ED Classification

Overview

Code

The ED Classification data mart implements the 2017 update to the original NYU ED Classification Algorithm to identify potentially preventable ED visits.

Instructions

This mart builds off of the Core Encounter table and classifies emergency department encounters into the following categories:

Classification	Description
alcohol	Alcohol Related
drug	Drug Related (excluding alcohol)
emergent_ed_not_preventable	Emergent, ED Care Needed, Not Preventable/Avoidable
emergent_ed_preventable	Emergent, ED Care Needed, Preventable/Avoidable
emergent_primary_care	Emergent, Primary Care Treatable
injury	Injury
mental_health	Mental Health Related
non_emergent	Non-Emergent
unclassified	Not in a Special Category, and Not Classified

dbt Examples

# Runs all marts
dbt build

# Runs only the ED Classification mart
dbt build --select tag:ed_classification

Data Dictionary

summary

ED Classification as a cube that can be summarized.

Column	Data Type	Description	Terminology

Analytics

Analyzing ED claims data helps identify high utilizers of emergency services, often indicating overuse of EDs for conditions that can be managed with proper primary care.

ED Classification

select 
      coalesce(s.ed_classification_description,'Not Classified') as ed_classification_category
    , count(*) as visit_count
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
    left join ed_classification.summary s 
        on e.encounter_id = s.encounter_id
group by coalesce(s.ed_classification_description,'Not Classified')
order by visit_count desc;

Members with at least One Potentially Preventable ED Visit

with encounter as (
    select
          e.patient_id
        , left(year_month,4) as year_nbr
        , data_source
        , count(distinct e.encounter_id) as potentially_preventable
        , sum(e.paid_amount) as paid_amount
    from core.encounter e
    inner join ed_classification.summary s
        on e.encounter_id = s.encounter_id
    where ed_classification_description in (
          'Emergent, Primary Care Treatable'
        , 'Non-Emergent'
        , 'Emergent, ED Care Needed, Preventable/Avoidable')
    group by
          e.patient_id
        , data_source
        , left(year_month,4)
)

, member_year as (
    select distinct
          data_source
        , left(year_month,4) as year_nbr
        , patient_id
    from financial_pmpm.pmpm_prep pmpm
)

select
    my.data_source
    ,my.year_nbr
    ,sum(case when enc.potentially_preventable >=1 then 1 else 0 end) as members_with_potentially_preventable
    ,count(*) as total_members
    ,sum(case when enc.potentially_preventable >=1 then 1 else 0 end)/count(*) as potentially_preventable_percent_of_total
    ,sum(enc.paid_amount)/sum(enc.potentially_preventable) as avg_cost_potentially_preventable
from member_year as my
    left join encounter as enc
        on my.year_nbr = enc.year_nbr
        and enc.data_source = my.data_source
        and enc.patient_id = my.patient_id
group by
      my.data_source
    , my.year_nbr;

Primary Diagnosis Codes for Avoidable Categories

select
      coalesce(s.ed_classification_description,'Not Classified') as ed_classification_category
    , e.primary_diagnosis_code
    , e.primary_diagnosis_description
    , count(*) as visit_count
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
    left join ed_classification.summary s
        on e.encounter_id = s.encounter_id
where ed_classification_description in (
      'Emergent, Primary Care Treatable'
    , 'Non-Emergent'
    , 'Emergent, ED Care Needed, Preventable/Avoidable')
group by
      coalesce(s.ed_classification_description,'Not Classified')
    , e.primary_diagnosis_code
    , e.primary_diagnosis_description
order by
      ed_classification_category
    , visit_count desc;

ED Visits Trended

Trending ED Visit Volume, PKPY, and Cost

with ed as (
    select
          data_source
        , TO_CHAR(encounter_end_date, 'YYYYMM') AS year_month
        , COUNT(*) AS ed_visits
        , AVG(paid_amount) as avg_paid_amount
        , sum(paid_amount) as total_paid_amount
    from core.encounter
    where encounter_type = 'emergency department'
    group by
          data_source
        , TO_CHAR(encounter_end_date, 'YYYYMM')
)

, member_months as (
    select
          data_source
        , year_month
        , count(1) as member_months
    from financial_pmpm.member_months
    group by
          data_source
        , year_month
)

select
      a.data_source
    , a.year_month
    , b.member_months
    , ed_visits
    , cast(ed_visits / member_months * 12000 as decimal(18,2)) as ed_visits_pkpy
    , cast(avg_paid_amount as decimal(18,2)) as avg_paid_amount
    , cast(total_paid_amount as decimal(18,2))as ed_total_paid_amount
from  member_months b
left join ed a
  on a.year_month = b.year_month
  and a.data_source = b.data_source
order by
      a.data_source
    , a.year_month;

ED Spend as Percent of Total Spend

select
      data_source
    , year_month
    , sum(emergency_department_paid) as ed_paid
    , sum(total_paid) as total_paid
    , cast(sum(emergency_department_paid) as decimal(18,2))/cast(sum(total_paid) as decimal(18,2)) as ed_percent_of_total_paid
from financial_pmpm.pmpm_prep
group by
      data_source
    , year_month
order by
      data_source
    , year_month;

ED Visits by Member and Year

select
      data_source
    , year_month
    , sum(emergency_department_paid) as ed_paid
    , sum(total_paid) as total_paid
    , cast(sum(emergency_department_paid) as decimal(18,2))/cast(sum(total_paid) as decimal(18,2)) as ed_percent_of_total_paid
from financial_pmpm.pmpm_prep
group by
     data_source
    , year_month
order by
      data_source
    , year_month;

Frequency Distribution of ED Visits

with visits as (
    select
          data_source
        , patient_id
        , count(*) as ed_visits
    from core.encounter
    where encounter_type = 'emergency department'
    group by
          data_source
        , patient_id
)

, members as (
    select distinct
          patient_id
        , data_source
    from financial_pmpm.member_months
)

, members_total as (
    select count(*) as total_member_count
    from members
)

, members_with_visits as (
    select
          m.patient_id
        , m.data_source
        , coalesce(v.ed_visits,0) as ed_visits
    from members m
        left join visits v
            on m.patient_id = v.patient_id
            and m.data_source = v.data_source
)

select
      ed_visits
    , count(*) as member_count
    , count(*) / cast(max(total_member_count) as real) as percent_of_total_members
from members_with_visits
cross join members_total
group by ed_visits
order by ed_visits;

Count of ED NPIs

select 
      data_source
    , count(distinct facility_id) as ed_facilities_count
from core.encounter e
where encounter_type = 'emergency department'
group by data_source
order by ed_facilities_count desc;

Visit by Facility

select
     facility_id
    , count(*) AS ed_visits
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
where encounter_type = 'emergency department'
group by facility_id
order by ed_visits desc;

Admit Source and Type

select
      admit_source_code
    , admit_source_description
    , admit_type_code
    , admit_type_description
    , count(*) AS ed_visits
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
where encounter_type = 'emergency department'
group by
      admit_source_code
    , admit_source_description
    , admit_type_code
    , admit_type_description
order by ed_visits desc;

ED Diagnosis Grouping

The Tuva Project provides several ways of grouping diagnosis codes.

CCSR provides a hierarchy grouping of diagnosis codes, and is useful for recognizing patterns of care by what the patient was diagnosed with at the ED.

Chronic Conditions are a way of grouping members by conditions that they have been diagnosed with (within the relevant timespan, usually the last one or two years.)

ED Visits by CCSR Category and Body System

select
      p.ccsr_category
    , p.ccsr_category_description
    , p.ccsr_parent_category
    , p.body_system
    , count(*) as visit_count
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
    left join ccsr.long_condition_category p
        on e.primary_diagnosis_code = p.normalized_code
        and p.condition_rank = 1
where e.encounter_type = 'emergency department'
group by
      p.ccsr_category
    , p.ccsr_category_description
    , p.ccsr_parent_category
    , p.body_system
order by visit_count desc;

ED Visits by Chronic Condition Category

Since members often have more than one chronic condition, encounters are duplicated for each chronic condition causing the total amount to be inflated. The division of encounters by chronic condition is useful for comparision across disease states, and less so from the total standpoint.

with chronic_condition_members as (
    select distinct
    patient_id
    from chronic_conditions.tuva_chronic_conditions_long
)

, chronic_conditions as (
    select patient_id
    , condition
    from chronic_conditions.tuva_chronic_conditions_long

    union

    select p.patient_id
    , 'No Chronic Conditions' as condition
    from core.patient p
        left join chronic_condition_members ccm
            on p.patient_id=ccm.patient_id
    where ccm.patient_id is null
)

select
      cc.condition
    , count(*) as visit_count
    , sum(cast(e.paid_amount as decimal(18,2))) as paid_amount
    , cast(sum(e.paid_amount)/count(*) as decimal(18,2))as paid_per_visit
from core.encounter e
    left join chronic_conditions cc
        on e.patient_id = cc.patient_id
where encounter_type = 'emergency department'
group by cc.condition
order by visit_count desc;