OPPS Medicare Benchmarks

Summary

Is there a bias that arises when comparing commercial rates to OPPS medicare rates, because OPPS uses APC but private insurers do not? Does it matter?

Medicare relativity isn't a standardized metric and its distribution can vary a lot depending on the service.

Background

In the inpatient setting (MS-DRGs), nearly all codes have a median commercial-rate roughly 2x the medicare rate. Here's a plot showing the distribution of MSDRG median validated-rates as a % of medicare.

In contrast, variance is much wider for OPPS: the plot below shows the distribution of HCPCS median validated-rates as a % of medicare.

Here's an example of APC 5733. Majority of codes have medians < 500% of medicare, but ~a quarter have medians near 1000% of medicare or even above 2000% of medicare

Outliers

In the OP setting, we should be careful with using % of medicare to flag outliers. For some codes, we may see more than a quarter of validated rates > 3000% of medicare.

Instead, an approach that relies on the distribution of validated rates would be a better approach (e.g. log-iqr).

Appendix

More Plots By APC

Top 40 Codes with Highest Median as % of Medicare

code	description	median_canonical_rate	median_medicare_rate	%_of_medicare
0101T	EXTRACORPOREAL SHOCK WAVE INVOLVING MUSCULOSKELETAL SYSTEM, NOT OTHERWISE SPECIFIED	6787.07	239.934	28.2872
17340	CRYOTHERAPY (CO2 SLUSH, LIQUID N2) FOR ACNE	1423.27	59.454	23.939
51792	STIMULUS EVOKED RESPONSE (EG, MEASUREMENT OF BULBOCAVERNOSUS REFLEX LATENCY TIME)	1183.85	59.4495	19.9136
0563T	EVACUATION OF MEIBOMIAN GLANDS, USING HEAT DELIVERED THROUGH WEARABLE, OPEN-EYE EYELID TREATMENT DEVICES AND MANUAL GLAND EXPRESSION, BILATERAL	1029.23	57.9122	17.7722
0780T	INSTILLATION OF FECAL MICROBIOTA SUSPENSION VIA RECTAL ENEMA INTO LOWER GASTROINTESTINAL TRACT	1924.27	128.616	14.9614
76390	MAGNETIC RESONANCE SPECTROSCOPY	1267.23	87.9334	14.4112
57150	IRRIGATION OF VAGINA AND/OR APPLICATION OF MEDICAMENT FOR TREATMENT OF BACTERIAL, PARASITIC, OR FUNGOID DISEASE	844.157	59.2329	14.2515
0207T	EVACUATION OF MEIBOMIAN GLANDS, AUTOMATED, USING HEAT AND INTERMITTENT PRESSURE, UNILATERAL	1400.16	128.424	10.9026
58579	UNLISTED HYSTEROSCOPY PROCEDURE, UTERUS	1865.19	201.458	9.25847
72125	COMPUTED TOMOGRAPHY, CERVICAL SPINE; WITHOUT CONTRAST MATERIAL	948.138	106.279	8.92123
53661	DILATION OF FEMALE URETHRA INCLUDING SUPPOSITORY AND/OR INSTILLATION; SUBSEQUENT	1135.57	128.759	8.81933
67820	CORRECTION OF TRICHIASIS; EPILATION, BY FORCEPS ONLY	1121.11	128.749	8.70776
71250	COMPUTED TOMOGRAPHY, THORAX, DIAGNOSTIC; WITHOUT CONTRAST MATERIAL	915.655	106.267	8.61655
53601	DILATION OF URETHRAL STRICTURE BY PASSAGE OF SOUND OR URETHRAL DILATOR, MALE; SUBSEQUENT	1106.71	128.965	8.58145
72131	COMPUTED TOMOGRAPHY, LUMBAR SPINE; WITHOUT CONTRAST MATERIAL	904.527	106.301	8.50913
72128	COMPUTED TOMOGRAPHY, THORACIC SPINE; WITHOUT CONTRAST MATERIAL	887.13	106.203	8.35315
29999	UNLISTED PROCEDURE, ARTHROSCOPY	2000.73	239.54	8.35241
70450	COMPUTED TOMOGRAPHY, HEAD OR BRAIN; WITHOUT CONTRAST MATERIAL	872.114	106.292	8.20493
73700	COMPUTED TOMOGRAPHY, LOWER EXTREMITY; WITHOUT CONTRAST MATERIAL	869.824	106.262	8.18567
70486	COMPUTED TOMOGRAPHY, MAXILLOFACIAL AREA; WITHOUT CONTRAST MATERIAL	869.587	106.237	8.18532
72192	COMPUTED TOMOGRAPHY, PELVIS; WITHOUT CONTRAST MATERIAL	857.352	106.335	8.06273
73200	COMPUTED TOMOGRAPHY, UPPER EXTREMITY; WITHOUT CONTRAST MATERIAL	854.759	106.261	8.04396
31899	UNLISTED PROCEDURE, TRACHEA, BRONCHI	1531.87	192.772	7.94657
70480	COMPUTED TOMOGRAPHY, ORBIT, SELLA, OR POSTERIOR FOSSA OR OUTER, MIDDLE, OR INNER EAR; WITHOUT CONTRAST MATERIAL	828.809	106.243	7.8011
17360	CHEMICAL EXFOLIATION FOR ACNE (EG, ACNE PASTE, ACID)	1549.8	198.917	7.79119
74150	COMPUTED TOMOGRAPHY, ABDOMEN; WITHOUT CONTRAST MATERIAL	812.637	106.11	7.65843
70490	COMPUTED TOMOGRAPHY, SOFT TISSUE NECK; WITHOUT CONTRAST MATERIAL	804.321	106.001	7.58786
29720	REPAIR OF SPICA, BODY CAST OR JACKET	1171.94	158.081	7.41357
40899	UNLISTED PROCEDURE, VESTIBULE OF MOUTH	1695.7	230.435	7.35867
71275	COMPUTED TOMOGRAPHIC ANGIOGRAPHY, CHEST (NONCORONARY), WITH CONTRAST MATERIAL(S), INCLUDING NONCONTRAST IMAGES, IF PERFORMED, AND IMAGE POSTPROCESSING	1301.82	177.82	7.32099
70496	COMPUTED TOMOGRAPHIC ANGIOGRAPHY, HEAD, WITH CONTRAST MATERIAL(S), INCLUDING NONCONTRAST IMAGES, IF PERFORMED, AND IMAGE POSTPROCESSING	1261.85	177.917	7.09235
93226	EXTERNAL ELECTROCARDIOGRAPHIC RECORDING UP TO 48 HOURS BY CONTINUOUS RHYTHM RECORDING AND STORAGE; SCANNING ANALYSIS WITH REPORT	414.288	58.7557	7.05103
76498	UNLISTED MAGNETIC RESONANCE PROCEDURE (EG, DIAGNOSTIC, INTERVENTIONAL)	614.584	87.431	7.02936
67208	DESTRUCTION OF LOCALIZED LESION OF RETINA (EG, MACULAR EDEMA, TUMORS), 1 OR MORE SESSIONS; CRYOTHERAPY, DIATHERMY	2088.03	299.342	6.97539
67516	SUPRACHOROIDAL SPACE INJECTION OF PHARMACOLOGIC AGENT (SEPARATE PROCEDURE)	2303.71	330.345	6.97364
70498	COMPUTED TOMOGRAPHIC ANGIOGRAPHY, NECK, WITH CONTRAST MATERIAL(S), INCLUDING NONCONTRAST IMAGES, IF PERFORMED, AND IMAGE POSTPROCESSING	1238.8	177.964	6.96098
20999	UNLISTED PROCEDURE, MUSCULOSKELETAL SYSTEM, GENERAL	1656.69	238.681	6.941
69979	UNLISTED PROCEDURE, TEMPORAL BONE, MIDDLE FOSSA APPROACH	1613.37	232.527	6.93841
20561	NEEDLE INSERTION(S) WITHOUT INJECTION(S); 3 OR MORE MUSCLES	168.951	24.4465	6.91106
21085	IMPRESSION AND CUSTOM PREPARATION; ORAL SURGICAL SPLINT	1602.48	232.527	6.89158

Bias Example

How can this introduce bias when we calculate aggregate statistics?

Complete Overlap

When we have complete overlap, the relative comparisons between hospitals is still meaningful, even if the magnitude is not.

In the example below, code 72125 has a commercial median that is well above medicare (900%), whereas the other codes do not. Nevertheless, we correctly conclude that hospital Y's rates are higher than hospital Y on average.

However, it could be misleading to say hospital X's rates are on average 400% of medicare. This is true - but it's driven solely by code 72125.

code	median % of medicare	hospital X	hospital Y
72125	892.123%	1000%	1000%
25066	175.5%	200%	300%
53442	140.6%	200%	300%
25071	253.6%	200%	300%
average	365.456%	400%	475%

Bias in Missingness

When we do not have complete overlap, there's potential for the relative comparisons between hospitals to be inaccurate.

In the example below, we have a rate for 72125 for hospital X, but not for hospital Y. We incorrectly conclude that hospital X has higher rates relative to medicare.

code	median % of medicare	hospital X	hospital Y
72125	892.123%	1000%	NULL
25066	175.5%	200%	300%
53442	140.6%	200%	300%
25071	253.6%	200%	300%
average	365.456%	400%	300%

How often do we expect this to happen? It depends on the randomness of missingness. If we are looking at 1000 codes, there may be codes like 72125 that have high median % of medicare, but hospital Y has a rate whereas hospital X does not. And so the effect of this bias would "even out".

Solutions (WIP)

• For Analyze:
  • Evaluate bias in real reports by comparing against versions of reports that adjust for the OPPS-bias.
  • We may not want to surface this to customers, but one strategy would be to use national commercial medians as benchmarks. Log-transform to reduce skew and compute inverse-variance weighted average % of medians. This would standardize the various relative-to-benchmarks metrics.
    • How often does this metric conflict with un-weighted un-adjusted Analyze reports' conclusions?
• Don't use % of medicare to flag outliers for OP codes (we use IQR on log-scale in CLD)