1

An Adaptive Agent-Based Model for Observing

Smoking Cessation Patterns
Margarita Sordo,1,2 Andrew B. Phillips3,4

1 Brigham and Women’s Hospital Department of General Medicine, Boston, USA,

2 Harvard Medical School, Boston, MA, USA,
3 MGH Institute of Health Professions, School of Nursing, Boston, USA,
4 Spaulding Rehabilitation Network, Boston, MA, USA,
Corresponding author: M Sordo,
ORCID: M Sordo: 0000-0002-1494-8455; AB Phillips: 0000-0002-6322-471x

Abstract

Background: The complex effect of multiple influencing factors in the idiosyncrasies of how

people frame a purpose, and the decisions and actions they carry out to reach a goal have a

strong impact on repeated-occurrence behaviors, such as smoking cessation, where the perceived

benefit of behavioral change may wane in response to difficulties or setbacks. Purpose: We

applied a modeling cycle methodology to implement a knowledge-based ABM to assess the

impact of individual health preferences, and societal factors on modifiable behaviors to identify

time-related opportunities for non-pharmaceutical interventions to sustain the desired behavioral

changes. Methods: We gathered and encoded information about patient beliefs, preferences, and

societal factors as simple rules to roughly represent patient behaviors. Through ABM simulations

we looked at idiosyncratic patterns stemming from the complex effect of multiple influencing

factors. Results: Marked smoking/non-smoking fluctuations of women vs. slower, steadier

decline in smoking men highlight the complex effect of multiple influencing factors in the

idiosyncrasies of how people frame a purpose, and the decisions and actions they carry out to

reach a goal. Unintentional patterns of segregation underline the impact surrounding neighbors’

have on an agent’s behavior, blurring the line between individual motivation and collective

influence, leading agents to align with the surrounding majority. Conclusions: ABMs provide

insights on the impact multi-factorial, dynamic individual behaviors, and societal factors have on
 2

repeated-occurrence modifiable behaviors, e.g., smoking cessation, and pinpoint opportune

educational and motivational adjuvant interventions to counteract negative, more ingrained

behaviors, and external factors to improve compliance and success.

Keywords: Agent-based modeling, Individual health preferences, Repeated-occurrence

behaviors, Smoking cessation, Societal factors, Determinants of health

Statements and Declarations

Compliance with Ethical Standards

 Funding: The authors did not receive support from any organization for the submitted
work.
 Financial Interests: The authors have no relevant financial or non-financial interests to
disclose.
 Conflict of Interest: The authors have no conflict interest.
 Human Rights: This article does not contain any studies with human participants
performed by any of the authors.
 Informed Consent: This study does not involve human participants and informed
consent was therefore not required.
 Welfare of Animals: This article does not contain any studies with animals performed by
any of the authors.
 Data Transparency: Data for this study was randomly generated based on knowledge
gathered from literature. The defined variables and values are described in the
manuscript.
 Software availability: Analytic code used to run Agent-Based Model simulations
presented in this study are not available in a public archive. They may be available by
emailing the corresponding author.
 Materials: All materials used to conduct the study are described in the references section
of the manuscript and are publicly available.
Authors contributions
 M.Sordo and A.B.Phillips conceived the study and designed the model; M.Sordo
implemented software; M.Sordo, A.B. Phillips contributed to the implementation of the
conceptual framework and interpretation of results; M.Sordo wrote the manuscript;
M.Sordo, A.B. Phillips commented and approved the manuscript.
 3

1 Introduction

Simulation identifies and models behavior patterns in terms of factors, decisions and

actions that serve a purpose and fulfill a goal. Simulation requires understanding the purpose that

leads to a behavior. Understanding is the cornerstone for any simplifications, abstractions,

assumptions, and generalizations required during the modeling process to produce an accurate,

yet general representation of a behavior [1]. Individual preferences and external factors are

closely linked to decisions and actions. The complex effect of multiple influencing factors is

reflected in the idiosyncrasies of how people frame a purpose, and the decisions and actions they

carry out to fulfill it and reach a goal. Whether consciously or not, people are influenced by

multiple factors stemming from different dimensions of their lives.

These factors, called determinants of health (DOH) are generally classified into 5

domains: Individual, Social, Environmental, Genetics and Biology, and Medical Care; and to a

large extent, have a considerable impact on health [2–4]. Smoking falls into the individual

behavior category of DOH. It is well-documented that smoking has a strong negative impact on

health [5] and behaviors to thwart it are strongly encouraged by healthcare policies and

interventions [6]. Smoking is strongly influenced by individual preferences and societal factors,

and therefore impacted by decisions and actions believed to be aligned with a personal purpose

and goals.Agent-Based Models (ABMs) are composed of independent, autonomous agents with

their own unique characteristics and evolving behaviors within social and physical environments.

Individual behavior is driven by each agent’s own characteristics, preferences, beliefs, actions

and decisions, and its ability to adapt/react to stimuli. ABMs are a non-linear alternative to

modeling, evaluating, and understanding the complexity of individual health preferences.

Simulation of these characteristics, preferences, and behaviors over time produces emergent

patterns of behavior at both individual and system level.

 4

Our current research focuses on a methodology to implement ABMs to assess the impact

of individual health preferences on behavior patterns over a time continuum - with smoking

cessation being a specific example. Given the dynamics of health behaviors, it is important to

emphasize the need to observe the variability of such behaviors over a time continuum, and not

only at predefined time points that may miss meaningful variations [7]. Therefore, change over

time must be seen as a process with dynamic changes reflecting gains and loses [8].

2 Background

2.1 Agent Models

ABMs describe a system in terms of individuals, and their behavior. Individuals within

ABMs are unique, autonomous entities that locally interact with each other and their

environment. By being unique, individuals differ from each other in terms of properties or

attributes. While we may observe trends in both attributes and behaviors, their distinctive

characteristics and behaviors provide a rich, varied representation that allows for both general

trends and individual, more specific features in a model. This approach resembles the real world

in that not everybody is equal, and such variations should be considered when modeling. By

being autonomous, individuals behave independently of each other pursuing their own

objectives. Individuals do not interact with everybody in the system, but rather with their

neighbors within e.g., a geographic space or a network. Local interaction, however, does not

preclude the disseminating impact such local actions and behaviors may have over the whole

system. Consequences of local interactions may initially manifest as local foci eventually

extending, if circumstances are ripe, to the whole system.

Individuals strive to survive, reproduce, reach goals, etc. based on their individual

characteristics, beliefs, preferences, and actions. Through adaptation, individuals adjust their

behavior to their current state of themselves, the environment, and other individuals to improve
 5

the chances of reaching such goals. Therefore, as each individual adapts, so does everybody else

in the system, resulting in emerging system dynamics arising from how individuals interact with,

and respond to each other and their environment. It is this ‘local impacts whole system’ multi-

level interactions that makes ABMs suitable to answering questions about how a system’s

behavior emerges from, and is tightly linked to its individual components [9].

The ability of ABMs to model complex, emerging dynamics requires strategies for

designing and implementing models that reflect behaviors and processes we believe to be

important. These strategies should allow for relative ease of parametrization, validation, and

analysis of all those elements believed to be representative of the model at hand. Determining

what and how to include and exclude from the model requires iterative formulation and analysis

of such parameters. Given that a model starts from often simplified observational and

experimental assumptions upon which we build, we need to iterate through the modeling,

validation, and analysis cycle to assert the validity of those assumptions. We further discuss this

approach in the Methods Section.

2.2 Individual Preferences

In traditional models of care, both patients and clinicians presume clinician preeminence

in decision-making, and patients tend to defer to the judgment of the clinician. Even when

patients initially comply to recommendations, the burden of implementation, and adherence to

health-related interventions often falls upon them, with little support and guidance outside the

clinical setting. Agreement and compliance are strongly influenced by a patient’s individual

preferences.

In a healthcare context, individual preferences refer to the desirability of behaviors,

treatment options, or health states [10]. Patient preferences have a direct impact on

implementation and compliance of recommendations or interventions intended to improve their

 6

health. As health practitioners explain the benefits of a recommendation, patient preferences and

beliefs influence their perception of the positive value such recommendation may have in their

life. This perception will impact both their decision to accept the proposed intervention and

adhere throughout its duration. Adherence is particularly challenging in repeated-occurrence

behaviors [11] as the perceived beneficial value may fluctuate over the duration of the

intervention, as this perception is modified by behaviors, external influences and observed

benefits.

Understanding the role of individual preferences is vital as patients are more empowered

to actively participate in managing their health and take ownership of their health-related

decisions.

They are essential to the success of recommended interventions to improve non-acute

health states that require changes in behaviors and routines. It becomes imperative that patients

understand the benefits of an intervention to fully invest their efforts. It is equally important that

they receive the necessary help and support for the duration of the intervention to counteract

negative, more ingrained behaviors, and external factors, e.g., social pressure that could wane

their compliance and derail a successful outcome.

Individual determinants of health refer to habits and activities ingrained in individual

behaviors and circumstances. These behaviors and circumstances, strongly influenced by

individual preferences, directly impact our physical and mental health, and overall well-being.

Positive changes to individual behavior can reduce the risk of developing a disease, and many

health interventions are aimed at modifying individual behavior.

Smoking is an individual determinant of health and the leading preventable cause of

death in the United States [5]. Smoking is an example of a modifiable individual behavior that if

left unaltered can lead to harmful outcomes. Smoking behaviors depend on many individual and
 7

societal factors, e.g., perceived benefits/harms, stress coping, self-image, peer-pressure. Smoking

cessation requires concerted, multi-level interventions addressing these individual and external,

societal factors. Initial smoking cessation recommendations would be more effective if they

address individual preferences, behaviors, and societal factors, and provide motivational and

educational adjuvant interventions at appropriate times.

3 Methods

Our modeling approach, based on the modeling cycle described by Railsback [9] and

depicted in Fig. 1, is an iterative cycle involving hypothesis generation followed by experimental

testing and validation.

Fig. 1 Modeling cycle as a cyclic sequence of activities: a) formulating question(s); b)

assembling hypotheses for key attributes and behaviors; c) choosing an ppropriate variables,

processes, parameters; d) programming the mode; e) analyzing, testing, revising observed

behavior(s); f) communicating findings (From [9])

The modeling process starts with the problem formulation (Fig. 1a), i.e., a question that

guides us through the modelling process, allowing us to filter out spurious information. In the
 8

context of individual preferences and smoking cessation, the question is:

What individual and environmental factors influence a person’s resolve to stop smoking?

To answer this question, we identified relevant factors that impact and define behaviors,

as well as potential interdependencies among these factors. Known patterns are incorporated at

this stage (Fig. 1b). Approaching this hypothesis generation step in a simple manner allows for a

gradual, sounder understanding of the task at hand. Having a conceptually simple model

facilitates validation of the initial question and hypothesis. It might be necessary to cycle back

and review the initial question and hypothesis but having a conceptually simple model eases the

iterative refining process. We performed a literature review and identified several potential

factors influencing smoking cessation efforts [7–12]. Influencing factors stratified by gender

include desire to stop smoking; perception of health benefits resulting from smoking cessation;

susceptibility to (societal) peer pressure by neighbor agents; likelihood of relapse. They are

represented by three simple rules in our model [12,13]:

i. Social pressure limited to same gender

ii. Women are more vulnerable to social pressure to quit

iii. Women more likely to relapse

Once satisfied with our assumptions, we defined the structure of the model in terms of

states, entities, state variables, parameters and processes required to implement and simulate the

agent’s behavior and their environment (Fig. 1c). Our initial population is determined by size of

the environment, and consists of over 22,000 agents, all smokers. Agents are heterogeneous,

each with static (e.g., gender) and dynamic characteristics and preferences (e.g., satisfaction,

smoking status), all located within an environment (a discrete, homogeneous grid). Initial

individual characteristics and preferences are setup as follows:

a) gender: randomly assigned in a 50/50 male/female proportion (it remains unchanged).

 9

b) attractiveness of health: a global variable initially set the same to all agents. During

simulations, it adapts to the influence of neighbor non-smokers to reflect positive

social pressure. Values ranged from 0.1 to 0.9 in 0.1 increments.

c) satisfaction: it represents the perception of health benefits resulting from smoking

cessation. It is a global variable initially set to 1 for all agents. During simulations, it

adapts to reflect the agent’s attitude toward their individual perception of health, and

the influence of healthier (non-smoker) neighbors.

d) threshold: it represents the desire to stop smoking, and in combination with

satisfaction, it is used to trigger a change in smoking status. If the social pressure from

surrounding neighbor agents exceeds their satisfaction, the agent will switch their

status. Initial values ranged from 0.1 to 0.9 in 0.1 increments.

e) smoking status: one of smoking, non-smoking. Initially all agents are smokers.

In the implementation phase (Fig. 1d), we translated the description of the model into

agents with properties, attributes and behaviors, environment variables, and model dynamics

determined by logic functions and methods.

The dynamics of the smoking cessation model reflect knowledge gathered from literature

review and encoded in the three rules presented above. Gender is a strong component in all the

rules, with social pressure limited to same-gender neighbors (rule i). Susceptibility to peer

pressure by surrounding neighbor agents (rule ii) and likelihood of relapse (rule iii) are

determined by a weighted function of same-gender, non- and smoking neighbors, with women

having a stronger weight assigned to reflect their higher susceptibility to peer pressure.

With the model definition completed, we were ready to execute, and analyze both the

model, and our initial assumptions (Fig. 1e). We modeled this analysis activity in three phases

(see Fig. 2): a) in the initialization phase, we defined the agent’s preferences and attributes, e.g.,
 10

gender, smoking status, perception of value of healthy behavior, satisfaction with current

smoking status; b) in simulation phase, each agent interacted with surrounding neighbor agents

and; c) internal attribute values were recalculated in response to internal, adjusted beliefs and

preferences, and peer pressure. During simulation agents’ behavior changed and adapted to new

internal and external setup, and d) emerging patterns became apparent.

Fig. 2 Schematic view of modeling process. a) Initial setup of agent preferences, attributes, and

environment; b) Start simulation. Agents gather information of other agents and the

environment; c) Dynamic interactions change internal beliefs and preferences while agents

adapt; d) System reaches equilibrium as agents adapt, and emerging patterns of behavior arise

Previously unknown response-relapse patterns emerged during simulations shedding light

into the dynamics of the system as a whole, and the idiosyncratic behaviors that arose in

response to agents’ internal, other agents, and environment influences (Fig. 1f). Emergence

resulted from a multi-level approach where agents dynamically adapted to cope with changes

within themselves; and as they changed, their influence on other agents also changed, requiring

them to adjust. As agents exist within an environment, they are exposed to environment and

societal (from neighbor agents) factors to which they need to adapt. In the following section we

present results of our smoking cessation case study and discuss findings.
 11

4 Results and Discussion

4.1 Results

An initial population of about 22,000 smokers was split into 50-50 male/female. 810

Simulations using NetLogo [14] were run until the model reached equilibrium at about 2,000

iterations. For each combination of initial parameters (shown in Error! Reference source not

found.) 10 runs were performed, and results were averaged.

Table 1: Initial Parameter Value Settings for ABM Simulations

Parameter* Initial Value Setting

Attractiveness [0.1 to 0.9] in 0.1 increments
Satisfaction 1
Threshold [0.1 to 0.9] in 0.1 increments as seed of a normal distribution
Smoking status Smoker
Gender 50/50 randomly as one-of [male, female]
*: All initial parameter values changed during simulations except gender.

In Fig. 3, we present results from 810 stochastic simulations reflecting the dynamics of

agents interacting and adapting their behaviors, beliefs and preferences in response to exerting

influence from surrounding neighbor agents. For each simulation, 10 runs were performed, and

results averaged for each combination of Attractiveness and Threshold values.

Dynamics observed in Fig. 3 show final smoker/non-smoker ratios between genders are

similar with overall trends showing a percentage of the population of about half successfully

stopped smoking. If we think in terms of only beginning-end points, though important, this

finding may be considered unremarkable, in that it does not show significant differences between

genders. However, the purpose of the current research is to turn our attention to the

smoking/non-smoking fluctuations of women during the simulations and the slower, if more

steady decline in smoking men, and appreciate the cyclical pattern of marked fluctuations
 12

observed in women. These fluctuations provide opportunities for timely interventions to sustain

behavior change and reduce the likelihood of relapse.

Fig. 3 Smoking cessation patterns over time. 810 stochastic simulations reflecting the dynamics

of agents interacting and adapting their behaviors, beliefs, and preferences in response to

exerting influence from neighbor agents. 10 runs were performed, and results averaged for each

combination of Attractiveness and Threshold values. Overall trends show similar smoking

cessation rates. However, gender-based behavioral differences are evident in the cyclical pattern

of marked smoking/non-smoking fluctuations observed in women (light blue), and the slower,

steadier decline in smoking men (red)

These gender-based behaviors, though attributable to the knowledge gathered from

literature and encoded as simple rules, show their impact on the dynamics of the system, namely

women are more likely to quit (rule ii), and more likely to relapse (rule iii) given that they are

more susceptible to social pressure (rule i). These rules highlight the complex effect of multiple
 13

influencing factors in the idiosyncrasies of how people frame a purpose, and the decisions and

actions they carry out to fulfill it and reach a goal. This is particularly true in repeated-

occurrence behaviors where the perceived benefit may wane in response to difficulties or

setbacks.

Fig. 4 shows the distribution of agents within the environment after a simulation. It is

possible to observe scattered clusters of smokers and non-smoker agents highlighting

unintentional patterns of segregation. It is apparent how surrounding neighbors’ behaviors have

an impact on an agent’s behavior, blurring the line between individual motivation and collective

influence, leading agents to align with the surrounding majority. This aggregate segregation

reflects the emerging boundaries that create such clusters of local majorities of smokers or non-

smokers. Even though the results are aggregate, the decisions and behaviors leading to these

patterns are individual. A smoker who ceases to smoke and a non-smoker who relapses are

reacting individually to their individual beliefs and motivations, but also to an environment

consisting of other agents who behave in a similar manner. It is also possible to observe that

clusters of smokers and non-smokers mix and co-exist, but there is no single smoker or non-

smoker completely surrounded by neighbors with the opposite smoking status, reflecting a tacit

need for acceptance and sense of belonging, reinforcing the unintended segregation.
 14

Fig. 4 Distribution of agents within the environment after a simulation. It is possible to observe

clusters of agents with same smoking status. Smokers tend to gather with other smokers, and

similar behavior applies to non-smokers regardless of gender

4.2 Discussion

Traditional models of care tend to assign the role of decision-maker to the clinician

providing care, and the patient as the compliant recipient of treatments and recommendations.

For these recommendations and treatments to have the intended impact, it is important to include

patient preferences and beliefs in the decision-making process – as this would potentially

increase adherence and positive outcomes.

We modeled the impact of preferences and beliefs as factors influencing agents’ behavior

towards a desired health outcome: smoking cessation. Straightforward rules and procedures

allowed us to represent these preferences (satisfaction level), beliefs (health attractiveness),

behaviors (smoking cessation) and exerting influences (social pressure) in a simple manner to

identify and understand complex, emerging patterns of behavior[15,16]. Through stochastic

 15

simulations we confirmed the importance of preferences and beliefs as driving factors of

behavioral change. Gender-based idiosyncrasies are observable in the marked fluctuating

smoking/non-smoking dynamics of women during the simulations and the slower, steadier

decline in men. This difference highlights the importance of identifying repeated-occurrence

behaviors that require a sustained effort to overcome potential hurdles to reach the desired goal.

These fluctuations provide opportunities for timely non-pharmaceutical interventions to support

smoking cessation efforts to mitigate the chances of a relapse and increase the long-term success

of the desired outcome. For example, women may require more frequent follow-up interventions

and support to offset their susceptibility to social pressure and increased likelihood of relapse.

Our approach focuses on change over a time continuum as a process with dynamic changes

reflecting gains and loses [8], and not only at predefined time points that may miss the observed

meaningful variations.

Our results also show unintentional patterns of segregation arising from neighbors’

behaviors and interactions that to certain extent blur the line between individual motivation and

collective influence leading agents to align with the surrounding majority. This aggregate

segregation reflects the emerging boundaries that create such clusters of local majorities of

smokers or non-smokers. Even though the results are aggregate, the decisions and behaviors

leading to these patterns are individual. These clusters of local majorities of smokers or non-

smokers highlight the impact social pressure may exert on individuals that do not conform with

the surroundings, and their tacit need for acceptance and sense of belonging, reinforcing the

unintended segregation.

Results confirm how decisions and actions are strongly influenced by individual

preferences, beliefs, and societal factors. Understanding the role of individual preferences is vital

as patients become more empowered and actively participate in their health-related decisions.
 16

Smoking cessation recommendations that address individual preferences and behaviors, and

societal factors are more likely to succeed. It is equally important that patients receive the

necessary help and support for the duration of the intervention to counteract negative, more

ingrained behaviors, and external factors that could wane their compliance and derail a

successful outcome.

ABMs provide the means of formulating questions, assembling hypotheses, and building

models based on simple rules to discover emerging patterns of dynamic idiosyncratic behavior in

complex systems. Stochastic simulations provide diversity on individual attributes while

preserving general trends. This diversity results in a population of unique individuals, each with

their own specific attributes, preferences, behaviors, and nuances, therefore reflecting to some

extent those of individuals in the real world.

Results presented herein provide valuable insights for the timing of potential post-

recommendation interventions to improve adherence and efficacy of the recommendation.

Although exploratory, these results are suggestive of patterns of idiosyncratic behavior that could

benefit from non-pharmaceutical targeted interventions aimed at supporting patients’

perseverance at modifying repeated-occurrence behaviors (e.g., smoking) leading to beneficial

health outcomes.

5 Conclusions and Future Work

We have explored the impact that DOH, e.g., individual preferences and beliefs, and

societal pressure have on modifiable repeated-occurrence behaviors such as smoking. We found

that preferences, beliefs and societal factors impact such behaviors and may support or hinder

behavioral changes aimed at improving health. Smoking, as well as other determinants of health,

is a modifiable behavior that if left unaltered can lead to harmful outcomes. Hence, it is desirable

that smoking cessation recommendations not only consider individual preferences and behaviors,
 17

and societal factors, but also provide opportune educational and motivational interventions to

improve adherence and success.

Our study has potential limitations inherent to the nature of agent-based models. Quite

often, models can be considered either too trivial or too complex. Therefore, it is important to

find a level of complexity that reflects trends of behavior and not every single observed nuance.

At times, results from ABMs may seem obvious in that they confirm initial assumptions – as it is

the case of the three simple initial rules embedded in the model. However, during simulations, as

models evolve, adaptive patterns may surface providing valuable insights on the aggregate

dynamics of individuals as parts of the system – as could be observed in the women’s fluctuating

cessation/relapse patterns, as well as in the unintended patterns of segregation that create clusters

of local majorities of smokers or non-smokers. These clusters of unintended segregation

highlight the impact social pressure may exert on individuals that do not conform to their

surroundings, and the need of individuals for acceptance and a sense of belonging.

Despite these apparent limitations, ABM has the potential to facilitate and improve

research on the impact of emerging behavioral patterns as it is the case with determinants of

health, by surfacing idiosyncratic behaviors resulting from the interplay of multi-factorial

influences from DOH, other agents, and individual beliefs.

As we continue our research on the impact of determinants of health, we will expand the

current model to include timely interventions to reinforce positive behavior and mitigate the

likelihood of relapse. We will also gather and incorporate real-world demographic data to

populate our model. Additionally, we will expand our modeling efforts to incorporate work on

frailty [17], exercise [18,19], and other determinants of health, into more intricate models

following the modeling cycle [9] presented herein with the purpose of creating more accurate

representations of the real world.

 18

References

1. Chevance G, Perski O, Hekler EB. Innovative methods for observing and changing complex
health behaviors: four propositions. Translational Behavioral Medicine. 2021;11(2):676-
685. doi:10.1093/tbm/ibaa026

2. WHO | The determinants of health. WHO. Accessed February 19, 2022.

https://www.who.int/hia/evidence/doh/en/

3. Choi E, Sonin J. Determinants of Health Visualized. Accessed December 7, 2022.

https://www.goinvo.com/vision/determinants-of-health/

4. Frequently Asked Questions | Social Determinants of Health | NCHHSTP | CDC. Published

May 3, 2019. Accessed November 27, 2022.
https://www.cdc.gov/nchhstp/socialdeterminants/faq.html

5. Centers for Disease Control and Prevention (CDC). 2014 SGR: The Health Consequences of
Smoking—50 Years of Progress. Centers for Disease Control and Prevention. Published
March 5, 2018. Accessed December 9, 2022.
https://www.ncbi.nlm.nih.gov/books/NBK179276/pdf/Bookshelf_NBK179276.pdf

6. Cole BL, Fielding JE. Health Impact Assessment: A Tool to Help Policy Makers Understand
Health Beyond Health Care. Annu Rev Public Health. 2007;28(1):393-412.
doi:10.1146/annurev.publhealth.28.083006.131942

7. Scholz U. It’s Time to Think about Time in Health Psychology. Appl Psychol Health Well
Being. 2019;11(2):173-186. doi:10.1111/aphw.12156

8. Helmich MA, Wichers M, Olthof M, et al. Sudden gains in day-to-day change: Revealing
nonlinear patterns of individual improvement in depression. J Consult Clin Psychol.
2020;88(2):119-127. doi:10.1037/ccp0000469

9. Steven F. Railsback, Volker Grimm. Agent-Based and Individual-Based Modeling. Princeton

University Press; 2012.

10. Brennan PF, Strombom I. Improving Health Care by Understanding Patient Preferences. J
Am Med Inform Assoc. 1998;5(3):257-262. Accessed December 19, 2022.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC61299/

11. Dunton GF. Sustaining Health-Protective Behaviors Such as Physical Activity and Healthy
Eating. JAMA. 2018;320(7):639. doi:10.1001/jama.2018.6621

12. Christakis NA, Fowler JH. The collective dynamics of smoking in a large social network. N
Engl J Med. 2008;358(21):2249-2258. doi:10.1056/NEJMsa0706154

13. Westmaas JL, Langsam K. Unaided smoking cessation and predictors of failure to quit in a
community sample: effects of gender. Addict Behav. 2005;30(7):1405-1424.
doi:10.1016/j.addbeh.2005.03.001
 19

14. Wilensky U. Center for Connected Learning and Computer-Based Modeling, Northwestern
University, Evanston, IL. Published online 1999. Accessed December 9, 2022.
https://ccl.northwestern.edu/netlogo/

15. Holland, John H. Hidden Order: How Adaptation Builds Complexity. Addison-Wesley.;
1995.

16. Holland JH. Studying Complex Adaptive Systems. Jrl Syst Sci & Complex. 2006;19(1):1-8.
doi:10.1007/s11424-006-0001-z

17. Phillips AB, Sordo M, Lopez RP, Merrill JA. Developing a Framework for Evaluating
Influencers of Frailty in Aging. Stud Health Technol Inform. Published online 2017.

18. Phillips, AB, Sordo, M. Use of an Adaptive Agent-Based Model in Evaluating Patient
Preferences in Healthcare. Stud Health Technol Inform. Published online 2015.

19. Phillips AB, Sordo M. Agent Based Model to Evaluate Simple Behavioral Changes on
Individual Health Over Time. Stud Health Technol Inform. Published online 2019.

20. Wetter DW, Kenford SL, Smith SS, Fiore MC, Jorenby DE, Baker TB. Gender differences in
smoking cessation. J Consult Clin Psychol. 1999;67(4):555-562. doi:10.1037//0022-
006x.67.4.555

21. Jarvis MJ, Cohen JE, Delnevo CD, Giovino GA. Dispelling myths about gender differences
in smoking cessation: population data from the USA, Canada and Britain. Tobacco Control.
2013;22(5):356-360. doi:10.1136/tobaccocontrol-2011-050279

22. Smith PH, Bessette AJ, Weinberger AH, Sheffer CE, McKee SA. Sex/gender differences in
smoking cessation: A review. Prev Med. 2016;92:135-140.
doi:10.1016/j.ypmed.2016.07.013

23. Royce JM, Corbett K, Sorensen G, Ockene J. Gender, social pressure, and smoking
cessations: the Community Intervention Trial for Smoking Cessation (COMMIT) at
baseline. Soc Sci Med. 1997;44(3):359-370. doi:10.1016/s0277-9536(96)00149-9