RELIABILITY OF 30-DAY READMISSION MEASURES IN THE HOSPITAL - PowerPoint PPT Presentation

RELIABILITY OF 30-DAY READMISSION MEASURES IN THE HOSPITAL READMISSION REDUCTION PROGRAM Michael P . Thompson PhD, Health Services and Policy Research Group Department of Preventive Medicine, University of Tennessee Health Science Center 1

ACKNOWLEDGEMENTS  Co-Authors:  UTHSC – Teresa M. Waters, PhD; Cameron M. Kaplan, PhD  VCU – Gloria J. Bazzoli, PhD; Yu Cao, MS  Supported by: R01HS023783 (AHRQ; TM Waters, PI) 2

HOSPITAL READMISSIONS 1 in 5 beneficiaries are readmitted within 30-days $24 billion in hospital costs in 2011 3

RISK-STANDARDIZED READMISSION RATES  RSRRs for AMI, Heart Failure, and Pneumonia  Hierarchical Regression Model (HLM) 𝑍 𝑗𝑘 = 𝑐 0𝑘 + 𝐶𝑌 + 𝑓 𝑗𝑘 𝑐 0𝑘 ~ 𝑂(𝜈, 𝝊 𝟑 )  After risk-adjustment ( 𝐶𝑌 ), there are systematic differences in outcomes ( 𝝊 𝟑 )  Underlying differences in quality 4

MEASURE RELIABILITY  To what extent is the observed variation in RSRRs due to systematic differences between hospitals, rather than statistical noise? 𝑻𝒋𝒉𝒐𝒃𝒎 𝑺𝒇𝒎𝒋𝒃𝒄𝒋𝒎𝒋𝒖𝒛 = 𝑻𝒋𝒉𝒐𝒃𝒎 + 𝑶𝒑𝒋𝒕𝒇  Signal = Between-hospital variation in RSRR  Noise = Within-hospital variation in RSRR HLM random intercept variance p = hospital event rate, n = hospital case volume   𝑍 𝑗𝑘 = 𝑐 0𝑗 + 𝐶𝑌 + 𝑓 𝑗𝑘 𝑞(1 − 𝑞) 𝑂𝑝𝑗𝑡𝑓 = 𝑐 0𝑗 ~ 𝑂(𝜈, 𝝊 𝟑 ) 𝑜 5

ILLUSTRATING RELIABILITY 36 36 32 32 28 28 Within RSRR (%) RSRR (%) 24 24 Between 20 Between 20 16 Within 16 12 12 1 2 3 4 5 1 2 3 4 5 Low Reliability High Reliability 6

MEASURE RELIABILITY  Study Questions: What is the reliability of 30-day RSRRs initially 1. targeted by the HRRP? How many hospitals meet a predetermined 2. benchmark for reliability?  Reliably identifying low performing hospitals is critical for policy success of the HRRP 7

DATA SOURCES AND SAMPLE Table 1. Condition-specific cohort sample size and RSRRs  Data Sources Heart  HCUP SID Data (2009-2011) AMI Pneumonia Failure  AR, CA, FL, IA, NY, and WA Hospitals, N 603 860 896  AHA Annual Survey  Area Health Resource Files Patients, n 156,177 367,188 282,041 (AHRF)  CMS criteria Number of 31,658 86,759 57,796 Readmissions  ICD-9 Codes RSRRs (%), 20.3 23.2 19.7  Inclusion/Exclusion criteria Median (Range) (14.1-28.4) (14.8-45.6) (11.6-55.2) 8

STATISTICAL ANALYSIS HLM model for readmissions, adjusted for patient characteristics, to estimate 1. hospital-specific reliability  signal / (signal + noise)  Median reliability (interquartile range)  Also, adjusted for hospital and community-level socioeconomic factors Compare reliability estimates to commonly used benchmark (R=0.7), and estimate: 2.  Three-year case-volume threshold where R=0.70  Percent of hospitals with reliability ≥0.70 9

Table 2. Median reliability of 30-day readmission measures for AMI, Heart Failure, and Pneumonia cumulatively adjusted for patient, hospital, and community-level factors. Median Reliability (IQR) Hospitals, Condition N Patient * + Hospital † + Community-Level ‡ AMI 603 0.54 (0.41-0.63) 0.52 (0.39-0.61) 0.50 (0.38-0.60) Heart Failure 860 0.71 (0.61-0.78) 0.66 (0.56-0.74) 0.65 (0.54-0.73) 0.82 (0.76-0.86) 0.80 (0.72-0.83) 0.79 (0.72-0.83) Pneumonia 896 * Adjusted for age, gender, race, dual eligible status, median income category, number of Elixhauser comorbidities, length of stay, discharge disposition, index admission through ED, planned admission, weekend admission, and major OR procedure † Adjusted for hospital teaching status, bed size, safety-net hospital, hospital system member, for-profit status, rural hospital, and hospital Medicare proportion ‡ Adjusted for HPSA-primary care shortage area, % county below FPL, and % county uninsured 10

1.00 0.90 0.80 0.70 Reliability (R) 0.60 Table 3. Volume required to achieve moderate reliability 0.50 (R=0.70) and percent of hospitals exceeding that threshold AMI Condition Volume Threshold R≥0.7, % HF 0.40 674 11.4 AMI PN Heart Failure 243 54.9 0.30 62 88.5 Pneumonia 0.20 0 250 500 750 1000 1250 1500 1750 2000 Hospital Volume (n) Figure . Relationship between hospital volume and reliability of 30-day RSRRs for AMI (N=603), Heart Failure (N=860), 11 and Pneumonia (N=896) (note: observations of volume>2000 not shown).

PRINCIPAL FINDINGS  Reliability of RSRRs varies widely by condition  The AMI readmission measure is particularly unreliable:  Low volume  Low between-hospital variation: AMI: 14%-28%, PN: 12%-55%  A modest amount of the variation in RSRRs can be explained by hospital and community-level factors, beyond patient factors  Only 79 hospitals (8.8%) had moderate reliability on all measures (AMI, Heart Failure, and Pneumonia 12

LIMITATIONS  No Medicare claims data  CMS risk-adjustment models  Does not ensure all individuals are Medicare beneficiaries  Limited sample – selected states (AR, CA, FL, IA, NY, and WA)  Expanded list of measures – COPD, TH/KA, CABG 13

POLICY IMPLICATIONS  To the extent that their measures are unreliable, many hospitals may end up being penalized unnecessarily  As between-hospital variation in RSRRs reduces, measures become less reliable  Reducing readmissions requires investments into multifaceted interventions  Hospital investments may be misallocated 14

FUTURE DIRECTIONS: IMPROVING RELIABILITY 1. Raise minimum case volume for eligibility 2. Eligibility criteria based on minimum benchmark of reliability, rather than volume 3. Drop AMI measure, add other conditions 4. Employ composite Readmission measure  Condition groups  Hospital-Wide All-Cause Readmission Rate 15

16

ILLUSTRATION CREDITS Illustration Credits Man by Giovanni Tagliente  currency by Creative Stall from the Noun Project  Money by Marie Van den Broeck  signal by useiconic.com from the Noun Project  low signal by TMD from the Noun Project  Dice by Les Kleuver from the Noun Project  pressure horizontal by ImageCatalog from the Noun Project  trend by ChangHoon Baek from the Noun Project  17