OHDSI Home | Forums | Wiki | Github

Phenotype Phebruary 2023 - P4 - Acquired Neutropenia

Building on previous work in Phenotype Phebruary 2022

Clinical Description

Defects in neutrophil function can be quantitative, as seen in neutropenia or qualitative, as seen in neutrophil dysfunction. This phenotype is quantitative decrease in neutrophils in persons who previously had normal neutrophil count. Neutropenia is ANC <1500 cells/microL in an adult while agranulocytosis is a more severe form of neutropenia with ANC <200 cells/microL. Note: pancytopenia may include Neutropenia. In this definition, any person with a quantitatively low neutrophil count for any cause is considered to have Neutropenia. Neutropenia may occur due to cytotoxic or marrow toxic drugs (chemotherapy), or in setting of certain diseases like Antineutrophil or autoimmune neutropenia has been observed in autoimmune diseases like Rheumatoid arthritis (Felty syndrome), Inflammatory bowel disease, Sjogren syndrome, Chronic autoimmune hepatitis, Granulomatosis with polyangiitis, or infections like hepatitis, HIV, EBV, HPV, Nutritional deficiency like Hodgkin lymphoma. Presentation: The neutropenia is often asymptomatic or presents with an insidious onset. history of recurrent infections, infections by rare bacteria and fungi, opportunistic infections with observation of low to absent absolute neutrophil count. The clinical presentation is usually oral ulcerations with or without fever, but occasionally sepsis may be the initial presentation. Fever, mouth sores, or gingival disease inflammation. Plan: Find and treat/prevent exposure to offending agent. Granulocyte-colony stimulating factor (G-CSF) may be used is some scenarios. Prognosis: Neutropenia longer than 2 months is considered chronic neutropenia and since most chemotherapy induced neutropenia is expected to last about 3 weeks, we expect the median to be about 21 days in duration.
Disqualifiers: Normal neutrophil count or elevated neutrophil count that is concurrent is considered not biologically plausible and may indicate bad data. Also persons wtih Familial/congenital neutropenia, benign ethnic neutropenia all time - cannot develop acquired neutropenia (as they may never have had normal neutrophils). Persons post bone marrow transplant are also considered not eligible for aqcuired neutropenia.

Designated Medical Event - MedDRA PT terms:

Granulocytopenia, Has neutropenia as a component: Aplastic anemia, Bone marrow failure, Pancytopenia; Neutropenic colitis, Neutropenic infection, Neutropenic sepsis

Phenotype development:

Concept set expression creation: Terms Neutropenia vs Leukopenia: [Decision] Since Neutrophils are the major component of white blood cells (~75%) and the occurrence of having normal neutrophil counts in the presence of low white blood cell count is rare (i.e. we can expect neutropenia in all persons with leukopenia/granulocytopenia) – we decided to consider granulocytopenia and leukopenia as synonyms. However, we decided to not consider the terms lymphocytopenia, eosinopenia as synonyms for neutropenia – as they clearly represent granulocytes that are not neutrophils. An

Use of laboratory results in cohort definitions: [Decision] We decided to use the American Board Of Internal Medicine 2022 laboratory test reference range in the definition, which are as follows: Absolute neutrophil count (ANC) – 2000–8250/μL, 2000 to 8250/microL, 2 to 8.25 x 10/L (SI); Leukocyte count 4000 to 11,000/microL, 4.5 to 11 x 109/L (SI); Segmented neutrophils: 50 to 70 percent. Agranulocytosis < 200.

Congenital or Hereditary Neutropenia vs Acquired Neutropenia: [Decision] Persons with neutropenia may have neutropenia because of congenital or genetic conditions. Those persons cannot have acquired neutropenia. We decided to define Acquired Neutropenia as those persons who do not have a diagnosis of congenital or acquired neutropenia.

Out of scope: Single cell line vs Multiple cell lines: [Decision] Neutropenia may be part of multiple cell line penia’s like pancytopenia’s related to peripheral destruction or (more commonly) bone marrow failure/aplastic anemia. We discussed that persons with neutropenia limited to one cell line vs multiple cell lines were two different sub-types of neutropenia. However, phenotyping these subsets are out of scope of this submission.

Out of scope: Degree of neutropenia: [Decision] Because degree of severity of neutropenia correlates with their risk of infection, phenotyping neutropenia sub-types by degree of severity is important. Severe reduction of neutrophils is agranulocytosis. sub-types by degree of neutropenia is out of scope of this submission.

Data quality: [Decision] We observed that in rare occasions for a person to have a lab result corresponding to neutropenia, while another lab result on the same day is in the normal range. We decided that these may represent FALSE positives, potentially due to lab errors – and decided to remove those individuals.

Cohort Exit [Decision] We assumed that most neutropenia will resolve by around 3 weeks, hence we used a cohort exit persistence criteria of 21 days. We also assumed persons with another episode of neutropenia within 365 days were part of the original neutropenia (e.g. repeat chemotherapy).


We submit the cohort definition # 213 to the OHDSI phenotype library which is currently in peer review status. C213: [P] Acquired Neutropenia or unspecified leukopenia (21Pe, 365Era)
Logic Description: all events of neutropenia indexed on diagnosis or lab results with no congenital or genetic neutropenia. Events with concomitant neutrophilia or normal neutrophil counts would be considered bad data and removed.

Phenotype evaluation:

  • I observed that we have about 1.1 to 1.2 events per persons with 75%ile of person entries having time in cohort of 21 days. Since 21 days of neutropenia is biologically plausible, I think the cohort exit strategy that utilized 21 days is collapsing any follow-up visits when present.
  • Older ages had higher incidence rate compared to younger age, with females > males.

  • on review of characterization data, i observed that about 40% of persons may have received chemotherapy.

and a vast majority of persons had neoplastic disease

this suggests that the majority of the persons with neutropenia are probably neoplasm or antineoplastic treatment related neutropenia.

Also prevalent in this population were autoimmune and immunodeficiency conditions (note: congenital neutropenia are removed by design, implying these are all acquired immunodeficiency)

We also observe the use of filgrastim/pegfilgrastim in such persons

Finally we observe some index date misclassificaiton

Other finding in some data sources, the code of neutropenia was not used. It appeared the term of agranulocytosis was used to represent the idea of neutropenia. Reason for this is unclear.

Thanks so much, @Gowtham_Rao!

To begin, I want to clarify that we are talking about Acquired Neutropenia here (it took me way to long to figure it out).
Here are some comments broken down into the corresponding sections of the submission:

Clinical description

I appreciated an extensive description. When going through the description I kept thinking about the rationale behind the phenotype. As described, there are multiple different causes for neutropenia, including cancer or drug toxicity. Neutropenia can vary in severity and length. I’d ideally like to know in which studies this phenotype is meant to be used to better assess the appropriateness of it. For example, I wasn’t sure why bone marrow transplant was a disqualifier in clinical description (even though the cohort definitions did not exclude patients with bone marrow transplant).

I liked the section on strengtheners which are things to watch out for in patient characterization and which are indicative of neutropenia (drugs, procedures, associated conditions).

I liked the following section: “Data quality: [Decision] We observed that in rare occasions for a person to have a lab result corresponding to neutropenia, while another lab result on the same day is in the normal range. We decided that these may represent FALSE positives, potentially due to lab errors – and decided to remove those individuals. “

Would be happy to see such an assessment to be a part of the standardized approach that we recommend in measurement-based phenotypes.

Concept sets and logic

I looked at the 2 cohorts on https://data.ohdsi.org/PhenotypeLibrary/ (213 and 214, neutropenia and isolated neutropenia).

Concept sets

Concept sets were quite comprehensive except for ‘Congenital or genetic neutropenia, leukopenia or agranulocytosis’ where I didn’t see some of the disorders mentioned in the clinical description (such as Chediak-Higashi syndrome or chronic granulomatous disease).

Units: I liked seeing multiple options for units for neutrophil count, it seems that a data-driven approach was used there (highly recommend!).


Acquired neutropenia cohort excluded those who had a diagnosis of genetic neutropenia all time prior or 7 days after. Not clear why exactly 7 days after and why people who had a congenital diagnosis on day 8 would be included.

Patient characteristics and performance


Mostly saw plausible things as described in Gowtham’s post. Not a lot of autoimmune disorders at all. Cohorts 213 and 214 do not differ substantially in patient number or characteristics, which was to be expected.


PheValuator was ran on 7 databases. I wasn’t sure how to directly connect the cohorts in Phenotype Library to PheValuator output.

Compared to one previous study that reported performance (reported in the literature review) sensitivity is much higher across the board, which is nice. Some of the cohorts were execute only on some databases, not clear why exactly. I observed some database-specific patterns (jmdc has the highest performance and Optum EHR has the lowest performance). I would be interested in digging deeper in the phenomenon. If I’m not mistaken, optum ehr was the only EHR data source, so maybe it’s due to the difference in data capture.

Specificity was very high, sensitivity was on average 0.75 and was the lowest for Acquired Isolated Neutropenia (no leukopenia as entry) and the highest for Acquired Neutropenia (no leukopenia as entry) including Agranulocytosis, which makes me think that the latter is the best phenotype if you want to be inclusive of different types of neutropenia.

Honestly, PheValuator was a deal-breaker for me. Yes, Optum EHR was a weird outlier but there still was a clear trend in comparative performance and a clear winner (and much better performance to what previously reported). In fact, as soon as I saw that patient characteristics are similar across the definitions and one definition had sensitivity and specificity better than the others I was quite confident about choosing the best performing phenotype as the one for my hypothetical study.

Thank you @aostropolets , this is helpful to see your peer review of each of the components. You highlight ‘congenial or genetic neutropenia’, and that seems to be something that should be clarified in the clinical definition. It will warm @jswerdel 's heart to know that PheValuator results were helpful in your assessment in comparing alternative definitions.

Agree with you @aostropolets , adding measurements to the definition is always a heavy lift and we always seem to question the value of it. It is of note that when we developed this pheonotype for Neutropenia we first assessed the values and the concepts available in the data sources, where @Dymshyts helped us select the appropriate concepts and values. If this phenotype to be used in another data sources, one will need to make sure that the included concepts and values cover the new data.