NZ Hospital Performance (2001-09) Outputs, Inputs, and Productivity - PowerPoint PPT Presentation

NZ Hospital Performance (2001-09) Outputs, Inputs, and Productivity

Motivating questions  Have NZ hospitals become more/less productive in the past 9 years (2001-09)?  What is the evidence?  What are the limitations in drawing inferences from available data and statistical models?  Are there identifiable differences in hospital productivity across DHBs?  What is the evidence?  What factors explain differences?  What are the limitations in drawing inferences from available data and statistical models? 2

Hospital Outputs  Data  NMDS 2001-2009  all facilities, all discharges  Measures  Case mix weighted total discharges per month/year  using relative resource use by DRG (WEIS) as weights  Average length of stay  Proportion of day stays  Shortcoming  No outpatient & emergency department visits  ~25% (or more) of total hospital output 3

Hospital Output – over the years No. of discharges Average length of stay (days) 1000000 3.9 950000 3.8 No. of Discharges Ave Length of Stay 3.7 900000 3.6 3.5 850000 3.4 23% increase 16% decline 3.3 800000 3.2 3.1 750000 3 2.9 700000 2.8 2001 20022003200420052006200720082009 Year Year 4

Total Discharges per month Mean WEIS-weighted Mean discharges per month discharges per month 4100 4100 WIES weighted No. of Discharges 3900 3900 No. of Discharges 3700 3700 3500 3500 3300 3300 3100 3100 2900 2900 2700 2700 2500 2500 Time (2001 – 2009) Time (2001 – 2009) 5

Average length of stay & proportion day stays Change in Average Length of Stay Change in Proportion of Day Stays 120 200 110 180 Index (2001=100) Index (2001=100) 100 160 90 140 80 120 70 100 60 50 80 Year Year 6

Change in WEIS-weighted Annual discharges 150 140 Index (2001=100) 130 120 110 100 90 80 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year 7

Hospital Inputs  Data  DHB Provider expenditures 2001-2009 (MoH)  By month, disaggregated by type, with FTEs  Measures  Total real expenditures in 2001 NZD (1 st Qtr)  Deflation using GDP deflator (Stats NZ)  Proportions by type of expenditure (later)  Shortcoming  DHB-level aggregation  No breakdown by inpatient, outpatient, ED 8

DHB provider expenditure over the years DHB (real) expenditure per Total DHB (real) expenditure capita $6.0 1.4000 1.3500 $5.5 1.3000 $ thousands $ billions $5.0 1.2500 40% increase 1.2000 $4.5 26% increase 1.1500 1.1000 $4.0 1.0500 $3.5 1.0000 Year Year 9

DHB Expenditure breakdown Input Expenditure shares Personnel Expenditure shares Infrastructu Manageme re & nt Nonclinical Medical 17% Support supplies 25% 3% 19% Clinical Allied Supplies Personnel Health 14% 60% 16% Nursing 39% Outsource d Services 7% 10

DHB Expenditure variation Total Real Expenditures Real Expenditure per discharge 190 190 180 180 170 Index (2001=100) Index (2001=100) 170 160 160 150 150 140 140 130 130 120 120 110 110 100 100 90 90 01 02 03 04 05 06 07 08 09 01 02 03 04 05 06 07 08 09 Year Year 11

Productivity  Measure: Output per $ of (input) expenditure  NMDS Outputs = inpatient stays (case mix weighted)  incl. day stays counted as 0.5 days  Excl. outpatient & ED visits  DHB provider expenditures in 2001 NZD(MoH)  Deflated by GDP deflator (StatsNZ)  Results  12% decline between 2001 & 2009  BUT measure is based on inpatient stays so underestimates productivity  IF share of outpatient and ED increased over time then decline is over-stated DID SHARE increase by >12%?  Substantial variation across DHBs 12

Change in Productivity over time 110 105 100 Index (2001=100) 95 90 85 80 75 70 65 60 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year 13

Modelling variation in productivity  Data: Panel of monthly data on case-mix weighted hospital output (partial) and DHB provider expenditure for 9 years  Super-population perspective for statistical inference  Hospital productivity of DHB at time t is a function of …  time-varying DHB characteristics (eg. case mix, organization, resource allocation)  time-invariant DHB characteristics (eg. Population size and demographic composition, location)  time-varying DHB-invariant characteristics (policy directive)  Some characteristics observed, some not  Dynamic (changing) relationships – with past characteristics (including productivity itself) 14

Productivity variation: Regression models  Static Models  Specifications estimated (so far)  Pooled with robust standard errors for DHB clustering  Random effects  Fixed effects  Dynamic Models  Lagged dependent variables  Endogenous regressors (input expenditures, acute admissions, etc) 15

Model-based estimates of temporal variation in productivity Seasonal Variation (Base=Jan) Annual Variation (Base=2001) 0.025 0 0.02 Percentage Change (Base=2001) -0.01 Percentage Change (Base=Jan) 0.015 -0.02 0.01 -0.03 0.005 -0.04 0 -0.05 -0.005 -0.06 -0.01 Month 02 03 04 05 06 07 08 Mar Apr May Jun Jul Aug Sep Oct Nov Dec Year Monthly Data Yearly Data 16

Other (preliminary) results  Static models  Monthly data  DHB fixed effects specification best  Significant effects : Proportion Pacific (-ve), Proportion newborn (-ve), proportion arranged admissions (+ve), and significant month & year variation (previous slide)  Annual data  DHB random effects specification best  Only year dummies significant (previous slide)  Dynamic models (so far only with annual data)  Serial correlation in errors so dynamic model more efficient  Significant effects:  Lagged productivity (+ve): higher past productivity -> higher current  Economies of scale (+ve): larger admission volume -> higher productivity 17

Quality of hospital services  Composite measure (annual) based on patient safety indicators  based on 11 (of 20) provider-level patient safety indicators (PSIs) developed by AHRQ  Construction of measure Risk-Adjustment (for each component) 1. Patient-level Logistic regressions for each PSI to derive predicted values of the outcomes of  interest on full nine years of data in NMDS. Reliability Adjustment (for each component) 2. Need to adjust indicator for reliability by isolating true variability of indicator.  Combining 11 components (multi-dimensionality) 3. Need weighting system to combine indicators of different dimensions into a single composite  index. Equal Weights, Factor Analysis based weights, Expert opinion based weights  18

Change in Hospital Quality 150 Increase in index 140 => Index (2001=100) decline in 130 quality 120 110 100 90 80 2001 2002 2003 2004 2005 2006 2007 2008 2009 Year 19

Model-based results for hospital quality  Findings from preliminary panel econometric regressions Significant Effects Sign Time +ve Proportion Female +ve Proportion NZ European -ve Proportion Pacific -ve Clinical Severity +ve  Total DHB expenditures have little explanatory power Source: Bowden-Desai paper presented at NZAE (2011) 20

Ongoing work and next steps Further refinement of dynamic models for Productivity, Input 1. expenditures, Quality ……to analyze relationships between these 3 variables in a dynamic model setting 2. Examination of effects of DHB monitoring (MoH data) and changes in hospital output composition (DHB data) Further refinement of hospital (patient safety) quality index using 3. (ongoing) survey data on ranking of PSI by clinicians 21