Transcript N - University of Bristol

The Long-Run Impacts of Biomedical
Innovation: Evidence from the Sulfa
Drug Era
Work in progress, December 2010
Sonia Bhalotra, University of Bristol (Bristol, UK)
Atheendar Venkataramani, Washington University School of Medicine (St. Louis,USA)
Introduction

Growing literature on the impact of the in
utero and early childhood environment on
health and economic outcomes later in life
(Almond and Currie, 2010; Heckman 2007).
Focus on natural disasters and epidemics Famine e.g. Dutch, Chinese
 Pollution e.g. Chernobyl
 Disease e.g. influenza epidemic
A few recent studies analyse the long-run effects of
public programs

malaria & hookworm eradication; water quality
improvements – Lucas, 2010; Cutler, et al, 2010;
Bleakley, 2010; Venkataramani, 2010
In quantifying the longer run benefits of intervention,
these studies have clearer policy implications.
What We Do


Examine the long-run impacts of early life exposure to
infectious disease using the sulfa drug innovation as an
instrument for infectious disease.
Sulfa drugs (anti-microbial sulfonomides) were the first
pharmaceuticals effective at treating infectious diseases.
Contribute evidence on
Impact of birth-year exposure to infection on health,
cognitive and economic outcomes in adulthood
Differences by sex *race
Reinforcing (v compensating) parental investments
The [undocumented] long run returns to medical innovation
Sulfa Drugs- timing

Discovered in a German lab in 1932, evidence for their antimicrobial potential first published in 1935, first clinical trials
in 1936, 1937 (London, NY)

Jayachandran, Lleras-Muney and Smith (JLS 2010) identify a
structural break in trend in 1937 for diseases treatable by
sulfa drugs: strep infections, pneumonia, meningitis

The US witnessed unprecedented declines in mortality in the
20th century. There were no significant advances in treatment
of infectious disease before sulfa arrived. And nothing else on
the stage till antibiotics appeared in the mid-1940s.

We sample cohorts born 1930-1943.
Short run impact

JLS attribute a 25 % decline in maternal mortality [puerperal
fever] and a 13 % decline in pneumonia and influenza
mortality* between 1937 and 1943 to sulfa.

These declines a/c for 40-75% of the total decline in deaths
from these causes during the period.

No significant change in 1937 in rate of decline of mortality
from “control diseases” such as TB, diarrhea, cancer, heart
disease.

* pneumonia responded to sulfa but influenza did not. Some
75% of deaths from (p+i) were on account of p.
Prevalence and infections of children
Pre v post sulfa mortality rates per 1000 (JLS)
 Maternal mortality
6.5 – 3.6
 Influenza-pneumonia
1.2 – 0.8
Pre v post neonatal mortality rates per 1000 All causes: 3.6 - 2.4
 Pneumonia: 1.6 – 1.1
 Influenza: 0.2 – 0.2
Pneumonia was the leading cause of child death (8% v 44% pre)
Mortality rates proxy wider morbidity rates.
We analyse LR impact of exogenous declines in pneumonia and
maternal mortality rates at birth [and all-cause infant
mortality rate.]
Why Long-Run Effects?
Mechanisms: Pneumonia-exposure
Infectious disease results in the body redirecting nutritional
resources from physical and mental growth to fighting
infection.
Long run outcomes most sensitive to exposure in early childhood:
(a) rapid growth- greater nutritional demands
(b) immune system not fully developed
Under-researched potential role of
(a) reinforcing or compensating parental investments
(b) dynamic complementarities resulting in multiplicative deficits
if early life brain development is impaired together with
physiological growth.
Mechanisms- maternal mortality
Maternal mortality rates fell with sulfa because of
control of puerperal sepsis, a post-birth infection.
Likely paths for impact on offspring are
 increase in investments in girls as their life
expectancy improves (Jayachandran and LlerasMuney, 2009; Albanesi and Olivetti, 2010)
 Increased investment in both genders as more
mothers survive
Why gender and race heterogeneity




The pre-sulfa incidence of pneumonia and MMR was about
twice as high in the black population- so they stood to gain
more.
But there was racial segregation in medical care and black
Americans were more rural. For both reasons they were less
likely to benefit from new technology.
Boys are more sensitive to resource deprivation in the pre and
postnatal period (Waldron 1983, Stinson 1985). So they may
show greater gains in general – biological reasons.
Girls may show greater gains from improvements in maternal
mortality – parental investment reasons.
Extant Empirical Approaches




JLS 2010: Structural break in national trend in
treated-disease mortality at time of interventionMdt = α + β treatedd*postt*yeart + ..
Bleakley 2007: Intervention creates a decline in
mortality that varies across regions, decreasing
(continuously) in the pre-intervention level of
mortalityMjt= α + β postt*Mj(pre) + ..
Our Empirical Strategy

We effectively combine these approaches, exploiting
variation across treated/untreated diseases in states
with high/low pre-intervention mortality pre/post
sulfa.

Individual data from the US census files for 19702000. Cohorts born in 1937 are aged 33, 43, 53, 63.
Data collapsed to state*sex*race averages.
Later: longitudinal micro data on offspring of
exposed cohorts.


Estimated Equations

First stage
Mdst =αf +βf postt*Mds(pre)+ δsf + γtf +µrf + εstf
[treatment]
 Second stage
Yrst = αs + βs Mdst + Xsts´π +δrss + γrts +µras + εsts
Marginal impact on outcome of sulfa-induced
decline in mortality. Xst includes control
disease mortality rates.
Reduced form
Yrst = α + β*postt*Ms(pre) + Xsts´π + θrs + ηrt + λra
+ ergst ; βs = β / βf
Postt = 1 for birth cohorts 1937-43
Ms(pre) is the state-specific pre-intervention mortality
rate (1930-35).
Outcome equations include fixed effects for race*birth
state, race*birth year and race*census year.
Heterogeneity in treatment effects by gender*race.
Threats to Identification


State * cohort macroeconomic or disease shocks
Pre-existing trends
We assess stability of our results to inclusion of birth state * birth
year data on mortality rates from other infectious and noninfectious diseases, state macroeconomic characteristics and
state specific time trends.
TB and diarrhoea measure state-year variation in sanitation and
poverty. Heart disease and cancer deaths capture trends in
medical technology.
Mortality rates for sulfa-treated
diseases- trend break (JLS 2010)
Trend Breaks: treated v control
diseases
VARIABLES
TREATED
(1)
(2)
Maternal
Influenza &
mortality
Pneumonia
rate
(3)
Infant
mortality
rate
TB
Diarrhea.
(post==1)*year
-0.0999***
-0.214***
-1.063***
0.0117***
9.826**
-0.0108*** -0.0406***
(0.00592)
(0.0252)
(0.168)
(0.00388)
(4.971)
(0.00195)
(4)
UNTREATED
(5)
(6)
Cancer
(7)
Heart
(0.00583)
667
667
667
Observations
667
655
667
667
0.851
0.764
0.544
R-squared
0.635
0.297
0.720
0.799
48
48
48
Number of state
48
47
48
48
4.923
54.72
0.541
Mean of D.V.
0.937
289.6
1.038
2.402
1.959
16.77
0.328
S.D. of D.V.
0.298
318.8
0.326
0.767
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1, Sample period: 1930-1943. State fixed effects
included. IPR, TB, Cancer, Heart.:#deaths/1,000 pop. IMR, MMR: infant/maternal deaths per 1,000
Convergence in pneumonia-influenza
mortality rate post-sulfa
Convergence in pneumonia-influenza
mortality compared to tuberculosis
Convergence by treatability
-250
-200
-150
-100
-50
0
Flu & Pneumonia vs. Tuberculosis
0
100
Flu & Pneumonia
TB
200
i_disease
300
Flu & Pneumonia
fit line Tuberculosis
400
Convergence Regressions
TREATED
VARIABLES
(post==1)*base mortality
Observations
R-squared
Number of state
Mean of D.V.
S.D. of D.V.
UNTREATED
(6)
(5)
Cancer
Diarr.
(1)
IPR
(2)
MMR
(3)
IMR
(4)
-0.336***
-0.227***
-0.152***
-0.369***
-0.343**
0.0225*
0.169***
(0.0480)
(0.0396)
(0.0198)
(0.0146)
(0.0243)
(0.0123)
(0.0165)
667
0.792
48
0.937
2.980
667
0.863
48
4.923
1.959
667
0.810
48
54.72
16.77
667
0.784
48
0.541
3.283
655
0.510
47
289.6
318.8
667
0.743
48
1.038
3.265
667
0.845
48
2.402
7.665
TB
(7)
Heart
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1. State and Year FE, Period: 1930 – 1943, Base: 1930 to
1935. IPR, TB, Cancer, Heart dis.: no. of deaths among 1000 pop. IMR/MMR per 1000 births.
Data

Outcomes



log family income, educational attainment, college attendance, current
employment, number of children born, disability preventing work,
difficulty with mobility, self-care, cohort size
We compute (weighted) means for race X gender X birth state X birth
year X census year cells
State-year varying controls


Diarrhea, TB, heart disease and cancer mortality rates from US Vital
Statistics
State income per capita, number of hospitals, physicians, schools and
educational spending from various sources
Educational Outcomes
(1)
Education
(2)
Education
(3)
Education
(4)
College
(5)
College
(6)
College
0.245**
(0.114)
0.00466
(0.0167)
0.247**
(0.119)
-0.00596
(0.0171)
0.151**
(0.0718)
-0.0238***
(0.00791)
-0.0154
(0.0150)
-0.00224
(0.00197)
-0.00840
(0.0151)
-0.00179
(0.00195)
0.0151
(0.0132)
-0.00306*
(0.00171)
4,860
Education
4,763
Education
4,763
Education
4,860
College
4,763
College
4,763
College
0.214**
(0.0872)
0.0154
(0.0104)
0.201**
(0.0917)
0.0137
(0.0118)
-0.0454
(0.0717)
0.00141
(0.00927)
-0.0164
(0.0153)
-0.00338*
(0.00182)
-0.00928
(0.0119)
-0.00129
(0.00144)
0.000167
(0.0117)
-0.00176
(0.00196)
4,887
4,790
4,790
4,887
4,790
4,790
Males
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
Females
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
ControlsEvery column includes birth state and birth year fixed effects (interacted with race); base: 1930-1935, disease is per 1000.
Census Year x Race FE
Y
Y
Y
Y
Y
Y
Disease Controls (level)
N
Y
Y
N
Y
Y
Econ. Control and State Specific Trends
N
N
Y
N
N
Y
Income, Employment, Fertility
(7)
Ln(FamInc)
(8)
(9)
Ln(FamInc) Ln(FamInc)
(10)
(11)
(12)
Employed
Employed
Employed
Males
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
0.0471***
(0.0176)
-0.000554
(0.00181)
0.0468***
(0.0168)
-0.00132
(0.00188)
0.0404*
(0.0213)
-0.00392
(0.00370)
0.0139**
(0.00671)
0.000460
(0.000911)
0.0152**
(0.00746)
-0.000100
(0.000798)
0.0202*
(0.0104)
-0.00319**
(0.00156)
4,839
4,743
4,743
3,718
3,641
3,641
Chborn
Chborn
Chborn
Ln(FamInc)
Ln(FamInc) Ln(FamInc)
Females
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
0.0406**
(0.0179)
0.00464**
(0.00210)
0.0346*
(0.0183)
0.00310
(0.00210)
0.00262
(0.0223)
0.00196
(0.00345)
0.114
(0.0821)
0.00910
(0.0116)
0.105
(0.0840)
0.00876
(0.0115)
0.0518
(0.0570)
-0.00252
(0.00751)
4,875
4,778
4,778
3,652
3,577
3,577
Controls
Every column includes birth state and birth year fixed effects (interacted with race) ; control dis: level ; base: 1930-1935
Census Year x Race FE
Y
Y
Y
Y
Y
Disease Controls (level)
N
Y
Y
N
Y
Econ. Control & State Specific Trends
N
N
Y
N
N
Y
Y
Y
Disability
(1)
Work Disability
(2)
(3)
Work Disability Work Disability
(4)
(5)
(6)
Difficulty with
Mobility
Difficulty with
Mobility
Difficulty with
Mobility
Males
Post*Base(Pneumonia)
Post*Base(MMR)
N
-0.0188**
(0.00757)
-0.0194***
(0.00635)
-0.0141
(0.00895)
-0.0106**
(0.00515)
-0.00928*
(0.00493)
-0.0122
(0.00760)
-0.00262***
(0.000929)
-0.00182**
(0.000839)
0.000585
(0.00117)
-0.00178**
(0.000807)
-0.00154*
(0.000887)
0.000723
(0.000972)
3,718
3,641
3,641
2,440
2,392
2,392
Difficulty with
Mobility
Difficulty with
Mobility
Difficulty with
Mobility
Work Disability
Work Disability Work Disability
Females
Post*Base(Pneumonia)
Post*Base(MMR)
N
-0.0103
(0.00964)
-0.00854
(0.00871)
0.00884
(0.00717)
-0.00787
(0.00750)
-0.00517
(0.00747)
0.00538
(0.00912)
-0.00365***
-0.00261**
-0.000606
-0.00198
-0.00195
-0.000309
(0.00127)
(0.00126)
(0.00113)
(0.00134)
(0.00149)
(0.00170)
3,739
3,663
3,663
2,460
2,416
2,416
Controls
Every column includes birth state and birth year fixed effects (interacted with race) PNA, MMR: no. of death among 1,000pop/births; base: 30-35
Census Year x Race FE
Y
Y
Y
Y
Y
Disease Controls (level)
N
Y
Y
N
Y
Y
Y
Disability
(7)
(8)
(9)
Difficulty with Self-Care
Difficulty with Self-Care
Difficulty with Self-Care
-0.00158
(0.00370)
0.001000
(0.00340)
-0.00373
(0.00673)
-0.00168**
(0.000657)
-0.00125**
(0.000582)
0.000938
(0.000899)
2,440
2,392
2,392
Difficulty with Self-Care
Difficulty with Self-Care
Difficulty with Self-Care
-0.00305
(0.00486)
-0.00149*
(0.000790)
-0.00211
(0.00482)
-0.00130
(0.000882)
0.00175
(0.00684)
-0.000284
(0.00121)
2,460
2,416
2,416
Males
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
Females
Post*Base(Pneumonia)
Post*Base(Maternal Mortality)
N
Controls
Every column includes birth state and birth year fixed effects (interacted with race), PNA, MMR: no. of death among 1,000 pop/births) ; base: 30-35
Census Year x Race FE
Y
Y
Disease Controls (level)
N
Y
Y
Y
Econ. Control & State Specific Linear Time Trends
Y
N
N
IV Estimates
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Educ.
College
Log Income
Employed
Disab. Work
Diffi. Mob
Diff. Care
-1.016**
0.0591
-0.190**
-0.0655***
0.0904
0.0594**
0.0110
(0.495)
(0.0770)
(0.0817)
(0.0252)
(0.0596)
(0.0287)
(0.0145)
-0.0624
0.0114
-0.0109
-0.00506
0.0193**
0.0105**
0.00548**
(0.0845)
(0.0101)
(0.0122)
(0.00431)
(0.00854)
(0.00488)
(0.00251)
4,763
4,763
4,743
3,641
3,641
2,392
2,392
Educ.
College
Log Income
Child born
Disab. Work
Diffi. Mob
Diff. Care
-1.023**
0.0550
-0.185**
-0.559
0.00954
0.0445
0.0239
(0.444)
(0.0611)
(0.0834)
(0.461)
(0.0566)
(0.0307)
(0.0246)
-0.138**
0.00943
-0.0271**
-0.0800
0.0169**
0.0109*
0.00679*
(0.0658)
(0.00850)
(0.0124)
(0.0711)
(0.00708)
(0.00576)
(0.00395)
4,790
4,790
4,778
3,577
3,663
2,416
2,416
Male
Pneumonia
MMR
N
Female
Pneumonia
MMR
N
Controls all includes control diseases mortality (level) and birth state/year fixed effect (interacted with race). F test ~10
Simulation- women
Reduced form coefficients-
Educ.
College
Log income
Chborn
Pneumonia
0.201**
-0.00928
0.0346*
0.105
Maternal mortality
0.0137
-0.00129
0.0031
0.00876
10th percent pneumonia/1000
0.825
0.233
-0.014
0.044
0.130
0.388
-0.023
0.073
0.215
10th percentile MMR/1000
4.9
90th percentile IPR/1000
1.39
90th percentile MMR/1000
Effect at 10th percentile for
both
Effect at 90th percentile for
both
7.91
Simulation- men
Reduced form coefficients-
Educ.
College
Log income
Employed
Pneumonia
0.247**
-0.0084
0.0468***
0.015**
Maternal mortality
-0.00596
-0.00179
-0.0013
-1E-04
0.175
-0.016
0.032
0.012
0.296
-0.026
0.055
0.020
10th percentile pneumonia/1000
0.825
10th percentile MMR/1000
4.9
90th percentile pneumonia/1000
1.39
90th percentile MMR/1000
Change at 10th percentile from
both
Change at 90th percentile from
both
7.91
Comparison of effect sizes
A state with the mean pre-sulfa pneumonia mortality rate saw a
post-sulfa education increase of 0.25 years and an income
increase of 4%.
 Influenza and pneumonia death rate pre/post 1937: 1.1 – 0.79
Almond (2006) estimates that the cohort exposed to the influenza
epidemic of 1918 had 0.25 years less education and income
lower by 6% percent
 Influenza &pneumonia death rate 1917-1918 in %: 1.16- 4.91
 Influenza death rate 1917-1918 in %:
0.17 – 2.9
Suggests pneumonia more scarring than influenza.
ITT<ATT

We are estimating the intent to treat (ITT) i.e. the
effect of sulfa averaged across the population it is
supposed to help

This will be smaller than the ATT to the extent that
not everybody could afford or access sulfa drugs. e.g
r/u, m/f.
The cost of a complete course was $28-$100 (in
2008 US $) or $4.3 per patient per day.

Pooled sample: coefs on cohort*base
white females
white males
black females
black males
Other Results

Stratifying by race



Coefficients in preferred specification (with all controls and state
trends) generally smaller in magnitude for blacks vis-à-vis whites
among males; no consistent pattern with females
Consistent with whites having preferred access to medical treatment
(JLS).
Falsification check – placebo interventions in 1935 and 1939

Precisely estimates zeros for most outcome variables
Mechanisms?


Endowments alone? Or endowments + compensating or
reinforcing endowments? Effects only in adulthood or
differences seen in adolescence?
We examine impact of program on whether child attended
school in the two months prior to the enumeration date of the
1950 census (“marginal” cohort is13 years old)



Schooling is an outcome in itself; but mechanism for earnings, empl,
fertility.
Caveat 1: compulsory school laws generally in place by early 1930s,
so lack of attendance could be due to variety of factors and may be
thought of as an outcome in addition to a mechanism
Caveat 2: School attendance only available for a subset of the sample
so sibling FE not possible
Schooling of children 7-18 years old
(1)
School
(2)
School
(3)
School
(4)
High Grade
(5)
(6)
High Grade High Grade
0.00173**
(0.000704)
0.00171**
(0.000681)
0.00201***
(0.000547)
0.141***
(0.0355)
957
School
943
School
943
School
957
High Grade
0.00163***
(0.000465)
0.00163***
(0.000461)
0.00137**
(0.000636)
0.0534**
(0.0261)
0.0510*
(0.0293)
0.0165
(0.0418)
939
923
923
939
923
923
Males
Post*Base(pneumonia)
N
0.129***
(0.0337)
0.0535
(0.0496)
943
943
High Grade High Grade
Females
Post*Base(pneumonia)
N
Controls
Every column includes birth state and birth year fixed effects (interacted with race); control dis.: log; base: 1930-1936.
Pneumonia is deaths per 100,000pop; period: 1932-1943
Census Year x Race FE
Y
Y
Y
Y
Y
Economic and Disease Controls
N
Y
Y
N
Y
State Specific Linear Time Trends
N
N
Y
N
N
Y
Y
Y
Conclusions

There is some evidence of sulfa-induced declines in mortality
in early childhood exerting positive long-run effects on
income, educational attainment, employment and work
disability

The effects are fairly substantial though in cases they are
sensitive to controls for to state-year varying variables

The evidence is more robust evidence for men, especially
white men.

Black men record stronger effects on prob(poverty).
Conclusions contd

Impact from pneumonia reduction > impact from MMR
reduction for SES

MMR has more of an impact on disability; more of an impact
on black people for eg. their education.

Results for adolescent school attendance suggest reinforcing
parental investments

Implications for developing countries, where childhood
pneumonia remains a leading cause of death
Work in progress

Intergenerational effects



Currently looking at data from Collaborative Perinatal Project:
longitudinal data for the early 1970s that include information on
offspring of pre/post sulfa cohorts. Rich set of indicators.
Preliminary findings show association between conditions faced by
mothers during birth year on the birth weight, motor development and
IQ of their children
Mechanisms

These data allow us to look more carefully at parental investments
The (provisional) end
Please email us with any questions or
comments