HEALTH ECONOMICS

1
Marginal and average productivity
 We are moving on from the lifestyle effect
 Production function for health
 The characteristics of medical care as an input
 Uncertainty of outcomes
 Uncertainty in the choice of health interventions across
doctors
 Marginal productivity
 Typical decreasing marginal productivity in a production
process dH/dm < 0, where H=f(m)
 First, increasing marginal productivity due to specialization, then
decreasing
 Eventually, average productivity (H/m) will also fall

2
Extensive margin
 Marginal productivity of health care will fall as more
resources are used
 Iatrogenic (induced) illnesses => negative MP
 One way to think
 E.g. Breast cancer screening
 Epidemiological studies can characterize risks per age and per other
characteristics. True population ratio is f
 Yield of positive cases per 1000 screenings depends on population
and test accuracy
 False negatives (miss the cancer) & false positives (no cancer in
reality)
 p –probability of detecting true cases: sensitivity
 q –probability of false positives
 Yield = f x p + (1-f) x q

3
Extensive margin
 Extending the screening age limit from 50 to 40
 Marginal productivity declines as we extend: We may cause
cancer if we extend the age to 20
 E.g. Back Surgery (65-80% of Americans will have serious
lower back pain at some point in life)
 2% have surgery
 Success rate varies across locations
 Case selection methods vary. Some cases are clear: near unanimous
surgery decision. For these patients success rates are high
 Others may be due to doctor’s decision : By expanding the
extensive margin this way, more surgeries could result in more
back pain (marginal product of health care falls to negative)

4
Intensive (of use) margin
 Population held constant, frequency increased
 If women receive mammography too often or too
infrequent, marginal productivity will fall
 In summary, we gain a lot from medical care on
average, but not a lot from a little more
 We use health care to a point where marginal product is
declining

5
Aggregate data comparisons
(average productivity)
 Cross-country or cross regional comparisons of health
outcomes (necessarily broad measures like life
expectancy) show that per capita income, per capita
education, medical care use, and good health
outcomes move together
 Life expectancy, mortality, death rates for infants may
depend more on infrastructure rather than medical care

6
Marginal Productivity of Medical
Care
 It increases at first (especially in underdeveloped
countries or regions),
 If we continue to increase the amount of health care
then MP decreases or changes to negative with
iatrogenic illnesses outweighing the benefit.
 Productivity changes in the extensive and intensive
margin.

7
Example : Screening for Breast
Cancer
 Screening success: Test’s sensitivity (ability to detect true
cases) – with probability p,
 False positive rate – with probability q,
 Suppose the fraction of true positives (learnt from
epidemiology studies) is f,
 Yield rate of the test: f*p + (1-p)*q

 Breast cancer risk increases with age

 1- Start with the high-risk group (+50, women)
 2- Increase the extensive margin
 As you increase the extensive margin, you will see that the
true cases detected per 1000 test will be low.

8
Example: Intensive margin
 Keeping the population constant but increasing the
intensity of the treatment.
 Should women aged 50-60 have the test every 10 yrs, 5
yrs …. or 6 months or daily?

9
Evidence on Aggregate Productivity
Health Care

10
Interrelation of factors
 Per capita income- improved living conditions, sanitary
water supply, safer roads, better nutrition, more purchasing
power.
 Higher per capita income <=> higher per capita education
 Education=> better management of life, better utilization of
medical resources
 Better health => better utilization of school => higher income
 Exception: for white males income and mortality are
positively related (cross-state studies in the US)
 Could be due to consumption of “bads” –but complicated

11
Estimation of the Production of
Health
 Outcome: Life expectancy
 Input: Medical spending
 1950-2000
 Health elasticity of medical spending by age group
 High until age 15
 Decreases after age 55

12
13
Marginal cost of life and life year
saved
Note: The cost-
effectiveness
analysis suggest
$100,000 as a
cutoff for
decision making
for life year saved

14
Productivity of Specific Treatments
 Extending treatments to either extensive or intensive
margins
 Compared with “usual care,” uses fixed technology
 Table 3.1
 Incremental CE ratios in 2008 dollars
 Added cost of for using the intervention divided by the added
life-years
 Cost remains the same across cases (extended)
 Ratios shows MP: low CE rate => high MP

15
Lovastatin: Medication to treat high cholesterol
16
Rand HI Experiment
 Health producing effects of medical care
 Randomized control trial –RAND HIS (RHIS)
 Four cities, two rural sites, 20,000 person years of data, 3-5 years
 Two purposes: 1) Measuring the effect of insurance (price), 2) Measuring the
effect of medical care on health
 Random assignment to health insurance plans
 Some full coverage others with co-pay, varying the price of health care –not health care
per se per person (unethical)
 Price of medical care varied => utilization differences (low coverage group used
2/3 of medical care used by high coverage group)
 Health outcome measures in RHIS
 Activities of Daily Living, perceived physical and mental health
 Sick-loss days
 Modified physical for conditions health care is supposed to affect: weight, blood
pressure…
 All measurements led to a “health-status age,” a physiological measure

18
Does price for care affect health?
 RAND HIE:
 Generally, no
health
differences
between people
on free plan vs.
cost-sharing!
**Only statistically
significant
difference between
plans were in blood
pressure, myopia, &
presbyopia
Results and Criticism
 “Health-status” aged as expected
 Not much difference due to usage reduction to 2/3 of low
coverage group relative to full coverage group
 Except: Better vision and lower blood pressure for low
income full coverage group than partial coverage group
 Almost all improvement in the full coverage group for persons with
high health risks like obesity (%10 decrease in risk of dying) came
from lower blood pressure
 Conclusion: focus on reducing blood pressure with targeted
investments, not free health care
 Criticism: Too short a period
 No changes in health habits were detected (medical care
usage does not alter these habits, we don’t know what would )

20
Over-time productivity changes
 Changes in cost and outcomes for heart attack patients
(over 1984-1998)
Cost increase: $10,000 Life expectancy increase (of heart
attack patients): +1 year (Cutler et al., 1998)
 Changes in treatment of low-birthweight infants (over
1950-1990)
 Cost increase: $40,000 Life expectancy increase : +12
years (Cutler and Meara., 2000)

21
QALY: Quality Adjusted Life Years
 In general mortality/life expectancy is not the best
measure
 It may distort the best use of medical resources
 Improvements in mental health (except suicide) or
otherwise the quality of life (e.g. hip replacement
surgery) need to be accounted for
Weights are based on patients’ assessment. i.e. “If a
healthy day is 100, what is a day with migraine headache
(65)
 Derived from a utility maximization framework

22
DALY: Disability-Adjusted Life Years
 Created by Harvard University (Murray and Lopez,
1996), adapted by WHO and WB.
 Begin with highest achievable life span: (Japanese
women live 86.8 years )
 Subtract losses from death and disability (YLL: years of
life lost)
 YLD: Years lost to disability= Years lived with
disability*DW(Disability Weight)
 DWs are determined by a panel of experts
 Different weight based on age

23
DW

24
DALY Disability Weights

25
Medical Practice Disagreements

26
Disagreements among the doctors
 Disagreement seen in many countries
 US, Norway, Canada, UK…
 Mostly re: extensive margin (how many people should
receive various treatments)
 A series of studies showed this disagreement:
 Fix the population in question and measure rate of
hospitalization for a procedure, do not measure the rate
of use in a region, because a large university hospital
could attract many referrals as opposed to a rural
hospital

27
Coefficient of Variation (COV)
 Coefficient of Variation (σ/µ) is reported
 Free of country of origin of data or medical activity
 High COV → considerable disagreement about the
marginal productivity of a type of medical care
 Since populations are similar, disagreement is generally
on the subgroup to receive treatment
 Extensive margin
 Example: Tonsillectomies differed a lot in UK
 The absolute variation differs across studies, but
relative variations are stable
28
Examples of disagreement in
surgical procedures (Table 3.3)
 Relative agreement on hernia repair
 Although overall rates per 100,000 persons differ
considerably across regions: 113 England, 282 US
 Disagreements:
 Removal of Tonsil and Adenoids
 Removal of Hemorrhoid
 Higher COVs in other studies re: hospitals
 Within hospital dental extractions (0.73)
 False labor (0.75)

29
30
More on variation
 Variation also exists in hospitalizations of
 Urinary tract infections
 Chest pain
 Bronchitis
 Middle-ear infection
 Skin biopsy
 Pediatric hospitalizations (controlling for age-mix)
 Back problems
 Depression

31
Harvard v Yale
 Boston v New Haven
 Academic medicine –but still variation
 Cities are similar in terms of age profile, income, non-
white population etc. –
 Boston had 55% more hospital beds per capita
 22% more employees per bed, paid 5% more
 Age adjusted medical care use is higher in Boston –
uniformly
 Most variation in minor surgeries and medical diagnoses
in which variations in admissions rates are high (see
previous table)
32
More on variation
 In some types of major surgery, New Haven had higher
rates
 Also variation in admission rates other hospital market
areas

33
Disagreements re: intensive margin
of use
 Is Boston aggressive in extensive margin (hospitalize
people) but not in intensive margin (the length of
stay)
 No relationship found so far
 Coronary artery bypass (number of grafts unrelated to the
overall rate)
 Except: Roos et al. 1986: propensity to admit v length of stay
had weak negative relationship
 Marginal Productivity remains an open question

34
35
36
Finer Outcome Measures
 Quality of life v. length of life
 The US Medicare population had about 5-8 times more
invasive tests or heart surgery done within the first 30
days after the heart attack than Canadians (gap slightly
narrowed in 180 days), mortality rates improved from
22.3% only to 21.4% a month later (but was similar a year
later)
 Marginal productivity was low
 Revascularization interventions in the US were triple the
rate of those in Canada and life quality was better (21% v
34% chest pain and 45% v 29% shortness of breath)
(Mark et al., 1994 )

37
Moving on to Physician-specific
variation
 Up until now…
 Compare use of medical intervention across different
regions
 Must make sure differences in treatment rates is not due
to differences in illnesses
 Hence need large numbers in order to gain statistical
confidence
 In order to standardize age and gender differences an
epidemiological method called “indirect
standardization” is used

38
Indirect Standardization
 Standard (crude) death rate per 1000 persons for an
age group * Population of that age group in the Region
 Sum across age groups to obtain Expected Deaths
 Standardized mortality rate = SMR =
 Total Deaths in Region / Expected Deaths

39
Standardizing Physician case mix
 “No doctor in the US treats enough patients with a
single disease to make comparisons meaningful”
 Need to standardize doctors’ patient mix and severity
of illness to estimate a doctor’s style
 Too costly or not?

40
Physician styles (profiling for cost-
control)
 Distributions of medical resource use (resource units billed
by physicians) differed between states: Figure 3.5 (Oregon v
Florida)
 “Relative value was measured in relative-value units (RVUs),
according to the resource-based relative-value scale used by
Medicare in determining payments to physicians. The mean
number of RVUs per admission was then adjusted for the
physician's case mix according to the patients' assigned
diagnosis-related groups.”
 Attending physician is resident at that hospital
 Mean 30 (~work of 30 routine office visits) v 46
 On average lowest cost 10% used half of the medical care
resources than the highest cost 10%
 Table 3.5
41
42
43
Direct Standardization
 Regional (crude) death rate per 1000 persons for an age
group * (Population of that age group in the US / Total
US population)
 Sum across age groups to obtain standardized death rate
at a region

44
Physician style regression
 Phelps et al. (1994): Data from Blue Choice HMO,
expenses are authorized by PCP.
 Individual annual medical care spending on all areas
(excluding nursing homes) was regressed on
 Case mix
 Illness severity
 Age, gender, demographic variables (representing
patient mix)
 Dummy for each physician
 These fixed effects constituted physician styles (Figure 3.2)
 A score of -0.1 means 10% less spending than average

45
Variation in practice styles

What is the policy suggestion?

46
Other approaches to
understanding production of
health
 Dynamics (not fixed technology any more)
 Associate per case spending with per case changes in
health outcomes
 Cutler et al. (1998)
 Costs increase by $10,000 whereas life expectancy of heart
attack victims increase by one year
 Most change in both cost and health outcomes arise from
technological change
 Cutler et al. (2000)
 1950-1990 low birth infants
 $40,000 increase and 12 years => CE = $3,300

49
Cutler and Meara (2000)

50
51
52
53
54
55