Psychosocial phenotyping as a personalization strategy for chronic disease self-management interventions

Abstract

Background: As the U.S. population grows older and more diverse, self-management needs are increasingly complicated. In order to deliver effective personalized interventions to those suffer from chronic conditions social determinants of health must be considered. Therefore, psychosocial phenotyping holds strong promise as a tool for tailoring interventions based on precision health principles. Purpose: To define psychosocial phenotyping and develop a research agenda that promotes its integration into chronic disease management as a tool for precision self-management interventions. Methods: Since psychosocial phenotyping is not yet used in interventions for self-management support, we conducted a literature review to identify potential phenotypes for chronic disease self-management. We also reviewed policy intervention case reports from the Centers for Medicare and Medicaid Services to examine factors related to social determinants of health in people with chronic illnesses. Finally, we reviewed methodological approaches for identifying patient profiles or phenotypes. Results: The literature review revealed areas within which to collect data for psychosocial phenotyping that can inform personalized interventions. The findings of our exemplar cases revealed that several environmental or key SDOH such as factors realted with economic stability and neighborhood environment have been closely linked with the success of chronic disease management interventions. We elucidated theory, definitions, and pragmatic conceptual boundaries related to psychosocial phenotyping for precision health. Conclusions: Our literature review with case example analysis demonstrates the potential usefulness of psychosocial phenotyping as a tool to enhance personalized self-management interventions for people with chronic diseases, with implications for future research.

Introduction

According to a 2011 CDC report, 6 out of 10 US adults suffer from chronic diseases, at an estimated annual healthcare cost of 3.3 trillion dollars [1]. Changes in demographics, lifestyle, and environmental factors, as well as medical successes, have transformed once terminal diagnoses into chronic conditions. To effectively manage chronic diseases such as diabetes or hypertension, self-management and lifestyle modifications are important. Self-management has been defined as “the ability of the individual, in conjunction with family, community, and healthcare professionals, to manage symptoms, treatments, lifestyle changes, and psychosocial, cultural, and spiritual consequences of health conditions (particularly chronic diseases)” [2]. It is well-known that self-management and changes in health behavior are challenging for many people [3]. Effective chronic disease management requires significant self-management skills, including not only changes in health behavior but also maintenance, problem solving, and resource utilization, all of which must be integrated into patients’ daily lives in order to achieve benefits for health [4].

Individual differences play an important role in the adoption and maintenance of health behavior changes for self-management. For many decades, theory-based tailoring has been the focus of efforts to personalize interventions to improve patients’ self-management success. However, in 2015, the Precision Medicine Initiative ushered in a new era in the use of personalized medicine [5]. This research effort is leading to new approaches in pharmacology [6,7] and clinical medicine [8] that leverage extensive genetic, genomic, and clinical data for individualized care. In cancer management, for example, individualized treatment plans enabled by advances in the availability of data and in analytics are increasingly available [9].

Given the societal need to address chronic diseases and psychosocial challenges, new efforts to provide self-management support for these individuals might utilize precision self-management plans. We propose that rather than supplanting theory-based intervention tailoring with data-driven precision approaches, these two paradigms should work in concert. Existing theoretical evidence can be combined with novel insights gained from machine learning and big data to improve self-management science. While precision approaches have thus far focused on individual factors, including genetics, clinical diagnoses, psychology, and behavior, many health behavior theories that can be applied to self-management address the collective function of intra- and interpersonal factors that take place within family, community, and society such as social determinants of health (SDOH).

Thus there is a distinct need to obtain comprehensive information about individuals to create meaningful profiles, or phenotypes, that can inform the personalization of self-management interventions. In the context of chronic disease management, however, the science of phenotyping is in its infancy. For genetic phenotyping, we currently use genome sequencing; for clinical phenotyping, we use electronic health record (EHR) data. While there is some effort to merge individuals’ clinical data and genetics, e.g., eMerge [10], to provide insights for effective treatment, there are few approaches that include psychosocial factors that influence self-management behaviors.

Here we present a framework for synchronizing these seemingly divergent paradigms via “psychosocial phenotyping”. First we define psychosocial phenotyping; then we review the literature to identify the constructs that should be included in this new type of phenotyping; and finally we discuss the implications of this approach for self-mangement interventions, including methodologies, data sources, and implications for health disparities and ethics. Furthermore, we offer methodological insights and discuss the potential interplay between psychosocial phenotyping and other emerging approaches in self-management science such as digital phenotyping, health equity in the context of population science, and ethical implications.

Psychosocial phenotyping definition

Precision medicine efforts to personalize healthcare for individual patients have focused primarily on using genomic data or selective biomarkers to identify molecularly selective therapeutic targets [11]. More recently, data from clinical records to identify individual differences in prognoses [12,13] have been used. Although these efforts to integrate multiple sources of data into patient care are increasing, precision medicine approaches are used almost exclusively in treatment, despite their potential applicability in prevention and in behavioral interventions. There is extensive evidence from self-management science and behavioral research that different people respond to behavioral interventions in different ways [14]. Individual characteristics such as gender, personality, and cultural and contextual factors influence educational and behavioral interventions. Moreover, growing evidence indicates that environmental factors and SDOH exert strong influences on the prevention and management of chronic diseases.

Phenotyping based on genetic and clinical data has been used to identify characteristics to inform response platforms for treatment options in precision medicine. Phenotypes, sets of observable characteristics that reflect the interaction of an organism’s genes and environment, present an apt analogy for characterizing the combinations of attitudes, social influences, and personal agency that mediate individuals’ chronic disease management. However, these data are not typically represented in current approaches to precision health. We propose that psychosocial phenotyping can propel the use of precision health in self-management science for the future of chronic disease management.

Psychosocial phenotypes have been defined on the basis of the “psychological and social characteristics” of patients’ obtained from EHRs [15]. More recently, however, a patient’s psychosocial phenotype has been defined as “the combination of psychological and social characteristics that explain variations in behavioral response to an intervention” [16]. The National Institute of Diabetes and Digestive and Kidney Diseases refers to a related concept, the behavioral and psychological phenotype, as “a pattern of behavior or psychological characteristics that are measurable/quantifiable and distinct (explains individual variation)” [17]. All of these definitions are similar, but it is important that research on psychosocial phenotypes and psychosocial phenotyping methods focus not only on factors solely internal to the patient but also on external, structural factors that interact with the individual and influence the individual’s response to self-management interventions.

Based on prior work in the field [16,17], we propose a working definition of psychosocial phenotyping in the context of self-management science as a methodology to identify patterns of measurable, quantifiable, distinct behaviors or psychological characteristics that can explain individual variation in the context of the self-management of chronic conditions. This definition is based on the following theoretical premises: (1) as a variable, the psychosocial phenotype must represent the complex interplay between psychological and social determinants of health; (2) phenotyping is a valid, replicable way to identify the behavioral and/or psychological expressions (phenotypes) that meaningfully explain individual variability in behavioral or clinical outcomes in response to self-management interventions; and (3) the identification of phenotypes should improve the matching or tailoring of interventions or suggest novel targets for more efficacious individual and population-level approaches for the self-management of chronic conditions [18].

Genetic or genomic phenotyping has been validated using well-established sequencing protocols for the analysis of big data accumulated over the years. In this paper, we aim to develop an agenda for future research in psychosocial phenotyping for the self-management of chronic disease management on the population level. To fully understand and validate psychosocial phenotyping for use, multiple sources of large, diverse sets of data are necessary-self-reported data, community level data, data collected in clinical settings, and digital traces of behavior. Moreover, a theoretical framework and consensus regarding important factors for psychosocial phenotyping are needed. Finally, common data elements for psychosocial phenotyping must be identified, especially those that are potentially useful in the context of self-management behaviors of those with chronic illnesses.

Literature review method

We conducted an integrated literature review to identify the most common and measurable phenotypic characteristics of individuals that are predictive of individual variation in self-management processes and outcomes. By reviewing current psychosocial phenotyping efforts and including exemplars of psychosocial phenotyping methods in relation to chronic disease management, we were able to establish a basis for psychosocial phenotyping as well as the obvious gaps in the literature. We then present a framework to guide future research.

Because there is no scientific consensus regarding how to best derive psychosocial phenotypes, it is important to consider avenues traditionally used to target, account for, and better understand variations in response to interventions. Prior literature has explored myriad reliable psychosocial/nongenomic factors potentially significant for the development of composite phenotypes. By examining the literature for the types of factors that have been assessed and finding ways to integrate them within advances in data science, it may be possible to arrive at a synergy between traditional and contemporary methods.

We specifically reviewed evidence on chronic disease management in the context of metabolic, cardiovascular, and respiratory conditions as well as diabetes mellitus. We searched the English-language literature in PubMed and Web of Science through July 2019 using the following search terms and Medical Subject Headings: chronic disease, diabetes, hypertension, heart failure, asthma, COPD, obesity, intervention adherence, intervention compliance, intervention persistence, education, behavior, profile, characteristics, pheynotype, factor, determinant, predictor, correlate. Included studies were review articles-literature reviews of both psychosocial factors AND phenotypes/profiles, according to titles/abstracts-that examined at least two psychosocial factors for participants but not for the interventions themselves. Studies were excluded if they focused on pharmacological management, clinical phenotyping, pain, or genetic disorders, or if they were done with animal subjects. Articles were then reviewed by the authors to ensure that they met selection criteria, with any disagreements resolved via group consensus.

Selection of “exemplar” SDOH cases

To select pertinent cases that illustrate the promise of psychosocial phenotyping, we also considered several policy intervention case reports from the Centers for Medicare and Medicaid Services (CMS) that address social determinants of health (SDOH) of people with chronic illnesses and demonstrate the clinical utility of personalized interventions informed by social needs rather than simply by medical diagnosis [19-22]. Among them, we chose the best examples for each of the domains of factors that emerged from our initial literature review. The selected studies were reviewed and discussed.

Results

Search results, characteristics, and factor domains

Our search yielded 28,379 studies that were reduced to 105 through title and abstract screening. Full text screening further reduced those studies to 24, which consisted of systematic reviews and meta-analyses, the majority of which were published in the last 5 years (71%, n = 17). See Figure 1 for study selection flowchart. Eleven of the 24 studies (46%) were meta-analyses. All chronic conditions represented by the search terms were included in our sample, with cardiovascular disease (38%, n = 9 studies) the most frequent, followed by metabolic syndrome (25%, n = 6) and diabetes (17%, n = 4). The most frequent study outcomes measured participants’ adherence/engagement to interventions (67%, n = 16 studies), and nearly all of the studies evaluated associations between psychosocial factors and the stated outcome(s) of interest (92%, n = 22). The reviewed studies yielded five domains of factors important to the formation of psychosocial phenotypes: demographic, psychological, social, clinical, and environmental. Demographic and psychological characteristics tied as the most frequently included domains (79%, n = 19 studies) and only 3 studies (13%) examined all five domains of factors.

Figure 1.

Flow diagram of study selection process.

Demographic characteristics

The majority of the studies (79%, n = 19) included at least two of the commonly examined characteristics of gender, age, racial/ethnicity, marital status, educational attainment, income, and socioeconomic status [4,8,23-34]. Studies also examined specific characteristics that related to their specific research question(s): immigration status [8], number of children living in the home [4], languages spoken [24], insurance type(s) [1,28], religion [25], and other financial indicators such as out-of-pocket spending on prescription drugs or insurance copayments [3,28,29,34]. Of the three studies that did not include demographic characteristics, one was a review paper focused specifically on psychosocial predictors [31], and two were not review papers but described psychosocial phenotypes using other methods: one was a qualitative study [16]; the other, a multidisciplinary workshop [26].

Many of the studies examined in the 24 articles presented individual interventions in which specific demographic or individual characteristics were selected as factors in intervention design. The most frequently used individual characteristics were gender, level of education, access to regular care, and household or individual socioeconomic status. Among demographic factors, gender, age, and education were often linked to clinical outcomes or the process of self-management of chronic illnesses. For example, one meta-analysis [30] found that medication adherence interventions in hypertensive patients were most effective in older women with higher education and moderate to higher income. The authors of that article suggested that characteristics of interventions would need to be adapted for younger men with limited income to increase effectiveness.

Psychological factors

Multiple psychological factors are related to a person’s emotional and mental state. Such factors were as commonly examined in the literature as demographic characteristics (79%, n = 19 studies) [15,16,24-27,29,31-43]. Included psychological factors were as follows: self-efficacy (25%, n = 6) [16,25,27,32,33,36], depression (58%, n = 14) [24,25,27,29,31,33,35,36,38,39,41-44], stress [25,27,36], body image [36], and readiness/stage of change [36]. In 2 studies, self-efficacy [25] and stage of change [24] were derived from theoretical approaches to behavior change.

Several of the meta-analyses indicated that individual levels of self-efficacy explained significant variations in self-management outcomes among populations with chronic illnesses [24,25,28]. Similarly, levels of depression also influenced both processes (self-management behaviors) and outcomes of self-management in many studies [31]. For example, self-efficacy was a strong predictor for dietary adherence among people with Type 2 diabetes in a meta-analysis conducted by Brown et al. Also, people with depression were twice as likely to be observed with medication nonadherence as were people without depression [31].

One novel factor included in two studies [31,33] addressing cardiovascular disease was “distressed” or Type D personality. Type D personality-social inhibition and negative affectivity-was found to be related to medication nonadherence [2,24,45-47] and decreased consulting behavior despite increased symptoms [24,48,49].

Social factors

Ten studies (42%) [2,4,7,8,16,26,31-33,50] addressed factors such as social support (25%, n = 6) [2,4,16,26,31,33]. Other such factors, including social pressure [8], social norms [7], patient-child relationship [32], and social network size [2], were addressed by single studies in the sample. Results from 5 studies examining social support showed that it was associated with the outcome of interest and showed a positive relationship such that increased social support increased self-management outcomes (21%, n = 5) [2,8,24,26,33]. For example, a meta-analysis by Lemstra found that weight loss interventions with social support had higher adherence than did those without social support (RR, 1.29: 95% CI, 1.24-1.34) [35]. The social support in the studies in Lemstra’s meta-analysis included group sessions, peer coaches, social support contracts, and “buddy” programs [35].

Clinical factors

The clinical domain comprised factors related to the chronic disease of interest and any comorbidities and was represented by 14 studies in the sample (58%) [3,5,9,15,16,23,25,28,29,31-34,50]. The most common factors were duration of disease [5,9,25], weight status/BMI [16,23], number of medications [25,33], and comorbidities (25%, n = 6 studies) [3,25,29,31-33]. For example, people with disease conditions such as osteopororsis and hyperlipidemia had the highest rates of nonadherence to medication interventions in comparson with people with diabetes in one meta-analysis [28]. Three studies used clinical factors for their outcome of interest, with two examining hemoglobin A1C [5,9] and one examining cardiorespiratory fitness via maximal or submaximal incremental cardiopulmonary exercise on a treadmill or cycle ergometer [30].

Environmental and SDOH-related factors

The final domain of factors that emerged from the literature was the environment, which was least frequently assessed (33%, n = 8 studies) [1,3,4,7,23,31,34]. Environment was most often operationalized in the assessment of healthcare settings (17%, n = 4 studies) [1,3,4,31], with a few studies including more specific factors such as neighborhood environment [7,23] and community population type such as urban versus rural [31,34]. Our sample shows a clear gap in the traditional literature in addressing environmental factors within the context of chronic disease management.

Results from the exemplar case studies found that in addition to safety of the home and neighborhood environment, evolving literature on SDOH includes factors related to food security and access to transportation [51]. Because of information constraints in traditional research or literature regarding broad social factors, we examined recent efforts by the CMS to address SDOH. Many SDOH demonstration intervention projects are still in planning or undergoing implementation, so that full evaluation reports are not yet available, but preliminary reports on these programs do provide insight for identifying important social factors that may be useful in expanding our understanding of SDOH in those with chronic diseases. Overall, early reports indicate that programs addressing SDOH tend to produce better health outcomes, better health equity, and more economic strategies than do traditional programs based on medical diagnosis [52]. For example, under a 2012 Medicaid demonstration waiver, the state of Oregon introduced a coordinated care model in which organizations experimented with reimbursement for a wide variety of flexible services that supplement covered benefits, in order to address SDOH that affect individuals’ care processes. Examples of reimbursed flexible services provided by Oregon care organizations include (1) items (or devices) helpful for managing specific chronic conditions; (2) assistance with food and nutrition; (3) classes or memberships to promote wellness; (4) temporary housing and environmental improvements; (5) transportation; and (6) care coordination or case management programs [21]. Precise cost effectiveness data are not yet available, but this SDOH program for a CMS population as well as other health care plans that recognize SDOH are rapidly being expedited.

According to a recent evaluation of approaches to SDOH across 17 states, commonly reported topics for SDOH screening were housing instability and food insecurity [22]. Many of these programs are based on the definitions and framework for SDOH given in Healthy People 2020, which refer specifically to (1) economic stability (employment, food insecurity, housing instability, poverty); (2) education (early childhood education and development, enrollment in higher education, high school graduation, language and literacy); (3) social and community context (civic participation, discrimination, incarceration, social cohesion); (4) health and health care (access to health care, access to primary care, health literacy); and (5) neighborhood and built environment (access to foods that support healthy eating, crime and violence, environmental conditions, quality of housing) [53].

A psychosocial phenotyping approach that encompasses all the major factor domains including environmental and SDOH can hold the greatest potential in eliciting factors of most significance to chronic disease self-management. Specifically, the demographic, psychological, social and environmental factors discussed above are the likely candidates for inclusion in psychosocial phenotypes and deserve further exploration. See Table 1.

Table 1.

Characteristics of included studies

First author, Year	Study design	Chronic disease focus	Sample characteristics	Selected psychosocial factors & outcomes addressed	Factors linked to outcomes	Domains addressed
1. Alramadan, 2018	SR	T2DM	● 13 studies	Factors:	A1c:	Demographics √
			● Arabian Gulf Council countries	● Diet	- Longer T2DM duration	Psychological √
				● Med adherence	- Anxiety	Social
				● Self-mgmt	- Depression	Clinical √
				● Attitudes toward T2DM	- Lower education level	Environmental
				● Anxiety	- Poor med adherence
				● Depression
				● T2DM duration
				● Education
				● Gender
				Outcome:
				● A1c
2. Brown, 2016	SR/MA	T2DM	● 739 research reports	Factors:	A1c:	Demographics √
			● 533,445 participants	● Stress	+ Dietary adherence	Psychological √
			● Mean age 58.9 ± 6.2 years	● Depression	+ Coping	Social
			● 52% of studies in the US	● Anxiety	+ Stress	Clinical √
			● 73% published	● Self-efficacy	Adherence behaviors:	Environmental
			● 11% attrition rate	● Coping	+ Self-efficacy
				● Diet adherence	Fasting blood glucose:
				● Exercise adherence	- No associations
				● Med adherence
				● Diabetes duration
				Outcomes:
				● A1c
				● Fasting blood glucose
				● Adherence behaviors
3. Bryan, 2017	NIDDK workshop and exemplars	Physical activity and sedentary behavior	● Not applicable	Factors:	None assessed	Demographics
				● Reinforcing value		Psychological √
				● Affective response		Social
				Outcome:		Clinical
				● None assessed		Environmental
4. Burgermaster, 2018	Qualitative	Obesity prevention	● 18 children	Factors:	4 Psychosocial phenotypes related to outcomes identified:	Demographics
			● 5th grade	● Self-efficacy	+ Activated-successful at behavior changes	Psychological √
			● 88% racial/ethnic minority	● Self-regulation	- Inspired-motivated but not successful at behavior changes	Social √
				● Neighborhood environment	+ Reinforced-experienced at target health behaviors	Clinical
				● Social norms	- Indifferent-uninterested in behavior changes	Environmental √
				● Skills-nutrition label reading
				Outcome-increased:
				● Fruit & veggies
				● Exercise
				Outcome-decreased:
				● Sugary drinks
				● Processed snacks
5. Burgess, 2017	SR	Obesity	● 24 total studies	Factors:	Predictors of adherence:	Demographics √
			● 17 of 24 identified predictors of behavior change	● Self-efficacy	+ Mood	Psychological √
				● Psychiatric disorders	+ Male	Social √
				● Stress	+ Older age	Clinical
				● Anxiety	+ Early weight loss success	Environmental
				● Depression	Barriers to behavior change:
				● Social pressure	+ Physical limitations
				● Immigrant status	+ Low income
				Outcomes:	+ Poor motivation
				● Predictors of adherence	+ Social pressures
				● Barriers to behavior change
6. Cheen, 2019	SR/MA	Chronic diseases	● 33 studies in SR	Factors:	Med non-adherence:	Demographics √
			● 79% cohort studies in SR	● Age	+ Chronic disease type-osteoporosis	Psychological
			● 31 studies in MA	● English speaking	+ Younger age	Social
			● 519,971 participants in MA	● Insurance type	+ Number of co-medications	Clinical
				● Medicare status	+ High co-payment	Environmental √
				● Distrust of meds
				● Care setting
				● Alcohol consumption
				● Chronic dx type
				Outcome:
				● Med non-adherence
7. Choi, 2017	SR/MA	T2DM	● 33 studies in SR	Factors:	Med engagement:	Demographics √
			● 79% in US	● Age	- Women	Psychological √
			● 22 studies in MA	● Gender	- Depression	Social
			● MA included age, gender, depression and costs only	● Depression	- Out-of-pocket costs	Clinical √
				● Out-of-pocket costs	+ Older age	Environmental √
				● Comorbidities
				● Glycemic control
				● Pharmacy type
				Outcome:
				● Med engagement
8. Conn, 2015	SR/MA	HTN	● 101 studies	Factors:	Med adherence:	Demographics √
			● 88% published	● Setting	+ Female	Psychological
			● 8.1% attrition	● Discipline	+ Older age	Social √
				● Dose	+ Moderate to high income	Clinical
				● Duration	+ Duration of intervention	Environmental √
				● Social support for adherence
				● Socioeconomic status
				● Age
				● Gender
				Outcome:
				● Med Adherence
9. Crawshaw, 2016	SR/MA	Acute coronary syndrome	● 17 total studies	Factors:	Med non-adherence:	Demographics
			● 7,401 participants	● Depression	+ Depression	Psychological √
			● Mean age 61.8 ± 4.5 years	● Distressed personality type	+ Distressed personality type	Social √
				● Social network size	+ Treatment beliefs	Clinical
				● Social support	- Social support	Environmental
				● Treatment beliefs
				Outcome:
				● Med non-adherence
10. Fuentes, 2019	Cohort, Creation of two psycho-social profiles to explore associations with socio- economic variables	Obesity	● 4,519 participants	Factors included in psychosocial profiles:	Psychosocial profiles (adverse and favorable profiles) and socioeconomic variables:	Demographics √
			● Paris, France	● Weight dissatisfaction	- No associations	Psychological √
			● Mean age 53.3	● Weight locus of control		Social
			● 67.1% employed	● Perceptions of body		Clinical √
			● 21% obese	● Quality of life		Environmental √
				● Self-efficacy
				Outcomes:
				Socioeconomic variables
				● Childhood socioeconomic status
				● Socioeconomic status
				● Neighborhood socioeconomic level
11. Gundlapalli, 2013	Develop algorithms for psychosocial phenotyping	Not applicable	● Veterans Affairs data	Developed lexicon for homelessness:	Psychosocial concepts identified:	Demographics √
			● Pulled from 218 standard note titles	● 300+ terms identified from literature, experts, and medical records	+ 58,707	Psychological √
			● Used NLP pipeline v3NLP	Outcomes:	Hit rate:	Social √
				● Number of psychosocial concepts identified	+ 0.2 concepts per document	Clinical √
				● Hit rate	Precision:	Environmental √
				● Precision	+ 80%
				● Sensitivity	Sensitivity:
					+ 49% (95% CI 43-55%)
12. Kessing, 2016	SR/MA	Heart failure	● 65 studies	Factors:	Self-care:	Demographics
			● 31 studies from US	● Anxiety	- Depression	Psychological √
			● 72% cross sectional	● Depression	- Self-efficacy	Social
				● Self-efficacy	- Mental well being	Clinical
				● Distressed personality type	Med adherence:	Environmental
				● Mental well being	- None significant
				Outcome:
				● Self-care
				● Med adherence
13. Krass, 2015	SR/MA	T2DM	● 27 studies	Factors:	Med Adherence:	Demographics √
			● 67% cross-sectional	● Age	- Depression	Psychological √
			● 60% in US	● Education	- Healthcare costs	Social
				● Health status		Clinical √
				● Socioeconomic status		Environmental
				● Diabetes duration
				● Comorbidities
				● Number of meds
				● Types of meds
				Outcome:
				● Med adherence
14. Lemstra, 2016	SR/MA	Weight loss/obesity	● 27 studies	Factors:	Weight loss adherence:	Demographics √
			● 6,803 participants	● Age	+ Social support	Psychological √
			● 74% RCTs	● Income		Social √
				● Education		Clinical
				● Social support		Environmental
				● Depression/mood
				● Weight
				● Smoking status
				Outcome:
				● Weight loss adherence
15. Leung, 2017	Literature review	Weight mgmt/obesity	● 19 studies	Factors:	Weight mgmt. adherence:	Demographics √
			● 47% RCTs	● Self-efficacy	+ Older age	Psychological √
			● 15 studies BMI ≥30	● Depression	+ Higher education	Social √
				● Stress	- Depression	Clinical √
				● Previous weight loss attempts	- Stress	Environmental
				● Age	- Previous weight loss attempts
				● Employment status
				● Education
				● BMI
				Outcome:
				● Weigh mgmt. adherence
16. Lewey, 2013	MA	Statin use	● 53 total studies	Factors:	Statin adherence:	Demographics √
			● 2,663,638 participants	● Gender	- Women	Psychological
			● 55% in US	● Race	- Nonwhite race	Social
				Outcome:		Clinical
				● Statin adherence		Environmental
17. Mann, 2010	SR/MA	Statin use	● 22 total studies	Factors:	Med adherence:	Demographics √
			● 2,663,638 participants	● Race	- Lower and oldest age	Psychological √
			● 55% in US	● Income	- Women	Social
				● Depression	- Lower income	Clinical √
				● Dementia	+ Diagnosis of diabetes or HTN	Environmental
				● Health beliefs
				● Disease severity
				● Comorbidities
				● Costs
				Outcome:
				● Med adherence
18. Ofori-Asenso, 2018	SR/MA	Statin use	● 45 total studies	Factors:	Nonadherence:	Demographics √
			● 1,842,054 participants	● Age	+ Smoking status	Psychological √
			● 53% in Europe	● Gender	+ Women	Social
				● Smoking status	+ Depression	Clinical √
				● Comorbidities	- History of CVD	Environmental
				● Copayment	Discontinuation:
				● Depression	+ Smoking status
				Outcome:	+ Low income
				● Nonadherence	- History of CVD
				● Discontinuation	- Comorbidities of diabetes or HTN
19. Ombrellaro, 2018	SR/MA	Cardio-respiratory fitness	● 15 studies in SR	Factors:	Cardiorespiratory fitness:	Demographics √
			● 3 studies in MA	● Education	+ High education	Psychological
			● 93% cross-sectional	● Socioeconomic status		Social
			● 40% in US	Outcome:		Clinical
				● Cardiorespiratory fitness		Environmental
20. Oosterom-Calo, 2013	SR	Heart failure	● 11 studies	Factors:	Med adherence:	Demographics √
			● 64% in US	● Age	+ Hospital setting	Psychological √
				● Education		Social √
				● Social support		Clinical √
				● Setting		Environmental √
				● Comorbidities
				● Functional status
				● Depression
				Outcome:
				● Med adherence
21. Ritz, 2013	Literature review	Asthma	● 34 studies	Factors:	None assessed	Demographics √
				● Race		Psychological √
				● Parent-child relationship		Social √
				● Family conflicts		Clinical √
				● Anxiety		Environmental √
				● Depression
				● Urban environments
				● Illness beliefs
				● Physical activity level
				● Comorbidities
				Outcome:
				● None assessed
22. Sedlar, 2017	SR	Heart failure	● 30 studies	Factors:	Self-care:	Demographics √
			● 50% cross-sectional	● Age	- Depression	Psychological √
				● Health-related quality of life		Social
				● Gender		Clinical √
				● Education		Environmental
				● Depression
				● Left ventricular ejection fraction
				Outcome:
				● Self-care
23. Wu, 2008	Literature review	Heart failure	● 50 studies	Factors:	Med adherence:	Demographics √
			● 50% cross-sectional	● Income	+ Social support	Psychological √
				● Race	- Forgetfulness	Social √
				● Comorbidity		Clinical √
				● Depression		Environmental
				● Forgetfulness
				● Multiple meds
				● Social support
				Outcome:
				● Med adherence
24. Zeber, 2013	SR	Chronic diseases	● 24 studies	Factors:	Initial med adherence:	Demographics √
			● 50% cross-sectional	● Drug class	- Poor treatment response	Psychological
				● Age	- High copayments	Social
				● Socioeconomic status	- Middle and older age	Clinical √
				● Urban residency	- Comorbidities of long-term diseases like diabetes or psoriasis	Environmental √
				● Copayment	Overall med adherence:
				● Treatment response	- Younger age
				● Comorbidities	- Poor social support
				Outcome:
				● Initial med adherence
				● Overall med adherence

Discussion

Knowledge gaps and future directions

Overall, psychosocial phenotypes and phenotyping present an evolving science with room for theoretical refinement, including useful definitions and pragmatic conceptual boundaries. As for cross-validation and replication, psychosocial phenotyping is still in an exploratory phase. There is certainly a need for psychosocial phenotyping, which shows promise as a tool to enhance personalized self-management intervention strategies for people with chronic diseases.

Theory

Whereas the first wave of precision medicine research has focused on gathering genomic and biological data, truly personalized care will be possible only when an intervention or care practice integrates psychological, behavioral, environmental, social, and cultural determinants along with biological determinants [54]. For psychosocial phenotyping methods to actualize the full potential of precision health and advance self-management science as well as health equity research, a concerted effort among researchers to develop and refine a theoretical framework to guide this line of inquiry is needed. The role of big data in precision health is critical, and precison health models must provide a theoretical basis with pertinent multilevel conceptual domains as well as guide data collection from multiple sources.

Our literature review shows that theoretical models that integrate biological, social, psychological, and environmental determinants and are operationalized empirically to guide precision health interventions are rare. At present, many current precision health interventions that use genomic and biological data are predominantly guided by narrow disease-specific theoretical frameworks. Fortunately, however, several conceptual and theoretical perspectives and models in public and behavioral health hold strong potential to guide precision health interventions: the ecological model [55], the precede-proceed model [56], and the chronic care model [57,58]. Those models have been used to provide a theoretical orientation for multilevel projects rather than as empirical theoretical models to guide testable hypotheses. With rapid advances in technology and data science, however, such models can guide precision health interventions.

The development of theoretical models to guide precision health interventions or to support personalized healthcare decision-making should integrate ways to collect “genomic, biological, behavioral, environmental and other data on individuals on individuals” [59]. The integration of these dimensions into theoretical models and the accumulation of these data will enable useful psychosocial phenotyping to advance personalized health through self-management science, patient-centered care, and policy [54].

Methodology and measurement

Powerful methods for data capture and processing can help realize the promise of psychosocial phenotyping. Data sources should provide information that can encapsulate individuals’ internal characteristics such as health status, demographics, and self-management behaviors, as well as external characteristics such as social support or access to healthcare resources, among others. Technological advances have enabled the exploration of available, powerful personal devices and wearables with GPS units to implement personalized intervention with real-time assessment [59]. After the collection of multi-dimensional data, analytic tools must be sophisticated enough to provide a composite of the individual’s internal and external characteristics in order to develop a psychosocial profile for self-management. Our review suggests that four common methodological tools hold the potential to extract psychosocial phenotypes relevant to self-management.

Meta-analyses and related analytic tools

In the absence of well-validated psychosocial phenotyping in many populations, diseases, or environmental contexts, meta-analyses of patient characteristics that have been examined as mediators or predictors of chronic disease self-management behaviors can be an important methodology to inform the development of psychosocial phenotypes. Meta-analyses can include large enough numbers of patients for big data analysis and can be used to find patterns that inform the conceptualization of psychosocial phenotypes. Brown et al. [25] and Cheen et al. [28] conducted meta-analyses of studies that addressed predictors of disease-related behaviors and health outcomes in order to discover the relative impacts of certain psychosocial characteristics on disease-related self-management behaviors and related health outcomes in comparison with others (see Table 2). Psychosocial determinants such as depression, coping, or self-efficacy were found to have large effects on self-management behaviors such as physical activity or dietary adherence for diabetes [28] or heart failure [33]. Certain behaviors predictive of health outcomes such as dietary and medication adherence were strongly related to improved glycemic control, and glucose self-monitoring was related to fasting blood glucose [25]. In another systematic review of 53 studies (n = 2,663,638), Lewey and colleagues found that female gender and non-white race had higher odds of medication nonadherence. Such findings can inform interventions to improve medication adherence and self-care behaviors in especially vulnerable populations. A use case scenario with psychosocial phenotypes derived from these meta-analyses is provided in Table 2, along with benefits and limitations of using the methodology of meta-analyses for deriving psychosocial phenotypes.

Table 2.

Case Scenarios for Deriving Psychosocial Phenotypes from Selected Meta Analyses

Analytic Approach	Data Source	Use Case Scenario	Pros	Cons
NLP-based algorithm on EHR records	a) EHR, n = 316,355 documents	Terms related to homelessness phenotype that provided direct evidence of actual (e.g. sleeping in park), at risk (e.g. doubled up) or needs (e.g. needs socks) related to homelessness was extracted from the EHR	Sheer volume and variety of available data through EHRs	i) Potential inaccuracies in data quality and accuracy
a) Gundlapalli (2013)				ii) Static data that fails to capture of the dynamic evolution of health status of an individual temporally
Meta-Analysis	a) 31 studies, n = 519,971	a) Phenotype for medication non-adherence included younger age, higher number of concurrent medications, orthopedic practitioner specialty and higher co-payment	i) Marked increases in statistical power;	i) Restriction of variables to those that were measured by instruments in the included studies, but fail to capture unmeasured covariates that can act of potential confounders
a) Cheen (2019): Prevalence and factors contributing to medication non-adherence	b) 759 studies, n = 533,445	b) Phenotype for improved glucose control included self-efficacy, coping, dietary adherence and medication adherence	ii) Greater heterogeneity in subject demographics;	ii) Pooling of data may introduce uncertainty due to potential sampling errors or unmeasured covariates (Lemstra, et al., 2016)
b) Brown (2016): Predictors of diabetes outcomes			iii) Ppportunity to test hypotheses not considered in the original studies; and	iii) Failure to capture temporal influence of psychosocial factors due to static data measurement. (Lueng, et al., 2015)
			iv) Increased efficiency in both time and money incorporating the vast historical information already available
Population-level rich questionnaires	a) RECORD questionnaire with 6460 participants aged 30-79 years living in the Paris region between 2011 and 2014	i) Phenotype for adverse weight profile - negative body image, underestimation of the impact of weight in quality of life, low weight-related self-efficacy, and weight-related external locus of control;	i) Combination of multiple dimensions of socioeconomic characteristics in current socioeconomic status, economic status in childhood, and education status in the residential neighborhood allowed to assess the overall impact of a family of psychosocial mechanisms on obesity simultaneously	i) Eligibility to complete the survey was restricted to employed individuals which could have excluded more socio-economically deprived individuals
a) Fuentes (2020)		ii) Phenotype for favorable weight profile-positive body image, high self-efficacy, and internal locus of control		ii) Recall bias of participants to answer life-course related questions
				iii) Large proportion of observations with missing values (32%).
Qualitative	a) Interviews with 18 students participating in a school-based behavior change interventions	i) Activated psychosocial phenotype - successful behavior-changers with strong internal supports	Rich, and nuanced details that identify psychosocial characteristics of varying responses to behavioral interventions	Timing of the interviews precludes prospective identification of psychosocial phenotypes to assess their influence on intervention results
a) Burgermaster (2018)		ii) Indifferent psychosocial phenotype - uninterested in behavior change and only did target behaviors if family insisted

Structured and unstructured EHR documentation

EHRs are emerging as repositories of individuals’ data that can facilitate the derivation of psychosocial phenotypes for self-management. The proliferation of EHR usage makes it possible to gather data on thousands of patients, and the growing sophistication and robustness of analytic tools for natural language processing (NLP) and machine learning enable researchers to use EHRs as a viable data source to derive psychosocial phenotypes. Gundlapalli et al. [15] have used NLP on a large corpus of free-text clinical data including provider notes to unlock rich information to identify psychosocial phenotypes [15]. See Table 2 for a use case scenario with a derived phenotype, as well as benefits and limitations of using EHR records for deriving psychosocial phenotypes.

Qualitative data

Qualitative methodologies provide nuanced, rich analyses of patients’ contextual factors and can also be valuable for deriving psychosocial phenotypes related to self-management. For example, in the systematic review conducted by Burgess et al. [27], qualitative articles revealed factors relevant to individuals’ psychosocial contexts such as environmental, societal, and social pressures; negative thoughts/moods; socioeconomic constraints; and lack of enjoyment of exercise as barriers to behavior change. Burgermaster et al. [16] employed case-ordered meta-matrices to identify salient psychosocial phenotypes that helped in the derivation of four psychosocial phenotypes of responses to behavioral interventions to prevent childhood obesity (see Table 2).

Factor and cluster analysis

Factor and cluster analysis of large amounts of both structured and unstructured data can also serve as a tool for identifying and refining psychosocial phenotypes. Fuentes et al. employed multilevel population surveys administered to a city population cohort that included not only socioeconomic and demographic characteristics but also psychological health evaluations and health-related behavioral, psychological, cognitive, and attitudinal characteristics to derive psychosocial profiles for obesity [32]. The data from this large epidemiological cohort of adults enabled the use of sophisticated analytic methods such as factor analysis and cluster analysis to derive clear obesity-related psychosocial profiles. An interesting finding from this study was the clear relationship between the unified psychosocial profile and depression and gender but not other socioeconomic dimensions.

Emerging data types

Digital health data for psychosocial phenotyping is becoming ubiquitous owing to the proliferation of smart phones and the wide use of social media. Social media are increasingly becoming an avenue for gathering data on not only individuals’ health behaviors but also their attitudes toward those behaviors. Twitter tweets and Facebook posts and related responses can provide a cumulative understanding of trends in health behaviors at least among those who are active on social media. Thus, McIver et al. [60] characterized the profiles of individuals who posted in groups about sleep by identifying patterns in their tweets on Twitter. The use of social media as an additional data source to develop meaningful psychosocial phenotypes is still in its infancy and it is important to consider that the profiles developed from social media may not provide completely accurate representations of individuals.

Ethical implications

Psychosocial phenotyping based on social and behavioral data does have ethical ramifications with respect to the collection, storing, and analysis of such data. Implicit biases may result from overrepresentation of certain social or behavioral characteristics in a dataset and may exacerbate disparities. Datasets oftenunderrepresent minority communities due to disparities in healthcare access [61]. Incomplete data or data varying in accuracy may lead to confounding or misleading interpretations which could affect the utility of psychosocial phenotyping. Given the sensitive nature of social and behavioral data, it is imperative that researchers take steps to protect the privacy and confidentiality of individuals’ data and prevent misuse for commercial or legal purposes [62]. Finally, providing participants with informed consent that enables them to understand how their sensitive social and behavioral data will be protected, deidentified, and used is an important ethical consideration that will require continued and sustained efforts [62].

Health disparities implications

The U.S. population is growing older and becoming more diverse. As the population ages, so do its needs for care. The frequency and burden of chronic diseases are rising, and many subgroups within the increasingly diverse U.S. population are experiencing health disparities despite efforts to redress such gaps. We spend 1.5 trillion dollars on management and care of chronic conditions, yet more than half of our population lacks adequate disease manageme. For example, the recent COVID-19 pandemic has revealed that ethnic minorities from resource-scarce communities are more susceptible to COVID-19, with deaths disproportionately high among African Americans and other ethnic minority groups [23]. This unfortunate susceptibility among underserved populations is likely related to poor management of chronic diseases. Insufficient community health infrastructures create challenges in accessing care, as well as an inefficient environment for self-management support, leading to poor management of chronic diseases [23,50].

Limitations

The findings of this study have several important limitations, given the evolving nature of psychosocial phenotyping. First, we had to examine proxy variables rather than well-established phenotypes, because the science of psychoscial phenotyping has not yet matured. Second, our analysis is largely focused on the findings of meta-analyses rather than individual studies. Emerging data from digital platforms have the potential to provide equally strong insights, as do structured data from research projects that we could not include in this analysis.

Nevertheless, this paper suggests important implicatons for future research and practice in implementing precision health strategies for the self-management of chronic illnesses. Given that current psychosocial phenotyping is predominantly exploratory, with large volumes of unstructured data, our review suggests the need for a theoretically guided psychosocial phenotyping algorithm with adequate levels of efficiency and scientific sophistication. Ultimately, methodological insights will be translated in novel research to determine phenotypic characteristics of people with chronic conditions and deliver personalized clinical management and self-management interventions.

Conclusion

To fulfill the promises of precision health practice in the coming years requires a concerted effort in both research and practice. With this paper, we hope to raise awareness about the wide range of issues related to the application of precision health principles within self-management across various populations with chronic conditions. For the research community, we have suggested directions for future research, data collection, and methodology to advance self-management science by maximizing benefits of precision health and advanced technology. We hope that this paper will stimulate scientific discussions about the use of psychological, behavioral, social, and environmental information via useful phenotyping for effective personized self-management interventions in people with chronic conditions. Factors empirically validated as facilitating optimal management of chronic conditions can be used by researchers as common data elements to enrich a collective data pool [63] that will ultimately inform algorithms for valid psychosocial phenotyping and delivery of highly personalized interventions.

We also hope to stimulate a discussion of policy regarding future investment in the collection of pertinent data and the creation of infrastructures to support personalized interventions in populations with chronic conditions and/or limited resources. Current technology equips us to integrate large volumes of structured and unstructured data from multiple sources for phenotyping and subsequent personalized intervention delivery. This will rquire significant societal investment. The success of precision medicine in cancer treatment using genetic information to find precise therapies is based on years of societal investment in basic and clinical research, on the collection of significant amounts of data, on the building of technological infrastructures, and on the development of treatments. Thought leaders in behavioral medicine and population health are now calling for such investments “to provide patient-centered, personalized care, informed by the best combination of genomic, biological, behavioral, and social-environmental information” [54].

Given the increasing need of self-management support within various populations, precision interventions based on meaningful and practical psychosocial phenotyping will enhance the effectiveness of such programs, reduce financial and social costs, and improve the quality of life of the most vulnerable. Moreover, clinical practice equipped with useful phenotyping and technology-guided interventions will be an extremely powerful tool to improve health equity in populations with complex social needs and limited resources.

Disclosure of conflict of interest

None.