Version: 2.1

Methodology

We build off the analysis in Employer Aetna POS Rate Comparison, generalizing the approach to compare any two networks.

Methodology

Sample 100,000 rates that appear in both networks for the same provider-code combination. Filter to NULL modifiers and HCPCS/MS-DRG codes only.
For each provider-code combination, compute the max rate for network A and the max rate for network B.
Compute the absolute and percent difference between the max rates for network A and network B.
Report the 10th, 20th, ..., 90th, 95th, and 99th percentiles of the absolute and percent differences.
Store results in two tables containing pairwise comparisons:
- tq_dev.internal_dev_csong_sandbox.network_similarity_analysis
  - contains overall results
- tq_dev.internal_dev_csong_sandbox.network_similarity_analysis_by_class_and_code_type
  - contains results grouped by billing class and billing code type
Use python to generate matrix view of the results.

Limitations

Nothing major, but some things to consider:

We use a completely random sample. Revenue-weighted sampling may be more appropriate. It's possible that the rate differences occur where rates don't matter, or othat differences occur most where rates matter most.
We only compare max rates. Some networks may have more variability within provider-network-code than others.
This doesn't consider coverage. One network may cover more providers than another.

Matrix View

Aetna	OPEN ACCESS MANAGED CHOICE	OPEN ACCESS ELECT CHOICE	NATIONAL PPO	NATIONAL EPO
NATIONAL EPO	1	1.0	0.95	1
NATIONAL PPO	0.95	0.90	1
OPEN ACCESS ELECT CHOICE	1	1
OPEN ACCESS MANAGED CHOICE	1

Read as: "National EPO" and "OAMC" are 100% similar, while "National EPO" and "National PPO" are 95% similar.

View Python Code

# %%
df = pd.read_sql(f"""
SELECT 
    payer_id_a as payer_id,
    product_network_label_a as network_a,
    product_network_label_b as network_b,
    CASE
        WHEN max(max_diff_pct_p99) < 0.05 THEN '1.0'
        WHEN max(max_diff_pct_p95) < 0.05 THEN '0.95'
        WHEN max(max_diff_pct_p90) < 0.05 THEN '0.90'
        WHEN max(max_diff_pct_p80) < 0.10 THEN '0.80'
        WHEN max(max_diff_pct_p70) < 0.10 THEN '0.70'
        WHEN max(max_diff_pct_p60) < 0.10 THEN '0.60'
        WHEN max(max_diff_pct_p50) < 0.10 THEN '0.50'
        WHEN max(max_diff_pct_p40) < 0.10 THEN '0.40'
        WHEN max(max_diff_pct_p30) < 0.10 THEN '0.30'
        WHEN max(max_diff_pct_p20) < 0.10 THEN '0.20'
        WHEN max(max_diff_pct_p10) < 0.10 THEN '0.10'
        ELSE '0'
    END as similarity,
    max(max_diff_pct_p90) as max_diff_pct_p90,
    max(max_diff_pct_p95) as max_diff_pct_p95,
    max(max_diff_pct_p99) as max_diff_pct_p99
FROM tq_dev.internal_dev_csong_sandbox.network_similarity_analysis
WHERE data_source = 'core_rates'
GROUP BY 1, 2, 3
""", con=trino_conn)
df

# %% [markdown]
"""
Comparison Matrices
"""

# %%
bucket_to_val = {"1.0": 1, "0.95": 0.95, "0.90": 0.9, "0.80": 0.8, "0.70": 0.7, "0.60": 0.6, "0.50": 0.5, "0.40": 0.4, "0.30": 0.3, "0.20": 0.2, "0.10": 0.1, "0": 0}
df["similarity_val"] = df["similarity"].map(bucket_to_val)

matrices = {}

for payer_id, g in df.groupby("payer_id", dropna=False):
    nets = sorted(set(g["network_a"].dropna().unique()).union(set(g["network_b"].dropna().unique())))
    if not nets:
        continue

    mat = g.pivot_table(
        index="network_a",
        columns="network_b",
        values="similarity",
        aggfunc="first"
    )
    # reverse column order for upper-left triangle
    mat = mat.reindex(index=nets, columns=list(reversed(nets)))

    # fill diagonal with 1.0
    for n in nets:
        mat.loc[n, n] = "1"

    matrices[payer_id] = mat.astype(float)

    print(f"\n### Payer {payer_id} — Network Similarity Matrix\n")
    print(mat.fillna("").to_markdown())

Output Pairwise Comparison Tables

View SQL Query

SELECT 
    data_source,
    payer_id_a,
    payer_id_b,
    product_network_label_a,
    product_network_label_b,
    schema_month,
    max_diff_p10,
    max_diff_p20,
    max_diff_p30,
    max_diff_p40,
    max_diff_p50,
    max_diff_p60,
    max_diff_p70,
    max_diff_p80,
    max_diff_p90,
    max_diff_p95,
    max_diff_p99,
    max_diff_pct_p10,
    max_diff_pct_p20,
    max_diff_pct_p30,
    max_diff_pct_p40,
    max_diff_pct_p50,
    max_diff_pct_p60,
    max_diff_pct_p70,
    max_diff_pct_p80,
    max_diff_pct_p90,
    max_diff_pct_p95,
    max_diff_pct_p99,
    run_timestamp
FROM tq_dev.internal_dev_csong_sandbox.network_similarity_analysis
WHERE data_source = 'core_rates'
-- OR use core_rates_physician_groups
-- WHERE data_source = 'core_rates'
LIMIT 5

data_source	payer_id_a	payer_id_b	product_network_label_a	product_network_label_b	schema_month	max_diff_p95	max_diff_p99	max_diff_pct_p95	max_diff_pct_p99	run_timestamp
core_rates	7	None	NATIONAL EPO	OPEN ACCESS MANAGED CHOICE	2025_07	0	0	0	0	2025-09-12 08:36:22
core_rates	7	None	NATIONAL EPO	NATIONAL PPO	2025_07	5.14153	2858.55	0.0463063	0.889578	2025-09-12 08:37:29
core_rates	7	None	NATIONAL EPO	OPEN ACCESS ELECT CHOICE	2025_07	0	0	0	0	2025-09-12 08:35:10
core_rates	7	None	NATIONAL PPO	OPEN ACCESS ELECT CHOICE	2025_07	7.39286	2841.29	0.0557815	1.04107	2025-09-12 08:34:15
core_rates	7	None	OPEN ACCESS ELECT CHOICE	OPEN ACCESS MANAGED CHOICE	2025_07	0	0	0	0	2025-09-12 08:32:15

code

Visit network_similarity.py

Methodology​

Matrix View​

Output Pairwise Comparison Tables​

On this page:

Methodology

Matrix View

Output Pairwise Comparison Tables