Big Data, Social Determinants of Health, and Health ...

Big data, social determinants of health, and health inequities

Sandro Galea

Boston University School of Public Health

1. The trouble with the population’s health

expectancy vs. health expenditure over time (1970-2014) I..,.,._,"'"""'° Health pending mea ure the con umption of health care good and services, including per onal health care (curative care, rehabilitative care, long-term care ancillary ervice and medical goods) and collective ervice (pr vention and public health . rvic s as well a h alth admini tration), but ex luding spending on inve tment . Shown i total health e:q,enditur (financed by public and private ource ).

84

83

82

81

80

>, (_) 79 C ro +-' (_) 78 (].) Q_ X

77 w (].) -~ 76

75

74

73

72

71

70

~ Spain

,/ 11taly

South K?~f!el

Japan

• _, . France Australia" Luxembourg

~ nad~ Sweden

~ Switzerland

Our World in Data. “The link between health spending and life expectancy: The US is an outlier.”

<https://ourworldindata.org/the-link-between-life-expectancy-and-health-spending-us-focus> Accessed September 7, 2016

https://ourworldindata.org/the-link-between-life-expectancy-and-health-spending-us-focus

..c +-' L..

80

:': 76 co >, u C co +-' u Q) a. ~ 72 ~ .....J

68

Total Population

1980 1990 2000 Year

2010

I I •

\ \

2020

Total Population -- Latino -- White -- Black

Andrasfay T, Goldman N. Reductions in 2020 US life expectancy due to COVID-19 and the disproportionate impact on the Black and Latino populations. PNAS 2021 Vol.

118 No. 5 e2014746118

2. Understanding population health

“The health outcomes of a group of individuals, including the distribution of such outcomes within the group ”

Kindig D, Stoddart G. What is population health? American Journal of Public Health 2003; 93;380-383.

“The health outcomes of a group of individuals, including the distribution of such outcomes within the group ” Why? So that we may intervene

Kindig D, Stoddart G. What is population health? American Journal of Public Health 2003; 93;380-383.

tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt

t t t t

t t

t t t

t t t

t t

tttttt tttttt tttttt

3. Four principles that can help

a. Population health data as continuous, not binary

Figure 1. Distribution of BMI in two populations illustrating health as a continuum in the

population

Panel A

%

20

15

5

BMl=30

Mean=23

18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

BMI

Rose G. Sick individuals and sick populations. International Journal of Epidemiology. 1985;14:32-38

40

30

20

10

. ... '\ \

' \ \ \ \ \ t

'

- NQN .. C.HD

----CHO

... ... "'14 0 t ~ I 01 180 2.2.0 21 G 300 MO 3 B 0 ◄201 4EK> 490 500 -1-

S EAU M CHOLESTEROL

Figure 3 Percentag.e cli:st.r"b fon ol serum hoJe.s erol eveJs {1:ng/dl) in m · n a,ged 50-62 who did or did not suJbsequently velop coronary heart. dtseitJ,se (Framingha . 1ud-y 5)

Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. American Journal of Epidemiology 2004; 159:882-890.

OR= 1.5 .,- OR= 3.0 ,,--,, ,, , / /

{5 , ,

f5 , , , / / ,, ,, ,, ,, __ .,,,

"'"

-1 0 2 3 -1 0 2 3

OR =9.0 ✓,-- .... OR= 25.0 ,,.--.... ,, ....

' ,, / , -' ,

D ,

D .," , ' , ' ,, ,, / ....

' ,, ,, .,,, .... __ ,,., _,, ----1 0 2 3 · 1 0 2 3

OR = 70.0 ,,.-- .... OR= 350.0 .,.-- ..... ,, ' , / ' ' / ' / '

, i5 ' , o ,

' , , ' / / ' , / ,,,, ,,,,

,,, __ ., · 1 0 2 3 -1 0 2 3

FIGURE 2. Probability distributions of a marker, X, in cases (solid curves) and controls (dashed curves) consistent with the logistic model log~ itP(D = 1 IX) =a.+ ~X. It has been assumed that X has a mean of 0 and a standard deviation of 0.5 in controls so that a unit increase represents the difference between the 84th and 16th percentiles of X in controls. The marker is normally distributed with the same variance in cases. The odds ratio (OR) per unit increase in Xis shown.

b. Illuminating ubiquitous causes

Figure 1. A metaphor for ubiquity

The goldfish are surrounded by water and everything they do is influenced by the quality of the

water in which they live; therefore, water is a ubiquitous factor influencing the fish and needs to

be taken into consideration every time we may want to improve the lives of the fish.

September 17, 1989 Crack Babies: Crack's Toll Among Babies: A Joyless View

September 6, 1988 The Worst Threat Is Mom Herself

, Cocaine: Litany of Fetal Risks

By DouglBB J. Besharov

LAST WEEK in this city, Greater Southeast Cam· munity Hospital released a 7-w ek-old- ~by to her home!~. drug-addicted mother eveo·.thouab·

the child was ~t severe risk of pulmona,ry arreet. ?ht hOIJ)ital'" dplanation: "Because lth6 mothu} , demanded that the baby be released."

Grows

The boepital provided the mother with an ;tpnea man--. itor to warn her if the baby stopped breathing while aslttp, and trained her in CPR. But on thJ? very first nipt, the mother went out drinking and left the child at a friend'& house-without the monitor. Within seven hours, the baby was dead. Like Dooney Waters, the t;. year-old living in hi mother' · drug den, whose shocking story was reported in The Washington Post last week, this child was all but abandoned b·,-----------------------------------ities. c

I

.._,____.

Yo Can Sto S O lr .; ~ ~ '-~~·~-..

STOP DRINKING ALCOHOL

By C[a l9. .. Beck

http://www.yourweightmatters.org/portion-sizes-changed-time/

Food

Bagel

Muffin

C heesebu rg er

Pasta (Spaghett i & Meatballs)

French Fries

Soda

Theater Popcorn

Turkey Sandwich

Pizza

...... ~911

20Years Ago

140 calories ' •~ (3" diameter)

.· .·· •\~

210 calories (1.5 oz)

333 calories

:>,:~~ 500 calories



270 calories (5 cups)

320 calories

- 500 calories (2 slices)

Today

350 calories ( 6" diameter)

500 calories (4oz)

590 calories

1025 calories

610calories (6.9 oz)

250 calories (20 oz)

630 calories (1 tub)

820 calories

850 calories (2 calories)

c. The role of co-occurring causes

How much of our obesity risk is determined by our genes?

tttttttttttttttttttt tttttttttttttttttttt tttttttttttttttttttt tttttttttttttttttttt tttttttttttttttttttt

= GE+

t rt t t t rt

t t rt t t

t t rt t

t

= ENV+t

ttttt ttti tttt

-- tttt

= OB+

t t - t

-~ ~

~ ~ t ~- - t - - " -- -

~~ ~~ ~~ ~~

t

= ENV+

Scenario 1

= OB+= GE+

t ntttntttttttttttt tt t ntttttt t ntt ntttttt r: tttt n tttttt nttf t ttttt tttt_nrftttttt ti nt nttt ntttttttttttt

t t t

= ENV+

Scenario 1

= OB+= GE+

t t tttt ttt tttttt tt t t tttttt t t ttt tttttti t ttttt tttttttttttfft ttttt tttttt ttttttt ft t ttt tttt tttttttttttt

t t t

= ENV+= OB+

Scenario 2

= GE+

t t t r: - ttt t t t t ttti

t f t

= ENV+= OB+

Scenario 2

= GE+

t t t f tttt t t t t ttti

t t t tt

t f t

Therefore under a very plausible assumption of co-

occurring causes, the gene-obesity association can only be

understood if we understand the urban factors that create

the conditions for disease

Rampersaud E, Mitchell BD, Pollin TI, Fu M, Shen H, O'Connell JR, et al. Physical Activity and the Association of Common FTO Gene Variants With Body Mass Index and

Obesity. Arch Intern Med. 2008;168(16):1791-1797.

29.5

29.0

28.5

28.0

27.5

~ 27.0

26.5

26.0

25.5

25.0

Genotype - AA - GA

GG

24.5-+------r---~-~--+-----r-------r-------r--------. -2.0 -1 .5 -1.0 -0.5 0 0.5 1.0 1.5 2.0

Residualized PhysjcaJ Activity

figure 3. Predicted body mass index (BMI) * calculated as weight in kilograms divided by height in meters squared, as a function of residualized age- and sex-specific In-transformed physical activity accelerometer counts according to FTO rs1861868 genotypes. On the left side of the plot (low physicaJ activity) 1 BMI levels are strikingly dissimilar between rs1861868 genotypes. In contrast* on the right side of the plot, similar BMI levels can be seen across genotypes* particularly in subjects mth very high levels of physicaJ activity.

Meigs JB, Shrader P, Sullivan LM, McAteer JB, Fox CS, Dupuis J, Manning AK, Florez JC, Wilson PW, D'Agostino RB Sr, Cupples LA. Genotype score in addition to

common risk factors for prediction of type 2 diabetes. N Engl J Med. 2008 Nov 20;359(21):2208-19

B 25

Cl)

~ CV 20 ..c ns ·-C

4-0 CV 15 u C

~ ·-u C 10 -CV > • ... ns -:::J 5 E :::J u

0

-

P<0.001

-

-

-_,....

- --

I

--

--

I

16- 20

Genotype Score

--

--

I

Meigs JB, Shrader P, Sullivan LM, McAteer JB, Fox CS, Dupuis J, Manning AK, Florez JC, Wilson PW, D'Agostino RB Sr, Cupples LA. Genotype score in addition to

common risk factors for prediction of type 2 diabetes. N Engl J Med. 2008 Nov 20;359(21):2208-19

A 100

90 P 0.001

80 (I) .... u 70 CV ·-.c ~ 60 V')

\,t-0 50 CV t)O ns

40 .... C CV ~ 30 £ Diabetes 20

10 No diabetes

0 I I I I I I 0 8 1012141618 20 22 24 26 28 30 32 34 36

Genotype Score

d. The centrality of health equity

Bor J, Cohen G, Galea S. Population Health in an Era of Rising Income Inequality: United States, 1980-2015. The Lancet. 2017; 389(10077):1475-1490

-~ "' (1)

.2:)

.c. -"' (1)

-0 1tl (1) C'l

"' -0

t 0,.

~

B 90

85

80

75

70----Ql Q2

82

80

78 78

Q3 Q4 QS Ql Q2

1930

Birth cohort

Figure 5; Widening income-related inequalities in survival across birth cohorts

89 88

83

Q3 Q4 QS

1960

46National Academy of Medicine. The growing gap in life expectancy by income: Implications for federal programs and policy responses. 2015.

http://www.nap.edu/download.php?record_id=19015#

lnequaUty in life expectancy widens for women

Wealth ier women can expect to Ii e longer than th ir parents did, while life

expectancy for poor women may have declined.

90

• 85

83.1 Upper middle •--------------------0 • •

1980

Life ex ectancy for 50- ea '-<>Ids in a gi n y ar, b

ye rs

79. 7 Lower middle

78.3 Poorest

2010,

uintil of incom over th previous 10

Sou re : a ional Ac d mi of Sci nc , En in rin nd M ,cit e

Figure 1. Gain ing overall population heath

0

intervention

lnterv ntion adding: 1 DALY

Intervention adding

10 DALY

DALY-50

DALY= 25

DALY= 51

DALY= 25.5

DALY= 60

DALY - 30

In uali o 25 DAY

I equa ·, o 25.5 0 L

Figure 2. Gaining overa ll population health while creating health inequalities

t DALY = 50

Before } No intervention inequality

t DALY =50

High SES

t DALY =60

After } Inequality of 5

intervention DALY

Low SES t DALY= 55 •••

4. Methods for population health

tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt tttttttttttttttttttttt

Populations are

1. Heterogeneous, ie have diversity of agents

2. Characterized by nonlinear dynamics

3. Characterized by contact structure, networks,

organization

4. Have feedback, adaptation, learning, evolution

5. Stochastic with important tails

6. Display emergent properties

http://i.telegraph.co.uk/multimedia/archive/00796/crowded-britain_796405c.jpg

Populations are

1. Heterogeneous, ie have diversity of agents

2. Characterized by nonlinear dynamics

3. Characterized by contact structure, networks,

organization

4. Have feedback, adaptation, learning, evolution

5. Stochastic with important tails

6. Display emergent properties

Complex systems

Foresight - Tackling Obesities: Future Choices:<http://www.foresight.gov.uk/OurWork/ActiveProjects/Obesity/KeyInfo/Index.asp>

Figure 5.2: The full obesity ystem map with thematic cluster (see main text 5.1.2 for discuss1onJ17 ' 8 Var ab es ,He represented by bo es positive causal re at1ons,1Ips are ~ "' :: d by solid arrows and neg at v relat1ons'11ps b dott d mes The central ergIne Is h ghhgh1ed in ore. g at e centre of the map

Maps

Full Generic Map Thematic Clusters (fi lled)

Food Prod

SOcJol

I ( Psychologlul)

( ]I . Economk .

( Food j ( Ac1Mty ~ ( lnfrutructvre

(oeve loptM.nttl

_P_o_,1t_1v_e_ln_n_u_•n_o_• __.► ( 8'ologic;, I

cgative lnftuence J ..................... ( Mod,n l j

Causes and counterfactuals

Observed

Counterfactual (parallel universe)

'.1/ /

'"(~>~ \::;}


Observed


-- ~(1:1 :~ ~; -=-- _, ;;

·:: ;~, ~~-<:;•;::, i ,,. __, ,. .,

, ~ ..L 4' , . . .., ~ ,.


Observed


-- ~(1:1 :~ ~; -=-- _, ;;

·:: ;~, ~~-<:;•;::, i ,,. __, ,. .,

, ~ ..L 4' , . . .., ~ ,.

' '

“Treatment”Simulation

“Control”Simulation Exposed UnexposedTreatment Control

Population of Interest

Idealized Randomized Controlled Trial

Simulation Study(Agent-Based Counterfactual) Observational Study

TreatmentAssignment

ComputerSimulation

Study Design

Manipulation

Disease Distribution

CausalEffect?

Stratification or G-methods

Exposure or TreatmentDistribution

Yes Yes† Yes*

Marshall BDL, Galea S. Formalizing the role of complex systems methods in causal inference and epidemiology. American Journal of Epidemiology. 2015;181(2):92-99. .

' ' ' ' ' ' ' ' ' ' -----:----'

---:----

' ----~---' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' ' '

' ' ' --------1--, ' ' ' ' '

D






TreatmentAssignment

ComputerSimulation

Study Design

Manipulation


CausalEffect?



Yes Yes† Yes*


' - - - - - - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

D

• . -: . . . · . . . · . . ·· .






TreatmentAssignment

ComputerSimulation

Study Design

Manipulation


CausalEffect?



Yes Yes† Yes*


' - - - - - - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

D

• . -: . . . · . . . · . . ·· .

“ Everything should be made as simple as possible, but not simpler ”

Attributed to Albert Einstein.

Simple approaches, a foundational myth

The effectiveness of simple approaches?

Davey Smith G, Ebrahim S. Epidemiology-is it time to call it a day? International Journal of Epidemiology. 2001;30:1-11.

150 140 130 120 110 100 90

,B 80 ~ 70 0 60

50 40 30 20 10

Pump closed

1--- Onset of Fatal Cases I 0 -l--...----,.-------,----------.....!....--.--.....:......:::::=:~:::::;:=:::::2::=:::;=,..:-==~~~>-,

Aug Sep 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 12th 13th 14th 31st 1st

( 668 Deaths: Onset date unknown = 127)

5. The big picture

Institutions

Neighborhoods and communities

Living conditions

Social relationships

Individual risk factors

Genetic/ constitutional factors

Pathophysiologic pathways

ividual/populat health

Figure 2.1 Levels of influence on the health of populations. Source: Modified from Kaplan, G.

What's wrong with social epidemiolog)7j and how can we make it better? Epid R ev. 2004; 26: 124-135.

twitter/@sandrogalea

[email protected]

Big Data, Social Determinants of Health, and Health ...

Documents

Transcript of Big Data, Social Determinants of Health, and Health ...