Do Health Care Funding Levels Affect Patient Outcomes?

1
Do Health Care Funding Levels Affect Patient
Outcomes?

• Margaret Byrne
• University of Pittsburgh
• Laura Petersen, Kenneth Pietz
• Baylor College of Medicine

2
Background

• Health care costs continue to rise
• Debate over how much spending is appropriate
• One approach is to look at differences in
  outcomes with different levels of funding

3
Background

• Health expenditures and utilization vary greatly
  across different countries and regions
• Research not conclusive as to difference in
  outcomes

4
International (OECD) studies

• Insignificant relationships found between
• per capita medical care expenditures and mean
  age at death (LeGrand 1987)
• health expenditure and mortality rates (Judge et
  al. 1998)
• U.S. doesnt have better health outcomes despite
  higher spending (Andersen 2000)

5
Life Expectancy and Spending
6
International (OECD) studies

• Hitiris and Posnett (1992), only study with cross
  sectional and time series data
• Find increases in health expenditures per capita
  significantly lower mortality rates, although
  effect is small

7
U.S. regional variation in

• Medicare payments to physicians per beneficiary
  in 1989
• Miami 1874
• San Francisco 872
• Welch et al. 1993, New Engl J Med

8
U.S. regional variation in

• Age, sex, race adjusted total beneficiary
  spending in Medicare in 1996
• Miami 8,414
• Minneapolis 3,341
• Difference of over 50,000 over average lifetime
• Wennberg and Cooper.
• Darthmouth Atlas of Health Care 1999

9
US regional variation in
10
Expenditures and outcomes

• Fisher et al. Annals of Internal Medicine 2003
• Medicare hospitals categorized to different
  levels of spending based on end of life spending
  (unrelated to illness level)
• Assigned to quintiles of EOL-EI

11
Expenditures and outcomes

• Cohorts of patients with
• Colorectal cancer
• Hip fracture
• Acute MI
• Random sample
• Patients in higher spending regions receive 60
  more care
• More inpatient care
• More specialist care
• More procedures and tests

12
Expenditures and outcomes

• Access and quality of care no better or worse in
  higher spending regions
• No difference in rates of decline of functional
  status over 5 years
• Small increase in mortality rates in regions with
  higher spending levels

13
Limitations to Fisher study

• Mostly cohort study
• Risk adjustment inadequate
• Used a proxy for spending, not actual spending
  levels
• Cross sectional design for measuring spending
  levels

14
Objectives

• Determine whether VA Networks risk adjusted
  funding levels differ
• Determine whether differences in funding levels
  affect mortality rates

15
Advantages of study design

• Complete cohort of VA patients
• Funding allocations rather than expenditure or
  expenditure proxy
• Comprehensive risk-adjustment methodology
• Longitudinal design

16
Outline of analyses

• Correlation between Network funding and mortality
• Cross sectional logit regressions of effect of
  risk adjusted funding levels on mortality in 22
  VA Networks
• 4 year fixed effects logit to distinguish effect
  of funding level from Network and year effects

17
Cohort of users

• All users of the VA in each fiscal year 1998-2001
  with patient level costs
• Excluded veterans with ages lt17 or gt120 years
• Males and females analyzed separately

18
Funding levels

• VA headquarters allocates to the 22 regional
  Networks
• Allocations based on a 3 category capitation
  system
• Assignment of veterans to a capitation class
  based on most serious diagnoses and some
  utilization
• Allocations adjusted for education, equipment,
  non-recurring maintenance
• Not adjusted for local labor costs
• Deflated to 1998 dollars

19
Illness burden measure

• Calculated for each VA user
• Diagnosis based risk adjustment methodology (DxCG
  software)
• All diagnoses used to develop a relative risk
  score

20
Funding per illness burden

• Funding per unit of illness burden calculated
  as
• Where Nnumber of veterans in that Network
• Funding per illness burden calculated for each
  Network for each year

21
Mortality

• Determined using BIRLS file (death benefits
  applications) and checking with Patient Treatment
  File

22
Average RRS
23
Funding per illness burden
24
Differences in FIB
FY98 FY99 FY00 FY01
Average 6476 6346 6697 6627
St. dev. 505 558 522 619
Minimum 5615 5498 5828 5552
Maximum 8021 8075 7531 7478
Size of Range 2406 2577 1703 1926
25
Mortality and FIB FY98
26
Mortality and FIB FY99
27
Mortality and FIB FY00
28
Mortality and FIB FY01
29
Cross sectional regressions

• Dependent variable
• Mortality
• Independent variables
• Network FIB level, scaled to show the effect of a
  1000 change
• Individual RRS and age (including squared and
  cubed terms)

30
Cross section regressions
  FY98 FY99 FY00 FY01
All Males
All Women
Males, age lt 60
Males, age gt 80
Males, RRSlt0.4
Males, RRSgt2.0
31
Cross section regressions
  FY98 FY99 FY00 FY01
All Males .866 .854, .878 .905 .894, .917 .944 .930, .958 .981 .969, .993
All Women
Males, age lt 60
Males, age gt 80
Males, RRSlt0.4
Males, RRSgt2.0
32
Cross section regressions
  FY98 FY99 FY00 FY01
All Males .866 .854, .878 .905 .894, .917 .944 .930, .958 .981 .969, .993
All Women
Males, age lt 60 .890 .859, .921 .916 .888, .946 .914 .889, .941 .960 .938, .982
Males, age gt 80 .817 .790, .845 .885 .860, .911 .948 .916, .982 .987 .962, 1.014
Males, RRSlt0.4 .905 .885, .925 .841 .823, 8.59 .950 .926, .975 .988 .967, 1.01
Males, RRSgt2.0 .809 .790, .828 .930 .911, .949 .924 .902, .946 .972 .953, .991
33
Cross section regressions
  FY98 FY99 FY00 FY01
All Males .866 .854, .878 .905 .894, .917 .944 .930, .958 .981 .969, .993
All Women .923 .820, 1.039 1.027 .926, 1.140 .891 .792, 1.003 1.013 .924, 1.111
Males, age lt 60 .890 .859, .921 .916 .888, .946 .914 .889, .941 .960 .938, .982
Males, age gt 80 .817 .790, .845 .885 .860, .911 .948 .916, .982 .987 .962, 1.014
Males, RRSlt0.4 .905 .885, .925 .841 .823, 8.59 .950 .926, .975 .988 .967, 1.01
Males, RRSgt2.0 .809 .790, .828 .930 .911, .949 .924 .902, .946 .972 .953, .991
34
Fixed effect logit regressions

• Model 1 lump all 4 years of data together
• Model 2 include year fixed effects
• Model 3 include year and Network fixed effects

35
Fixed effect logit regressions
  Model 1 Model 2 Model 3
FIB .947
Age .845
Age2 1.002
Age3 .999
RRS 2.349
RRS2 .956
RRS3 1.001
FY99
FY00
FY01
Model 3 includes Network fixed effects Model 3 includes Network fixed effects Model 3 includes Network fixed effects Model 3 includes Network fixed effects
36
Fixed effect logit regressions
  Model 1 Model 2 Model 3
FIB .947 .934
Age .845 .843
Age2 1.002 1.003
Age3 .999 .999
RRS 2.349 2.344
RRS2 .956 .956
RRS3 1.001 1.001
FY99 0.994
FY00 .965
FY01 .896
Model 3 includes Network fixed effects Model 3 includes Network fixed effects Model 3 includes Network fixed effects Model 3 includes Network fixed effects
37
Fixed effect logit regressions
  Model 1 Model 2 Model 3
FIB .947 .934 1.037
Age .845 .843 .843
Age2 1.002 1.003 1.003
Age3 .999 .999 .999
RRS 2.349 2.344 2.344
RRS2 .956 .956 .956
RRS3 1.001 1.001 1.001
FY99 0.994 1.017
FY00 .965 .968
FY01 .896 .926
Model 3 includes Network fixed effects, results not shown here Model 3 includes Network fixed effects, results not shown here Model 3 includes Network fixed effects, results not shown here Model 3 includes Network fixed effects, results not shown here
38
Summary of results

• Funding across the VA Networks varies both across
  Networks and over time
• Maximum differences in FIB among Networks ranged
  from 2532 in FY99 to 1791 in FY01

39
Summary of results

• Correlations between contemporaneous funding
  levels and Network-level risk adjusted mortality
  rates
• Association holds for all 4 years of the study

40
Summary of results

• For cross section regressions of all men and all
  subsample of males, FIB was significantly related
  to mortality over all 4 years
• FIB not significantly related to mortality for
  women
• Network fixed effects in the 4 year panel logit
  regression nullify the effect of FIB

41
Conclusions

• Strong funding levels in the past create physical
  and human capital in Networks
• Build up human and physical capital
• Stability in hiring
• Ability to purchase and use new technology
• Allows Networks to provider higher quality acute
  care which reduces mortality
• Allows networks to provide more preventive care

42
Implications

• Higher funding or expenditures may not be
  associated with contemporaneous health outcomes,
  but will have an effect over the long term

43
Implications

• Research on the relationship between
  funding/expenditures and outcomes must use panel
  data to distinguish fixed effects and direct
  effects of funding

44
Correlations of FIB over time
  FY99 FY00 FY01
FY98 0.909 0.657 0.614
FY98 0.000 0.001 0.002
FY99 0.810 0.731
FY99 0.000 0.000
FY00 0.96
FY00 0.000
45
Cross section regressions
  FY98 FY99 FY00 FY01
All Males 2,841,181 .866 .854, .878 .905 .894, .917 .944 .930, .958 .981 .969, .993
All Women 134,978 .923 .820, 1.039 1.027 .926, 1.140 .891 .792, 1.003 1.013 .924, 1.111
Males, age lt 60 1,318,974 .890 .859, .921 .916 .888, .946 .914 .889, .941 .960 .938, .982
Males, age gt 80 139,976 .817 .790, .845 .885 .860, .911 .948 .916, .982 .987 .962, 1.014
Males, RRSlt0.4 1,715,045 .905 .885, .925 .841 .823, 8.59 .950 .926, .975 .988 .967, 1.01
Males, RRSgt2.0 267,442 .809 .790, .828 .930 .911, .949 .924 .902, .946 .972 .953, .991
46
VERA payment levels FY00

• Basic unvested (single outpatient)
• 636,696 patients
• 105 per patient
• Basic vested 18 classes
• 2,882,051 patients
• 3,249 per patient
• Complex 24 classes
• 139,607 patients
• 42,153 per patient

47
DCG Models

• DCGs (Ash et al, 1989 Ellis et al 1996)
• Diagnoses first classified by clinicians into 543
  clinically homogeneous DxGroups.
• Cluster DxGroups into 118 Condition Categories
  (CCs) according to expected similarities in
  future costs (eg there are 8 Neoplasm CCs.
  Neoplasm1 is metastatic cancer, Neoplasm8 is
  benign neoplasms).

48
DCG Models (2)

• To avoid reimbursement for separate coding of the
  same condition (eg metastatic cancer neoplasm
  of lung, brain, bone), Hierarchical Condition
  Categories (HCCs) are imposed
• Next, assign a patient to one of many mutually
  exclusive DCGs

49
DCG (25)
HCC (118)
CC (118)
Dx Groups (545)
ICD9 (15K)
CLASSIFYING PATIENTS BY GROUPING DIAGNOSES
MSG
50
Appendix 2 CCs with Hierarchies
51
How do DCGs work?
DCG
MSG
52
Example of HCC Classification
ICD-9-CM DxGroup CC HCC
250.13 IDDM, uncontrolled with ketoacidosis 23.01 Diabetes with acute complications 14 Diabetes with Acute Complications 14
250.01 IDDM, not stated as uncontrolled, without mention of complications 22.01 Diabetes without complications 15 Diabetes with No or Unspecified Complications 14
MSG
53
HCC Payment Example All Patient Model Basic
Patient, capped _at_ 1 W/REGISTRY FLAGS

• Patient has HCC013Diabetes with Chronic
  Complications (N about 89,000) with Payment
  Average Cost of 807 often (with Ngt35,000) also
  has
• HCC004 (Other Infections) with Payment of 856
• HCC018 (Other Endocrine) with no Payment (part
  of NoHCC31)
• HCC026 (Other Musculoskeletal) with no Payment
  (part of NoHCC31)
• HCC052 (Chronic Heart) with Payment of 281
  (Constrained)
• HCC057 (Hypertension) with Payment of 281
  (Constrained))
• HCC072 (High Cost Eye) with Payment of 281
  (Constrained)
• HCC092 (Other Dermatological) with no Payment
  (part of NoHCC31)
• HCC100 (Minor Symptoms) with Payment of 601
• HCC117 (Screening) with no Payment (part of
  NoHCC31)
• HCC Cost Prediction 807 856 281 281 281
  601 3,107
• Payment 3,107 Falls into DCG Payment Group 9
  3,061

MSG
54
HCC Payment Example All Patient Model Complex
Patient, capped _at_ 1 W/REGISTRY FLAGS

• Patient has HCC040Quadriplegia (N9451) with
  Payment Average Cost of 5731 and VSCIFLG with an
  average cost of 6237 often (with Ngt2500) also
  has
• HCC075 (Low Cost Ear) with Payment of 281
• HCC080 (Other Urinary) with Payment of 1216
• HCC091 (Chronic Skin Ulcer) with Payment of
  3591
• HCC092 (Other Dermatological) with no Payment
  (part of NoHCC31)
• HCC097 (Other Injuries) with Payment of 1060
• HCC099 (Major Symptoms) with Payment of 1149
• HCC117 (Screening) with no Payment (part of
  NoHCC31)
• HCC Cost Prediction 5731 6237 281 1216
  3591 1060 1149 19,265
• Payment 19,265 Falls into DCG Payment Group 17
  26,120
• If THIS patients actual cost exceeded 70,000,
  payment would be 26,120 (actual cost-70,000)

MSG
55
Network fixed effects
56
Fixed effects logit bias Katz 2001