Background

Adults, and particularly those who are older, often rate their health as very good despite multi-morbidity [1, 2], demonstrating their perception that the determinants of health and well-being extend beyond diagnosis and treatment of illness [3, 4]. If health is more than the absence of disease [5] what might maximize this state of well-being? We will explore the attributes of Canadians whose subjective rating of their health (SRH) and satisfaction with it are high, to identify characteristics of this group and, particularly, any that could be fostered via social policies or programs.

A strength of the indicator, SRH, is its potential to capture more than objective counts of diagnoses. Although such counts do predict longevity they give a limited picture of day-to-day health and well-being. Such a picture more clearly depicts how life is lived rather than only when it might end. SRH, at times, does present a composite picture of disease burden, mental health status, mortality, and the impact of social circumstances [3–9]. The subjective nature of the measure can allow for individualized responses that incorporate aspects of well-being and context [4]. However, that subjectivity also inserts an elasticity into characteristics considered in rating of health, producing inconsistent interpretations across individuals and groups. For example, SRH is often more predictive of mortality in men than women [10, 11]. Perhaps the experiences women consider when rating their health embrace social and contextual factors while men focus more specifically on number and nature of diagnoses [12, 13] The alignment of SRH and longevity or number of serious diseases also tends to vary with age. Older adults report relatively high SRH despite morbidity and looming mortality [1, 10]. These inconsistencies suggest complex variations in how perceptions of health, counts of diseases and lived circumstances differ among groups or individuals. Such inconsistencies can have a very real impact on subjective ratings of health.

What, then, might be a more universal measure of current health and well-being, one that is subject to less variability in interpretation of the question? There is evidence that satisfaction with life and health measures something different from SRH [14] but also aligns with current and subsequent health outcomes [15] and mortality [16]. Of importance from a policy perspective, life satisfaction is a health asset [17] that can be modified to improve physical health and well-being [18, 19].

Adding responses about satisfaction with health (SH) to SRH may insert aspects of lived realities overlooked by some when rating their health, and thus correct for inconsistencies noted earlier (e.g. by age and sex/gender) in how subjective health is reported. We use the term, sex/gender here as there is almost certainly an interaction between biologic sex and the lived social realities associated with being a man or a woman that we will refer to as gender [20, 21]. Combinations of SRH and SH might provide a broader and deeper picture of characteristics of those with highest reported current health and well-being, characteristics that augment health assets and that can be fostered.

There is extensive evidence that economic deprivation is among the strongest risks to health [22, 23], although less so for women than men [24] Equalizing socio-economic status (SES) is, however, neither politically nor practically straightforward [25]. Of particular interest to policy-makers are health promoting circumstances or qualities that are malleable. For example, there is a nascent literature demonstrating that individual resilience is tied to satisfaction with life and health, can promote health, well-being and longevity [26–29] and can be fostered throughout life. Multiple definitions of resilience exist. In this context we define it as the process of positively adapting to adversity, trauma, threats, or significant distress [28].

Determining the characteristics of individuals reporting high SRH and high SH, particularly those that can be augmented via individual or social interventions without large political shifts, could guide policies and practices that extend beyond medical prevention and treatment. Despite a large body of evidence on predictors of SRH, to the best of our knowledge there is no study that directly identifies characteristics of individuals with high levels of both SRH and SH.

We aimed to identify medical and sociodemographic characteristics of Canadian adults with the highest SRH and SH, that is, those whose health and perceptions of well-being in the broadest sense, are greatest. This is novel and exploratory research, that by classifying a whole population into subgroups with the highest probability of the outcome, examines whether that subgroup’s characteristics can be fostered in others. Our methodology was intentionally chosen to make no a priori assumptions, treat all factors equally, recognize the interconnections of characteristics, and let the data ‘speak’.

We used data from the Canadian General Social Survey (GSS) to 1. examine the interplay between self-rated health and satisfaction with health, and 2. identify potential predictors of a high rate of both SRH and SH. The GSS is a well-established national survey that has been used to examine the impacts of caregiving [30], social capital [31], physical activity [32], and immigration status [33] on various aspects of health. To the best of our knowledge, no multi-dimensional models acknowledging the interconnectedness of SRH and SH have been described using GSS data. We performed Classification and Regression Tree (CRT) modelling to identify predictors of a high rating of this combined measure of SRH and SH. This exploratory methodology can identify interconnected (or intersecting) and non-linear relationships among social and medical predictors [34–36] by classifying the survey population into any combination of characteristics that predict the outcome of interest. As an example, it is possible to determine whether different combinations of SES, resilience and ability to work shape SRH/SH differently. This analytic approach enabled identification of specific behaviours, groups, or sub-groups for whom interventions could improve overall health and wellbeing.

Methods

Results

We present details of findings here and a summary of their meanings in the discussion. Out of the 19,609 original participants, 124 participants were excluded because of missing data on the outcome variable. We started the analysis with 19,485 data points. Table 1 shows descriptive statistics for the variables used in this study. About 55% of the survey population included were female. 29% older than 65, and 78% lived in urban settings. Notably, 85% and 72% of people surveyed reported SRH and satisfaction with health (SH) as good, respectively. Before combining SRH and SH variables to generate a composite measure for our outcome, we examined their bivariate relationships with sex. Both showed significant associations according to the results of Chi-square test (p. value for SRH <0.001 and for SH = 0.003).Bivariate analysis (Table 1) demonstrated that the two outcome groups differed significantly in the distribution of all explanatory variables, with the exception of sex, in particular, but also population centre indicator, and importance of religion/spirituality.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Distribution of variables for the whole population, those with good SRH and high SH, and those with poor SRH, SH, or both.

https://doi.org/10.1371/journal.pone.0293976.t001

We next examined congruence between reports of SRH and SH by performing a simple cross-tabulation (Table 2). Most participants (81%) had congruent assessments of SRH and SH, 69% reported good for both measures and 12% reported poor self-assessments of the two indicators. However, a small proportion had either good SRH, but were dissatisfied with their health (16%); or poor SRH, but were satisfied with their current health status (3%).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Cross-tabulation of self-rated health and satisfaction with health.

Greyed categories were grouped for the CRT analysis.

https://doi.org/10.1371/journal.pone.0293976.t002

Guided by Table 2 because there were relatively small numbers of individuals with discrepancies in their reported SRH and SH we concluded that generating a regression tree with four outcome groups, though technically possible, would have been very unstable and hard to interpret and thus excluded from our analysis plan.

The predictors of good SRH/SH identified by the regression tree and their relative importance are shown in Fig 1. Healthy eating was the first splitter and identified as the most important factor for the outcome. Occupation, despite appearing first in the third level of the Tree was pivotal in generating many nodes. The regression tree correctly classified 93.3% of participants with positive outcomes and resulted in 24 terminal nodes, ten of which were deemed important. Correct classification for those with negative outcome was 35% yielding a total correct classification of 76%. We defined important nodes as those subgroups with a frequency of reporting good for both indicators more than 20% different from the rate for the whole sample. Since 68.9% of the total population reported good SRH/SH any nodes that reported the rates of this positive outcome larger than 83 or smaller than 55 were deemed important. Summary characteristics of these 10 important subgroups are described in Table 3.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Variable importance to the model.

Assigned importance and normalized importance to the model for each variable included in the final CRT model.

https://doi.org/10.1371/journal.pone.0293976.g001

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Description of important subgroups identified by CRT.

https://doi.org/10.1371/journal.pone.0293976.t003

Eight important nodes identified the subpopulation with poor SRH/SH with health (between 19.6% and 46.6% reported the positive outcome lower than the whole population rate of 68.9%) in comparison to the whole survey population and two nodes (node 35 and 38) identified subpopulations with reported better SRH/SH than the whole sample (68.9%; node 0, Fig 2). The first five branch points of the tree were: (1) healthy eating; (2) mental/psychological disability; (3) perceived social class; (4) ability to work in the past 12 months; and (5) resilience. The remaining variables identified by the regression tree were measures of socioeconomic status such as education and living arrangement, as well as behaviours such as use of the internet, smoking and alcohol consumption. These were lower down in the tree indicating they were of less importance to ratings of health and satisfaction with health.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Classification regression tree analysis.

Outcome is defined as good SRH/SH vs. any poor rating of SRH or SH. Maximum number of branches capped at 6. Minimum parent node size of 100. Minimum child node size of 50.

https://doi.org/10.1371/journal.pone.0293976.g002

Two pathways that lead to terminal nodes that identified subgroup with higher than total population rates of good SRH/SH were mostly defined by perceived healthy eating, absence of psychological disability, middle social class, and high resilience (nodes 35 and 38). Other factors such as smoking and internet use also had some, although limited predictive impacts.

Perception of dietary health emerged as an important predictor. Six (out of eight) subgroups with lower perceived health reported poor diet. Subgroups with good diet but lower SRH/SH suffered from other health issues such as mental disability, inability to work due to illness, and low resilience (nodes 10 and 28). Notably, the characteristics of limited statistical importance, that is, of limited predictive value, were sex, population centre indicator, smoking, and religion/spirituality activities.

Discussion

We found that most Canadians surveyed had good SRH and congruent satisfaction with their health. However, the approximately 30% who reported either poor SRH and/or poor SH could benefit from interventions that improve health and well-being. This study is the first to identify predictors of SRH/SH considered together and in a general population. Others have used ordinary multivariate regression analyses to examine the independent importance of medical, behavioural, social and economic variables [3, 38–40] in the perception of health. In view of the complexity of relationships among characteristics that shape SRH use of simple regression has been criticized [41] regression tree techniques are a means of better exploring this non-linear complexity [42].

Combining SRH and SH has the potential to explain inconsistencies and lack of reproducibility of SRH when subgroups such as women and men are considered, and to demonstrate the complex interplay of multiple health-related, behavioural and social characteristics. A comprehensive review of sex differences in all studies of SRH is well beyond the scope of this paper. In general, though, findings of differences in men’s and women’s SRH have been consistent but also confusing in two particular ways. First, while sex differences in SRH are the norm, uniformity in which group reports greater subjective health varies [1, 6, 8]. Using GSS data there were significant differences in bivariate analyses of sex and SRH, alone, (data not reported) as there were with sex and SH. However, sex was not significantly related to the outcome of SRH combined with SH in bivariate analysis (p = 0.295) nor was it identified in the regression tree as a predictor. As a sensitivity analysis we also constructed sex-stratified regression trees. Results were similar to those for the whole sample (data not shown). The lack of significance of sex as a predictor of the combined outcome of SRH/SH suggests that adding satisfaction with health may correct for sex/gender differences in interpretation of the meaning and, hence, rating of subjective health. The second way in which adding SH to SRH may deepen meaning has to do with women’s frequent although not universal reports of poorer SRH relative to men, alongside their greater longevity [10–13]. This paradox also raises questions as to the meanings men and women attribute to the measure, SRH [43]. In keeping with the findings of others, perceived healthy eating [38, 39], absence of mental health issues [44], and ability to maintain a meaningful occupation were the strongest predictors of good SRH and good satisfaction with health. Others have demonstrated that perceptions of healthy eating are strongly associated with socioeconomic status [45, 46], suggesting that this first branch in our CRT analysis might be a proxy measure for socioeconomic effects. This is reinforced by the finding that employment can overcome perceived poor eating habits and tip the balance toward reporting of good SRH/SH (node 12).

Also described by others, resilience, or the ability to overcome life’s challenges and thrive emerged as an important predictor of good SRH/SH [47]; however, this effect was more nuanced than has been previously noted. High resilience is responsible for both important subgroups that identified subgroups with high rates of good SRH/SH (nodes 35 and 38). Even with low-moderate resilience, we found that participants might overcome other predictors of poor SRH/SH by having an active occupation or a perceived social status greater than middle class (node 30). Consistent with the literature [40], this suggests that predictors of poor SRH/SH might be modified by one’s socioeconomic status. Another related social factor that we identified was occupation. With a similar moderate and low level of resilience those with an active occupation showed 80% rates of good SRH/SH (node 18) whereas only 20% of those with no occupation secondary to illness rated their health/satisfaction as good (node 28).

Behavioural variables such as smoking, alcohol consumption and internet use were identified further down in the regression tree. Due to their limited importance they will not be discussed here to avoid potential over-interpretation of an exploratory analytic method.

Conclusions

Sex differences in subjective health ratings disappear when measures of SRH and SH are combined. This combination may capture day to day experience along with medical circumstances and produce a more comprehensive picture of well-being. SRH/SH seems to correct for varying interpretations of the meaning of self-rated health, alone. The disappearance of sex differences suggests that components of men’s and women’s different reporting of subjective health may be an artefact of definition. We do acknowledge that despite the intersectional analysis inherent in decision tree designs there are almost certainly some gender differences (that is, intersections of sex and social circumstances) that predict SRH/SH and have gone unmeasured in this study. Our findings were that the interplay of physical health, mental health, behaviour and socio-economic status, but not sex, shape perceived health and satisfaction with it. Particularly diet, resilience, and ability to work and cope with life stressors were strong predictors of good health and satisfaction with it. There is growing evidence that policies requiring limited political or economic upheaval can foster resilience among adults [49]. CRT analysis allowed us to identify complex, non-linear relationships that would not have emerged using classical multivariable regression analyses. Future studies of SRH and SH could continue exploring these nuanced, intersectional relationships to better guide public health policies and avoid putting individuals into an ‘all or none’ basket, that is, avoid assuming homogeneity within groups such as women or men.

Supporting information

Acknowledgments

We acknowledge the participants of the GSS survey, without whom this research would not be possible.