Open Forum Infectious Diseases

MAJOR ARTICLE

Heterogeneity Between States in the Health and Economic

Impact of Measles Immunization in the United States
Angel Paternina-Caicedo,1 Julia Driessen,1 Mark Roberts,1 and Willem Gijsbert van Panhuis2
Departments of 1Health Policy and Management and 2Epidemiology, University of Pittsburgh Graduate School of Public Health, Pittsburgh, Pennsylvania

Background.  Vaccines have been used successfully for disease elimination programs in many countries. Evidence on the impact
of vaccination programs can support decision-making among medical practitioners and policy makers to improve immunization
rates. We estimated the health and economic impact of measles vaccination for each of the 48 contiguous states and the District of

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

Columbia since 1964.
Methods.  For each state, we fitted multiple time-series models to prevaccination data and used the best-fitting model to predict
counterfactual cases that would have occurred in the absence of vaccination. We then subtracted observed from counterfactual mea-
sles cases, deaths, and related costs to estimate the impact of vaccination.
Results.  We estimated that 149 million children were vaccinated against measles in the United States between 1964 and 2014, at
a cost of $12.2 billion, and that vaccination prevented 29.8 million cases, 32 000 deaths, and $25.8 billion in societal costs. The impact
exceeded the national average in 70% of Western and Northeastern states, compared with only 24% of Southern and Midwestern
states.
Conclusions.  The significant health and economic benefit of measles vaccination in the United States should encourage contin-
ued investments to sustain and expand vaccination programs globally.
Keywords.  measles; measles vaccine; vaccine impact.

Measles vaccination is one of the most successful public health enacted legislation to end personal belief exemptions [9], and
programs worldwide and has prevented an estimated 17.1 mil- Pennsylvania reduced the time allowed for children to get vac-
lion deaths between 2000 and 2014. In the United States alone, cinated to 5 days from school entry, from 8 months previously
vaccination has prevented an estimated 0.5–3.8 million measles [10]. Several studies have assessed the epidemiological and eco-
cases per year [1–4]. Despite a declaration of measles elimin- nomic impact of measles vaccination in the United States at the
ation in the United States in 2000, vaccine hesitancy and rein- national level [1–4], but information about the heterogeneity of
troductions of the virus have led to continued outbreaks [5, impact between states is limited.
6]. At least 40% of measles cases in the United States from 18 We estimated the number of cases and deaths prevented,
outbreaks that occurred between 2000 and 2015 were unvac- and cost savings, by measles vaccination for each state since the
cinated [7]. In 2017, Minnesota experienced its largest measles introduction of this vaccine in 1964.
outbreak since 1990, 79 cases occurred in a community where
vaccination rates had dropped to 42% [8]. METHODS
Vaccine regulations can vary between states, leading to heter-
Data Sources
ogeneity in vaccination coverage rates and in the risk of disease
Project Tycho is a repository for disease surveillance data that
outbreaks. For example, California and Vermont have recently
contains data for all notifiable diseases in the United States
that have been reported provisionally by states to the Centers
for Disease Control and Prevention (CDC) on a weekly basis
Received 19 April 2018; editorial decision 30 May 2018; accepted 13 June 2018 since 1888. Measles cases were available from Project Tycho
Previous presentations.  MIDAS Meeting, Atlanta, May, 2017. The Health and Economic
Impact of Measles Immunization in the United States: A State-Level Analysis From 1931 to 2014. for 1931–1992, and we used measles cases reported by annual
Correspondence: W. G. van Panhuis, MD, PhD, 130 De Soto Street, Office A737, Graduate CDC surveillance summaries for 1993–2014 [11]. We excluded
School of Public Health, University of Pittsburgh, Pittsburgh, PA 15261 (wilbert.van.panhuis@
pitt.edu). Hawaii and Alaska because these states were not included in
Open Forum Infectious Diseases® the national surveillance system until the 1950s. We found that
© The Author(s) 2018. Published by Oxford University Press on behalf of Infectious Diseases 32% of weekly Project Tycho data were missing, mostly between
Society of America. This is an Open Access article distributed under the terms of the Creative
Commons Attribution-NonCommercial-NoDerivs licence (http://creativecommons.org/licenses/ 1980 and 1992, when case counts were low [1]. Without impu-
by-nc-nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any tation, nationally aggregated Project Tycho data overestimated
medium, provided the original work is not altered or transformed in any way, and that the work
is properly cited. For commercial re-use, please contact journals.permissions@oup.com
CDC national data for some years that had both sources avail-
DOI: 10.1093/ofid/ofy137 able. We imputed missing weekly Project Tycho counts for each

Measles Vaccination in the United States  •  OFID • 1

state with the average of counts for the preceding and following to estimate counterfactual case counts during the vaccination
weeks for which data were available (linear imputation). Our period using a variety of counterfactual models [1, 22, 23].
imputed Project Tycho counts likely overestimated CDC data in A vaccine impact model based on prevaccination data compares
the vaccination period and would lead to conservative estimates a population with a vaccination program (vaccination period)
of vaccine impact. with a population without a vaccination program (prevaccina-
We collected the annual number of measles deaths from the tion period), and thus estimates an overall impact of the vaccine
National Vital Statistics System (NVSS) for 1961–2014 and that includes the direct and indirect (ie, herd immunity) effects
from US vital mortality statistics reports [12] for 1931–1960. of the vaccine [24].
We used national-level cost data for all-cause and mea- For each state, we fitted 72 different ARIMA models to the
sles-specific hospitalization in 1995 [2] and state-specific cost first 20 years of the prevaccination data using time as the inde-
data for all-cause hospitalization between 1969 and 2014 [13, pendent variable and measles incidence rates as the dependent
14] to estimate the direct health care costs of measles disease. variable (1931–1950). We used each model to predict (out of

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

We used the health care inflation rate to impute missing cost sample) the last 10 years of the prevaccination data (1951–1960)
data between 1931 and 2014. We assumed a higher cost of mea- using time as the independent variable. The 72 ARIMA mod-
sles hospitalization and measles-related encephalitis compared els comprised every combination of: (1) 6 specifications of the
with nonhospitalized measles. We used the probability of mea- autoregressive component (0–5), (2) 2 specifications of the dif-
sles hospitalization as reported by CDC surveillance [15] and ference component (0–1), and (3) 6 specifications of the moving
the probability of measles-related encephalitis from Zhou et al. average component (0–5). We defined the best model for each
[2]. We used state-specific annual income data between 1931 state as the model with the lowest mean squared error between
and 2014, as reported by the Bureau of Economic Analysis [16], the observed and predicted values for the 1951–1960 testing
to estimate indirect costs of measles. period (Supplementary Table 2 and Supplementary Figures 2
Nationwide coverage data for the measles, mumps, rubella and 3). We then fitted the best model for each state to data from
(MMR) vaccine were available for all years between 1931 and the entire prevaccination period (1931–1960) and predicted
2014, except between 1986 and 1990 (these missing data years the counterfactual case incidence or death rates. We used the
were imputed using linear interpolation) [17]. State-specific lower and upper bounds of the ARIMA 80% confidence interval
vaccination coverage rates for 1 dose of the MMR vaccine as the lower and upper uncertainty bounds for counterfactual
were available from the National Immunization Survey (NIS) estimates (Supplementary Figures 4 and 5). The ARIMA confi-
for the years between 1995 and 2014 [18]. In the absence of dence interval became wider over time as predictions for years
2-dose coverage data from NIS, we assumed that 1-dose cover- far from the prevaccination period were less certain vs years
age reported in the NIS was similar to 2-dose coverage. We shortly after the prevaccination period.
used the ratio of national and state-specific vaccination cover- The ARIMA model for deaths in South Dakota did not con-
age rates during the years for which both were available to verge due to an extreme observation of 120 deaths in 1934,
estimate state-specific vaccination coverage rates for the years compared with an average of 6 deaths in other years. Instead of
1964–1994 (Supplementary Figure 1). We used the cost per ARIMA predictions, we used the mean prevaccination measles
vaccination dose for commercial entities to estimate the cost of mortality rate and its 95% confidence interval as the counterfac-
vaccination (Supplementary Table 1). In the absence of more tual estimate for South Dakota.
detailed information, we assumed that commercial vaccine
Costs of Measles Disease
prices would be representative of the cost of vaccines and also
We used nationwide cost data for 2 types of measles-specific
of vaccine delivery in the United States, given that most US vac-
hospitalization: (1) nonencephalitis measles (NEM), that is,
cines are purchased at discounted prices. Previous studies also
uncomplicated measles or measles with diarrhea; and (2) mea-
used commercial pricing to represent the cost of vaccination
sles encephalitis (EM). We used national cost data available for
[19–21]. We provide additional details about the study method-
1995 to compute the proportion of all-cause hospitalization
ology in Supplementary Text.
costs that was spent on each type of measles. We then multi-
Estimation of Counterfactual Cases plied these proportions by the all-cause hospitalization costs for
We used a time-series autoregressive integrated moving aver- each state and year to estimate annual state-level hospitalization
age (ARIMA) model to estimate the counterfactual number of costs for each type of measles (Supplementary Figure 6).
cases and deaths that would have occurred in the absence of We computed the cost related to NEM hospitalizations for
vaccination between 1964 and 2014. We separately estimated each state by multiplying the annual number of measles cases
counterfactual cases and deaths. ARIMA models represent by the probability and cost of NEM. We did the same for EM
autocorrelation, seasonal patterns, and trends over time and hospitalizations. We added the costs of NEM and EM hospi-
have been used previously to model infectious disease time-se- talizations to obtain the total direct costs related to measles
ries data [22]. Previous studies have used prevaccination data hospitalization.

2 • OFID • Paternina-Caicedo et al
 We estimated indirect costs related to measles using the by the vaccination coverage and the vaccine price per dose
human capital approach [25]. We assumed that 1 caregiver (Supplementary Text).
would be unable to work for the duration of the average
Health and Economic and Impact of Measles Vaccination
hospitalization period for NEM and EM [2]. We computed
We estimated the epidemiological impact of measles vaccination
the average hourly and daily income per state from annual
by subtracting the estimated number of counterfactual cases or
income information (Supplementary Figure 7) [26], assum-
deaths from the observed number. We calculated the costs pre-
ing 8-hour work days and 40-hour work weeks. We then
vented by subtracting the total societal costs of observed measles
multiplied each measles case by the average number of days
during the vaccination period and the costs of the vaccination
hospitalized and by the average daily income (Supplementary
program from the total societal costs of counterfactual cases.
Text).
We reported all costs in this study in 2014 dollars. We stratified
We estimated the average direct and indirect cost of a mea-
our impact estimates by phase of the vaccination period: (1) the
sles case per state and year based on 1000 Monte Carlo sim-
vaccine introduction phase, (2) the 1-dose phase (starting when

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

ulations of the annual direct and indirect costs of measles,
coverage reached 60% in 1971), and (3) the 2-dose phase (start-
per state, using a gamma distribution. We then multiplied the
ing with introduction of the second dose in 1990).
Monte Carlo cost averages with ARIMA counterfactual case
We computed the nationwide health and economic impact of
projections to obtain the total direct and indirect cost per state
vaccination as the sum of all state-level cases, deaths, and costs
and year. We multiplied the Monte Carlo cost average with the
prevented.
ARIMA counterfactual 80% uncertainty bounds to represent
We used the Wilcoxon rank-sum test to estimate the asso-
uncertainty bounds of cost estimates. We defined total societal
ciation between the number of cases and costs prevented and
costs related to measles as the sum of direct and indirect costs.
state-level income, and between vaccination coverage and state-
We estimated the costs of measles for both the observed and
level income.
counterfactual scenarios.
Sensitivity Analysis
Costs of the Measles Vaccination Program We estimated the impact of measles vaccination for 8 additional
The US immunization program included 2 doses of the mea- scenarios, each using a different imputation method for missing
sles-containing vaccine from 1989 onwards. We assumed that data (substitution with 0s, substitution with a random count of
vaccination coverage represented a 1-dose vaccination sched- the same week in a different year), and a different model for esti-
ule between 1964 and 1989 and a 2-dose vaccination schedule mating counterfactual measles cases or measles-related deaths
after 1989. We estimated the costs of the vaccination pro- (linear model, prevaccination mean rates) (Supplementary
gram per state and year by multiplying the number of births Figure 8).

Table 1.  Observed and Prevented Measles Cases, Deaths, and Related Costs in the United States, With 80% Uncertainty Range

Prevaccination Vaccination

(1931–1963) Introduction (1964–1970) 1-dose Vaccine (1971–1989) 2-dose Vaccine (1990–2014) Entire Period (1964–2014)

Cases, millions
Observeda 16.81 1.14 0.39 0.05 1.57
Prevented — 2.49 (0.30 to 8.35) 10.12 (3.19 to 31.89) 17.17 (5.59 to 57.60) 29.78 (9.08 to 97.84)
Deaths, thousands
Observed 45.52 1.19 0.28 0.11 1.59
Prevented — 1.46 (–1.18 to 10.50) 10.51 (–0.28 to 76.48) 19.61 (–0.11 to 334.61) 31.57 (–1.57 to 421.59)
Health care costs, USD, billionsb
Estimated 4.27 (4.26–4.28) 0.28 (0.28 to 0.28) 0.16 (0.16 to 0.16) 0.03 (0.03 to 0.03) 0.47 (0.47 to 0.48)
Prevented — 0.65 (0.09 to 2.14) 4.86 (1.55 to 15.29) 16.71 (5.44 to 56.07) 22.22 (7.08 to 73.50)
Lost income, USD, billionsb
Estimated 3.43 (3.42–3.44) 0.34 (0.34 to 0.34) 0.16 (0.16 to 0.16) 0.02 (0.02 to 0.02) 0.53 (0.52 to 0.53)
Prevented — 0.90 (0.15 to 2.90) 4.59 (1.47 to 14.23) 10.20 (3.36 to 33.60) 15.69 (4.99 to 50.72)
Vaccination costs, USD, billions
Estimated — 0.56 (0.45 to 0.68) 2.05 (1.64 to 2.46) 9.54 (7.63 to 11.45) 12.15 (9.72 to 14.58)
Societal cost, USD, billions
Prevented — 0.99 (–0.32 to 4.48) 7.40 (0.97 to 27.47) 17.36 (–0.74 to 80.13) 25.75 (–0.08 to 112.07)
a
Observed cases, as reported by the US Centers for Disease Control and Prevention (CDC; data from Project Tycho and the CDC).
b
Estimated costs due to hospitalization or lost income associated with reported measles cases based on state-level cost estimates.

Measles Vaccination in the United States  •  OFID • 3

RESULTS phase), only 4 measles-related deaths were reported, on aver-
Since 1964, 149.4 million (M) children have been vaccinated age, per year for the entire United States.
against measles in the United States at a cost of $12.2 billion, We estimated that $0.6 billion was invested in measles
preventing 29.8M cases and 32 000 deaths and saving $25.8 bil- vaccination during the vaccine introduction phase, when
lion in societal costs (Table 1). 12.7M children were vaccinated; $2.1 billion was invested
during the 1-dose phase (45.1M children vaccinated), and
Nationwide Impact $9.5 billion during the 2-dose phase (91.5M vaccinated)
The national average prevaccination measles incidence rate (IR) (Table 1).
was 344 cases/100  000 (Table  1, Figure  1). During the vaccine Measles vaccination saved $0.72 per person in the entire
introduction phase, the annual IR dropped by 76% to 83/100 000, population of the United States during the vaccine introduction
and by a further 99% to 0.6/100 000 during the 2-dose phase. period. As vaccination coverage improved, cost savings more
Before vaccination, the national measles-related mortality than doubled during the 1-dose period to $1.71 per person, and

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

rate (mMR) was 0.93 deaths/100 000, representing an average of tripled to $2.44 per person during the 2-dose period (Table 1).
1379 deaths per year (Table 1). In the 6 years following vaccine On average, these cost savings represent a return of $2.12 dol-
introduction, the mMR dropped by 91% to 0.09 deaths/100 000. lars per $1 invested by the United States in vaccination between
In the 25  years between 1989 and 2014 (2-dose vaccination 1964 and 2014.

A Measles incidence rate, observed B Measles incidence rate, counterfactual

MW MW
NE NE

S S

W W
1931 1950 1970 1990 2010 1931 1950 1970 1990 2010

0 1k 2k 3k 4k per 100 000 pop. 0 1k 2k 3k 4k per 100 000

E Measles costs, observed F Measles costs, counterfactual

MW MW
NE NE

S S

W W
1931 1950 1970 1990 2010 1931 1950 1970 1990 2010

$0 $5 $10 $15 $20 per capita $0 $5 $10 $15 $20 per capita

C Measles death rate, observed D Measles death rate, counterfactual

MW MW
NE NE

S S

W W
1931 1950 1970 1990 2010 1931 1950 1970 1990 2010

0 10 20 30 40 per 100000 pop. 0 10 20 30 40 per 100000 po

Figure 1.  Annual state-level observed and counterfactual measles cases, deaths, and societal costs between 1931 and 2014 for each state, ordered by region: Midwest
(IA, IL, IN, KS, MI, MN, MO, ND, NE, OH, SD, and WI), Northeast (CT, MA, ME, NH, NJ, NY, PA, RI, and VT), South (AL, AR, DC, DE, FL, GA, KY, LA, MD, MS, NC, OK, SC, TN,
TX, VA, and WV), and West (AZ, CA, CO, ID, MT, NM, NV, OR, UT, WA, and WY). A, the annual observed measles incidence rate (IR). B, The annual observed and counter-
factual measles IR, C, The annual observed measles death rate. D, The annual observed and counterfactual measles death rate. The red line indicates the year of vaccine
introduction (1964).

4 • OFID • Paternina-Caicedo et al
Table 2.  Absolute Number of Measles Cases and Deaths Observed and Prevented in the United States Between 1964 and 2014, by
Region and State

Cases, 1000s Deaths

Observed Prevented (80% Uncertainty Range) Observed Prevented (80% Uncertainty Range)

Midwest 499 7350 (2040 to 24 417) 332 4855 (–331 to 68 557)

Iowa 52 199 (11 to 943) 10 635 (–10 to 1962)
Illinois 66 1135 (292 to 3936) 97 672 (–97 to 8740)
Indiana 52 396 (69 to 1590) 37 196 (–37 to 6501)
Kansas 11 184 (23 to 1065) 19 160 (–19 to 1736)
Michigan 113 1701 (511 to 5182) 34 317 (–34 to 13 296)
Minnesota 10 262 (63 to 986) 18 531 (–18 to 3162)
Missouri 12 202 (44 to 794) 35 490 (–35 to 9700)

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

North Dakota 15 80 (10 to 353) 2 306 (–1 to 1036)
Nebraska 5 77 (12 to 381) 3 347 (–3 to 1203)
Ohio 63 1204 (327 to 4027) 44 321 (–44 to 15 378)
South Dakota 3 12 (–1 to 139) 11 271 (–11 to 1885)
Wisconsin 97 1898 (680 to 5022) 22 607 (–22 to 3958)
Northeast 277 6750 (2279 to 20 682) 208 9364 (–207 to 39 863)
Connecticut 27 613 (224 to 1608) 12 856 (–12 to 2548)
Massachusetts 39 1287 (521 to 3108) 14 111 (–14 to 2574)
Maine 11 215 (49 to 846) 15 65 (–15 to 1097)
New Hampshire 5 66 (15 to 255) 1 7 (–1 to 982)
New Jersey 39 1814 (590 to 5429) 25 1718 (–25 to 5531)
New York 98 1987 (730 to 5218) 77 6422 (–76 to 20 003)
Pennsylvania 45 487 (69 to 3207) 62 –50 (–62 to 4078)
Rhode Island 8 94 (11 to 529) 1 28 (–1 to 2285)
Vermont 5 187 (71 to 482) 1 207 (–1 to 764)
South 502 7096 (1596 to 29 657) 783 10 551 (–777 to 237 461)
Alabama 29 244 (45 to 964) 44 65 (–44 to 4549)
Arkansas 6 110 (9 to 978) 26 –16 (–26 to 286)
District of Columbia 2 42 (9 to 181) 2 28 (–2 to 549)
Delaware 3 60 (11 to 276) 1 –1 (–1 to 2959)
Florida 31 704 (195 to 2332) 40 2983 (–40 to 22 400)
Georgia 6 233 (42 to 1126) 51 54 (–51 to 6544)
Kentucky 34 600 (105 to 2975) 48 344 (–48 to 5217)
Louisiana 6 67 (11 to 297) 41 883 (–41 to 8709)
Maryland 11 507 (135 to 1821) 16 1118 (–16 to 4452)
Mississippi 15 –3 (–14 to 127) 63 446 (–63 to 3396)
North Carolina 9 96 (–2 to 1676) 50 220 (–45 to 10 910)
Oklahoma 8 159 (41 to 560) 31 –15 (–31 to 4123)
South Carolina 10 254 (73 to 819) 47 1464 (–47 to 10 887)
Tennessee 55 512 (166 to 1397) 47 344 (–47 to 7675)
Texas 204 2131 (375 to 9257) 223 –102 (–223 to 130 708)
Virginia 37 926 (319 to 2560) 35 2435 (–35 to 8418)
West Virginia 34 452 (76 to 2310) 18 298 (–18 to 5681)
West 291 8589 (3160 to 23 086) 264 6823 (–257 to 75 043)
Arizona 22 818 (267 to 2419) 37 1523 (–37 to 28 291)
California 125 4590 (1884 to 10 922) 148 2077 (–148 to 29 211)
Colorado 19 680 (205 to 2160) 12 206 (–12 to 3864)
Idaho 12 120 (37 to 347) 5 128 (–5 to 1077)
Montana 17 193 (64 to 534) 6 366 (–6 to 1167)
New Mexico 7 214 (62 to 696) 28 –15 (–28 to 123)
Nevada 3 150 (65 to 338) 3 208 (–3 to 3348)
Oregon 21 469 (162 to 1288) 11 655 (–11 to 1931)
Utah 13 376 (73 to 1732) 6 580 (0 to 2317)
Washington 48 894 (312 to 2422) 7 1057 (–7 to 3221)
Wyoming 3 84 (30 to 227) 1 38 (–1 to 492)

Measles Vaccination in the United States  •  OFID • 5

State-Level Impact Uncertainty bounds for the number of deaths prevented are
Between 1964 and 2014, Western states prevented 8.6M mea- wide due to variability in prevaccine count values and uncer-
sles cases with vaccination (80% uncertainty interval [UI], tainty in ARIMA counterfactual predictions. After adjusting for
3.2M–23.1M), compared with 7.4M in the Midwest (80% UI, population size, Northeastern states prevented the most deaths
2.0M–24.4M), 7.1M in the South (80% UI, 1.6M–29.7M), (0.36/100  000), followed by the West (0.26/100  000), the South
and 6.8M in the Northeast (80% UI, 2.3M–20.7M) (Table  2; (0.24/100  000), and the Midwest (0.16/100  000). Interestingly,
Supplementary Table 4). California prevented the most cases of a Midwestern state (North Dakota) prevented the most deaths
all states (4.6M), followed by Texas (2.1M) and New York (2.0M). per capita with 0.92/100  000, followed by Montana with
After adjustment for population size, states in the West and 0.86/100 000 (Figure 2). Our models predicted a minor increase
Northeast prevented 329 and 257 cases/100  000, respectively, in measles-related mortality for Arkansas (0.013/100  000),
compared with lower rates in the Midwest and the South, with Delaware (0.002/100  000), New Mexico (0.019/100  000),
236 and 159 cases prevented per 100 000, respectively (Figure 2). Oklahoma (0.009/100  000), Pennsylvania (0.008/100  000), and

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

Wisconsin prevented the most cases per 100  000 (745), fol- Texas (0.011/100 000). For these states, our ARIMA model pre-
lowed by Vermont (670.4/100  000) (Figure 2). Mississippi was dicted close to 0 counterfactual deaths after vaccine introduc-
the only state where our models indicated an increase in cases tion based on already low mortality in the prevaccine period.
(2/100 000), but only during the introduction phase of the vac- Outbreaks continued to occur in the early years of the vaccina-
cination period (Figure  3). This estimated increase was likely tion period, and deaths that occurred in these years exceeded
due to underreporting of cases before vaccination (45/100  000 the very low counterfactual mortality.
vs 344/100 000 nationally). Underreported prevaccination case Between 1964 and 2014, Southern states invested the most
counts led to a low number of counterfactual cases and an in measles vaccination, $4.4 billion, followed by the West
underestimated impact during the vaccine introduction phase. ($2.8 billion), the Midwest ($2.8 billion), and the Northeast
During the 1- and 2-dose phases, vaccination prevented cases ($2.2 billion) (Table  3; Supplementary Table 6). The largest
in all states, including 2 and 8 cases/100  000, respectively, in cost savings due to vaccination occurred in Western states,
Mississippi (Figure 3). with a total of $8.5 billion saved in total societal cost. After
We estimated that vaccination prevented 10  551 measles-re- adjusting for population size, Western states invested the
lated deaths in the South, 9364 in the Northeast, 6823 in the West, most in vaccination with $1.06 per person, followed by
and 4855 in the Northeast (Table  2; Supplementary Table 5). Southern states with $0.99 per person. Utah invested the

A
Cases avoided per state
(per 100k pop.)

800
Cases

400
Avoided

0
WIWVNJ AZCOWY MI CA KYNMTX ID MDTN RI KS IN DC AL FL NE PA GA LA MS

B
B. Deaths avoided per state
(per 100k pop.)

1.0
Deaths

0.5
Avoided
0.0
VTMTMA UTCTWAMEORVA NDOHNY NV IL DE S C IA NHMNOKARMOS DNC

C
(USD per 1k pop.)

C. Costs avoided per state

$10k
Costs

$5k
Avoided
$0k
WIWVNJ AZCOWY MI CA KYNMTX ID MDTN RI KS IN DC AL FL NE PA GA LA MS

Figure 2.  Measles cases, deaths, and costs avoided by vaccination in the United States between 1964 and 2014, by state. A, Ranking of states by the number of avoided
cases/100 000 (WI, VT, WV, MT, NJ, MA, AZ, UT, CO, CT, WY, WA, MI, ME, CA, OR, KY, VA, NM, ND, TX, US, OH, ID, NY, MD, NV, TN, IL, RI, DE, KS, SC, IN, IA, DC, NH, AL,
MN, FL, OK, NE, AR, PA, MO, GA, SD, LA, NC, and MS). B, Avoided deaths/100 000 for states ranked by number of avoided cases/100 000. C, Avoided measles-related.

6 • OFID • Paternina-Caicedo et al
 A Vaccine introduction phase B One-dose vaccination phase
$18k $18k
$15k $15k
$12k $12k

Avoided costs

Avoided costs
per 1000 pop

per 1000 pop

$9k $9k
$6k $6k
$3k $3k
$0k $0k
–2 0 2 4 6 8 –2 0 2 4 6 8
$–3k $–3k
$–6k $–6k
Avoided cases Avoided cases
per 1000 pop per 1000 pop

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

C Two-dose vaccination phase
Midwest
$18k South
$15k Northeast
West
$12k
United States
Avoided costs
per 1000 pop

$9k
$6k
$3k
$0k
–2 0 2 4 6 8
$–3k
$–6k
Avoided cases
per 1000 pop

Figure 3.  Relationship between measles cases and costs avoided by vaccination during different phases of the vaccination period, by region. The right top quadrant of each
plot represents cases prevented and cost savings, the right bottom quadrant represents cases prevented at a cost, and the left bottom quadrant represents no cases prevented
at a cost. A, Cost savings (/1000 people) by the number of cases avoided by measles vaccination in the United States during the vaccine introduction phase (1964–1970). B,
As in (A), but during the 1-dose vaccination phase (1971–1989). C, As in (A), but during the 2-dose vaccination phase (1990–2014).

most in vaccination with $1.42 per person, and West Virginia Sensitivity Analyses
invested the least with $0.74 per person (Figure 2). The largest We conducted a sensitivity analysis to compare estimates for
cost savings occurred in states that prevented the most cases, cases and costs prevented using different counterfactual models
that is, Wisconsin ($8.81 per person) and Vermont ($7.72 per and different imputation methods for missing data (Figure  4;
person). Four states prevented measles cases but not costs Supplementary Figure 8). The lowest number of cases prevented
(Georgia, Louisiana, North Carolina, and South Dakota). (21.7M) and the lowest cost savings ($15.5 billion) resulted
We estimated that these states prevented between 8.1 (North from imputing missing data with 0s and using the mean pre-
Carolina) and 1.7 (Georgia) cases/100  000 at a cost ranging vaccine IR as counterfactual. The highest number of cases pre-
from $0.27 (Arkansas) to $0.96 per person (North Carolina) vented (43.9M) and largest cost savings ($43.9 billion) resulted
(Figure 2). from imputing missing data with linear interpolation and using
We found substantial differences in the health and eco- the mean prevaccine IR as counterfactual.
nomic impact of measles vaccination among states. The
DISCUSSION
difference in impact was associated with income. During
the 2-dose phase, states with a per-capita income above the We found substantial heterogeneity in vaccine impact between
national level prevented 12% more cases (95% confidence states. We found that high-income states prevented more mea-
interval [CI], 11%–12%), 43% more deaths (95% CI, 40%– sles cases and deaths and more measles-related costs compared
43%), and 28% more costs (95% CI, 28%–28%) vs low-in- with states with lower income. States with higher incomes would
come states. A higher impact in high-income states is likely also have higher cost savings, all else being equal, as high-in-
due to stronger vaccination programs: High-income states come households would lose more income when a parent
had 0.5% higher vaccine coverage vs low-income states (95% stayed home with a sick child. We found substantial variation
CI, 0.1%–1.7%). in the average cost of all-cause hospitalization between states,

Measles Vaccination in the United States  •  OFID • 7

Table 3.  Costs From Measles, Investments in Measles Vaccination, and Cost Savings by Vaccine in the United States Between 1964
and 2014, by Region and State

Total Societal Costs From Measles Vaccine Investment (80% Total Societal Cost Savings (80%
(80% Uncertainty Range), USD, Millions Uncertainty Range), USD, Millions Uncertainty Range), USD, Millions

Midwest 327 (326 to 327) 2761 (2209 to 3313) 6363 (–50 to 27 108)
Iowa 31 (31 to 31) 122 (98 to 147) 1299 (–84 to 987)
Illinois 47 (47 to 47) 556 (445 to 668) 8510 (–172 to 4264)
Indiana 31 (31 to 31) 261 (209 to 314) 2230 (–155 to 1611)
Kansas 9 (9 to 9) 120 (96 to 144) 1007 (–88 to 1144)
Michigan 76 (76 to 76) 406 (325 to 487) 16 620 (244 to 5856)
Minnesota 9 (9 to 9) 211 (169 to 253) 1508 (–122 to 1161)
Missouri 8 (8 to 8) 237 (190 to 285) 19 (–183 to 703)
North Dakota 9 (9 to 9) 28 (23 to 34) 820 (–8 to 439)

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

Nebraska 3 (3 to 3) 78 (62 to 94) 195 (–61 to 396)
Ohio 40 (40 to 40) 484 (387 to 581) 9184 (–82 to 4153)
South Dakota 3 (3 to 3) 36 (28 to 43) –192 (–36 to 165)
Wisconsin 61 (61 to 61) 221 (176 to 265) 22 433 (696 to 6229)
Northeast 202 (201 to 202) 2213 (1770 to 2655) 6623 (781 to 25 425)
Connecticut 18 (18 to 18) 138 (110 to 165) 6745 (165 to 1988)
Massachusetts 26 (26 to 26) 259 (207 to 311) 14 099 (418 to 3803)
Maine 6 (6 to 6) 47 (37 to 56) 2019 (13 to 929)
New Hampshire 4 (4 to 4) 47 (37 to 56) 440 (–26 to 309)
New Jersey 28 (28 to 28) 352 (281 to 422) 20 986 (445 to 7062)
New York 80 (80 to 80) 824 (659 to 989) 18 269 (135 to 6342)
Pennsylvania 30 (30 to 30) 483 (387 to 580) 789 (–414 to 3810)
Rhode Island 5 (5 to 5) 41 (33 to 49) 731 (–24 to 592)
Vermont 3 (3 to 3) 22 (18 to 27) 2151 (68 to 590)
South 273 (272 to 273) 4399 (3519 to 5279) 4299 (–2262 to 32 289)
Alabama 14 (14 to 14) 192 (153 to 230) 840 (–127 to 852)
Arkansas 3 (3 to 3) 113 (91 to 136) 163 (–102 to 1167)
District of Columbia 2 (2 to 2) 30 (24 to 36) 278 (–18 to 221)
Delaware 2 (2 to 2) 34 (27 to 41) 460 (–18 to 343)
Florida 20 (20 to 20) 600 (480 to 720) 3051 (–339 to 2385)
Georgia 4 (4 to 4) 376 (301 to 451) –937 (–323 to 989)
Kentucky 18 (18 to 18) 170 (136 to 204) 5053 (–48 to 3361)
Louisiana 4 (4 to 4) 205 (164 to 246) –1281 (–191 to 127)
Maryland 8 (8 to 8) 231 (185 to 277) 4490 (–50 to 2249)
Mississippi 8 (8 to 8) 133 (106 to 160) –1283 (–140 to 24)
North Carolina 5 (5 to 5) 356 (285 to 427) –2376 (–356 to 1991)
Oklahoma 5 (5 to 5) 152 (122 to 182) 261 (–105 to 481)
South Carolina 5 (5 to 5) 172 (137 to 206) 1156 (–86 to 751)
Tennessee 26 (26 to 27) 238 (190 to 286) 3564 (–25 to 1340)
Texas 107 (107 to 108) 1032 (826 to 1239) 16 582 (–470 to 10 514)
Virginia 23 (23 to 23) 296 (237 to 356) 8906 (124 to 2974)
West Virginia 18 (18 to 18) 69 (55 to 83) 4064 (13 to 2521)
West 199 (199 to 200) 2780 (2224 to 3336) 8467 (1448 to 27 254)
Arizona 13 (13 to 13) 229 (183 to 275) 7918 (110 to 2789)
California 97 (97 to 98) 1586 (1269 to 1903) 44 747 (928 to 12 808)
Colorado 12 (12 to 12) 182 (145 to 218) 7293 (97 to 2731)
Idaho 8 (8 to 8) 60 (48 to 72) 1082 (–4 to 418)
Montana 12 (12 to 13) 36 (29 to 43) 2152 (51 to 652)
New Mexico 4 (4 to 4) 81 (65 to 98) 1731 (–5 to 739)
Nevada 3 (3 to 3) 79 (63 to 95) 1299 (13 to 390)
Oregon 12 (12 to 12) 131 (105 to 157) 4996 (95 to 1583)
Utah 8 (8 to 8) 134 (107 to 161) 3060 (–43 to 1885)
Washington 28 (28 to 28) 240 (192 to 288) 9339 (183 to 2932)
Wyoming 2 (2 to 2) 22 (18 to 26) 1056 (24 to 326)

8 • OFID • Paternina-Caicedo et al
 1,000

800

Measles incidence rate

(per 100k pop.)
600

400

200

1931 1940 1950 1960 1970 1980 1990 2000 2010

Year

Observed Linear modeling, imputation with 0s

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

Base-case analysis (ARIMA best model) Mean rates, imputation with 0s
Linear modeling, imputation with interpolation ARIMA best model, imputation with random counts
Mean rates, imputation with interpolation Linear modeling, imputation with random counts
ARIMA best model, imputation with 0s Mean rates, imputation with random counts

Figure 4.  Observed measles incidence rate and counterfactual incidence rates resulting from our default and alternative imputation and counterfactual models. All inci-
dence rates (IRs) resulted from combining state-level rates into a national-level overview. Observed measles IRs (/100 000) and fitted values from the best autoregressive
integrated moving average (ARIMA) models are shown for the prevaccination period (pre-1964). For the vaccination period, observed values and counterfactual IRs are shown
resulting from our default model and various combinations of alternative imputation and counterfactual models. The 80% confidence interval of counterfactual IRs from our
default ARIMA models are shown in shaded blue.

from $3104 per hospitalization in the District of Columbia to some have used an expansion factor to account for underre-
$4891 per hospitalization in Wyoming for the 1964–2014 vac- porting. Limited historical data about underreporting at the
cination period. We did not, however, find a statistically sig- state level have prevented us from using an expansion factor,
nificant association between hospitalization cost and income leading to more conservative estimates by our study.
for states between 1964 and 2014, meaning that differences Our study had several limitations. A  lack of impact was
in hospitalization cost would not explain the heterogeneity of most likely due to underreporting of measles IRs before vac-
vaccine impact between states. Higher income may explain a cine introduction and to uncertainty in the ARIMA model fit.
higher economic impact of vaccination in high-income states For example, the reported prevaccination IR in Mississippi was
but would not explain the higher number of cases and deaths 87% lower (44/100  000) compared with the national average of
prevented. We found that high-income states also had higher 344/100  000. When we substituted prevaccine measles IRs for
vaccination coverage compared with low-income states (in the states with IRs below the fifth percentile of the national distri-
2-dose phase for which state-level vaccination data were avail- bution, with the national average IR, the nationwide impact
able), suggesting a more effective vaccination program in such increased to 30.8 million cases and $26.2 billion prevented. As
states. Other factors could also explain heterogeneity in impact done in previous vaccine impact studies [1, 22, 23], we based
between states, such as population density, level of urbaniza- our impact model on comparing measles incidence rates before
tion, and improved clinical treatment. Indeed, the number of and after vaccine introduction, assuming that all other factors
cases prevented did not follow the same pattern as the num- remained unchanged before and after vaccine introduction.
ber of deaths prevented among states due to differences in the Other factors, such as better health care or reducing birth rates,
prevaccination measles case fatality rate (CFR) among states. have likely contributed to the decline of infectious diseases, but
The prevaccine CFR ranged from an average low of 0.06% in a lack of detailed information about such factors has limited the
Wisconsin to a high of 3.66% in Mississippi (Supplementary possibility of disentangling the impact of vaccination from other
Figure 9). Differences in CFRs can be caused by a variety of factors. Recent studies using mathematical modeling of nation-
factors, including heterogeneity in access to health care between al-level data have started to disentangle the effect of demographic
states and heterogeneity in the decline of measles mortality changes and vaccination on the decline of measles, showing that
before vaccination due to improvements in nutrition, housing, almost half of the decline in measles incidence in high-income
sanitation, and other factors [23, 27]. countries could be explained by the reduction in fertility rates
Previous studies at the national level have estimated that [32, 33]. Future studies should be able to disentangle vaccination
measles vaccination in the United States has saved $8–$11 from demographic and other effects at the state and local levels
billion and has prevented 0.5–3.8 million cases per year as well, when detailed historical information about demographic
[2–4, 28–31]. Many studies have estimated counterfactual cases changes and social determinants of health become available for
based on the average IR during the prevaccination period, and research. Although extrapolations are difficult to make without

Measles Vaccination in the United States  •  OFID • 9

sufficient information, even if half of the decline in measles in 9. Mello MM, Studdert DM, Parmet WE. Shifting vaccination politics–the end of
personal-belief exemptions in California. N Engl J Med 2015; 373:785–7.
the United States could be attributed to demographic changes, 10. PA Department of Health. School immunizations. 2017. Available at: https://www.
the prevention of 15 million cases, 15 thousand deaths, and $13 pabulletin.com/secure/data/vol47/47–9/377.html. Accessed 16 February 2018.
11. Centers for Disease Control and Prevention. Summary of Notifiable Infectious
billion in cost could be attributed to vaccination. Diseases. Atlanta, GA: US Centers for Disease Control and Prevention. Available
The US measles vaccination program was cost saving. Other at: https://www.cdc.gov/mmwr/mmwr_nd/index.html. Accessed 1 September
2016.
medical interventions, such as screening programs, can avoid
12. Centers for Disease Control and Prevention. Publications and Information
disease, but often at a cost. For example, the breast cancer Products from the National Center for Health Statistics. Vital Statistics of the
screening program is estimated to cost a net of $17  050 per United States. Atlanta, GA: US Centers for Disease Control and Prevention.
Available at: https://www.cdc.gov/nchs/products/vsus.htm. Accessed 1 September
life-year saved [34], and combination antiretroviral therapy for 2016.
HIV-infected patients costs $29  000 per quality-adjusted life- 13. US Census Bureau. Statistical abstracts series. Available at: https://www.cen-
sus.gov/library/publications/time-series/statistical_abstracts.html. Accessed 8
year gained [35]. Measles vaccination saved $821 per case pre- November 2016.
vented instead of costing money. 14. Kaiser Family Foundation. Hospital adjusted expenses per inpatient

Downloaded from https://academic.oup.com/ofid/article-abstract/5/7/ofy137/5039595 by guest on 01 October 2018

day. Available at: http://kff.org/other/state-indicator/expenses-per-inpa-
In conclusion, the substantial human impact and cost savings of tient-day/?currentTimeframe=0&sortModel=%7B%22colId%22:%22Loca-
measles vaccination in the United States should motivate parents tion%22,%22sort%22:%22asc%22%7D. Accessed 1 June 2016.
15. Centers for Disease Control and Prevention. Public-sector vaccination efforts in
and policy makers around the world to participate in, sustain, and
response to the resurgence of measles among preschool-aged children—United
expand vaccination programs toward measles elimination. The States, 1989–1991. MMWR Morb Mortal Wkly Rep 1992; 41:522–5.
differences in vaccination impact across states should encourage 16. US Department of Commerce Bureau of Economic Analysis. Bureau of Economic
Analysis. Available at: https://bea.gov/. Accessed 8 November 2016.
all of us to strive for equal vaccination coverage and equal access to 17. Hamborsky J, Andrew Kroger M, Charles Wolfe M. Immunology and Vaccine-
vaccination services throughout the United States and worldwide. Preventable Diseases—Pink Book—2015. Washington, DC: Public Health
Foundation; 2015.
Supplementary Data 18. Centers for Diseases Control and Prevention. Vaccination Coverage | NIS Child.
CDC; 2016.
Supplementary materials are available at Open Forum Infectious Diseases
19. De la Hoz-Restrepo F, Castañeda-Orjuela C, Paternina A, Alvis-Guzman N.
online. Consisting of data provided by the authors to benefit the reader, Systematic review of incremental non-vaccine cost estimates used in cost-effect-
the posted materials are not copyedited and are the sole responsibility of iveness analysis on the introduction of rotavirus and pneumococcal vaccines.
the authors, so questions or comments should be addressed to the corre- Vaccine 2013; 31:C80–7.
sponding author. 20. Jit M, Brisson M, Laprise JF, Choi YH. Comparison of two dose and three dose
human papillomavirus vaccine schedules: cost effectiveness analysis based on
Acknowledgments transmission model. BMJ 2015; 350:g7584.
We would like to thank Anne L.  Cross and Michael Sharbaugh at the 21. Zhou F, Reef S, Massoudi M, et al. An economic analysis of the current universal
2-dose measles-mumps-rubella vaccination program in the United States. J Infect
University of Pittsburgh Graduate School of Public Health for their support
Dis 2004; 189:S131–45.
with data preparation and analysis. We would like to thank Dr. Elizabeth
22. Pezzotti P, Bellino S, Prestinaci F, et al. The impact of immunization programs on
Van Nostrand for her editorial suggestions to improve the manuscript.  10 vaccine preventable diseases in Italy: 1900–2015. Vaccine 2018; 36:1435–43.
Financial support.  This work was supported by research awards 23. van Wijhe M, McDonald SA, de Melker HE, et al. Effect of vaccination programmes
from the Bill and Melinda Gates Foundation (Grant 49276, “Evalation on mortality burden among children and young adults in the Netherlands during
of Candidate Vaccine Technologies Using Computational Models”) and the 20th century: a historical analysis. Lancet Infect Dis 2016; 16:592–8.
from the US National Institute of General Medical Sciences (Grant U54 24. Halloran ME, Struchiner CJ. Study designs for dependent happenings.
GM088491, “Computational Models of Infectious Disease Threats”). The Epidemiology 1991; 2:331–8.
funders had no role in the study design, data collection and analysis, deci- 25. Drummond M. Methods for the Economic Evaluation of Health Care
Programmes. 4th ed. Oxford: Oxford University Press; 2015.
sion to publish, or preparation of the manuscript.
26. US Department of Commerce Bureau of Economic Analysis. Bureau of Labor
Potential conflicts of interest.  We declare no conflict of interest for
Statistics Data. Washington, DC: US Department of Commerce. Available at:
any author. All authors have submitted the ICMJE Form for Disclosure of https://www.bea.gov. Accessed 1 August 2016.
Potential Conflicts of Interest. Conflicts that the editors consider relevant to 27. Merler S, Ajelli M. Deciphering the relative weights of demographic transition
the content of the manuscript have been disclosed. and vaccination in the decrease of measles incidence in Italy. Proc Biol Sci 2014;
281:20132676.
References 28. Whitney CG, Zhou F, Singleton J, Schuchat A; Centers for Disease Control and
1. van Panhuis WG, Grefenstette J, Jung SY, et al. Contagious diseases in the United Prevention. Benefits from immunization during the vaccines for children program
States from 1888 to the present. N Engl J Med 2013; 369:2152–8. era—United States, 1994–2013. MMWR Morb Mortal Wkly Rep 2014; 63:352–5.
2. Zhou F, Reef S, Massoudi M, et al. An economic analysis of the current universal 29. Zhou F, Shefer A, Wenger J, et al. Economic evaluation of the routine childhood
2-dose measles-mumps-rubella vaccination program in the United States. J Infect immunization program in the United States, 2009. Pediatrics 2014; 133:577–85.
Dis 2004; 189(Suppl 1):S131–45. 30. Fontanesi J, De Guire M, Kopald D, Holcomb K. The price of prevention. cost of
3. Bloch AB, Orenstein WA, Stetler HC, et al. Health impact of measles vaccination recommended activities to improve immunizations. Am J Prev Med 2004; 26:41–5.
in the United States. Pediatrics 1985; 76:524–32. 31. Zhou F, Santoli J, Messonnier ML, et  al. Economic evaluation of the 7-vaccine
4. Roush SW, Murphy TV; Vaccine-Preventable Disease Table Working Group. routine childhood immunization schedule in the United States, 2001. Arch
Historical comparisons of morbidity and mortality for vaccine-preventable dis- Pediatr Adolesc Med 2005; 159:1136–44.
eases in the United States. JAMA 2007; 298:2155–63. 32. Merler S, Ajelli M. Deciphering the relative weights of demographic transition
5. Kennedy A, Lavail K, Nowak G, et  al. Confidence about vaccines in the United and vaccination in the decrease of measles incidence in Italy. Proc Biol Sci 2014;
States: understanding parents’ perceptions. Health Aff (Millwood) 2011; 30:1151–9. 281:20132676.
6. Godlee F, Smith J, Marcovitch H. Wakefield’s article linking MMR vaccine and 33. Trentini F, Poletti P, Merler S, Melegaro A. Measles immunity gaps and the pro-
autism was fraudulent. BMJ 2011; 342:c7452. gress towards elimination: a multi-country modelling analysis. Lancet Infect Dis
7. Phadke VK, Bednarczyk RA, Salmon DA, Omer SB. Association between vaccine 2017; 17:1089–97.
refusal and vaccine-preventable diseases in the United States: a review of measles 34. Melnikow J, Tancredi DJ, Yang Z, et al. Program-specific cost-effectiveness ana-
and pertussis. JAMA 2016; 315:1149–58. lysis: breast cancer screening policies for a safety-net program. Value Health 2013;
8. Minnesota Department of Health. Measles (Rubeola). St Paul, MN: Minnesota 16:932–41.
Department of Health. Available at: http://www.health.state.mn.us/divs/idepc/ 35. Neumann PJ, Cohen JT, Weinstein MC. Updating cost-effectiveness–the curious
diseases/measles/stats.html. Accessed 1 June 2017. resilience of the $50 000-per-QALY threshold. N Engl J Med 2014; 371:796–7.

10 • OFID • Paternina-Caicedo et al