Master Thesis : Is neighborhood socioeconomic status associated with health outcomes? An empirical study of a Dutch surgical cohort
Ramesh Marapin (2317001) MSc Business Administration - Health
Faculty of Economics and Business, University of Groningen First supervisor: prof. dr. J.O. Mierau
Second supervisor: prof. dr. M.J. Postma Word count: 10611
___________________________________________________________________________
Abstract
Socioeconomic status (SES) has been empirically associated with myriad health outcomes.
Namely, in low SES populations health outcomes are relatively worse. While the effect of individual SES on health outcomes is well known, the effect of SES at the neighborhood level (NSES) remains relatively unexplored. This thesis investigates the association between NSES and health outcomes to garner new insights for policymakers in their mission to ameliorate health disparities. Using data from 585 patients, I analyze the association between NSES and both the odds of a favorable postsurgical (textbook) outcome, and pre-existent comorbidity.
The results of this thesis do not provide evidence of a significant association between NSES and the studied health outcomes. Moreover, I discover higher age to be significantly associated with lower odds of a textbook outcome, while I discover both higher age and the male gender to be significantly associated with higher pre-existent comorbidity. The findings in this thesis are in contrast to previous literature in this field, which report an inverse association between NSES and health outcomes. This discrepancy could be due to hospital-level factors (i.e., hospital volume), which potentially modify this association. Strategies to reduce health disparities should be tailored to the individuals’ socioeconomic context and characteristics.
These strategies should focus on lifestyle-related risk factors, health literacy and healthcare engagement, as well as the neighborhood context, such as educational and recreational facilities. Ultimately, concerted efforts addressing these issues are required in order to structurally mitigate the socioeconomic gradient in health outcomes in the Netherlands.
___________________________________________________________________________
2
Abstract
Socioeconomic status (SES) has been empirically associated with myriad health outcomes.
Namely, in low SES populations health outcomes are relatively worse. While the effect of individual SES on health outcomes is well known, the effect of SES at the neighborhood level (NSES) remains relatively unexplored. This thesis investigates the association between NSES and health outcomes to garner new insights for policymakers in their mission to ameliorate health disparities. Using data from 585 patients, I analyze the association between NSES and both the odds of a favorable postsurgical (textbook) outcome, and pre-existent comorbidity.
The results of this thesis do not provide evidence of a significant association between NSES and the studied health outcomes. Moreover, I discover higher age to be significantly associated with lower odds of a textbook outcome, while I discover both higher age and the male gender to be significantly associated with higher pre-existent comorbidity. The findings in this thesis are in contrast to previous literature in this field, which report an inverse association between NSES and health outcomes. This discrepancy could be due to hospital-level factors (i.e., hospital volume), which potentially modify this association. Strategies to reduce health disparities should be tailored to the individuals’ socioeconomic context and characteristics.
These strategies should focus on lifestyle-related risk factors, health literacy and healthcare engagement, as well as the neighborhood context, such as educational and recreational facilities. Ultimately, concerted efforts addressing these issues are required in order to structurally mitigate the socioeconomic gradient in health outcomes in the Netherlands.
Keywords: neighborhood socioeconomic status, postoperative outcomes, surgery, Dutch cohort JEL classification: I10, I14
3
Table of Contents
1. Introduction ... 4
2. Literature Review ... 6
2.1 Components of socioeconomic status ... 7
2.2 Pathways and mechanisms ... 8
2.3 Neighborhood socioeconomic status and health disparities ... 9
3. Data and methods ... 11
3.1 Study design and cohort ... 11
3.2 Socioeconomic Analysis ... 11
3.3 Dependent variables of interest ... 13
3.5 Statistical Analysis ... 14
4. Results... 15
4.1 Descriptive statistics ... 15
4.2 The socioeconomic gradient in postoperative outcomes ... 16
4.3 The socioeconomic gradient in comorbidity ... 17
4.4 Statistical analysis ... 18
5. Discussion ... 22
6. Conclusion and implications... 25
7. References ... 27
4
1. Introduction
Currently, an empirically supported association exists between socioeconomic status (SES) and several health outcomes and health behaviors (Adler et al., 1993; Duncan et al., 2002; Feldman et al., 1989; Venkataramani et al., 2016). While no formal definition exists, SES can be considered as a composite construct which commonly includes measures such as occupation, educational status, household income, and so on (Oakes & Rossi, 2003). In low SES contexts, health outcomes commonly manifest as higher morbidity and mortality, and reduced access to health care (Ancona et al., 2004; Picciotto et al., 2006; Van Lenthe & Mackenbach, 2006). For example, cardiac related morbidity and mortality figures are systematically higher among people with lower SES, which is the result of a higher prevalence of cardiac risk factors in this population (Schröder et al., 2016). Therefore, it is evident that socioeconomic disparities are fundamental causes of societal health inequalities and are items which rate high on the political agenda of policymakers.
Despite considerable attention given to the problem of health inequalities since the 1980s (Murray et al., 1999), striking differences in health unfortunately still exist between and within countries today (Braveman et al., 2000). Interestingly, in the past few decades there has been growing interest in the adverse health effects of living in neighborhoods characterized by poor socioeconomic conditions, i.e., low neighborhood socioeconomic status (NSES; Anderson et al., 1997). There is an increasing body of evidence supporting the hypothesis that living in such areas has adverse effects on health outcomes, regardless of individual SES levels (Curtis &
Jones, 1998; Diez-Roux, 1998; Michael G. Marmot, 1998). This may have important implications for public health policymakers. Namely, this evidence suggests that health policies should not only focus on individuals, but to also consider the socioeconomic environment in which they live as a key determinant influencing the health of the population.
The economic ramifications associated with poor health outcomes are salient through multiple
channels, including increased healthcare costs and productive time losses resulting from illness
(McIntyre et al., 2006; Muka et al., 2015). To this end, the economic costs of health inequalities
in the European Union have been estimated at 1 trillion € per year, which is 9.5% of its GDP
(Mackenbach et al., 2011). In addition to this societal impact, other important reasons to
address the socioeconomic gradient in population’s health include moral concerns and
undermining notions of justice and fairness (Whitehead, 1992). Particularly, health disparities
5
are morally and ethically problematic as all individuals should have equal capability to be healthy, as health is crucial for individual agency – the ability to live a life we value. Taking this multifaceted impact into account, the association between NSES and health outcomes therefore warrants further investigation to better understand the underlying pathways.
Several papers discuss the adverse impact of SES on health outcomes in the context of postsurgical outcomes. While surgical interventions often serve as life-saving alternatives, the outcomes of these interventions are not necessarily equally distributed among the patient population. In fact, there is increasing recognition that a low SES leads to worse postsurgical outcomes. For example, Birkmeyer and colleagues found that patients with lower SES have higher incidences of postoperative mortality compared to patients with higher SES across a wide range of surgical interventions (Birkmeyer et al., 2008). In another study conducted in patients post-cardiovascular surgery, socioeconomically disadvantaged people seem particularly vulnerable to mortality after surgery (Agabiti et al., 2008). Finally, in a Dutch population based study investigating patients with pancreatic cancer, low SES was found to be an independent risk factor for poor survival (Van Roest et al., 2016).
Although there is ample evidence on the impact of individual socioeconomic factors on health outcomes in surgical populations, the impact of NSES in these populations is relatively unexplored, even more so in the European context. More importantly, the mechanisms underlying this association remain poorly understood. Policymakers are thus faced with the daunting task of developing appropriate strategies and policies to mitigate these societal health inequalities, essentially while lacking the necessary framework and roadmap to do so. By addressing this literature gap, with this paper I aim to contribute to the conceptualization of socioeconomic disparities in health. Currently, the literature consists mainly of studies addressing the impact of NSES on surgical populations outside of Europe, warranting further research on the impact of NSES on a European surgical context. To the best of my knowledge, this paper is one of, if not the first, to investigate the association between NSES and postoperative outcomes in a European healthcare setting.
To this end, the theoretical contributions of my paper are twofold. First, I seek to fill the gap in the theory of socioeconomic disparities in health by empirically investigating the relationship between NSES and postoperative outcomes in patients in a European context.
Second, I assess whether there is a relationship between NSES and patients’ pre-existing
6
comorbidity scores. With this study, I thus contribute to the conceptualization of socioeconomic disparities in health with novel insights on the effects of neighborhood characteristics on health outcomes in a European context. Due to the sociocultural differences between Europe and the rest of the world, garnering insights from the European context is fundamental to policymaker efforts of devising strategies aimed at their societal needs.
Furthermore, by drawing attention to the health-related outcomes associated with the social structure and ecology of neighborhoods, innovative approaches to society level interventions can be fostered.
Based on the literature gap I identified in the Introduction and the aims of my research, I have formulated the following two research questions. First, the scant empirical evidence of the association between patients’ NSES and postoperative outcomes in a European context has led to my first research question:
RQ1: What is the relationship between patients’ neighborhood socioeconomic status and
postoperative outcomes in a Dutch surgical population?
Second, my aim is to investigate whether there is a relationship between NSES and patients’
pre-existing comorbidity scores. This leads to my second research question:
RQ2: What
is the relationship between patients’ neighborhood socioeconomic status and their pre-existing comorbidity scores in a Dutch surgical population?
2. Literature Review
In this section, I will first briefly describe the components that traditionally define socioeconomic status. Next, I will describe the pathways by which SES influences health.
These pathways contain perspectives from both the individual and neighborhood level, in order
to provide a comprehensive understanding of the role of the social environment in relation to
health. Omitting either construct will result in an incomplete and possibly biased understanding
of this pathway (Pickett & Pearl, 2001). Afterwards, I will shift the focus to the association
between NSES and disparate health outcomes. Finally, I will appraise the current evidence on
the association between SES and postoperative outcomes.
7 2.1 Components of socioeconomic status
Before characterizing the components that make up socioeconomic status, it is necessary to provide a definition of health disparities and how this relates to socioeconomic disparities.
While a general consensus is lacking, health disparities can be conceptualized as a particular type of health difference. Namely, health disparities are differences that exist among specific population groups in the attainment of maximum health potential. These disparities can be measured by differences in incidence, prevalence, mortality, morbidity, and other adverse health conditions. In this regard it is important to consider that especially disadvantaged social groups (e.g. the poor, racial/ethnic minorities, or other groups) who have continually experienced social disadvantage or discrimination, systematically experience worse health outcomes compared to more advantaged social groups (Braveman, 2006). As such, while health disparities result from differences due to biological and social disparities, here I focus on the latter as the effect is greater but also avoidable and particularly unjust as seen from an ethical perspective.
Conventionally, SES has been defined by level of education, income, and occupation. Each of these components exhibit different relationships to various health outcomes, and thus require different appraisal strategies and policies to address underlying disparities. The first component, education, furnishes important knowledge and life skills that allow better-educated persons to gain more access to resources to promote health (Ross & Chia-Ling Wu, 1995). To this end, policies focusing on the improvement of education offer several benefits, ranging from improved population health to collateral virtues such as increased productivity and decreased healthcare costs (Adler & Newman, 2002).
The second component comprises income, where the distribution of income within countries and states has been linked to mortality rates (Kaplan et al., 1996). While still a subject of contention, one explication for this phenomenon is that underinvestment in public goods and welfare and the experience of inequality are both more pronounced in more stratified communities. In turn, this can affect health outcomes in these communities. Importantly, this phenomenon is not only restricted to incomes below the poverty level (Backlund et al., 1999).
However, this burden is particularly salient for individuals living in poverty.
8
The final component encompasses occupation. One way to conceptualize occupation is whether or not somebody is employed. Occupation is a pertinent component to study, as employed persons have better health compared to the unemployed (Ross & Mirowsky, 1995).
Furthermore, examining the employed shows that there are differences between occupations, such as job characteristics and qualifications. Crucially, these different indicators reflecting occupational status are linked to mortality risk (Gregorio et al., 1997). This link also holds for non-mortality related measures, such as the incidence of coronary heart disease, which has been shown in the landmark Whitehall study in British civil servants (Marmot et al., 1997).
2.2 Pathways and mechanisms
Recently, many scholars have attempted to explain the mechanisms by which socioeconomic status influences health outcomes. In their work, Adler and Newman describe pathways and mechanisms by which SES influences health (Adler & Newman, 2002). First, SES has been shown to influence the environmental exposure pathway. Specifically, exposure to noxious environmental agents (e.g., asbestos) varies with SES. Occupying a lower position on the socioeconomic ladder is associated with a higher probability of working and living in worse physical environments. Additionally, less affluent neighborhoods are disproportionately located next to highways and industrialized land, as land there is less expensive.
Furthermore, housing quality is worse for families with low SES. Consequently, inhabitans from poor households demonstrate significantly higher rates of high blood lead levels compared with their counterparts from higher income households (Adler & Newman, 2002).
While the underlying mechanisms remain unclear, possible explanations for this include a decline in housekeeping as a result of increased adult work hours, and deteriorating housing stock (Cookson & Moffatt, 1997).
Socioeconomic status also exerts its effect on the social environment pathway. Lack of social
engagement and isolation in social networks are strong predictors of health (Adler & Newman,
2002). For example, individuals living in social isolation have a relative risk of mortality
ranging between 1.9 to 5 times higher compared to individuals with better social cohesion
(Berkman & Glass, 2000). Additionally, disease risk is also influenced by patterns of social
interaction. Architectural features of communities and institutions have also been found to
potentially promote social integration, which in turn improves health.
9
Moreover, access to health care and the quality of health care received can vary by SES. In the United States, the majority of the uninsured reflect low-income families (Monheit & Vistnes, 2000). This has important ramifications, as people who lack insurance receive less medical care, including treatment, compared to those with medical insurance (Hafner-Eaton, 1994).
This problem also extends to countries that provide universal coverage, as people with lower education and less income do not necessarily use health care services in the same way that their wealthier and better educated counterparts do. For example, in a study done in Canada it was found that citizens with lower SES were less likely to get specialty care when needed (Dunlop et al., 2000). Another example where universal coverage insufficiently offsets social disparities can be seen in England (Smith et al., 1990). Thus, this underscores the notion that higher disease incidences and toxic exposure remain dominant forces by which SES affects health.
2.3 Neighborhood socioeconomic status and health disparities
The influence of neighborhoods on the health of its inhabitants has been a popular item of research in the first decade of this century (Li & Chuang, 2009). Accordingly, it is increasingly recognized that individual health is not solely influenced by individual characteristics, but also by the neighborhood socioeconomic context to which one belongs (Berkman & Glass, 2000).
Typically, these neighborhood contextual variables are either acquired at the individual level, e.g., median household income, or are central and exclusive to the neighborhood (Diez-Roux, 1998). An example of the latter is the number of recreational and educational facilities in a neighborhood (Macintyre et al., 1993). In this regard, NSES influences health opportunities and outcomes through its effect on individuals' achieved income, education, and occupation (Diez-Roux, 2001). For example, differences within and between neighborhoods could be in terms of access to (health) resources, which includes services that promote healthy behavior, and how groups and peers within neighborhoods influence each other’s behavior.
Also, as described previously, less affluent neighborhoods are disproportionately negatively
affected by their environment (e.g., higher concentrations of harmful air pollutants and worse
housing conditions), which leads to worse health outcomes for its inhabitants (Brook et al.,
2010; Hajat et al., 2013; Roy et al., 2014). Thus, when considering how to reduce health
disparities, the context in which these risk factors originate needs to be better understood.
10
Ample evidence has demonstrated an association between NSES and health. Importantly, however, the direction of this association remains inconsistent, with several studies reporting either an association or no association at all between NSES and health. Indeed, living in neighborhoods with lower NSES has been shown to be associated with worse health and patient reported outcomes (Al Adas et al., 2019; Ana V. Diez Roux & Mair, 2010; Jerath et al., 2020a;
Jones et al., 2019; Patrick et al., 2020; Pickett & Pearl, 2001). Furthermore, findings from a Dutch study pointed to a higher prevalence of social disintegration, poor housing conditions, adverse psychologic characteristics, and unhealthy behaviors in neighborhoods with lower NSES (Bosma et al., 2001).
On the other hand, several studies have found no association between living in socioeconomically deprived neighborhood and health outcomes (Obeng-Gyasi et al., 2020;
Powers et al., 2019; Zarzaur et al., 2010). Potential reasons cited for this discrepancy revolve around the fact that high-volume centers and standardized pathways might overcome the adverse health outcomes due to socioeconomic disparities. However, as underscored by Pickett and Pearl 2001, the evidence for neighborhood effects on health is fairly consistent. Thus, the main takeaway here is that studies in various surgical patient populations have demonstrated a significant correlation between SES and health outcomes, however, this association is not always found.
Based on what I discuss above, I expect to find a similar effect for NSES with regards to its association with health outcomes. Specifically, I hypothesize:
H1a: The relationship between neighborhood socioeconomic status and adverse postoperative
outcomes is inverse (negative correlation coefficient).
H2a: The relationship between neighborhood socioeconomic status and pre-existing
comorbidity scores is inverse (negative correlation coefficient).
11
3. Data and methods
3.1 Study design and cohort
In this investigation, I perform a retrospective cohort analysis of patients who underwent hepatopancreatobilliary (HPB) surgery. The data I use in this study was retrieved from the Managed Clinical Network, which is a database of HPB surgery from the northeastern part of the Netherlands. This network consists of five collaborating hospitals (University Medical Center Groningen [UMCG], Medisch Centrum Leeuwarden [MCL], Isala Zwolle, Medisch Spectrum Twente [MST], and Tjongerschans Hospital). This network contains data from approximately 3000 patients, who were operated between 2013-2017. Due to feasibility and ethical reasons related to accessing this data, I only use data from the UMCG for this analysis.
This dataset consists in total of 585 patients.
The protected patient electronic health records contain level 6 postal code data, which were collected in collaboration with a researcher at the UMCG who has access to these records.
These postal codes were used determine the NSES for each patient, as I explain more in depth below. Patients were included if they underwent an oncologic HPB procedure in one of the MCN affiliated centers between 2013 and 2017, and if there was sufficient data for analysis.
Patients who were under the age of 18 (on the date of the surgery), underwent diagnostic laparotomy or laparoscopy without resection, registered an objection to scientific research, and had no available postal code data were excluded from this analysis. In the case of missing data, these were obtained from the respective patients’ electronic health records.
For patients included within the UMCG, consent for medical research was checked through means of the “no objection” register. As this is a retrospective cohort study with the primary goal of improving the quality of patient care and includes a large proportion of deceased patients, the Medical Ethics committee (METc) of the UMCG deemed this to be sufficient. For this reason, I did not actively seek consent from patients to include them in this study.
3.2 Socioeconomic Analysis
For this analysis, I use a composite index measure for socioeconomic status, where I aggregate
several variables related to socioeconomic status at the neighbourhood level in order to derive
a unidimensional measure of NSES. I opted for this approach, since single SES related
12
measures (e.g., income and level of education) may not be sufficient or relevant to adequately differentiate between different levels of household SES. Furthermore, using myriad separate single measures may lead to collinearity and cumbersome results (Pickett & Pearl, 2001).
I obtained data on the variables needed to derive the NSES from Statistics Netherlands (in Dutch: Centraal Bureau voor de Statistiek), a Dutch governmental institution that gathers statistical information about the Netherlands. Statistical Netherlands provides SES related variables at the neighborhood level, which was matched to patients using their level 6 postal code data retrieved from their electronic health records.
Using the socioeconomic status related variables, I applied the principal components analysis (PCA) technique to derive a NSES index. Principal component analysis is a dimensionality- reduction statistical method that reduces the dimensionality of large data sets, by transforming a large set of variables into a smaller, composite variables. Essentially, PCA creates non- correlated linear combinations of the variables with maximal variance, which allows for the best contrast between statistical units. Principal component analysis was run using the Rstudio statistical software package (Version 1.2.5042). In order to overcome missing data, I used the NIPALS (Nonlinear Iterative Partial Least Squares) PCA algorithm, as this allowed me to include more neighbourhoods in this analysis.
The data used for this study was collected in 2016 by Statistics Netherlands, this
Neighbourhood dataset contains data from 12,822 neighbourhoods. Of these, 1,318 have no
socioeconomic data as a result of having low numbers of inhabitants. This constituted 0.1% of
the total dataset and was removed from the data. All other neighbourhoods were included in
the analysis. The variables used for the PCA were neighbourhood income statistics (percentage
people in lower 2 quintiles of national income, percentage people in highest quintile), social
welfare reliance (percentage of people on benefits, disability benefits, or unemployment
benefits), and housing market characteristics (average estimated housing-price, percentage
owner-occupied, percentage council housing). The scores on the first principal component
(factor) defined the NSES score. Following the method of Asaria et al. (2016), I ordered the
neighborhoods by NSES score and divided them into roughly equal-sized quintiles (with
NSES1 being the most deprived neighborhoods and NSES5 the most affluent; NSES3 was
chosen as the reference category).
13 3.3 Dependent variables of interest
In this study, I use two health indicators, namely postoperative outcomes and pre-existing comorbidity scores (CCI). First, postoperative outcomes were determined using variables registered in the Dutch Pancreatic Cancer Audit (DPCA) and the Dutch Hepatobiliary Audit (DHBA). These two databases belong to the Dutch Institute for Clinical Auditing (DICA), which acts as a national facilitating body for cross center quality assessment research. The DHBA and DPCA encompass patients undergoing oncologic liver and pancreas resections.
These postoperative outcomes were subsequently classified into “textbook” (TO) and “non- textbook” (NTO) clinical outcomes.
Textbook outcome is a theoretically grounded composite measure reflecting the “ideal”
surgical outcome, which consists of factors such as postoperative complications, no readmission to the hospital within 30 days, discharge from the hospital within two weeks, and no mortality. There is no universal definition of a “normal” postoperative course, and many classifications have been proposed which all remain subjective and significantly overlapping between other classifications. Therefore, I operationalized a “normal” postoperative course using the “textbook outcome” definition which was forged trough an international expert consensus (Van Roessel et al., 2020). Postoperative complications were scored in accordance with the Clavien Dindo classification, which classifies complications based on severity into one of five predetermined categories.
Second, patients' pre-existing comorbidities were scored in accordance with the Charlson Comorbidity Index (CCI). The CCI consists of 19 predetermined comorbidities, each assigned a score (1-3) based on their severity. Furthermore, an additional point is added for each decade above the age of 40.
3.4 Contextual factors of interest
In this study, I include five potential explanatory contextual (confounding) factors in the
analysis between NSES and postoperative outcomes. The first two variables I included were
age and body mass index (BMI), as both these variables have been shown to be independent
predictors of complications and worse outcomes in general surgery patients (Gajdos et al.,
2013; Yanquez et al., 2013). This data was collected from the electronic patient records.
14
Furthermore, I included gender and patient’s diabetic status as covariates in this analysis, as these have also been shown to be independently associated with worse postoperative outcomes (Browne et al., 2007; Komatsu et al., 2020; Yang et al., 2014; Zimmerman et al., 2011). Finally, I included the CCI score as a covariate in the analysis of NSES association with textbook outcome, as pre-existing comorbidity is associated with postoperative outcomes.
3.5 Statistical Analysis
Health and contextual characteristics of NSES quintiles were analyzed by descriptive statistics.
In order to understand the association between NSES and postoperative outcomes in a Dutch surgical cohort, I use a logit regression analysis. In this analysis, I aim to assess the role of the explanatory predictor and contextual variables of interest in the variation of TO. As such, the model I use for this analysis takes the following form:
Pr[TO=1|𝑥𝑖]=Λ(t)(𝛽0+ 𝛽1NSES𝑖+𝛽2𝑎𝑔𝑒𝑖+𝛽3diabetes𝑖+𝛽4gender𝑖+𝛽5BMI𝑖+𝛽6CCI𝑖+𝜀i)
where i represents the individual and Λ(t)=𝑒𝑡/(1+𝑒𝑡) is the logistic function with values ranging from zero to one. “TO” is a dummy variable indicating if the individual i has a textbook outcome or not.
Furthermore, a multiple linear regression model was used to assess the role of the predictor and contextual variables in the variation of CCI. The linear model used for this analysis takes the following form:
𝑦 =
𝛽0+ 𝛽1NSES𝑖+𝛽2𝑎𝑔𝑒𝑖+𝛽3diabetes𝑖+𝛽4gender𝑖+𝛽5BMI𝑖+𝜀iwhere i represents the individual and 𝑦 represents the linear outcome CCI.
15
4. Results
4.1 Descriptive statistics
The descriptive statistics on neighborhoods characteristics by quintiles of neighborhood socioeconomic status is shown in Table 1 below. As can be observed, the majority of participants had a textbook outcome after surgery (62.05%), i.e., a “positive” postoperative outcome. This was also the case for each NSES quintile separately, where the lowest ranked quintile (Q1) had the highest rates of textbook outcomes, and the fourth quintile (Q4) had the lowest rates of textbook outcomes. The second outcome variable of interest, the Charlson Comorbidity Index (pre-existent comorbidity), had a mean score of 7.00 (± 3.69). Furthermore, the mean NSES was -0.0049 (± 0.0079). The majority of the sample is male; the first NSES quintile is the only subsample of which the majority gender was female. Regarding diabetes status, 15.73% of the sample has diabetes; the highest prevalence of diabetes is in the lowest NSES quintile (Q1), namely 20.51%. With regards to nutritional status, the average BMI of the population was 26.50, which falls under the category of “overweight”. More information on the contextual variables included in this analysis can be found below.
Table 1. Descriptive statistics on neighborhoods characteristics by neighborhood socioeconomic status quintiles.
NSES-Q1 NSES-Q2 NSES-Q3 NSES-Q4 NSES-Q5 Total/Mean
Patients
117 117 114 120 117 585
Age (SD)
65.06 (11.08)
64.36 (10.67)
64.13 (9.92)
64.06 (10.34)
61.06 (13.44)
64.00 (11.20)
Gender: % male46.15 50.43 51.75 50.83 53.86 50.60
TO, yes (%)65.81 64.10 58.77 58.33 63.25 62.05
CCI (SD)6.61
(3.69)
6.92 (3.44)
7.04 (3.59)
6.81 (3.63)
7.12 (3.97)
6.90 (0.15)
NSES (SD)-0.016
(0.0037)
-0.0089 (0.0014)
-0.0050 (0.0013)
-0.0011 (0.0013)
0.0063 (0.0040)
-0.0049 (0.0079)
Diabetes, yes (%) 20.5116.24 16.67 14.67 11.11 15.73
BMI (SD)26.76
(5.42)
26.93 (5.85)
26.26 (4.46)
26.43 (4.96)
25.93 (4.26)
26.50
(4.80)
Abbreviations: BMI: body mass index; CCI: Charlson Comorbidity Index; NSES: neighborhood socioeconomic status; SD: standard deviation; Q= quintile.16
4.2 The socioeconomic gradient in postoperative outcomes
The neighborhood socioeconomic gradient in postoperative outcomes can be conceptualized in this paper as the inverse correlation between neighborhood socioeconomic status and adverse postoperative outcomes, the latter of which is assessed using the outcome measure textbook outcome. As can be seen in Table 1, an initial descriptive analysis of the whole population data fails to confirm the hypothesis that there indeed exists an inverse relationship between NSES and adverse postoperative outcomes (H1a). Namely, the highest rate of TO (i.e., “positive postoperative outcomes) is seen in the lowest NSES quintile (Q1), whereas the lowest rate of TO is seen is the second highest NSES quintile (Q4). The data suggests that the rate of textbook outcome decreases up until Q4; the rate of textbook outcome increases from Q4 to Q5.
Fig 1. The association between NSES and textbook outcome. Textbook outcome is denoted in percentage. Abbreviations: NSES = neighborhood socioeconomic status.
17 4.3 The socioeconomic gradient in comorbidity
In a similar manner to the previous socioeconomic gradient described, the neighborhood socioeconomic gradient in comorbidity can be conceptualized as the inverse correlation between neighborhood socioeconomic status and pre-existing comorbidity scores; as explained previously, these comorbidity scores are assessed using the outcome measure CCI. As can be observed in Table 1, a descriptive analysis of the data at population level fails to confirm the hypothesis that there indeed exists an inverse relationship between neighborhood NSES and pre-existing comorbidity scores (H2a). Surprisingly, the CCI is seen in the highest NSES quintile (Q5), whereas the lowest CCI is seen is the lowest quintile (Q1). Based on these findings, the data suggests that there exists a positive correlation rather than an inverse correlation between NSES and CCI. Below, I look at the relationship between the outcome variables of interest (TO and CCI) and predictor and contextual variables in greater detail.
Fig 1. The association between NSES and CCI. CCI represents the pre-existing comorbidity scores.
Abbreviations: CCI = Charlson Comorbidity Index; NSES = neighborhood socioeconomic status.
18 4.4 Statistical analysis
As previously described, in order to understand the association between neighborhood socioeconomic status and textbook outcome (favorable postoperative outcome), I conducted a logit regression analysis to further explore this association. To explore the association between NSES and CCI scores (pre-existing comorbidity), I conducted a multiple linear regression. The principal aim here is to analyze these associations while controlling for the relevant contextual variables of interest described previously.
The coefficients reported in both these analyses are in the units of log odds. Specifically, these coefficients indicate the amount of change expected in the log odds when there is a one unit change in the predictor variable with all of the other variables in the model held constant.
Model 1 depicts the results of the regression analysis while looking solely at the association between the predictor variable of interest (NSES) and the outcome variable of interest (TO or CCI), while excluding the effects of the contextual variables. Model 2 presents the results of these regressions by including the contextual variables in the analysis. In both these analyses, NSES Q3 was chosen as the reference category.
The first analysis, namely looking at the association between NSES and TO, will be discussed here. For the results of this regression analysis, I refer to Table 2. The first model shows that, with Q3 as the reference level, Q1 yields the highest (log) odds of a TO, as was also observed in the previous section and Figure 1. From the table we can observe that the predictor variable of interest, NSES, shows no significant association with TO. This is in accordance with the results presented in Figure 1, which does not seem to suggest a clear relationship between these variables.
In the second model, it can be observed that age has a significant negative correlation with TO
(-0.019; 95% CI [0.036, -0.0015]; p < 0.05). When comparing the Pseudo R
2for both models
(0.0028 and 0.017 for model 1 and 2, respectively), which is a goodness-of-fit measure for
logistic regression models, we see that model 2 has a higher Pseudo R
2value. By virtue of this
higher value (albeit still quite low), model 2 thus constitutes the better fit model.
19
Table 2. Logit regression model results between textbook outcome and the predictor and contextual variables of interest.
Variables of interest Textbook Outcome Model 1 Textbook Outcome Model 2
NSES quintile=1
0.30 0.33
[-0.23, 0.83] [-0.22,0.87]
NSES quintile=2
0.23 0.21
[-0.31, 0.76] [-0.33,0.75]
NSES quintile=4
-0.019 -0.033
[-0.54, 0.50] [-0.56,0.49]
NSES quintile=5
0.19 0.16
[-0.34, 0.72] [-0.38,0.70]
Age
- -0.019
*[-0.036,-0.0015]
Gender
- 0.073
[-0.28,0.42]
Diabetes
- 0.19
[-0.30,0.68]
BMI
- 0.038
[-0.00019,0.077]
CCI
- 0.0145
[-0.036,0.065]
Constant
0.35 0.38
[-0.018, 0.73] [-1.15,1.92]
Observations
585 585
Textbook outcome is denoted in terms of coefficients scales in log odds. 95% confidence intervals are shown in the brackets. Gender is a binary dummy variable, where males are denoted by 0 and females by 1. Model 1 Pseudo R2 = 0.0028; Model 2 Pseudo R2 = 0.017.
Abbreviations: BMI = body mass index; CCI = Charlson Comorbidity index; NSES: neighborhood socioeconomic status; TO = textbook outcome.
* p < 0.05, ** p < 0.01, *** p < 0.001
20
In the second analysis, I explore the association between NSES and CCI. For the results of this regression analysis, I refer to Table 3. The first model shows that, with Q3 as the reference level, Q1 yields the highest coefficients for the association with CCI, and this correlation is negative. To explicate, this suggests that, compared to the reference level of Q3, individuals in the Q1 of NSES have a lower value of CCI (pre-existent comorbidity). This observation is in line with what has been explained in the previous section and Figure 1.
Conversely, the highest quintile Q5 is shown to have a positive association with CCI. From Table 3, we can also observe that the predictor variable of interest, NSES, shows no significant association with CCI scores. This is in accordance with the results presented in Figure 1, which does not seem to suggest a definite relationship between these variables.
In the second model, it can be observed that both the age and gender variables demonstrate
significant associations with CCI (p < 0.001). First, age is shown to be positively associated
with CCI (0.11; 95% CI [0.089, 0.14]). Second, the female gender is shown to be negatively
associated with CCI (-1.20; 95% CI [-1.75, -0.64]). When comparing the adjusted R
2for both
models (-0.0045 and 0.16 for model 1 and 2, respectively), which is a goodness-of-fit measure
for linear regression models, we see that model 2 has a higher value adjusted R
2value and thus
represents the better fit model.
21
Table 3. Logit regression model results between CCI and the predictor and contextual variables of interest.
Variables of interest CCI Model 1 CCI Model 2
NSES quintile=1
-0.42 -0.46
[-1.37, 0.53] [-1.34, 0.41]
NSES quintile=2
-0.11 -0.15
[-1.06, 0.84] [-1.02, 0.72]
NSES quintile=4
-0.23 -0.22
[-1.17, 0.72] [-1.08, 0.65]
NSES quintile=5
0.085 0.42
[-0.86, 1.03] [-0.45, 1.29]
Age
- 0.11***
[0.089, 0.14]
Gender
- -1.20***
[-1.75, -0.64]
Diabetes
- -0.072
[-0.85, 0.71]
BMI
- -0.044
[-0.015, 0.10]
Constant
7.04
[6.36, 7.71]
-0.82 [-3.22, 1.58]
Observations
585 585
CCI is denoted in terms of coefficients scales in log odds. 95% confidence intervals are shown in the brackets. Gender is a binary dummy variable, where males are denoted by 0 and females by 1.
Model 1 adjusted R2 = -0.0045; Model 2 adjusted R2 = 0.16
Abbreviations: BMI = Body Mass Index; CCI = Charlson Comorbidity Index; NSES: neighborhood socioeconomic status.
* p < 0.05, ** p < 0.01, *** p < 0.001