Analyzing Physician-Patient Referral Network Topology

About Us

Welcome to APPARENT, our interactive dashboard for analyzing healthcare networks. By providing access to precomputed descriptions of patient referral networks, we aim to help users better understand, analyze, and improve care delivery in the United States. Our mission is to put advanced analytical methods from graph theory, network science, and machine learning at the fingertips of healthcare professionals. Use APPARENT to explore local or global trends in referral networks with our built-in dashboards, or design your own SQL queries to directly access the underlying database. For more advanced users, we also maintain the APPARENT Python library for custom construction and analysis of referral networks.

Overview

What is a Hospital Service Area?

This database is organized by Hospital Service Area (HSA), the geographic designation within which patients receive most of their hospital care.

There are nearly 3,500 HSAs in the United States, an average of ~70 per state. Each HSA is assigned a unique identification number, or HSA_ID. Check out our map to find the HSA to which you belong.

Who defines Hospital Service Areas?

The HSA designations are maintained by the Dartmouth Atlas of Health Care, which provides a wealth of HSA-level data.

For an example of this data and a visualization of the HSA boundaries in the US, see: Medical Discharges by HSA.

More details on HSA designations can be found here.

Find Your HSA

What is a Physician Referral Network?

For each HSA, we construct a referral network: a graph/web with vertices representing the individual providers and edges connecting providers who have referred patients to one another. These referral networks are built based on data from a Medicare claims.

These networks can be studied to provide insights into the connectivity and flow of patients through the local healthcare system.

Additional Resources on Physician Referral Networks

To learn more about how referral networks are built and the mathematics behind their analysis, see the following paper:
Characterizing Physician Referral Networks with Ricci Curvature, by Jeremy Wayland, Russel J. Funk, and Bastian Rieck (2024).

For developers, the code repository for constructing and analyzing these referral networks is hosted here on GitHub:
Analysing Physician-Patient Referral Network Topology

Dashboards

State-Level Analysis

View aggregated data for Hospital Service Areas (HSAs) by state.

Go to Dashboard

HSA-Level Analysis

Enter an HSA_ID to investigate an individual Hospital Service Area in detail.
To find an HSA_ID, reference the map or the State-Level Analysis dashboard above.

Go to Dashboard

SQL Queries

Example SQL Queries

We provide the following example queries to help you get started.

Tailor these to your liking or develop your own queries to explore the data in more detail.

Query the Database

Network Feature Data

        SELECT
          /* SELECTING AND NAMING RELEVANT COLUMNS */
          forman_mean as care_convergence_index,
          average_clustering as mutual_care_index,
          nnodes as num_providers,
        FROM
          referral_network_features
        WHERE
          /* FILTERING ROWS */
          /* Only returns desired years */
          date IN (2015, 2016, 2017)
          /* Will ask for user to input state code (e.g., VA) */
          AND HSA_ID IN (
            SELECT
              hsa
            FROM
              hospital_atlas_data
            WHERE
              TRUE
              AND pstate = :state
          )
        ORDER BY
          /* SORTING */
          HSA_ID, date
        LIMIT
          /* NUMBER OF ROWS RETURNED */
          10000

Demographic Data

        SELECT
          /* SELECTING AND NAMING RELEVANT COLUMNS */
          population_census.hsa as HSA_ID,
          population_census.year as date,
          population_census.median_hh_income,
          (race_black_pop * 1.0 / race_total_pop) * 100 AS percent_black,
          (hispanic_pop * 1.0 / hispanic_total_pop) * 100 AS percent_hispanic,
          (population_census.employment_pop * 1.0 / population_census.employment_unemployed_pop) * 100 as unemployment,
          (population_census.education_nohs_pop * 1.0 / population_census.education_pop) * 100 as percent_without_hs_diploma
        FROM
          population_census
        WHERE
          /* FILTERING ROWS */
          /* Only returns desired years */
          date IN (2015, 2016, 2017)
          /* Will ask for user to input state code (e.g., VA) */
          AND HSA_ID IN (
            SELECT
              hsa
            FROM
              hospital_atlas_data
            WHERE
              TRUE
              AND pstate = :state
          )
        ORDER BY
          /* SORTING */
          HSA_ID,
          date
        LIMIT
          /* NUMBER OF ROWS RETURNED */
          10000

Patient Outcomes Data

        SELECT
          /* SELECTING AND NAMING RELEVANT COLUMNS */
          hsa as HSA_ID,
          year as date,
          cr_pt_asr_med_readmit AS medical_readmissions,
          oecr_pt_asr_med_readmit AS expected_ratio_medical_readmit,
          cr_pt_asr_sur_readmit AS surgical_readmissions,
          oecr_pt_asr_sur_readmit AS expected_ratio_surgical_readmt,
          cr_pt_asr_med_aoervis AS percent_medical_er_followup,
          oecr_pt_asr_med_aoervis AS expected_ratio_medical_er_followup,
          cr_pt_asr_sur_aoervis AS percent_surgical_er_followup,
          oecr_pt_asr_sur_aoervis AS expected_ratio_surgical_er_followup
        FROM
          post_discharge_records
        WHERE
          /* FILTERING ROWS */
          /* Only returns desired years */
          date IN (2015, 2016, 2017)
          /* Will ask for user to input state code (e.g., VA) */
          AND HSA_ID IN (
            SELECT
              hsa
            FROM
              hospital_atlas_data
            WHERE
              TRUE
              AND pstate = :state
          )
        ORDER BY
          /* SORTING */
          HSA_ID,
          date
        LIMIT
          /* NUMBER OF ROWS RETURNED */
          10000

Pricing and Payment Data

        SELECT
          /* SELECTING AND NAMING RELEVANT COLUMNS */
          hsa as HSA_ID,
          year as date,
          ar_stdprice_physician AS std_physician_reimbursements,
          oe_stdprice_physician * 1.0 AS expected_ratio_physician_reimbursements,
          ar_stdprice_hospitalsnf AS std_hospital_reimbursements,
          oe_stdprice_hospitalsnf * 1.0 AS expected_ratio_hospital_reimbursements,
          ar_stdprice_total AS std_total_reimbursements,
          oe_stdprice_total * 1.0 AS expected_ratio_total_reimbursements
        FROM
          standard_pricing
        WHERE
          /* FILTERING ROWS */
          /* Only returns desired years */
          date IN (2015, 2016, 2017)
          /* Will ask for user to input state code (e.g., VA) */
          AND HSA_ID IN (
            SELECT
              hsa
            FROM
              hospital_atlas_data
            WHERE
              TRUE
              AND pstate = :state
          )
        ORDER BY
          /* SORTING */
          HSA_ID,
          date
        LIMIT
          /* NUMBER OF ROWS RETURNED */
          10000

Joining Network Feature and Patient Outcome Data

        SELECT
          /* SELECTING AND NAMING RELEVANT COLUMNS */
          /* Network Feature Data */
          referral_network_features.hsa as HSA_ID,
          referral_network_features.year as date,
          referral_network_features.forman_mean as care_convergence_index,
          referral_network_features.average_clustering as mutual_care_index,
          referral_network_features.nnodes as num_providers,
          /* Patient Outcome Data */
          post_discharge_records.cr_pt_asr_med_readmit AS medical_readmissions,
          post_discharge_records.oecr_pt_asr_med_readmit AS expected_ratio_medical_readmit,
          post_discharge_records.cr_pt_asr_sur_readmit AS surgical_readmissions,
          post_discharge_records.oecr_pt_asr_sur_readmit AS expected_ratio_surgical_readmt,
          post_discharge_records.cr_pt_asr_med_aoervis AS percent_medical_er_followup,
          post_discharge_records.oecr_pt_asr_med_aoervis AS expected_ratio_medical_er_followup,
          post_discharge_records.cr_pt_asr_sur_aoervis AS percent_surgical_er_followup,
          post_discharge_records.oecr_pt_asr_sur_aoervis AS expected_ratio_surgical_er_followup
        FROM
          /* JOINING TABLES ALONG HSA AND YEAR */
          referral_network_features
          JOIN post_discharge_records ON post_discharge_records.hsa = referral_network_features.hsa
          AND post_discharge_records.year = referral_network_features.year
        WHERE
          date in (2015, 2016, 2017)
          /* Change year as needed, or remove */
          AND HSA_ID IN (
            SELECT
              hsa
            FROM
              hospital_atlas_data
            WHERE
              TRUE
              AND pstate = :state
          )
        ORDER BY
          /* SORTING */
          HSA_ID,
          date
        LIMIT
          /* NUMBER OF ROWS RETURNED */
          10000;