Phenotype Phebrurary 2023 - P2 - Anaphylaxis

[Gowtham_Rao]

Target Clinical Description

(Original ideas for this phenotype was described in a 10 minute video here 10-Minute Tutorials: Cohort Diagnostics (Gowtham Rao, June 8 OHDSI Community Call) - YouTube )

• Anaphylaxis Anaphylaxis - StatPearls - NCBI Bookshelf 1

Summary: Acute potentially life-threatening multisystem syndrome due to mast cell mediator release most often via IgE. Presentation: Anaphylaxis is highly likely when sampson criteria is satisficed. Anaphylaxis is a clinical diagnosis, and labs are not used as part of diagnosis. Plan: Preserve airway and breathing, circulation, mentation. Evaluate for airway compromise. Epinephrine IM or IV depending on severity of presentation. Antihistamines as adjunct. Bronchodilators for bronchospasm. Prognosis: Complete recovery common when anaphylaxis recognized early in the clinical course, to fatal if treated too late or with inadequate intervention. Allergist consult recommended after recovery.

Sampson Criteria

Second symposium on the definition and management of anaphylaxis: summary...

Additional case definitions are discussed here Anaphylaxis: Case definition and guidelines for data collection, analysis, and presentation of immunization safety data - ScienceDirect

MedDRA PTs: anaphylactic reaction, anaphylactoid reaction, anaphylactic shock, anaphylactoid shock

Phenotype Development: We made design choices in the cohort definition based on the following reasons.

1. Concept Set expression creation: We chose the following concept id based on lexical matching. Only one concept id is from the condition domain, and others are from observation domain.
2. No visit restriction: we decided to not restrict persons based on visit.
3.Exit strategy: Because most persons with anaphylaxis are expected to not continuously suffer from anaphylaxis for more than 1 day, we decided to exit persons at 1 day after start of anaphylaxis.

Submission 259 We submit one cohort definition that is currently in pending peer review status in OHDSI PL. id 259 and name ‘Anaphylaxis all cause’.

Cohort Definition Logic Description 259: we identify multiple events per person for anaphylaxis, i.e., one person may have more than one record. Because the median expected duration of anaphylaxis is 1day, we decided to define the end date of the phenotype after 1 day from the last date of continuous care for anaphylaxis. Because we opined that any further anaphylaxis within 30 days may reflect ongoing care for a prior anaphylaxis event, we used a collapse strategy that joined any new event to be part of the old event. For detailed logic please read human readable text on data.ohdsi.org/PhenotypeLibrary shiny application for cohort id 259.

Phenotype Evaluation 259:

• I observed that we have about 1.1 to 1.5 events per persons with 90%ile of person entries having time in cohort of 2 days. This suggests that our cohort exist strategy is correct.
• On incidence rate plot, I observe most data sources to have similar incidence rate - with most differences explainable by data source heterogeneity all within an order of magnitude of each other. However, the data sources that had the lowest incidence rate appear to be those that do not have inpatient data capture. Younger age had higher incidence rate compared to the other age groups.
• On reviewing the index event breakdown, about 20% of persons appear to enter the cohort based on food anaphylaxis or toxin induced anaphylaxis.
• More than 70% of visits that started on index date were outpatient visits. Because I expected that persons with anaphylaxis were being managed in emergency setting, this observation of people being managed in an outpatient or office setting made me think of specificity errors. However, when I reviewed the characterization of this cohort on Day 0 - I observe that more than 50% of persons have received epinephrine, along with other drugs commonly used to manage anaphylaxis like antihistamines, steroids. This allays my concern of specificity errors and suggests that these are true cases.
• In the time window immediately prior (-1 to -30d) I observed some use of drugs for the treatment of anaphylaxis - suggesting index date misclassification.

---

Cohort Definition id 221:
Justification for second cohort definition 221: I presented this, potentially conflicting diagnostics to my collaborators, specifically the presence of higher incidence in younger age groups, observation of outpatient visits and the utilization of drugs. Our collaborators focused the discussion on the finding that more than 20% of persons indexed on food anaphylaxis. After discussing we decided to explore if it was possible to phenotype a subset of anaphylaxis – that we decided to call “Anaphylaxis when a person was not known to be exposed to certain likely or known allergens” (a new phenotype), i.e. in this phenotype we are using a modified Sampson’s criteria as a case definition i.e… we want to remove persons who were reported to have anaphylaxis after an environment allergen (likely or known) e.g. bee stings. We decided to call this phenotype as ‘Anaphylaxis non environment exposure related’ and specifically indicate that persons with environmental exposure to a likely or known allergen would not meet this new case definition.

Submission 221: Based on the justification above, we submit the second cohort definition which is currently pending peer review in OHDSI PL and has the id 221 with the name ‘Anaphylaxis Non Environmental exposure related’.
Logic Description 221: Compared to 259 (all cause) we allow persons to enter on a narrower set of concept set that excludes concepts related to food, and environmental exposure. In addition, to remove persons who may have both a more ‘non-specific’ anaphylaxis code and a specific anaphylaxis code, we added additional rules that remove events were persons are found to have food or environmental anaphylaxis.

Phenotype Evaluation 221: Cohort counts: Compared to 221 (all causes), the new 259 (non-environment) had a considerable reduction in counts – sometimes more than 50%. This implied that persons with food and environmental anaphylaxis accounted for many persons in most data sources and it is important to evaluate this sub-type of anaphylaxis as its own phenotype.

Notice in the attrition table below – the use of inclusion rules was not the reason for this reduction in count.

It was mostly from persons not entering via the modified concept set above. i.e., a lot of persons with anaphylaxis are coded with only the food or environmental anaphylaxis concept, and the number of events/persons with both a non-specific anaphylaxis concept and a food/environmental concept at the same time is relatively small - only about 8%.

I then compared the cohort in 259 and 221 – and found the population level characteristics of treatments very similar on index date and in the period immediately after. However, there were differences in incidence rate with younger age group - the newer restrictive definition had the same incidence rate in all age groups, while the all anaphylaxis had higher incidence among younger ages. This is explained by prior knowledge that younger individuals are more likely to have food and environment related anaphylaxis compared to older adults.

Overall 259: We submit 259 to be a subset of 251 but a different phenotype (i.e. non environmental). On review of the treatment patterns, the characteristics of 259 is similar to 251, and so both cohorts have anaphylaxis.

Submission 258: Finally – we are submitting a third cohort definition. This is based on a previous OHDSI work discussed here Phenotype Phebruary Day 24 - Anaphylaxis as part of Phenotype Phebruary 2022. This cohort definition attempted to replicate the guidance from the Food and Drug Administration (FDA) and is detailed in the referenced post.This new cohort definition appears to capture most of the person in 258, but not all. But it also captures persons who are not in 258. Interestingly, unlike 221; 258 is trying to model all anaphylaxis and not attempting to avoid environmental allergen related anaphylaxis. It appears to have lower sensitivity compared to 259.

Overall: We have submitted three cohort definitions for the consideration for peer review. We expect 258 to have more sensitivity error compared to 259. We however argue that 221 (non-environmental) is attempting to model a slightly different clinical idea (phenotype) that we describe as “Anaphylaxis Non-Environmental exposure related”. We observe some evidence of index date misclassification in the -1 to -30d window that is residual and not corrected.

[Andrea_Noel]

Thanks @Gowtham_Rao and @Azza_Shoaibi for the opportunity to weigh in on the anaphylaxis phenotypes for Phenotype Phebruary 2023.

My brief background for other data enthusiasts - I’m a physician working on the Clinical Informatics team at Epic Systems Corp, and for several years I’ve been involved with phenotyping for single healthcare organizations using Epic EHRs as well as phenotyping on aggregated datasets from multiple Epic organizations.

I am assuming we are trying to define acute anaphylaxis events with these phenotypes and not follow up care for anaphylaxis.

As you know, anaphylaxis is a clinical diagnosis, and happens within minutes to hours of exposure to an offending agent and resolves within hours to days. Symptoms range from mild to life threatening. The clinical targets outlined in algorithms like NIAID/FAAN and WAO criteria are unlikely to be cleanly captured in many real-world datasets and can have high false positives for other conditions. Recent updates to the Brighton Criteria may be somewhat more in reach for RWD identification and provide some guidance as to level of certainty about anaphylaxis, but each of these criteria could be its own phenotype! Anaphylaxis: Revision of the Brighton collaboration case definition - ScienceDirect.

Acute anaphylaxis care patterns may involve emergency transport/ambulance services, emergent/urgent evaluation (+/- workup), acute treatment, and at resolution most patients will receive an outpatient prescription for rescue treatment and a recommendation for follow-up care or further outpatient workup.

Because the clinical criteria above may be challenging to reliably ascertain from RWD, I support diagnosis as a leading method to identify a cohort’s acute anaphylaxis events, but we may consider inclusion/exclusion of the following elements depending on the cohort’s use case: Diagnoses setting, vitals/signs/symptoms/exam findings, labs/results (positive/high serum mast cell tryptase or histamine drawn in same encounter), medications, allergy reaction type, allergen type, proximity of anaphylaxis diagnosis to identifiable offending agent/allergen

So, I think we have opportunities to increase sensitivity and specificity of the proposed phenotypes if we are trying for acute anaphylaxis events…

Opportunities to increase specificity:

Encounter: The encounter type where diagnosis was placed matters a fair amount. Anaphylaxis diagnoses are really common in outpatient encounters like office visits and refill encounters where physicians are associating the diagnosis to follow-up care, allergy testing, or EpiPen refills/prescriptions. If we dive into the pattern of epinephrine prescription vs administration in non-emergent encounter types, you will likely see significantly more prescriptions for EpiPens instead of epinephrine administrations. Without an encounter restrictor on your cohort, I suspect that more than 50% of your cohort may be follow-up care instead of acute events. I strongly suggest limiting the encounters for consideration to inpatient, emergency care, urgent care, or surgery if we want to capture the acute event.

Treatment: In a true anaphylactic event, patients should receive urgent treatment. Common treatments include medications like epinephrine, other H1/H2 antihistamines, bronchodilators, steroids, normal saline, or even methylene blue. Epinephrine is the gold standard treatment, and the primary route is Intramuscular (IM) and dose is 0.01 mg/kg (maximum dose of 0.5 mg) per single dose; the most common concentration is 1mg/ml, and common doses you will see include 0.15mg, 0.3mg, and 0.5mg. IM epinephrine is rarely used for other conditions, so presence of IM epinephrine at these doses near an anaphylaxis diagnosis is a decent sign of that doctors are trying to treat for anaphylaxis. If patients self-treated with an EpiPen before presentation, you may see the other medications without epinephrine. Another benefit to adding a requirement for treatment is to exclude patients who use the emergency department for primary care. If patients present to the emergency room solely to get an EpiPen prescription refilled because they do not have primary care doctor who can fill script, then you may see diagnosis, plus prescription without any treatment. I’m not sure how often this scenario will affect data but may impact datasets from US partners more than datasets from countries with universal healthcare.

Misdiagnosis: Mimics of anaphylaxis are broad, and common mimics are acute generalized urticaria, angioedema, acute asthma exacerbations, syncope, and anxiety/panic attacks. There is no real downside to treating for anaphylaxis while working up these other conditions, as a missed opportunity to treat anaphylaxis may be a greater risk to patients than unnecessary treatment. So even with a diagnosis and evidence of treatment for anaphylaxis in an emergent visit, the cohort is likely to be an over-estimation. Of note, anxiety is correlated with higher rates of EpiPen prescriptions and anaphylaxis diagnoses, and anxiety may have been a significant factor in some of the reports of anaphylaxis surrounding COVID vaccination due to heightened emotional, social, and political situation around COVID vaccines: Misdiagnosis of systemic allergic reactions to mRNA COVID-19 vaccines - Annals of Allergy, Asthma & Immunology (annallergy.org). Quantifying or overcoming this problem may be difficult. We may consider sensitivity analyses for subpopulations with other pre-existing conditions? This could help users of this phenotype to have an idea how often their cohort may not be true anaphylaxis events.

Time window for cohort inclusion may not include the date of anaphylaxis: Depending on how data partners have attributed the diagnosis to encounter, the date of diagnosis may follow the anaphylaxis event by a calendar day. This is most important for any encounter which crosses a midnight. In an emergency encounter, the encounter date will be the date patient presents to ED, but if they are monitored across a midnight, the clinician may not enter their clinical diagnosis until the subsequent day. In this case, evidence of treatment initiation or the encounter date may be a better marker for the date of anaphylaxis event. A billed diagnosis may however get attributed to the encounter date, so in the case of an inpatient encounter, if anaphylaxis happened in the middle of an admission, the billed diagnosis might actually pre-date the onset of the event. If data partners have included medication administration times, they may increase their accuracy on event date by favoring treatment dates instead of diagnosis dates.

Restricting to specific allergens: While occasionally the diagnosis will be specific to the allergen, the most common diagnosis will be “anaphylaxis, unspecified”. If data partners have access allergy/allergen/reaction data, these could be layered onto the current phenotype to help find or exclude a specific population when the diagnosis is generic. A patient’s allergy list should not be used in isolation to define anaphylaxis events, as it can be added any time after events, but the proximity of timing to treatment may be a clue when the diagnosis is not specific to agent.

• When looking for drug/vaccine anaphylaxis, for example, it may help to look for signs of the offending agent (either administration of agent in near time to anaphylaxis diagnosis or a specific documented allergen/reaction in same encounter as diagnosis).
• For the sets that exclude environmental causes, consider additional exclusion if an allergy is documented to an environmental allergen with the same encounter or a reasonable follow-up period after the cohort exit.

Adding additional data types: It will be prohibitive for some datasets that may be missing these features, but layering clinical criteria like exam, symptoms, and lab findings may further increase specificity. I don’t think this is necessary addition for the current definitions though.

Opportunities for more sensitivity: Where could this definition be missing patients?

Patients who do not present to the emergency room: While many patients do seek emergency services for anaphylaxis symptoms, many don’t. Reasons include unawareness of mild symptoms, patient self-manage outside of ED with antihistamines or EpiPens, inability to seek emergency care due to a variety of factors including financial, location, etc. It’s unclear how many patients do not present to healthcare for their first event, but one academic allergist’s opinion was that approx. 30% of patients he sees for a new anaphylaxis diagnosis did not present to emergency room for their first event. There are plenty of community stories about families with kids with known anaphylactic allergies sitting in cars outside the ED after giving an EpiPen to their child and if the child gets better, they don’t go into the ED. This behavior may even be supported by recent literature supporting lowering ED utilization after self-administering EpiPens: Retire the Advice to Send Patients to the Emergency Room After Epinephrine Use for Observation - Annals of Allergy, Asthma & Immunology (annallergy.org). Possible ways to quantify the number of initial events missed (either care was sought outside of dataset or patient didn’t seek care at all) could be to evaluate number of patients with a first observed outpatient diagnoses for anaphylaxis or an outpatient EpiPen prescription and see how many lacked evidence of emergent care (treatment or emergency room diagnosis).

Diagnoses considered are too narrow: A significant portion of patients receive anaphylaxis treatment in emergency visits and their diagnoses are less specific like “Allergy, unspecified” or “Rash, unspecified”.

[Patrick_Ryan]

Thanks @Andrea_Noel for your very thoughtful post. It’s really wonderful to have you and your team at Epic-Cosmos meaningfully engaging with the the OHDSI community in this important area.

I appreciate you identifying potential ideas for potentially improving specificity and sensitivity in the candidate definition that @Gowtham_Rao proposed. This could offer the opportunity for you or anyone else in the community the change to author an alternative definition and evaluate its performance, comparing against the definition that is currently on the table. And clearly, if someone were to identify a new definition that was empirically observed to have better sensitivity AND specificity than an older definition, it would be reasonable for that new definition to be preferred across all analytic use cases.

My question for you, same as I prompted @Evan_Minty on the pancreatitis phenotype: Since you are serving as the role of the peer reviewer, if I were to frame as most peer-review journals do: Has the submitted phenotype been 1) accepted for ‘publication’ (as a designated entry in our phenotype library), 2) rejected, or 3) are you recommending a ‘revise-and-resubmit’?

[Kevin_Haynes]

@Andrea_Noel this is a fantastic review, thank you for the detailed clinical review of the clinical phenotype. I had two thoughts.

@Andrea_Noel

A patient’s allergy list should not be used in isolation to define anaphylaxis events, as it can be added any time after events,

In EPIC are entries into the allergy list date stamped such that one might be able to use this as some ‘causal’ assessment of you had a food allergy and then a few days later, next office visit, or close out of an ED stay someone updated your record thus affirming the potential allergen to the ever present ‘unspecified’ billing code?

@Andrea_Noel

self-manage outside of ED with antihistamines or EpiPens,

I know these things expire and we probably all see a spike in dispensing (or in your EHR use case, orders) for these things in May for camps and Aug/Sept for school. But a refill of these could signal a potential for a missed outcome. Can we program my EPIC MyChart app to have a patient MAR (medication administration record) to start to be as smart as my apple watch picking up my false ski accidents? Why Apple Watches Keep Calling 911 - The New York Times
Maybe this could ping my insurer to have them pay me a copay for NOT using expensive ED services for a condition which resolved given treatment and still yield us researchers with the observational nuggets.

Data linkage between EHR/EMR, claims, and the patient directly and indirectly through digital integration will greatly change and enhance our phenotype capabilities both from a better data collection over long periods of time (change in health system change in insurance) and richer data over discrete periods of time (inpatient record with claims pre-post admissions).

[Andrea_Noel]

Thanks, @Patrick_Ryan.

@Gowtham_Rao, I would recommend revise and resubmit if the goal of the phenotype is to capture acute anaphylaxis events. Please consider investigation and addition of several elements from my prior post to increase specificity and/or sensitivity. My guess is across datasets, we may get the highest impact to the definition with:

1. Encounter location restriction (with the understanding that some datasets may have limited ability to identify emergency services from outpatient visits)
2. Medication administration (with understanding that some datasets may have limited ability to recognize medication administration)
3. Adjusted time window to account for diagnosis dates vs treatment dates

If ODHSI collaborators are willing to take up a call to action to investigate a few of these ideas with deeper evaluation and diagnostics, that would be phenomenal.

[Gowtham_Rao]

Thank you @andrea_noel for your review and perspectives on anaphylaxis.

assuming we are trying to define acute anaphylaxis events with these phenotypes and not follow up care for anaphylaxis.

You are correct. After reading your review, I think you bring up three possible scenarios:

1. Directly observed active anaphylaxis (x1): persons are still in anaphylaxis and they are being provided direct care in hospital/ER/urgent care encounter setting. These persons are probably going to get full treatment of epinephrine, steroids, anti-histamine, bronchodilators, fluids etc.
2. Recent anaphylaxis and recovered (x2): they may have had an anaphylaxis in the morning or yesterday, then recovered (probably because they self administered an epi-pen, or had a mild anaphylaxis) - and are now seeking medical care. These individuals may either be monitored or get control measures such as steroids, anti-histamines to prevent short-term reoccurrence (but no epinephrine administration). They are likely to go home with an epi-pen.
3. History of anaphylaxis coming for follow-up care (x3): These individuals are those that have not had a recent anaphylaxis, but had a history of anaphylaxis some point in the past. They may already have an epi-pen at home. The purpose of their visit is to ensure they are prepared for another future possible event (e.g. have an anaphylaxis action plan)

From data coding perspective, the diagnosis codes of ‘anaphylaxis’ may be used for all three of the scenarios. In 1 and 2, the diagnosis may be used to justify the treatment of active or recently resolved anaphylaxis. For 3 it may be to justify the creation or update of the anaphylaxis action plan.

Based on the clinical description (target) - we are looking for 1 and 2, but not 3. i.e.

• if we are missing persons with 1 and 2 in our cohort, we will have SENSITIVITY errors.
• If we are seeing persons with 3 in our cohort we will have SPECIFICITY errors.
• And if we are seeing 2 in our cohort, we may have INDEX DATE MISCLASSIFICATION.

I have noted two additional interesting issues you have brought up that may be source of measurement errors.

1. Another issue you bring up is the potential for overdiagnosis because of the conservative nature of the practice of medicine (x4).

There is no real downside to treating for anaphylaxis while working up these other conditions, as a missed opportunity to treat anaphylaxis may be a greater risk to patients than unnecessary treatment.
A significant portion of patients receive anaphylaxis treatment in emergency visits and their diagnoses are less specific like “Allergy, unspecified” or “Rash, unspecified”.

2. Mild cases not presenting (x5)

While many patients do seek emergency services for anaphylaxis symptoms, many don’t. Reasons include unawareness of mild symptoms, patient self-manage outside of ED with antihistamines or EpiPens

Here 1 may lead to specificity errors, while 2 may lead to sensitivity errors. Specifically

• For 2 (mild cases) - I think getting an estimate of persons who are first observed in a TRUE physician office visit (x5 below) for a consultation on anaphylaxis without acute care (as evidenced by first diagnosis of anaphylaxis and/or first dispensation of epi-pen but no ER or Inpatient visit temporally close) maybe a reasonable surrogate for sensitivity errors.
• For 1 (over diagnosis) - I would argue that some of these over diagnosis may be TRUE anaphylaxis, and thus not all wrong. In other words, if a licensed clinician had enough probable cause to administer a definitive treatment for anaphylaxis to a person with ‘Allergy, unspecified’ or ‘Rash, unspecified’ - then I would based on the popular ‘Duck test’ I would find it hard to refute or second guess the judgment of the professional at the time of care.

So to detect and potentially estimate such errors, one approach may be to build cohorts of persons who satisfy the x1:x5 above - and quantify by how much x1:x5 overlap with the submitted cohortIds 259 and 251. Logically - the cohorts would be

x1: persons with diagnosis of anaphylaxis that overlaps with an inpatient or ER or urgent care.(potential TRUE positive)
x1a: (persons with diagnosis of anaphylaxis that overlaps with an inpatient or ER or urgent care) AND (an IM administration of epinephrine in that care event).(potential TRUE positive)

x2: persons with diagnosis of anaphylaxis that overlaps with an inpatient or ER or urgent care who received either fluids, steroids, anti-histamine, bronchodilators and a prescription/dispensation of epi-pen but not administration of epinephrine.

x3: persons with diagnosis of anaphylaxis that overlaps with a TRUE outpatient physician office visit and no ER or inpatient visit (potential TRUE negative)

x4: persons without diagnosis of anaphylaxis but have a diagnosis of rash, allergy, swelling that overlaps with an inpatient or ER or urgent care who received an IM administration of epinephrine. (potential False negative)

The theory here is that if we are able to identify such cohorts, AND check for temporal co-occurrence of such cohorts (x1:x5) with the primary cohort id 259 or 251, we may get directional estimate of sensitivity, specificity and index date misclassification errors.

@Andrea_Noel

If ODHSI collaborators are willing to take up a call to action to investigate a few of these ideas with deeper evaluation and diagnostics, that would be phenomenal.

What do you all think OHDSI community - is this a reasonable plan? If yes, would you like to collaborate and lead an effort to build such cohorts? Do you have another cohort design in mind that may help answer @andrea_noel review? Do you want to be part of a OHDSI network study?

Looking for collaborators who want to:

1. Build cohort definitions based on the logic description above. or, propose an alternative logic description and build cohort definitions (you will gain experience building cohorts for a real clinical question)
2. Execute the cohort on your data sources and return results (you will gain experience running OHDSI software like CohortDiagnostics, PheValuator, Atlas Cohort Characterization functionality and FeatureExtraction, CohortIncidence)
3. Be part of an OHDSI network study - ofcourse, 1 and 2 above is a OHDSI network study. Maybe we can collaboratively write our collective experience and contribute to the scientific body of work on anaphylaxis.

[david_vizcaya]

@Gowtham_Rao

What do you all think OHDSI community - is this a reasonable plan? If yes, would you like to collaborate and lead an effort to build such cohorts? Do you have another cohort design in mind that may help answer @andrea_noel review? Do you want to be part of a OHDSI network study?

@Gowtham_Rao @Andrea_Noel Challenge accepted!! I can run Cohort diagnostics in MScan CCAE and MDCR and OPTUM Claims once you have the Jsons ready.
CCing a couple of Spanish colleagues who will (hopefully) come along with me @mmayer @Marcela

[hripcsa (George Hripcsak)]

I like the design. I suspect we can run it.

[mmayer (Miguel A. Mayer)]

Thanks @david_vizcaya @Gowtham_Rao we can run in our instance in Parc de Salut Mar (Barcelona), it is a hospital-based EHR, so point 2 (point 3 as well). I would like helping in point 1 (building cohort definitions), but I am not sure what the dynamics of this point is or how to proceed…

[Gowtham_Rao]

@mmayer

(building cohort definitions), but I am not sure what the dynamics of this point is or how to proceed

Thank you @mmayer

Everyone - we now have a PhenotypePhebruary2023 study package (DO NOT RUN IT YET ) Our next steps are:

1. This Friday February 10th at 9am EST as part of the OHDSI Phenotype Development Evaluation Workgroup we will attempt to collaboratively develop the proposed cohort definitions. This meeting may have to be longer than usual.
2. The cohorts we build will go to https://github.com/ohdsi-studies/PhenotypePhebruary2023 (target Saturday).
3. Execute network study (target Monday February 13th)
4. Review results (target Friday February 17th)

Looking forward to working with you all.

[Gowtham_Rao]

Today @Andrea_Noel @Evan_Minty @Gowtham_Rao will collaborate to attempt to create the cohort definitions for x1:x5 at 12:30pm EST. The meeting invite is here

[Gowtham_Rao]

@Andrea_Noel , @Evan_Minty , @Gowtham_Rao , @mmayer , @david_vizcaya , @bebray and another 15 collaborators empirically evaluated if we have evidence of specificity errors as anticipated by @Andrea_Noel . Specifically, as described below

@Gowtham_Rao

History of anaphylaxis coming for follow-up care (x3):

We wanted to check if there are persons getting the diagnosis code of anaphylaxis - to justify the creation, update of an anaphylaxis action plan (e.g. a person with allergies getting an epi pen and a diagnosis of anaphylaxis), i.e. they are not TRUE cases of anaphylaxis and introduce specificity errors.

To quantify this type of specificity error - we built a cohort definition that attempts to identify persons receiving care in a ‘TRUE outpatient physician office’ with the diagnosis code of anaphylaxis. To prevent counting persons who may be having a follow-up visit immediately after an acute anaphylaxis event, we also limited to persons who did not have an inpatient or visit visit or care within +/-1 day.

Detailed analysis to follow - but briefly (full output here https://data.ohdsi.org/PhenotypePhebruary2023/) [network study data partners we are still refining cohort definitions and study package - so please do not run it yet)

1. We observe that the number of persons/events that potentially met the this ‘TRUE outpatient physician office’ to be quite large - in many cases about 70%. See original definition - ‘[P] Anaphylaxis all cause (1Pe, 30Era) C259’ vs. new ‘[P] x3 Anaphylaxis In A TRUE Doctors Office C314’
   

   

2. We then built a cohort definition for events that belonged to the rest of 30% i.e. ‘C316: [P] x1 Anaphylaxis In an inpatient or emergency setting’. These two cohorts are not perfectly mutually exclusive (there was a small <3% overlap), but they were sufficiently mutually exclusive to draw reliable inference. On review of the compare characteristics the two cohorts appear very different - especially for drug treatments as shown in scatter plot and table below were different
   

   

Our main insights here are:

1. We more likely than not to have specificity errors.
2. Quantitatively this is large enough that we need to investigate and attempt to improve the cohort definitions.

What do you all think we should do next?