity overall when including JMDC and MDCD at some acceptable loss of sensitivity/PPV.
1781844 then adds further specificity by including both med criteria (PD meds in support and drug-induced meds to exclude) with further loss of some sensitivity/PPV.

I believe these 4 unanimity cohorts are ok from clinical logic.
@Gowtham_Rao I believe these can be reviewed for technical logic.

I will continue to review/refine the tiered consensus cohort definitions w/wo specialty criteria and post when I feel those are also aligned with current comments.

[Gowtham_Rao]

See video recording here Meeting in General-20230215_120115-Meeting Recording.mp4

@CraigSachson could you please post it to Phenotype Pheburary 2023 homepage when possible Phenotype Phebruary 2023 – OHDSI

[Gowtham_Rao]

@allanwu The shiny app should be updated based on today’s discussion related changes

[allanwu]

Quick question for Atlas gurus: @Gowtham_Rao @Azza_Shoaibi @Patrick_Ryan
I am reviewing the current Cohort Diagnostics run on the unanimity and tiered cohorts (as they existed yesterday) and I want to make sure I have the correct Atlas configurations (in spite of discussion yesterday).

My question is what is the correct way to code Atlas to match the correct bold/italics sections below?
And which of the two cohorts criteria below correctly reflects the idea that we want to capture two PD condition occurrences separated by at least 30 days? (and within the last 3 years)?

I am using Cohort Diagnostics to know exactly what was run and copying from the Cohort Defintion within Cohort Diagnostics to help interpret counts/attrition:
https://data.ohdsi.org/PhenotypePhebruary2023_P8_ParkinsonsDisease/

The unannimity cohort (1781748) criterion 3 (two PD conditions reads as follows:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting in the 30 days prior to ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date; allow events outside observation period.

The tiered consensus cohort 1781732 criterion 3 (same two PD conditions)
reads as follows:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting anytime up to 30 days before ‘Parkinson’s Disease (PD)’ start date; allow events outside observation period.

(I realize that the PD condition used in the two cohorts are different and I’ll work on fixing that in the tiered cohorts during this week, so these criteria cannot be directly compared yet in the Cohort Dx tool.)

[Gowtham_Rao]

@allanwu

And which of the two cohorts criteria below correctly reflects the idea that we want to capture two PD condition occurrences separated by at least 30 days? (and within the last 3 years)?

Apologies @allanwu i may not be fully understanding your question - but regarding the rule and its text: these two are the same

From Atlas

From CohortDiagnostics:

3. has 2 encounters with PD that are at least 30 days apart

Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting in the 30 days prior to ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date; allow events outside observation period.

[Chris_Knoll]

@allanwu

Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting anytime up to 30 days before ‘Parkinson’s Disease (PD)’ start date; allow events outside observation period.

Just to add a little technical context:

This print-friendly view is combining the nested criteria into a single sentence for print-friendly purposes.

The outer part is saying you have at least 1 PD in the past 3 years of the cohort entry event, and the inner part is saying that the PD that was found must have at least 1 PD in the prior 30d not including the date of the PD event we found.

How that resolves in print friendly:
Outer part:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period;

Inner part: having at least 1 condition occurrence of ‘Parkinson’s Disease (PD)’, starting anytime up to 30 days before ‘Parkinson’s Disease (PD)’ start date; allow events outside observation period.

What I bold above is how you know the context of the statement. The outer statement is indexing on the cohrt entry event, the inner context is indexing on the PD event that you found in the 1095d prior to cohort entry.

[allanwu]

Thanks for replies. Sorry I was not more clear.
I understand the nested nature and will try to clarify the ask here.

I am asking Atlas gurus @Gowtham_Rao @Azza_Shoaibi @Patrick_Ryan to confirm the logic for the criterion that there should be two PD condition occurrences separated by at least 30 days, both within 3 years.

I remain unsure of which of the following 3 options (all of which have seen some use in Phenotype Pheb’s 2022 and 2023 for PD). I believe the correct one is actually version 3 which has not been implemented for Pheb 2023.

version 1: revised based on discussion on 2/14/2023 - was used for unanimity cohorts that have been run
version 2: early version of this criterion shortly after P8 started, revised to version 1 form after after first discussion 2/6/23
version 3: adapted from criterion suggested by Patrick Ryan in 2022

Details:
version 1:
this seems to capture two PD events that are WITHIN 30 days of each other, not separated by 30 days of each other.

Description version 1:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting in the 30 days prior to ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date; allow events outside observation period.

version 2:
this one captures a 30 day separation between PD occurrences but goes back in time from the outer index event - which risks the earlier event within the nested criteria to be PRIOR to the 3 year time limit imposed by the outer criteria.

Description version 2:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting anytime up to 30 days before ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date; allow events outside observation period.

version 3:
this goes forward in time and seems to assure that the initial outer index is within 3 years and then ensures there is another PD condition separated by 30 days in the future which should be ok.

Description version 3:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting 30 days after ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date; allow events outside observation period.

So, I think I favor version 3. If we agree, then I will change all the atlas-demo phenotypes to this version. Version 1 is in use in the current unanimity phenotype and it is showing a very large dropoff of counts between the 1 PD criterion and the 2 PD criterion which seems unusual.

[Patrick_Ryan]

@allanwu , the temporal ordering was why I was investigating this earlier.

I understand the use case you want is 2+ visits with >30d between then, in which case you are correct that versions 2 and 3 both do this, and version 1 does not (because as you say, it looks for 2 visits WITHIN 30d).

To decide if we are looking ‘before’ or ‘after’ really depends on how we are going to set the cohort start date. When you started this, you were looking to find the latest event, which the idea that this gave you the most information to classify persons with the disease, but the consequence of that approach was that the index date was clearly ‘wrong’ in that it didn’t represent time of disease start. Assuming that you’ve settled on how to ‘find the people with the disease’, then our task is just to correct for index date misspecification. My initial impression is that we’d probably assign ‘disease start’ to be the first time the person was initially diagnosed, since that was the first suspicion by some clinician, which was later corroborated by the criteria you’ve implemented with drugs and neurology visits, etc. If that’s the case, then I would think our new algorithm would be using the earliest event of the condition in the entry event, and then we’d have an inclusion criteria requiring 2+ visits >30d apart to be after the index. (Note, this isn’t quite your version 3, because you are looking 3 years before for a diagnosis and then requiring that diagnosis to have another visit >30d after)

[Kevin_Haynes]

@Patrick_Ryan

My initial impression is that we’d probably assign ‘disease start’ to be the first time the person was initially diagnosed, since that was the first suspicion by some clinician, which was later corroborated by the criteria you’ve implemented with drugs and neurology visits, etc.

This is a great discussion of the PD phenotype. I’m forever challenged when we utilize the future to ascribe knowledge/wisdom to the past. Clearly on the day the first clinician suspected and recorded a diagnosis of PD for billing there was only a workup of PD. The future would confirm or refute our diagnostic uncertainty. Now if I gold standard pulled a medical record for that index date of disease and presented the case to a neurologist they might say no disease on this date. Recall you can’t give them information that hasn’t happened yet in the chart review since the treating clinician on that date doesn’t know the future. But alas biologically this is the better index date. @jswerdel does PheValuator use the future in it’s index prediction modeling of disease. If I’m not diagnosed again with PD for 2.5y you only know that I have PD because of this future event. What about all of the person-time of PD if I switched insurance at year 2.25 (as I aged into Medicare). Now I truly have full blown PD but the algorithm won’t classify me until my second dx (in reality my third). Is our underlying health care delivery system dragging up the age of onset by design of our data availabilities?

Oh future Medicare…be ready in 20 years when I enroll I’m sending you a link to my historical administrative claims data in the OMOP CDM for you to load for future researchers…

[allanwu]

Thank you for your replies @Patrick_Ryan, @Chris_Knoll @Kevin_Haynes!
I agree we ideally would have cohort entry criteria be first occurrence of any sort of parkinsonism (that is what we do for our epidemiological purpose) and, within what is in the database, apply the “specific PD” cohort criteria to the most recent/latest 3 years (in our proposed definition here) of parkinsonism-related criteria. It is ok to be rejected from cohort if there is only 1 year of observation time for a given
person because parkinsonism was coded correctly as earliest suspicion but only 1 PD-specific occurence was available. This same person after 5 years of followup will have many PD specific code occurrences in the last 3 years and is correctly PD).

I just couldn’t figure out how to use Atlas to capture entry criteria as earliest occurrence (to be accurate with incidence) and make sure that the timeframe of 3 years (or shorter if there was <3 years observation time) was used as time criteria to be specific about the later diagnosis.

It’s almost as though we need two nested Atlas cohorts. One cohort captures the entry criteria of earliest (this is actually represented correctly in xSens):
#1781859 [Pheb2023][ucepd] Persons with Parkinson’s disease xSens (Atlas Demo, 2/17/23)

Then this cohort entry could have a nested cohort within the persons captured to have (sub)cohort entry of latest occurrence of condition with the subsequent criteria needed to assess a lookback time of 3 years (in our use case here).
I could not see how to do this in Atlas, though I’m sure we could find a way to code it this way manually

so for now, after noodling with Atlas further, I am going to stick with the latest occurrence for the cohort for PD identification so we have the right specificity being represented (but not the right incidence).

Re-reviewing all the comments, I think the most sensical criteria is indeed version 3 in my 2/17 post 34.
the logic holds for me - the outer criteria captures the last 3 years based on cohort entry criteria of “latest occurrence”
The inner criteria “index” is based on any of the outer criteria occurrences (wihtin 3 years) and finds one of those occurrences that has another occurrence 30 days later. It safely keeps the criteria applied within 3 years which is what is desired.
Description version 3:
Entry events having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting between 1,095 days before and 0 days before cohort entry start date; allow events outside observation period; having at least 1 condition occurrence of ‘Parkinson’s Disease [ucepd][Pheb2023]’, starting 30 days after ‘Parkinson’s Disease [ucepd][Pheb2023]’ start date ; allow events outside observation period.

i’ve completed a re-work of all the proposed criteria using this version 3 criteria for 2 codes and included a xSens and xSpec so we can get some basic comparison data across databases/cohort defintions which I will post next.
(atlas-demo is down right now, so will post this after clinic)

[Mark_Danese]

Just keep in mind that any survival analysis results will suffer from immortal time bias if you use the first occurrence as your index date but you require 2+ occurrences to define the cohort. This may not be applicable to this specific cohort or the intended analyses, but I didn’t have a chance to try to decipher exactly how the cohort was going to be constructed.

[Gowtham_Rao]

Please join us tomorrow February 22nd at 1pm EST to further the conversation. Meeting invite

Join conversation