Social Stability Relates Social Conditions to the Syndemic of Sex, Drugs, and Violence

Abstract

The distribution of violence, sexually transmitted infections, and substance use disorders is not random, but rather the product of disease, behavior, and social conditions that co-occur in synergistic ways (syndemics). Syndemics often disproportionately affect urban communities. Studies of syndemics, however, rarely apply consistent measures of social conditions. Here, the construct of social stability (SS) (housing, legal, residential, income, employment, and relationship stability) was evaluated as a consistent measure of social conditions related to sex, drug, and violence exposures in a new population in a Mid-Atlantic urban center. Lower SS predicted greater likelihood of any and combinations of risk. The magnitude varied based on specification: odds of sex-drug-violence exposure were greater for low vs. high latent SS class (OR = 6.25; 95%CI = 2.46, 15.96) compared with low vs. high SS category (OR = 2.64; 95%CI = 1.29, 5.39). A latent class characterized by residential instability was associated with greater likelihood of risk—a relationship that would have been missed with SS characterized only as an ordinal category. SS reliably captured social conditions associated with sexual, drug, and violence risks, and both quantity and quality of SS matter.

Electronic supplementary material

The online version of this article (10.1007/s11524-020-00431-z) contains supplementary material, which is available to authorized users.

Introduction

Violence, sexually transmitted infections (STI), and substance use disorders individually have grave implications and economic burden for public and individual health. Deaths and injuries due to opioids, alcohol use disorders, violence, and epidemics of STI have markedly increased in the past 5 years, and HIV incidence has remained constant (40,000/year) [1–4]. US total lifetime costs of injuries and violence were estimated at $671 billion in 2013 [5], the opioid epidemic is estimated at over $1 trillion [6], and STI cost $16 billion annually [7]. Non-metropolitan areas are experiencing unprecedented increase in substance use epidemics; however, urban areas continue to bear the greatest burden [8].

These conditions do not occur in isolation, rather they are likely to co-occur, especially within the context of adverse social determinants of health (SDH). In fact, this phenomenon, described as a syndemic (synergistic epidemics), was referred to as SAVA (substance abuse, violence, AIDS) by Singer and Merrill [9] in the 1990s, where the presence of one exposure was found to predict another, and this co-occurrence often exacerbated negative consequences of each condition. (The syndemic referred to below is SDV: sex, drugs, and violence.) A unique feature of the syndemic approach to conceptualize public health and epidemics is the “emphasis on the situation and circumstances in which individuals live. In other words, syndemics fundamentally rely on context.” [10](p. 881)

Yet, most research on links between social conditions and health, including syndemics, does not consistently apply specific measures of social conditions. The construct of social stability (SS) may be a useful representation of social context. First referenced in the 1950s, SS captures multiple manifestations of social determinants of health at the individual level in a defined period of time. SS is described as the structure and routine that create circumstances that allow people to “maintain connections with social resources and societal expectations.” [11 (p. 3)] These attributes may protect against certain risks and facilitate healthy behaviors and outcomes. SS definitions vary and other constructs are used to connote similar meaning (structural vulnerability, social disadvantage, etc.) [12–14] German, in her study of women with heightened risk for heterosexually acquired HIV, defined SS as stability in housing, residence, income, employment, relationships, and freedom from imprisonment over 6 months [11, 15].

One advantage of SS over other terms used to operationalize SDH are its key tenets. SS considers the following: (1) domains are inter-connected and more than one is required to create stability; (2) a defined period of time for measurement of the duration of stability in each domain; and (3) both quantity (accumulation) and quality (patterns of co-occurrence) matter in interpretation of SS.

The current study extends the validation of German’s definition of SS as a unifying construct to partially represent the all-important aspect of social context in relation to syndemic manifestations of co-occurring risk exposures in a population that is under-examined in both syndemics and SS research: men and women with heterosexual risk for HIV in an urban setting. Prior studies focused primarily on women only, men only, people living with HIV, or people with substance use disorders [16–18].

This study examines whether the presence of one exposure (sexual risk behavior, problematic substance use, or violence) signals the presence of another, and if co-occurrence is more likely in certain social contexts (e.g., low social stability), thus satisfying conditions for syndemic consequences.

Finally, this study contributes a robust approach to examine relations between risks and SS by modeling two different composite SS characterizations: as a summative ordinal variable and a latent class variable. The summative variable does not consider the pattern or type of SS as all indicators are considered equal (= 1). In this approach, we assess whether the amount (ordinal category) or type (latent class) of SS is related to the likelihood of experiencing individual or combinations of risks. The relation of risk behaviors to demographic covariates was also examined given the potential for differential exposure based on biological and social contextual factors of gender, age, race, and level of education. Our findings will have implications for public health practitioners, researchers, clinicians, and policy makers in both urban and non-urban settings if the social stability representation of social context allows for better understanding, assessment, and addressing of multi-dimensional and inter-connected social conditions that create favorable environments for syndemics.

Methods

Design, Data, and Sample

This is a secondary analysis of data from the BESURE study, the Baltimore, MD site of National HIV Behavioral Surveillance Study (NHBS), specifically, the 3rd cycle investigating heterosexual risk for HIV: BESURE-HET3 [19, 20]. Original data were collected between September and December 2013, the most recent cohort data available at the time of investigation. The original and secondary studies were approved by the relevant institutional review boards.

Sample

Recruitment of the sample occurred via respondent-driven sampling (RDS), a technique to reach hidden or vulnerable populations [21]. Initial study recruits or “seeds” (n = 10), selected by the research team, were required to reside in census tracts of the Baltimore, MD metropolitan statistical area (MSA) with high prevalence of poverty, formerly incarcerated individuals, and heterosexually acquired HIV [22]. Data were collected in community-based setting with seeds recruiting persons within their social and sexual networks. Non-seed participants must reside in the MSA and must have engaged in heterosexual sex in past 12 months [23]. (Persons reporting non-heterosexual sex were not excluded. Of participants identifying as male, 14% reported same-sex partners (15 in last year), and 17% of women reported same-sex partners (all (n = 42) in the past year).) Recruits would bring in coupons that they had been given by the seed contacts. Once determined eligible, informed consent was obtained prior to commencing interviews. including informing potential participants about the sensitive nature of questions asked and their right to refuse to answer questions or to stop participation.

Only non-seed data were included this study [24]. The RDSAT software [25] was used to apply bootstrap simulation analysis (assuming Markov chain connections between recruiters and recruits) to estimate whether the sample is reflective of the underlying population on specified variables [26]. Results showed no significant bias on demographic and SS indicators (prevalence within the 95% confidence interval of the population level estimates); thus, data were not RDS-weighted in the analyses below.

Measures

Social Stability

The indicators applied here are adapted from German [11] to replicate and validate this definition of SS. Dichotomous variables were created based on the coding of BESURE-HET3 participant survey responses referencing the past 12-month period (details in supplemental materials): (1) housing stability (no homelessness); (2) legal stability (no detention in legal custody for 24 h or more); (3) residential stability (not moving residence); (4) income stability (annual income at or above $10,000, based on the NHBS pilot study where income less than $10,000 was associated with greater HIV risk) [27]; (5) employment stability (employed full or part-time); (6) relationship stability (last sexual partner considered main partner and has been with this partner for at least 12 months.)

Three composite SS measures were calculated: a summative score created by adding the responses to the six SS indicators (0–6 possible); ordinal SS categories of high (= 5–6 SS score), medium (= 3–4 SS score), and low (= 0–2 SS score); and a latent class model (see “Results”). Both simple logistic regression and latent class regression (LCR) were applied to evaluate whether different characterizations of SS were consistently associated with risk exposures.

Individual and Composite Measures of Risk Behaviors

Individual sexual, substance use, and violence risk variables were selected based on indicators from studies of heterosexual HIV risk including heightened sexual risk indicators [22, 28], substance use (drug) indicators associated with increased risk for HIV or overdose death [29], or indicators of violence based on contribution to HIV risk and SAVA syndemic [30, 31]. (Details in supplemental materials.) Each risk indicator was analyzed individually, then used to create dichotomous composite variables for the presence (yes or no) of each type of risk behavior (e.g., “any sex” risk, “any drug” risk, or “any violence” risk). An “any risk” composite variable was created to denote (yes or no) the presence of any of the eight risk indicators. Another set of composite variables was created denoting the presence of two or more types of risk behavior (sex-and-drug, sex-and-violence, drug-and-violence, and sex-drug-and-violence).

Sociodemographic Variables

Dichotomous demographic variables were created based on participant self-identification of gender (male, female), race (black, non-black), age (at or above median, below median), and level of education (at least high school/GED or no high school/GED). These were analyzed for covariance with risk variables.

Statistical Analysis

Using IBM SPSS version 24 [32], SS and risk behavior variables were analyzed for prevalence, co-occurrence, and bivariate relations. Next, composite risk variables were regressed on three variations of composite SS variables to assess whether the likelihood of risk exposure varied based on the amount or type of SS. First, simple logistic regression (SLR) was applied to assess the influence of SS as an ordinal variable on risk type. Next, 3-step LCR analysis was applied using Mplus 7 [33], regressing the odds of class membership based on the presence of the auxiliary risk variables. In step one of 3-step LCR, the likely number of latent classes is estimated for best model fit. In step two, one assigns individuals to their most likely class. In step three, each class is assigned a fixed logit probability (based on resulting logits of class membership in step one), so that the error (or misclassification) in class assignment is accounted for and the measurement model will not change excessively with addition of covariates in the third (regression) step [34]. Finally, one adds the risk variables to the regression equation (including the error term) to estimate whether the likelihood of class membership or class structure changes based on their presence.

Results

Sample Description

Non-seed BESURE-HET3 (n = 503) participant ages ranged from 18 to 60 years, with mean age of 38 (SD = 12.3, median = 37), and a bi-modal distribution including peaks at 27 and 49 (Table 1). Half of the sample identified as male (51%), most identified as black/African-American (88%), and 61% had completed 12th grade or GED.

Table 1.

Demographic and social stability characteristics of (non-seed) participants BESURE-HET3 (n = 503) Baltimore, MD (2013)

Demographic characteristics	n (%)
Age: mean = 37.9 (SD = 12.3) Median age (37 years) or older	257 (51)
Male	259 (51)
Black/African-American	444 (88)
12th grade/GED or higher	308 (61)
Housing stability: no homelessness in past 12 months	418 (83)
Legal stability: no incarceration in past 12 monthsa, g	404 (80)
Residential stability: no move in past 12 monthsa	341 (68)
Income stability: $10,000 or higher in past 12 monthse	273 (54)
Relationship stability: with main partner for > 12 monthsa	160 (32)
Employment stability: full-time or part-timea, e	123 (25)

Superscript letters indicate significant chi-square association with covariates: a, age; g, gender; e, high school education

Prevalence of Social Stability

The BESURE-HET3 sample reported more stability than instability with most reporting stability in housing (83%), legal (80%), residential (68%), income (54%), but less than half with relationship (32%), and employment (25%) stability in the past 12 months (Table 1). (Data were missing for 9 individuals on the income variable only. These individuals were assigned dichotomous values (0, 1) after imputation using Mplus missing data function.)

The mean SS score was 3.42 (SD = 1.2, mode = 4.0) out of maximum 6 points, with 18% of participants in high SS category (SS score 5–6 pts), 60% in medium (SS score 3–4 pts), and 22% in low category (SS score 0–2 pts). In analysis of covariates, completing high school/GED was significantly associated with the SS category (χ² = 8.07, p = 0.004) specifically, highest SS category vs. other categories in Bonferroni adjustment (alpha (0.05/6) = 0.008; adjusted χ² = 7.84, p = 0.005).

Using latent class analysis (LCA), underlying patterns of responses to each of the original six dichotomous SS indicators (no covariates) were analyzed to identify potential sub-groups within the population. A 3-class solution yielded better entropy (0.78) over the other classes. Criterion indices, Lo-Mendell-Rubin, and bootstrapped log likelihood ratio tests showed 3-class model to be significantly better fit than 2-class or 4-class solutions (Table 2). Based on guidance from Dziak and colleagues for appropriate sample size for latent class analyses using bootstrapped likelihood test, our size was sufficient to estimate the number of classes we proposed with the degree of variation present in SS indicators [35].

Table 2.

Model fit evaluation information: social stability latent classes BESURE-HET3 (n = 503)

N	LL1	Npar2	BIC3	ABIC4	LMR-RT5	BLRT6	Entropy
1-Cluster	− 1732.26	6	3507.84	3488.79	--	--	1
2-Cluster	− 1710.94	13	3502.74	3461.48	p = < 0.001	p = < 0.001	0.73
3-Cluster	− 1696.48	20	3517.37	3453.89	p = 0.018	p = < 0.001	0.78
4-Cluster	− 1689.89	27	3547.73	3462.03	p = 0.16	p = 0.6	0.75

¹Log-likelihood

²Number of parameters

³Bayesian Information Criterion (BIC)

⁴Adjusted Bayesian Information Criterion (ABIC)

The 3-class solution yielded a “high” stability class with 55% of participants (class members were likely to endorse stability in most indicators), a “moved” stability class with 25% of participants (class members likely to endorse stability but had moved at least once), and a “low” stability class with 20% of participants, and class members were less likely to endorse stability in most indicators. (Proportions are based on most likely class membership.) See class profiles and distribution in Fig. 1.

Fig. 1.

Profiles of latent 3-class solution: probability of endorsing social stability (SS) indicators. BESURE-HET3 2013 (n = 503).

Covariates of age (> median 37) OR = 1.76 (95%CI: 1.13, 2.77) and education (high school/GED completion) OR = 9.37 (95%CI = 2.75, 31.82) were significantly associated with the likelihood of being in higher SS class compared with lower.

Prevalence of Risk Behaviors

Almost three-fourths (74%) of the sample reported exposure to at least one of the eight risk indicators in the past 12 months (this exposure is represented as composite variable, “any risk,” n = 370). Prevalence of individual risk indicators range from 9% reporting forced sex and injection drug use (IDU) to 37% reporting multiple partners. (Prevalence and association with covariates are displayed in supplemental materials). Each individual risk indicator was significantly associated with each other risk indicator in bivariate analysis except for binge drinking and IDU. Aggregating the eight individual risk behaviors into three types (sex, drug, and violence), 53% reported any sex risk, 46% reported any drug risk, and 36% percent reported any violence risk. Figure 2 portrays the quantity of risk variable exposures (sex, drug, violence and co-occurring risks) and proportion of exposure by covariate (age, gender, education, and race/ethnicity) and chi-square results. Sexual risks are significantly associated with gender (χ² (1, n = 503) = 20.81; p < 0.01); more male respondents reported sexual risks. Drug risks were significantly associated with age (χ² (1, n = 503) = 10.31; p < 0.01); older participants reported more drug exposure, and gender (χ² (1, n = 503) = 5.43, p = 0.02); more males reporting drug risks). Violence is associated with age (χ² (1, n = 503) = 11.51, p < 0.01) with more younger participants reporting violence exposure.

Fig. 2.

Proportion of participants endorsing sexual, violence, or substance use risk or combination risk exposures by demographic covariates. BESURE-HET3 (n = 503) Baltimore, MD (2013). *Significant in chi-square analysis p < 0.05

Co-Occurring Risk

There was frequent co-occurrence of combination risk types: 30% reported one type of risk, 26% two types (sex-drug, sex-violence, drug-violence), and 18% reported all three types of risk. While there was low representation of non-black participants (n = 59), race was significantly associated with all combinations of risk, with non-black participants reporting fewer risk exposures.

Social Stability and Risk Behaviors

Regression Results

Regression analyses were conducted to estimate the influence of social stability on “any” sex, drug, and violence risk and co-occurring risks. A continuous social stability score variable was applied as covariate in relation to risk exposures in SLR, followed by regression with the SS categories. Finally, SS latent classes were analyzed in relation to risk exposures via latent class regression. Given higher prevalence and conceptually normative status, the higher stability group was denoted as reference in SLR and LCR.

Logistic Regression

Tables 3 and 4 depict, in descending order, the SLR results of models comparing representations of SS as summative score, SS categories, and SS latent classes (LCR) in relation to sex, drug, violence, and co-occurring risk variables. (For graphic depictions of the unadjusted odds ratios in forest plots, see supplemental material.)

Table 3.

Comparison of logistic regression results for social stability variables on sexual, drug, and violence risks. BESURE-HET3 (n = 503) Baltimore, MD (2013)

Composite risks	Any risks	Any sex risks	Any drug risks	Any violence risks
Social stability	AOR*	AOR*	AOR*	AOR*
Summative SS	1.27 (1.06, 1.51)	1.085 (0.93, 1.26)	1.45 (1.23, 1.70)	1.31 (1.12, 1.53)
Category (SLR)
Low vs. high (ref.)	3.26 (1.55, 6.85)	1.45 (0.82, 2.57)	3.45 (1.93, 6.21)	2.30 (1.28, 4.14)
Low vs. medium (ref.)	3.03 (1.62, 5.88)	1.56 (0.99, 2.44)	2.78 (1.75, 4.36)	2.08 (1.33, 3.23)
Medium vs. high (ref.)	1.05 (0.62, 1.76)	0.93 (0.59, 1.50)	1.29 (0.78, 2.11)	1.10 (0.66, 1.83)
Latent SS class (LCR)
Low vs. high (ref.)	5.52 (1.39, 21.76)	3.44 (1.07, 11.02)	4.91 (2.12, 11.47)	4.13 (1.7, 10.07)
Low vs. “moved”(ref.)	2.71 (0.63, 11.59)	2.87 (0.84, 9.78)	2.61 (1.08, 6.36)	2.81 (1.05, 7.46)
Moved vs. high (ref.)	2.04 (1.16, 3.56)	1.2 (0.74, 1.93)	1.88 (1.13, 3.13)	1.47 (0.87, 2.48)

SS, social stability

Italicized text = significant differences

*AOR, odds ratio adjusted for gender, race, age, and education

Table 4.

Comparison of logistic regression results of social stability on syndemic sexual, drug, and violence risks. BESURE-HET3 (n = 503) Baltimore, MD (2013)

Syndemic Risks	Sex-drug	Sex-violence	Drug-violence	Sex-drug-violence
Social stability measure	AOR*	AOR*	AOR*	AOR*
Summative SS (SLR)	1.31 (1.12, 1.53)	1.29 (1.09, 1.54)	1.39 (1.16, 1.66)	1.36 (1.15, 1.60)
SS category (SLR)
Low vs. high (ref.)	2.92 (1.58, 5.40)	2.21 (1.16, 4.24)	3.03 (1.52, 6.02)	2.69 (1.29, 5.59)
Low vs. medium (ref.)	2.58 (1.61, 4.14)	2.15 (1.32, 3.49)	2.27 (1.39, 3.70)	2.61 (1.53, 4.46)
Medium vs. high (ref.)	1.13 (0.66, 1.93)	1.03 (0.57, 1.86)	1.34 (0.71, 2.50)	1.03 (0.52, 2.04)
SS latent class (LCR)
Low vs. high (ref.)	6.79 (2.29, 20.09)	6.79 (1.9, 24.29)	4.97 (1.99, 12.43)	8.52 (2.94, 24.78)
Low vs. moved (ref.)	4.2 (1.57, 11.25)	6.16 (1.34, 28.22)	3.59 (1.23, 10.38)	7.9 (2.2, 28.22)
Moved vs. high (ref.)	1.62 (0.88, 2.97)	1.1 (0.56, 2.18)	1.38 (0.73, 2.64)	1.08 (0.48, 2.41)

Italicized text = significant differences

*AOR, odds ratio adjusted for gender, race, age, and education

“Any Risk” Exposures

In SLR, a one-unit decrease in SS score resulted in expected 1.27 (95%CI = 1.06, 1.51) greater odds of experiencing any risk in a model adjusted for covariates (unadjusted results were essentially identical). Alternatively, incremental increase in stability was associated with ~ 25% reduced odds of experiencing any risk.

For sub-groups of stability, the likelihood of risk exposure was consistently associated with belonging to a lower vs. higher SS grouping (whether category or latent class). Compared with the highest category, the lowest category was three times as likely to report any risk behaviors (adjusted odds ratio (AOR) = 3.26; 95%CI = 1.55, 6.85). Similarly, in LCR, the low vs. high SS class was associated with greater odds of any risk (AOR = 5.52; 95%CI = 1.39, 21.76). (Wide confidence intervals in LCR are likely due to sample size and the inclusion of an error term (potential error in class assignment) perhaps contributing to uncertainty in estimate of relationship. The guidance for adequate sample size for power in LCR analysis is limited; however, given this sample’s distribution of outcomes, a sample size closer to 600 would yield more accurate estimates of the effect of SS exposure for logistic regression which likely holds true for LCR as well (per Dartmouth Power Calculator, www.dartmouth.edu/~eugened/power-samplesize.php).

Being in the low vs. medium stability category was predictive of risk exposure; AOR for low vs. medium for any risk was 3.03 (95%CI = 1.62, 5.88). Therefore, accumulating more stability at the lower end of the SS scale, going from low category (SS score of 0–2) to medium category (SS score of 3–4) could be protective, but at higher levels of stability, going from the medium to the high stability category did not apparently convey significant reduced risk exposure. In contrast, in LCR, those presumed to be in the “moved” latent class were more likely to report any risk than those in the high stability class (AOR = 2.04; 95%CI = 1.16, 3.56.) This relationship was mostly due to increased odds of substance use for those in “moved” vs. high latent class (AOR = 1.88; 95%CI = 1.13, 3.13).

Co-Occurring Risk Exposures

Continuous. A one-unit decrease in SS score resulted in greater odds of co-occurring risk with greatest magnitude of risk associated with drug-violence (OR = 1.39; 95% CI = 1.16, 1.66) and sex-drug-violence exposures (OR = 1.36, 95%CI = 1.15, 1.60) (Table 4). Similarly, being in the lowest SS grouping compared with highest or the medium groupings was significantly associated with reporting co-occurring risk whether categorical or latent characterization (Table 4). Categorial. The likelihood of co-occurring risk was particularly great for those in low vs. high category in relation to drug-violence risk (AOR 3.03; 95%CI = 1.52, 6.02). Latent. The magnitude of odds were greater for each co-occurring relationship for the latent classes compared with the categorical representations of SS. In particular, the greatest magnitude AOR was found for the lowest vs. highest latent SS class in relation to the sex-drug-violence variable (AOR = 8.5; 95%CI = 2.94,24.78). (Forest plot depictions of unadjusted odds ratios in supplemental materials). This wide confidence interval is likely due to insufficient sample size and the error estimate included in LCR. To achieve accurate estimate of the effect of stability exposure with only 18% of sample reporting all three (SDV) risk types, a sample size closer to 700 would be needed (Dartmouth Power Calculator, www.dartmouth.edu/~eugened/power-samplesize.php).

Discussion

This study makes several contributions to research in social determinants of health and syndemics. First, social stability, as defined by German [15], worked well to characterize social context in a new population. Second, the risk exposures/behaviors were explored as co-occurring in a new population: a social network that is under-explored and more generalized (men and women who engage in heterosexual sex) compared with other studies of syndemics. Third, this study offers further evidence that the theoretically related concepts of sexual risk behavior, problematic drug use, and exposure to violence may be considered syndemic exposures contributing to syndemic outcomes based on their common co-occurrence, significant bivariate associations, and the greater likelihood of occurrence and co-occurrence in adverse social conditions (lower social stability). Finally, whether operationalized as a continuous, a categorial, or latent variable, social stability reliably predicted multiple types of risk. Therefore, this study contributes evidence to the utility of social stability as a meaningful and practical representation of inter-connected social conditions to be considered in future studies.

Quantity of Stability Matters

Lower SS was consistently related to increased likelihood of reporting risk exposures. The BESURE-HET3 sample has at least double the proportion of persons experiencing poverty, incarceration, unemployment, and homelessness relative to the general population of Baltimore City. Even within this socio-economically vulnerable population, accumulating a small amount of stability was protective. Ostensibly, if a participant accumulated one to two more indicators of stability, they would have reduced likelihood of experiencing “any” and combinations of risk exposures.

Quality of Stability Also Matters

The presence of a latent class that was identified by the indicator of moving in the past year is a further unique contribution of the analysis. The relationship of the act of moving households to overall stability merits greater conceptual exploration independent of relation to risk exposures. Further, in our era of substance use epidemics, moving may be a contributing destabilizing influence. For those in the “moved” vs. high stability class, the increased likelihood of reporting of problematic drug use would be missed if SS was treated only as a summative variable. Substance use has been associated with increased residential mobility, especially among youth [36, 37]. One study even describes a profile of persons experiencing homelessness as the transient substance user, someone who had not lived long in the location of study was more likely to engage in substance use [38]. Furthermore, among an already substance using population, transient individuals were more likely to engage in HIV-risk activities such as needle sharing or frequenting a shooting gallery [39]. Whether substance use precedes residential mobility or whether the stresses of increased mobility lead to increased substance use needs further exploration, ideally in longitudinal studies.

An additional contribution of the study is that using observed, additive variable approaches in simple logistic regression could underestimate the magnitude of risk relationships if the latent characterization of SS was not examined: subtypes of SS may be driving particular relationships and may be missed if not properly evaluated. Feingold et al. [40] state this misspecification of true associations is due to reduction of data to approximated observed variables (in SLR) instead of considering actual individual data pattern clusters (with internalized error terms) related to covariates or distal outcomes (in LCR).

Limitations

This is a cross-sectional study without the benefit of longitudinal data to determine the timing of stability or risk. However, social determinants usually precede behaviors in life course studies. Another limitation is the failure to explore interactions or potential synergistic relationships of risk behaviors and SS to distal disease outcomes such as HIV or hepatitis C due to low prevalence or missing data for these conditions. Future (longitudinal) studies with larger samples should include analysis of syndemic relations between upstream influences and downstream outcomes. Furthermore, we did not consider multi-level influences of social network, neighborhood, or policy environment. However, BESURE-HET3 represents a social network at risk for heterosexual HIV based on their relational and residential context [22]. Thus, these results may be valid for similar networks as, per RDS analysis, the sample and population level markers of SS were not significantly different. Lastly, this population may not be representative of different socio-structural or geographic contexts. Important next steps would be to investigate this definition of SS among similar and diverse populations (i.e., non-urban) and in relation to different risk exposures.

Conclusions

Just a small increase in social stability was associated with significantly reduced odds of any and combination risk behaviors. In addition, the type of stability, especially instability of residence, may be important in the context of likelihood of risk behaviors. Therefore, assessing and addressing the level but also the nature of SS in clinical practice, public health agencies, and in policy considerations could lead to information that can inform interventions to reduce risk exposures/behaviors and improve health outcomes. Research. These results contribute to the theoretical and quantitative justification for using SS as a partial representation of social context in future scholarship on syndemics. Applying this definition of SS in studies of different populations will be important to further establish its validity. Clinical Practice. Screening for social determinants of health is currently recommended for inclusion in electronic health records by the Centers for Medicare and Medicaid Systems [41]. Important next steps would be to integrate the construct of SS into existing screening tools, especially incorporating its main tenets: (1) screening for more than one domain, (2) assessing duration of stability in each domain, and (3) quantifying responses and characterizing the quality/patterns of responses. This practice can help healthcare teams identify and prioritize interventions at the individual level. Urban Health. The construct of social stability, as applied in this and other studies, has demonstrated that social conditions operate in conjunction. In an urban environment, policy makers and public health practitioners must consider interventions that impact SS as a whole. Of note, just completing 12th grade or GED was the greatest predictor of potential membership in the high stability class. The quality of education received by respondents is not known. However, a low hanging fruit for increasing SS may be for healthcare and other professionals to engage in advocacy for evidence-based upstream interventions that create conditions for increased high school completion. One promising initiative is Communities In Schools bridging schools and communities to achieve graduation rates of 90% for students most at risk of dropout [42].

Yet, one-off interventions on individual SS indicators are not likely to change the equation between social conditions and health/behaviors. For example, initiatives that only provide housing have mixed outcomes, while permanent supportive housing initiatives—with holistic services addressing other needs—reduce costs and improve health outcomes [43]. However, some hospital systems are finding that just paying rent is both cost saving and can reduce high healthcare utilization [44] Therefore, a multi-sector approach ameliorating broader urban social contexts is more likely to achieve widespread and long-term increases in SS, thereby reducing the conditions that favor syndemics and increasing conditions that favor individual and population health.