Luca Corazzini
(luca.corazzini@unive.it)
MUNI, Brno, May 13th 2021
Behavioral Economics and Health Behaviors: Insights from Cancer Screening Programs

Behavioral Economics and Health Behaviors: Insights from Cancer Screening Programs
•The Good Outcomes of Bad News. A Field Experiment on Formatting Breast Cancer Screening Invitation
Letters
•
•Cervical Cancer Screening Invitations in Low- and Middle- Income Countries: Evidence from Armenia

The Good Outcomes of Bad News. A Field Experiment on Formatting Breast Cancer Screening Invitation
Letters
American Journal of Health Economics, 6(3), 372-409, 2020
Marco BERTONI
(UNIPD)
Luca CORAZZINI
(UNIVE)
Silvana ROBONE
(UNIVA)

Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Motivation (1)
•
•
•Among women, breast cancer is the most common neoplastic disease worldwide and the second most
common cause of cancer mortality in developed countries (IARC 2012)
•
∘1.7 million new cases in 2012, accounting for 25% of all new cancer cases in women. Estimated
number of 521,900 deaths in 2012
∘
∘Breast cancer is associated with an overall cost of about 0.5-0.6% of the total health care
expenditure of developed countries (OECD 2009)
•

•
•A mammography every two years lowers the risk of dying through breast cancer by up to 40%, namely
8 deaths prevented per 1,000 screened women (Lauby-Secretan et al. 2015).
•
•This measure is also highly cost-effective (Cutler 2008, Moore et al. 2009)
•
•As of March 2014, screening programs based on EU indications were active in almost all the EU28
member states.
•
•Unfortunately, take-up rates were still below the EU target rate (75%) in many states (Altobelli
and Lattanzi 2014)
Motivation (2)

This paper
•We run a randomized field experiment in the province of Messina (Sicily) to study how the take-up
rate responds to costless manipulations of the invitation letters
•
•Manipulations along two dimensions: framing (either negative or positive) and enhancing or not
information
•
•In comparison to the baseline, when combining the loss frame with enhanced information on the
negative consequences of not taking the mammography, the take-up rate increases by 25% (2.5 pp). No
effect for the other manipulations.
•The effect is larger among subjects with lower baseline take-ups: those living farther away from
the screening site, residing in municipalities with low education, or with no recent screening
experience
•
•Mechanism: perceived importance and urgency of the screening. Women exposed to the letter
combining the loss frame with enhanced information on the negative consequences of not taking the
mammography are less luckily to reschedule/post-pone the screening visit

Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Related literature – the framework
•
•“Information disclosure” as a form of nudging: providing individuals with more information about
the consequences of their choices (Sunstein 2014) may affect their behavior and promote good
practices.
•
•“Gain-loss framing” (Tversky and Kahneman 1981) in health persuasion. Rothman and Salovey (1997)
distinguish between:
1.Prevention behaviors: non-risky and help in maintaining good health (a gain);
2.Detection behaviors: risky and serve to identify illnesses (a loss)
•
•Gain-framed messages are more effective at promoting prevention behaviors and loss-framed messages
at promoting detection ones
•Nudge (Thaler and Sunstein 2008): any aspect of the choice architecture that alters people’s
behavior in a predictable way without forbidding any options or significantly changing their
economic incentives

Related literature – relevant examples and our contributions
•Meyerowitz and Chaiken (1987): loss-framed messages are more effective than gain-framed messages
in motivating college students to perform breast self-examinations
•
•Banks et al. (1995): loss-framed videos are more effective than gain-framed videos in enhancing
women aged 40+ to self-report mammography utilization
•
•Bourmand et al (2016): providing a 12-page information leaflet does not increase take-up rate for
breast cancer screening.
•
•Goldzahl, Hollard and Jusot (2017): (i) a new logo on the envelope; (ii) patient-approved clarity
in the letter’s content; (iii) a combination of the two previous treatments; (iv) information on
the number of women receiving mammograms in the recipient’s area of residence do not affect take-up
rate for breast cancer screening
•
àOur contributions:
àcombine information & framing - wholly new and effective form of nudging,
àfield experiment built within the actual LHA’s screening program,
àadministrative data on actual take-up rate,
àassess heterogeneous effects.

Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Institutional context (1)
•
•EU recommendations (European Parliament, 2003, 2006):
•
i.women aged 50-69 should take a mammography every two years;
ii.the invitation letter must provide information about the screening program;
iii.qualified radiologists and modern dedicated X-ray equipment/image receptors;
iv.double reading procedure
•
•ITALY: national breast cancer screening program included in the Basic Healthcare Parameters
(Livelli Essenziali di Assistenza) since 2001
•
•LHAs responsible for the implementation of the screening program
•
•Take-up rates are still relatively low and exhibit a strong North-South gap

Institutional context (2)
•The Province of MESSINA:
(i)92,048 women aged 50-69 targeted by the screening program;
(ii)
(ii)8 districts: Messina, Taormina, Milazzo, Lipari, Barcellona Pozzo di Gotto, Patti, Mistretta,
Sant’Agata Militello;
(iii)
(iii)5 health care centers: the Ospedale “San Vincenzo” in Taormina, the Poliambulatorio in
Messina, the Ospedale “Barone Romeo” in Patti, the Presidio Ospedaliero in Sant’Agata Militello
(Sant’Agata Militello and Mistretta), the Presidio Ospedaliero “G. Fogliani” in Milazzo (Milazzo,
Barcellona Pozzo di Gotto and Lipari)
•
•Starting with a pilot study in 2014 and reaching population-level coverage in 2015, the Messina
LHA has implemented the national breast cancer screening program by inviting all women aged 50-69
to take a free mammography every two years
•
•Very low take-up rate, below 15% in both 2015 and 2016. Pressing need to intervene.
•

Institutional context (3)



Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Experimental design: the manipulations

INFORMATION INCLUDED
INFORMATION EXCLUDED


GAIN FRAME
“Enhanced - Gain”
“Scientific studies demonstrate that participating in breast cancer screening programs can have
relevant positive effects on the treatment of an early diagnosed disease: it reduces the mortality
rate, allows for less extensive surgeries, more effective treatments, with higher chances of
recovery.”
“Restricted - Gain”
“Scientific studies demonstrate that participating in breast cancer screening programs can have
relevant positive effects on the treatment of an early diagnosed disease.”


LOSS FRAME
“Enhanced – Loss”
“Scientific studies demonstrate that not participating in breast cancer screening programs can have
relevant negative effects on the treatment of a lately diagnosed disease: it increases the
mortality rate, implies more extensive surgeries, less effective treatments, with lower chances of
recovery.”
“Restricted – Loss”

“Scientific studies demonstrate that not participating in breast cancer screening programs can have
relevant negative effects on the treatment of a lately diagnosed disease.”




•Screening centers provide a yearly stock of slots – enough to satisfy the needs of the target
population – evenly distributed throughout the year
•
•Women randomly allocated to screening slots by LHA computer system
•
•Our manipulations affected women invited for screening during Feb 13 – Mar 19 2017
•
•We sent a different letter to women invited in each week
•
•
•
•
•
•
•
•Letters sent 3 weeks before the mammography by a professional private mail company – no
information available on actual letter delivery
Experimental design: procedures

•Administrative archives of Messina’s Local Health Authority
∘Screening take-up
∘Hospital providing the screening
∘Demographic information
∘Previous screening experience (invitation and take-up)
•
•Administrative archive of the mail company managing the delivery of the invitation letters
∘Date of invitation à treatment status
∘Home address à home-hospital travel time (STATA georoute module)
•Total: 6,194 subjects
•
à  The experiment interests <10% of the target population, each letter received by <2% of it �
spillover effects are unlikely
Data

Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Results - descriptive statistics



Results - descriptive statistics



Results - balancing



Main results



Main results



Robustness tests (1)
•Only one week per treatment: seasonality?
•Redo the analysis on data for same screening weeks in previous years, when all subjects received
the same letter
•
•

Robustness tests (2)
Bertoni Corazzini Robone  – The Good Outcomes of Bad News – 26
•Remaining identification concern are group-specific time effects
∘Example: “Breast cancer awareness month” takes place in October every year. It could have raised
take-up rates independently of the experiment, had the latter taken place in October
•Special festivities or public holidays: not in our experimental weeks
•Concurring campaigns on breast cancer: indirect evidence - Google Trends data
•

Robustness tests (2)



Robustness tests (3)
•Ambiguous effects on take-up
•Increases travel time
•Decreases opp. cost of leisure
•
•Empirically, random rain unrelated to take-up
•
•As expected, effects unchanged by its inclusion as a control
Slight imbalance in E-G week
Yet, 5mm/day cannot cause disarray

Robustness tests (4)
•
•8th March General Strike took place during the “enhanced-gain” manipulation
•
∘Modest participation (<25% interested workforce, mostly concentrated in school sector)
∘No significant heterogeneity in daily take-up rate within the strike week
∘
∘
∘
∘
∘
∘
∘If we assumed take-up rate for the 8th of March was as high as the highest take-up rate during the
week (11.7% for Thu 9), weekly take-up rate would be 11.1%, and difference with baseline would be
1.2 pp, below Minimum Detectable Effect
∘

Robustness tests (5)
•Statistical Inference: a problem of multiple testing?
•
•We are simultaneously testing the effects of four manipulations against a baseline
•
∘If a single test is performed at the 5% level of confidence and the null hypothesis being tested
is true, we expect a 5% chance of incorrectly rejecting it
∘If N=4 independent tests are simultaneously carried out and all corresponding null hypotheses are
true, the probability of at least one incorrect rejection is equal to 18.5%
∘
•List, Shaikh and Xu (2016): resampling-based method for testing multiple null hypotheses
simultaneously in experimental settings with multiple treatments
•“Enhanced-loss” effect still significant with p = 0.082
•Standard Bonferroni-Holm procedure (more conservative) gives p = 0.096

Heterogeneous effects




(1)
(2)
LPM
LPM



Enhanced - Loss
-0.323***
-0.319***
(0.068)
(0.068)



Observations
6,194
6,194



Covariates
No
Yes
Postponement conditional on participation
Mechanisms: perceived importance and urgency of the screening

Scheme of the presentation
•Motivation
•
•Related literature
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Conclusion
•Loss frame and enhanced information increase the take-up rate by 25% (2.5pp) relative to the
baseline. Other treatments are instead ineffective.
•
•The effect is stronger for subjects living farther away from the health care centers: letter
formatting can overcome the cost to take the mammography
•
•“Enhanced – Loss” letter increases the perceived importance and urgency of participating in the
screening program (psychological “unpacking” effect, see Van Boven and Epley 2003 and Angelini et
al. 2017)
•
•Back of the envelope calculations. Screening saves 8 out of 1,000 screened women (Lauby-Secretan
et al. 2015). Extending our results on the overall target population in Messina (90,000 women): 18
lives saved AT ZERO COST
•
•Concerns about over-diagnosis: “breast cancers that would never have been diagnosed or never
caused harm if women had not been screened”… BUT over-diagnosis in Italy is low - between 1 and
4.6% (Puliti et al. 2012)
•

Implications for policy makers
•
•
•European guidelines for quality assurance in breast cancer screening and diagnosis: invitations to
the screening program should be positively framed (e.g. 9 out of 10 recalled women are found to be
normal rather than 1 out of 10 recalled women will have cancer)
•
•Our experimental findings do not lend empirical support to this advice, and would call for an
update of the guidelines to ensure that the highest possible take-up rate is achieved, at least for
areas comparable to Messina
•
•Future research: re-do the experiment in areas with higher baseline take-up rate (but very hard to
convince LHAs to participate)
•

Behavioral Economics and Health Behaviors: Insights from Cancer Screening Programs
•The Good Outcomes of Bad News. A Field Experiment on Formatting Breast Cancer Screening Invitation
Letters
•
•Cervical Cancer Screening Invitations in Low- and Middle- Income Countries: Evidence from Armenia

Cervical Cancer Screening Invitations in Low- and Middle- Income Countries: Evidence from Armenia
Social Science and Medicine, 273, 113739, 2021
Marco BERTONI
(UNIPD)
Luca CORAZZINI
(UNIVE)
Armenak ANTINYAN
(Zhongnan U)

Scheme of the presentation
•Motivation
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

•Collaboration between Armenia National SDG Innovation Lab (joint initiative of UN and the
Government of the Republic of Armenia, supported by the UNDP) and the Ministry of Health of the
Republic of Armenia
•
•The study is aimed at increasing the uptake of a cervical cancer screening program that runs in
the Republic of Armenia
•
•To the best of our knowledge, the first RCT in the developing world that studies the impact of
invitation letters and reminders on (cervical) cancer screening uptake
∘
•
∘
∘
∘
The project in a nutshell

•CC is the fourth most frequent cancer among women in the world, with roughly 570,000 new cases in
2018 (9.3% of all female cancers) (GLOBOCAN, 2018)
•
•Yearly, around 90% of deaths occur in low- and middle-income countries (LMIC):
•
∘Absence of organized screening programs or low participation if a program is present  (e.g.,
Gakidou et al., 2008; O’Donovan et al., 2019; Sankaranarayanan, 2001)
∘
∘In the last 40 years Sharp decline of cervical cancer incidence in high-income countries due to
organized screening programs
•
•
•
Cervical cancer (CC)




•Luckily, CC is one of the most preventable among the relevant human cancers.
∘Mono-causal genesis: infection of the uterine cervix with human papillomavirus (HPV) needs to
persist for many years to generate cancer.
∘
•
•Main prevention devices (European Commission, 2015):
∘Population-based HPV vaccination of girls aged 12+
∘Population-based Pap-test screening of women aged 25-64, every 3 years
∘This has been recently substituted by the introduction of HPV testing every 5 years for women
above 30
CC screening

•Despite large benefits, lack of infrastructures and scarce health care resources limit the
possibility to implement adequate screening activities in LMICs (Lazcano-Ponce et al, 1999; Rao
2012)
•
•When in place, low participation in these programs (WHO, 2002) due to:
∘Information gaps, cultural and socio-economic barriers
•
•In HICs, invitation letters and reminders stimulate participation in CC screening programs (Decker
et al., 2013; Eaker et al., 2011; Radde et al., 2016; Tavasoli et al., 2016)
•
•Some evidence that framing of letters also matters (Bertoni et al., 2020)
•
•Lack of research on how these results extend to LMICs
•
•
•
CC screening programs in LMICs

•Absence of insurance and low income
∘patients frightened to detect any illness as they would it impossible to get treated if cancer is
detected - which in turn may deter attendance
•
•Traditional cultural values (even about medical exams)
•
•Distrust toward the medical system (corruption and low quality)
•
•
•Response to screening programs and various invitation strategies can be different between HIC and
LMIC
Why invitation letters and reminders may not work in LMICs?

•We worked with the Health Ministry of Armenia and Armenia SDG Innovation Lab to evaluate the
effects of invitation letters and reminders aimed at enhancing screening participation
What do we do?

Scheme of the presentation
•Motivation
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

•Post-communist, middle income country in transition
•
•Population: about 3,000,000
•
•GDP per capita: 4,000 USD (2017)
•
•Poverty: 25.7% (2017)
•
•CC Incidence:
•Armenia= 8.4, Europe=11.2
•
•CC Mortality:
•Armenia= 5.6, Europe=3.8
•
•
Image result for map of armenia
Armenia

https://www.cia.gov/library/publications/the-world-factbook/geos/am.html)
https://www.adb.org/countries/armenia/poverty

•Screening
∘
∘Up to 2014: opportunistic screening not through PAP testing (pay out of pocket if you want to do
it)
∘
∘Since 2015: "Disease Prevention and Control Project in Armenia" project funded by the World Bank
(2015-2020)
∘
∘One free screening slot for each woman aged 30-60 every three years
∘
∘No invitation system. Mostly advertised using classical advocacy tools as TV and radio programs,
leaflets in supermarkets
∘
∘As of Feb 19, participation was not satisfactory for the Government
∘
CC screening and treatment in Armenia

-44% of the population below the poverty line (the highest poverty rate in Armenia)
-Population 251,941 (2011 Census)
-
-Urban: 146,908 (58.3%)
-
-Rural: 105,033 (41.7%)
The region of interest

At this point we do not have the medical map in English, but we will try to produce it for the
paper.
The big green area in the middle of the photo is the capital city of the region (Gyumri), in which
6 medical institutions are concentrated. The green circles are the medical institutions, the black
circles are the communities/villages/cities served by these medical institutions. The other medical
institutions are mainly concentrated in major cities of the region. I THINK THIS MUCH IS ENOUGH,
but if you need more info about medical institutions let me know.

Scheme of the presentation
•Motivation
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

•We manipulate
•
∘Presence of a letter
∘
∘Presence of a reminder on top of the letter (Altmann & Traxler, 2017, Calzolari & Nardotto, 2016)
∘
∘The frame of the invitation letters and reminders (Positive framing; Negative Framing; Concerned
for others framing) (Rothman and Salovey, 1997; Bertoni et al 2019; Du, Li, Lu & Lu, 2019)
∘
∘
•8 treatments (different invitations) + 1 control (no invitation)
•
•
•
Experimental design

•Neutral (slightly positive invitation):
∘
∘Please note that scientific studies demonstrate that participating in cervical cancer screening
programs can have relevant positive effects on the treatment of an early diagnosed disease.
•
•Negative Framing:
∘
∘Please note that scientific studies demonstrate that not participating in cervical cancer
screening programs can have relevant negative effects on the treatment of a lately diagnosed
disease: it increases the mortality rate, implies more extensive surgeries, less effective
treatments, with lower chances of recovery.
•Concern for Others:
∘
∘Your family members, relatives and friends expect you to live a long and healthy life with them.
Detecting and curing a potential cancer at early stages can help you fulfil their expectations. Go
to the screening for your loved ones!
The letter frames

•Shirak target population: about 36,000 eligible women aged 30-60 who have not attended the program
as of Feb 2019 (or attended in 2015/2016)
•
•
•
•
•
•20,800 people receiving letters
∘Letters received 3 weeks before the scheduled week
∘Reminders received 1 week before the scheduled week
•
•Those individuals who did not receive letters are kept as the «control group»
•
•
•
•
•
•
•
Assignment to treatments

•We opted for individual-level randomization, stratified by GP
•Select a share of patients per GP in letter sample equal to share of patient per GP in the
population
•
•Each letter type was equally represented within GP
•
•Day of letter delivery also independently and individually randomized
Randomization

The RCT was implemented in Shirak province between May-July, 2019
Sample allocation and treatments


Implementation



Data
•Internal records of the hospitals
•
•Background data (date of birth, place of residence, GP id)
•
•Take-up:
∘For the time being: we measure take-up until 19 July 2019
∘
•

Scheme of the presentation
•Motivation
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Prima facie evidence: regional data



Screening Participation by Treatment Group



Econometric specification

Considering that receiving letter and reminder depends on observable and unobservable individual
characteristics (area of residence, say, or co-residence with others, time spent at home, and so
on), we elect not to estimate Equation (3) by OLS. Instead, we use Two-Stage Least Squares, and
take the two dummies Lij and Rij for the randomly assigned type of invitation sent to each subject
as instrumental variables. The corresponding first-stage equations model letter and reminder
reception as functions of letter and reminder dispatch, and are specified as follows:
LetterReceivedij =αj + ρ10Lij + ρ11Lij Χ Rij + v1ij (4) LetterReceivedij ΧReminderReceivedij = αj +
ρ20Lij + ρ21Lij Χ Rij + v2ij (5) TOT effects in Equation (3) can be obtained as the ratios of the
ITT effects in Equation (2) to the first-stage effects in Equations (4) and (5), i. e. φ0 = δ0/ρ10
and φ1 = δ1/ρ12.




Scheme of the presentation
•Motivation
•
•Institutional context
•
•Experimental design
•
•Results
•
•Conclusion

Conclusions
•
•To the best of our knowledge the first RCT that studies the impact of invitation letters and
reminders on (cervical) cancer screening participation in the developing world
•
•We find huge impact of invitations letters on cancer screening participation in LMICs
•
•An invitation letter is particularly effective if followed by a reminder
•
•Framing of the letters does not seem to matter
•
•
•

Thanks a lot for your attention!
(luca.corazzini@unive.it)