Leapfrog’s Resource Utilization Measures & Severity-Adjustment Models April 25, 2008.

Leapfrog’s Resource Utilization Measures & Severity-Adjustment

Models

April 25, 2008

2

Townhall Call Overview• Introductions

• Resource Utilization Measures

– How Leapfrog is measuring resource utilization

– Applicable procedures and conditions

– Hospital reporting requirements

• Relationship to Efficiency of Care Scores

• Severity-Adjustment Models for Resource Utilization Measures

• Calculations & Data Provided Back to Hospitals

• Website Resources

• Q & A

3

Resource Utilization Measures• Measure: Severity-adjusted average length of stay inflated

by readmission rate

• Length of stay associated with resource utilization

• Readmission used as inflator to avoid “perverse incentive”

(inappropriately releasing patients too early)

• Measurement is specific to a condition--added to complement quality measures for four procedures/ conditions: CABG, PCI, AMI, and Pneumonia

4

Resource Utilization Reporting Requirements

• For each procedure/condition, hospitals are asked to report:- the average length of stay (logarithmically transformed

—GEOMEAN),- the number of cases followed by any readmission to

that hospital within 14 days for any cause,- a count of cases with certain risk factors present

• The clinical information (risk factors) and LOS/Readmission statistics needed to report these data can be accessed from the hospital’s administrative data system; no chart abstraction is necessary

5

Relationship to Efficiency of Care Scores

• For each of the four procedures/conditions, the quality score and resource utilization score will be combined to create a efficiency of care score (see scoring algorithm for details)

• Leapfrog will report on its website the efficiency of care scores, with a drilldown of quality and resource utilization scores

Severity-Adjustment Models

7

Objectives

• Construct length of stay severity-adjustment model

• Steps

– Obtain representative national hospital discharge dataset

– Review and propose candidate risk factors

– Fit model for four classes of discharges

– Document results in white paper

8

Data Source for Models• National Hospital Discharge Survey (NHDS)

• 2003, 2004 and 2005

• De-identified, ongoing public data

• 1 million total discharges

– 20,000 each for AMI & PCI

– 8,000 for CABG

– 33,000 for pneumonia

9

Methodology

• Literature review of prior modeling risk factors

• SAS coding of discharge groupings and candidate risk factors based on diagnostic and procedure codes

• Exploratory analysis of LOS and linear regression modeling of Ln(LOS)

• Demonstrate risk adjustment impact

10

Literature Review of Risk Factors

• Specifications for preliminary risk factors for each discharge group were provided by Leapfrog

• Literature review supported 50 of 67 risk factors and suggested using six for other discharge groups

• All preliminary and suggested risk factors were included as candidates in modeling

11

Exploratory Analysis – LOS Histogram

0%

5%

10%

15%

20%

25%

30%

35%

40%

45%

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

LOS

Per

cent of D

isch

arges

All AMI PCI CABG Pneumonia

12

Exploratory Analysis – Means & Medians

AMI PCI CABG Pneumonia Item

Primary Primary Secondary Primary Secondary Primary Secondary Count 20,083 19,258 1,124 6,338 1,415 27,606 5,423 LOS Median 4.00 2.00 6.00 7.00 9.00 4.00 7.00 Mean 5.82 2.80 8.21 9.18 13.51 5.98 10.35 Std Dev 6.90 3.45 10.43 7.17 12.27 5.86 11.51 ln(LOS) Median 1.39 0.69 1.79 1.95 2.20 1.39 1.95 Mean 1.38 0.69 1.65 2.05 2.33 1.51 1.95 Std Dev 0.85 0.74 0.96 0.53 0.72 0.73 0.89

• “Heavy Tail” outlier impact, especially for small facilities

• Transform LOS by taking logarithm, Ln(LOS)

13

Model Format

• Ln(LOS)] = α + β1RF1 +…+ βmRFm + ε, where,

– Ln(LOS) is the observed log LOS for the discharge

– RFi is one if the risk factor i is present, else zero

– α is the expected log LOS absent any risk factors

– βi is the effect of risk factor i on the expected log LOS

– ε is the “error” term for the model

• Estimate parameters (α and β’s) from individual

discharges available from NHDS

14

Application of Model at Hospital Level

• Linear form of the model provides similar format at hospital level:

• Avg[Ln(LOS)] = α + β1 Avg(RF1) +…+ βm Avg(RFm) + ε*, where,

– Avg[Ln(LOS)] is the observed average log LOS for the hospital

– Avg(RFi) is the percent of discharges with risk factor i for the hospital

– α is the expected average log LOS absent any risk factors

– βi is the effect of risk factor i on the expected average log LOS

– ε* is the “error” term for the facility-level model

• Ln[Geometric Mean(LOS)] = Avg[Ln(LOS)], that is,

Ln[(Y1∙Y2 ∙…∙Yn)1/n] = [Ln(Y1)+Ln(Y2)+…+Ln(Yn)]/n

15

Results – AMI

Variable Label Parameter Estimate

Standard Error

t Value Pr > |t|

Intercept Intercept 0.90309 0.01407 64.20 <.0001

rf17 CABG 1.06319 0.01865 57.02 <.0001

rf33 Congestive heart failure 0.47231 0.01183 39.93 0.0037

rf32 Stroke or transient ischemic attack 0.58837 0.03061 19.22 0.0201

rf08 Chronic renal disease 0.28977 0.01844 15.72 <.0001

rf01 Age GE 55 0.19802 0.01379 14.36 <.0001

rf06 Cancer 0.23670 0.03378 7.01 <.0001

rf09 chronic liver disease 0.28350 0.08068 3.51 0.0034

rf16 PCI 0.02017 0.01196 1.69 <.0001

R-Square 0.2416

Root MSE 0.74180

16

Results – PCI

Variable Label Parameter Estimate

Standard Error

t Value Pr > |t|


rf19 AMI 0.59360 0.00992 59.81 <.0001

rf33 Congestive heart failure 0.62010 0.01423 43.59 <.0001

rf17 CABG 1.27329 0.04593 27.72 <.0001


rf11 COPD 0.24593 0.01655 14.86 <.0001

rf06 Cancer 0.31939 0.03827 8.35 <.0001


R-Square 0.2925

Root MSE 0.66190

17

Results – CABGVariable Label

Parameter Estimate

Standard Error

t Value Pr > |t|



rf19 AMI 0.21230 0.01448 14.66 <.0001


rf01 Age GE 55 0.15761 0.01644 9.59 <.0001

rf02 Is Male -0.10828 0.01342 -8.07 <.0001

rf11 COPD 0.08532 0.01752 4.87 <.0001

rf12 Cardiomyopathy 0.13713 0.03490 3.93 <.0001


R-Square 0.1502

Root MSE 0.53464

18

Results – PneumoniaVariable Label

Parameter Estimate

Standard Error

t Value Pr > |t|


rf45 Respiratory failure 0.45802 0.01144 40.02 <.0001

rf44 Septicemia 0.32223 0.01284 25.10 <.0001


rf43 Pleural effusion 0.37170 0.01890 19.67 <.0001

rf01 Age GE 55 0.14158 0.01044 13.56 <.0001

rf34 Kidney disease 0.11281 0.01123 10.04 <.0001

rf36 COPD 0.06800 0.00878 7.75 <.0001

rf32 Stroke - transient ischemic attack 0.29821 0.04448 6.70 <.0001

rf31 Cirrhosis or chronic hepatitis 0.20157 0.04315 4.67 <.0001

R-Square 0.1245

Root MSE 0.72431

19

Severity Adjustment LOS Impact

– AMI

0.70

0.75

0.80

0.85

0.90

0.95

1.00

1.05

1.10

1.15

1.20

1.25

1.30

0.70 0.75 0.80 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20 1.25 1.30

Hospital Average LOS / Overall Average LOS

A/E Ratio

20

Severity Adjustment Ranking Impact

– AMI

-

10

20

30

40

50

60

70

80

90

100

- 10 20 30 40 50 60 70 80 90 100

Unadjusted Rank

Risk-Adjusted Rank

21

Recommendations

• Initially employ the basic linear models

• Update the model coefficients each year using the most recent three years of NHDS data

• Additional recommendations can be found in the white paper

Calculations and Resources

23

Calculations• Actual length of stay – reported by hospitals

• Expected length of stay – calculated from model: – y-intercept + Σi βiRFi, where βi = parameter estimate of risk

factor i & RFi = proportion of cases with risk factor i– Antilog of expected length of stay then taken

• Standardized length of stay – (Actual LOS /expected LOS) x all-hospital average expected LOS

• Standardized length of stay is then inflated by the readmission rate (1+readmission rate)

• Overall score based on quartile ranking,cutpoints @6/30/2008

24

Data Provided Back to Hospitals• The following data points will be provided back to hospitals

– Actual length of stay– Expected length of stay– Standardized length of stay– Readmission rate– Overall standardized length of stay, inflated by the

readmission rate

• How data will be provided to hospitals still a work-in-progress

• Detailed scoring document available to hospitals in early July

25

Website Resources for Resource Utilization and Severity-Adjustment

Models To assist hospitals in completing and understanding the resource utilization measures and the severity-adjustment models, Leapfrog makes the following tools available on the survey website:

– Resource Measure Specifications– Fact Sheet on Efficiency of Care & Resource

Utilization– White Paper on severity-adjustment for LOS– Automated worksheet to calculate “GEOMEAN for

LOS” (see next slide)– Scoring Algorithms (more scoring details added

in July)

26

GEOMEAN Calculator

27

Questions?

Leapfrog’s Resource Utilization Measures & Severity-Adjustment Models April 25, 2008.

Documents

Transcript of Leapfrog’s Resource Utilization Measures & Severity-Adjustment Models April 25, 2008.