Drug Accuracy Scores

work-in-progress

Eventually, the goal is to use a provision-informed methodology to select the best drug rate. We would infer high cost OP % of charge provisions and use this to select the best rate.

In the meantime, we need to have a methodology that will offer a way to distinguish scores 2-4 (5 are validated, 1 are outliers).

Scoring Criteria

• Score 5: Validated
• Score 4: Between 1.25 and 3.5 (10th–90th percentiles)
• Score 3: Between 0.9 and 4
• Score 2: < 0.9 or > 4
• Outlier:
  • Hospital MRF: < 0.8 or > 4
  • Payer MRF: < 0.8 or > 10

Distributions

Validated

	hospital	payer
count	7092	2908
mean	2.24731	2.13091
std	0.811092	0.786044
min	1.2	1.2
1%	1.21	1.21
5%	1.24	1.23
10%	1.28	1.26
20%	1.43	1.374
30%	1.63	1.55
40%	1.83	1.758
50%	2.09	1.97
60%	2.37	2.14
70%	2.7	2.42
80%	3.05	2.82
90%	3.52	3.43
95%	3.76	3.77
99%	3.96	3.96
max	4	4

Code

# %%
df = pd.read_sql(f"""
SELECT canonical_rate_source, canonical_rate_percent_of_medicare
FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
WHERE 
    is_drug_code = True
    AND provider_type = 'Hospital'
    AND canonical_rate_score = 5
    AND canonical_rate_class != 'Impute'
ORDER BY RANDOM()
LIMIT 10000
""", con=trino_conn)

# %%
print(
    df
    .groupby('canonical_rate_source')['canonical_rate_percent_of_medicare']
    .describe(
        percentiles=[
            0.01, 
            0.05,
            0.1,
            0.20,
            0.30,
            0.40,
            0.50,
            0.60,
            0.70,
            0.80,
            0.90,
            0.95,
            0.99,
        ]
    )
    .T
    .to_markdown()
)

All Non-Outlier Rates

	hospital	payer
count	2253	7747
mean	2.03734	1.79309
std	0.910745	1.57546
min	0.8	0.8
1%	0.83	0.83
5%	0.92	0.9
10%	1	0.94
20%	1.15	0.98
30%	1.35	1
40%	1.59	1.04
50%	1.83	1.11
60%	2.14	1.25
70%	2.5	1.57
80%	2.97	2.12
90%	3.47	3.55
95%	3.75	5.44
99%	3.96	8.6908
max	4	10

Code

df = pd.read_sql(f"""
SELECT canonical_rate_source, canonical_rate_percent_of_medicare
FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged
WHERE 
    is_drug_code = True
    AND provider_type = 'Hospital'
    AND canonical_rate_score > 1
    AND canonical_rate_class != 'Impute'
ORDER BY RANDOM()
LIMIT 10000
""", con=trino_conn)
df

print(
    df
    .groupby('canonical_rate_source')['canonical_rate_percent_of_medicare']
    .describe(
        percentiles=[
            0.01, 
            0.05,
            0.1,
            0.20,
            0.30,
            0.40,
            0.50,
            0.60,
            0.70,
            0.80,
            0.90,
            0.95,
            0.99,
        ]
    )
    .T
    .to_markdown()
)

Consistent Rates

	pct_of_medicare
count	6852
mean	1.53906
std	0.886572
min	0.8
1%	0.86
5%	0.94
10%	0.98
25%	1
50%	1.43
75%	1.63
90%	2.05
95%	2.89
99%	5.63
max	9.37

Code

%%

df = pd.read_sql(f""" WITH df AS ( SELECT provider_id, network_id, APPROX_PERCENTILE(i.canonical_rate_percent_of_medicare, 0.05) as p05, APPROX_PERCENTILE(i.canonical_rate_percent_of_medicare, 0.25) as p25, APPROX_PERCENTILE(i.canonical_rate_percent_of_medicare, 0.5) as p50, APPROX_PERCENTILE(i.canonical_rate_percent_of_medicare, 0.75) as p75, APPROX_PERCENTILE(i.canonical_rate_percent_of_medicare, 0.95) as p95, COUNT(*) as n_rates FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged i WHERE i.provider_type = 'Hospital' AND i.is_drug_code = True AND i.canonical_rate_class != 'Impute' AND i.canonical_rate_score > 1 AND i.canonical_rate_source = 'hospital' GROUP BY 1,2 ), consistent AS ( SELECT * FROM df WHERE p95 - p05 < 0.25 AND n_rates >= 30 ) SELECT i.canonical_rate_percent_of_medicare as pct_of_medicare FROM tq_dev.internal_dev_csong_cld_v2_2_2.prod_combined_abridged i JOIN consistent USING (provider_id, network_id) WHERE i.canonical_rate_class != 'Impute' AND i.is_drug_code = True AND i.provider_type = 'Hospital' AND i.canonical_rate_score > 1 AND i.canonical_rate_source = 'hospital' ORDER BY RANDOM() LIMIT 50000 """, con=trino_conn)

%%

print( df['pct_of_medicare'] .describe(percentiles=[0.01, 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95, 0.99]) .to_markdown() )