Inpatient Claims

Attribute Details
Source Name claims_inpatient
Data Source De-identified medical billing claims from Delphi health system partners
Geographic Levels county, hrr, msa, state, hhs, census_division, census_region, nation
Temporal Granularity Daily, by date of service
Reporting Cadence Daily
Temporal Scope Start 2019-10-01; the other-ARI signal begins 2022-08-01
Temporal Scope End Ongoing
Extra Key Columns None
License CC BY 4.0

This source reproduces the legacy V4 hospital-admissions source. It keeps the same claims feed and definition of a respiratory admission, but replaces the day-of-week and backfill adjustments with a plain 7-day windowed ratio, adds influenza and other-ARI signals, and serves revision history through the /archive/ and /snapshot/ endpoints. See Relationship to V4.

Table of contents

  1. Overview
  2. Indicators (Signals)
  3. Estimation
    1. Metric Definition
    2. Qualifying Conditions
      1. Signal Composition
    3. Smoothing
    4. Temporal Handling
    5. Geographic Handling
  4. Relationship to V4
  5. Schema
    1. Columns
  6. Missingness & Privacy
  7. Limitations
  8. Lag & Backfill
  9. Source and Licensing

Overview

This source measures the share of inpatient hospital admission claims that carry a respiratory illness diagnosis. It is built from de-identified medical billing claims contributed by Delphi health system partners and covers COVID-19, influenza, and other acute respiratory illness (ARI).

Indicators (Signals)

Indicator Name Pathogen or Disease Metric Type Description
claims_inpatient_adm_pct_claims_covid COVID-19 Percentage, 7-day window Share of inpatient admission claims with a COVID-19 diagnosis.
claims_inpatient_adm_pct_claims_flu Influenza Percentage, 7-day window Share of inpatient admission claims with a confirmed influenza diagnosis.
claims_inpatient_adm_pct_ari_other Other ARI Percentage, 7-day window Share of inpatient admission claims with an acute respiratory illness diagnosis not attributed to COVID-19 or influenza. History for this signal begins 2022-08-01.

Estimation

Metric Definition

For location \(i\) and service date \(t\), let \(W_t = \{t-6, \dots, t\}\) be the trailing 7-day window, \(N_{is}\) total admission claims on day \(s\), and \(Y_{is}^k\) claims carrying diagnosis group \(k\). The published percentage is

\[\hat{p}_{it}^k = 100 \cdot \frac{\sum_{s \in W_t} Y_{is}^k}{\sum_{s \in W_t} N_{is}},\]

computed only when at least 5 days contribute and \(\sum_{s \in W_t} N_{is} \ge 100\).

Qualifying Conditions

The raw claims feed categorizes unique daily hospital admissions into five mutually exclusive count categories based on primary ICD-10 diagnosis codes. When a patient admission carries multiple qualifying diagnosis codes on the same date, priority is assigned in descending order of diagnostic specificity: Flu, COVID-like, Flu-like, then Mixed.

Category Primary ICD-10 Codes Description
Denominator All Total count of all unique inpatient admission claims.
Flu J09*, J10*, J11* Confirmed influenza diagnoses (* includes all subcodes).
COVID-like U07.1, U07.2, B97.29, J12.81, Z03.818, B34.2, J12.89 Confirmed or suspected COVID-19 diagnoses.
Flu-like J22, B34.9 Unspecified acute lower-respiratory infections and unspecified viral infections. Correlated with respiratory illness but not specific to influenza; tracked COVID-19 early in the pandemic before specific codes were widely adopted.
Mixed Z20.828, J12.9 Suspected exposure to other viral communicable diseases and viral pneumonia, unspecified.

Signal Composition

These raw categories map to the three published V5 indicators as follows:

  • COVID-19 (pct_claims_covid):
    • Before 2021-06-01: Sums COVID-like and Flu-like counts over the denominator. The Flu-like (unspecified lower-respiratory) category tracked COVID-19 closely during the early pandemic while specific codes were not yet in widespread use.
    • From 2021-06-01 onward: Uses COVID-like counts only over the denominator.
  • Influenza (pct_claims_flu): Uses the confirmed Flu count over the denominator across the entire series.
  • Other ARI (pct_ari_other): Published from 2022-08-01 onward. Combines the Mixed and Flu-like counts over the denominator, capturing acute respiratory illness encounters not attributed to COVID-19 or confirmed influenza.

Smoothing

For each geography, a 7-day trailing sum is taken of every diagnosis count and the total claim count. The window must contain at least 5 of 7 days to produce a value.

Temporal Handling

Dates refer to the clinical date of service. Claims dated to the first calendar day of any month are dropped, because submission batching inflates that day. Service dates before 2019-10-01, and service dates in the future, are also dropped.

Geographic Handling

Claims arrive at two native geographic levels: county (county) and Hospital Referral Region (hrr). HRR values are served as reported.

All other geographic levels (msa, state, hhs, census_division, census_region, nation) are built by directly summing raw daily county counts (numerators and denominators are aggregated separately). Aggregation is performed on raw daily counts before computing 7-day sums, without imputation, so fill_method is always source.

Relationship to V4

The V4 hospital-admissions source used the same health system claims and published a single COVID-19 percentage, through smoothed_covid19_from_claims and its day-of-week-adjusted twin smoothed_adj_covid19_from_claims.

V4 Signal V5 Signal Notes
smoothed_covid19_from_claims claims_inpatient_adm_pct_claims_covid Replaced smoothed Jeffreys estimator with 7-day trailing ratio
smoothed_adj_covid19_from_claims (not available) Day-of-week adjustment discontinued
smoothed_covid19 (not available) Discontinued in V4
smoothed_adj_covid19 (not available) Discontinued in V4
(not available) claims_inpatient_adm_pct_claims_flu New in V5
(not available) claims_inpatient_adm_pct_ari_other New in V5 (available from 2022-08-01)

What changed in V5:

  • Expanded Geographies. V4 served nation, state, county, HRR, and MSA. V5 adds HHS regions, census divisions, and census regions.
  • Broader Signal Set. V5 adds influenza and other-ARI percentages next to COVID-19, and drops the combined electronic-record-and-claims variant that V4 froze in 2020.
  • Revised Estimator. V4 applied a Jeffreys-style \(\frac{Y + 0.5}{N + 1}\) correction, a backwards-padding step that widened the averaging window until a 500-visit threshold was met, a Poisson day-of-week adjustment, and a Gaussian linear smoother. V5 replaces all of it with the 7-day trailing sum from Metric Definition.

Schema

Columns

Column Key Type Data Type Description
signal Primary Key string The name of the requested indicator.
report_time Primary Key date The publication or release date (YYYY-MM-DD).
geo_type Primary Key string Geographic level (e.g., county, state, nation); see the summary table above for the full list.
geo_value Primary Key string Unique code for the location (e.g., pa for Pennsylvania, 42003 for Allegheny County).
fill_method Primary Key string Imputation method used during geographic aggregation, always source for this source.
reference_time Primary Key date The date or surveillance period represented by the observation (YYYY-MM-DD, date of service).
value Value Column float The recorded percentage of admission claims over the trailing 7-day window.

Missingness & Privacy

To protect patient privacy, observations with fewer than 100 total claims across the trailing 7-day window are suppressed and omitted from publication. Windows with fewer than 5 days of data are withheld due to data sparsity and are likewise omitted. Missing days within an otherwise-qualifying window (at least 5 days and at least 100 claims) do not block publication.

Unobserved dates or geographies without reporting claims produce no records. Both suppressed and unobserved values are omitted entirely from query responses rather than reported as null.

Limitations

This source only captures admissions from healthcare facilities that partner with Delphi, which does not cover all hospitals in the United States. Coverage density varies across states and counties.

The data reflects insured patients admitted to participating inpatient facilities. Uninsured individuals and admissions at non-participating institutions are not represented.

Standard errors and sample sizes are not published for this source.

Differences in local coding habits, billing practices, and facility participation mean that absolute levels are not always comparable across different regions. Trends within a single location over time are generally more reliable than spatial comparisons.

Lag & Backfill

Hospital admissions are reported with several days of lag, so estimates for a given date of service typically become available 3 to 7 days later.

Insurance claims experience substantial backfill over several weeks as facilities and payers submit and process claims on varying schedules. Delphi runs daily updates across historical dates to incorporate late-arriving records.

Estimates for the most recent 7 to 13 days change significantly across subsequent revisions, often showing median shifts of 10 percent or more. Data points older than 57 days are generally stable, with subsequent median revisions under 1 percent. Full revision histories can be queried through the /archive/ and /snapshot/ endpoints. Delphi documented the empirical behavior of this backfill process in a blog post on claims backfill.

Source and Licensing

De-identified medical billing claims contributed by Delphi health system partners. This dataset is made available under CC BY 4.0.